ML20199B027

Vol 1, Appendices E,F & G to DOE/ID-10541. Page F-4 of Incoming Submittal Was Not Included

ML20199B027
Person / Time
Site:	05200003
Issue date:	10/31/1997
From:	Theofanous T CALIFORNIA, UNIV. OF, SANTA BARBARA, CA
To:
Shared Package
ML20199B025	List: ML20199B027 ML20199B046 NSD-NRC-97-5434, Forwards Vols 1, Appendices to DOE/ID-10541, 2, Addenda to DOE/ID-10541,-10503,-10504 of DOE Rept,Dtd Oct 1997 & White Paper, Effects of Containment Spray on AP600 Level 2 PRA Hydrogen Analysis
References
NUDOCS 9711180193
Download: ML20199B027 (463)

Text

__ -. . _ . _ _ . _ . . . _ _ _ _

lC '

VOLUME 1 APPENDICES E, F, AND G October 1997 TO DOE /ID-10541 T. G. Theofanous, et al.

O l

Advanced Reactor Severe Department of Energy Acciderit Program O

= wuna,"

A pm ,

f' i

I ARGONNE NATIONAL LAB [$dTORY l 9700 South Cass Avenue, Argonne,blinois 60439 Phone: (630) 252-4595 l Fax: (630) 252 4780 September 25,1996 - - .- . -- - _ - - . _

O RECElVED '

RIACTQn E!:GlHEat,G r P/;s:cN '

-D'OECTOT5 OFF:U.~

Dr. L. W. Deltrich Reactor Engineering Division SEP 261996 -

Argonne National Laboratory fcm _ , gg ta ,

9700 South Cass Avenue - 7 ____

Argonne, Illinois 60439 pm.gi , t -- -F g- ;-

Dear Walt:

.. m "

[

In response to the request made in your letter of June 17, 1996, I have uviewed the report

" Lower Head Integrity Under In-vessel Steam Explosion loads" by T. G. Theofanous et al. You indicated that this report and a companion document together " intend to demonstrate the effectiveness of 'in-vessel retention' as a severe accident management concept for a reactor like the AP600". You further indicated that "the purpose of this review is to assess whether this intent has been achieved to a sufficient degree for the results to be of use in the regulatory / licensing area". Based on my r-view of the report, I fmd that additional supporting work would be needed if the concluskas of the report were to be used in the regulatory / licensing area.

On page 91, the authors state that "Me'hodological:y, the assessment involved only a slight O scenario dependence, principally on the permanence of the blockages preventing direct downward, through the lower core support plate, relocation", and that "thus the assessment is of Grade B, in the ROAAM scale". I think the scenario dependence is more than slight, so the assessment may be more of Grade C than Grade B in the ROAAM scale. Suppose a steam explosion would take place in the downcomer region or in the lower plenum, as described in the report. The l

explosion may not be strong enough to fail the lower head, but it may be energetic enough to l mechanically disrupt the blockages formed at the lower end of the core. Further, the explosion would likely expel some water from the lower plenum so that the Icwer core support plate may no longer be in contact with water (i.e., the ability to cool the core support plate would be lost).

Thus, the initial explosion, while not failing the lower head, could severely weaken the blockages l mechanically as well as thermally. It would seem possible that a relatively'small initial explosion l would be followed by a massive downward relocation of core melt through the core support plate, i setting the stage for a secondary explosion probably involving a much larger melt mass. The lower head may well survive such a secondary explosion, but a separate assessment of this possibility would defuitely be needed, i

Based on the code calculations performed, the report coactudes that the saturated coolant condition in the lower plenum leads to highly voided premixtures that have a dampening effect on the resulting explosion energetics. While I am not judging the validity of the calculations, I fmd it difficult to recone'le this conclusion with available experimental evidence. Exp:rience tells us that triggering of a steam explosion would be more difficult with saturated water than with highly submled water. However, once triggered, the explosion energetics does not seem to depend on E-37 Operated by The University of Chicago for The United States Department of Energy

i O

VOLUME 1 APPENDICES E, F, AND G October 1997 TO DOE /ID-10541 T.G. Theofanous et al.

O Center for Risk Studies and Safety Departments of Chemical and MecLaical En;incering University of California, Santa Barbara Santa Barbara, CA 93106 Prepared for the U. S. Department of Energy Idaho Operations Office Under ANL Subcontract No. 23572401 v

l 4

+

t CONTENTS APPENDIX E. REVIEW COMMENTS IN THE ORIGINAL - . . . . . . . . . . . . . - . . . E-l APPENDIX F. POINT BY.POIKr RESPONSE'iD REVIEW COMMENTS . . . . . . . . . . F 1 APPENDIX O. COMMENTS AND RESPONSES CLASS!FIED BY TOPIC . . . . . . . . . . O.1 .

2 2

s I

s e

6 i

U l

lil 1

>~

?]?

- . n, , , , , + .m , , , . - , , , , - y . . . , a ,w,. , , .., , --,n, s,---

APPENDIX E .

O REVIEWER LETTERS IN THE ORIGINAL TABLE OF CONTENTS E.1  : Bankoff, S. G. (Northwestern) .................... E-3 E.2 Berthoud, G. (CEA Grenoble) . . . . . . . . . . . . . . . . . . . . . E-8 E.3 Burger, M. (U Stuttgard) . . . . . . . . . . . . . . . . . . . . . . . E-15 E.4 Butler, T. (LANL) . . . . . . . . . . . . . . . . . . . . . . . . . . . E-30 E.5 Cho, D. H. (ANL) . . . . . . . . . . . . . . . . . . . . . . . . . . E-37 E.6 Corradini, M. L. (U Wisconsin) . . . . . . . . . . . . . . . . . . . . E-39 E.7 Fauske, H. and Henry, R. E. (FAI) . . . . . . . . . . . . . . . . . . . E-47 .

E.8 Fletcher, D. E (U Sydney) . . . . . . . . . . . . . . . . . . . . . . E-50 E.9 Jacobs, H. (KfK-INR) ..................... . . E-69 E.10 Mayinger, E (U Manchen) . . . . . . . . . . . . . . . . . . . . . . E-78 O

(/ E.11 Moody, E J. (GE) . . . . . . . . . . . . . . . . . . . . . . . . . . E-85 E.12 Sehgal, B. R (RIT) . . . . . . . . . . . . . . . . . . . . . . . . . . E-94 E.13 Shewmon, R (OSU) . . . . . . . . . . . . . . . . . . . . . . . . E-112 E.14 Turland, B. D. (AEA) ....................... E-113 ,

E.15 Young, M. R (SNL) ........................ E-129

ri dj ,

E-1

. . __ _. _ _ _ _ _ .m _-_ _ _ _ . _ - _ - . _ _ _ _ . _ . _ _ . _ _ _ _ _ . , _ . . _ .

e 4

NORTHWESTERN UNIVERSITY 2145 sHERIDAN ROAD EvANsToN ILLINOIS 60208-3120 usa

s. oEoRoE BANKoFF TELEPHONE: (847) 491 5267 WALTER P. MURPHY PRoPESsoREMERrIVs FAX: (847)491 3728 E Mail (gbankoff 9 casbah acas.aws.ede)

CHEMICAL ENoINEERINo DEPARTMENT August 23,1996 Dr. L W. Deitrich, Director Reactor Engineering Division Argonne National Laboratory 9700 South Cass Avenue Argonne,IL 60439

Dear Walt:

I enclose herewith my review of DOE /ID-10541. I had to read the suppciting documents as well in order to get the necessary perspective. In the process I spent 32 hours3.703704e-4 days <br />0.00889 hours <br />5.291005e-5 weeks <br />1.2176e-5 months <br />, which convinced me that in-vessel retention is a valid concept for the AP-600.

With kind regards, W '

S. G. Bankoff SGB:hke .

Encl.

c RECElVED i REACTOR ENGINiiERING DIVISION .

-DIRECTOR'S OFFICE-O ACTION:

AUG 2 91996 Ls) i E-3 33tq,w,7,an; ca pyx _ e,,g .

kc%D a

._.._ . - _. If

REVIEW OF DOF1ID-10541 AND SUPPORTING DOCUMENTATION S. G. Bankoff

GENERAL COMMENT

S The principal documents which were read by the author were DOESD-10541 (June 1996), DOE-10460 Vols. I and 2 (July 1995), and DOE 10849 (Jan.1995), as well as various papers published and/or presented by Prof. Theofanous and one or more of his co-authort.

l i

My general impression is that this is a massive piece of work, which attacks all aspects of the l steam explosion problem in the Westinghouse AP-600 reactor, and conclusively demonstrates that failure of the vessel, to say nothing of the containment,is physically unreasonable. If no failure occurs in the reactor vessel, essentially no release can occur to the inside of the l

l containment building, and hence the threat to the public health and safety is eliminated. In the' process of developing the evidence in terms of focused experiments, covelopment of new and improved codes, tying in work done around the world, and developing a methodology for assessing the safety goals and margins for rare, but high-consequence, hazards, a set of tools .

has been developed which represents a huge step forward in examining severe accidents in new types cf advanced nuclear reactors and ir existing nuclear reactors.

In other words, in execution, scope and potential consequences, the total of this work represents a very important achievement.

SPECIFIC COMMENTS l 1. I think that the ROAMM approach makes very good sense for rare, but high-c;.' sequence, events. I believe that a similar approach has been used before, but never so E-4 i

I

l explicitly and clearly stated. In particular, the_ recognition that_ there are " intangibles" which

- will never be known in advance, conservatively bounding them at each stage, and then -

L enveloping the pdf passed on to the next stage, makes the uncenainties clear.

2. . There is no estimate of the conservatisms induced by ignoring fluid-structure E interactions, assuming plastic flow with no strain hardening, and using a much lower yield

. stress (330 MPA) compared to the measured yield stress (450 MPA). Such estimates would

be helpful.- .

3. - All the codes (ESPROSE.m. ABAQUS and PM-ALPHA) appear to be well-tested ,

against available data, and have been conservatively adjusted.

4. I would think the result that the greater the load localization, the smaller the effective impulse, has practical limits. For a perfectly plastic material, a delta function load (point i

~

f- source) will always penetrate continuously. The introduction of bending moments to spread the load, of course, represents the real situation. However, how far does the use of the

l. effective impulse go for failure criterion with a very concentrated load?

5. The net result, from the axisymmetric and non axisymmetric calculations, that the details of the loading pattern are not particularly important is, in itself, very important. This result would seem to hold not only for the AP-600 reactor, but for all reactors, and greatly reduces the probability of local failure.

6. All the premixing work that I am familiar with (including my own) assumes that a i

corium jet falls into the lower pool by melt-through of the'LSP (lower support plate). A

- great-deal of work has been done (and is contemplated for the future) on jet breakup, with various materials, jet diameters, temperatures and velocities, that is predicated on this assumption. The unexpected result that for the AP-600 melt-through will occur through the E-5

side, rather than the bottom, of the melt pool, is therefore of Hrst importance. Of course, this happened at TMI, but in AP-600 there is a thick stainless-steel reflector around the core. The 1 code capability allows calculation of the subsequent premixing and melt / water dynamics.

There can be no large melt jet in transit; stratification appears very likely even in transit; and the whole scenario of damaging multiple explosions disappears.

7. The simple models for blockage formation and blockage coolability, leading to non-availability of downwards relocation paths, and transition to a molten pool, are made credible because of the relatively flat radial power distribution in the AP-600 design. This lumped approximation would have to be re-examined for other reactor designs.

8. The calculations of melt length scales and local void fractions lead to quantitative results which are more realistic and detailed than previously available. As expected, liquid water is rapidly depleted from regions of high fuel concentration, and the boundaries of such regions can be quite sharp. Board-Hall thermodynamics theory for steady plane shocks, previous multiphase calculations of the flow fields behind the shock front (Sharon and

.Bankoff) agree that regions of large void content cannot sustain shock propagation at supercritical pressures. This is the principal reason that the SERG-2 panel felt that the cx-mode failure was not physically reasonable. Precisely the same results are obtained by the PM Alpha and ESPROSE.m calculations. .

9. The residual uncertainties proposed in NUREG-1524 were jet breakup triggering,2D vs. 3D codes, and chemical augmentation. r or the AP-600, jet breakup is no longer a major concern, as discussed above. The 2D vs. 3D controversy is no longer relevant, since validated 3D codes are now available. Chemical augmentation w'th  ; the real corium produced in the reactor will have no important effects. Triggering is the sole intangible which will O

E-6

~

1 never be known for a real accident. However, it is irrelevant if it is assumed that triggering alwayt occurs at the worst time and place, and the result is evaluated by' energetics, which is 1

to say the validated codes. The approach taken of triggering by setting one mesh to a high

-initial pressure seems to me to be a perfectly valid procedure.

SUMMARY

' Convincing arguments have been addressed, backed up by a huge volume of high-quality experimental, analytical and computational work, that the AP-600 reactor will not fail 1

l In the course of a severe accident. This implies that all later scenarios of containment-building pressurization and beat-up are no longer necessary. In my opinion, this closes the severe accident scenario for the AP-600, and leads to consideration for licensing. The consideration of other reactor types, on the other hand, does not appear to be so -

straightforward, and further work needs to be done.

.1:

w E-7

. - . ... _. - . , -- . = - ~ . - -

%DdE9?

bk NOR.'HWESTERN UNIVERSITY 2145 sHERIDAN ROAD EVANsToN,!!11NolS 60208-3120 usa .

- S. oBoRoE BANKoPP TELEPHONE: - (847) d915267 ' ,

WALTtWt P. MURPHY PRoPES$oR EMERITUS FAX: (847) 491 3728 CHEMICAL ENotNEERINo DEPARTMENT E-Mail (gbankoff ecasbah.acns.awu edu) 5 October 22,1996 Professor Theofanis Theofanous Dept. of Chemical and Nuclear Engineering -

. . University of California

' Santa Barbara, CA 93106

Dear Theo:

Enclosed is my review of the two additional documents, which I spent an additional 6 hours6.944444e-5 days <br />0.00167 hours <br />9.920635e-6 weeks <br />2.283e-6 months <br /> in reading and comparing reviews. All together, it is an outstanding

_ piece of work.

-. Sincerely,

.I b S. G. Bankoff alf 'f SGB:hke Encl.

1 E-7.1

, , , , - . . y - - - , - _ . _ - - ,_ , . . _ . - - - . ,-,-,m.r,

SUPPLEMENTARY COMMENTS ON j

THEOFANOUS, ET AL. SEVERE ACCIDENT DOCUMENTS i

1. DOE /ID-10503 " Propagation of Steam Explosion: ESPROSE.m Verification Studies", by T. G. Theofanous, et al. Aug.1996, and update of Sect. 4.2.1 This is a convincing document, laying out the evidence that ESPROSE.m has the-capability successfuhy to predict various shock and vapor explosion scenarios. These range from simple steady, one-dimensional shocks propagating through single-phase liquid and homogeneous gas-liquid mixtures, for which exact (or nearly exact) solutions can be found, to experimental shock and expletion data in the SIGMA and KROTOS facilities. The 1-D wave dynamics were tested for shock speed, fluid velocity, reflection at a rigid wall and reflection at a free interface with venting for single-phase, liquid-air and liquid-vapor cases, using ESPROSE.m and CHAT. The small deviations between the analytical solutions and the codes can be attributed, at least in part, to the fact that the analytical solutions used an assumed constant sound speed in the liquid, taken as 1500 m/s and 2000 m/s, while ESPROSE.m used the real properties of water. In some cases, as in the deviation in reflected shock speed, the differences in the analytical predictions for the two assumed sound speeds is considerable, but ESPROSE.m gives a smooth function of shock pressure which interpolates between the two limits, and is hence more credible than either of the two analytical shock reflected shock speeds. As a check on the wave dynamics with reflection / transmission at interfaces between two materials of different acoustic impedance, which govems the unloading-explosion coupling near a free interface, the CHAT code, using the method of characteristics, was written. For large amplitudes the quasi-linear code CHAT-QL cvaluates the coefficients in terms of the local fluid properties. These codes were then compared with ESPROSE.m for O

E-7.2

g pressure and velocity distribution for 1-D single-phase venting and for shock speed, Guid 1

i ')

velocity, reflected shock speed and shock amplitude for a 10% void non-condensible J steam / water mixture, with e3 :llent results. This sort of independent cross-checking lends considerable ccnfidence in tie basic structure of the ESPROSE.m code, which is a Snite- i difference code in laboretory coordinates. Exact sohtions in 2D geometries are then given for infinit: pool, cylinhical open pool and cylindrical closed pool geometries, by superposition of an infinite array of sources and sinks in order to obtain reflection and transmission behavior at rigid and free boundaries. These solutions are in themselves impressive, and likewise the general agrer: ment between the code and analytical predictions for pressure as a function of time over the two-dimensional region. In fact, the agreement between the two predictions for the centerline pressure distribution as a function of time and distance is remarkable.

V There is also good agreement between the ESPROSE.m prediction and the experimental data in the SIGMA facility with all liquid, and with a liquid-air mixture in the expansion section. The key parameters of time of arrival and amplitude of pressure waves at several locations are well-predicted, particularly in view of some high-frequency ringing in the experimental transducers. The same is true for multi-region runs with high pressure (AP

= 68 and 1:6 bars) with different initial void fractions.

One aspect of vapor explosions which is difficult to model properly is the presence of strong energy sources, especially near a free boundary, which are caused by local explosions (rapid mixing) of fuel drops produced in the course of jet breakup. These sources can distort venting and reflection phenomena near free boundaries. This was modeled first by

/T characteristics solutions with single internal heat sources, with the energy assumed to be

• D .

E-7.3

going only into the vapor, with heating rate increasing with velocity:

Q, a Cu'd (3.2) with C being an empirical constant. More explanation is needed for the assumed form of this equation and the magnitude of the exponent, based on energy dissipation. However, excellent agreement is obtained for various assumed values of C beiween CHAT and ESPROSE.m.

Ilowever, comparisons with data for single exploding drops are lacking.

At the other extreme is a plane shock wave moving into a fuel / steam / water nixture at low pressure. This is the scenario envisaged by Boarc and Hall, following the one-dimensional theories of Lifshitz and 2L.avvich. In the B-H theory the average specific volume of the mixture is plotted against pressw t, starting with adiabatic compression to a peak pressure (von Neumann point) at which the steady mass-continuity condition for the flow behind the shock front is satisfied with minimum entropy generation. This results in the Chapman-Jouguet (C-J) condition of tangency to the reaction adiabat, er Hugoniot curve, leading to a minimum of shock speed with pressure. The ESPROSE.m calculation shows the development of the shock into a steady-state detonation, using a fragmentation rate given by Fr=ph E

^* (4.1) where pi is the local macroscopic fuel density, C is the " expected" shock speed (1500 m/s), and Ax is the grid size, Why the fragmentation inte should be a function of ax is not clear, nor is the form of this equation. More explanation is needed. However, detailed comparisons between the 2D and 3D versions of ESPROSE.m are encouraging, and the ESPRGSE.m P-V lines are close to the expected Rayleigh line, and bounded by the shock adiabat.

O E-7.4

These results are excellent back-up for the interpretations of the KROTOS experiments O given in DOE /ID-10489. All in all, even at this stage of limited comparison with integral explosion tests, one has confidence in the predictions of pressure-vs-ume at various locations on the pool boundaries, and consequently of the initial kinetic energy of large masses impacting on reactor structure.

Typos:

p; l 1 verified Fig. 1 , viscosity 9

4 3 and ff Hugoniot von Neumann C-J point O 2. DOE /ID-10504 " Premixing of Steam Explosions: PM_ ALPHA Verification Studies",

Sept.1996.

This is, once again, a thorough, and highly professional, document on the verification of the PM_ ALPHA code against available experimental data and known physics. The agreement with a wide range of data, from single particles settling in water, to panicle swarms, both cold and hot (up to 2000* C), to integral tests with prototypic materials at high pressure, is rather remarkable. The 'oreakup constant , has been chosen to fit the integral data from several tests, but it is used consistently. The Richardson-Zaki exponent for a monodisperse system of spherical particles has been used without modific ion for thermal effects, with excellent results. The FARO experiments, which gave very little usable data on i- the jet breakup and ' dispersion, has been well-approximated for the measured steam flov> and pressure. The overall result is that the code seems to be well suited for licensing purposes.

. E-7.5

llowever, some specific comments may be made and questions raised:

1. Eq. 2.3 This equation is incorporated into the code, but no independent check on the R Z exponent is made. On the other hand, the R Z exponent was chosen by comparison with a large body of data on systems of spherical particles.

2. Figs. 9 and 10 (p. 210) are !nteresting in showing decaying oscillations, and an attractor above, but close to, the steady drift flux / particle volume fraction curve. This physics appears to be new, and should be further investigated.

3. p. 216. The careful treatme.it of the radiation boundary condition with slight subcooling is noteworthy.

4. Figs. 8 and 9 (pp. 2 30/2A1). The comparisons between the predictions for front poiition and the da,a for the QS - Q11 experiments in the QEOS series is ren.arkable. The level swell is not well predicted in the first 0 ls, but this may be due to experimental uncenainty, For the important range t > 0.2s the agret :nt is excellent.

5. The explanation for the absence of a pressure hump at early times compared to measurements for Q17, as being due to the extra radiant heating before impact.seems to me to be reasonable.

6. Turning to the MIXA experiments, there is reasonable agreement with the pressure data, and excellent agreement for the cumulative steam flow.

7. Similarly, there is remarkable agreement with FARO L 14 water level swell, pressure, and pressurization rate, especially considering the complex geometry of the equipment. The additional information from the code on local void O

r )

fraction, melt temperature, melt volumetr;c fraction and melt location seems to )

i 1

me to be very useful, in view of the inherent limitations of the experiment, j

' My own view is that the cost benefit ratio for further experiments of this sort is l i

i large. .

a ,

1

. i i

b I

l e

T 4

. .v f

.i j

+

L E-7.7 i e

5

.. ; . _ _u__...,.._...__..__.,__. ___________._.._-___.___...___.-._._.,....,__.j

i

-- A twnee asessem I DeeGCDON DSS kAACT9Ung pm aamag comtutNT og THenecawouutout ti or rmsout sttvict et THttucenouuuout Pova u s mucAtoN$ INDusTRituts (AnoaArom triTuots roNDAMENTALf5 b.W. DCIlfiCh Argonne National Laboratory

'swM '**" " ' Building 208, Room C213 ARGONNE, Illinals 60439

,,l ,,

m' USA ftL F4 H 30 64 '

9A2 F4 H 30 3e November 20,1996

Dear Doctor Deltrich,

Please find enclosed my review concerning the document a tower Head Integrity under In vessel Steam Explosion Loads e by T.G. Theofanous et al.

I must confess that it took me more time than the allocated 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br />.

I am also sending an invoice corresponding to a 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br /> work to the University of California to M. Vaughn Boyle.

Sincerely, G. Berthoud Chef du Laboratoire d' Etudes Fondamentales Service de Thermohydraulique pour les Applications Industrielles N E C E I V E T-i- .

j REACTOR ENGINEERiflG D!V: ';ll 1, -D!RECTOT3 0 ." CE- ,

NOV 261996 l

{ ACTicPh _ _ jg _

lhf0AMATl0N A w y) 4 }{ *~ -

e -

"2'* W ' s- t n .>> _ , _

~ ~ ~

E-8 ---- -- - .

et A / omnoble . it we oss manms cenas

Comments about the document c Lower liead Integrity under in vessel Steam Expthsl[n Loads o by T.O. Theofsnous et al.

G. Berthoud CEA/Grcnoble O

O This document presents an analysis of the potentiality oflower head failure of the AP600 resulting fro:n a Steam Explosion. The conclusion that the risk is negligible (a physically unreasonable *) is quite convincing and is based on:

1. the fact that water will be saturated and at I bar due to complete depressurisation to the containment pressure and that these conditions will lead to large and rapid voiding which is not favorable for large S. E.

2. the fact that we have permanent blockages at the bottom of the core that will impeed any coherent relocation through the core support plate

3. the fact that relocation will occur sideways throu'gh the reflector and core barrel an'd so that the Steam Explosion will occur in a 3D geometry without any large constraint allowing large sustained pressure

4. the fact that - even if reflooding is taken into account - when the melt will be ejected sideways, we will have enough time to heat the added water up to saturation and so to prevent good mixing.

The vahdity of the conclusion is then linked to the validity of the above four arguments.

As for the first argument,there is no doubt that water will be saturated as far as reflooding is not taken into account, the fact that the pressure will be atmospheric cannot be discusred here as this is justified in another report (IVR Report - table 7.3) however, I think that this has to be justified as the voiding will be less important at pressures a little bit higher, around some bars. At these pressures, we can also recall that it was found it was easier to trigger an explosion in the single iron oxide droplet experiments of Nelson in Sandia.

WATER SUSCOOUNG (K) 70 s7ps 130 148 1se tta 11 ,i, I, "l i i

~

1.4 -

Fia. 8 : Reistion entre le 1.4 , -

" trigger" ndeessaire au declen-1.a , - chement d'une interaction et la pression ambiante (cas de gxttes sanosion d'oxyde de for de 2,9 mm ,de i 1.0 0.8 3.4

/

-- * *eo --- -- -

diam &tre 1 2230*K tombant dans de l' eau a 298'K) 10.4 -s , f

/ -

p)

( 0.2 le e- g--- g--

e / 800 aRMossoses

ho .

0.0 O.0 0.2 0.4 0.8 0.s 1.0 1.2 1.4 AussNT PRESSURE (wa) E9

The fact what we have up to now no evidence of explosion in experiments using reactor like materials (Krotos)(and that this is due to the non occurrence of good mixing) is stressed by the i I

authors. But once again these Krotos experirnents are performed at i bar pressure while experiments at pressures of 50 and 20 bars, with saturated w4ter some mixing was obtained. In a near future a Faro experiment using initially saturated water at 5 bar will be performed and we will then have an indication of the quality of mixing at small pressure.

L will now go through the different chapters trying to analyze the justifications which are presented to suppon the crucial arguments mentioned previously.

Chnpter 2: Problem definition end over all approach

• It is mentioned that it is only recently that pressures in the kbar range wcre observed experimentally in constrained one dimensional geometry. However, I think that a pressure peak of the order of the kbar amplitude nd millisecond duration was measured in the Sandia FITS RC2 experiment which was well vented (initially open at the top and later vented at the bottom as the vessel left the ground). But this was obtained using iron alumina thermite and subcooled water.

• Another important argument is that a because of extensive voiding, we need only be concemed about the first relocation event, and only for early trigger in it n. This seems to be justified by the premixing and explosion calculations presented later but I wonder why, after a first event, when water is sloshing back a second event cannot occur at about the same location where the structure will has been already dynamically loaded and eventually already deformed by the first event.

g Chapter 3: Structural failure criterin in this chapter, it is stated that a the time-duration of the loads of interest here is less than the stmetural frequency n, so it is expected that a peak strain would be basically independ of the details of the pressure pulse shape n. However, nothing is said about the estimated value of the natural frequency of the R.P.V. which seems to me to be of the order of magnitude of some msee so not so far from the load duration.

However, all the analysis is made with the analytical solution of Dufley and Mitcheli which assumes a shon pressure pulse n and allows to evaluate the plastic equivalent strain with incorporation of strain rate e!Tect by formula (3.2). But the comparison of the analytical results to ABAQUS calculations shows that the analytical relation gives conservative results for the plastic r. train evaluation (fig. 3.2).

Another mitigating factor is investigated: the effect ofload localization which shows that for a given impulse, the equivalent plastic strain is smaller when the loading is smaller. Use of these results is l

then made by assuming that a fraction p of the impulse is used for bending energy so that only an l

a effective impulse > is applied for the evaluation of the equivalent plastic strain. We are told that is a material and geometric a constant n but I have not found any indication ofits evaluation i.e. how the results shown on fig. 3.8 and fig. 3.9 are obtained. As fig. 3.9 is used to evaluate th loads calculated in Chapter 6, I think that it should be a little more explained.

E-10 1

It also seems to me that the localized loadings are applied en the axis cf the hemisphere (see table 3.2), does the fact that these localized loadings will occur on the side of the hemisphere with

• !ngularity where the sphere is linked to the cylinder will modify the conclusions we can draw from n "g.3.9.

Ud02 N.B. There are some errors in table 3.2 as for the valte of Ao which does not correspond to 4

as written in the caption Chapter 4: Quantification of melt relocation characteristics This is a very important task as most of the boundary conditions for premixing and explosion calculations are obtained from such an evaluation.

• The downward relocation path (arguments 2) is not envisaged: a we expect this path to be blocked by molten cladding and the blockage be robust n. This a expert judgment n is supported

- by the large heat sink associated with the large amount of a cold a materials in the lower part of the core. As it is said that due to be big stainless steel reflector, the first relocation will be deleyed compared to what occurred in TIM and that at this time, we will have a large oxide pool, it is important to know if this molten pool will reach the region of the lower fission gas polonium where the heat sink is not very large and where we can have a breakdown of the supporting material. However, in the paper, the blockage is said to occur in the region of the 7 cm a lower Zr m plug and lowermost spacer grid n. Some calculations are presented to show that the plugging time of this region by melt with negligible superheat is of the order of seconds. For this calculation I

(") have some trouble with formula 4.2 where, as for me, 2 is not the same as in the Carslaw and Jaegger text book but I did not try to perform the calculation. We can also make another remark:

if we have some breaking down of the fission gaz plenum region, when the molten pool arrives we may have superheated molten material from this pool that with flow in the lower blockage region for some times before plugging. It would be interesting to know what amount of molten materials can be transfered in the lower plenum through the holes in the core support plate before plugging of the passages in the blockage region. As for this plugging time which is crucial to support argument 2 - it would be interesting to see more realistic calculations including the influence of the interface thermal resistance between the cmst and the solid wall that will slow down the freezing process and then increase the plugging time.

• As for the blockage coolability: .

- the stable blockage thickness should be sensible to the radiation factor fr which is set to 0.7 without any explanations

- the cooling of this blockage is ensured for about 100 mn which is the time required to vaporize the water which fills all the volume between bottom of active core and bottom of core support plate. It is later estimated that meltthrough of the reflector by the molten oxidic pool will occur between 76 and 91 mm according to the amount of oxidation (80 to 95 mm in the calculation a without n preheating). If we add the time require to melt through the core barrel, we get 73

(") timing of the release of the same order of magnitude than the insurrance of blockage coolability.

As all these calculations are ord:r of magnitude ones, I think that argument 2 (no downward relocation path) may be questioned.

E-11

o Molten pool form: tion

- In the initial heat up calculation, are the reflector and core barrel in contact everywhere as it is shown fig. 4.8 and 4.97 In that case the cooling effect will be overestimated and the melt . perheat underestimated, g During the transient heat up calculation, what happens to the molten cladding and how the calculation with the effective therr ial conductivity is enventually modified?

e Molten pool calculation Such a calculation is performed for the oxidic and metallic pool, and there is a crucial hypothesis which is the presence of a stable oxidic crust at the upper surface of the oxidic pool.

In the document, it is mentionned that it is assumed that the clad drains but is it fully true?

Cannot we have some metals included in the moving down oxidic pool? What happen to the part of fission products which are released at fuel melting? Will they modify the molten pool behaviour for the stability of the upper crust and the evaluation of the differents fluxes?

e Melt through and melt release calculations It is said that rapidly the metallic pool will melt through the reflector but it is assumed that the metal a will be gradually draining n into the space between the flats of reflector and the core barrel. Cannot we have some kind of metallic jet impacting on the core barrel with some rapid meltthrough '.eading to a steam explosion between metals and water?

yr e f

mebike

/ /

f- &

M. . n . : oMe vw.4 f

oa W _r 1 yp.,

e From the above analysis it is concluded that when the oxidic melts through the reflector, there is no metal on it and that failure of the core barrel occurs soon after. First, it would be interesting to evaluate the time required for core barrel meltthrough (if there is an open space between the two of them).

But there is another problem if the space between the flats of the reflector and the core barrel is already filled with the metals from the metallic layer, how the oxidic pool can rapidly go through the core barrel. This situation may be a promoter for downward relocation if this added metal may increase the time for meltthrough.

• As for the location and size of the failure, most of the information is obtained from expert judgment and should be further justified:

- the failure a ir expected n to be local azimuthally and very near the top of the oxidic pool. I would agree with that statement as even, if the calculation is 2D cylindrical, once a flat will fail, E-12

the rapid relocation will impeed failure cn other flats. Cut I would not be able to give any probability for 2 quasi simultaneous failure, or 3....

p for the size of the breach, it is said that 0.4 m a would appear geometrically a good upper

() bound on the first breach width n and that a a 10 cm axlal gap is believed to be conservative n.

there is no mention of the rapid increase of the size of the breach during the melt release as it has been observed in experiments. However, as only short duration premixing triggering scenario are taken to be ofinterest, this enlargment would not be important. But, if we take into account steam explosion occuring when water is sloshing back after a first event, this has to be taken into account.

Chapter 5: Quantification of Premixtures Given the melt release conditions (flow rate, location, temperature and composition), the premixtures are calculated with an improved version of PM ALPHA which is now 3D and includes a melt fragmentation law (which was lacking up to now) as it is recognized that it is interesting a to know the distributions of the melt length scale n. However, this fragmentation law is not described and this should be done and justified as fragmentation is responsible for voiding (a the rate of voiding increasing rapidly with the rate of breakup n). I would also like to know why the a breaking law is operative only for as long as the coolant has a void fraction ofless than 50% n. If the fragmentation was slways operative, voiding would be larger so there must be a good reason for doing so but I do not see why.

_ The melt entrance conditions into water are also specified and not calculated:

C) entrance velocity whose evaluation is correct

- distribution of the melt a over an efTective radial width of 10 cm a with a melt volume fraction evaluated to get the correct mass flow rate. This distribution is crucial in determining the amount of vapour which is produced as the larger the entrance area, the longer you are in the film boiling regime in which the steam praduction is at maximum.. This behaviour was observed in MC3D recalculations of FARO test:, where a doubling of the pressare increase (linked with vapour production) was obtained with a doubling of the diameter of the melt flow. Recently CHYhES 2 recalculations showed the same trend.

- initial droplet diameter which is set to 20 mm (a large enough value to represent a minimally broken up melt stream). This parameter is also important for vapour production. It would be interesting to see sensitivity calculations with diameters varying from 10 to I cm.

I am not so sure that the melt will be transformed in a droplet population'before entering the water.

We may have a large melt stream on the wall with subsequent fragmentation into the water but with a different law than droplet fragmentation. Would it lead to a a benign evolution as it is mentionned. This is again an expert judgment.

As forjet fragmentation calculations with THIRMAL, I cannot trust them if the fragmentation is still governed by Kelvin Helmholtz type calculations. Moreover, in FARO experiments with 10 cm melt jet, it took more than 2 m of water to break the jets in a 50 bar atmosphere for which voiding is smaller, g

E 13

Ch:pter 6: Q:::tificatloa cf explosloa lo:ds Nothing is said about the parameters used in ESPROSE m 3D but as the trigger uses a 100 bar steam release, we may think that the hydrodynamics fragmentation law will be correct. Due to the &

small amount of melt involved in explosion calculations, there is no problem with the energetics of W the explosion and we are only interested in dynamical loadings of the RPV, This is done by the estimation of the impulse and of the local area ofloadings from ESPROSE-m results. I have some d0 problems to understand how D,y is estimated page 6.3 from the area evolution as shown 6.5.a.

In the text, it is said that peak impulses are around 0.1 and 0.2 MPa.s with effective area around 0.1 m2 (which gives g- 0.15) and from fig. 6.5 c where I find a 0.2 Mpa.s impulse, I do not 2

understand how I get Ao - 0.1 m from the area evolution which is shown.

Apart from this problem, if! accept the figures mentionned in the text, I agree when coming back to figure 3.9 that there is no risk for I ~ 0.2 and 3L- 0.15 as UP- - 0. This is confirmed by the D.S c ABAQUS calculations of the two most energetic explosion calculations obtained by PM ALPHA ,

plus ESPROSE m.

As a conclusion, I can say that if we accept the scenario which is retained by the authors, I think that

- whatever my remarks about premixing quantifications - the AP 600 RPV cannot be cha steam explosien. However, I would like to have more established confirmations of this scenario by mechanistic calculations when possible or parametric calculations when it is not.

The main thing to be confirmed is the impossibility to have, a downward relocation i.e

- the possibility to have a break down of the lower fission gas plenum rather than a continuous draining. This will give a sudden access of the core support plate holes to the melt

- the influence of the already relocated metallic pool on the oxidic release it may take a longer time to break through and the blockage integrity may then be challenged. The influence of an interfacial resistance between the oxidic solid crust and the wall specially at the top of the pool - will also participate to an increase of the time of break throu3hand of the evaluation of melt superheat.

Other branches for the scenario should also be evaluated:

- the possiblity of the metallic melt to rapidly go through the core barrel leading to metal water steam explosions

- the possibility to have steam explosion at later times (in the oxidic case) when water'is sloshing back after a first small scale (i.e low energetic) event.

As for reflooding scenario, the fact that water will be closed to saturation should also be evaluated.

O E 14

^ ~~ ~ ~ ^~ ~ ~

UNIViiHSITAT Oldi'1GAHl g INSTITUT FOR KERNENERCETIK UND ENERGIESYSTEME I Abteilung Reaktorsicherheit und Umwelt 1

O IKE Pfaffenwaldring 31 70550 Stuttgart Stuttgart Valhingen, den 02.10.1996 Pfaffenwaldring 31 Dr. L. W. Deitrich Director, Reactor Eng. Division Telex 7 255 44s univ d Argonne National Laboratory M y jjj' @ $ o4 9700 South Cass Avenue l l

Argonne, Illinois 60439 Unser Zeichen:IB i U.S.A. Ihre Nachricht vorn: l lhr Zeichen: 1 Review of the report ,, Lower Head Integrity under in. Vessel Steam Explosion loads "

Dear Dr. Deitrich:

by order of my colleague Manfred Burger I send you encrosed a copy of the above review. Mr.

Burger is on holidays until October 21. If you need further information during this time, you can contact me or my colleague Eberhard von Berg (same phone and fax number as Mr. Burger).

With best regards

~'~p *g - ' , .: . -

REisCTOR ENGINEER'l,7, D?ci ' 0.1 -

-D!RECTOTS Oren- ,

i 0CT' : 1995 Michael Buck

/ctc:n _ _ ./) L g,,,, ,

.~

l. 4 7: n id.:

L(C~h, Q%sm.G~~ ' ' te '

T

([

,-  := =. -

o Enclosures  ?.5.5 'I e iclead a dia.cRe w'e %c rtviev b WeseJd (A3 erd (ormaf, E-15 h.Ingl

Review of the report '

i I

,, Lower Head Integrity under In Vessel Steam Explosion Loads" l by T. G. Theofanous et al., DOE /ID-10541, June 1996 h Manfred Burger Institut fur Kemenergetik und Energiesysteme, Universitut Stuttgan l Pfaffenwaldring 31, D 70569 Stuttgan, Gennany Tel. 449 711685 2368 Fax +49 711-685 2010, E mail buerger @ike. uni stuttgart.de l

1. Purpose, Procedure and Main Conclusions of the Study Tha purpose of the work is to show that the lower head of a reactor like the AP600 withstands the load of steam explosions. According to the ROAAM philosophy, all physically meaningful causal paths that could lead to failure have to be investigated. The decomposition yields the following ,

central areas of analysis:

h, i

1. Since pressures in the kilobar range have been obtained in the KROTOS experiments

. (although not yet with corium and in one dimension), a direct exclusion of lower head failure cannot be done. Thus, detailed calculations of possible explosion loads are required, taking into account the specific geometry with respect to venting effects. This is done by use of a 3D version of ESPROSE.m.

2. The possible spectmm of melt / coolant mixtures developing in the lower head due to an assumed core melting must be determined. 'Ihis is done by use of a 3D version of PM-ALPHA.

3. Possible timings and strengths of triggers have to be considered. Due to the uncertainties, an envelopping approach is pursued here, conceming the timing as well as the strength.

4. Since close to the wall pressures in tne kilobar range are obtained in the calculations, specific investigations on failure criteria are required. This is done by a simple estimate and also by h E-16

., l i

means of the ABAQUS code. Consideradons on the possible interaction with thermal loads are also required. l O- 5. Considerations on the melting and relocation process in the core and the release to the lower plenum have been considered as necessary for restricting the possible spectmm of melt / coolant mixtus:s. This is done by separate estimates.

' The main arguments in the report are: .

i lt is assumed that a pool of ceramic melt surrounded by crusts forms in the core due to the  :

cooling capabilities of remaining water in the lower plenum (at the beginning of melt moti-ons with level at ~25% of active' core height) and the large heat capacity of the lower part of f fuel bundle with lower Zr plugs and the lowermost spacer grid. The key points are then that a downward relocation path of melt through the core support plate is excluded and melt-through of reflector and core barrel is assumed yielding finally a sidewards relocation-through the downcomer.  ;

I This sidewuds relocation is restricted in extent assuming' failure at the upper end of the pool based on the analysis of heat transfer from the pool and assuming plausible failure sizes.  ;

Further, it is argued that only one failure location is available within relevant times for pre-i rnixing of the relocated melt in the water and triggering. Strong voiding of the mixtures under the expected conditions of saturated water is expected and calculated by 3D PM-ALPHA. Thus, only small amounts of melt in the lower tens of kg are considered to be po-  :

tentially explosive. This is taken to directly exclude large break possibilities for the lower ,

' head. Various calculations with ESPROSE.m assuming sufficiently strong triggers are addi- t tionally taken to exclude also local threats to the RPV. This is finally done by comparison with failure criteria for the 'APV wall yielding directly (without application of the probabili- ,

stic framework) the conclusion that failure is physically unreasonable. 4 Further, reflood scenarios are evaluated to even mitigate the possibility of vapor explosion ,

threats, due to cooling and preventing melt outflow. Mixing with the melt in the pool is not considered as effective (small yield of stratified explosions). In addition, preventing outflow by reflood would also mean to exclude mixing of melt with highly subcooled water, which is -

E 17 1

.__m,,,ju_y.._,,,Ji _,_. ,,,.r, ,-a.~,, , ' . , - . - , _ , , , _ _ , _ _ , . _ - -. _ _ _ . _____-._____.-.,.._,....__._.,_m... ~ ~ . - -

considered as the only case with a potential to chall:nge the 1:wer head due ta increased penetration depths without excessive volds.

Cased with thermally weakened RPV walls are restricted to later phases of melt outflow.

Then the water and mixtures are already assumed as strongly volded. Reflood FCIs are con-sidered in the report in stratified configuration, i.e. water above a metallic layer. A threat is excluded due to rapid spontaneous interactions with the subcooled water and rapid freezing of the metal surface before a thick wate sayer establishes which could yield sufficient con.

straint for strong pressure buildup.

Based on these considerations, the major conclusion of the study is that steam explosion-Induced lower head failure in an AP600 like reactor is .. physically unreasonable".

2. General Comments .

The procedure as well as the general arguments are convincing. *Ihis concems especially:

• The argument that a strong cold trap at the core support plate, c>pecially if still connected with water, can prevent the downwards release path to occur before sidewards release at the upper region of the melt pool. This yields a significant reduction in melt flow rates to the water, especially to a possible downwards release in multiple streams. lf the cold trap at the botto strong enough, no downwards release will occur until all melt is released sidewards due to a continous failure progression.

• Then, the saturated coolant condition prevents larger premixtures without high voids, since larger premixtures could only develop within longer times. The geometrical conditions of the flow through the downcomer also favors this.

• The strong voiding of mixtures calculated with PM ALPHA is thus plausible.

• With the small mixtures (small melt masses) of not extensive void plausible from the above statements. the ESPROSE results are also plausible (the obtained pressures even appear astonishingly high - probably due to the restricted venting).

O E-18

• Thus, also the conclusions on the threats are plausible. .

• ne high number of calculations with PM ALPHA and ESPROSE covering a wide range of conditions can also be taken as supponing.

In spite of this agreement in principle, there remain problems in the details of the argumentation and performance of the analysis. Improvements may be performed to even better confirm the statements and conclusions as a basis for use in licensing actions. This will be discussed as follows in some detail. Since, in my opinion, the statements on the relocation path are a most critical point, I firstly concentrate on this, then considering the subsequent analyses on premixing and explosion. I will not consider the aspects of structural failure'eriteria which appear to be well established. I will only give few' arguments on possible funher scenario aspects concerning reflood.

3. Comments on Melting and Relocation Firstly, it is shown by estimating plugging times from a freezing model based on semi infinite heat conduction that the plugging takes place in a range of seconds (for the lowest initial rod temperatures 0.6 s with Zr and 2.4 s with UO,). However, the melt flow is not taken into account in thesa considerations. This means, that the freezing zone may extend over quite a distance. E.g.,

i

-0.6 m would result for 1 s with 0.6 m/s as a typical velocity from CORA experiments. Higher l velocities would result with thicker melt films. Thus, the final blockage formation should require some more time and distance (need of additional melting and melt flow or compaction to a crust by remelting and relocation of upper parts of a partly blocked region). Funber, this process may yield local incoherencies of the crust formation, i.e. also weaker regions, although the cold traps at the bottom give certainly a unifying trend. Thus, in order to further verify the statement of rapid blockage formation calculations with a core melt code would be desirable. These could also yield i a more detailed perspective on related (subsequent) important questions, especially the heatup of the cold trap regions and the water level development.

Secondly, assuming an existing blockage with an overlying melt pool, it is checked whether a steady state with a stable crust below melting point (~2800 K for oxidic or 2100 K for metallic E-19

material respectively) can exist. Metallic and ceramic crusts are considered altematively, with a heat aux from the molten ceramic pool above of s 0.02 hnV/m', a volumetric heaUng of - 0.5 MW/m' in the ceramic crust and cooling from below via radiation. Here, it appears not clear to me why the fraction of fuel volume is only taken as -30% (p. 4 6). His seems to be a value for g

intact stmetures. However, if the metallic pans ue all relocated during establishment of the ceramic crust, then this may consist essentially of UO/2.rO, (-80/20 wt ratio). The local shape factor should also not be decisive due to the crust formation from upper material. Taking a value of decay heat of 300 W per kg fuel (p. 418) this would yield ~2 hBV per m' of the UO/2.rO, Crust.

For the downward heat flux from the molten corium pool above, a maximum value of 0.02 knV/m' (fully developed) is assumed, his is derived from Eq. (4.14) based on the Steinbemer.

Reinecke conclations for a rectangular geometry (typing error in (4.14): exponent 0.095 instead of 0049). However, this correlation is only confirmed for Ra' < 510*. For the conditions considered here, I obtain a value of Ra' ~ 10", assuming H = 1.8 m and Q = 2 hBV/m'.

The correlation is also derived for non isothermr) lateral boundary conditions, in contrast to the present assumptions. The influence of the lateral boundary conditions appears to be small, however. For a case with venical cylinder and melting point temperature at all boundaries THEKAR calculations (1) also yield a rather similar correlation (Nu, = 0.935 Ra' **), but only for Ra' numbers below 10'.

But, the main question is to me whether -in view of the above arguments and at least some lateral cooling potential - the assumption of a rectangular pool geometry is a too strong idealization and other geometries closer to hemispherical shapes can really be excluded. Such geometrical variations would yield significant variations in the heat transfer to the lower boundary. The influence of the lateral boundary would increase (natural convection influence versus stable stratification). For a hemisphere (certainly an extreme under 9e given flat radial power shape) even a mean heat aux of 1.05 hnV/m' would result at the curved lower boundary according to (5.28) from the IVR report and at the center still 0.1 hBV/m' according to (5.30a) from IVR. With a thermal load from the melt pool of 0.1 hnV/m' and Q = 2hnV/m' in the crust only 3 cm of stable ceramic crust would result trom equations (4.3) and (4.4), with 0.02 hnV/m' about 5 cm.

E-20

Punhor, the first blockage should be metallic and a ceramic cmst should settle above. Den, the-combined system of ceramic and metallic cmsts should be considered. his yields a lower bottom temperature, thus lower radiative heat removal. herefore, the crusts should become even thinner.

If, due to heatup of the lower structures, the lower region of the metallic crust remelts and relocates, this yields a further decrease of downwards heat removal from the ceramic crust region, j thus inducing funher remelting. j I

Finally, the downwards relocation path appears not yet as surely excluded as stated. It is also to be mentioned here that local melt streams into the melt pool could strongly enhance the local heat l

- transfer to the bottom crust as shown in (1). Dus, together widi the uncertainties of the process of crust formation considered above and the smaller crust thicknesses of the above estimates, local i inhomogeneities of the crust may become imponant and may induce local failure at the bottom.

However, the basic idea that significant cooling potential is provided from the remaltdng water in  ;

the lower head and the massive core support plate is promising. Perhaps, some funher calculations j related to the above objections could yield further suppon, But, the steady state consideration for the crust may not be sufficient in general. Calculations on the time development of melting and crust development with available codes may be necessary for better confumation.

t The maln _ statement is that sidewards melt through - occurs significantly before possible downwards relocation, within the time of - 100 minutes during which effective cooling from remaining water above the core suppon plate is available. The basic statement is that sidewards ,

cooling is much less effective than downwards cooling. The evaluations in the repon take the sidewards boundary condition as adiabatic, i.e. no lateral heat removal is assumed. His appears ;o  :

i - be a too strong restriction. On one hand, heat removal by the produced steam should be taken it.to f

account. On the other hand, heatup of the RPV wall by radiation from the barrel and outside vessel cooling by flooding should be considered. Taking a temperature difference of ~ 500 K over ,

the barTel and reflector and an outer barrel temperature of - 1000 K as given from the calculations

, (Figs. 4.8 7 4.12) nearly half of the heat flux through barrel and reflector could be radiated to the ,

b RPV wall (if taken at saturation),

The calculations on core heatup and melting essentially yield the timing for melt pool formation -

(- 42 -57 trdnutes from core uncovery to 20%) to be related to the times for evaporation of. water

-'- E-21 i  !

j

above the lower core support plate and for heatup of reflector and barrel. The further heatup of the ceramic pool and the overlying metallic melt layer resulting from renector melting as well as the heatup and melting of reflector and barrel is calculated by means of equations (4.10) - (4.15), of which (4.14) has been questioned above (questioning the assumption of rectangular pool shape).

With the lateral heat flux in the ceramic pool an additional time of 34-38 minutes is calculated for reflector melting. This is taken to verify lateral melt release at a time with still effective cooling from below (water above lower core plate). But, it has to be remarked again that lateral heat removal is neglected. At melting temperature, at least half of the lateral heat flux could be radiated to the RPV wall (if this is not taken to be superheated sufficiently). Funher, heatup and melting of the sidewards ceramic crust as well as of the barrel is not considered.

The considerations on the overlying metal layer resulting from the melting reflector seem not to yield important effects with respect to the final melt release. Although earlier melt-through o reflector can be expected in this range, this only means that the reflector melt is essentially relocated into the gaps between the reflector and the barrel. But then, the refrozen material must melt again to get break-through. Certainly, freezing heats up the still solid reflector and barrel material. But, the material and energy redistribution by these processes may yield some azimuthal homogenization. Thus, the assumption of a local azimuthal failure may not be justified by the g considerations on geometrical inhomogenities in the report. In general, the assumptions on failure locations and size are problematic, although the bounding assumptions apear to be reasonable. In my view, the main objections could be, on one hand, those of above, questioning 'he exclusion downwards failure, and, on the other hand, the exclusion of several failure locations within a certainly short time frame.

The latter point indicates a further deficit: the further course of melt release is not considered sufficiently. Even with an outflow rate of 400 kg/s (see below) the time of outflow of the whole corium melt pool would last some minutes. During this time, failure could occur at multiple locations, overlapping in time. Further, local melt / coolant interactions could yield additional and also larger failures. A question is whether failures of the bc,ttom of the cmst by such interactions can be excluded. On the other hand, the strong voiding with the resulting necessity of early triggering to get explosions restricts :he possibilities for coinciding events. Enhanced evaporation of the water pool also acts in this direction.

E-22

Scenados of ex vessel reflood are considered with respect to the time of vessel flooding but not conceming the establishment of effective lateral cooling as discussed above. De considerations on vessel flooding before or just about the time of reflector melt through consider only the cooling aspects and thermal effects of focusing by thin metal layers. But, embrittlement due to rapid quenching may favor failure of the pool surroundings at any location. Funher, especially ,

entrapment explosions in the gaps may yield such failures and thus more extensive melt release.

4. Comments on Breakup and Prembdng From the assumed failure location and size, melt flow rates of 200 to 400 kg/s are estimated, yielding -5m/s entrance velocity into the sr.turated water pool at a level in the range of the core suppon plate. Hen, the next main peint is in my view the fragmentation process. It is stated that adequately bounding the effect of various degrees of' breakup leads to extensive voiding developing rapidly in all cases. This voiding of premixtures is calculated with a 3D version of PM AI.PHA. The melt stream is assumed to be broken up initially into drops of diameter 20 mm

("large enough value to represent a minimally broken up meit stream"). However, as compared to a coherent stream of -Ilcm diameter (with 400 kg/s and Sm/s), this yields a surface of factor 8.4 higher and correspondingly a higher heat transfer and steam production Transient breakup could thus yield significantly less steam production. On the other hand, the breakup may then not be sufficient for explosive premixtures. A factor of 6 still results for two jets of melt with l

l correspondingly smaller diameters which fcciliates breakup again to some extent. Thus, mixtures

'with smaller void may result from transient breakup and assuming several jets with smaller l diameters. On the other hand, there remains certainly a limitation to tareakup due to the time consuming process of breakup, In my opinion, these contrary effects with respect to getting an effective mixture, i.e. too less breakup or too strong volding with stronger breakup, should be explored more for getting the inherent limitations to explosive mixtures. Although the statement of strong steaming appears to be plausible, it may not be possible to demonstrate it for all possible cases, as indicated above. A U smaller window for explosive mixtures may becue plausible taking into account the above E-23

effects of time requirement for breakup and too coarse breakup combined with weaker voiding.

Perhaps, some additional variations with PM ALPHA could be done to show tiils, e.g. by considering plausible time laws for breakup of coherent jets together with varying breakup length scales.

Indeed, the THIRMAL calculations give some perspective on this, showing the extreme cases of little stripping of small fragments for a thick jet (7.3 cm diameter) and coarse breakup for the smaller jets (2.9 cm and 1.8 cm) due to long wave instabilities. However, certainly cases in between these extremes of breakup beha flour should be considered. Further, the present state of jet breakup modeling canuot be taken as verified. 'Ihis is also indicated by the sigrJficaat differences between results based on Kelvin Helmholtz instabilities and on the theory taking into account velocity profiles (Miles) which have been obtained with IKEJET,4.g. [2].

Since multiple Jets may occur from one hole by some separation effects (e.g. connected with the failure mode) or from several failure locations, the restrictign of mass assumed in Appendix D, p.

D-6 for the case with thinnerjets appears not to be justified. Also, the concluding statement of p.

512, that both length scale and void fractions are well, encompassed by the PM ALPHA calculations" appears to me as too rough, tri view of the variations of cases indicated with the THIRMAL calculatiens and considered above. On the other hand, I agree in principle to the h

expectation of strong voiding based on the situation considered, with melt into saturated water and with the necessity of breakup for explosive mixtures. Further variations may even better demonstrate this, as indicated above. I think, this could also be done for the situation of bottom failures of the melt pool, excluded in the report. The exclusion may also be better based by

.considering additionally the cold trap properties of the lower spacer grid and the Zr plugs quantitatively for conditions after boil-off in this region and with melt relocation to thh egion before lateral melting of reflector and barrel occurs (or: improved considerations on t!. timing of the events, with respect to the above discussions).

9 E 24

5, Comments on Explosion, together with Premixing .

g b A trigger of sufficient strength is applied to the mixtures in ESPROSE calculations to quantify explosions. The chosen trigger appears to be sufficiently strong to produce strong escalations as in the KROTOS experiments, but its strength is not assessed with respect to possible trigger trengths. I agree that with a suf0clently strorig tdgger the escalation dynamics may no longer depend on the trigger strength (if overdriven cases are excluded as unrealistic). Thus, the results of the numerical tests may indicate such a limiting strength and no need for further variation. In view of the effects in the KROTOS experiments, the chosen trigger can also be considered as strong encugn to yield mitjor effects. Looking at early rather than later times for bounding the effect of trigger timing appears also appropriate in view of the strong voiding (excluding other possibilities of melt reicase as discussed in the previous chapters).

'N sets given in the report show s.lg;uficant differences in the maximum local pressures, the mtximum impulses as well as loaded areas and e.t;nes of loading depending on the chosen melt mass flow, the breakup behaviour and the time of triggering. E.g., with the higher melt flow a rnaximum pressure > 5000 bar, impulse 100 120 kPas find maximum area of 5 m' results (v) choosing p = 10 for fragmentation and an instant of triggering at 0.05 s (case C210(0.05)), as com; ared '.o nearly 'O' bar, 150 200 kPa s and 3.5 m' for = 20 and 0.12 s (case C2 20(0.12)).

On the other ' .md, there seem also to be similarities or bounding trends. E.g., for case Cl-10(0.05) with the lower melt flow, p = 10 and 0.05 s, the results are rather similar to C2 10 (0.05),

with somewhat s.naller impulses and areas in Cl 10(0.05). But small shifts in trigger time give also strons differences, e.g. pages C.3 16 to C.3-20 in the report for case C2-20. The same is valid for the comparison of C210 and C2 20 with similar trigger times (pages C 313 and C.3-16).

This is certainly due to the relation between time development of breakup and voiding, producing optimum mixture configurations et different times. This is one cause of uncertainties in getting explosive events or not (together with triggering time).

Conceming this problem of sensitivity, the large number of calculations performed is convincing.

They yield maximum events but in a limited range and not as singular cases. Some questionirig I still have with this respect concerns the choice of for premixing breakup and perhap! the

{J E 25

. N underlying time law of breakup (not given in the repon). Since the maximum loads appear to be obtained with case C2 20(0.12) as compared to the lower p, it is not quite convincing to jump to nb and not to coissider cases between. Other time dependences may yleid funher variations. This concems the questioning of above conceming the premixing process a: well as the melt flows.

Conceming the latter point, it is to be ren srxed that the main effect of multiple melt streams into the water - if taken as saturated would be that a larger region is loaded by the explosion (perhaps also some funher escalations in more extended premixtures may be possibL, this could be checked by ESPROSE calculations) and that thus the venting will be f.sther limited. Also the pressure relief in the vessel wall will then be limited. Thus, it is important to funher confirm the exclusion of such multiple events (small windows for this!) or to check the coincidence effects.

It remains to formulate some general questioning conceming the verification state of PM ALPHA and ESPROSE. Although a let of work has been performed on this, I think that severe ques ons remain. Even if numerical aspects may be considered as well established, also with repect to 3D, there remrdn open areas conceming the physical formulations. These are e.g.:

• Checks with MAGICO were - to my knowledge - restrict'ed to relatively small volume parts of spheres.

• In general, correlations for exchange processes in three phases are uncertain and need further clariilcation.

• The uncenalntles onjet breakup have already been mentioned above.

• The microinteractions formulation for hydrodynamic fragmentation in thermal detonation waves needs funher clarification and development, for single drop a well as finally for drop assemblies. This concems the conclusions based on theory and on single effect experiments as well as on KROTOS experiments.

With respect to FARO and KROTOS analyses, different premixing and explosion codes have shown the capabilities to calculate the experimental behaviour. But the underlying physical formulations and thus the physical interpretations differ strongly. Further, even no convincing comprehensive understanding has up to now been gained on the differences especially in premixing behavior between UO/ZrO, and Al,0 in KROTOS. Thus, funher work is nec 3

E-26

b u

i get approved understanding, models and codes. In general, the results would be more convincing l.

If supported also by other codes based on a common physical undentanding and conosponding l O formuistions.

6. Comments on Reflood Scenarios  :

a As already remarked unde: 3. of my comments, the ex' vessel reflood should also be taken into account with respect to the considerations on the cooling aspects determining the conclusions on l

.t i

the relocation path. In the context of vessel reflood also the possibility of embrittlement and thermal strosres favouring failure should be considered. It should be shown that also under these I

conditions iarger melt release and in this case contact with subcooled water is avoided or not  ;

threatening. Entrapment explosions, e.g. In the gaps between reflector and barrel, should also be f addressed concerning a possible increase in failure and melt release. , f Concerning the interaction of reflood water with melt pools, I agree to the argument of rapid small scale interactions, rapid solidification and in general small effectivity of stratified l explosions. However, it should be addressed whether relevant effects of mixing due to the water  !

impact and due to smrJi-scale interactions (also taking into account the falling-back of expelled j water) can be excluded. 'Ihe situation of reflood under conditions of still existing melt / water ,

mixtures in the lower head may be even more important than the extreme of the melt pools in the .

lower head, if the reflood vzater could enhance mixing again, collapse steam, favor funher melt l release by the above pr~~m. and this under the conditions of already settled melt, i.e. therral

' load at the bottom. 1

7. Some Conunents for Formal Improvements Some typing errors of relevance are given below (I had no time for funher detailed checking), 'l together with some need for detailed descriptions:

• Ra' in Nomenclature: factor g is missing, g also missing in Nomenclature.-

' ab '

t

.# -. >,..-.a--..w-.- -

,--w. ... ,c,erw,-.-.. ., , , ..,.-.....,_,m.#,,,,wm.w._,m,..,.,,.,,_,,..m,y, ,,v . m -, m- w --,--mm-w.%-,.,,---.m--y,

o P. 4-6, second line from below: 0.2 MW/m' instead of m'. l

• P. 418: effective power density .. of 0.26 W/g: does this mean of core material in contrast to

' I fuel?

• P. 4-21, eq (4.14): Ra'"".

• P. 5 2: giving the breakup law would be helpful.

• P. 5 4: some further informations on this and on the other color figures would be helpful, e.g.

length scales, quantitles of melt volume fraction.

'

• P. 5 5: z directed to top, but in subsequent results inversely. ,

• P. 5 10,13th line from above: ,, melt" instead ,, coolant".

• P. 5 10, second line bom below: ,,two slower pours" seem to be better characterized by pours with smaller diameter.

• P. 6-1, second line trom below: ,, propagation intensity is basically independent of the magnitude of the trigger." - It is not quite clear to me'what is meant, e.g. with ,, propagation intensity".

• P. 6-1, end of second paragraph: (Theofanous et al,1996a)?.

• P. 6 4: 6th line from below: 0.1 m'

• Fig. 6.2: should be better characterized: length scales, pressure scales

• Fig. 6.4: locations are identiDed in Fig. 6.27

• Fig. 6.5: location of peak loading: where?, not included in 6.4!

S P. B.1-; etc.: color characterization is not quite clear to me: fuel vold, red lines, blue isolines in jet.

• Appendix C: color pictures: case? charactenzations?

• choice of parameters in the ESPROSE calculations, especially conceming micro-interactions?

• P. C.313 etc.: locations?

• P. D-4,6th line from above: R,,, instead R',,, s R, O

E-28

References .

t O (1) Mayinger, F. et al.: ,,Untersuchung tiwioodynarnischer Vorgunge sowie Waren**=a-h in der Komschmelse. Teil 1: Zusammenfassende Darstellung der Ergebnisse." Institut fur Verfahrep +t% Technische Universitut Hannover, AbschluBbericht BMPT - RS 48/1, July 1975

[2] Burger, M., Cho, S.H., v. Berg, E., Schatz, A.: ,, Breakup of Melt Jets as Precondition for Prem1xing: Modelling and Experimental Verification." Nucl. Eng. and Design 155 (1995) 215-251 0

t

.O Stuttgart, September 30,1996

.,1{ . Cp

' Manfred Burger-E-29

Los Alamos no,M "*"

PS .Eetl!d'" O National Laboratory Los Alamos, New Mexico 87545 Engineering Sclerces & Applications ESA.EA. Engineering Anaysis, MS P946 Dr. L W. Detrich Argonne National Laborstory 9700 S. Cass Avenue , Building 208 Argonne,IL 60439

SUBJECT:

REVIEW COMMENTS Enclosed are my review coremeats on the report entitled 1.ower Head Integrity Under in Vessel Stream Explosion Loads," by T. G. Theofanous, et al. These comments represent the final deliverable under Memorandum Purchase Order No. 063094. If Ican be of further help please contact me by e-mail at butlerOlani. gov or by 16fephone at (505)667 6578.

Sincerely.

~

I R E C E l V E ii~ ':

Thomas A. Butler, PE l REACTOR ENG!NEER!NG D!VP':0:1 -

-D' RECTOR'S OFF:CE- ,

TAB:rbh .

! 1  ;

DEC 13 996

Enclosures:

A/S , ge7;cn: , , ,, ,_ ,

Cy: T. G. Theof anous (w enclosures) pgynuEi t. e . # A~ I~wW,[

' ' " M N5'- -

W. M. Walsh (w/o enclosures) ' --

CIC 10, MS A150 (w/o enclosures) ,, , ,

ESA EA File (w enclosures)

O E-30

4 r Review Comments on Report DOE /ID 10541, " Lower Head Integrity Under k In-Vessel Steam Explosion Loads" i Tom Butler Group ESA EA Los Alamos National Laboratory i December 2,1996 INTRODUCTION AND

SUMMARY

The comments contained in this review are restricted only to a review of Section 3 of the subject report. The report's authors have done a goodjob of scoping the possibilities of falling the lower vessel head under the assumed loading conditions. Well '

established analytical approximations were used to establish the validity of the finite element model that was developed to study local failure of the head. A more detailed model needs to be developed to include transverse shear effects and to simulate failure of damaged q elements during the course of the calculation. This lack of simulating progressive failure is U the weakest point of the analysis. Appropriate simulation of progressive failure has to be included in order to obtain defensible results that can be included in probabilistic evaluations.

SPECIFIC COMMENTS Finite Element Model:

Use of the shell elements in ABAQUS is acceptable for determining the distribution through the thickness of all components except for transverse shear. In the ABAQUS thick shell elements transverse shear is approximated by constant shear through the section. This is not judged to be adequate for evaluating the possibility of a shear type of failure. A better method for getting good approximations for all of the strain components would be to use several continuum elements through the thickness rather than the thick shell element.

Use of many more elements would make the runs longer, but use of the explicit version of ABAQUS would help in this regard (see below). In addition, the use of an axisymmetric O finite element model would afford the opportunity to use a much more dense mesh in the O

E-31

analysis with run times that are still relatively shon The structure and all of the loads that were considered are axisymmetric.

The mesh should be considerably more dense in order to resolve fine details in the strain distribution, especially those details relating to strains other than in plane strains.

Refening to Figures 3.5a-c, even the in plane strains vary from their maximum levels to just hdf that level overjust one or two elements.

Although not stated in tha repon, I assume that the implicit version of the AB AQUS wde was used for these calculations. The implicit version is always stable but may not always be convergent. There is no indication in t? mpon as to whether the time step was varied to ensure a convergent solution. A better altemative may be to use the explicit version of AB AQUS for the shon transient solutions that are requiud for the types ofloads being considered here. The explicit versio< af ABAQUS also offen the opponunity to use a failure model that would give more realistic failure predictions (see below).

The statement that the time duration of the loads is less than the natural frequency of the head may not be correct. A handbook solution of the frequency of a full sphere with the same dimensions as the hemispherical head gives a natural period of 1.5 ms, very near to the 2 ms pulse duration used in this study. It is no wonder that, as stated, the impulse time is "non negligible."

Load Strain Behavior:

O Use of the Bodner and Symonds approximation for the dynamic yield stress is a reasonable approximation. However, use of the values assumed for the constants D and p should be justified more thoroughly. The values used here are for mild steels, and may not be appropriate for the alloy steel that is used in the pressure vessel being evaluated.

Obtaining a good approximation for this relation is particularly imponant because the maximum strain is very dependent upon it. I used an axisymmetric model with 15 continuum elements through the thickness to replicate some of the calculations in the repon.

The results shawed that the maximum strain went from 0.52 to 0.16 with addition of the rate model for the yield stress. Considering the magnitude of this difference, one should cenainly be very carefulin the selection 'f the rate parameters.

Dexter and Chan (1990) address

• e effects of strain rate and temperature or. A533B steels. This alloy is close to A508 steel md may provide some usefulinformation in developing an appropriate dynamic yield stress model.

O E-32

. l l

4 Failure Criteria and Fragili*y: l l l This is probably the most difficult aspect of modeling the response of the vessel j head. 'the failure criterion that is used in the repart is probably realistic and conservative -

i l

except for one imponant aspect. 'the model, as reponed here does not remove the load l

carrying capability of elements that have exceeded the failure criterion. Maintaining the  !

load canying capacity of damaged elements can give signirwant over estimates of the capacity of the structure. I used the explicit finite element model mentioned above to look

at this aspect of the problem and found that, depending on the parameters used for the AB AQUS failure model, the head could fail for the loads that are reponed. I strongly suggest using some sort of failure criterica embedded in the computational model for future [

calculations.  !

i 'Ihe subject report briefly mentions the effect of stress anisotropy ont the failure  !

strain. This is an important issue and needs to be more fully evaluated. The work  ;

referenced in the repon by Pao and Gilat was performed on Charpy bars (roughly unlaxial -

. strain) and by Shockey et kl. was performed in pure shear (no hydrostatic component).

Therefere these data don't address the important effects coming from multi-dimensional ,

stress fields. Data summarized by Ju and Butler (1984) show that A533 alloy steel when in  ;

equal blaxial tension fails at an equivalent strain equal to about one third the strain for

!- uniaxial tension. Equal biaxial tension is the stress state at the " pole" of the lower head  ;

^ '

where failure would first be expected. The alloy content of A533 steel is similar to that of the A508 steel considered in the subject repon. Mirza,' Barton, and Church (1996) reported

> ~t he effect of the stress .ield on failure strain and its effcets in transitioning fmm ductile to brittle failure characteristics. Johnson and Cook (1985) also discuss the effects of the stress field on fracture of ductile metals. Other references that raay be of help include Jones .

and Shen (1993) and Ferron and Zeghloul (1993). 1

' As previously mentioned, the head would have to be modeled with continuum  ;

- elements to accurately predict transverse shear strains. In addition, a failure criterion for

. transvuse shear needs to be established, it is unlikely that the failure criteria discussed in the above references are adequate. They may however give some guidance in establishing

. - the appropriate criteria. It is possible that when the loading conditions are investigatti more closely, the load cases that 1:ad to the highest shear load (such as case 1+) can be eliminated obviating the need for this criterion.

e 3

O

E-33

. . . = . .. =. - - .. - - . . . . . - . . _ . . . - - - - .

Miscellaneous:

1) The use of the higher yield stress 450 MPa is justified based on actual data from Server O

and Oldfield (1978) where the average yield stress is approximately 440 MPa for A508 steel (very close to the Japan Steel Works Ltd, value of 450 MPa). This it one parameter with a:vple data to support the use of the actual, as tested value.

2) The statement is made that A533B steel has a carbon content of 0.19 vs 0.16% for A508. Information from Server and Oldfield (1978) and the ASME Code show that A533 has a carbon content of 0.25% maximum and A508, Class 3 has the same upper limit for carbon content. Actual analyses show carbon content from 0.21 to 0.25% for both steels.

3) Chapter 3 in the subject aport does not mention radiation embrittlement effects. If they can be dismissed, the reasons sh3uld be given.

41 For SA508 the. transition from ductile to brittle behavior starts at about room temperature. The report should give the approrimate material temperatures during the postulated event to show that it is well above room temperature.

5) The presence of flaws is not addressed. I assume that in service inspection will have identified any that are significant in affecting ductile fracture.

References O

Dexter, R. J. and K. S. Chan, "Viscoplastic Characterization of A533B Steel at High Strain Rates," Journal of Pressure Vessel Technology, Vol 112,(1990) 218-224.

Ferron, G. and A Zeghloul, " Stain Localization and Fracture in Metal Sheets and Thin-Walled Structures," in Structural Crashworthiness and Failure, N Jones and T. Wierzbicki, Eds. (Elsevier Applied Science, London and New York,1993), Chap. 4, pp.131-163.

Johnson, G. R. and W. H. Cool:, " Fracture Characteristics of Three Metals Subjected to Various Strains, Strain rates. Temperatures, and Pressures," Journal of Engineering Fracture Mechanics, Vol. 21 No.1, (1985) 31-48.

Jones, N. and W. Q. Shen, " Criteria for the Inelastic Rupture of Ductile Metal Beams Stbjected to Large Dynamic Loads," in Structural Crashworthiness and Failure, N Jones and T. Wierzbicki, Eds. (Elsevier Applied Science, London and New York,1993), Chap.

3, pp.95-130.

E-34 )

l

Ju, F. D. and T. A. Butler, " Review of Proposed Ductile Failure Crites'.a for Ductile

- Materials," Los Alamos National Laboratory repon LA-10007-MS (NUREG/CR-3544),

April 1984.

Mirza, M. S. and D. C. Barton, "The Effect of Stress Triaxialhy and Strain-Rate on the Fracture Characteristics of Ductile Metals," Journal of Materials Science, Vol. 31 (1996) 453-461.

Server, W. L. and W. Oldfield, " Nuclear Pressure Vessel Steel Data Base," Electric Power Research Institute report EPRI-NP-933, December 1978.

O -

\

4 O

E-35

i n:L

%cu l Los Alamos o: aanua e,i. 7 Refer to: ESA E 97 000

,g i National Laboratory Los Alamos, New Mexico 87545 ILLj Engineering Sciences & Applicatlocs ESA.EA, Engineering Analysis, MS P946 1

l Dr. L W. Deltrich Argonne National Laboratory '

9700 S. Cass Ave., Bldg. 208 Argonne,IL 60439

Dear Dr.Deitrich:

The purpose of this letter is to clarify the applicability of comments that I made in the attachment to my previous letter to you dated December 1996, regarding review of the report entitled " Lower Head integrity Under in-Vessel Stream Explosion Loads," by T. G. Theofanous, et al.

The comments made on that attachment were linued to Chapter 3 of the subject report and, consequently, affect the fragility curves developed in that chapter. The fragility curves are subsequently referred to in reaching conclusions in Chapters 6 and 7 of the report it b important to make clear, however, that the fragility curves in question do not have major efft.ct on the conclusions reached in those chapters. The loads developed in Chapter 6 and applied in Chapter 7 are low enough that the vessel resoonse is definitely below the lowest probabili%; level used in defining the O

4 curves (10 ).

I should also point out that I concur that the probability levels used in developing the fragility curves are conservative. The association of these levels with strain magnitudes through the vessel wa3 are acceptable. However, the calculated strain levels used to develop the detailed curves above the 10'8 level may not be conservative. If the curves are ever used for evaluating higher loads; they should be reevaluated based on the review comments that I previously submitted. Implementation of the information in these comments will affect the shape of the curves and could shift them toward lower levels of impulse load (to the left in figure 3.11 of the subject report).

If I can be of further help please do not hesitate to contact me.

Sincerely, a

Thomas A. Butler, PE TAB;rbh f E 1 V ~:'. l

}. REACTOR ENGINEER!N: * ~ %

Cy: T. G. Theofanous - Digge70e 3 p? -

W. M. Walsh CIC 10. MS A150 ESA.EA File

' AcTicm ,

JAN 131997 g E-36  ! -..~.....' '

g,vo w gm -

.T. '. ~.'.

ARGONNE NATIONAL LABORATORY 9700 South Cass Avenue, Argonne, Illinois 60439 Phone: (630) 252-4595 Fax: (630) 252-4780 September 25,1996 -- . . - - - - . . . . . _ . . .

RECEIVED  !

REACTOR E.ElHEI.T!W C.im:CN . ,

-D!."fCTOR'5 O.WC1-Dr. L. W. Deitrich Reactor Engmeering Division $EP 261993 .

Argonne National Laboratory Acm:n _ _ , gg ; _t a ,

9700 South Cass Avenue Argonne, Illinois 60439 L to 9, T%.,_[ '

p ita w n s

~

Dear Walt:

In response to the request made in your letter of June 17, 1996, I have reviewed the report "IAwcr Head Integrity Under In-vessel Steam Explosion Imads" by T. G. 'lheofanous et al. You indicated that this report and a companion document together " intend to demonstrate the effectiveness of 'in-vessel retention' as a severe accident management concept for a reactor like the AP600". You further indicated that "the purpose of this review is to assess whether this intent has been achieved to a sufficient degree for the results to be of use in the regulatory / licensing area". Based on my review of the report, I find Qat additional supporting work would be needed if the conclusions of the report were to be used in the regulatory / licensing area.

On page 9-1, the authors state that " Methodologically, the assessment involved only a slight scenario dependence, principally on tre permanence of the blockages preventing direct downward, through the lower core support plate, relocation", and that "thus the assessment is of Grade B, in the ROAAM scale". I think the v:enario dependence is more than slight, so the assessment may be more of Grade C than Grade B in the ROAAM scale. Suppose a steam explosion would -

take place in the downcomer region or in the lower plenum, as described in the report. The explosion may not be strong enough to fail the lower head, but it may be energetic enough to mechanically disrupt the blockages formed at the lower end of the core. Further, the explosion would likely expel some water from the lower plenum so that the lower core support plate may no longer be in contact with water (i.e., the ability to cool the core support plate would be lost).

Thus, the initial explosion, while not failing the lower head, could severely weaken the blockages mechanically as well as thermally. It would seem possible that a relatively~small initial explosion would be followed by a massive downward relocation of core melt through the core support plate, setting the stage for a secondary explosion probably involving a much larger melt mass. The lower head may well survive such a secondary explosion, but a separate assessment of this possibility would definitely be needed.

Based on the code calculations performed, the report concludes that the saturated coolant condition in the lower plenum leads to highly voided premixtures that have a dampening effect on the resulting explosion energetics. While I am not judging the validity of the calculations, I fm' d it p

V difficult to reconcile this conclusion with available experimental evidence. Experience tells us that triggering of a steam explosion would be more difficult with saturated water than with highly subcooled water. However, once triggered, the explosion energetics does not seem to depend on E-37 Operated by De University of Chicago for ne United States Depanment of Energy

l.

L. W. Deitrich 2 September 25,1996 the coolant temperature that much. Consider, for example, the results of the KROTOS tests Nos.

28,29, and 30 [H. Hohmann et al., "FCI Experiments in the Aluminum Oxide / Water System,"

Nucl. Eng. Design 155 (1995) 391-403). In these tests, approximately 1.5 kg quantities of A12O3 g

melt at 2300-2400*C were poured into a column of water and steam explosions took place. In KROTOS 28, the water was nearly saturated at 87'C while in KROTOS 29 and 30, the water was highly subcooled at 20'C. The energy conversion ratio was estimated to be 1.3%,0.8%, and 1.25%, respectively, for KROTOS 28,29, and 30. It thus appears that the explosion with the nearly saturated water was at least as energetic as those with the highly subcooled water. Similar findings regarding the effect of water temperature on the explosien energetics were also made in our recent ZREX experiments. Such experimental evidence would need to be considered when discussing the explosion energetics. Perhaps additional parametric calculations in terms of the breakup and trigger timings might be useful.

In all supporting calculations, the water was considered to he saturated with the primary system completely depressurized to 1 bar. Even in a large-break LOCA, the containment back pressure would remain in the range of 2 4 bars for a long period of time. It would appear that a system pressure somewhat higher than 1 bar (e.g., 3 bars) would have been more realistic for the supporting calculations.

Reflood FCIs were discussed in Chapter 8. I suspect that reflood FCIs in stratified configurations would be of secondary importance compared to the premixed explosions addressed in the rest of the report. Nevertheless, reflood FCIs need to be considered for completeness, particularly in view of the potential for vessel wall thinning due to chemical attack by the metallic melt. The authors should be commended for making an effort here. I would le e to say that this effort represents a best-estimate assessment based on engineering judgment. At present, there is no adequate database or computational tool for large-scale stratified explosions.

On page 7-1, the authors state that "Also in this chapter, we would normally present a series of arbitrary parametric and sensitivity calculations, to illustrate, for cases where the base results happen to be beoign, the margins to failure" and claim that "This, in effect, has already been done, too, by the breakup and triggering calculations, in the course of bounding the behavior".

I believe additional work would be needed to make this claim fully valid, and I am confident that the authors will succeed in doing that.

Finally, the authors are to be conunended for conducting such a detailed evaluation of a very complex issue.

Sincerely,

_/l,,,

D. H. Cho Reactor Engineering Division

/at E-38

SEM SY:UW-NADISON  : 1- I-97 1 15:18 1 COLL OF ENGR. ADMIN + 1 630 252 4180is 2

( vc.s 6.wdL G.L.

Nuclear Ennineering and Enaineerings Phvalca copertment of sneineenne Phyeles, universny W wissensin MM Isueiner sessey neeseesh center .

1000 Jolmeen Ddwo stodeen WI83700 .

I hone:(808) 306 319454001 -

Pam (808)3884707/3 4400 August 13,1996 Prof. T.O.Theofamous Chemical and Mechanical Engr. Depts.

Director, Center for Risk Studies and Safety University of California, Santa Barbara Santa Barbara, California 9310G1070 .

Dear Theo,

I roccived your letter of July 9th,1996. and the enclosed reports. I am still waiting for the two verification reports (DOE /ID-10503 and 10504]. Since Ifirst roccived the initial request by Mr.

Steve Sorrsil, I have been thinkingabout this review process and have come to a couple of decisions that I wantedto inform you about. First, because of the depth of your IVE report and O * > > <i S.4 - < it - iii t * - < ** 2 4 *

'*Ihus, I will only be able to begin the review in that time period,and plan to continue oe ix vi - r *.

extent I feel is necessary to adequately understand the analysis you and your colleagues have done. Second, because of my association with the Nuclear Regulatory Commission in research and associated consulting, it is not appropriate that I be paid for this review.

I will follow the timetable that has been set for this review as best as I canand will send copies of my comments to the USNRC research staff. .

Sincerely, .

Michael Corradini Nuclear and Mechanical Engr. Depts.

Wisconsin Distinguished Professor xc: Dr. Sud Basu E-39

_ _ _ _ . . _ _ _ _ _ _ _ _ _ - - _ _ _ - - - _ _ _ ----- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ~

i 1- 9-91 1 15t19 iCOLt OF EN % 'ADMle 1 630 252 4180is 3 SENT SYtW-MADISON INITIAL COMMENTS sad QUESTIONS fc "EO/ID 10541

1) he authors do a goodjob in giving a context for their wodc. Howeyw, I amnot sure if this analysis which is provided is a failure analysis for the AP600 reactor prusure vessel or a design analysis for the RPV. %

former implies thetit would be a 'best estimate' analysis, whus the latter must account for factors of safety to assure survival. Can the authors clarify what is the ultimate intent of the work?

2) la Chapter 3 the authors denne the failure criteria and the fragility curve for the reactor prusure vessel. IfI understand the approach a strainfauurs limit is used and the amociated analysis suggests a RPV lower wall failure probability ranging from 0 to 1 for a spectrum ofloading paherns withlmpulus between 200 - 400 kPesec. Structural failure limha are a complex arsaof work and I have asked colleagues to help review this chapter, Currently, I wonder how this failure envelope compares to that of previous LWR plantanalyzed for an in-vs.nel steam esplosioni c.g., the ZIP study in the early 1930s?

3) In Chapter 4 the authors' m4cr point is that the core and vessel designis sufficiently different from past LWRs, such that the core melt behavior is quite different. Two aspects are emphasised: first, the lower corssupport plate and the non active fuel length above it (30cm) is large enoughin sise to delay the enre melt progreeston downward; second, the core stal reflectorin the radial direction is aho thicker (over 15cm], also delaying and changingthe details of radial core melt progression. In essence, the ' race' to thebreach of the cori m melt crucible, which is formed dudng the meltdown. downward or radially outward is governedby these boundaries h authors use a specific 3BE core melt accident sequence toillustrate this behavior. If one accepts this premlee about a radically diffarent core geometrical design, a few questions arise:

e) What is the sensitivity of meltdown timing to downward boil off of water? More exampics are needed.

b) is the core reelt event timing essentially independent of accident sequence? No guidance is given hers.

c) h corium exit flowrate seems to be not by the ' rip' in the reflectorsions the radial edge of the core region at the very top of the pool. Is thisa realistic estimate, since it is not much more than that one would calculate from adiabatic heatup and meltdown of the corst e.g., as evidenced at TMI2 7 % authors suggest that 200to 400 kg/sec " appears to be a reasonable rance physically to bound the behavior"t but I wonder if we really know that much about this core melt failursprogression in a radically new geometric design that this flowrate is a

'remonable bound'? Morejustification is needed for me to 'huy' the argument.

d) This last question really leads me to the key question of this whole analysis;i.e., the authors leave me with the impression that there is a good deal of certainty in the melt progression and I have signifkant trouble acceptingthis premise. Speci5cally, the whole analysis hinges on the fact that the melt crucible which forms

' during the melt prostussion has a structural integrityof enough certainty that it would release the melt radially through a pour area no larger than 0.02 to 0.04 sq. meters. This admate also seems to berobust enough that it would be a " bound" even with coolant esflood into thecore region and any possible disruptive events that may occur. I am very dubious about this and need to see more analysis tonceept this as a ' reasonable bound *,'11ds inelt failure location and pour rate is the key determinant in limittag subsequent FCl energottes.

4) In Chapter 5, the authors detail their multidimensional premixing analyses.As stated previously, the melt flowrate of 200-400kg/s seems to predetermine thebenign nature of the FCI energetics, but mixing is also part cf the process. A few questions arise here:

a) Why has the effect of RPV pressure been neglected? Premixing will occur atslevated pressures not 1 bar

[2 5 bars) and this will affect the mixing process. Also,will the rise in pressure locally during mixing will cause the pool to subcool and this has been neglected. Were calculations done to ' bound' these effects?

b)'The authors seem to have only considermi the pumixing process as the moltfalls through the limited water pool Bom its surface to the curved RPV bottom.Would not mixing continue u the melt continues to fall along

• the wall. Thisscems to have been neglected. Is this premixing process of no importance or is the premixing E-40 l

SENT SY:UW-IIADISON 1- g-87 i 15:20 iCOLL OF ENGR. ADMIN-* 1 830 252 4700is 4 analysis with PM ALPHA not valid for thesalonger dmes?

c) The biggest effect of thsee srnall pours in my mind is that they may cause local PCIs which do not harm the RPV but totally change the melt pooring characteriodes for subsequent melt poursi f.e., those small pours and 4 MM PC! swill damage the core melt crucible and markedly insmese the melt Sowrase or change its location. The authors have sons to great pains to doestmine the Angility of the RFYwall, but totally ignore the fragility of the melt encible and the anect of these PCIs. I wouldsuggest that larger melt pours will be indoned ham the le;ttom of tbs emelbleas well as along its radially edges with larger holes, al! caused by early small PCIs. How have these events been considered or conservadvoly bounded for RPV survival?

• d) Finally, the PM ALPHA model has a parametric fuel breakup model that ismentioned briefly, but has yet to be asesssed against asperiments. Fcr thesesmall pour rates, the modal effect is not of great interest, but would be for larger pours in these complex geometrin!: this model discussed in the suppost documents?

4 5] In Chapter 6, the amhors use ESPROSE.m in a sovised 3D version to simulate theexplosions within the RPY lower planum. Given the premixed mass of fuel we have a range of results given in Table 6.1. Ont.v a couple of quesdons arise:

a) Why is the trigger time to short; 1.e., much less than 1 see? Is it due to the dme to the RPV wall?Why

cannot further mixing along the PPV wall cause larger explosions?

b) Why is the impulse largest for tbs mid range value of ' beta'? Is the impulseof 200 kPa-sac near the failure limit? or am I reading this prediction correedy?

c) The detailed calculadonal results in Appendix C abruptly stop in manycases at 1 or 2 or 3 milliseconds.

Why? Is this an indication of something numerically fatal in the ESPROSE.m simuladon or what is up?

l '

l 8 6

SENT SY:UW-NADIS0N i 1- 8-87 i 15:20 iCOLL. OF ENGR. ADilI P ' 1 s30 252 478018 5 Nuclear Engineering and Eneineering Phvales ,

empenment et aneinsertas physise, uemerehr er wissensin moeteersesser passeersh censer  !

isesJohneen avove bleeleen.WI B5798 Phone toes)3e6419853001 Pan (sos)ssa.eform.seos

• January 3rd,1997 Prof. T.O.Theofanous l Chemical and Machmatem! Engr. Depts.

Director, Cantar for Risk Studies and Safety University of California, Santa Barbara

• Santa Barbara, California 93106-1070 -

DearTheo,

I am sorry for my tardincap in reviewing your reports. DOE /ID-10503/04. After receiving the PM-ALPHA report in latter October, I had to finish my class and administrative obligations and I thus was unable to give it the proper attention. I have read them over the Christmas break and have some comments. On balance both reports are well written and address the verification of the computer models. If this were a model/ code review for the NRC, we would also need a complete code manual from each for review. Appendix A gives excerpts, but I wonder if we h-can get the complete code theory / users-manual? Attached are my comments and questions. '

H:pe to see you at the next CSNI meeting in spring.

R gards.

Michael Corradini '

Wisconsin Distinguished Professor Nuclear and Mechanical Engineering

SENT BY!UW-MADISON i 1- I-97 i 15121 iCOLt OF ENGR. ADNie 1 830 252 4780is s O

INITIAL COMMENTS and QUESTIONS for DEO/ID 10504  !

The overall report is quite informative, but I do have specific comments / questions that need to be addressed.

1] %e analysis of the QUEOS experiments are very interesting. Por any of the expedmonts  !

(Q5,6, g,10,11] the visual image is compared to the code, and the leading edge, level swell, steaming tute, steam produced and pressuru is compared. My first question is what is the criterion to determine the leading edge? In the pictures for the tests, specifically Q10 and Q11 it seems to me that PM-ALPHA is predicting the movement of the front to be faster _than the data indicates. Yet in the plots the opposite is represented. Either there is a contradiction or I am observing numedcal diffbslon in the images and the researchers have a definition of the leading edge that " corrects or compensates" for this. I have seen the same behavior with IPCI and therefore, am sensitive to it. His needs to be sorted out before I would say that the agreement in the kinematics is acceptable. De MIXA results in Section 3 r, cam to indicate the same behavior to me and thus I am worried about this numerical diffusion. Dere was also no study of the nadalization effects in Section 4 and this is surprising given the results in Section 2. Als

! seems to be a logical thing to do and really should be done.

LO 2] he second comment about QUEOS relates to the radiative hest transfer model. On page 2-16 the report states that the radiative model had to be changed from what is normal in PM-ALPHA to accommodate the experiments. Later on page 2-19, the report states that the tests do not meet the ' fitness for purpose' criteria, and one reason is that the temperature is too Iow (2000C compared to 3000C). I am troubled by this empirical "fix" to model the test and thus, am wondering about the " mixed" transport model in PM ALPHA. his is known to be a tough problem, clearly, but to arbitrarily change it seems too rough. Also, I disagree that the tests are not " fit for purpose". ney are more fit than others and thus, savery relevant. nus, the proportion of the radiative transfer that goes into bulk heating versus steam production is important to consider and impmvc upon.

3) I would also like to see a calculation of PM-ALPHA /3D for QUEOS if indeed there is a benefit to a 3D calculation. It seems that the QUEOS tests are the largest and highest temperature simulant tests to date with solid particles; thus, it may be of use.

_ - - - _ _. . _ , _ - , . - . - _ y

~~

SF.NT SY:UW-MADISON I 1- 9-9f i 15i21 iCOLL." Or Usu. ~A d # ~i'aiw eas'e/ow~.o i

4) he report finally examines the FARO-LWR test L-14 as a comparison with a large prototypical simulant melt poured into water %is seems like a reasonable compadson test, but I am surprised about what data is compared. Dere is an enormous amount of data available over the first twenty seconds of the test [the first 54 seconds is reasonable before heat loss comes significantly into play) and yet the data comparison is sparse at best. I would suggest the  !

following vadables he displayed and compared over the first 54 seconds:

a) the total pressure and pressure dse rate (done now) b) the steam and water temperature at a few locations since its 2-D c) the kinematics of melt entry and arrival at the chamber base and settling d) the surface area generated by the breakup as a funcdon of time e) the mean particle diameter as a function of time f) the energy flow to vapor and coolant liquid and loss by fuel 31 the level swell of the pool (done but not for long enough times)

Also I am concemed about the arbitradness of the dynamic breakup model [p value = 50), that is used and described in pages A-34/35. nis whole procedure is a matching exercise for some value of beta unless the results begin with a jet of -10cm and break up to a size that is consistent with the post test debds data [as well as the amount left as a ' cake']. It would seem advisable to compare the 'frosen' rnodel to other FARO tests to prove that results can be consistently predicted for LO6, LOS, Lll, L19 and L20; all of which were high pressure tests for quenching. Also the ' mixed' transport including radiative transport wouldhave to be held constant in these compadsons to prove the match of L14 has some limited ' universality'.

h E-44

SENT 'tY:tM-NADIS0N 1 1- 9-17-1 15:22 .iCot.L OF ENGR. ADMIN

• 1 830 252 4780ls 8-i n

s O

i t

.. INITIAL COMMENTS and QUESTIONS for DEQ/ID 10503 L The repost is very well organized and describes in sufficient detail the ability of ESPROSE.m to L perform shock propagation calculations for gas / water ane'. vapor / water situations. I do not ,

completely understand the origin of the CHAT [or CHAT-QL) code comparisons. Are these standard code models or a formulation of the authors to Jo a code cross-comparison? I understood them to be the latter, and thus I wonder thout th need to compare to actual

. asperimental data on shock propagatiou in single phase and multi-phase systems. %1s is a minor point, but I think for completeness a link to data 'un best. My main comments are about the comparison to the KROTOS data.

1) he initial statement is made that the KROTOS tests are a challenge since they have imperfect characterization of the laitial conditions. One ouestioa may be if thers are any other i p tests which give them more insight? After my own sea.ch, however flawed these tests are,

V these and other one di=amional experiments are the best we have. My other comment is about

the initial conditions. De comments on page 4-26 indicate that the fuel mass and ficwrste, but there is a problem as far as I can tell, he mass is correct, but the initialjet size is not I cm but j 3 cm and I think the flowrate of 1 kg/sec is too low by at least 50%. Finally, the fuel particle temperatures are different for each of the tests noted as is the location and timing of the trigger.

I am not sure that the authors are aware of this. I can send them this information if needed, but in the case of KROTOS 38, the initial conditions are not correct; e.g.,'the jet elze is 3 cm and the trigger time is 1,12 sec at or near K3 and not at the leading edge, with a pour time of about l 0.75 sec.

2] ne concept of using the parametric mixing model for a p value of 30 or 50, again rains the question of what is appropriate and why, ne kinematics in Figure 2 don't have any comparison to the thermocouple data for position of the melt as a function of time and give no indication what 50 is "better" or more correct than 30 for a value. Also what is the time evolution of the particles as the jet breaks up from 3cm to what size? None of this is discussed at all.

i

?O e4s e,

~

ftENT'BY:UW-MADISON j 1- b-W[ i biH iGOLL ~ ur cNen. Au:una . i 049 m *(ouoe ,

1 i

3) h Saal point is the use of the parameter, x f= 0.5 to 1.0. Does this pcrameter mean that when the value is 1.0 all the fuel is TwH as it is fragmented with some fraction of water and steam? If that is the correct interpretation then, the pressure plots do not seem to make

sense to me. ' Ibis is especially the case, since the predicted void in by Figure 2 and Figure 3 is very small. 'thers is something minains in the description;'since 1.5 kg of molten alumina has the energy of 6-7 MJ and thus must be quenched by almost all the 35kg of coolant if there is to i be such a 'small' pressurization with such little void. How much water is " assumed" to be irtermixed with this fuel to give the pressure signature we see? This is never discussed and it is

the esost crudal part of the model % complete picture is missing and thus, I am not prone l i to tgree this is a amanaamhle prediction until all the ' parametersare speciSed and explained.

Also, comparisons to mass than one test is eW ' Ibis has been done with other PCI Mia.

. l I

i O

1

\

t l

l l

1 4

E-46

.. -- l l

I fi E C EI V E D

~

l , REACTOR ENGINEERING DIVlCIC;l 1

-DIRECTOR'S OFFICE-V' i' JAN 1 ? 1997 3

rae..,one.

i ACTION: *M rau s . u s --

' INFoRMATION: T h .U m ,j, '

d.?%7 " ' ' % C January 14, 1997 Dr. L. W. Deitrich Director, Reactor Engineering Division Argonne National Laboratory 9700 South Cass Avenue Argonne, Illinois 60439

Subject:

Review of "Irwer Hemd Integrity Under In-Vew1 Steam Explosion Loads" by T. G. Theofanous et al. ,

As requested in your letter dated June 17,1996, the following comments are; offered in the areas of Meltdown / Relocation Phenomenology and Steam Explosion Imads.

h Meltdown / Relocation Phenomenology - We agree completely that a . downward relocation path of the melting core material through the core support structure (and resulting large fuel pour rates) is " physically unreasonable". Furthermore, the predicted relocation off to the side and from a fully developed melt pool leading to a molten fuel pour rate into the lower reactor vessel plenum of about 200 kg/s, is consistent with the Three Mile Island Unit 2 Core Relocation as described by Epstein and Fauske (Nuclear Technology Vol. 89, p.1021-1035, -

December 1989). Fuel pour rates of this magnitude by themselves eliminate concerns relative to global vessel failures, even if an energetic steam explosion is postulated. As illustrated by Epstein und Fauske (1989) and Theofanous et al. (1996), such low fuel pour rates limit the fuel

. that can be found in transit within the lower plenum to values at least an order of magnitude less than the required for incipient lower head failure (3 to 5 tons). Quoting Epstein and Fauske (1989) "A key aspect of the relocation is, then, that significant quantifies of corium melt were not mixed with water at one time. The slow melt relocation phenomenon is, perhaps, the most important piece of information gained from TMI-2 studies and should figure prominently in future assessments of steam-explosion-induced containment failure as well as lower reactor vessel plenum failure due to fuel debris overheating." This is clearly the case in the current assessment

"~" '

provided by Theofanous et al' .

Steam Explosion Loads - Having eliminated the potential for glotial vessel failure, Theofanous et al. proceed to evaluate the potential for localized damage, by considering local shock loading, with peak amplitudes in the Kbar range, as a result of a steam explosion occurrence. Again, the conclusion is that failure is " physically unreasonable". This conclusion l

U E-47 16WO70 West 83rd Street

• Burr Ridge. Illinois 60521 * (630) 323-8750 Teletat (630) 986-5481 = E-Mail- tai @!auske.com

, is further supported by noting the following observations.

The above loadings are produced by subjecting the limiting premixtures at atmospheric l

pressure to triggers resulting from releasing steam at 100 bar. Quoting the authors, "our triggers are chosen sufficient to initiate explosions, and they have no relation to what might arise l'

spontaneously during a pour." We agree with this observation, and in fact believe that the occurrence of spontaneous steam explosions with the molten corium-saturated water system at The atmospheric pressure considered by Theofanous et al. is " physically unreasonable".

enormous film boiling heat flux (~ 3 Mw/m )2and corresponding vapor flux resulting with this 2

system (several times the critical heat flux of ~ l Mw/m ) promotes separation and prevents physical contact between the molten corium and water, a prerequisite for steam explosion given a fuel-water pre-mixture. Temperatums (~ 2000*C) which are well below the melting temperature of corium (- 2700'C), would be required in order to reduce the vapor flux in connection with film boiling to fall below the fluidization vapor flux. The above considerations are consistent with the noted absence of "explosivity" for the corium-water system (I. Huhtiniemi et al., "FCI Experiments in the Corium/ Water System", NUREG/CP-0142,1712-1727,1996).

This is in sharp contrast to the noted "explosivity" with the often used molten alumina (Af230 )-

2 water system. Here the estimated film boiling vapor flux (~ 0.5 Mw/m ) is well below the fluidization vapor flux allowing physical contact while the alumina is still molten. While the noted effkiencies are quite low, the super critical pressures observed with the alumina-water tests in the KROTOS faciilty (Hohmann, H. D. et al., Nuclear Eng. & Design, 155, 391-403, 1995), apparently encouraged Theofanous et al. to model such events and apply them to the.

LWR system.

We also reviewed the approach taken to assess the steam explosion created impulsive loads. Certainly the efforts performed by the authors are impressive in the number of analyses performed and the detailed graphical presentation of the results. The calculations are based on the microinteractions model which appears to be applied in a self-consistent manner. Our question is whether this is the only way that the relevant experimental information can be interpreted. The underlying supposition in such models are that dynamic fragmentation and intermixing of melt and water can occur on an explosive timescale. Certainly the available information shows that fragmentation can occur during an explosion. However, we are not convinced that the elements associated with fragmentation and intermixing on such a rapid timescale have been demonstrated. In particular, the SIGMA tests performed with molten aluminum indicate virtually no fragmentation for melt temperatures where numerous large scale studies have observed explosive events.

Furthermore, the detonation concept is compared to the KROTOS molten tin-water and molten aluminum oxide-water tests. With the agreement from this comparison, the authors, in a previous DOE report (DOPJID-10489), conclude that " low void fraction geometries can produce highly supercritical, energetic detonations." Our analyses show that there is an alternate explanation to the KROTOS experiments that requires no melt fragmentation. If this is the case, the comparison of the microinteractions model with the KROTOS experiments indicates nothing more than that ESPROSE approach is consistent with the experimental observations. It does not provide justification for the microinteractions phys'.al concept.

E-48 g <

+

Therefore, we are of the opinion that the approach taken in this document is conservative ,

la that it overmales the possible Iceds that could be created as a result of thermal explosions.

. O .

If a design evaluation uses this model and concludes that the boundgy wc::!d nd be challenged, we believe that the conclusion is sound. However, if the modeling approach is used and the resulting loads exceed the capabilities of the structures, we do not believe that this represents an actual challenge to the system integrity.

Before such a fragmentation approach can be recommended for realistically assessing the structure loads, it should be proven that a relatively small pressure increase would be

sufficient to self-trigger a coarsely fragmented and intermixed system. In particular, it should

. . be demonstrated that a coarsely mixed system could calcalate from a small triggering event into i an event like that characterized in these evaluadons. Assuming that a single grid is filled with steam at 100 bars as a triggering mechanism, it is far too coarse to provide such a definitive representation.

- In summary, we believe the mo6eing of fuel relocation and quantification of premixtures to be reasonable and consistent with experimental observations including the TMI-2 incident.

On the other hand, the assessment of steam explosion loads appear to be very conservative. The corium saturated water system is not likely to exhibit "explosivity". 'Iherefore, it very strong case can and has been made for the effectiveness of "in-vessel retention" as a severe accident management =~*pt for a reactor like the AP600.

Sincerely yours, O a x.G 42LS E :.

ans K. Fauske Robert E. Henry C

President Senior Vice President I

L p HKI': lab l

l l

w - i-.e% -e ,,% --- ., -w - - r- er m - + .-

i

_- h THE UNIVERSITY OF SYDNEY

- -- DEPARTMENT OF CHEMICAL ENGINEERING NSW2006, AUSTRALIA g

TELEPHONE: 61 7 9331 4147 FAX: 612 9J512:34 DX: lis4 SYDNEY EMAIL:

Dr. David F. Fletcher davidf@ chem. cog usyd.edu.au senw Rm mhfellow Dr. L.W Deitrich, Argonne National Laboratory, 9700 South Cass Avenue, Argoane, IL 60639, USA.

Dear Walter,

Please find enclosed my review of the DOE project on " Lower Head Integnty Under in-vessel Steam Explosion Loads" by Theofanous and co-workers. As you will see from the review I jadge it to be an excellent study in terms of its depth, scope, technical quality and shear volume of work. I fully agree with the conclusions drawn by ,

the authors.

. Yours sincerely, l

9 '

l , David F. Fletcher (Dr.)

1

! cc. Prof. T.O. Theofanous l R E T EI V 5~D -

a REACTOR ENGINEERING D!VC:;.'l

! -DIRECTOR'S OFFICE- ,

NOV 201996  !

. c.u..

~ ~ -

u3 _.

j ,

I iM untA 7.0t.: 'M5T UD E-50, 7 you ~.IE h

'F i r\.

! J Review of DOE /ID-10541

" Lower Head Integrity Under In-Vessel Steam Er.plosion Loads" by T.G. Theofsnous, W.W. Yuen, S. Angelini, J.J. Sienicki, K. Freeman, X. Chen and T. Salmassi Reviewer: D.F. Fletcher, Department of Chemical Engineering, University of Sydney, NSW 2006, Australia.

November 18,1996 Summary This review rovers the study of lower head integrity under steam explosions performed at UC'SB by Theofanous and co workers, together with the code validation reports for PM-ALPHA and ESPROSE.m. The study and validation reports contain a massive amount of very high quality work. The depth of the study and extremes to which the authors have gone to use validated tools is second to none world-wide. For example, no one else la performing

, O-O 3D premixing and propagation calculations.

The work is of very high quality and in my view the conclusion that steam explosion indured lower head failure is unphysical ls completely justified. The technical arguments support this with a high degree of redundancy.

1 Introduction Firstly, I believe it is important to comment on both the quantity and quality of the docu-mentation supplied for this review.The very complete verification manuals for PM ALPHA and ESPROSE.m are unique. A minor semantic point but they are much more than veri-fication (whleh implies that the code does what it should) manuals but are also validation manuals na they examine how well the rode represents real experiments.

Secondly, I wish to record that I wasimpressed by the scope, depth and quality of this study.

It provides a very comprehensive basis for rejection of steam explosion induced failure of the lower head.

The remainder of this. document presents sperilir comments on the Study and the two validation reporta.

/ C O- E-51 1

2 The Study (DOb/ID-10541) g This section deals with the main document of the study (DOE /lD-10541) and pays particular attention to the steam explosion part of the study.

2.1 Introduction This sectien gives a brief summary of earlier work on lower head failure. It discusses three earlier studies by Bohl et al, Theofanous et al and Turland et al, all of which highlight the need for mechanistic pressure loading calculations before the lower hevi issue could be addressed adequately. This is the first such study in which this approach has been possible.

2.2 Problem Definition and Overall Approach This section sets out the methodology to be used. Essentially, the now established ROAAM procedure is used in which the overall event in split up into well-defined physical processes that can be modelled, combined with intangible parameters (such as triggering time). The proposed squence of events and the split between physical processes that can be quantified using a entidated model and those which must be treated in a parametric manner seems correct to me. In particular. I believe that the flow chart shown in Figure 2.3 gives a correct and well-judged progression of ents. Details of the modelling will be discussed later. How-ever, it is important to emphas ze that the identification of a sound methodology is very important and I believe that the authors have done a good job at this stage of making the process transparent.

2.3 Structural Failure Criteria This section deals with quantifiertion of the likelihood of vessel failure for a transient, localized load. The material la presented in a clear manner and there is a step-by-step progression from an axisymmetric model to the examination of localized loads. The analysis presented in equation (3.10) and Figure 3.8 provides a neat means oidetermining the effect of localized loading and the performance of equation (3.10)in correlating the data is impressive.

Also 1 believe that the failure criteria given in Table 3.3 are sensible and fit the presented database.

This chapter is important in that it sets up the basis for the determination of whether a particular explosion toading will or will not fall the lower head. There would appear to be significant conservatism in the analysis, as noted on page 3-1 and from Figure 3.8 at the high impulse end. and therefore it provides the required function for this study.

E-52 2

- - - - - - -- --- ~ - . . - - - - ---

.1 a

i i

N 2.4 Quantincation of the Melt Relocation Characteristics i

This section presents an_ analyals of the melt relocation characteristics, it is important to -

note that the analys'i4 does not une a system code but instead a number of highly specific 1 l

models have been developed to address the physleal processes i eemed to be important. This -

was the approach followed in the Sitewell B study med neem to me to be the correct way to  ;

proceed. Based on my participation la the Sizewell B study I believe that the methodology ~

used and the conclusions drawn are correc.t.

The melt flow rates and release conditions are consistent with those found d in tie Size

study. In partkalar,I believe that massive pours of many tonnes per second ha te been rule out on the correct physical baals, in the section on retooding the authors do not consider the possibility that a steam explosion may occur as the water refloods the molten pool, it is covered in a later section and it would perhaps be wise to have given .s forward reference here. .

W 2.5 Quantification of the Premixture i This section addresses the determinattor. of the premixture configuration. Firstly, it is im -

' portant to note that the highly 3D nature of the pour has been taken luto account via the extension of the Pht-ALPHA code to 3D. Thus the localized, rather than smeared in 2D, characteristics of the melt water interaction process can be simulated. Secondly, it should be noted that melt breakup has been taken into account in a parametric manner. At first sight this may seem like a weakness, as many proposed breakup models exist. However, given that none of these has been praperly validated it seems appropriate that the effect of breakup be addressed in a parametric manner. As pointed out in the report,in the event that the melt enters the water pool and runs along the vessel wall, there will be less mixing than calculated here and t.herefore the explosion energetics will be reduced.

Based on my experience of premixing experiments and modelling I have no difficulty in believing that only tens of kilogrammes of melt are likely to be mixed in the gisen config-uration. Clearly the high voiding rate is a consequence of the water pool being naturated. I was left wondering whether in the event that the melt pour occurred during the reflooding piocesses whether there would be aufficient subcooling present to increase these masses sig-nificantly? .\ ly expectation is that the incrrase would be by no more than a factor of two, which would still result in small mixture zones.

. '2.6 QuantiReation of Explosion Loads This section deals with the determination of the magnitude of the possible explosions that could.be generated from the premixtures calculated using Phl-ALPilA. It is iniportant to

p-r note that these calculations, performed using ESPROSE m are fully 3D and can therefore account properly for explosion venting. The validation of the model lx discussed in a separate E-53 .

3

section, it is sufficient here to note that the code has been subjected to a very significant validation effoit which I believe shows that it is ' fit for purpose'.

I agree with the approach adopted regarding triggering. Specifically, triggering at differen times and looking for the maximum load is clearly conservative,in addition, the effect of the premixing breakup pr rameter J is consistent with experimental observations and hig the fact that the unccrtainties in breakup can be taken into account in a parametric manner.

Given the premixture contigurations determined using PM ALPHA I am not the least sur-prised that none of the explosions challenges the integrity of th lower head.

2.7 Integration and Assessment This very brief section explains that as a consequence of the methodology and results the is no need te continue with the probabilistic approach because of the enormous mismatch between explostca loads catrulated and those required for failure. In order to show that this is not an artifact of the approximate structural treatment, full ABAQUS calculations showed there to be no problem.

I agree that the only way to obtain a significant explosion is to have extensive mixing whic requires highly subcooled water. I believe the arguments against this are sound, especially if one keeps in mind that the enormous amount of heat which would be stored in the lower core support ' structure would be availa' ole to remove subrooling.

2.8 Consideration of Reflood FCIs This is an important nection, as the above analysir. has clearly shown that premixed 'explo-sions cannot cause failure of the lower head. I agree with the view taken that you need a very substantial overlyirg water pool to provide sufficient inertial constraint to generate a high pressure explosion. As in the previous stenario everything is against this, viz. the low water addition rates, the case with which '.he melt surface freezes and the fact that as film boiling orruts the overlying pool will develop voids reducing its ability to constrain. The analysis rules out to my satisfaction the possibility that stratified explosions could fail the vessel.

2.9 Conclusions The ennelusions contain a summary of the results presented in the earlier chapters and presents a concise summary of the important physical features of the system and the physical mechanisms which lead to the conclusion that f4 nre of the lower head by a steam explosion is unphysical. I really appreciated this carefully presented summary.

E-54

- - . - . _ . - , . ~ - -- .- . . - .-. - -

. s 3 - - PM-ALPHA Verification Studies (DOE /ID-10504) _

s

- This section presents a review of the Pht ALPHA verincation studin report. It is important

to note up front that Pht ALPHA hu been the subject of continuous development and peer review (at conferences) over an 8-10 year period. It is therefore a mature piece of software.

3.1 Introduction The main point of interest in this section is Figure 1 which lays out the verification and validation approach. This is very comprehensive and covers numerical aspects, comparison with other codes and analytical _ solutions and with experimental data. I can suggest no improvements to this validation matrix it is also worth noting that this section highlights the new feature of Phi ALPHA, namely extension to 3D which in clearly needed in.the Study. This ciently representa a massive amount of work but the new insights gained are.

definitely worth the effort.

3.2 Multifield Aspects This section deals with the testing of the multiphase constitutive relations and the mod-7 elling for the. sedimentation of particles or clouds of particles. PM-ALPHA computational T

results are compared with experimental data and analytical models (based on the ddft flux approximation) for the sedimentation of single particles and clouds. In a!! cases agreement is excellent. A novel feature of this p'.esentation is that the trajectory of the solution in drift flux volume fraction phase space is presented. These results show that the solution is-c approached in a variety of ways and helps to explain wh'y multiphase numerics prove to he no complex. These results confirm that the code can reproduce the correct particle fall L ,

' speed, an important feature the steam explosion study.

L

{

Numerous refereed papers have been presented showing that PM-ALPHA can simulate the MAGICO tests, where in most cases there is also pha.~e transformation. These simulations also show good agreement with localdata on mixture composition and void fraction. This is g

l important as PM-ALPHA must predict the correct mixture composition if the calculations of explosion propagation are to be reliable.

Comparisons of PM ALPHA simulations with data from the QllEOS tests are generally good. There is evidence of numerical diffusion in, for example. F:gure 6 but the authors are aware of this and are planning runs on finer grids. In the hot cases I agree with the . ,

authors that both the relatively low melt temperature (making radiation absorption a surface phenomenon) and th_e gravity-induced subcooling are important. If the explanation of the difference in stmm production advanced in the textla correct (namely the superheating of a layer of water during the fall stage)it means that interpretation of experiments of this type, p

- where there la relatively little steam production, will always be very complicated. Given the

. short time available to the ' authors to analyze this data and the experimental uncertainties

~y l feel that PM ALPH A performed as well u could be expected.

E-55 .

5

3.3 Integral Aspects The code comparisons with CHYh!ES and between the 2D and 3D versions of the code give a high degree of confidence that the basic numerical algorithm is correctly coded and that the 2D and 3D approaches are consistent.

The comparison with dat.'. Aom the MIXA06 experiment is at least as good as that achieved by the experimenters using the CHYhiES code. The lack of melt spreading in the simulations is very similar to that found using CHYMES. The level swell and steam production data are well reproduced given the experimental uncertainties. Again it is fair to say that this test is well simulated given that there are severalimportant experimental uncertainties regarding particle breakup and the steam flow rate.

The comparisons of code calculations with data from the L 14 FARO experiment are also good. In this experiment there is no local data and only global quantities, such as vessel pressurization and level swell, are available for code comparison. The choice of parameters to match these data seems very reasonable. I found the figures illustrating the non local absorption of radiation interesting and these clearly illustrated the importance of this phe-nomenon for high temperature melts. To my knowledge these are the first calculations to include this feature, which is clearly ofimportance in high temperature melt applications.

3A ' Breakup Aspects

' O I completely agree with the chosen approach to breakup. As more tests are analysed it will he possible to increase the degree of confidence in the chosen values for the parameters.

Clearly, given that the melt surface area transport equation is already coded it would he a simple matter to include a mechanistic model should a validated breakup model become available. However, the analysis presented in the study shows that the overall predictions of loading are insensitive to the choice of these paramt:ers. Therefore the lack of a detailed model does not in any way effect the conclusions of this study.

3.5 Numerical Aspects .

The authors are clearly aware of the need to avoid numerical differencing errors and the presented calculations show that they are taking care to address this problem.

3.6 Concluding Remarks I think this section identifies the correct r.reas for future focus. If I were the authors I would have made more of the fact that this is the most comprehensive validation effort to date and that the code has performed extremely well.

E-56 g

N 3.7 -Appendices (V -

Appendix A provides a comprehensive description of the constitutive laws and Appendix B provides a detailed paper on the hlAGICO tests.The reviewer is familiar with the material la the Appendices and this has not been reviewed in detail.

4 ESPROSE.m Validation Studies (DOE /ID-10503)

Firstly,it is important to tackle head on the ESPROSE.m formulation, which I believe it is fair to say has not been widely accepted. I find it hard to understand why this is the case.

Essentially, the novel feature in ESPROSE.m is the inclusion of an additional fluid (the m-fluid) whleh represents the fragments and the fluid in intimate contact with them which is being heated. The need for such an approach seems beyond doubt to me following the very careful experimental analysis of Balnes [1] and my own attempts to analyze KROTOS-like tests using CULDESAC [2]. The authors have provided comprehensive experimental data for appropriate pressure loadings to show Se finite mixing rate. They identify the need for an enlarged database but it should be recognl ed that the ESPROSE.m fonnulation is conservative in the sense that by mixing the fmgments trith only a fmetion of the coolant they genemte high local pressuru. This point should be kept in mind when examining the use of ESPROSE.m results.

O) tv The remainder of this section contains detailed comments on the various chapters of the verification report.

4.1 Introduction

~

The main feature of this chapter is Figure I which gives the validation strategy. This is very l

extensive and to my knowledge is the first model to be subjected to specific wave dynamics and explosion coupling verification studies against anslytical and experimental data. It also covers the two main experimental programmes KROTOS and ALPHA.

l 4.2 Wave Dynamics The ID solutions for the shock speed and particle velocity (important in relative velocity l

fragmentation) are excellent. The same applies to wall reflection studies, the effect of void l

and the effect of non condensable gas. The venting calculations also show good agreement with the CHAT resnits. I was curious to know why the calculations were performed for j

a pressure step of 40 bars over a base pressure of 100 bars and over a space dimension L

of only 1.4 cm. I would have preferred to see venting on the 0.1 m scale (with a 1 cm mesh) and a pressure difference of say 10 bars venting to atmosphere. Figure 9b shows that

]n the ESPROSE.m results only exhibit dispersion at the first few time steps and that the numerical diffusion is modest.

E-57 .

The 2D comparisons are impressive and show that ESPROSE.m captures the wave dynamics very well. The only point that this section raises for me is why in the type B behaviour the ESPROSE.m results have a spike at the origin (as expected from the source description) but the analytic solution does not (see Figures 7,13 and 19),is this simply a plotting omission?

The experimental comparisons with data from the SIGMA facility are interesting and show that ESPROSE.m is capturing the average wave behaviour well. Clearly, the pressure trsns-ducers are picking up many local reflection events which are due to the inhomogeneous nature of the ' mixture' and cannot be modelled via a continuum approach. I am surprised that ESPROSE.m has done so well for this system with the only apparent systematie dir-ference is the tendency for a ~1 ms time lag.

This section provides very solid verification for the code algorithm and the choice of solution parameters. l l

4.3 Explosion Coupling l

This section contains test cases in which energy is input into the gas phase via a parametric l

relationship in which the energy input into the gas phase is proportional to the fluid velocity to the power 1.5. This is done to represent the fact that in ESPROSE.m the energy is input l In the m-fluid. Results for calculations for bo'.h cases considered are in excellent agreement with the CilAT simulations. Figure 5 for simulations on a larger space scale, examines the effect of grid size. The comparisons av good with differences being confined to the interface region.

4.4 Integral Aspects The analytical tests show that ESPROSE.m can perform well at the extreme limit assumed  ;

in the Board Hall model. These calculations are interesting as they show explicitly the effect of the fragmentation rate and entralament factor on the propagation characteristics. Figure 6 is interesting in that it shows dispersive-like behaviour but if the grid is as described earlier these are real rather than numerical. Could the authors commer'.?

The confirmation that the 2D and 3D models give similar results is thorough and convincing.

I agree with the authors that the 1(ROTOS tests are too poorly characterized for real validation studies and therefore I do not think this section is central to the validation case.

The point about melt freezing is very interesting and the fact that the code under-predicts freezing times is important. This effect will be compounded by the fact that that the melt is assumed to be at a uniform temperature, whereas an outer shell will freeze first. Surface freezing provides the most convincing hypothesis (to me) of the non-explosive behaviour of l'0 in 2 l{ROTOS.

E-58 h

e

p Numerical Aspects Q 4.5 I agree with the conclusions drawn.The presented calculations clearly show that the authors are aware of the need for adequate spatial and temporal resolution. In addition, the results show a good compromise between difusive and dispersive errors. 4 4.6 Concluding Remarks This is a very important section and I believe the authors have judged the current situation very well. I agree entirely with the conclusions they have drawn from the very comprehensive sets of calculations performed to dste. There is a clear need for the "-h temperature SIGMA data and I am aware that plans to obtain this are well advanced I personally doubt that it will every be possible to characterize the KROTOS experiments much better and my experience with the MIXA tests tells me that there will always be sometMng left to be measured. Therefore I agree that this is s lower priority. The comments on secondary pressure waves are interesting and clearly of a very fundamental nature. I do not believe that such eEcets could be addressed easily within the continuum model but I would certainly encourage their investigation.

Finally, I agree com pletely with the closing paragraph: moving to large-scale, multi-dimensional

_ [m') experiments will only add confusion.

%J 4.7 Appendix A 1 have no specific comments here. I am generally famliiar with the modelling approach taken and I believe appropriate modelling choices have .been made from the available database of constitutive laws, it should be recognized that it is in the formulation stage that the ESPROSE.m model diRers fundamentally from othersin thatit is 3D and uses the microin-teraction concept to allow for thermal disequilibrium within the coolant.

. 4.8 Appendix B This section contains a description of the CH AT code used to provide analytical solutions for code comparisons. The model is formulated for the case of homogeneous flow of liquid and coolant (no slip but diRerent tempeintures). Thus the system has only real characteristics and therefore can be solved in an elegant and accurate manner. It provides an excellent means of testing ESPROSE.m.

p d E-59 9

4.9 Appendix C O

This appendix is a reprint of a conference paper which describes the mictointeraction data 5

and its implementation into ESPROSE.m. I am familiar with this work (from t ie paper and visiting the facility) and believe it to be both unique and of a high quality. Whilst at present results from low temperature melts have to be extrapolated to the reactor case, plans are well advanced to produce the required data.

4.10 Appendix D This appendix also contains a reprint of a conference paper which discusses the manner in which the *: cal world' differs from the Board-Hall model. It is very interesting as it shows how the inclusion of microinteraction physics produces propagation behaviour which is very different from the Board Hall model and other propagation models which do not a!!ow for micro-mixing. Essentially, it allows propagation in systems which are melt lean' because the energy from the melt is transferred to only a fraction of the water present. It provides an interesting perspective on which to end the ESPROSE.m validation report and clearly illustrates what a significant advance the mictointeraction concept has been in propagation modelling.

References

[1] Baines. M. (104). Preliminary measurements of steam explosion work yields in a con-strained system. Inst. Chem. Eng. Symp. Strirs, 86. 97-10.9

[2] Fletcher. D. F. (1991). An improved mathematical model of melt / water detonations-II. A study of escalation. Int. J llent Man Transfer,34,2449-2459.

10

m Fcracnung2z:ntrurn ter:2runa I

Technik und Umwelt

~

Institut for Neutronenphysik und Reaktortechnik t n.e emr. or.-ins. or. Ac. o. x.ai.e O; ce u.a-. 4 =.a.o noei-Dr. L. W. Deitrich Reactor Engineering Division h,'N,],'e'"of,'"

Argonne National Laboratory T.w.n o7247 s s2 2443 9700 Soth Cass Avenue T.i.e so7247tsa 3824 or 4874 Argonne, Illinois 60439 ihm um.aung

. USA. .

'i RECEIVED REACTOR ENGINEERING DIVIOL ,

i

-DIRECTOR'S OFFICE-1 Nov 251996 1 Review of report DOE /ID-10541

• " _~ ~k 9, __

b k ,,j, letFORMATTON: ._

4

.i._.......... - - -

Dear Dr. Deitrich,

please find enclosed a signed copy of the final version of my review of tr 3 report.

Sincerely

- Y* & .

Dr.H.Jacobs

" *"*" **"""*""*""""*""~#

' M" N'

. =- = ._"W_E-69 _ _ _ . _ . = = - = = . ,

T.='4#lJ* "'

F5_"= , 54=r: -  :" Tem L"::%i;!*M' M .,w""'" '::*#.Td.u= o==> L' %2 0 *= % T = sa

H.Jacobs Karlsruhe,20 November 1996 O

Review of Report DOEllD 10541: Lower Head Integrity Under in vessel Steam Explosion Loads by T. G. Theofanous et al.

1. Introductory remark in order to put my comments to follow into the right perspective, I must state first of all that I fully agree with the general approach to the problem taken by the au-thors,i.e. the ROAAM. To what extent probabilities are used within this approach may depend on the purpose end the problem of the study. However, dividing the problem into its physical aspects, treating them in separate parts of the study that can be scrutinized by other experts and linking them in a well defined and verifl.

able way defines a clear path towards the resolution of the full problem.

Similarly I fully support the basic approach taken to treat the steam explosion problem. The material presented is based on and incorporates a lot of pioneerinp and exemplary work in this field. I do not want to shed any doubt on that. The only question I'm discussing Is: Is the state of development sufficient to finally answer the question under discussion. This forces me to elaborate on potential weak points in the argumentation, if a technical field isn't developed sufficiently, even a ' peer review' cannot finally ensure the correctness of an evaluation.

Quite obviously, steam explosions are not phenomena that are well understood in the scientific sense, especially if we are concerned with such large scale events as are discussed in connection with reactor safety analysis. Unfortunately, such events lie far outside the parameter range that can easily be studied exper-Imentally. This is true of the initial temperature and the composition of the melt as well as the masses involved (as mentioned above).' This dilemma forces us to largely rely on codes for extrapolating from the accessib8e parameter range to that of the envisaged accident situations, ideaHy this extrapolation requires full knowledge and oppropriate modeling of at relevant phenomena. Here again we are confronted with gaps, the relevance of which is difficult to judge. The concept of ' fitness for purpose' may be helpful in areas in wh'th the consequences of ne-glecting something can be estimated. But how about problems which have not yet' been identified or the importance of which has not yet been perceived? In the present state of knowledge bad surprises cannot be excluded. The (only?) way to deal with this difficuty is to account for all (known) possible traps in the analysis (take a conservative approach) and to require a large safety factor. To some ex-tent, this principle is followed in the study discussed here. But in my judgement not to a sufficient extent.

From the point of view of quality assurance, a peer review like this one can be-come fully effective only if at least the background material was published s!nce quite some time so that a thorough discussion of it has been possible among the experts, in the present case an important part of the background material was delivered very late during the review ptocess. This reduces the relevance of the present review process.

E 70 1

i l

2. Scope of this review This review is concerned with the steam explosion aspects of the study. The contribution of this pad of the study to the positive final conclusion,l.e. interacting masses that are insignificant from an overall energetic standpoint and even local loads that lead to elastic strain only, can be attributed to small pouring rates, a strong volding of the premixing zone and early explosions. The first of these are to some extent a consequence of the melt water mixing scenarlos chosen and al-though core melt down is not my proper field of experience I must make a few comments on this because the way in which melt and water are brought into con-tact is basic for the subsequent events. The possibility of a small steam explosion inducing a larger one is neglected altogether. The second point, i.e. the pro.

posed strong volding of the zone in which corium melt and water are intermixed prior to an explosion (the mixing zone or premixture),is instrumentalin two ways:

In addition to tiie small pouring rates it reduces the interacting masses, At the same time, this volding seems to be one reason for the dying away of the ener-gefics of explosions with increasing time of triggering which is is the most con-vincing argument for considering early explosions. Of course, this finding,also depends on the third point,l.a. the way in which the steam explosion proper is modeled. The above three aspects,l.e. scenarlos and modeling of premixing and explosion are discusses one after the other below.

3. Technical evaluation 3.1 Melt relocation scenarios By the scenarios it is defined how the melt relocates into the lower plenum and this gives the rates at which corium is fed into the lower plenum. Therefore this (V) is an important aspect that must be scrutinized during the review step of ROAAM.

I am not really an expert in this field, mydelf, but I must raise the question whether it is really possible to exclude with sufficient certainty a downward relocation that could lead to much higher corium flow rates depending on the number of holes in the core support plate through which corium flows into the lower plenum. My doubts in this respect e me from the agreement of the experts in this field that the late phase of core melt down,i.e. the melt relocation phase,is not well understood and from the virtual aosence of mechanistic models for growth and especially ra-dial expansion of molten pools. The study that is under discussion here tries to bridge this gap using simple and clear estimates of conditions influencing the thermal stability of a metallic crust. But in these estimates, e.g. no consideration is given to the possible formation of cutectics which might drastically reduce the' molting temperature and thus crust stability. One might also speculate that some hot material could drop into the water remaining below the corium pool, thus de-creasing the time until it is aporated and thus the time of crust stability. In the present study the evaporation time 'happens to be just about equal the time it would take to melt through the reflector and core barrel.' Of course, there is in addition the thermal inertia of the core support plate. But as soon as its top falls dry,its surface temperature will increase and thus reduce the effect of radiative heat transfer.

Another possible uncedainly is the stability (leak tightriess) of a sideways (radial-ly) adsancing crust. This process might induce transverse forces on the support.

(~') ing stubs of fuel pins which these cannot withstand in their damaged condition, v So the crust could fall and the oxydic melt could flow freely towards the core E 71 2

_.___------m __-__ - ____ _ _ _ _ _ _ _ ..______..__-m_ _ - _ . . - _ _ _ _ _ . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ , _ _ _ _ __m_

suppori plate and possibly through it. (Table 4.1 Indicates that the ' cold trap' la not likely to stop flowing oxydic corium.) Here one may recall that processes of this nature occurred during the TMI.2 accident [1] although, in that case, the i whole melt pool was submerged. As witnessed by several tonnes of corium that '

solidified within the core support assembly, a large amount of corium has flown down through about 4 peripheral fuel elements around core position RO. Another downward relocation occurred at core position K8. The latter may have been i brought to a stop above the core support assembly. Out we do not know how and i by what margin. l Finally, the possibility of a large coherent steam explosion that is Induced by a smaller one (e.g. one of those considered)is completely left aside. Such event might proceed in different ways. The common starting point of these would be the machanical destruction of ties crust keeping the melt pool. This might be caused directly by the action of the pressure of the first steam exploslon or indirectly by the pressure of another mell cooknt interaction due to the addittion of some water into ths upper zone or on top of de melt pool. The induced steam explosion would then occur either within the core volume (if there were still water left) or in the lower plenum after the melt relt,ased from the broken melt pool has drained through the still open holes in Ine lower grid plate, it is sometimes argued that such melt couldn't encounter wate?In the lower plenum because that would have been driven away by the initial steam explosion. However, the first (weak) explo-slon might have caused essentially a sloshing movement of the water so that this could mix very effectively with the corium streams when returning, in this context one should also keep In mind that with a large molten corium mass available and melt water interactions occurring, large amounts of mechanical energy may be-come available. So it is often hard to argue that some process was unlikely.

3.2 Modeling of premixing Premixing is the process that is thought to be required to set the stage for any large scale coherent steam explosion, it is, at the same time, expected to inher-ently limit the masses participating in an explosion by the ' water depletion' effect, i.e. removal of liquid water from the premixture by large amounts of steam that are created due to fast heat transfer. As these processes are difficult to simulate directly in experiments, recourse is taken to numerical modeling with the code PM ALPHA.30. For the scenarios considered, this code predicts strong volding of the volumes accessed by melt, in combination with a cut off of propagation that is effective at high volding this gives a strong limitation of the melt masses that can interact. And this is the second pillar on which the final result of the study is resting.

While there are good arguments for the concept of ' water depletion' and also some experimental observations that appear to support the idea in principle, there remains the quer.'.lon whether the quantification given by PM ALPHA.3D is suffi-ciently rellable. The program predicts 'the major porilon of it [i.e. the fuel] being in a highly volded region (2 > 80%)' and also that the vold fraction ' gradient is veri steep', i.e. the vold fraction increases from values around 20 % to more than 80 % within a short distance Such behavior, however, was not seen in the pre-mixing experimerts that are being conducted at Forschungszentrum Karlsruhe in oroer to study the phenornenon and to collect data for code validation [2], [3],

[4]. It is too early to draw final conclusions from these experiments, but the void fractions in the surrounding of broken up ' fuel' appear to be smaller than ex-E-72 3

a fi pected. One may also draw attention to data reported of the KROTOS exper- l iments [5]. In these tous molten alumina was poured through an orifice with 3 .!

O cm diameter into a 10 cm wide tube filled with water, it mixed with the water and strong steam explosions occurred either spontaneously or following an external  :

{'

l trigger. The melt temperature was high, typically 2600 K, but the water was sub-cooled which, of course, tends to reduce voiding, in the KROTOS tests #28 and i

1. 2g the water was subcooled by 10 K and 80 K, respectively, in both cases the i steam volume fractions within the reaction tube were 4 % only. But as these are  !

mean values over the whole tube which may contain some regions occupied by i water only, it may be more relevant to point out that the steam volume was only about half the melt vo!ume. In test #30, subcooling was again 80 K but the melt l mass was larger and its breakup was more Intensive, in this case the steam vol.  !

ume fraction reached 23 % but this is again only 1.3 times the melt volume. So l

. we must check how well the above cited calculational results of PM ALPHA.30 are founded which imply steam volume fractions that are larger than the melt volume ,

fractions by well over an order of magnitude. j U

The original PM4PHA was one of the two pioneering codes that used three vo-  !

locity fields for describing the separate motions of melt, liquid water and steam  :

at the cost of adding considerable complexity to it.? already quite complicated  !

two field description of two phase flow. But this is the only way in which one can hope to develop a reasonable description of the phenomena during a steam ex-  !

plosion. The fairly standard multiphase equations used provide compilance with i the conservation equations only. All the controlling and very complicated physics in the three phase (and at least) three-component mixture must be described by constitutive relations. Here the difficulty arises that one of the main purposes of  !

l .

such codes is to extrapolate from the experiments that are possible in practice to

- the envisaged accident situation. This implies extrapolation from simulation me- .!

terials (sometimes even solid spheres) to the expected (but still quite uncedain)  ;

molten corium, from often quite low ' melt' temperatures to temperatures around 3000 K, and from the mostly very small scale of experiments to the reactor size.

There are a few experiments in which one or the other of the above Inlllal condi- i

~

tions is not as bad as indicated here but as the experimental difficulties, grow enormously as the expected accident conditions are approached, the exper-

<- Imental information on the inillal conditions and details of the processes is often poor in these cases so that a successful comparison of calculational results with  !

Integral experimental results doesn't necessarily Indicate correctness of the the-  :

oretical model, indeed, one can expect a code to perform the required extrapo- j lations only, if all relevant mechanisms are modeled mechanistically and with, sufilclent accuracy. However, the constitutive relations used in PM ALPHA are '

often heuristic, sometimes parametrical. The latter is described in the report for - i the melt breakup model but is true as well for one formulation of the evaporation  !

rate. The other formulation looks more physical but still does not allow for the  ;

possibility that evaporation and condensation occur concurrently in the same In-tegration volume (calculational mesh) due to limited subcooling of the water and -

Intensive local-(radiant) heat flux to the vapor / liquid Interface where the melt ,

drops are covered by a thin vapor film only, as e.g. on those paris of their sur-faces that are oriented towards the direction of motion. So any extrapolation to ,

accident conditions must be affileted with large uncerialntles.

Validation of the original PM ALPHA code by comparison'with experiments was  !

first described in Reference [6].which is also reproduced as Appendix B in the O

r special verification report [7]. An appeal of the general agreement reached may -

E-73 l

4 w- g-.m-uiv-MPvy+'d--'-V7 -yWy,, y-i pqe g y%e -y* v W y y yup ggN ypog- p 9+e-%r+.p-g-gp+ni.yp---g---:g9--wev-p*- -rw-p-iyy-p.e .We yg-- 4m 9 y- y e :-ry9pg*9 9 gpc,g 9 g w g, gap 9,g .99,inig- g 9,9---t-es6

h l be obtained from the data on the leading edge advancement. With cold spheres this agreement is mostly reasonable. With sphere temperatures of about 1600 K the data are reproduced within about a factor 2. In the ' production runs' of the present study the inillal temperature of the melt will have been beyond 2900 K so that the uncertaintles will certainly have increased quite considerably.

Here we are mainly interested in the high vold fractions that have been measured and predicted during the verification process. The data given in [6] have been obtained with the MAGICO experiment and have been described as highly rele-vant ('the measurement not only provides insight into premixing, but represents probably the most important test for computer codes'). Hence our expectation to find high local vold fractions in our own experiments. However, the local vold data presented in [6] have been measured in a position or better line or 'small region' (of unknown size) 15 cm below the Initial water level. This depth is only two thirds of the equivalent diameter of the pour. We may guess that the meas-uring volume was centered with respect to the particle jet (the pour). How its width compares to the width of the pour is not known. The measurement was performed at 0.35 sec, i.e. just after the end of (or behind) the pour, probably in order to avoid the presence of many spheres at the level of the measurement.

These circumstances appear to have produced the observed high void fractions possibly without too much contribution of steaming, it is our observation from the QUEOS. experiments [3] [4] in which streams of spheres are poured into a water pool in a similar way, that the particle cloud is always followed by a gas filled chimney - with cold spheres as well as with hot spheres. This is largely a conse-quence of the momentum transfer between the particles and the water while ther-mal effects are of secondary importance - they essentially influence the way in which the gas chimney is closed again. That this is also true in the MAGICO ex-periments is clearly shown by Figures 14 and 15 in Reference [6] which illustrate a ' cold' run. This means that the reported high vold fractions have little to do with the so called ' water depletion

• cffect and there is no experimental support for the high vold fractions calculated in the ' production' runs at positions far away from tne melt entrance. One. might add that corresponding to our observations in the QUEOS experiments, thermal effects just start to be detectable in an overall sense (beyond local effects around each individual sphere) at sphere temperatures as low as 1600 K. Even at the much higher temperatures _beyond 2300 K that have been reached in QUEOS, no high vold fractions could be observed outside the Initial gas chimney produced by the entering clouds of spheres (essentially by momentum transfer).

In the main body of the verification report [7] global estimates of the water con-tent within the mixing zone in QUEOS are used for further checking PM ALPHA.

Unfortunately this type of data is hardly suited for a quantitative comparison with code calculations. The difficulty is that the result very much depends on the choice of the outer radius of this zone because, due to the weighing with the ra-dius squared, it is this region that dominates the Integration over the total volume, in the experiment this difficulty can be overcome to some extent by preclsely de-termining the shape of the mixing zone from high-quality photographs - at least to the extent that a quclitative result can be obtained. However,in code calculations, the calculational mesh is not able to sufficiently resolve this outer boundary. So, what is given in [7]is the 'PM ALPHA result for the central region of the mixture, containing the main portion of the particle cloud.' As a consequence, the calcu-lated value is somewhat ambiguous and Figure 13 in Chapter 2 of Reference [7]

unavoidably compares quantitles with different definitions.

E-74 h

5

)

It remains that the code in this case predicts low volding (in contrast with tho production runs). But here the code appears to have gone to the other extreme h

V due to its inability to describe evaporation in the presence of subccoled water which even leads to the reported underestimation of evaporation (steam flow) rate and pressure rise. To explain these discrepancies by possible liquid superheat of the water in the experiment is probably inappropriate in the presence of large free surfaces.

Another uncertainty of the calculational results is due to modeling the corium breakup. The surface of a certain amount of material varies linearly with the (In-verse of the) parilele radius. Therefore modeling the corium as Individual drop-lets with 2 cm diameter from the very beginning gives it already a quarter of the surface that it would have with drop diameters of 0.5 cm whlch can certainly be considered as well prefragmented (broken up). In the calculations presented, this Inlllal diameter is combined with an entrance volume fraction of 25 % only so that there is an intensive thermal Interaction from the very beginning. However, in the PREMIX experiments being performed at Forschungszentrum Karlsruhe [2], we have observed that a melt let can penetrale to quite some depth into saturated water (e.g. 0.5 m for a jet diameter of about 4 cm) before it starts to break up and to Interact more violently (s*Ill not explosively). In these cases the melt is molten alumina at about 2600 K the density of which is only about one third of that of corium. So this behavior is even more probable (should be more pronounced) with corium. Such dynamic breakup process with virtually no breakup in the be-ginning that allow the melt to penetrate deeply into the water followed by more rapid fragmentation that breaks the melt into medium sized drops (which might be the most dangerous configuration) cannot be bounded by the parametric breakup model that was employed. Such bounding would require to model as (gj well the entrance of coherent melt (melt being the continuous phase) that is not premixed with water artificially (by assumption) from the very beginning. In this context it is also important to note that breaking the melt into very small droplets (e.g. 0.2 cm) may be very optimistic because these small drops produce a lot of l vapor, i.e. high volding, and may already stari to freeze so that they can no longer participate in an explosive interaction. The importance of freezing for the benign explosion results reported is not discussed.

3.3 Modeling of explosions The most important finding of the calculations in this area is the cutoff that occurs at higher vold fractions. However, the model used to describe explosive Inter-l actions - the microinteraction model- has been developed on the basis of exper-l imental observations in a situation with viriually zero volding. The parameters of I the model have been fixed using these experiments and it has been shown that the model can be made to give results looking reasonable (by proper parameter choices) by simulating a KROTOS experiment in which the local vold fraction was assumed to be between 25 and 40 E lt has been the declared purpose of the microinteraction model to explain the occurrence of strong pressure increases in the presence of large amounts of water (Iow fuel to water mass ratio). And as such it is highly interesting from a eclentific point of view and may be very rele-vant in this special situation. But one cannot expect this same model (with the same parameter settings) to work properly in a completely different situation in which there is very little water present. The failure of this special interaction mo-A del to predict strong steam explosions under condition: for which it wasn't de-() signed does not necessarily say anything about the occurrence of steam explo.

E-75 6

slons in situations as suggested by the premixing calculations should these ever occur. Especially in the case of larger melt masses (and possibly smaller overall  ;

void fractions) the lower plenum of a pressurized water reactor might provide j enough external confinement for completely different interaction mechanisms to l become effective. These mechanisms may need more time for their development but might in the end arrive at similarly effective interactions. An important ex. l ample of mechanisms that may contribute to such alternate types of interactions l are the thermal tragmentation mechanisms that may not need much water and are i completely left aride in the prment study. This might explain why the most effi- l cient explosions are obtained very early (prior to 0.12 sec) followed by much less l efficient interactions at later times in all cases with a finite breakup parameter.

The picture is less clear in the cases in which additional breakup was assumed not to occur. As outlined in the previous section these might be the most Inter-stM cases in this study. Here no clear maximum of explosivity has been found an,Jng the cases considered and it is argued that 'slightly broken up premixtures .

remain very benI D n.' However, Table 0.1 shows that in the case C2 nb the maxl. l mum peak local Impulse is 30 kPa.s which may already be viewed as a low to ink termediate value and that it occurs at the le trigger time considered,l.e.1.0 sec.

l Nothing in the results presented supports the idea that the value might not be larger (and maybe important) at even larger triggering times.

There is a further and independent argument for early triggering. It states that early triggering is due to the interaction of melt (jets) with structures. This widely used contention, however, dou not egree with the observations from the PREMIX experiments at Forschungszent.um Karlsruhe. We have now performed 11 such tests and in 4 of these the melt was forced to Interact with structures (vertical 'let' .

on horizontal plate - In one case even equipped with compartments). Only one  !

of these tests (the last one performed on 21' August 1996) lead to a violent thermal ,

interaction (a weak steam explosion) about 0.8 see (almost a full second!) after melt structure contact [8). One may also make reference to the KROTOS tests,in which the otherwise very explosive alumina melt settled at the bottom of the re-action vessel copying its shape when solidifying, in cases in which the water was saturateu and no external trigger Was applied [5]. So, mell-structure interaction does not necessarily provide early tiggering.

4. Summary The affirmative final result of the study follows from three findings: low corium-water mixing rates, very high void fractions in the premixture, and, partly de -

pending on that, effective explosions being possible only during a subsecond pe-riod at the beginning of premixing. I have serious doubts about all three of these.

With respect to the melt relocation scenarios I doubt that the present state of knowledge allows to definitely exclude downward relocation paths that could lead to much larger relocation rates. Not really being an expert in this field I must '

leave the judgement to those experts, provided they can positively defeat my ar-guments, in addition, processes that are induced by a first (weak) steam explo-slon might lead to a more effective melt water mixing and thus to a larger steam explosion. With respect to premixing, the very high void fractions predicted by the code PM ALPHA even outside the gas channel that immediately follows a mass plunging into water don't seem to be supported by experimental evidence.

The code itself is not provided with sufficiently mechanistic models and is not sufficiently validated to support the high vold fractions by itself. With respect to E 76 h

7

i the explosions, the failure of the code ESPROSE.m. l.a. the peculiar Interaction model in it (the microinteraction model), to predict efficient explosions in highly

(~'T volded premixtures, doesn't prove that such explosions were not possible on the V base of different Interaction mechanisms, even if highly volded states would oc-cur.

Literature .

[1] J. M. Broughton, P'ul Kuan, D. A. Petti, and E. L. Tolman, A scenarlo of the Three Mlle Island Unit 2 accident, Nuclear Technology 87 (1989) 34 - 53 l

[2] F. Huber A. Kaiser, M. Steinbrock, and H. Will, PREMIX, Documentation of the Results of Experiments PM01 to PM06, Forschungszentrum Karlsruhe Report, FZKA 5756 (March 1996)

[3] L. Meyer and G. Schumacher, QUEOS, a Simulation Experiment of the Pre- '

mixing Phase of a Steam Explosion with Hot Spheres in Water, Base case Experiments. Forschungszentrum Karlsruhe Repor1, FZKA 5612 (April 1996)

[4] L. Meyer, The Interaction of a falling mass of hot spheres with water,1996 ASME/AIChe/ANS National Heat Transfer Conference. Houston, TX, August 3-6, 1996: ANS Proceedings, HTC-Vol. 9, pp.105-114

[5] H. Hohmann, D. Magallon, H. Schins and A Yerkess, FCI experiments in the aluminum oxide / water system, Proc. CSNI Specialist Meeting on Fuel-Coo-lant Interactions, Santa Barbara, CA, January 5-8,1993, U.S. Nuclear Regu.

l latory Cornmission Report NUREG/CP 0127, NEA/CSNI/R(93)8 (March 1994)

I pp.193 201 -

[6] S. Angelini, T. G. Theofanous, and W. W. Yuen, The mixing of particle clouds plunging into water, Proc. 7th Int. Mig on Nuclear Reactor Thermal Hydrau.

lics NURETH 7, Saratoga Springs, NY, September 10 15, 1995, NUREG/CP 0142, Vol. 3, pp.1754 - 1778

[7] T. G. Theofanous. W. W. Yuen, and S. Angelini, Premixing of Steam Explo-sions: PM ALPHA Verification Studies, Report DOE /ID-10504 (September l

1996)

[8] H. Will, private communication (to be presented at the OECD/NEA/CSNI Spec. Mtg on Fuel Coolant Interactions, Tokal, Japan,19 21 May 1997) 1

. (4-Author's address: Dr. Helmut Jacobs FZK-INR Postfach 3640 Phone: +49 7247 82 2443 D 76021 Karlsruhe Fax: +49 7247-82 3824 Germany E mell: helmut.jacobs@inr.fzk.de O

Q E 77 8

TECHNISCHE UNIVER$liAT MONCHEN LEHRSTUHL A FOR THERMODYNAMIK Prof. Dr 4r.g. Dr.4ng. E.h. F. Moyinger n,enadH A . T. din;xh. un new mond.n . so290 MA s0333 w =Jin O

Artiswone 21 n IMM 2105 3435 Diroctor of Reactor Engineering Division im strab Dr. L. W. Deitrich fuIMM 21os 34si Argonne National Laboratory 4_ ,,,,,, %.4 9700 South Cass Avenue Argonne, Illinois 60439 Neue Telefonnummer U.S.A.

(r.'89) 289-16215 (089) 289-16218 Fax k Z.;d.n k> Unw Zeiden Unw h M/ba 03.12.1996 Review on the report DOE /ID 10541

" Lower Heed Integrity under in Vessel Steam Explosion Loads"

Dear Dr. Deltrich,

Please find enclosed my review comments on the report DOE /ID-10541, which is titled

" Lower Head Integrity under.In Vessel Steam Explosion Loads".

Sincerely yours,

~ ~~ .

~

s k  % N~l Fi E C E I V 5 D i REACTOR ENGINEF. RING O!W.':":1

-D! RECTO?.'3 OTF.CE- ,

Prof. Dr. Ing. Dr. Ing.E.h. F. Mayinger i DEC 10 096 .

t A CTi0th __ .. , . ,,

law'tAfCN._ L ^Jn* 4 N >2J 7d5' .

s

1. % 7 _ _ , L .~.'.L.J- - -"

E-78 1

1

q t

TECHNISCHE UN!VElt$liAT MONcHEN l LEHRSTUHL A FOR THERMODYNAMIK f Prof. DrAng. DrAng. E.h. F. Moylae*r i

Review  !

i on the report DOE /lD 10541 f

Lower Head integrity under in Vessel Steam Explosion Loads l I

i

- - - Not being an expert in structural mechanics, I shall concentrate my review on the thermo-  ;

fluiddynamic part of the report, trying to give an overall assessment.

For my rev6ew, I also took into account the report DOE /ID 10503 ' Propagation of Stcom Explosions: ESPROSE.m Verification Studiss', a paper by S. Angelini u.a. on the Mixing of Particle Clouds Plunging into Water /1/ and another paper by Chen u.a. on the Constitutive  !

Description of the Microinteractions Concept in Steam Explosions /2/. l 4 r

.L A 9hle m There are many papers in the intomational literature dealing with the phenomena and the 1 effects of steam explosions. They differ widely in their statement on explosion loads  :

depending on assumptions or predictions for premixing, heat transport between molten fuel and conversion of thermal energy into mechanical loads. Experiments were made with various melts, representing a variety of boundary conditions (fror.1 one dimensional to mul-tidimensional) and a wide range of scale, e

The report under discussion here does deliberately not make the hopeless attempt to find l an agreement or en average between the wide spreading results of the literature. It further-more is based on carefully planed experiments, performed by some of the authors and on constitutive descriptions of phenomens, involved in steam explosion processes.

Object of the study is the advanced pressurised water. reactor AP600 or respectively the integrity of its pressure vessel against hypothet! cal loads of steam explosions.  ;

o Entering the jungle of phenomena and effects connected with and resulting from steam explosions with the aim to come to a quantitative and physically reasonable result with respect to the mechanical behaviour of a pressure vessel is a task, which cannot be ful- 3 filled in a complete, best estimate way on the basis of today's overall knowledge. This is the case in spite of the fact, that numerous research work has been performed world wide  !

and that the authors of the report, being under discussion here, made excellent contribu-tions, analysing steam explosion phenomena and effects in a theoretical and in an ex- i perimental way. There are many intangibles in steam explosion processes. Being forced to l E-79 {

1  ;

t

dernonstrate the safety margins of a pressure vessel against stearn explosion loads in a way, which is resistant against critical questions, it is quite obvious to apply conservative assumptions.

The design of the AP600

• invites

• such conservative assumptions, because, besides the low power density, the core is not only surrounded by a pressure vessel with a rather thick wall, but also by a stainless steel reflector inside the core barrel. So the AP600 design can

" tolerate" conservative assumptions. By doing this and regarding the results, one has to be very careful with any attempts to transfer the data, obtained for the AP600, to other pres-surised water reactors. Conservatisms, assumed when calculating the thermo and fluld-dynamic situations during steam explostons, could lead to predictions with respect to pres-sure vessel failures, which are far beyond the physical reality under such an hypothetical accident. Therefore, inspite of the fine work presented in the seport DOE /ID 10541, there is still a lot to do to obtain a still more realistic basis for safety analysis and realistic pre- ~

dictions. However, we must also be aware of the fact, that there always will remain many Intangibles within the scenarios of hypothetical severe accidents.

2 Melt relocation characteristics Melt relocation characteristics are influenced by the heating up of the uncovered core, the transition to a molten pool, the availability or non availability of downward relocation paths and several melt release conditions. The authors very carefully analysed all processes, preceding or being involved in melt relocation, including blockage coolabilities and the resistance of the reflector and the core barrel against melt through. The conclusions, drawn from the calculations and physical considerations, are convincing. The two main conclusions, namely that the failure itself can be expected, that it will be local azimuthally and very near to the top of the oxidic pool and that the release will occur within a time period, which is within the coolability of the lower blockage, are prosented in chapter 4 of the report (see page 4 25) and give the good feeling, that the maximum amount of melt, which can interact with the water in the lower plenum, forming a steam explosion, is limited and by this also the energy release and the mechanical load onto the pressure vessel wall would be within a reasonable frame. So the limitation of the energy scenario, by carefully studying melt relocation characteristics, is a very important and very commendable contribution of this report to the state of art in steam explosion analysis.

A further, very important result in this chapter is, that 're flood scenarios" need no further consideration from a steam explosion standpoint (lower head integrity). This conclusion should and could have consequences for future planning of accident management activities for existing pressurised water reactors, also. It means, that any effort should be under.

taken to add water again into the pressure vessel after a beginning core degradation, be-cause it would be of advantage for preventing a further escalation of a sesere accident.

E-80 2

i i

2. Quantificatlan of oramixturma  !

O The authors of the report came to the result, that for the APS00 the amount of molt, pour.

Ing into the lower plenum through the downcomer, would be in the order of a few hundred  !

kg/s. Based on Als information, they determined the range of premixtures of molt, water {

and steam and their distribution on the way to the bottom of the vessel. Their calculations +

are based on fundamental aspects of the premixing phase, which a part of the authors I studied seriously in experiments (the MAGICO 2000), involving well-characterised particle ,

clouds mixing with water /1/. In these experiments, they performed detailed measurements  ;

on external and internal characteristics of the mixing zones. Mixing in saturated and in sub-  !

cooled water was studied. The results of these measurements found entrance into the i PM ALPHA code, which they at first used for interpreting the experimente,I resu'ta and  !

which is the basis for the analysis c8 quantifying premixtures during a hypothetical steam [

explosion econerlo in an AP600. Interesting phenomena tiny found were the formation of densely packed regions and of instabilities at the penetrating front (isothermal conditions) and local voiding in the mixing zone, as well as global voiding through the level swell (hot  ;

pours),

it should be mentioned here, that the original 2D PM ALPHA code was extended to a >

three dimensional version called PM ALPHA.3D version. The results, predicted for the <

AP600, showed, that premixing mainly takes pisce in the downcomer and at its lower end to the lower plenum. The average mixture zone and voidage zone is mostly shorter than 1  ;

metre and the average fuel length scale varies between a few millimetres and 2 cm. It takes a few tenth of a second until enough small molten particles are formed during the mixing process, i

, This gives hope, that a very first steam explosion will occur, before a larger amount of finely dispersed molten liquid is mixed with the water and that this very first steam explo-sion produces such a high voldage (steam) in the waterpool, that a further large steam ex.

+

plosion can be avoided. it is obvious, that the authors do not study this possibility, because it cannot be quantified, but it may be allowed to mention it in this review. Roughly speak-ing, one could perhaps say, that early, small steam explosions are the best guarantors, that large dangerous steam explosions probably won't occur in case of mixing hot melt with water. ,

l Another fact, which limits the momentum of a steam explosion, is the high voidage in the mixing zone, extending over a large part of it. This voidage has a strong damping effect on the migration of pressure pulses, because it offers a compressible volume.

The mixing deliberations and calculations, presented in the report, are physically well  :

based and deserve a high grade of credibility. . '

4. Quantification of evnlonlon loads f

There are two key phenomena influencing loads of steam explosion. These are E-81 ,

3

- the mixing of particle clouds plunging into water and

- the microinteraction between water and melt.

The first phenomenon was discussed in the chapter before. For describing the mictoin-O teractions between melt and water, the authors followed two ways. For describing the microinteraction and for simulating the propagation of steam explosions, they used the computer code ESPROSE.m. This code is based on a series of experiments the second parailal way which were performed in the so called SIGMA 2000 facility /2/. Originally the formulations for the mictointeraction were based on the astumption, that the rate of coolant nilxlng between debris and water is proportional to the melt fragmentation rate.

This is a reasonable assumption and by this it was possible to produce consistent com-parisons from available experiments for a wide range of steam explosion loads, starting from week propagations to supercritical detonations. The first formulations were mainly based on experimental results, obtained in the KROTOS facility. This first formulation was done for two dimensional geometries and could especially also demonstrato the mitigating effect of " venting", due to wave reflection at a free liquid surface. Supercritical detona-tions were observed in the KROTOS facility with aluminium oxide melt only, pouring at very high temperatures into water, in a next step, the constitutive equations wrJe assessed by using experimental results, ob-tained in the above mentioned SIGMA 2000 facility. These experiments were carried out with molten tin drops, having temperatures up to 18000 C, impinging into water. Of course one can argue, that there are scaling effects, if one wants to draw conclusions from the measured and evaluated data, gained in this small experimental set up, to the steam explosion loads to be expected during a severe accident in an AP600 reactor. Ac-cording to the reviewer's opinion, these scaling problems however are mainly with the mixing of particle clouds, plunging into water, a problem which was discussed in the chap-ter before and which was solved by the authors with the help of the computer code PM-ALPHA.3D.

The SIGMA 2000 facility was experimentally very well equipped and special measuring techniques, like radiography, gave very good quantitative information about the fragmenta-tion of the drop mass and its distribution. The fragmentation, measured with X ray flash, was reproducible within less than 20%, which is a very good accuracy for such types of experiments, in addition the fragmented melt was collected after freezing and was sub-jected to slave analysis. Very fine fragmented particles were analysed via scanning electron microscope photographs. Generally speaking these experiments are a very reliable basis for assessing a computer code like ESPROSE.m 3D, according to the opinion of the reviewct, in the SIGMA 2000 facility, not only the fragmentation rate, but also the pressure signals of the steam explosions were recorded by using high speed pressure transducers. Due to the small scale of the facility, these pressure signals may be conservative when applied to a large scale geometry, like the downcomer or the lower plenum of the AP600, in a large volume, in which fragmentation of a hot melt starts, there are always voided areas, damp-ing pressure propagation.

F-82 4

._ _ __- - _ _ . ~

I ne venneedon of the EspMOSE.m code is very wen d>cumented k the report D0540- l

^

f 0003. This report e>cuments how the vedous eMeets k steem expenalen progress, ne i wave dynamics, explaston coupNng and kregraf behaview were essessed, ne report elemenstrates how the coale is handlkg pressure waves k shgle and k two-phase flWde 1 and this not only k e onennmenolonel, but k a two alimenelonel geometry. Special etten- l 6en was given to renection and trenemisalon behaview. The compedson between pre- i altcred elete (ESMr0St.m coalel and expedmonial results showed very good agreement for l e wiele vedery of thermo and tlWadynamic parameters. The local situations and the tem-  ;

poret behaview are wen prenVcted. So, the code is k a cenaVtion, that snows to pree6ct steem exploalon behavicw also beyond the experkwnteNy vedfled seen. I The extrapolation from the small acele to the large geometry of the reactor were done by using the baalc equations for wave dynamica in multiphase media and constitutivs laws for . .

mictointerat.tions. The lattet ones were refined via experiments in the SIGMA facility, elao. l The combined theorotmal and experimental efforts are a very good basis for predicting and  !

simulating large scale conditions, also.  ;

}

Finally one has to ask the question on " substance scaling" l.a., the applicability of the i data, measured with modelling melts to liquid corium. The experiments wars mainly per-  ;

formed with tin and with aluminium oxide. Especially aluminium oxide is very likely to l produce supercritical steam explosions when it is mixed with water. The authors of the j report DOE /ID 10541 write on page 21 (chapter 2 " problem definition and overall i approach"):

{

"Also, it is important to note, that within the limited experience with reactor fuel material (UO2 , ZrO 2), we have no evidence of explosions, but rather ex- '

tonalvely voided premixtures (Huhtiniemi et al., 1995), nor is it known j whether or under what conditions such premixtures can' be triggered to - '

explode". j With respect to "suustance scaling" the date, presented in the report DOE /ID 10541, on explosion loads, originating from steem explosions are on the safe side withcut any doubt, because a corium molt / water interaction will produce much softer pressure pulaea than ex- j l perienced in the experiments with aluminium oxide molt / water interactions.

f

5. Intaaration and annessment i i

in the chapter 7 " integration and assessment", there are two very important statements, 7 namely that - j

- from a more global perspective, the only way "to potentially produce a sig- l niltant structural challenge on the lower head, would be by  !

having a highly subcooled pool in it" and

- "even a postulated rapid reflood scenario could not produce the condition of- ,

concom...". E-83  !

2 l

5 q-

After depressurlaing the primary system, following an hypothetical, severs accident, there is always and everywhere saturated (not subcoolcd) water in the lows r plenum of the pres-sure vessel. This would be true not only for the AP600, but also for all other pressurised water reactors.

So as long as one can guarantee, that the cooling of the lower core support structure is good encugh to prevent it from failing and core melt flows from the side to the lower plenum, steem explosions, originating from it, st.ould not be a problem.

The second statement is as important as the first one, because it eliminates doubts, exist-ing up to now, whether it would be advisable to try to flood a degraded core again after a certain escalation of a severe accident. This point was briefly discussed already in a former chapter of this review. Therefore in future accident management planning, there should be given more effort to In-vessel cooling also after a partial core disintegration.

6. Conclusions i fully agree with the conclusions presented in chapter g of the DOE /ID 10541 report, to the statement of the authors, that 'because of the wide margins, due to these controlling physics, it has been possible to band uncertainties to a sufficient degree...", I would like to add, that these ' wide margins' are still on the conservative side and the mechanical loads onto the pressure vessel and its lower plenum would be lower in case of a hypotheti-cal severe accident, than predicted in the DOE /ID 10541 report.

Finally I would like to congratulate the authors to this fine work, attacking a very difficult but important problem and solving it to a great extend from an engineering point of view, but based on controlling physics and on reliable constitutivu laws for the fluiddynamics to be expected in steam explosion scenarlos.

. ] ^\

M0nchen, November 20th,1996 Prof. Dr. Ing. Dr. Ing.E.h. F. Mayinger 1 S. Angelini, T.G. Theofanous and W.W. Yuen, The Mixing of Particle clouds Plung.

ing into Water, NURETH 7, Saratoga Springs, NY, September 1015, 1995, NUREG/CP 0142 Vol. 3,17541778 2 X. Chen, W.W. Yuen and T.G, Theofanous, On the Constitutive Description of the Microinteractions Concept in Steam Explosions, Procendings NURETH.7, Saratoga Springs, NY, September 1015,1995, NUREG/CP 0142 E-84 h 6

~

l i

i Decernber 30,1996 -

l i

i L. W. Deitrich, Director  ;

Reactor Engineering Division l Argonne National Laboratory i 9700 South Cass Avenue l Argonne,IL 60439 l t

Dear Dr. Deitrich,

i I have enclosed my review of, ,

l

" Lower Head Integrity under In Vessel Steam Explosion Loads," >

by T. O, Theofanous, et. al., DOE /ID.10541,~ June,1996. l A copy already was sent to Prof. Theofanous to help in preparation of his responses to be  !

included in a forthcoming document, which will include e< aments from all the reviewers.  ;

Thank you for an opportunity to review this work. I think it is one of the most  !

significant pieces of research I have ever reviewed. It is of both current and long term importance to the nuclear industry. l O

Since I have spent my career in the nuclear energy business, I personally appreciate i

!i your long range viewpoint for energy needs, which is obvious from your support of this program. '

Yours truly,

$p&- (

FrederMk J. Moody 2265 Sunrise Drive.

San Jose, CA 95124

'(408) 377 7900 (H)  ;

. (408) 925 6414 (W) ,

a copy: - T. G. Theofanous

. . a._ - -- , , - - - , ,. . . . . - ... . .. -

" Lower IIcad Integrity under in Vessel Steam Explosion Loads" I

by T. G, Theofanous, et, al., DOE /ID 10541, June,1996 Reviewed by: F. L Moody 1

The purpose for reviewing the subject report, with several other companion documents, was to assess whether "in. vessel retention"is demonstrated to be an effectin l severe accident manactment concent for a reactor like the AP600.

I have reviewed the work, and conclude that in vessel reter. tion has been shown to be an effective severe accident management concent for reactors with cepJnetry. fluid quantities. event sequencing. and thermonhvsical oronenies similar to those certainine to the AP-600.

The documents provided for this review describe the steps taken to understand and predict the complex, multi faceted subject of steam explosions. Associated phenomena have been closely simulated by experiments, and predicted with deterministic theoretical formulations (causal relations) to a degree of accuracy that makes confident predictions possible for full size AP 600 systems. It appears that all controlling physical effects have been included, even without the need for a complete understanding of the exact tmiing and conditions necessary to trigger steam explosions. Already known or conservatively estimated ranges have been placed on parameter, timing, and scenario path uneenainties, and stillit has been shown that the expected range oflower head steam explosion pressure loac J do not intersect the vessel fragility curve.

I was asked specifically to review the material on steam explosion loads, as discussed in

" Propagation of Steam Exp!osions: Esprose.m Verification Studies" by T. O. Theofanous, W. W. Yuen, K. Freeman, & X. 'Chen, DOE /ID 10503, August 1996.

The documents provided for this review collectively lay an extensive foundation of information, which testifies to the technical stature, competence, thoroughness, and integrity of the investigators. Indeed, the overall work is monumentalin its scope and achievement, and it is communicated in a writing style which is one of the most scholarly to be found in reactor safety studies Doth the authors and sponsors should be commended for a carefully formulated investigative strategy (strong, in depth, well-blended steps) resulting in the highest value obtained for the time and resources spent.

Beyond steam explosions the progress and understanding achieved in this work are likely to exert a major beneficial influence, both methodological and technical, on other significant and complex thermal. hydraulic issues.

E-86 h

. I 2

m SUMhiARY ,

l

1. The RO AAM has shown that vesselloads, result;ng from a compt thensive I range of severe accident scenarios, melt conditions, relocation flow, timing of release from l the core region, and thermal hydraulic processes between the melt and surrour.dirs water, lead to the conclusion that vessel failure is " physically unreasonable" in an AP-600 type reactor. Parameters including pool geometry, melt release rate, shock explosive formation and propagation, and venting yield load distributions on the vessel wall which were compared with the fragility curve in order to arrive at this conclusion. It is my opinion that even though all the mechanisms contributing to steam explosions are not fully understood, results embrace the extent of refinements which could eventually be made by further experiments and theoretical model (causal relation) development.

2. I agree that it would be useful to obtain datr. from the QUEOS experim:nt for a fully saturated water system, sithough it would not change the conclusion that vessel failure in AP 600 type reactors is " physically unreasonable." The value in such a test is to fill in a parameter range to giv: a more complete data base, and permit the technoloby to be extended to non AP 600 type systems.

3. One potential benefit of the RO AAM procedure is that it conceivably could be used in reverse. Suppose it was concluded that a system failure probability was larger

/ than acceptable. The ROAAhi could be employed to display which parameter (s)

\ dominate the outcome, thus pointing the way for desigr. or procedural changes to reduce the failure probability.

~

4. How does the ROAAhi accommoaate different causal relations, such as Pht ALPHA and ESPP.OSE.m. at different stages in the methodology if they might be strongly coupled through common variables? That it, the behavior of two systems alone may be altogether different when they are coupled together (like two spring mass systems). The probability distributions of the parameter: involved may combine differently when the separate systems are strongly coupled, leading to different probability ranges on the variables which determine succea or failure of a system or process.

5. The source term for area production in Appendix A of DOF/lD 10503 is besed on the assumption of particle number density remaining constant, while their size changes. A bit more explanation or justification would help. Wouldn't it make more sense to predict interfacial area growth by the formation of more particles as the melt decelerates in water? Taylo: instability was employed to obtain the Bond number caiterion in interfacial area growth. Could that model be employed to obtain a fastest growing wave length and dreplet formation?

6. It appears that in the heat transfer predictions of Phi ALPHA in DOE /ID.

10504, flow regimes are identified by steady state correlations. Are these likely to be n nonrepresentative for such transient events as fragmentation, and not provide a

() conservative characterization of tl.e actust heat transfer?

E-87

3

7. Convective and radiative heat transfer from the fuel to the coolant is estimated in much detail, drawing from various experimental studies between coolant and heated solid surfaces is there a backup analysis to show that for the rapid heating associated with steam explosions, the heat transfer is not limited to how fast it can escape from the molten particles? Are there potential droplet sizet relative velocities, and fluid properties wh$re ir.ternal cor. duction (or convection) might limit the heat exchange rate?

STRATEGY The severe accident management strategy addressed involves the retention of core materialin the reactor vest.el following a postulated severe accident in a reactor like the AP 600 design Inability to cool the core leads to melting of core material by decay heat, and relocating it in stages to the reactor pressure vessel (RPV) lower plenum. Molten core debris, which may flow to the bottom of the lower plenum can melt through the RPV wall and undergo release to the containment. liowever, flooding the cavity to submerge the RPV bottom head is expected to be a means of arresting the downward relocation of molten core debris.

Even if downward relocation of molten debris is arrested, there is the possibility that some mass of debris could drop into water present in the lower head region, causing a steam explosion and further damage. Part of the overall study shows that failure of the bottom head by exceeding its structural integrity is " physically unreasonable" O

TiiE RISK ORIENTED ACCIDENT ANALYSIS METliODOLOGY (ROAAM)

A primitive method of handling uncertainties in power systems came in the early 1960's (Moody, F. J.. " Probability Theory and Reactor Core Design," GE Report #

GEAP 3819, US AEC Contract AT(04 3) 361, January,1962). One of the greater concerns for a nuclear core during normal operation was reaching the " burnout" condition, where a hot spot in the fuel could exceed design limits, and cause fuel damage.

The fuel temperature could be expressed as a function of several variables and parameters (causal relations), each with its own degree of uncertainty. If one chose the most pessimistic limit of each variable and parameter, the " burnout" limit could be exceeded.

The most optimistic limits the " burnout" limit would not be exceeded. It was suggested that probability methods could be applied to give a reasonable assessment of the likelihood of exceeding the " burnout" limit. Data from power plant operatin;, logs was gathered to obtain probability distributions for certain variables and parameters. Wherever data was not available," expert opinion" was solicited. The results were then combined by the method proposed in an ASME paper (Kline, S. J., and McClintock, F. A.," Describing Uncertainties in Single Sample Experiments," Mechanical Encineerinc. January,1957),

which resulted in the expected mean and standard deviation for the hot spot temperature.

Comparison with the established design limit showed that it was " physically unreasonable" to expect " burnout" in most cases.

E 88 h

4

!g )

V The ROAAM is an extensive, operational methodology which ;n more termed than any ofits primitive predecessors. It has the capacity for incorporating causal relations (describing equations relating the variables and parameters), based on well-understood physics for the applicable phenomena, with specified parameter uncertainties, scenario bifurcations, and even a diversity of expen opimon. The process leads to a rationally-based prediction of those properties which determine the success or failure of a system or process.

The structure of ROAAM embraces the current phenomenological state of the art, built in activation response of safety and control systems, man machine interfaces, and procedural understanding. As new information becomes available, the ROAAM can accommodate it. Where expert opinions may ha diverse, the ROAAM provides a means of focusing further research to narrow the disagreements. That is, when experts strongly disagree on the range of a parameter, the ROAAM can be employed as a tool to display the sensitivity, showing if the parameter dominates the outcome, or is only a minor percentage effect on the overall result.

One question about use of the ROAAM involves the caum! relations for various phenomena. If the parameters in a causal relation are independent, their probabilities can be combined in a certain way to obtain the expected mean and standard deviations of that function. If the parameters are not independent, the combination is more complicated.

O The question involves how the ROAAM accommodates the possibility that some C/ parameters appearing in more than one causal relation may not be independent. How would results from ROAAM compare with one deterministic mega-computation where all the parameters are treated by something like a monte carlo process to obtain the distribution of variables which determine success or failure of a system?

ROAAM APPLICATION 1 have seen the ROAAM work in two separate campaigns to close severe accident issues, namely the direct containment heating (DCH)'ssue for one series of PWR's, and the Mark I liner melt issue for one class of BWK cont ainment. It is appropriate that this methodology should be applied to reach a conclusion on the in vessel retention severe accident management concept.

Application to in vessel retention embraces possible scenarios, melt conditions, coolant states, structural properties, debris mixing with water, triggering, explosion wave dynamics, and lower head fragility. Parameter ranges are associated with the amount of panicipating substances, the timing of events, event paths, and state propenies of various subsystems. Several analytical tools, based on physical models, provide the causal relations employed, namely PM ALPHA for enveloping the effect of melt breakup in water ESPROSE.m for enveloping the effects of fragmentation and microinteractions on

,m steam explosions, and ABAQUS v5.5 for enveloping the lower head failure criteria. The

() computer programs used for causal relations to envelope important variables have been E-89

5 compared with other analyses and experimental data to a level where their predictive capability of the tested parameters does not introduce uncertainties which are significant h

enough to consider.

The following comments are offered to help substantiate my conclusion that in-vessel retention has been shown to be an effective severe accident management concept for systemslike the AP 600.

t/ILT INTRODUCTION AND FRAGMENTATION Harly predictive models provide core melt scenarios and relocation rates with and without refinod, which can arrest the melt progression. However, the melt state which may reach water in the RPV, and the subsequen: breakup and penetration largely determine the rate of heat transfer, steam formation rate, and possible shock pressure loads. A quantity of melt arriving at the water can undergo Taylor unstable breakup or droplet formation a; the leading edge and Helmholtz breakup or droplet stripping on those surfaces with parallel velocity components. The PM ALPHA model has been developed to incorporate the melt and coolant propenies, and provide an envelope for the expected range of momentum, heat transfer, and phase change interactions associated with breakup for premixing considerations.

Single particle and particle cluster experiments hase been employed to test predictive capabilities of particle motion and energy transfer dynamics in water (the MAGICO and QUEOS experiments). Particle cloud elongation, steaming, spreading, and mixing with surrounding water are captured by the PM ALPHA code, which is employed as a causal celation in the ROAAM Comparisons include particle cloud distonions associated with release door opening time, panicle, and void volume fraction contours.

Of panicuh , mterest is the pinching of the vapor volume b-hind moving panicles, caused by condensation for the partic.': introduction into subcooled water. Since the condensation acts to reduce .nechanical energy transfer, I agree that it would be useful to condact QUEOS experimer,ts in fully saturated water.

Oc.c of the mr :t imponant considerations in fragmentation is the formation of new melt heat transfer area. Appendix A in DOE /ID 10503 describes the " source term" for interfacial area production. Equation (3.69)is based on a change in size of particles fr the same particle number density. It seems thit before panicles have reached a stable size, they ivould undergo t) formation of new panic!es. This assumption needs more explanat ion.

STEAM EXPLOSION The 'nechanics of steam explosionnte described in DOE /ID 10503, detailing melt introduction to water, interfacial breakup .pd premixing of debris partir.les with water, the E-90

6 O' effect of voiding around the panicles nn heat transfer, the triggering of explosions, and propagation of pressure waves with reflutions frorn rigid mechanical and gas liquid interfaces. It was earlier found that 1.0 GJ of energy could fail the lower head. However, funher understaading has led to a teexamination of the mechanics of steam explosion force generation to determine a more reahstic criterion for lower head failure.

It was determined that the AP 600 could withstan:1500 bars of pressure for milliseconds without failure. Computations with the ESPROSE program displayed the difficulty in generating such pressure impulses with attenuating phenomena like voiding, which resists triggering, and pressure venting from the water surface. Extensive  ;

development of ESPROSE have been performed with both data from the SIGMA and KROTOS experimental facilities. Simpler analytical models have provided assurance that ESPROSE accommodates detonations, shock propagation, and reflection.

Significant effects embraced by ESPROSE result from the physics incorporated.

which are consistent with experiments. Calculatior,s show the strong attenuation of shock pressure loads with distance, and time by venting from the water free surface in the AP.

600 systems. It is also realized that venting may not significantly reduce loads if the water depth is high in the lower head. Strong evidence is supplied that ESPROSE incorporates the appropriate physics, and can be used with confidence to provide the causal relation for ,

enveloping the effect of trigger time on steam explosion severity.

The physical mechanisms considered by ESPROSE.m include shock pressure propagation from a trigger, which collapses voids, forces liquid onto the melt, producing fragmentation and microinteractions, escalated heat transfer, further steam formation, and rapid expansion (explosion). A statement on page A 18 of DOE /ID 10503 needs further clarification. Where the pressure increases rapidly ahead of an explosion front, why does the vapor become instantaneously subcooled? (if saturated steam is rapidly compressed,it would tend 'o follow an isentropic path off the vapor dome into the superheated region, not subcooled.) On the same page,it is stated that behind the explosion front where pressu're is decreasing, the liquid can become superheated. (if you decompress saturated water, the path drops into the steam dome.) It would make better sense to me (I can't speak for others) to note that the nonequilibrium states lag behind a steady state in the superheated or subcooled region.

LOWER HEAD RESPONSE Dynamic response of the lowei head is based on well established physics of shells, modeled by the ABAQUS program. Mechanical failure of a shell depends not only on the magnitude of an applied load, but also on the frequency content, It is stated that the shock pressure loads which lie in the steam explosion envelope have a short period relative to the e

structural response, so that the peak strain would be essentially independent of the f pressure pulse time profile.

t E-91

l l

7 ,

The report has provided some " screening fragility" curves which would be used to l determine if predicted steam explosion loads were of such a character that the failure l

criteria envelope and fragility curve need to be further blended to provide a failure likelihood. It was concluded from the range of pressure loads and the lower head fragility l i

curve, that for all relevant severe accident scenarios, melt conditions, and timing of release from the core region, with ensuing mixing and explosion wave dynamics, steam explosion induced lower head failure in an AP600 like reactor is " physically unreasonable."

i REVIFW OF STEAM EXPLOSION LOADS The verification of E5 PROSE.m, based on stepwise experimental measurements and comparison with simplified theoretical methods shows that reasonably conservative assessments of steam explosions are possible in the present version.

The discussions of DOE /ID 10503 provide foundational support of the physical modeling and numerical pracedures to predict steam explosion properties for given melt addition rates and states. The basic physics involve wave dynamics, including sound wave propagation and shock development and propagation in a water filled region. Two.

dimensional calculations performed by ESPROSE.m are compared with simplified computations using the method ofimages and solutions similar to classical waterhammer.

Some comparisons are included based on characteristic solutMs. The results form a strong basis far concluding that the code is producing reasonable predictions for the expected range ofinput parameters. Pressure propasstion speeds, attenuation from wave interaction at free surfaces, and wave amplification by reflection from rigid surfaces have all played a role in the verification.

Numerous two-dimensional ESPROSE calculational surfaces are compared with solutions obtained from the method ofimages, and found to be sufficiently similar, leading to the conclusion that basic physics of explosions are included in the model. Several geometric parameters were varied, as was the source velocity function. Good comparisons were consistently achieved.

The SIGMA tests involved a melt droplet which was triggered at a specific position, leading to local pressure traces. Comparison of the pressure traces with ESPROSE calculations showed reasonable tracking of pressure waves originating from the droplet region to the rigid end of the test section, and reflection back toward their origin. Additional evaluation with the method of characteristica were provided. The wave dynamics, indeed, appear to be properly described in ESPROSE.

One piece ofinformation lack pointed out in the report is that the data base needs expansion for microinteractions with reactor materials.

E 92 I

l

_ . _= _ _

8 d THE NEXT STEP 1 understand that a number of experts are providing reviews of the documents provided. Some may believe (as I do) that even without a enmplete understanding of all the phenomena, the remaining uncertainties, processed by the ROAAM, still permit a strong statement about failure likelihood being" physically unreasonable." Some experts may feel that the uncertainty of a given parameter should be broader. This is a simple exercise in ROAAM, which would then provide output with a range that accommodates the panictlar variable uncertainty. Other experts may wish to change the causal relations to reflect various " bottom up" or fine structure effects. This is always a possibility, but may be unnecessary, since the causal relations are based on macroscopic formulations of basic principles. Ifit were recommended that nonequilibrium models be employed for causal relations, we would be farther behind than using ROAAM in its present structure, because nonequilibrium models would have to be verified by experiments.

When strong disagreements have resulted in physical modeling, small working groups have been formed to reuh agreement on acceptable formulation, with appropriate modifications in ROAAM.

Finally, it is possible that some would disagree with the ROAAM structure itself, suggesting that it skews results, or simply blurs our ignoi.aice of phenomena. I would (3 argue strongly that the ROAAM blends (not blurs) uncertainties (not ignorance)in a way V that makes it possible to reach cor.clusions with a known level of confidence.

OVERALL CONCLUSION As a curious person who enjoys formulating better theoretical models, based on more complete experimental understanding, I recommend additional experiments (e.g.,

QUEOS experiments with fully : tturated water) to help close the few remaining gaps in our understaading of steam explasion phenomena.

However, I believe that the studies provided for this review give substantial,in.

depth evidence to help conclude that in vessel retention is supponable as a severe accident management strategy in AP 600 type reactors without additional work to close the irsue.

O b E-93

""PJf" KuNcLTEKNISKA HOGSKOI.AN Royallnuitute of Technology "F I997 02 26 Div. of Nuclear Power Safety Dr. Walter Deitrich Director, Reactor Engineering Prof. B.R. Sehgal Argonne NationalI aboratory Tel: 46-8 790 654i Argonne, !!. 60439 Fax: 46 8 790 7678 email: sehgal@ne.kth.se USA

Subject:

Review of the Reports DOE /ID 10541, DOE /ID 10503 and DOESD 10504

Dear Walt,

I have enclosed reviews of the three subject reports in the attachment, ne particulars of the three reports me:

1) T.O. Theoufanous et.al," Lower Head Integrity under In Vessel Steam Explosion Le ,ds",

DOE /ID 10541, ARSAP Program (June 1996)

2) T.O. Theofanous et.al, "Propogation of Steam Explosions: ESPROSE-m in Verification Studies" DOE /ID 10503, ARSAP Program (August 1996)

3) T.O. Theofanous et.al, " Premixing of Steam Explosions PM ALPHA Verification Studies",

DOE /ID 10504, ARSAP Program (September 1996)

I have structured the reviews in the following order:

a) Overall approach, b) PM ALPHA verification report DOE /lD 10504 c) ESPROSE M verification report DOE /ID 10503 d) Assessment of the lower head steam explosion loads DOE /ID 10541 e) Concluding remarks I believe the above order ibilows the scenario of a steam explosion, as well as considers the method verification before the method application.

Thanks for providing me the opportunity to review these reports. I enjoyed reading them. I am sorry, it took so long to accomplish these reviews. I hope that you and Prof. Theofanous are happy with them.

g

• 'mp ~"C g-C' ;0 With best regards*

P.EACTOP. Ei:GINEER'NG Divi;;cN

' ~ E'2'E CE-Sincerely yours,  :

hi MAR 101997 ,

Bal Raj Ichgal ' Ac m __ ,,, , , , , , , , _

E 94 ..--.~.

Professor p.rc, marc..: r.,, a CJi - g--

2

• 11 ' n s w n 100 44 $TOCKHOLM 5%tDEN -46 8 790 60 00 ,_ ,,

3

( Review of the Reports DOE /ID-10504, DOE /ID-10503 AND DOE /ID-10541 (June - September 1996) by Professor T. Theofamous and Co Workers I. Review of the Overall Approach This is the fourth time I have had the opportunity to review a body of work that Professor Theofanous and co workers have produced for the resolution of a specific safety issue, or a specific concern. I believe, this is the most complex of all the issues (or concerns) so far and I believe, Professors Theofanous Yuen and co-workers have done their f*mest work so far. This body of work is of greater, and of more lasting, value, than earlier efforts, since a major part of this work is the development and i,crification of the methodology to describe the steam explosion phenomena, and to predict the loads imposed by the postulated occurrence of a steam explosion. Thh methodology, and the codes developed, could be applied to other accident scenarios, than the one considered in the present application.

I believe, some comments are in order on the overall a 3 roach followed in these three reports, complemented, of course, with the ROAAM method, and the previous work that Professor Theofanous and his teams have performed,(e.g., for the Alpha mode failure of an L #R q containment during a severe accident).

y/

Professor Theofanous and co workers, with their accumulated experience m steam explosion modeling s 1 applications, have developed a very well focussed overall approach in the body of worx pr a ited in the three reports. it is clear that an in-house experimental program was structured to p: wie the key ebservations, for the ideas ~needed,:o no,n m the steam explosion modelini, , J.e point where some meaningful predicticns can be mede. The innovative experiments performed in the M AGICO facility provided the gerv.ane ideas on steam depletion, and on the difficulty of obtaining pre-rdxtures, which would lead to very large steam explosions. Likewise, the experiments performed on the SIGMA facility provided the basis for the micro luteractions concept for the steam explosion itself, i.e., the concept and treatment of the m fiuld. I believe, the experimental underpinning of the ideas and concepts employed, and the further verification of the methods used in the codes against the integral

! experiments, has provided great strength to the overall approach.

L The overall approach followed, in the application report, conforms to the ROAAM method and employs the PM. ALPHA and the ESPROSE-m methodology. The extremely high values for the fragility curve made the task much simpler than the earlier applications of the

! ROAAM methodology, but it is well justified and credible.

Perhaps, the two points of possible short coming in the overall approach, which have also i

been admitted by the authors, should be stated:

!k E 95

2 First, is the question of maturity. Clearly, there is not enough separate-effect and integral-effect data to provide sufTicient validation of the steam explosion methodolog) developed.

This methodology employ *, many many correlation and submodels, whose ledi vidual verification is a monumental task. Nevertheless, an experimental verifi.atbn r.4 atrix should be developed, with priorization ofimportant effects, and executed, to provide greater verification of the methodology, thereby providing it greater maturity.

Second, a mechanistic treatment of the initial phase of the steam explosion scenario, i.e., the break up of the melt jet, and its sequential fragmentation, has not been included in the methodology developed so far. The authors claim that this phase of the steam explosion process can be conservatively bounded parameterically. Perhaps, the authors have done that successfully in this study, however, a more general treatment of the break up phase, and its linking with the pre rnixture phase, should be pursued to provide greater assurance that all the initial condition effects have been taken into account.

The above two points, in no way, diminish the value of the overall approach, and the results achieved. The above two are outlines of funher work to solidify the validity of the overall approach followed here, as, I believe, the authors have themselves identified. The prerent treatment of the physics is the ' State of Art'. I believe, rapi.' advances in understanding and modeling will follow the germ ofideas that the authors have provided here. Some of those advances will surely be accomplished by Professors Theof anous, Yuen and co workers.

II. Review of the Report doe /ID-10504 (Sept.1996) " PREMIXING OF STEAM EXPLOSIONS: PM ALPIIA VERIFICATION STUDIES" by T.G.Theofanous, W.W. Yuen, S. Angelini h

This report is the verification document for the Code PM ALPHA, which treats the pre-mixing phase of the steam explosion scenario. The report has two important appendices: (a) which describes the PM ALPHA models and (b), which describes a set of experiments in the MAGICO 2000 facility, in which several kilograms of high temperature particles of a specific material, and of specific diameter, are dropped into weler to obtain observations and data on the pre-mixing geometries and void fractions. The front parts of the report provide the comparisons of the predictions with the PM ALPHA code against the data from selected experiments, in the following paragraphs, I will provide comments on the main secticus of this report.

11.1 Armemfis B: " MIXING OFPARTICLE CLOUDS PLUNGING INTO WATi~R "

I am very impressed with the MAGICO facility. I believe the authors have performed outstanding experiments using quite high temperatures and respectable masses of the hot particles. The video pictures are outstanding. I am a bit disappointed with the quantitative data that could be obtained. The X ray pictures (in reproductions) do not communicate any information and the void fraction data shown in Figares B.23, B.25 and B.27 is rather meager as a validation standard.

E-96

3 The comparisons of the PM ALPHA predictiens to the measured data, shown in Appendix B, for the cold runs, show substantial differences in 6 advancement of the panicle front. It appears that a central part of the particle cloud tunnels through the water. His is not predicted (v] well by the code. For the hot runs, it appears from Figures B.26, that the calculations predict that the dense particle cloud also leaves the steam region behind, if a slight subcooling (3* C) is present. There are no comparisons shown for the hot runs, as shown for the cold runs in the Figs. Bl4 and B.15.

The concluding remarks state that the hot tests quantified local voiding in the mixing zone and global voiding through the level swell. Figures B.25 and B.26 Indicate that the voiding front is coincident with the panicle front, only, for the zero sabcooling case. The part2.s front is substantially ahead of the voidins front for a slight (3' C) subcooling of the coolant.

'Ihe extensive steran generation, indicated by P: 'al void profile also may increase the local subcooling by pressurization. I wish there wa.s . .ne quantitative data for the particle volume fractions, to compare in Figures B.25 and B.26. Was it not possible to obtain quaniification of the spatial panicle volume fractions from the X-ray pictures?

In this context,if the PM ALPHA predictions for the advancement of the particle 'ront lagged behind the measurements in the cold runs (cf. Figs. B.14 and B.15), would they not do the same for the hot runs, since same modeling is employed for both hot and cold runs. I do expect that the steam generation, caused by the radiative heat flux or. *e coolant from the particle cloud, will retard the advancement of the particles. I believe, this effect is represented in the code, since a raMation heat flux model is employed, however, I can not quantify its effect, on the differences in the particle cloud distributien between the hot and the cold runs.

(U) The subcooled coolant is important. The only data shown for the 18' C subcooling case is the lack of measured level swell. I would be interested in the axial void fraction and the particle volume fraction profiles, to understand if there are significant phenomenological differences between the saturated and the subcooled cases, and if these differences can be predicted by the PM ALPHA Code.

All in all, I believe the MAGICO experiments are relevant for the ideas, and data, on the mixing zone and the premixing conditions. I would like to connect the meltjet particulation to the particle cloud water interaction. This may be in the next phase of authors' experimental investigations.

11.2 Avvendix A: "PM ALPIL4: A COMPUTER CODE FOR ADDRESSING THE PREMIXING OFSTEAM EXPLOSIONS" PM ALPHA is a three (melt, coolant and vapour) field code employing separate mass, momentum and energy equations for each field. Thus, it is a very detailed code - more detailed than the codes RELAP 5 and TRAC. It also employs two and three dimensional geometry. Thus, it has capabilities beyond those of the conventional CFD codes, which, generally, employ only a single field. PM ALPHA is a very advanced and detailed computer code, indeed. There are other codes, currently in development, in Europe, e.g. IVA (Siemens, Germany) and MC 3 D (CEA, France), which are also incorporating similar capability, in (9

V order to treat the very complex, and very dynamic, physics of melt-water interaction and steam explosions.

E-97

4

{

lt is a general rule that more detailed the formulation for the description of a process, more detailed the information required to bring closure to the formulation; and more intuitively intelligent approximations have to be made to obtain credible solutions from the formulation.

This is quite apparent for PM. ALPHA, when a whcle page (A 20)is needed, to show the dimer.tional groups that appear in the constitutive laws for the fuel to coolant heat transfer.

This can not be avoided, however, the collective constliutive laws may provide reasonably-correct predictions for a particular set of pre-mixing circumstances, and not for another set. I believe, that verification on an even less integral level than the MAGICO experiments should be considered by thinking-through, and devising, a set of separate-effect experiments. They should be prioritised, so that the most important are performed first.

In die following paragraphs, I will provide some detailed comments.

11.2.1 PM-ALPHA 1 ormulation The modeling approach is logical and well thought. The authors admit that the formulation so far, emphasizes the multifield aspects of pre-mixing. The meltjet and particle break up are treated parameterically.

Two length scales are employed for the fuel field: one large, encompassing the original fuel drops, or fuel-melt jet, whici may break up but still are considered as fuel; and the other small enough to be called a debris, which mixes with water and gets quenched. The decisions about the amount of the ' fuel' and the ' debris

• are made with a correlation for the fragmentation rate.

The debris part:cles assume the same temperature, and velocity, as the coolant,instantly.

O They are not allewed to sedinwnt down with gravity, as they would normally do. This assumption is justified for the time interval considered, if the particles are of micron size.

The large-length scale fuel particles are assumed to have unifonn temperature. There is no treatment of the heat caduction from the fuel particle to the coolant. For the prototypic binary-oxide mixture men. P. is important to determine the solidification front growth into the particle, .-ince it may either prevent fragmentation, or reduce the rate of frag'nentation, thereby changing the heat source to the coolant.

Another facto: ir@ ement of the fuel particles, is the change in physical properties that occurs, as the fuei carticles cool down from abcve liquidus to below-solidus temperature. The increase in viscosity and surface tension affect the fragmentation characteristics, which in tum affect the terms in the debris mass equation, and in the liquid and debris momentum and energv equations. A paper submitted by Okkonen and Sehgal in the forthcoming FCI meeting in Japan discuss the two factors mentioned above for the behaviour of the fuel drops.

Recently, we at R yalInstitute of Technology (RIT), have performed some experiments on the interaction of .elatively low temperature cerrobend (an alloy with density of = 9 000 kg/m'

) jets with subcooled water. We have found that thejet brer.ks-up into small particles. There is a distribution to the particle size or mass, however, there were no particles oflength- scale comparable to the jet diameter. In these experiments the jet breaks-up completely. The FARO E-98

experiments show a melt ' cake

• at the bottom, however, it is not clear whether it is the unbroken jet or an agglomeration of melt droplets belonging to some size distribution, which,

/3 perhaps, does not contain length-scales approaching the meltjet diameters.

D Summarizing the above discussion,I believe, the treatment of fuel as having two length scales in the PM ALPHA formulat on i is valid. However, the source terms in the equations should be reviewed again. The variatio of properties of the fuel drops, with temperature, should also be taken into account; and the change in .c tert prature of the fuel drop should be calculated employing conduction equations. Mel jet, or ; rop, interactions with subcooled coolant may produce atomization, with no large partdes of itze similar to that of the meltjet, 11.2.2 Interfacial Momentum Transfer in PM-ALPHA The drag correlation used in PM-ALPHA for fuel-coolant interface distinguishes between the dispersed and the dense fuel regimes. The latter is taken as that for flow of gas through a densely packed bed. This correlation, perhaps, should be checked, since predicted penetration of the fuel cloud in the MAGICO experiments is less than the measurements.

Also, comparisons could be made with the isothermal tests in the BILLEAU and the QUEOS facilities. The logic diagrams on pages A-16 and A-17 were helpful.

11. 2.3 Interfacial Heat Transfer in PM ALPHA There are many regimes of convective heat transfer and many correlations. The authors use the best that they can find . Then, there is the large effect of radiation heat transfer, which

(

i was found to be important for the comparisons to the QUEOS test data.Their synergism, and effects of one regime on another, may need further exploration. For example, radiation-absorption will produce vapour which will change the convective flow patterns of the coolant, and, perhaps, change the heat transfer regime. Some separate- effect tests could be designed to test the synergism and the cifect of different convective regimes on each other, in order to test the heat transfer correlations package employed,

11. 2.4 Fuel Break-Uo and Fracmentation Modeline in PM-ALPHA I have referred to this earlier in the comments on the PM ALPHA formulation. The interfacial area equation (3.73) assumes spherical particles on break-up and fragmentation. This may not be appropriate. Perinaps, data from FARO or other fragmentation-break up experiments could be employed to develop a more prototypic interfacial area representation. In some of our experiments with cerrobend in subcooled water, we do not find spherical particles. Perhaps, in saturated water, with large flows of steam, the particle shapes may be spherical.

The model for fragmentation of fuel drops is based on the Bond number. I believe, data on hydrodynamic and thermal fragmentation oflarge-size melt droplets may be available in near future. The model could be checked against such data, when available.

The model for jet and large fuel drop-break up is parametric with an input -specified parameter p, whose value is varied in analysis. This approach is, perhaps, adequate for the

(\ present. However, it will be desirable to have a phenomenological/ mechanistic model.

E-99

6 l

l The authors distinguish between fragmentation and break-up as two separate processes. In some of our melt jet-water interaction experiments, we were not able to separate the two processes. The jet breaks-up (or fragments) into particles having a size distribution ranging from submillimeter to 3-4 millimetres. The process appears to be concunent and not sequential, as assumed in the parametric models described here, h

11.3 VERIFICATIOrt OF the PM-ALPHA CODE The >'ront part of the report doe /ID 10504 describes the verification pursued for the PM-Al Pila code by performing analytical tests, and by comparing with the data measured in wveral experimer.ts. This was a very large effort, and I believe, it has largely achieved its purpose. I will comment on a few comparisons of the data with the code predictions.

11.3.1 OUEOS Exoeriment These experiments are similar to the MAGICO experiments. The comparisons shown in Figuies 4 to 13 are remarkably good for such a dynamic process. The comparisons appear to be better than those for the MAGICO tests, it is not clear to tr.e what the experimental image actually implies, in terms of the distribution of hot particles, and of void. The pictures in Fig. 7 at 0.3 and 0.4 seconds seem to show that the experimental hot particle image may be not as advanced as the calculated contour. This also appears to be the case in Fig 6. at 0.3 and 0.4 seconds. T he graphs in Fig.

8, however, show very good agreement between measured and calculated front-advance locations versus time, 11.3.2 MIXA Exneriments The MIXA experiments employ a Uranium-Molybdenum thermite melt of several kilograms, at 3600K, poured into near-saturated water pools. The meltjet was broken into 6 mm diameter droplets. The MIXA-6, analysed here, used 3 kg melt pour into very nearly (s 1 K difference) saturated water. This, thus, is a prototypic experiment, albeit with small mass.

The comparisons are very good. I am somewhat concemed about the sensitivity of the results to the break-up-cut off void fraction and particle size. The authors recognise this, still, a change of only 5 % (85 % to 80 %), with the particle size of 1 mm, decreases the calculated pressure rise from = 0.4 bars to = 0.2 bars. Increasing the particle size from 1 mm to 1.2 mm at the 85 % cut-offlevel decreases the pressure rise from = 0.4 to = 0.28 bars. Thus, the break-up and fragmentation models appear to be very influential in the very high temperature, prototypic material experiments.

E-100

-.- -- -- - - - . . .- - - - . ~ ~ . _ - _ - - - - - - -.-...

_1. h i

- 11.3.2 FARO Experiments

• p V These are, perhaps, the most important experiments, since they use substantial quantities - '!

(it 100 kg) of prototypic materials;. and there are several experiments already performed and  !

more are underway. ,

The comparisons shown are very good indeed. Unfortunately, FARO does not produce any data on the mixing region, thus the colour figures, presented, show only calculations and no data.

I did noi understand why the initial particle size is chosen as 4 cm for ajet diameter of 10 cm.

--The p value chosen is 50, while for the MIXA test it was chosen as 20. The minimum particle size chosen is 1 mm, while in the MIXA test is was chosen as 1.2 mm.

.One experimental result, which FARO produces is the fraction of the jet material deposited as a ' cake' on the bottom plate. This is not provided by the authors from their analysis with the PM ALPHA code.

HA Numerical Aspects

- The authors do not provide a discussion on this topic. I believe, this is an important topic.

The ICE technique is known to have significant numerical diffusion. It is not clear whether

. any advanced space-time discretization scheme was employed. Node sizes of several ,

centimetres are generally not fine enough The authors, perhaps, by now, have investigated O. the numerical aspects further, and I would welcome a greater discussion of this topic.

l

- III, Review of the Report: Propagation of Steam Explosions: ESPROSE.m VeriGestion Studies by T.G. Theofamous, W.W. Yuen, K. Frceman and X. Chen c

This report deals with the next phase in the steam explosion process, after the pre mixing has been achieved. The report, therefore, deals with the explosion process and develops a methodology to describe the process, and evaluate the energetics, which are then employed to assess the damage potential of the explosion on structures, which surround the explosion. A trigger is assumed, which starts the explosion process, in which intimate contact of the fuel and the coolant leads to production oflarge amounts of vapour, and the supercritical explosion.

r The report consists of the front part, where the results of the verification calculations are compared to the observations, and data,' obtained in the SIGMA facility at U.C. Santa Barbara. The report also contains four important appendices in which the code models, a 1-D characteristics model, constitutive laws for micro-interactions and thermal 4tanations are discussed.

f in the following paragraphs, I will comment on each of the major sections of this report.

E-101 s

rii.--+.,.. ~-...w,-

, ,a  :~ , . . , - , - . < -

Illi Appendix A: THEESPROSE. m Af0DELS The overall approach of the model development is brilliant. Recognising that the dynamics of O

a pressure wave, generated by a trigger, coupled with fuel fragmentation, micro (or local) mixing and heat transfer result in energetic steam explosions, the authors have concentrated on those aspects. Perhaps, the SIGMA experiments provided the key observations towards the development of the micro-interaction concept and the m fluid, where the fuel-coolant heat transfer occurs. The energy transferred is then employed in the multifluid treatment to calculate the pressure fields as a function of time and space (2-D/ 3D). The damage potential is, th:n, evaluated with the calculated dynamic loading imposed in terms of kilo Pascal secor.ds.

The modeling approach is similar in most respects to that employed for the PM ALPHA code, i e., solution of a set of multifield conservation equations, with specified constitutive relatior.s. The fields chosen this time are fuel, liquid and the m fluid. There is an additional mass conservation equation for the debris, i.e. the fragmented material. The m fluid equations contain rource and sink terms, which are based on a picture ofliquid entrainment and phase change. Fuel fragmentation is included, which contributes to the increase in interfacial areas.

The heat transfer across the fields is included in the energy equations. The system of equations appears to be complete. The constitutive relations between the 3 fields for interfacial drag, heat transfer and phase change, again involve many correlations and dimensionless numbers.

I believe the comments that I had made regarding the complexity of the constitutive relations for the PM ALPHA code also apply here, and the possibility of checking the synergisms between the momentum and heat transfer processes through separate effect experiments, should be explored. New data may have to be obtained and some prioritisation should be performed. The fuel fragmentation is treated as in the PM-ALPHA code and is controlled parameterically through p . There is another parameter which enhances the fragmentation for thermal effects. Both of these parameters are user specified. The entrainment ofliquid in the m fluid is controlled through the parameter E, which is taken as a function of the fragmentation rate.

believe, the parametric treatment is very intuitive, and the authors admit that it is an important component of the micro-interactions concept with somewhat speculative constitutive laws. Since, the m fluid interactions are the basis of ESPROSE-m , I hope that the authors have already obtained additional data from the SIGMA facility to provide greater support for the experimental basis of the parametric treatment.

111.2 Appendix C: Constitutive Laws ofhiicro-interactions This appendix describes the experiments performed in the SIGMA-2000 facility with gallium and molten tin, subjected to high pressure waves,in order to derive the constitutive laws for the micro-interactions, needed for the m fluid.

E-102

9 The experiments are described, ney are really very difficult, but precise experiments. Son.e results are shown as movie, X-ray and SEM images for the change in pr -mixing volumw as o a function of time.

The results of experiments are used to derive the values for pr, y i and f , the entrainment factor. For example, Fig. C-13 shows fe=7,8 and 12 give best fits, respectively, for three isothermal Gallium tests i.e., G/204/45, G/68/45 and G/272/45. The more conservative value fe=7 is then used to determine the value of pr=9. De value ofy is derived from Fig C-10, while keeping fe=7, and pt=9. It appears that'yi varies from 1.4 at 68 bar pressure to 4.2 at 204 bar pressure.

The above is a logical but highly empirical determinatien of 3 parameters from a small number of tests. Perhsps, more data has been obtained from SIGMA to confirm the choices made for these key parameters. Obviously, more data is needed from SIGMA or'another shock tube. I believe, different materials should also be tested, in particular, melt drops of binary oxides. Their fragmentation behaviour maybe different, due to changes in properties they experience with a change in temperature.

111. 3 Appendix D. On the Existence of Thermal Detonations This is a very interesting re-examination of the Board-Hall model for steam explosions. The micro-interactbn model and the concept of the m fluid is employed to show that supercritical l steam explosions can be obtained with lean mixtures in highly voided regions; conditions for lq which the Board-Hall model will predict only very weak explosions.

' V My understanding of the micro-interactions concept, introduced by the authors, is that they take place in the m fluid in a limited volume. I believe, this results from the observations made from the Gallium drop (also perhaps the tin drap) experiments conducted in the SIGMA facility. The previous concept was that the pressure wave.will fragment a melt drop into fine dreplets, which will mix with tne whole coolant volume. The SIGMA experiments showed that this does not oc:ur in the time frame of the pressure-wave melt drop interaction. The heat

! transfer to the m firid's coolant, in the limited volume occupied by the m. fluid, generates

! very high pressure!. The shock wave then travels into the non-participating fluid around the m-fluid, increasing its pressure to sustain the propagation. This makes possible the supercritical steart explosion with a fuel-coolant mixture, which is lean on an overall-volume basis, but :s net so lean on the m fluid volume basis. (Cr. Figures D.8 and D-9, where high pressures are abtained for the coolant to debris mass ratio fe=1 in the case of tin at l 1500'C and for fe=2 to 8 for the case of UO2 at 3300*C)

I believe, the authors have provided a very logical explanation and frame work. I am, however, a bit concerned about the value ibt fe, which was chosen as 7 in Appendix C, based

! on the data from the gallium experiments in the SIGMA facility. In Fig D 8, a value of fe=7 will not produce a supercritical steam explosion. Thus, the value of fe may be material dependent, and more information is needed to choose an appropriate value.

O l V E-103 i

i 1

l l

. 10 l

l 111.4 VERIFICATIONSTUDIES l

The front part of the ESPROSE. m report describes the analytical tests, the SiOMA i experiments, explosion coupling, integral aspects, numerical asp: cts, and finally, a i comparison with the KROTOS tests. I will comment on these, briefly, individually.

1 111.4.1. Analvtical Tests 1

These are very valuable exercises and show that the modeling in ESPROSE-m can predict I pressure wave propagation. There are many figures. I wish there were more explanations i e.g., Figures 17 and 18, both show very good comparison between the analytical and the ESPROS m pressure distributions for early times, but deviate at later times. Is there an explanation? Similarly, there is a crater in the middle of the pressure wave in Fig 19. Is there a physical explanation for that? This section may be improved by the authors, through some explanatory text. It is very valuable, otherwise.

111.4.2 SIGMA Exneriments Thcse experiments, specially conducted in the SIGMA shock tube provide data for verification of the ESPROSE-m models for pressure wave propagation. The comparisons are excellent. There are some differences for the inhomogeneous cases, which, perhaps, are difficult to fix. Allin all, it is a splendid performance for the code for these separate-effect tests.

Ill. 4.3 Comoarisons with KROTOS Exoeriments KROTOS experiments provide the most appropriate data for the verification of the ESPROSE-m models. The KROTOS facility has performed steam explosion experiments by triggering the pre mixtures of water with several different material melts. The initial conditions, e.g. melt mass, melt temperature, melt superheat, pressure, water subcooling have been varied to provide a reasonably extensive data base. The test program is continuing, and could provide the data base needed for the ESPROSE-m valida'!on. Unfortunately, as in most of these melt water interaction integral experiments, the data obtained is integral anj the premixing and the steam explosion processes are not delineated. Thus, detailed verification and validation of the ESPROSE-m (or any other steam explosion code) may not be possible.

The document provided on the analysis of the KROTOS tests speculates that the melt break-up and quick freezing may be a reasonable explanation for the non-explosivity of the Uranium oxide tests. We reached similar conclusions, and also, evaluated the effects of the change h the surface tension and viscosity of the binary-oxide melt, as it cools down below the liquidus temperature. This has been reported in the 1995 ICONE meeting, and additional viork will be reported in the forthcoming FCI meeting.

Coming back to the comparisons of ESPROSE-m (using PM-ALPHA pre-mixtures) predictions against the measured data, the authors admit difficulties of representing particle freezing correctly in the PM-ALPHA formulation. The fuel- participation factor chosen affects the result greatly. The pressure wave-shapes versus time appear to be reasonable but E-104

It there are differences e.g. for KS there appears to be an earlier venting of pressure wave. I believe, revision of the PM ALPHA numerical scheme and/or modeling of the heat O

V conduction in the fuel particles (as was mentioned earlier in the comments on PM ALPHA modeling) may resolve this difficulty. The sensitivity to fuel participation factor is very large, indeed.

III 4.4 Numerical Asoccts I have similar comments as I had for this topic in the PM ALPHA document. The authors should provide more discussion and, perhaps, comparisons of the use of the ICE technique for similar problems. The numeical diffusion issue is quite important when, tracking pressure waves and/or interfaces. Recently, special numerical schemes have been devised to reduce or eliminate numerical diffusion. The ICE technique does not, compare well to such schemes, in term ofits performance, and with respect to numerical diffusion. Perhaps, the authors have implemented another scheme in developing the ESPROSE-m 3D code.

IV. Review of the Report: Lower HeadIntegrity UnderIn VesselSteam Explosion, DOE /ID-10541 by TG Theofanous, W.W. Yuen, S. Angelini, J.J. Slenicki, K. Freeman, X. Chen and T. Salmassi This report is concemed with answering the question: "Will the lower head of the advanced O passive reactor AP-600 fail, under the dynamic loading imposed by an in-vessel steam explosion, ifit were to occur?" This is an important issue for the accident management strategy chosen for the AP-600,i.e. retention of the core melt in the lower head, by employing extemal cooling of the vessel.

The methodology used to .esolve this issue is the ROAAM method developed by Prof.

Theofanous, employed most recently to respond to the companion question "Is it possible to retain the molten core of the AP 600 reactor,in the lower head by cooling the vessel extemally?" This question was .nswered in the affirmative by employing the ROAAM method 'Ihe ROAAM method has been extended and further clarified by Prof. Theofanous in a recent publication, attached as Appendix A in this report. .

Besides the ROAAM philosophy and procedures described in Appendix A, the detailed pre-mixing and explosion results are described in Appendices B and C respectively. Appendix D provides additional pre-mixing perspectives from the THIRMAL code, prepared by Drs. Chu and Sienicki of Agronne National Laboratory. The important chapters, in the main body of the report, are concemed with structural failure criteria, melt relocation characteristics, quantification of pre-mixtures and explosion loads and fmally the assessment of the integrity of the lower head of AP-600.

p d E-105

' 12 I

In the f:llowing paragraphs, I hava provided comments on the cppendices, chapters and j conclusions of the repon in the order:

- Chapter 3: Stmetural failure criteria I

Chapter 4: Melt relocation characteristics

- Chapter 5: Quantification of pre-mixtures g

- Appendix B: Detailed pre mixing results

- Appendix D: Additional pre mixing perspectives from the THIRMAL code

- Chapter 6: Quantification of explosion loads Appendix C: Detailed explosion results

- Chapter 8: Consideration of reflood FCIs

- Chapters 7 and 9: Integration, assessment and conclusions JY.1 Chapter 3: Structural Failure Criteria This is an important chapter, since it establishes the fragility curve, giving the probability of the lower 'iend failure for dynamic-loads ofincreasing magnitudes. The impulse loading, of interest, is in the range of 100 to 300 kilo Pascal seconds.

The authors have employed a commercial structural-analysis code, whose results they have compared with a simple analytical solution. ABAQUS is a 3-D finite element code, able to model the hemispherical lower head and the dynamic loadings imposed. The code provides .

the strain as a function of time for the assumed loading. These calculated results are, then, converted to a fragility curve, assuming probabilities oflower head failure, when strains of greater than 11 % are reached over certain fractions of the lower head wall thickness. g The ABAQUS calculations are performed for various loading pattems on the lower head. The non-uniformity ofloading was found to decrease the strain for a specific impulse. The colour pictures provide very nice strain morphologies.

This chapter provides clear and transparent results. The ABAQUS 'results are confirmed against a simple model for uniform loading. He fragility curve makes good sense.

I am a bit concerned about the very local non homogeneous loadings of the type predicted, later, in the report. Perhaps, a few ABAQUS calculations could be performed to establish the fragility curve for such a local-loading pattem.

IV. 2 Chapter 4. Melt Relocation Characteristics This chapter provides the initial conditions for the scenario of melt-water interaction in the lower head. He chapter, therefore, deals with the melt pool formation in the original core boundaries and, later, relocation of the melt from the in-core location of the lower head. The quantities needed are the rate of melt addition to the water in the lower head, thejet geometry (diameter, velocity and location in the vessel), the melt composition and superheat and, finally, the timing of this event relative to the other events in the core melt-progression process.

E-106

-- .. . - - - - - -- . .. . - - - . - _ - . - . - ~ _ . . _ - -

E 13 ,

The authors, first' point out the differences in the AP 600 core configuration from that of the conventional PWR. The AP-600 has some features which are quite favourable in terms of the melt releases conditions. Dese are the massive 36 cm thick core support plate, the core v - reflector, the gap between the core barrel and the reflector on the flat sides of the reflector; ,

and the long unheated section in the fuel elements at the bottom.

The authors have developed a credible scenario of melt pool formation, melt attack on the reflector and the core barrel. It is supported by enveloping models of appropriate ' complexity, which provide physical insight and transparency. The authors are wise not to use one of the myriad codes,' which provide user-motivated results. The analysis is brilliant and quite comprehensive. The melt release conditions of 200 to 400 kg/sec should be bounding values.

The melt superheat of 180 K also should be a good bounding value. The location of the release, near the top of the core in the vessel downcomer, may also be credible. Thejet -

velocity of few meters /second also appears to be sound. I, however, would like the authors to consider the following cautionary points: ,

~(i) ne timing of the melt release 76 to 91 minutes is much too close to the timing of = 100 minutes for evaporation of water in the bottom 25 % of the core height by the radiative heat flux imposed.

(ii) The core plate is massive but it is also loaded heavily. If the core plate temperatures go beyond 700'C, the yield strength will deteriorate.

(iii) The melt pool with =40 to 60 % unoxidized zirconium and some stainless steel, will probably form a primarily metal layer on the top. This layer is , thin and will focus the heat LO. flux to the sides. Recent work at RIT has evaluated the heat transfer from the metal layer to the vessel (which is of a thickness similar to that of the reflector) with a two-dimensional code, and found that the highest heat flux is still at the comer of the oxide pooljust below the l metallic layer. Thus, the failure could be below the metal layer.

I;

' (iv) While, I agree with the authors that the flat part of the reflector being closest to the core centre is most likely to be attacked first by the pool. The oxide pool however may not be axially symmetric and there may be azimuthal regions in the core, where fresh fuel and high

. power are dominant. Evaluation of a possible attack on the non-flat parts of the reflector should be considered.

l, .

i (v) The draining and freezing of the metallic layer into the well between the flat part of the reflector and the core barrel, without participation in any melt-water interaction,- is very

,- likely, but sounds too convenient. Additionally, in the absence of water above the core plate in l= the well, the thermal loading imposed by the superheated metallic melt on the core plate, or on E the core barrel region directly: above the core plate should be evaluated.

- Summarising, I believe, the authors' estimates for the range of melt release-characteristics is y credible, however, additional evaluations may help to put these estimates on a more solid i footing. l E-107 I

f

_ -. .~ ;-... _ _ . . , _ , . ._ - - - -

14 IV. 3 Chapter 3 Quantification ofPre mixtures Appendix B: DetailedPre-mixing Results The chapter 5 develops the rationale for the pre mixing that results from the release of the O

UO2 ZrO2 melt from near the top of the core, through the downcomer, into the water pool of the lower head. The water level is assumed to be a few centimetres above the top of the core support plate. Melt release rates of 200 and 400 kg/sec, reaching the velocity of 5 m/sec at entry into water are considered. The melt superheat is assumed as 180K.

The oxide meltjet is distributed over an effective radial width of 10 cm in the downcomer, with an initial melt volume fraction of = 25 % at water impact. This would translate to a melt stream of dimensions = 10 cm x 16 cm for the release rate of 200 kg/sec and = 10 cm x 32 cm for the release rate of 400 kg/sec.

The expanded melt jet is then allowed to traverse 20 mm in water, before break-up ensu'es.

The break-up rates are parameterized from no break up to very rapid break up (forming 2 mm size particles within 10 cm of travel in water.)

The above initial conditions were employed in the PM. ALPHA code to provide results on pre mixture characteristics i.e. the melt and the void volume fractions and the fuel length scale, as a function of time, and position. The integral quantity ofinterest is the number of kilogram of melt mixed with coolant, beforc the triggering and explosion.

The Appendix B presents a number of colour pictures and many graphs giving detailed results. The graphs of fuel length scale, fuel volume and void fractions are presented for more p values and for times upto = 1 sec. These pictures and graphs provide good back- up for the results, and arguments, presented in chapter 5.

I believe the authors have presente J a clear method of evaluation and the results are credible. I do have the following comments.

(i) The melt-through failure of the reflector and core barrel are assumed to be near the top of the melt pool in the original core boundary. If the failure is lower, the starting velocity for the meltjet would be higher, and so will be the velocity at water impact. This may be beneficial for break up.

(ii) T he initial impact area on the water surface is quite large. The jet going through the 2 meter steam region should not break up, to that extent.

(iii) Both the very fast and the no-break-up cases show (Cf. Figs. 5.4 (a) and 5.4 (b)) that for the initial =0.1 see the fuel front is more advanced than the void fraction front. This was also observed in the PM-ALPHA verification report. Later on, the void fraction front seems to catch up with the fuel front. For the C 1-10 case at = 0.4 seconds (Page B.3-3) a large frnction of fuel seems to be hung up in the voided zone. The same is true for C 1 - nb case (Page B.3-5). In the Cl-10 case, there would be a large steam flux rising, which could retard the descent of the fuel particles. For the C 1 - nb case the steam flux should be smaller, and the fuel particles of 2 cm should be ahead of the void fraction front.

E108 h

ts.

> . Summarising, I believe the break up assumptions, both, in the steam during descent ikom j

_ the original core boundary, and during water interaction, play a crucial role and, perhaps,--

X this part of the pre-mixing analysis could be strengthened. The no-break-up case appears to. ,

produce approximately the same results as the high break up case. This has been recognised,

' also, by the authors (Page 5 10). Perhaps a physical explanation of why these cases produce -

such similar results may be provided by the authors. ,

IV.4 Appendix D: Additionall' remixing Perspectivesfom the THIRMAL Code

-.In this appendix, the THIRMAL code has been used by C.C. Chu and 11 Sienicke of Argonne National Laboratory to provide a perspective on premixing. The code had to be modified to describe the melt jet-water interaction in the confmed geometry of the down comer. The calculations were performed for melt release rates of 14 to 220 kg/sec, with corresponding jet diameters of I g mm to 73 mm. The 220 kg/sec case resulted in median droplet size of 2.75 mm, with a mixing zone radius and void fraction at pool surface of 160 ram and 74 %, respectively.

These results are not too different from what were obtained from the PM ALPHA Code, although the jet entry conditions are different. THIRMAL calculates jet entry diameter of 6 cm (i.e., no break-up in the down comer steam zone). Models for break-up in THIRMAL must be quite different from the parametric model employed in the PM-ALPHA Code.

IV.5 Chapter 6. Quant {fication ofExplosion Loads. 1 Appendix C: DetailedExplosion Results The chapter 6 and Appendix C present the results of explosion-propagation calculations performed with the ESPROSE-m code, using, as initial conditions, the pre-mixture configurations calculated with the PM ALPHA code. The trigger time is chosen as very short, since during the early time the void fractions of the coolant around the fuel particles are relatively low. Later, the void fractions increase substantially, and would inhibit fuel break up

- and triggerability.

The results are presented for the C-1 and C-2 scenarios with three values of and a set of trigger times. For the no break-up case these times vary from 0.05 see to.l.0 sec, while for the break up cases, they vary from 0.04 to upto 0.19 seconds.

The results on pressure, impulse and effective area are shown for various locations in the lower head. Peak loadings histories are also shown as a fur.ction of trigger times. The extreme

~

- sensitivity to trigger time is evident from Table 6.1. If the trigger is delayed by 0.06 seconds for the C1-10 and C2-10 cases, there is only a very weak explosion. For the Cl-20 and C2-20 .

cases, there appears to be a time interval of only 15 30 msec for the trigger to generate a supercritical explosion. Thus, triggering time appears to be the deciding factor. A physical explanation for this extreme sensitivity should be provided by the authors.

The Appendix C gives very nice pictures of the pressure wave traversing through the lower head. The pressure signals at various points in the , lower head are shown, and the reak E-109 e)

< r -P . - ' - ' ' > - * - - - ' ww e er-;>v , - - - - - - - _ _ _ _ - - - - - - - - - __m. __

' 16 j

pressures and impulse loadings are shown as graphs versus time. Theses pictures and graphs were very helpful 2n the review of Chapter 6.

Summarising, I can say that the authors have performed logical analyses of the loadings imposed by the steam explosion, and have provided very nice results. I have not understood the reasons for the extreme sensitivity of the calculated results to the trigger time. The peak loadings, shovm in Table 6.1 are, in ger.eral, modest. The highest loading is found to be

=200 k. Pa. s. Is it possible that for p = 30, a higher value than 200 k. Pa. s. is calculated?

lY.6 Chapter 8. Consideration o,fthe reflood FCI's This chapter deals briefly with the stratified steam explosions that may result, if the reflood is effective, and a layer of water is brought on top of the melt pool, which has a metallic layer, at top.

It wa. found that the stable water layer may not exceed 10 cms, due to the low reflood rate and the time to freeze the upper metal layer. Any stratified explosion will be easily vented.

I b. lies e, the authors have a good argument. Certainly the peak pressures in such an explosion should be low and reflood FCI's may not be a problem.

IV.7 Chapter 7: Integration. Assessment Chapter 9: Conclusions These chapters combine the results achieved in the previous chapters and appendices to proside an overall assessment. This work was already practically done by the results achieved, since the maximum impulse loading was below the minimum of the fragility curve.

This was also confirmed by performing ABAQUS calculations for the peak loading for the actual cases and finding that the lower head strains were very low.

The authors conclude that for the saturated water case, the lower head integrity can not be compromised by a steam explosion. Having highly subcooled water is the only possible way to, potentially, involve a larger mass of melt, and produce a more energet.c explosion. The authors conclude that obtaining highly-subcooled water, even in reflood scenarios for the AP-600 is not credible.

V. Concluding Remarks in this section, I would like to provide a few concluding remarks after the review of the three reports.

I must congratulate the authors for producing such a fine and comprehensive body of work treating the tricky and controversial area of steam explosions. While, most of the researchers in this area are still trying to understand the fundamentals, the authors have leaped ahead with new concepts, advanced codes and considered-judgements to provide a reasonably robust estimation of the damage potential of a steam explosion. They have combined this E-110

.n -

with structural analysis to show that AP-600 lower head can withstand the dynamic loeds imposed.

Ths' authors have, also, noted the peculiarities of the AP-600 configuration and employed the advantages and disadvantages they confer on the analyses. Some of these peculiarities (differences) provide great advantages e.g. in the core melt progression and the melt release i characteristics. These sound a little bit too convenient and, perhaps, should be re-visited.

The authors have modelled the fuel break-up and fragmentation process only parametrically.

This may be a weak point in the whole development; since those processes provide the initial conditions for both the pre-mixing and the propagation phases of the steam explosion.

Perhaps, the analyses are well-bounded for these processes; howcVer, the sensitivity of the results to the break-up and the fragmentation modeling is very large.

Then, there is the question of maturity and of validation versus verification. I believe the methodology and the data presented, robust as they are, t.re still very new. The comparisons - -

presented against test data are not extensive, and I think, the authors recognising this, have wisely titled the reports as verification reports. Further experience with this methodology and further comparisons with separate-effect (e.g. SIGMA, MAGICO, BILLEAU and QUEOS) data and integral effect (e.g. FARO and KROTOS) data would provide validation and maturity to this methodology. In particular, the constitutive relations, being so many for such complicated phenomens, need greater experimental back-up. I believe, the authors are already busy in achieving such experiments in the MAGICO and SIGMA facilities.

l Lastly, I must say that I have enjoyed reading the reports and teamed much from them. I O think, I now understand the concept of micro-interactions and the m fluid. I have made constructive (hopefully) critical comments at places, to provide input to authors towards l Improvement of the reports. I believe, they have largely achieved the objective they had set out to achieve.

E-111 u

L_ __ - .. .- . - . -. - . . -, ,

PAUL SHEWMON 2477 Lytham Rd, Columbus, OH 4322o Tel. 614 457-4378, Fax 614 457 2995 ocail: sHEWMoN.1@osu.EDu August 6,1996  ;

Re: Evaluation of report by T.G. Theofanous, et al., " Lower Head Integrity l Under in Vessel Explosion Loads", DOE /ID 10541 The analysis of head failure sets up a model of the lower head using ABAQUS (a well established finite element code) to relate the stress pulse from steam explosion to local strain. The vessel material (ferritic SA508 steel) will undergo large amounts of strain (elongations of 50 to 100%) before fracture occurs. Whether or not the vessel undergoes  !

any plastic strain depends on the yield stress of the metal and the impulse from the steam. ,

For reason never explained or discussed, the authors chose 330 MPa for the yield stress of the vessel. They state that the conservative ' Code Allowable' is 345 MPa, and the actual value (found in a conventional tensile test) is 450 MPa. The choice of 330 MPa introduces a large conservatism (safety margin) since a best estimate should use 450 MP.A.

The impulse applied to the steelin the lower head would have a rise time of a few milliseconds. When ferritic steels are loaded this quickly their yield stress is substantially greater than that observed in a normal tensile test. The authors quote references that show the yield stress at this strain rate is about 40% greater than that found in a tensile test. They take full credit for this strain rate increment, which is justified and appropriate, in summary, the analysis of head failure seems to be competently and conservatively done, and the conclusions drawn are appropriate. I have also looked at the discussion of loads and loading. I am less of a specialist in this area, but it also seems to be well done.

Though no mention of radiation effects is made in the report, the analysis should be made for the vessel at end of life (40 years?). The temperature of the head during the eccident considered would be less than 212 F. This is beneath the RNDT for the beltline of some of the vessels now in service, i.e. such material might well behave in a brittle manner durir.g an accident of the type considered here. I considered this, but feel such radiation effects are not germane in the case of the AP600 for at least two reasons:

1) The fast neutron and hard gamma flux in the lower head will be at least a couple of orders of magnitude less than that in the beltline region of the vessel, so radiation effects should be negligible.

2) The steel to be used in the AP600 vessel should be appreciably lower in the elements than have load to radiation embrittlement (copper, e.nd phosphorous) in the older vessels now of concern in plants in the U.S.A.

With this in mind, I blieve there is every reason to believa that the material in the lower head would behave in a ductile manner and that the aridlysis given in the report is appropriate for (would apply to) a vesselin the AP600 after 40 years of service.

Or e ci i vud b b.W. O.dvic k .

Paul Shewmon E-112 h

• I M

% . g ,r # N 1 L6 l From: BrKn Tartand L Sent: Friday, October 25,199611.05 AM To: Reactor Engineering

Subject:

Re[2): Review of DOEllD 10541 s_ ~

Walt Dedrich Reactor Engineering Argonne National Laboratory Dear Walt.

I attach a Word 6 document containing my review of DOE /ID-10541. This does not contain any comments on the PM-ALPHA Verification Studies. I am generally happy with the PM-ALPHA report. I will send some formal comments later, but I now have a period of leave, courses and foreign travel that takes me through until November 7.

I hope you find my comments on the main document and the supporting document on ESPROSE.m verification studies useful. Please confirm that you are able to read the document successfully. .

Thank you for asking me to do this work. If I can be of help in the future, please let me know.

Best regards Brian Turtand AEA Technology plc Phone +441305 203029 Fax +441305 202508 e mail brian.turlind@aeat.co.uk E-113

REVIEW OF DOE /ID-10541: 4 LOWER HEAD INTEGRITY UNDER IN-VESSEL STEAM EXPLOSION l l

LOADS 4

by T G Theofanous, W W Yuen, S Angelini, J J Slenicki, K Freeman, X Chen and T Salmassi.

REVIEWER: B D Turland (AEA Technology plc, UK)

OVERALL COMMENTS This report and its associated documents represent the culmination of several years work by Prof. Theofanous arv' las colleagues. They have now demonstrated that the basic framework for a Gam explosion assessment in realistic geometzy is in place. This is a major achievement.

The reliance on detailed modelling codes makes the reviewer's task difficult - in the end one can look at the validation offered and consider whether the results presented look reasonable. In the supporting documents the authors make good use of the available experimental data to benchmark their calculational models. However, it is accepted that some of the constitutive physics use

2. The convnent that the break up parameter p set to 10 produces very rapid b.mak up 1 in ~10 cm of water suggests that the medelling is somewhat more efdciene at producing fragmentation thsn originally desired (break up in a specified fall distance taken as le smaller of the actual fall distance or p.Ds ). This also depends greatly on the assigned value of Di- here set to the initial particle size (20 mm). If the melt was assumed to fall as a thinning sheet (quite possible) then the initial penetration of the water mty be more local than represented in the PM-ALPHA calculations. However, I ara happy with the range chosen for p.

3. Dlease note that in Figure 5.2 and Appendix B the void is .:epresented by shading,

% fuel by contours. Explain the contours that follow the domain boundary,

s. Specify the boundary conditions for the calculation. What pressurisation is pr dicted?

5. The it gth scale increese referred to on page 5-5 is not evident in Figure 5.4. The area averaged over is nat clear, it is obviously not the whole cross-section. Since n riting this I found the 1% fuel volume fraction limit on the region considered in the text - for clarity add to caption of Figure 53,

6. i,3ddle of page 5-10: 'Only a very small fraction of the coolant is found to co exist with the water' - I know what you mean! It is clear though that here we have the key result anticipated for the mixing codes. This implies that the key region to seek validation of the code is in the production of the high void fraction.

7. In my view the THIRMAL calculations raise as many ques.tions as they answer, because of the poor validation status of any jet break up model. However, I do not think this is a key part of the argument.

Chap;er 6: Quantification of Explosion Loads

1. Where is the trigger cell?

2. At what time was the effective area evaluated - that cf peak pressure? If not, you obtain larger effective areas than ~0.1 m2,

3. The question raised by the calculations is how far is one from the danger zone?

Could we get there by a modest increase in system pressure (what value was assumed?) and/or varying the value of p?

E-118 I

Chapter 7: Integration and Assessment

1. 'Ihe conclusions reached are justified on the basis of the analysis presented. On the basis of current knowledge I am still not comfortable with the observation that downward relocation scenarios are ' physically unreasonable'.

2. I agree that there is a greater threat from subcooled conditions. It is not obvious, though, that a ' highly subcooled pool' is necessary. Perhaps this might be -

illustrated by a calculation with modest subcooling (eg 10 degrees) to show there is no threshold effect.

Chapter 8: Consideration of Reflood FCIs

1. This chapter has not been considered in any detail. The arguments presented appear persuasive provided that are no other me'ans of fast reflooding not considered by the authors and that crust formation prcceeds in the way that they envisage.

/7 Chapter 9: Conclusions V

1. I have indicated above that my principal reservations lie in the areas of the downward blockage and in ensuring that there are no operator actions that may prejudice the assumptiens made in the analysis. I agree with the authors that consideration of additional pathwr.ys is unlikely to change the conclusion.

2. For this application, the supporting analysis ought to concentrate on the melt relocation scenario. This would Mclude obtaining a better understanding of melt

! relocation in TMI-2 (eg why did it occur after reflooding the vessel?), to demonstrate that the processes are indeed understood. -

3. It would have been useful to have an indication of the effects of uncertainty in the constitutive laws (eg microinteractions) to determine where confirmatory studies are required.

Comments on DOF/ID-10503: Propagation of Steam Explosions: ESPROSE.m Verification Studies Only a limited time was available to review this supporting document.

\

E-119

Much of the document is concemed with the ability of the ESPROSE.m code to represent the wave dynamics correctly for single and two phase regions in one and two dimensions.

The information presented, along with the comparisons with the SIGhM experLnents

[ with a voided expansion region, indicate that this part of the code is doing its job correctly, even when relatively coarse (-0.01 m) meshes are used. This does not surprise me.

Numerical studies we perfonned when extending CULDESAC from .ine to two dimensions indicated good capabilities to capture the wave dynamics with relatively simple numerical schemes (the numerics of propagation are simpler than those of pre-mixing). I am therefore satisfied with the code's capabilities in this area and would expect l that the 3-D version of the code would also perform satisfactorily in this respect.

While Chaptei 2.1 uses a homogeneou:, model for the two. phase behaviour (by forcing large drag between the phases), it is unclear whether the calculations reported in Chapter 2.2 still use this model. If not, it would be interesting to compare how much better the full model performs against the experimental data, compared with the homogeneous model.

The authors of ESPROSE.m have implemented an, at the time, novel approach to cover lack of thermal equilibrium in the coolant during the propagation. This approach is physically based and can be considered to be well-justifi'ed.

l The application of the ESPROSE.m code to steam explosiens depends on the assumed constitutive physics. As Appendix D (particularly figures D8 and D9) illustrates, the l

cssumed parametcrs of the microinteraction model can have a major impact on the prediction (eg changing the parameter for coolant entrainment can change the C-J h

pressures by two orders of magnitude). Appendix C provides results from a series of experiments with one high temperature simulant, that has been used to modify a i hydrodynamic fragmentation model to take account of thermal effects. This approach is

ccceptable, but the range of uncertainty in the model parameters needs to be allowed fcr l in any assessment.

l The authors note that 'the main need identified is for constitutive laws for microinteractions with reactor materials' [ Abstract] -I agree. They also claim that

' reasonably conservative assessments are possible' - however the main report does not indicate what parameters were used to oba.in a sufficiently conservative assessment.

l I would have expected to see more discussion of the comparison with KROTOS experiments in the report as origmally supplied, rather than a reference back to the study.

Although there are some limitations on knowledge of the initial conditions and, in most of the tests with explosions, some loss of data, these provide the greatest confidence in the cpplication of any model to the steam explosion propagation phase. The calculations for KROTOS-38 provided as a supplement are useful. With current knowledge it is more E-120

important to be able to demonstrate conservatism in the calculations rather than good m agreement through parameter adjustment. Recently I saw calculations with TEXAS-IV for this test, where a very different melt distribution was calculated that le'd to very good agreement with the observed pressurisation following the trigger. Until there is a visual record of such tests it is not possible to determine which simualtion is doser to reality. .

In reading the material, I noted a number of examples where detail was not clear to me.

These are listed here for convenience, but have no impact on my overall assessment of the methodology:

1. In chapter 2.1 what value is used for Pi? Figure 6 a implies 100 bar, but elsewhere finite results are given when Pais only 10 bar.

2. In figs 7 and 8 of Chapter 2.1 a is shown as varying. I assume a is a void fraction -

of which region?

3. Chapter 2.1 precents results with and without phase change of the gas. It would have been instructive to see a direct comparison to illustrate the importance, or otherwise, of the phase change on wave propagation.,

4. I had difficulty understanding the location of the pre-voided region discussed in Chapter 2.2. Note that Fig 7 is incorrectly referred to as Fig 8 in the text. If for Fig 3 the pre-voided region stretches to the base of the tube, I do not understand the respective difference in timings of (1) the time between the shock arriving at PT3 after PT1 and (2) the time between the shock arriving at Pr3 and its reflection from the base arriving at Pr3. Note that you have offsti the pressures in the figures for ease of presentation, i

l V E-121

L0

% 4L , TL.

We%%:rMddi(GGM25%Vs5MGiW%iBh?GiVJf?k'GhW2OlB From: Brian Turtand Sent: Friday, November 29,1996 9:59 AM To Reactor Engineering SuNect: Review of DOE /ID 10541 Ireese Walt Deltrich .

Reactor EnginecGg Argonne National Laboratory

Dear Watt,

When i sent my formal comments on DOE /ID-10541, I promised to send some additional comments on the supporting document concemed with verifiaction of the pre mixing code, PM-ALPHA. These additional set of comments are attached c.s a MS WOR 06 file.

I had used my full a! location of effort for the review in preparing the original set of comments, including a preliminary read-through of the PM ALPHA report. Re-reading the PM ALPHA report,some investigative work, and preparing the written comments has taken an additional 8 hours9.259259e-5 days <br />0.00222 hours <br />1.322751e-5 weeks <br />3.044e-6 months <br />. I am prepared to cover this in my own time, but it would clearty be prefetable for myself and my orgnisation if I could be paid for the work. I believe from Stephen Sorrell that some additional fundign may be available.

The additional time spent has arisen In part frorP the volume of the mater!d supplied for review, the fact that different reports came at different times, and the time I spent uncovering some inconsistencies in the structural anatysis (communicated to UCSB) earlier.

Thank you, again, for asking mo to perform this review, Best regards, Brian Turiand Phone +441305 203029 Fax +441305 202508 e mail: brian.turland@aeat.co.ux E-122 h

Additional Comments on DOE /ID-10504 '

Premixing Of Steam Explosions: PM-ALPHA Verification Studies by T G Theofanous. W W Yuen and 8 Angelini Reviewer: B D Turland, AEA Technology plc Date of review: 29 November 1996 INTRODUCTION This document represents the cuhnination of a substantial piece of work to develop a mixing code for steam explosion studies and to validate it against the experimental data. The report makes good use of the (still rather limited) experimental data available for this purpose. The report con:entrates on the presentation of results rather than their evaluation. It would benefit from a leading chapter on the philosophy of the verification /nlidation process, accompanied by a matrix indicating which of the code's models are tested, and to what extent, by the comparisons reported. It would further benefit from a longer concluding chapter that draws together the results in the context of this matrix.

It is noticeable that efforts are made to compare isothermal particle-water predictions with accepted correlations. There ought to be scope to include similar matedal on two phase flow in the absence of particles; this is probably moa. important in establishing the reliability of the code to predict voiding behaviour.

While there are many detailed comments below, these should not detract from the achievement of the authors. The comparisons performed indicate that the code has the ability to make reasonable predictions for reactor conditions.

i However, the results should still be used cautiously, as the data currently do not exist to provide full validation of the model.

SPECIFIC COMMENTS Chapter 2 Single narticle settling While tracking a representative particle in a Lagrangian fashion gives the expected analytic result, melt mass is usually tracked through the volume fraction. This can be much more diffusive.

l I

O L123 AEA Technology 1 1

Settlina of oarticle clouds I have tried to check the consistency between the drag law for particles given by equations 3.14,3.21 and 3.22 of Appendix A wph the ddft flux formulation, but have been unsuccessful. There appear to be inconsistencies between equation 2.4, and Figurce 2 and 3. Taking v., = 0.487 i m/s, gives the liquid superficial velocity for et = 0.5 as 0.093 m/s. Figure 2 )

shows this as 0.12 m/s, while Figure 3 indicates 0.19 m/s. This suggests that it is not the superficial velocity that is being plotted in Figure 3 but the flow '

velocity, which would be 0.186 m/s from equation 2.3. My evaluations of the drag coefficient given in Appendix A for this case give a relative velocity of 0.286 l m/s, or a superficial velocity of 0.143 m/s. However, PM-ALPHA has produced,  :

according to Figure 3 a value close to 0.2 m/s. My hand calculations indicate that the PM-ALPHA model is not as close to the drift flux model as implied by Figure 3.

Settling of narticle clouds The comparison with the drift flux modelis clearly important as it goes some way to establishing the reliability of the drag coefficient

modelling in PM ALPHA (although it should be noted that the particle volume fraction is unlikely to exceed 20%, where the enhanced drag due to particle-nasticle effects is not that significant). It is less clear what one is expected to leam from the matedal presented on transient analysis regarding the validity of the code's models. It would have been usefulinstead/in addition to perform the same comparison with the drift flux model for gas water interactions where the form of the drag coefficient is rather different.

Ssstjon 2.2.2: MAGICO experiments It would have been useful to have a short synopsis of the conclusions drawn about the model from the analysis of the MAGICO tests. Besides the qualitative agreement (and general quantitative agreement) on the natur of the interaction, I think the most significant finding is the prediction and measuament of large void fractions (greater than 70%)

illustrated in Figure B23). It would be useful to provide a statement on the specille code models that these observations are believed to validate (eg water-steam drag, film boiling, radiative heat transfed?).

The OUEOS Exneriments: This looks a very interesting analysis of these tests.

The presentation of results in Figure 4 etc gives an excellent way of qualitatively comparing code results and experimental observations. Perheps some comment should be made about the apparently coherent release oflarge gas / steam volumes, seen eg at 0.41 s in Figure 4; also on the water spout effect predicted at this time (this seems to provide the mass difference between Meyer's interpretation of the water fraction in the mixing region and the PM-ALPHA values). The acceptability, or otherwise of numerier.1 diffusion, is a complicated matter, because of non-linear feedbacks through the drag laws; it is very easy to underpredict the peak particle volume fraction. Figure 5 does not give units for the liquid flux. Condensation in PM-ALPHA looks too effective at later times in Figure 6 compared with the experimental image.

E-124 AEA Technology 2 h

el Chapter S Camnarison with CHYMES:1It is only fair to note that this comparison was only .

i possible by turning off sub-coohng in PM ALPHA. Much of the detail of the PMc ALPHA predictions depend on the modelling of sub-cooled boiling. He -

observation that PM-AIEHA often only produces any void somewhat behind the  :

3 particle front, whereas other codes tend to produce some voiding wherever there are hot particles can have significant implications on the initial flow of water. For instance, we did not reproduce the so-called ETHICCA effect with CHYMES. In addition CHYMES drag laws were modified for the comparison. However, the main result - water depletion is predicted by both codes (at least for low pressure systems close to saturation temperature) - is robust.

ne MIXA Erneriments: I will try to clarify the question of time origins for the -

- data The experimental report, which I have, has unequivocal timings, with an ,

origin starting at the ignition of the pyrofuse for the thermitic reaction. On this timing the melt first contacted the water at 3140 ms, the peak (measured) ,

steaming rate was at 3810 ms and the peak pressure occuned at 4215 ms. The authors have adopted a timescale (their figure 4) where the time of first melt contact is taken to be f.ero. This is the same timescale used in Figure 1 of Fletcher and Denham for the measured pressure in the gas space - so the comparison given for pressure in the top frame of Figure 6 (page 3 21) is correct.

However, the transient steaming rate figure (figure 8 of Fletcher and Denham) does not use this time basis - this is because it was derived from the CHYMES calculations with the experimental data over-plotted). There is a significant outflow of gas before the melt reaches the water surface as shown in this figure.

i This may be due to (t) preheating and expansion of the gas in the test vessel: (11) evaporation of a water film on the test vessel wall (the favoured explanation for similar observations in FARO), and/or (iii) evaporation from the water surface.

De experimental data on the middle and lower frames of Figure 6 should therefore be shifted to the left by about 0.32 s (error on this is only from my l_ reading of the graph in Fletcher and Denham- it is no more than 0.02 s). The L effect of this is to move the measured peak steaming rate ahead of the measured peak pressure. However, I now believe that the measured steaming rates become increasingly unreliable (as quoted) due to carTy over of a two phase mixture; similar behaviour has been observed in PREMIX, Unfortunately, while the experimenters noted water cany-over post test, and observed a reduction of water height in the vessel post-test of 25 mm (the measured steam would '

produce a reduction of only about 4 mm), there is no information to determine how much of this occurred because of evaporation during the heating of the water. The same comments apply to Figures 7 to 10.

j .While the PM ALPHA calculations are as good as or 6ter than any I have seen.

for MIXA-06, I am not convinced that the real behaviour in the test is being LO E-125 Mrrh w 3

captured. The most noticeable features are the radial expansion of the melt as it enters the water and the apparent lack of any visual record of droplet break up.

Both of these effects seem to bs connected with sudden expansions of the melt g

region, due to enhanced steam generation, giving much more coupling between melt and steam than accounted for in CHYMES, and, by the look ofit, in PM-ALPHA. I conjecture that droplet fragmentation is occurring dudng these rapid events. The formation of smaller paruculate then encourages another process of melt spreading. Smaller parucles are carded upwards by the central steam flow, move outwards, and fallin the pedphery, thus extending the melt envelope outwards.

There is no visual evidence of the predicted extensive voiding atound the melt region - the leading droplets appear to be falling through water - the steam generating region is large because of the spread of the melt droplets.

I agree on the sensitivity of calculations to assumpuans on break up, lias the predicted mera particle size been compared with the experimental value of about 3mm?

This section should contain discussion / conclusions on implicadons of the comparison for model validation.

IhnJMRO Exnedments: Clearly the intual melt droplet size is very uncertain, as is the spread of the melt. L 14 appears to be the test in which the melt stream was best collimated, but one cannot tell whether the stream contracted as it poured through the gas space, or underwent a mild expansion (in 1-11 the melt stream appeared to undergo a major expansion). Ifit is believed that the meltjet h

contracted (note typoi steam for stream 4 lines from end of page 3 25), ther. the sjial meshing with Ar = 5 cm is too small. The cholce of break up parameters spears arbitrary - presumably these were selected to give reasonable agreement with the expenmental data. More detailed modelling of the melt release vessel indicates that the melt exit velocity was close to 3 m/s for most of the pour; this will not be replicated by the model shown in Figure 2. I am surpdsed that a Weber number critedon did not limit the droplet size: with the CHYMES implementatica of tu: cdterion we almost always get mean parucles close to that e observed in experiments (typically 3 - 5 mm). The commen*. on the absence of significantly superheated steam in the experimental data seems to me to be spetal pleading - it might be dght, or the steam flow might be much less conces:trated on axis than predicted by PM ALPHA. giving steam closer to saturatloc condluons. It is difncult to relate the scales on the coloured contour plots in Figures 10 and 11 to the colour scale, particularly because of interpolation effects, is break up sult occurring after the particles have settled funless they have solidified)?

Again, this section should be supplemented by an evaluation of the implications for the reality of PM ALPHA predictions. I think a word of caution is necessary, E 126 AEATechnology 4

4 -

l as although PM ALPHA, with the assumpUons used, performs well against  !

~

expedmental data, it predicts a highly two dimensional configuration. i Altematively, good cenparisons against the data have also been produced with i the one dinN.asional code, TEXAS IV. Until we see the natuce of the interacUon i sone (I expect it to be between these two computational extremes) then it is not i

possible to say that one simulation is better than the other.

Chapter 4 2

(

I agree with the general comments on break up modelling. As implied in my - f comments above, backing out break up behaviour from the experimental data  !

may compensate for other errors in the modelling. As I also noted, it is unclear, t even with the visualisadon, what break up processes were occurring in MIXA: 1 '

_. suspect the processes are much more dynamic than are currently embodied in '

the models, and coupled strongly with events of enhanced steam generation (coolant t *spping?). New FARO tests with visualisauon should provide  ;

informadon on the coherency of the initial pour, besides evidence of any subsequent break up.

Lesnerical aspects Our experience is not as comforQg as that presented by the authors. I think  ;

that numerical diffusion is probably not an important issue for large scale mixing calculadons. However, it becomes important in comparisons with smaller scale O ri- t -*i* r rt#de-i t4*vi di== d= < ret "#- "c>

compadsons that we have performed (extemal to CHYMES) show that upwinding r

schemes run below the matedal Courant condluon lead to very poor predicuons of peak particle fraction, u.d thus drag. Higher order schemes have to cater for possible discontinuides at the leading edge.14grangian approaches, as used by  :

the authors for their front tracking, provides much better accuracy, both for '

velocity and peak volume fractions. I believe that cu Tent schemes in the mixing codes can be improved substantially using physically based Iagrangian limiters, '

rather than mathematf.al limiters. Fully 1.agrangian approaches have the greater  :

benefit of handling a spectrum of particle sizes. This may be the best way to treat

. jet break up and is necessary if one is going to capture the role of the smaller droplets in spreading the melt, as observed in MIXA 06.

'Ihe current presence of numerical dillbston makes the code results difficult to interpret (eg how far back is the predicted peak concentration from the melt '

leading edge in the MIXA-06 calculadons?). Our experience with more refined meshes is that numerical diffusion is indeed reduced, but the calculauons are much more prone to matability of the resulting interface: this numedcal instability probably reflects the actual instability ofinterfaces observed in  ;

experiments, i

E-127 AEATechnology 5

= - - - - - - - - - z-n ,-n-, ,. ,.,~..,,.w . , - . , - , , , . , . , , ,n.,+. , , , , . . -. .. , ,n, .,

..w-. .mv.,.. ..n ,-- ,. s-m . , , = - , . , _ ,,_,mn.e.

)

)

l l

Concluding Remarks I would expect a more detailed technical evaluation of the calculations presented, I am surprised that questions related to the radiation transport nadelling, which h'

was clearly important in the FARO sirnulation, have not been highlighted. I would have liked to see more explicit bounds on models emerge form Ole work.

PM ALPHA Models lhe details of the correlations embodied in PM ALPHA will not be reviewed in detail.

I believe the modelling approach is sound. I note that reactor geometdes may impose strongly three-dimensional flow regions, so a 3 D code is needed for detailed applications (if found to be necessary). I get the impression that the modelling philosophy falls between two stools. At places it is admitted that the model necessarily contains many simpliflcations and constitutive physics that is uncertain, but only in the field of jet break up is a parametric approach used. I would prefer a broader approach to treatment of uncertainties. ,

With sub cooling implememed in CHYMES, it is closer in concept to TRIO MC rather than PM ALPHA. (EVA should be spelled IVA).

Elsewhere we have queried the use of the drag coefficients for droplet and bubbly flow. These are derived for bubbles rising at terminal velocity in a gravitational fleid. It is found that the shape factor for the bubble causes the drag per unit mass of gas to be independent oflength scale. It is noticeable that no effect of melt droplet shape appears in the corresponding formula for drag coefficient for the melt phase (equation 3.21). A completely different form for the liquid vapour drag is used for intermediate values of void fracuon; this may give large changes in drag when the transition vold fractions are crossed, it is not evident that there are such sudden changes in flow regime in plenum geometry.

I have not had the time to consider the radiation treatment in detail: also the relevant appendices are not included in the excerpt. For dense clouds of particles, the self-absorption effect will be very important. I would like assurance that this does not allow the region to emit more radiation extemally tl ut that of a black-body covering its surface at the same temperature.

E 128 AEATechnology 6

9 Sandia National Laboratories O - ' ' ~ - " - - "

Facsimile Transmittal -

Area V/6535 Please deliver the following message of 3 page(s)

(excluding cover sheet).

Date: sc 5'ev'44 Message To: At ke o cem .v 6fs u t Mm 4 .o 6 AWA-? Anr- a-o re A v e19 Phone: 4 % ->r, +tav FAX No: esa-m-hN Verify No:

Message From: /*We ' '/~~ r Org 4 /> <

Bldg /Rm. /rrc/n*

Phone: N N - 's ' N FAX: (505) 845-3117 Verify: (505) 845-3079 Note: e+W # fF 77..

Sent and Verified:

(Opemtor) (Date & Time) h w~

E-129

' ~ ^ ~ ' ~

(XMti '9o 13:bJ  !!> b N L WBtNJ thiewo osaN tu i Revicw ofI ower livad Integrity Under in. Vessel Steam lixploilon I.nads, by T. O.

ucofanous et al., l>OP/II)-10541.

Reviewed by Michael F. Young, Sandia National 1.aboratories, Org. 6421.

This is a massive piece of work which includes, in addition to sisam explosion loads, nome arear, with which I am not particularly familiar, such as probability methodologien and plugging behavior of molten materials. I will therefore comment mostly on the atcam explosion loading.

The approach taken in this report to determine sicam explosion loads is essentielly the one that has been recommended by most if not all steam explosion researchers: use of computational models valldated against experiments to determine bounding envelopes for reactor accident scenarios. Prof. Theofanous has taken an additional step here in simplifying the probabilistic framework with his ROAAM method; I think this is entirely in keeping with the use of these type of calculations in risk assessment and rulemaking.1 believe that the work described in this report has successfully accomplished the goal of cnveloping the steam explosion loading. The usefulnces of the results in rulemaking, however, therefore depends on the c.onfidence placed in the initial conditions of the accident scenario and in the analytic tools used.

In regard to the initial conditions, it is very important to the conclusions reached in the report that the melt be introduced through the side of the reflector, and that the lower core structure and support plate be plugged. /.s I mentioned before, plugging is not my area of expertise, so 1 will not comment further other than to point out again that confidence in the initial cm ':tionn in very important, in tegerd to the analytic tools used for calculating steam cxplosion loads,I have some commenta concerning possible gaps in the casen considered and in verification of certain parameters used in the models.

First, I see that trigger timing was varied parametrically but not trigger location; I assume that the cases were triggered near the bottom of the mixture region next to the wall; although I suspect that this is probably the most nevere case,I am wondering about the consequences of other trigger locations.

Scumd,in EPROSE.m.there are three parametern that must be set from experiment: an entrainment factor, a fragmentation constant, and a thermal enhecement factor. There uppears to be some dependence on the melt material for these factors, so the lack of data with reactor materials to set these parameters concems me. I believe this was also pointed out by Theofanous et al. in the " Concluding Remarks" section of the EPROSl!.m verification report, in regard to expanding the microinteractions database to reactor materials. Ilse of parameters that had been set from experiments using reactor materials would enhance confidence in this aspect of the calculation. In regard to the E 130 g

09 4 6 '96 13:53 ID S N L ORG 6423 FAX 50G8453tt7 PAGE 3 l

1 i

l i

O microinteraction model itself, I believe ht this model is sufficiently clone to reality ht cuperimental results can be extrapolatal to ructor scalc.

i i

't hM, the lack of a stratified mixing case bothers me, or rather, the lack of data to  !

properly model this ceae. I do not doubt that what the authors say is correct: if EP AO!!E.m were run with a stratified case, I? would probably produce a very '

nonenergetic stum explosion for b reasons stated. However,if memory serves, a stratined emplosion in a foundry involving water dripping into a " car" containing molten iron produced an explosion strong enough to take out some of the walls of the plant. This incident awms in centrast to what would be predicted with EPROSE.m. altho,gh the

• melt is different (iron versus reactor material) and the water was undoubtedly subcooled.

i The incident mentioned probably involves mixing of b motified material caused by b Initiat interaction. how well does EPROSLt.rn deal with such an ofrect? Or is this the province of PM AI.PIIA7 Maybe the authors could comment on this, or maybc it indicates the need for some stratified experiments.

Fourth, on page 6-4 there is a conclusion that the sire of the impulse does not depend strongly on the sits r f *.he mixture region. I think that this is in contrast to first principles, which vskin y? gest that it would he directly proportional, ignoring other efrects like the var >yri M%ction, and to the results in Table 6.1, which indicate a strong vwlation, ignoring G,5 timing, betwoon cases Cl and C2: 90 vs.120 for p = 10, and 120 vs. 200 for p e 20. Or did i misread the sentence? Also, how do the results compare in magnitude to the impulse of the initial trigger itself?

Fifth, in the section on stratified layer of molten stect and reflood, Chapter g, there appears to be one piece missing to make the case that the scenario is impossible: the thickness of the crust is not mentioned. Specifically,is this a " thick" crust that is stable, or in it a " thin" skin that could be broken?

All in all, this appears to be a very complete piece of work.

Following are minor points and typos.

1. In the discussion of the AllAQUS model of the lower head,it is referred to as a shell model; this is somcwhat confusing at first, as shell models are normally thought of as meaning thin shells,i.e., no bending moments are supported. This could be clarlflod by calling the model a thick shell, for instanec. ,

2. On p.4 4, there is a comment about approximately 25% of the fuel remaining uncovered. Is there a reference for this?

3. On pp. 5 6 through $ 9, it is hard to compue the graphs chosen because of varying x 1 axis scales and varying times for the plotted lines. For instance, the Cl nh plots start O E 131 ,

,yy -- y- -=-rsrr-- ' 4* ~y*

• T---9* --'m-wm'"'r

~~

. <w W4% to loix w a a tv t. u m o.c4 'rs.ww.o d ,

i at 0.4 s whereas the RCl nb plots end at 0.12 s. I neo that there arc other plots with overlapping times in the Appendix, so maybe one of these would be better.

4. On p.5 10. It says only a very small fraction of the coolant is found to coexist with the wates"; ahould this be rnelt?

5. In the graph for C2 10 a on p.ll2 7, the last time is give.n as 215 s instand of 0.21 $ s.

O l

f E 132 g

-. _ _._ _ _ - _ _ _ . _ . __._.__.__._..._.m__ _ . _ _ . . _ _ . _ _ _ _ _

l l

APPENDIX F EXPERTS' COMMENTS AND AUTHORS' RESPONSES TABLE OF CONTENTS General Comments and Point by Point Responses to:

El ~Bankon, S. G. (Northwestem) . . . . . . . . . . . . . . . . . . . . . F3 E2 Berthoud, G. (CEA Grenoble) . . . . . . . . . . . . . . . . . . . . . F-11 E3 Burger, M. (U Stuttgard) . . . . . . . . . . . . . . . . . . . . . . . F-22 E4 Butler, T. (LANL) . . . - . . . . . . . . . . . . . . . . . . . . . . F-39 E5 Cho, D. H. ( ANL) . . . . . . . . . . , . . . . ' . . . . . . . . . . . F-46 E6 Corradini, M. L (U Wisconsin) . . . . . . . . . . . . . . . . . . . F-50 E7 Fauske, H. and Henry, R. E. (FAI) . . . . . . . . . . . . . . . . . . . F-60 E8 Fletcher, D. E (U Sydney) . . . . . . . . . . . . . . . . . . . . . . F-64 E9 Jacobs, H. (KfK INR) . . . . . . . . . . . . . . . . . . . . . . . . . F-76 E10 Mayinger, E (U Munchen) . . . . . . . . . . . . . . . . . . . . . . F-91 E11 Moody, E J. (GE) . . . . . . . . . . . . . . . . . . . . . . . . . . F-99 E12 Sehgal, B. R. (RIT) . . . . . . . . . . . . . . . . . . . . . . . . . F-110 E13 Shewmon,P.(OSU) . . . . . . . . . . . . . . . . . . . . . .-. .

F-137 E14 'Ibriand, B. D. (AEA) . ...,................... F-139

- E15 Young, M. E (SNL) . . ........-.............. F-161 Notti the mark * * * *

• in the authors' responses indicates agreement, or no -

comment.

F _ _ - . _ _ _ . . . ..._ . _ ._ - _.

i F.1. Resoonse to S.G. BankofMNorthwestern)

General Cornment and Highlights General and unqualified agreement with the conclusions of the work under review.

Point-by-Point Responses

0. I enc %e herewith my review of DOE /ID 10541. I had to read the support-Ing documents as urllin order to get the necessary perspective. In the process  !

I spent 32 hours3.703704e-4 days <br />0.00889 hours <br />5.291005e-5 weeks <br />1.2176e-5 months <br />, which convinced me that in vessel retention is a valid concept for the AP-600. ,

4

1. The principal documents which were read by the author were DOE /lD.

10541 (June 1996), DOE 10460 Vols.1 and 2 (July 1995), and DOE-10849 (Jan. l 1995), as well as various papers published and/or presented by Prof. Theofanous and one or more of his co-authors. My generalimpression is that this is a masshe piece of uvrk, which attacks all aspects of the steam explosion problem in the Westinghouse AP-600 reactor, and conclushely demonstrates that failure of the vessel, to say nothing of the containment, is physically unreasonable. If no failure occurs in the reactor vessel, essentially no release can occur to the inside of the O containment building, and hence the threat to the public health and safety is eliminated. In the process of denloping the evidence in terms of focused experi-ments, development of new and improved codes, tying in uvik done around the uvrid, and developing a methodology for assessing the safety goals and margins for rare, but high-consequence, hazards, a set of tools has been developed which represents a huge step forward in examining severe accidents in new types of advanced nuclear reactors and in existing nuclear reactors.

In other uvrds, in execution, scope and potential consequences, the total of this uvrk represents a very important achievement.

2. I think that the ROAMM approach makes very good sense for rare, but high-consequence, ennts. I believe that a similar approach has been used before, but never so explicitly and clearly stated. In particular, the recognition that there are " intangibles" which will never be known in advance, consenatively bounding F-3

i l

Is su&lciently reliable. The program predicts 'the major portion of it (i.e. the fuel) being in a highly volded region (a > 80%)* and also that the void fraction

' gradient is very steep', i.e. the void fraction increases from values around 20 %

to more than 80 % witbin a short distance. Such behavior, however, was not seen in the premixing experiments that are being conducted at Forschungszentrum Karlsruhe in order to study the phenomenon and to collect data for code valida-tion [2), [3], [4]. It is too early to draw final conclusions fro,n these experiments, but the void fractions in the surrounding of broken up ' fuel' appear to be smaller than expected.

The QUEOS experiments were run under conditions quite different from those of M AGICO, from which the reviewer's " expectations" may have derived. We provided quantitative interpretations of the available QUEOS tests and see that this lower voiding should,in fact, have been expected. More imporantly, to this day we are not aware of any published, reliable vold fraction maps over the premixing zone in the QUEOS experiments. We have such detailed maps in MAGICO (see Appendix B of DOE /ID 10504 and the addenda to it), and show that even with very dilute pours (0.5%) we get void fractions in the 60-70% range, extended over the whole mixing zone. The QUEOS pours are too short, and too concentrated to reveal the important thermal interactions that lead to extensive and persisting voided premixtures. g

8. One may also draw attention to data reported of the KROTOS experiments (5). In these tests molten alumina was poured through an orifice with 3 cm diameter into a 10 cm wide tube filled with water, it mixed with the water and strong steam explosions occurred either spontaneously or following an external trigger. The melt temperature was high, typically 2000 K, but the water was subcooled which, of course, tends to reduce voiding. In the KROTOS tests #28 and #29 the water was subcooled by 10 K and 80 K, respectively. In both cases ,

the steam volume fractions within the reaction tube were 4 % only. But as these are mean values over the whole tube which may contain some regions occupied by water only, it may be more relemnt to point out that the steam volume was only about half the melt volume. In test #30, subcooling was again 80 K but the melt mass was larger and its breakup was more intenshv. In this case the steam volume fraction reached 23 % but this is again only 1.3 titrn the melt volume. So we must check how well the above cited calculational results of PM-ALPHA.3D are founded which imply steam volume fractions that are larger than the melt volume fractions by well over an order of magnitude.

F-81

support plate). A great deal of work has been done (and is contemplated for the

) futuse) on Jet breakup, with various materials, Jet diameters, temperatur~s and velocitics, that is predicated on this assumption. The unexpected result that for the AP-600 melt-through will occur through the side, rather than the bottom, of the melt pool, is therefore of Brst importance. Of course, this happened at TMI, but in AP-600 there is a thick stainless-steel reBector around the wrc. The code capability allows calculation of the subsequent premixing and melt / water dynamics. There can be no large melt jet in transit; stratllication appears very likely even in transitt and the whole scenarlo of damaging multiple explosions disappears.

8. The simple models for blocimge formation and blockage coolability, leading to non-amilability of downwards relocation paths, and transition to a molten pool, are made credible because of the relatively Bat radial power distribution in the AP-600 design. This lumped approximation would have to be re. examined for other reactor designs.

9. The calculations of melt length scales and local void fractions lead to quan-titative results which are more realistic and detailed than previously available. As expected, liquid water is rapidly depleted fr.um regions of high fuel concentration, and the boundarles ofsuch regions can be quite sharp Board Hall thermodynam-ics theory for steady plane shocks, previous multiphase calculations of the Bow

• fields behind the shock front (Sharon and Bankoff) agree that regions oflarge vold content cannot sustain shock propagation at supercritical pressures. This is the principal reann that the SERG-2 panel felt that the a mode failure was not pin'sically reasonable. Precisely the same results are obtained by the. PM Alpha and ESPROSE.m calculations.

10. The residual uncertainties proposed in NUREG-1524 were jet breakup triggering, 2D vs. 3D codes, and chemical augmentation. For the AP-600, Jet breakup is no longer a major concern, as discussed above. The 2D vs. 3D controversy is no longer r+.evant, since wildated 3D codes are now available.

Chemical augmentation with the real corium produced in the reactor will have O

O F-5

no !mportant effects. higgering is the sole intangible which wul never be known for a real accident. However, it is irrelevant ifit is assumed that triggering always occurs at the worst time and place, and the result is evaluated by energetics, which as to say the validated codes. The approach taken of triggering by setting one mesh to a big , initial pressure seems to me to be a perfectly valid procedure.

11. Convincing arguments have been addressed, backed up by a huge volume of high quality experimental, analytical and computational work, that the AP-600 nactor will nct fallin the course of a severe accident. This implies that all later scenarios of containment building pressurization and heat up are no longer necessary. In my opinicn, this closes the severe accident scenario for the AP-600, and leads to consideration for licensing. The consideration of other reactor types, on the other hand, does not appear to be so straightforward, and further work needs to bn done.

12. 1. DOE /ID 10503

• Propagation of Steam Explosion: ESPROSE.m Ver-incation Studies", by T. G. Theofanous, et al. Aug. 2006, and update of Sect.

4.2.1 This is a convincing document, laying out the evidence that ESPROSE.m has the capability successfully to predict various shock and vapor explosion scenar-los. These range from simple steady, one dimensional shocks propagating thvough single. phase liquid and homogeneous gas-liquid mixtures, for which exact (or nearly exact) solutions can be found, to experimental shock and explosion data in the SIGMA and KROTOS faculties. The 1 D wave dynamics were tested for shock speed, Buld velocity, reRection at a rigid wall and reBection at a free interface with venting for single-phase, liquid-air and liquid-vapor cases, using ESPROSE.m and CHAT. The small deviations between the analytical solutions and the codes can be attributed, at least in part, to the fact that the analyti-cal solutions used an assumed constant sound speed in the liquid, taken as 1500 wls and 2000 m/s, while ESPROSE.m used the real properties of water. In some cases, as in the deviation in reDected shock speed, the differences in the analytical predictions for the two assumed sound speeds is considerable, but ESPROSE.m gives a smooth function of shock pressu~ which intcrpolates between the two lim-its, and is hence more credible than either of the two analytical shock reRected F6

shock speeds. As a check on the wave dynamics with reflection / transmission at interfaces beturen tuo materials of different acoustic impedance, which governs the unloading-explosion coupling near a free interface, the CilAT code, using the l method of characteristics, was writ ten. For large amplit udes the quasi linear code CHAT.QL cvaluates the coeRicients in terms of the local Buld properties. These  ;

codes were then compared with ESPROSE.m for pressure and velocity distri-bution for 1.D single-phase venting and for shock speed, Buld velocity, reflected  !

shock speed and shock amplitude for a 10% vold non cont lensible steam / water mixture, with. excellent results. This sort of independent cross-chec. king lends considerable conndence in the basic structure of the ESPROSE.m code, which is a Bnite dliference code in laboratory coordinates. Exact solutions in 2D geome-tries are then ghen for infinite pool, cylindrical open pool and cylindrical closed pool geometries, by superposition of an inBnite array of sources and sinks in or-der to obtrin reBection and transmission behavior at rigid and 'ree boundaries.

These solutions are in themsehrs impresshv, and likewise the general agreement beturen the code and analytical predictions for pressure as a function of time over the two-dimensional region. In fact, the agreement between the two predic-tions for the centerline pressure distribution as a function of time and distance is p remarkable.

b There is also good agreement between the ESPROSE.m prediction and the exper-imental data in the SIGMA facility with all liquid, and with a liquid air mixture in the expansion section. The key parameters of time of arrival and amplitude of pressure waves at severallocations are util-predicted, particularly in view of some high frequency ringing in the experimental transducers. The same is true

, for multi-region runs with high pressure (hP = G8 and 136 bars) with different initial void fractions.

13. One aspect of vapor explosions which b dlBicult to model properly is the presence of strong energy sources, especially near a free boundary, which are caused by local explosions (rapid mixing) of fuel drops produced in the course ofjet breakup. These sources can distort venting and reRection phenomena near free boundaries. This was modeled first by characteristics solutions with single internal heat sources, with the energy assumed to be going only into the vapor, a with heating rate increasing with velocity:

N V F-7 l

i 4

Q, a Cu' * (3.2) with C being an empirical constant. More explanation is needed for the assumed form oI this equation and the magnitude oi the exponent, based on energy dissi.

pation. llowever, excellent agreement is obtained for various assumed values of C between CilAT and ESPROSE.m.110 wever, comparisons with data for single exploding drops are lacking.

Here we addressed a potential numericalissue, and the form of Eq. (3.2)is immaterial. All we wanted was to create a source which increases in intensity with velocity, so we pick a rather strong dependence on u to the 1.5 power. Through the coefficient c, we can further adjust the " base level" of the intensity of such a source. We then show how one can have different degrees of interaction between such a source and venting, and we show also that ESPROSE.m performs quite well against this rather major computational challenge. We do not see any need for special experiments in this area.

14. A t the ather extreme is a planc shock wave moving into a fuel /stcam/ water mixture at low pressure. This is the scenario envisaged by Board and 11all, fol-lowing the one-dimensional theories of Lifshitz and Zeldovich. In the B-H theory the average specific volume of the mixture is plotted against pressure, starting with adiabatic compression to a peak pressure (von Neumann point) at which the steady mass-continuity condition for the flow behind the shock front is satisfied with minimum entropy generation. This results in the Chapman Jouguet (C-J) condition of tangency to the reaction adiabat, or Hugonlot curve, leading to a minimum of shock speed with pressure. The ESPROSD.m calculation shows the development of the shock into a steady state detonation, using n fragmentation rate given by Fr = p} h (4.1) where p} is the local macroscopic density, C is the " expected" shock speed (1500 m/s), and ax is the grid size. Why the fragmentation rate should be a function of bx is not clear, nor is the form of this equation. More explana-tion is needed. However, detailed comparisons between 2D and 3D versions of ESPROSE.m are encouraging, and the ESPROSE.m P-V lines are close to the expected Rayleigh line, and bounded by the shock adiabat.

There is nothing special in Eq. (4.1), except that we wanted to be sure the fragmentation rate is fast enough, so as to be comparable with the " infinite" rate assumed in the steady F-8

l l

7x detonation theory. The form of Eq. (4.1) assures that fragmentation occurs within the I time it takes the wave to traverse one computational cell. We have since that time done calculations with very high, constant, fragmentation rates also. By improving our steady state theory (see Appendix D) to account for finite compression trajectory effects on the Hugoniot, the agreement now is even better than before.

!5. These result 8 are excellent back up for the !nterpretatIons oithe KROTOS experiments ghrn in DOE /ID 10489. All in all, even at this stage of limited comparison with integral explosion tests, one has confidence in the prediction of pressure vs thne at ntrious locations on the pool boundaries, and consequently of the initial kinetic energy oflarge masses impacting on reactor structure.

16. Typos:

p. 1 1 verifled Fig.1 viscosity 4-3 und if flugonlot von Neumann C-J point V(3 Iypos corrected.

17. 2. DOE /iD-10504 " Premixing of Steam Explosions: PM. ALPHA Verin.

cation Studies", Sept.1996 This is, once again, a thorough, and high professional, document on the verin.

cation of the PM ALPHA code against available experknental data and known physics. The agreement with a wide range of data, from single particles settling in water, to particle swarms, both cold and hot (up to 200TC), to integral tests ,

with prototypic materials at high pressure, is rather remarkable. The breakup constant 0, has been chosen to fit the futegral data from several tests, but it is used consistently. The Richardson Zaki exponent for a monodisperse system of spherical particles has been used without modification for thermal effects, with excellent results. The FARO experiments, which gave very little usable data on the Jet breakup and dispersion, has been well approximated for the measured steam flow and pressure. The owrall result is that the code seems to be well suited for licensing purposes.

f)

F-9

18. llowever, some speellic comments may be made and questions raised:

1. Eq. 2.3. This equation is incorporated into the code, but no independent check on the R Z exponent is made. On the other hand, the R Z exponent was chosen by comparison with a 1:.rge body of data on systems of spherical particles.

2. Figs 9 and 10 (p. 2-10) are interesting in showing decaying oscillations, and an atttactor above, but close to, the stcady driit. flux / particle volume fraction curve. This physics appears to be new, and should be further investigated.

3. p. 2-16. The careful treattnent of the radiation boundary condition with slight subcooling is noteworthy.

4. Figs. 8 and 9 (pp. 2 30/2.41). The compar! sons between the predictions for front position and the data for the QS - Q11 experiments in the QEOS series is remarkable. The level swellis not well predicted in the first 0.1 s, but this may be due to experimental uncertainty. For the important range t > 0.2s the agreement is excellent.

5. The explanation for the absence of a pressure hump at early times compared to measurements for Q17, as being due to the extra radiant heating before impact, seems to me to be reasonable.

6. Turning to the hilXA experknents, there is reasonabic agreement with the pressure data, and excellent agreement for the cumulative steam flow.

7. Shnllarly, there is remarkable agreement with FARO L-14 water level swell, pressure, and pressurization rate, especially considering the complex geom-etry of the equipment. The additional information from the code on local mid fraction, melt temperature, melt volumetric fraction and melt location seems to me to be very useful, in view of the inherent limitations of the ex-periment. hIy own view is that the cost-benefit rat 10 for further experiments of this stit is large.

F-10

F.2. Response to G Berthoud (CEA Grenoble)

,q V General Comment and Highlights The review is generally agreeable, but reluctant to fully accept that a bottom relocation path is physically unreasonable. This is a key point of our evaluation, and we provida responses to the reviewer's specific questions. We expect these, plus our responses to similar questions from other reviewers, to be helpfulin funher focusing the issues towards resolution.

Point by-Point Responses

1. This document presents an analysis of the potent!ality oflower head failure of the APG00 resulting from a Steam Explosion. '. .e conclusion that the risk is negligibir (< physically unreasonable :>) is quite convincing, and is based on:

1. the fact that water uill be saturated and at 1 bar due to complete depres-surlsation to the cont.dnment pressure and that these conditions willlead to large and rapid voiding which is not favorable for large S. E.

2. the fact that we have permanent blockages at the bottom of the core that will Impeed any coherent relocation through the core support plate T

{& 3. the fact that relocation will occur sideways through the reRector and core barrel and so that the Steam Explosion will occur in a 3D geometry without any large constraint allowing large sustained pressure

4. the fact that - even if reBooding is taken into account - when the melt will be ejected sideways, we will have enough time to heat the added water up ta saturation and so ta prevent good mixing.

The validity of the conclusion is then linked to the validity of the slave four arguments.

As for the Brst argument, there is no doubt that water will be saturated as far as reBooding is no*. taken into account, the fact that the picssure will be atmospheric cannot be discussed here as this is justined in another report (IVR Report - table 7.3) however, I think that this has to be lustined as the voiding will be less important at pressures a little bit higher, around some bars. At these pressures, we can also recall the it was found it was easier to trigger an explosion in the single Iron oxide droplet experiments of Nelson in Sandia.

C'N F-ll

The fact what we have up to now no evidence of explosion in experiments using reactor like matcrials (Krotos) (and that this is c'ue to the non occurrence of good mixing) is stressed by the authors. But once again these Krotos experiments ase performed at 1 bar pressure while in Faro experiments at pressures of 50 and 20 bars, with saturated water some mixing was obtained. In a near future a Faro experiment using Initially saturated water at 5 bar will be performed and we will then have an Indication of the quality of mixing at small pressuse.

WAttn suscoouwo tK) 70 87 99 130 148 108103 11 1

I.

l. .

1.s ,

1.4 .

s -

y 1.s rXnossowS 3 l1.0 U- -l: , .

0.8 -.,8-------*--***>o-~~-

0.4 N

e , j

/ -

l . / NO EXK0SiONS 0-i---~!

0.2 .6 o

• -o ~ i~o '- ~ i i ei i

,,,0.0 0.2 0.4 0.4 0.4 1.0 IJ 1.4 AMBENT PRESSURE (MPa)

Figure 8. Relation entre le " trigger" decessaire au dbclenchement necienchement d'une interaction et la pression ambiante (cas de gouttes d'oxfde de fer de 2, 9 mm de diambtre a 223(PK tombant dans de l' eau a 29s*K).

Because of the passive design of the AP600 the pressures applicable to the severe accident management window are in the range 1.1 to 1.7 bar. This is too narrow a range to bring in significant pressure effects, and for completeness, we have run some calculations and satisfied ourselves that indeed this is so (see addendum to Chapter 5).

2. I will now go thwugh the different chapters trying to analyze the justiBca-tions which are presented to support the crucial arguments mentioned previously.

F 12

Chapter 2: Problem definition and over all approach

\s' e it is mentioned that it is only recently that pressures in the kbar range were observed experlinentally in constrained one dimensional geomet"y. However, I think that a pressure peak of the order of the kbar amplitude and millisecond duration was measured in the Sandla FITS RC2 experiment which was well vented (initially open at the top and later vented at the bottom as the vesselleft the giound). But this was obtained usingiron alutnina thermite and subcooled water.

We are aware of the experiment, and its energetic natures (from movies), but we do not have indications in the literature of reliable pressure measurements.

3. e Another important argutnent is that < because of extensive widing, we need only be concerned a' ut the first relocation etrnt, and only for early trigger in it >. This seems to be justlRed by the premixing and explosion calculations presented later but I wondet why, after a Brat event, when water is sh>shing back n second event cannot occur at about the same location where the structure wil:

has been aircady dynatnically loaded and eventually already deformed by the first event, n As seen in Chapter 7, the structure response remains within the clastic limits. Subsequent U relocations in the scenario proposed by the reviewer would be too gradual, and water would be quickly depleted from the lower plenum.

4. Chapter 3: Structural fallute criteria in this chapter, it is stated that < the titne-duration of the loads ofinterest here is less than the structural frequency >, so it is expected that < peak strain would be basically independ of the details of the pressure pulse shape >. flowever, nothing is said about the estimated vnlue of the natural frequency of the R.P.V.

which sectus to me to be of the order of magnitude of some msec 50 not so far from the load duration.

Ilowe*:cr, all the analysis is made with the analytical solution of Dulley and Mitchell which assumes < short pressure pulse > and allows to evaluate the plastic equivalent strain with incorporation of strain rate effect by formula (3.2).

But the comparison of the analytical results to ADAQUS calculations shows that the analytical relation gives conservaths results for the plastic strain evaluation (fig. 3.2).

O C) F 13

i

)

No, all the analysis is not made with the analytical solution; it is nude with ABAQUS.

The analytical solution is used to obtain insights helpful to the task of coming up with the g lower head fragility.

5. Another mitigating factor is imvstigated: the effect of load localization which shows that for a given impulse, the equhalent plastic strain is smaller when the loading is smaller. Use of these results is then made by assuming that a fraction 0 of the impulse is used for bending energy so that only an <elfccthc impulse > ls applied for the evaluation of the equhalent plastic strain. We are told that 6 is a material and geometric < constant > but I have not found any indication of its evaluation i.e how the results shown on fig. 3.8 and fig. 3.9 are obtained. As flg. 3.9 is used to evaluate the oficct ofloads calculated in Chapter 6, I think that it should be a little more explained.

The B is chosen so as to fit the " data" (i.e., the results of ABAQUS simulations). The value is 0.05. This was added on Figure 3.8.

6. It also seems to me that the localized loadings are applied on the axis of the hemisphere (see table 3.2). Does the fact t! these localized loadings will occur on the side of the hemisphere with singularity where the sphere is linked to the cylinder will modify the conclusions we can draw from flg. 3.9.

The cylinder juncture to the lower head is quite a bit higher than where the water level is, and the further below location where the premixture develops. Moreover, in Chapter 6, we have full vessel simulations (see Eq. (6.2)).

7. N.B. There are some errors in table 3.2 as for the value of Ao which does not correspond to & as written in the captlon.

Typo corrected.

8. Chapter 4: Quantification of melt relocation characteristics This is a very important task as most of the boundary conditions for premixin s and exploslors calculations are obtained from such an evaduation.

e The dowmvard relocation path (arguments 2) is not envisaged: < we expect this path to be blocked by molten cladding and the blockage be robust >. This exl;ert judgment is supported by the large heat sink associated with the large amount of < cold > materials in the lower part of the core. As it is said that F-14

due to be big stalnkss steel reBector, the Brat relocation wul be delayed compared

(

G to what occurred in TIM and that at this time, we wHl have a large oxide pool, it is important to know if this mtiten pool will reach the region of the lower Rssion gas plenum where the heat sink is not very large and where we can have a breakdown of the supporting material. However, in the paper, the blockage is said to occur in the region of the 7 cm < lower 2r plug and lourrmost spacer grid >.

Some calculations are presented to show that the plugging time of this rcgion by melt 'b negligible superheat is of the order of seconds. For this calculation I have some troubic with formula 4.2 where, as for me, X is not the same as in the Carslaw and Jaegger text book but i did not try to perform the calculation. We can also make another remark: if we have some breaking down of the Rssion gas plenum region, when the molten pool arrives we may have superheated molten innterial from this pool that with Row in the lower blockage region for some times before plugging it would be interesting to know what amount of molten materials can be transfered in the lower plenum through the holes !a the core support plate before plugging of the passages in the blockage region. As for this plugging thne

- which is crucial to support arguinent 2 - it would be interesting to see more realistic calculations including the inBuence of the interface thermal resistance p between the crust and the solid waH that wlH slow down the freezing process and O then increase the plugging time.

See addendum to Chapter 4.

9.

• As for the blockage coolabHity:

- the stable blockage thickness should be sensible to the radiation factor fr which is set to 0.7 without any explanations

- the cooling of this blockage is ensured for about 100 min which is the time re-quired to vaporize the water which Rils all the volume between bottom of acth'e core and bottom of core support plate. It is later esthnated that meltthrough of the reBector by the molten oxidic pool wlH occur between 76 and 91 min accord-ing to the amount of oxidation (80 to 95 min in the calculation < without >

preheating).1(we add the time require to melt thmugh the core barrel, we get timing of the release of the same order of magnitude than the insurrance of block-age coolability. As all thase calculations are order of magnit ude ones, I think that argument 2 (no downuvi relocation path) may be questfored.

g) k F 15

All relevant factors were evaluated with some conservative blas, and these are not order of magnitude calculations. Rather, we would characterize them as basic-principles based conservative estimates, clear and supportable in every respect. So, to dispute the quanti.

h tative result, one would need to be specific about which input or aspect of the calculation is being questioned. In the report, we also emphasized that beyond the 100 min, the heat capacity of the core support plate provides further margins to failure-this is a key point.

In any case, some further elaboration on uncertainties is provided in an addendum to Chapter 4. In this addendum, an estimation of the additional margin, due to the thermal inertia of the core support plate,is also included.

10. e hfolten pool formation In the inillal heat up calculation, are the reRector and core barrel in contact everywhere as it is shown fig. 4.8 and 4.97 in that case the cooling cifect will be overestimated and the melt superheat underestimated.

No. The melt superheat is only controlled by the pool convection processes. The crust thickness adjusts itself only to the losses, and this is not important.

11. - During the transient heat up calculation, what happens to the molten cladding and how the calculatIan with the elfectIve thermal conductivity is even-tually modified?

The effective conductivity is not modified. The anoxidized cladding is assumed to drain.

12. e biolten pool calculation Such a calculation is performed for the oxidic and metallic pool, and there is a crucial hypothesis which is the presence of a stable oxidic crust at the upper surface of the oxidic pool. In the document, it is mentioned that it is assumed that the clad drains but is it fully true? Cannot we have some metals included in the moving down oxidic pool? What happen to the part of fission products which are released at fuel melting? Will they modify the molten pool behaviour for the stability of the upper crust and the evaluation of the differents Buxes?

The stability of the upper crust is not important here. If it is distorted and broken by some vaporization from below,it will for n again right away. We know that the metallayer will

" cat through" the reflector rather quickly, and the exact timing of it does not make any difference. Certainly it makes no significant difference to the treatment of the oxidic pool either.

F-16

13.

• hielt through and melt release calculations C\

U lt is said that rapidly the metaille pool will snelt through the reRector but it is assumed that the metal < will be gradually draining :s> Into the space between the Bats of reRector and the core barrel Cannot un han some kind of metallic jet impacting on the core barrel with some rapid nieltthrough leading to a steam explosion between inctals and water?

$ meWe mektke

't

[ P"I

/

/, g. . n .^ whe sud W* )' Y ns{\<W No. Because of the highly erosive property of the metalit could not accumulate to signifi-cant depths to produce the stream needed to penetrate the " cold" core barrel. The process will be more like a gradual overflow.

14.

• From the above analysis it is concluded that when the oxidic melts through the reRector, there is no metal on it and that failure of the core barrel occurs soon after. First, it would be interesting to evaluate the time required for core barrel meltthrough (if there is an open space between the two of them).

But there is another problem if the spa e between the Bats of the reRector and the core barrelis already Riled with the metals from the metallic layer, how the oxidic pool can rapidly go through the core baric 1. This situation may be a promoter for downward relocation 1i this added metal may increase the time for meltthrough.

As noted on p.4-23, the total capacity of the spaces between the reflector and the core barrel is ~10 tons of steel, and this corresponds to 50% of the reflector corresponding to the pool height. The oxidic pool will fall the reflector and the core barrel well before such an extensive melting of the reflector; that is, before the spaces are completely filled with metals. If this went not the case, the failure of the core barrel would occur at one of the comers of the reflector, making the failure more localized and hence producing a more bemgn pour in any case, there can be no significant enough delays in melting the core barrel as compared to the times necessary to lose lower blockage integrity.

F-17

15.

• As for the location and size of the failure, most of the information is obtained from expert judgment and should be further Justified:

- the failure < is expccted > to be local azimuthally and very near the top of the axidic pool. I would agree with that statement as even, if the calculation is 2D cylindrical once a Bat will fall, the rapid relocati:n willimpede failure on other flats. But I would not be able to give any probability for 2 quasi simultaneous failure, or 3....

- for the size of the breach, it is said that 0.4 m x would appear geometrically a good upper bound on the first breach width and that a < 10 cm axial gap is believed to be conservathw >.

- there is no mention of the rapid increase of the size of the breach during the melt release as it has been observed in experiments. Ilowever, as only short duration premixing triggering scenario are taken to be ofinterest, this enlargment would nat be important. But, if we take into account steam explosion occuring when water is sloshing back after a first event, this has to be taken into account.

See response to previous question about the postulated " series" of steam explosions.

16. Chapter 5: Quantification of Premixtures Given the melt release conditions (flow rate, location, temperature and compo-sition), the premixtures are calculated with an improved version of PM ALPilA which is now 3D and includes a melt fragmentation law (which was lacking up to now) as it is recognized that it is interesting < to know the distributions of the melt length sca'e >. However, this fragmentation law is not described and this should be done and justified as fragmentation is responsible for voiding (< the rate of voiding lucreasing rapidly with the rate of breakup >). I would also like to know why the < breaking law is operathc only for as long as the coolant has a void fraction ofless than 50% >. If the fragmentation was always operative, voiding would be larger so there must be a good reason for doing so but I do not see why.

The breakup (not fragmentation) law is described in the PM ALPHA verification study, which was also supplied to the reviewers. There is a void fraction limit (conservatively set to 50%) to express the fact that, breakup occurs significantly due to the presence of water-inertia effects and melt water thermal interaction effects. See, for example, our interpretation of MIXA and FARO data.

F 18

li. The melt entrance conditions into water are also speeltied and not calcu.

O lated: l

- entrance velocity whose evaluation is correct  ;

a distribution of the melt < over an effecthe radial width of10 cm :> with a melt volume fraction evaluated to get the correct mass flow rate. This distribution is  :

crucialin determining the amount et upour which is produced as the larger the

- entrance area, the longer you are in the film bolling regime in widch the steam  ;

production is at maxhnum. This beharlour was observed in MC3D recalculations

- of FARO tests, where a doubling of the pressure increau (linked with vapour pro- j

' duction) was obtained with a doubling of the diameter of the melt flow. Recently l CHYMES 2 recalculations showed the same trend. l Steaming rate should not be confused with voiding of a premixture. Here we are prin-clpally interested in the extent of voiding, not steaming rates. Moreover, independently of the specified radial width, the melt will quickly disperse to cover the relatively narrow '  ;

downcomer width.

18. . Initial droplet diameter which is set to 20 mm (a large enough value to l represent a minhnally broken up melt stream). This parameter is also important >

O for vapour production. It would be Interesting to see sensitivity calculations with l diameters varying from 10 to 1 cm.

The initial melt length scale is absolutely unimportant, as long as it is large enough to represent a minimally broken melt stream as we expect to be the case here in the travel through the vapor space. We have demonstrated that liy the 2 cm no break'up case, which is less volded, but still rather benign, a larger length scale will still be more benign. Now, due to the breakup, in the advanced portions of the jet, the vapor, like a chimney, also ,

would cover the less brc, ken portions and give still more benign exploslor

19. I am not so sure that the melt will be transformed in a droplet population ,

before entering the water. We may have a large melt stream on the wall with subsequent fragmentation into the water but with a different law than droplet fragmentation. Would it lead to a < benign evolution ;> as it is rnentionned.

This is again an expert judgment.

B O'.- F-19 B

,y --u,, , _,-- , - ,- .~,. - , - .v..,.. - = . - . , ..... 2-.--

l This concern would be applicable if we used a mechanistic law for droolet breakup, but i for thir, and other reasons we don't. Rather, we vary parametrically the law, and envelop the behavior regarding extent of voiding as related to the law (and extent) of breakup. h) 1

20. As for jet fragtnentation calculations with THIRhiAL, I cannot trust them l! the fragmentation is still governed by Kelvin Hehnholtz type calcula-tions. hforcover, in FARO experknents whh 10 cm melt jet, it took more than 2 in of water to break thejets in a 50 bar atmosphere fo; which volding is smaller.

At 50 bar the voiding is smaller and the steam velocities generated are also smaller (than at 1 bar), and this is why in FARO we expect less breakup. Still, however, the premixture was highly volded, and the melt extensively fragmented according to our interpretations of the tests. Ours are fully consistent calculations using the proper non local radiation absorption law. The THIRMAL calculations are offered as another perspective, still parametric, not mechanistic. We do not believe anyone can do mechanistic calculations that can be trusted; moreover, we do not believe such would be a good approach even for the future (it would run into serious verification problems due to inherent problems with directly quantifying this aspect in experiments). We do not know where the reviewer's assertion about the FARO tests derives from.

21. Chapter 6: Quantincation of explosion loads Nothing is said about the parameters used in ESPROSE-m 3D but as the trigger uses a 100 bar steam release, we may think that the hydrodynamics fragmenta-tion law will be correct Due to the small amount of melt imvhed in explosion calculatlons, there is no problem with the energetics oithe esplosion and we are only interest.cs k' dynamicalloadings of the RPV. This is done by the esthnation of the knpuht and of the local area of ladings from ESPROSE-m results. I have some problems to understand how h is esthnated page 6.3 from the area ewlution as shown on Bg. 6.5.a.

In the text, it is said that peak impulses are around 0.1 and 0.2 h!Pa.s with 3

enecthv area around 0.1 m (which gi vs b ~ 0.15 ) and from Bg 6.5 c where I Bnd a 0.2 hipa.s impulse, I do not understand how I get Ao ~ 0.1 m3 from the area cwiution which is shown.

We take the area over which the main impulse is delivered. For example, in Figure 6.5(a),

this occurs in the time interval from 0.3 to ~0.5 ms. Over this time interval the area is then seen to be ~0.1 m2 . Then using the equation on p. 6-3 we find d,/D, ~ 0.16. These F-20

)

estimates of areas and effective times are presented in a more refined manner in the new (j calculations.

22. Apart from this problem, ifI accept the Bgures mentionned in the text, I agree when coming back Rgure 3.9 that there is no risk for i ~ 0.2 and ft, ~ 0.15 as ifG ~ 0 This is coniltmed by the ABAQUS calculations of the tsvo most energetic explosion calculations obtained by PM ALPHA plus ESPROSE m.

23. As a conclusion, I can say that if we accept the scenario which is retained by the authors, I think that - whatever my remarks about premixing quantifica-tions - the AP 600 RPV cannot be challenged by a steam explosion. However, I would like to have more established confirmations of this scenario by mechanistic calculations when possible or parametric calculations when it is not.

The main thing to be conRrmed is the impossibility to have a downward relocation i.e the possibility to have a break down of the lower Rssion gas plenum rather than a continuous <* alning. This will give a sudden access of the core support plate holes to the melt

- the innuence of the already relocated metallic pool on the oxidic release. It may take a longer time to break through and the blockage integrity may then be challenged. The I:.stuence of an interfacial resistance between the oxidic solid crust and the wall- specially at the top of the pool- will also participate to an increase of the time of break through and of the evaluation of melt superheat.

Other branches for the scenario should also be evaluated:

- the possibility of the metallic melt to rapidly go through the core barrelleading to metal water steam explosions the possibility to have steam explosion at later times (in the oxidic case) wie water is sloshing back after a Brst small scale (i.e low energetic) event.

As for reRooding scenarlo, the fact that water will be closed to saturation should also be evaluated.

This is a summary of points made and responded to point by-point.

O D F-21

F.3. Enponse to M. Burger (U. Stuttgart)

General Comment and Highlights Many specific issues are raised in this revie w, about almost every aspect of the analysis and supporting documentation. However,it is also stated that in an overallivay the analysis is convincing and that the spirit of the criticism is to help provide further supporting evidence. To the extent possible, this is done in the point-by point responses below.

Point.by-Point Responses

1. 1. Purpose, Procedure and Main Conclusions of the Study The purpose of the work is to show that the lower head of a reactor like the AP600 whhstands the load of steam explosions. According to the ROAAM phl-losophy, all physically meaningful causal paths that could lead to failure have to be investigated. The decomposition yields the following central areas of analysis:

1. Since pressures in the kilobar range have been obtained in the KROTOS experiments (nithough not yet with corium and in one dimension), a direct exclusion oflouvr head failure cannot be done. Thus, detalled calculations of possible explosion loads are required, taking into account the specific ge-ometry with respect to venting c((ccts. This is done by use of a 3D version of ESPROSE.m.

2. The possible spectrum of melt / coolant mixtures developing in the lowcr head due to an assumed core melting must be determined. This is done by use of a 3D version of PM-ALPHA.

3. Possible timings and strengths of triggers have to be considered. Due to the uncertainties, an emvloping approach is pursued here, concerning the timing as urli as the strength.

4. Since close to the wall pressures in the kilobat range are obtained in the calculations, specific imvstigations en failure criteria are required. This is done by a simple estimate and also by means of the ADAQUS code. Con-siderations on the possible interaction with thermalloads are also required

5. Considerations on the melting and relocation process in the core and the release to the lower plenum have been considered as necessary for restricting the possible spectrum of melt / coolant mixtures. This is done by separate estimates.

F-22

1 The main arguments in the report are:

(

O It is assumed that a pool of ceramic melt surrounded by crusts forms in the core due to the cooling capabilities o. :maining water in the lov.er plenum (at the beginning of melt motions with level at ~25% of active core height) and the large heat capacity of the lower part of fuel bundle with lower Zr plugs and the Jowermost spacer grid. The key points are then that a downward relocation path of melt through the core eupport plate is excluded and meltthrough of reRector and core barrel is assumed yleiding Rnally a sidewards relocation through the downcomer.

This sidewards relocation is restricted in extent assuming failure at the upper end <

of the pool based on the analysis of heet transfer from the pool and assuming plausible failure sizes. Further, it is argued that only one failure location is available within relevant times for premixing of the relocated melt in the water and triggering. Strong voiding of the mixtures under the expected conditions of saturated water is expected and calculated by 3D-PM ALPHA. Thus, only small amounts of meh in the lower tens of kg are considered to be potentially explosive.

This is taken to directly exclude large break possibilities for the lower head.

Various calculations with ESPROSE.m assuming sutRelently strong triggers are additionally taken to exclude also local th.msts to the RPV. This is Rnally done by comparison with failure criteria for the RPV wall yleiding directly (without application of the probabilistic framework) the conclusion that failure is physically unreasonable.

Further, reRood scenarios are evaluated to even mitigate the possibility of vapor explosion threats, due to cooling and preventing melt outRow. Mixing with the melt in the pool is not considered as effective (smallyield of stratined explosions).

In addition, preventing outflow by reRood would also mean to exclude mixing of -

melt with highly subcooled water, which is considered as the only case with a potential to challenge the lowe? head due to increased penetration depths without ,

excesshe voids.

Cases with thermally uvakened RPV walls ere restricted to later phases of melt outflow. Then the water and mixtures are already assumed as strongly voided.

ReRoad FCis are considered in the report in stratined connguration, i.e. water above a metallic layer. A threat is excluded due to rapid spontaneous interactions with the subcoole.l water and rapid freezing of the metal surface before a thick t

g i L F-23

wates layer establishes which could yield sunicient constraint for strong pressure buildup.

Based on these considerations, the major conclusion of the study is that steam explosion-induced lower head failure in an AP600-like reactor is " physically un-reasonable"

2. 2. General Comments The procedure as well as the general arguments are convincmg. This concerns especially:

• The argument that a strong cold trap at the core supj> ort plate, especially if still connected with water, can prevent the downwards release path to occur before sidewards release at the npper region of the melt pool. This yields a signiBcant reduction in melt Bow rates to the water, especially to a possible downwards release in multiple stn.nms. If the cold trap at the bottom is strong enough, no downwards release will occur until all melt is released sidewards due to a continuous failure progression.

• Then, Ihe saturated coolant condition prevents larger premixtures without high vcids, since larger premixtures could only develop within longer times.

The geometrical conditions of the Bow through the downcomer also favors this.

e The strong voiding of mixtures calculated with PM-ALPHA is thus plausible.

• With the small mixtures (small melt masses) of not extw he void plausible from the above statements, the ESPROSE results are also plausible (the obtained pressures even appear astonishing]y high - probably due to the restricted venting).

• Thus, also the conclusions on the threats are plausible.

• The high number of calculati- as with PM-ALPHA and ESPROSE covering a wide range of conditions can also be taken as supporting.

In spite of this agreement in principle, there remain problems in the details of the argumentation and performance of the anrJysis. Improvements may be performed to even better conBrm the statements and conclusions as a basis for use in licensing F-24

i l

actions. Tids will be discussed as follows in some detail. Since, in my opinion, the G statements on the relocation path are a most critical point, I Brstly concentrate on this, then considering the subsequent analyses on premixing and explosian. I will not consider the aspects of structural failure criteria which appear to be well established. I will only give few arguments on possible further scenario aspects concerning renood.

3. 3. Comments On Melting and Relocation Firstly, it is shown by estimating plugging times from a freezing model based on semi-infinite heat conduction that the plugging takes place in a range of seconds (for the lowest initial rod temperatures 0.6 s with Zr and 2.4 s with UO2 ). How-ever, the melt Row is not taken into account in these considerations. This means, that the [teezing zone may extend over quite a distance. E.g., ~0.6 m would result for 1 s with 0.6 m/s as a typical velocity from CORA experiments. Higher velocities would result with thicker melt films. Thus, the final blockage formation should require some more time and distance (need of additional melting and melt now or compaction to a crust by remelting and relocation of upper parts of a

(% partly blocked region). Further, this process may yield localincoherencies of the

! crust formation, i.e. also weaker regions, although the cold traps at :he bottom give certainly a unifying trend. Thus, in order to further verify the statement of rapid blockage formation calculations with a core melt code would be desireble.

These could also yield a more detailed perspecthe on related (subsequent) im-portant questions, especially the heatup of the cold trap regions and the water level development.

To our knowledge there is no mechanistic code calculation out there that shows the pos- ,

sibility of direct melt relocation through the bottom of the core. Even for BWRs, with much more open geometry and a much larger metallic component in the core, there are significant doubts about such a direct relocation scenario. Given the results of our further evaluations reported in the addendum to Chapter 4, we do not think such calculations are necessary,

4. Secondly, assuming an existing blockage with an merlying melt pool, it is checked whether a steady state with a stable crust below melting point (~2800 K for oxidic or 2100 K for netallic material respectively) can exist. Me:allic n

U g.u

and ceramic crusts are considered alternath'ely, with a heat flux from the molten ceramic pool above of S 0.02 htW/m', a volumetric heating of ~0.5 hfW/m in the ceramic crust and cooling from below via radiation. Here, it appears not clear to me why the fraction of fuel volume is only taken as ~307i (p. 4-6). This seems to be a value for mtact structutes. However, if the metallic parts are all relocated during establishment of the ceramic crust, then this may consist essentially of UO2 /ZrO (~80/20 2 wt ratio). The local shape factor should also not be decishc due to the crust formation from upper material. Taking a value of decay heat of 300 W per kg fuel (p. 4-18) this would yield ~2 hiW per m of the UO2 ZrO2 crust.

Existing and relocated fuel volumes and relocated metallic melt volumes were consistently taken into account, The relocated material is taken at the radially local, axially averaged core power, while the fuel stubs that support the blockage is taken at the local peaking factor - but without releasing its volatile fission products as it is found in a cold, solid state. See additional results in the addendum to Chapter 4.

5. For the downward heat flux from the moltcn corium pool above, a maxi-mum value of 0.02 h!W/m3 (fully developed) is assumed. This is derived from Eq. (4.14) based on the Steinberner-Reinecke correlations for a rectangular ge-ometry (typing error in (4.14): exponent 0.095 instead of 0.049). However, this correlation is only cowirmed for Ra' < 5 10*. For the conditions considered 3

here, I obtain a ndue of Ra' ~ 10", arsuming H = 1.8 m and Q = 2 AIW/m .

The correlation is also derived for non isothermal lateral boundary conditions, in contrast to the present assumptions. The influence of the lateral boundary conditions appears to be small, however. For a case with vertical cylinder and melting point temperature at all boundaries THEKAR calculations [1] also yield a rather similar correlatica (Nu u = 0.935 Ra**), but only for Ra' numbers below 109.

The value of 0.02 MW/m 2is absolutely insignificant in comparison to the heat flux gener-ated due to the heating in the blockage itself (see also the addendum to Chapter 4), which is in the 0.1 to 0.2 MW/m 2range. Thus the 0.02 value could be varied by +100% or more without changing the results - such uncertainty is not expected in the kinds of questions asked here.

F-26

G. But, the main question is to me whether - in view of the above arguments

( and at least some lateral cooling potential- the assumption of a rectangular pool

()

geometry is a too strong idealization and other geometries closer to hemispherical shapes can really be excluded. Such geometrical variations would yield signin.

cant wriations in the heat transfer to the lower boundary. The inBuence of the lateral boundary uvuld increase (natural com ection inhuence versus stable strat-ilication). For a hemisphere (certainly an extreme under the given i?at radial power shape) even a mean heat Bux of 1.05 hiW/m3 would result at the curved lower boundary according to (5.28) from the 1VR report and at the center still 0.1 hfW/m' according to (5,30s) from IVR. With a thermalload from the melt 3

pool of 0.1 hiW/m3and Q = 2hfW/m in the crust only 3 cm of stable ceramic crust would result from equations (4.3) and (4.4), with 0.02 hiW/m3 about 5 cm.

The scenario described in the report does not lead directly to a rectangular pool. It is rather the final state, evolving through intermediate shapes such as that in TMI. Exactly because the fluxes on the vertical and slanted boundaries are greater, and including the non-coolable nature of these intermediate blockages there will be a gradual expansion of the pool, all the way to the reflector radially, and eventually to the final cold trap at the core bottom, where the blockage can be stabilized by radiation cooling. Neither the 0.1 2 3 I MW/m nor the 2 MW/m values are appropriate. Nevertheless, even a 3 cm blockage at the lowermost extremity of the core would be more than sufficient to contain the melt pool.

7. Further, the Erst blockage should be metallic and a ceramic crust should settle above. Then, the combined system of ceramic and metallic crusts should be considered. This yields a lower bottom temperature, thus lower radiative heat removal. Therefore, the crusts should become even thinner. If, due to heatup of the lower structures, the lower region of the metallic crust remelts and relocates, .

this yields a further decrease of downwards heat removal from the ceramic crust region, thus inducing further remelting. ,

This sequence of events describes one part of the phenomena associated with the pool expansion phase (as discussed in the answer to the previous question). The fact remains that this expansion will be stopped radially by the reflector, and axially by the cold traps at the lowermost ends of the bundles. There, the blockages will be coolable and we have shown them to be so both for metallic as well as oxidic compositions. Because of the tight V p.27 I

1 geometry even a few centimeters of blockage would be adequate to support the molten  !

P%1-1

8. - Finally, the downnards relocation path appears not yet as surely excluded as stated. It is also to be mentioned here that local melt streams into the melt pool could strongly enhance the local heat transfer to the bottom crust as shown in [1]. Thus, together with the uncertainties of the process of crust formation considered above and the smaller crust thicknesses of the above estimates, local inhomogeneitics of the crust may become important and may induce local failure at the bottom.

This mechanism is not appropriate here. Here the pool forms from the middle and top with a downwards progression. By the time of interes;. when the blockages have reached the lower extremities, there can be no supply of long-duration cold streams of melt. Also, the pool is rather deep (compared to the experiments mentioned) and would mix, and stop well before the cold plume could reach the bottom crusts.

9. However, the basic idea that signincant cooling potentialis provided from the remaining water in the lower head and the massive core support plate is promising. Perhaps, some further niculations related to the above objections could yield further support, But, the steady state consideration for the crust may not be suflicient in general Calculations on the time-development of melting and crust development with available codes may be necessary for better conBrmation.

Following the whole melt progression process, as suggested here, is a very complicated matter, and subject to much greater uncertainty than the basic-principles approach taken in the report. Such calculations may be useful in providing another perspective, but we don't believe we could defend them as the basis of our case. We are aware that this reviewer may have such capability, and would like to see related results be added here for such fusher support.

10. The main statement is that sidewards melt-through occurs signincantly before possible downwards relocation, within the time of ~ 100 minutes during which etlecth e cooling from remaining water above the core.cupport plate is avall-able. The basic statement is that sidewards cooling is much less ellecthe than downwards cooling. The evaluations la the report take the sidewards boundary condition as adiabatic, i.e. no lateral heat removal is assumed. This appears to be a too strong restriction. On one hand, heat remoni by the produced steam F-28

n should be taken into account. On the ather hand, heatup oithe RPV wall by radi.

T (j ation from the barrel and outside vessel cooling by Booding should be considered.

Taking a temperature difference of ~ 500 K over the barrel and reBector and an outer barrel temperature of ~ 1000 K as given from the calculations (Figs. 4.8 -

4.12) nearly half of the heat flux through barrel and reBector could be radiated ta the RPV wall (if taken at saturation).

The radiative heat sink to the vessel wall and through it to the outside water was taken into account. Because of the low pressure, steam cooling (through the reflector holes) was found to be negligible.

11. The calculations on core heatup and melting essentially yield the timing for melt pool formation (~ 42 -57 minutes from core uncovery to 20%) to be related to the times for evaporation of water above the lower core support plate and for heatup of reRector and barrel. The further heatup of the ceramic pool and the overlying metallic melt layer resulting from reBector melting as well as the heatup and melting of reRector and barrelis calculated by means of equa-tions (4.10) - (4.15), of which (4.14) has been questioned above (questioning the assumption of rectangular pool shape). With the lateral heat Bux in the ceramic I / pool an additional time of 34-38 minutes is calculated for renector melting. This l 'v is taken to verify lateral melt release at a time with still effecthe cooling from below (water above lower core plate). But, it has to be remarked again that lateral heat remo ~l is neglected. At melting temperature, at least half of the lateral heat Bux could be radiated to the RPV wall (if this is not taken to be superheated suniciently). Further, heatup and melting of the sidewards ceramic crust as well as of the barrelis not considered.

Both questions (on rectangular pool and lateral heat removal by radiation and conduction) have been responded to above.

12. The considerations on the overlying metallayer resulting from the melting ,

reRector seem not to yield important effects with respect to the Snal melt release.

Although earlier melt-through of the tcBector can be expected in this range, this only means that the reRector melt is essentially relocated into the gaps between the reRector and the barrel. But then, the refrozen material must melt again to get break-through.

(-

C F-29 i

Not so, because the axial path now is of large dimension, and the relocated metal goes to the bottom of these spaces and builds up from there. hl I

13. Certainly, freezing heats up the still solid reRector and barrel material.

But, the material and energy redistribution by these processes may yield some azimuthal homogenization. Thus, the assumption of a local azimuthal failure may not bejustined by the considerations on geometricalinhomogeneities in the report. In general, the assumptions on failure locations and size are problematic, although the bounding assumptions appear to be reasonable. In my view, the main objections cou'ld be, on one hand, those of above, questioning the exclusion of dowmvards failure, and, on the other hand, the exclusion of several failure locations within a certainly short time frame.

Note that once a relocation begins, by local meltthrough, the melt height drops and there is less opportunity of other melt breeakthroughs azimuthally. Rather, we think the path, once opened, will continue to enlarge and melt will be released basically from the same location. This, however, will have to be very gradual. Also, regarding time coherence, as we understand from the mixing explosion dynamics, has to be seen in the context of a few 100s of milliseconds, while we would have a hard time visualizing melting coherence even on a time scale of a few seconds.

14. The latter point indicates a further deBcit: the further course of melt release is not considered sufficiently. Even with an outBow rate of 400 kg/s (see below) the time of outRow of the whole corium melt pool would last some minutes. During this time, failure could occur at multiple locations, overlapping in time. Further, local melt / coolant interactions could yie.'d additional and also larger failures. A question is whether failures of the bottsm of the crust by such interactions can be excluded. On the other hand, the strong voiding with the resulting necessity of early triggering to get explosions restricts the possibilities for coinciding ewnts. Enhanced evaporation of the water pool also acts in this direction.

Subsequent events are obviously impossible to predict in exact sequence, but the train of though explained here by the reviewer is similar to ours. That is, as melt relocation will continue the water pool will remain with a lot of voids, and the water in it will deplete rather quickly. We would take issue only with the statement that failure will occur at multiple lo:ations. As explained above,it is the nature of the process, with the upper 30 F-30

to 50% of the core barrel thinned out by melting, once a relocation begins it would tend to

-( n) remain focused on the same path, enlarging it downwards.

15. Scenarios of ex-vessel renood are considered with respect to the time of vessel Booding but not concerning the establishment of eB'ecthe lateral cooling as discussed abme. The considerations on vessel Booding before or just about the

. time of reRector melt through consider only the cooling aspects and thermal ef-fccts of focusing by thin metallayers. But, embrittlement due to rapid quenching may favor failure of the pool surroundings at any location. Ebrther, especially en-trapment explosions in the gaps may yield such failures and thus more extensive melt release.

Disagree. We consider all three types of reflood scenarios, and show that the two that are relevant (in terms of their timing) to the melt progression process, actually could lead to men arrest within the core barrel boundaries. Embrittlement cannot lead to failure under these high temperature conditions. Moreover, as shown i Figure 4.15, in the " fast" scenario, the core barrel and reflector would be cooled well be y could heat up by the melt, and in the " medium" scenario, one would require a totahy singular coincideace between core barrel meltthmugh and water level traversing the downcomer length.

A

4. Comments on Breakup and Premixing Q) 16.

From the assumed failure location and size, melt Bow rates of 200 to 400 kg/s are estimated, yiciding ~5m/s entrance velocity into the saturated water pool at a level in the range of the core support plate. Then, the next main point is in my view the fragmentation process. It is stated that adequately bounding the effect of various degrees of breakup leads to extenske voiding develcping rapidly l In all cases. This voiding of premixtures is calculated with a 3D version of PM-1 ALPHA. The melt stream is assumed to be broken up initially into drops of diameter 20 mm ("large enough value to represent a minimally broken-up melt stream"). However, as compared to a coherent stream of ~11 cm diameter (with 400 kg/s and 5m/s), this yields a surface of factor 8.4 higher and correspondingly a higher heat transfer and steam production. Transient breakup could thus yield signincantlv less steam production. On the other hand, the breakup may then not be suRicient for explosive premixtuses. A factor of 6 still results for twoJets of melt with correspondingly smaller diameters which facilitates breakup again to some extent. Thus, mixtures with smaller void may result from transient breakup and assuming severaljets with smaller diametcrs. On the other hand, i V O F-31

there remains certainly a limitation to breakup due to the time consuming process of breakup.

l In my opinion, these contrary cfTects with respect to getting an elfective m!xture, i.e. too less breakup or too strong voiding with stronger breakup, should be ex-piored more for getting the inherent limitations to explosive mixtures. Although the statement of strong steaming appears to be plausible, it may not be possible to demonstrate it for all possible cases, as indicated above. A smaller window for explosive mixtures may become plausible taking into account the above effects of time requirement for breakup and too coarse breakup combined with weaker voiding. Perhaps, some additional variations with Phi-ALPHA could be done to show this, e.g. by considering plausible time laws for breakup of coherent jets together with varying breakup length scales.

Indeed, the THIRhiAL calculations give some perspective on this, showing the extreme cases oflittle stripping of small fragments for a thick jet (7.3 cm diame-ter) and coarse breakup for the smallerjets (2.9 cm and L8 cm) due to long-wave instabilities. However, certainly cases in between these extremes of breakup be-haviour should be considered. Further, the present state ofJet breakup modeling cannot be taken as veriBed. This is alo indicated by the signincant ditferences between results based on Kelvin-Helmholtz instabilities and on the theory taking into account velocity proBles (blifes) which have been obtained with IKEJET, e.g. [2].

Since multiple Jets may occur from one hole by some separation effects (e.g.

connected with the failure mode) or from several failure locations, the restriction of mass assumed in Appendix D, p. D-G for the case with thinner jets appears not to bejustined. Also, the concluding statement of p. 5-12, "that both length scales and void fractions are well encompassed by the Pht-ALPHA calculations" appears to me as too rough, in view of the variations of cases indicated with the THIRhfAL calculations and considered above. On the other hand, I agree in principle to the expectation of strong miding based on the situation considered, with melt into saturated water and with the necessity of breakup for explosive mixtures. Ehrther variations may even better demonstrate this, as indicated above.

The key point is that one cannot have simultaneously enough interfacial area for a strong exph%n and low enough void fractions to pmduce energetics. This was amply F-32

, demonstrated by the calculations made already. We have made additional calculations, C/ and we have provided further interpretations on this compensating mechanisra in the l addenda to Chapters 5 and 6.

17. I think, this could also be done for the situation of bottom failures of the melt pool, excluded in the report.

We also believe that bottom failures could not jeopardize lower head integrity, but prefer to not open up this direction of thinking arbitrarily; that is, without a viable blockage failure mechanism.

18. The exclusion may also be better based by considering additionally the cold trap properties of the lower spacer grid and the Zr plugs quantitatively for conditions ,1 tr boil-off in this region and with melt relocation to this region before lateral melting of redector and barrel occurs (or: improved considerations on the timing of the events, with respect to the above discussions).

The explicit consideration of blockages in the Zr plug region is now provided in the addendum to Chapter 4.

p 19. 5. Comments on Explosion, together with Premixing

( 'l A trigger of sullicient strength is applied to the mixtures in ESPROSE calcula-tions to quantify explosions. The chosen trigger appears to be sufficiently strong to produce strong escalations as in the KROTOS experiments, but its strength is not assessed with respect to possible trigger strengtlas. I agree.that with a sufficiently strong trigger the escalation dynamics may no longer depend on the trigger strength (if overdriven cases are excluded as unrealistic). Thus, the re-suits of the numerical tests may indicate such a limiting strength and no need for further variation. In view of the elfects in the KROTOS cxperiments, the chosen -

trigger can also be considered as strong enough to yield major effects. Looking at early rather than later times for bounding the elfcct of trigger timing appears ,

also appropriate in view of the strong voiding (excluding other possibilities of melt release as discussed in the previous chapters).

• * * *

• c * * * * * * * * * * * * * * * * *

20. The results given in the report show signincant differences in the maxi-

mum local pressures, the maximum impulses as well as loaded areas and times of loading depending on the chosen melt mass Bow, the breakup behaviour and the

(

C') F-33

l time of triggering. E.g., with the higher melt Row a maximum pressure > 5000 bar, impulse 100-120 kPa s and maximum area of 5 m' results choosing 0 = 10 for fragmentation and an instant of triggering at 0.05 s (case C2-10(0.05)), as compared to nearly 104 bar,100 200 kPa s and 3.5 m3 for O = 20 and 0.12 s (case C2-20(0.12)). On the other hand, there seem also to be similarities or bounding trends. E.g., for case C110(0.05) with the lower melt flow, f = 10 and 0.05 s, the results are rather similar to C2-10(0.05), with someuhat smaller impulses and areas in Cl-10(0.05). But small shifts in trigger time give also strong differences, e.g. pages C.3- 16 to C.3-20 in the report for case C2-20. The same is valid for the comparison of C2-10 and C2-20 with similar trigger times (pages C.3-13 and C.316). This is certainly due to the relation between time development of breakup and voiding, producing optimum mixture configurations at different times. This is one cause of uncertainties in getting explosive events or not (together with triggering time).

Concerning this problem o[ sensitivity, the large number of calculations performed is convincing. They yield maximum events but in a limited range and not as singular cases. Some questioning i still have with this respect concerns the choice of f for premixing breal.up and perhaps the underlying time law of breakup (not given in the report) Since the maximum loads appear to be obtained with case C2 20(0.12) as compared to the lower 0, it is not quite convincing to jump to nb and not to consider cases between. Other time dependences may yield further variations. This concerns the questioning of above concerning the premixing process as well as the melt flows.

Point is well taken. See addenda to Chapters 5 and 6 for further insights into relating premixture characteristics to resulting energetics and for consideration of intermediate cases. Through these interpretations a more coherent picture regarding the origins and implications of the relatively narrow " explosion-sensitive" region can be gained.

21. Concerning the latter po:nt, it is to be remarked that the main eIfect af multiple mcis streams into the water - if taken as saturated - would be that a larger region is onded by the explosion (perhaps also some further escalations in more extended premixtures may be possible, this could be checned by ESPROSE calculations) and that thus the venting will be further limited. Also the pressure relief in the vessel wall will then be limited. Thus, it is important to further F-34

P conBrm the exclusion of such multiple events (small wirdows for this!) or to

.,Q Q check the coincidence elfects.

Actually, our pour characteristics correspond to multiple melt }et streams, rather closely spaced within the lateral space dimension assumed in the meltthrough. To widely separate these Jets would be tantamount to assuming a much larger azimuthal coherence, which as discussed above, we do not consider appropriate.

22. :It remains to formulate some general questioning concerning the verin-cation state of Phi ALPHA and ESPROSE. Although a lot of work has been performed on this, I think that severe questions remain. Even if numerical as-pects may be considered as well established, also with respect to 3D, there remain open areas concerning the pigsical formulations. These are e.g.:

• Checks with htAGICC were -ta ny knowledge - restricted to relathely small volume parts of spheres.

Quite obviously this is the most interesting condition. See also addendum to Appendix B of DOE /ID-10504.

23.

• In general, correlations for exchange procestes in three phases are un-

, certain and need further clarincation.

We have an extensive data basc on film boilirg 'n steam-water flows, and a rather elaborate non local radiation transport model. Also, ou.i drag laws are well based on existing fundamental knowledge. Further, a wide range of tests on the multifield aspects show that the code performs well. Of course, there is always room for further developments, but the issue here is whether the reviewer sees any specific limitations or concerns.

24.

• The uncertainties on Jet breakup have already been mentioned above.

Our approach is specifically based on explicitly recognizing, and bounding, these uncer-tainties. .

25. e The microinteractions formulation for lydrodynamic fragmentation in thermal detonation waves needs further clarincation and development, for single drop a well as Snally for drop assemblies. This concerns the conclusions based on theory and on single cifect experiments as well as on KROTOS experiments.

n.

i1 V F-35

This was already recognized in the report, leading us, therefore, to take a conservative approach in the microinteractions parameters. h 2G. With respect to FARO and KROTOS analyses, ditferent premixing and explosion codes have shown the capabilities to calculate the experimental be-haviour. But, the underlying physical formulations and thus the physicalinterpre-tations differ strongly. Further, even no convincing comprehensive understanding has up to now been gained on the ditferences especially in premixing behavior between UO2 /?rO2and Al 0s 2 in KROTOS. Thus, further work is necessary to get approved understanding, models and codes. In general, the results would be more convincing if supported also by other codes based on a common physical understanding and corresponding formulations.

Clearly, this is so, but we cannot be responsible for the codes of others. KROTOS was not designed and is not appropriate for understanding premixing behavior. What is needed is better diagnostics on the premixtures triggered (not predictions) so that predictions of energetics can be made on a more secure basis.

27. G. Comments on ReBood Scenarios As already remarked under 3. of my comments, the ex-vessel renood should also be taken into account with respect to the considerations on the cooling aspects determining the conclusions on the relocation path. In the context of vessel renood also the possibility of embrittlement and thermal stresses favouring failure should be considered. It should be shown that also under these conditions larger melt release and in this case contact with subcooled water is avoided or not threatening. Entrapment explosions, e.g. in the gaps between reRector and barrel, should also be addressed concerning a possible increase in failure and melt release.

There is no mechanism for entrapment explosions, and all other items noted here have been discussed already above.

28. Concerning the interaction of renood water with melt pools, I agree to the argument of rapid small-scale interactions, rapid solidification and in gen-eral small etlectivity of stratined explosions. However, it should be addressed whether relenmt effects of mixing due to the water impact and due to small-scale interactions (also taking into account the falling-back of expelled water) can be F-36

excluded. The situation of reflood under conditions of still existing melt / water

() mixtures in the lower head may be even more important than the extreme of the melt pools in the lower head, if the reflood water could enhance mixing again, collapse steam, favor further melt release by the abow processes, and this under the conditions of already settled melt, i.e. thermal load at the bottom.

Reflood scenarios in this pressure reactor are not arbitrary in their timing, but rather well-defined, as explained at the end of Chapter 4 and in Chapter 8. On this basis we have considered all that needs to be considered.

29. 7. Some Comments for Formal Improvements Some typing errors of relevance are given below (I had no time for further detailed checking), together with some need for detailed descriptions: ,

e Ra' in Nomenclature: factor g is missing, g also missing in Nomenclature.

• P. 4 6, second line from below: 0.2 MW/m' instead of m .

• P. 4-18: effecthe power density .. of 0.26 W/g: does this mean of core materialin contrast to fuel?

e P.4-21, eq.(4.14): Ra'".

O]

e P. 5-2: giving the breakup law would be helpful. ,

o P. 5-4: some further informations on this and on the other color figures would be helpful, e.g. length scales, quantities of melt volume fraction.

• P. 5-5: z directed to top, but in subsequent results inversely.

• P. 5-10,13th line from above: " melt" instead "coohmt",

o P. 5-10, second line om below: "two slower pours" - seem to be better .

characterized by pours with smaller diameter.

e " 'i-1, second line from below: " propagation intensity is basically indepen- .

d .t of the magnitude of the trigger." - It is not quite clear to me what is n ' ant, e.g. with " propagation intensity".

• F 6-1, end of second paragraph: (Theofanous et al,1996a)?.

• L 6-4: 6th line from below: 0.1m' e Fig. 6.2: should be better characterized: length scales, pressure scales

( I kJ F-37

6 Fig. 6.4: locations are identified in Fig. 6.27

• Fig. 6.5: location of peak loading: where?, not included in 6.41

• P. B.1-1 etc.: color characterization is not quite clear to me: fuel void, red lines, blue isolines in jet.

• Appendix C: color pictures: case? characterizations?

e choice of parameters in the ESPROSE calculations, especially concerning micro interactions?

e P. C.3-13 etc.: locations?

e P. D4,6th line from above: Rana instead Rk, & R1 All typos have been corrected and requested information was supplied to the text. " Prop-agation intensity" refers to the strength of the explosion. We appreciate the reviewer's attention to detail.

30. References l [1] hlayinger, F. et al.: "Untersuchung thermodynamischer Vorgiinge sowie Wiirmeaustausch in der Kernschmelze. Teil 1: Zusammenfassende Darstel-lung der Ergebnisse." Institut Enr Yetfahrenstecimik, Tecimische Universitiit flannover, AbschluBbericht Bh!FT RS 48/1, July 1975

[2] Borger, bl., Cho, S.H., v. Berg, E., Schatz, A.: "B cakup of hielt Jets as Precondition for Premixing: blodelling and Exper: mental Verification."

Nucl. Eng. and Design 155 (1995) 215-251 F-38

1 m F.4. Response to T. A. Butler (LAND

td' General Comment and Highlights -

Two reports were provided. The first addresses the fragility and indicates that the portions of the curves above the threshold failure (10-3 probability) level may rise faster than given

- in the report, because we neglected progressive failure effects (through the wall thickness).

The second report examines the loads in relation to the fragility and concludes that, since the two do not intersect, the above criticism on the fragility does not affect the conclusions of the report. Based on this, the present response does not address this criticism. We plan to carry out the kinds of calculations suggested and will include the results in an addendum to Chapter 3,in the final report. Also, this reviewer provides evidence that the appropriate yield stress is 450 MPa (rather than the 330 value utilized), and that the strain rate effect may not be as strong as taken in the calculations. These variations are mutually compensating as intended, to begin with.

Point-by-Point Responses

1. INTRODUCTION AND SUhlbfARY The comments contained in this review are restricted only to a review of Section 3 cf the subject report. The report's authors have done a goodjob of scoping the p

Q possibilities of falling the lont vessel head under the assumed loading conditions.

Well estabilshed am lytical approximations were used to establish the validity of the finite element m > del that was developed to study local failure of the head.

A more detailed model needs to be developed to. include transverse shear etfects and to simulate'fai.ute of damaged elements during the course of the calculation.

This lack of simulating progresshe failure is the weakest point of the analysis.

Appropriate simulation ofprogresshe failure has to be included in order to obta!n defensible results that can be included in probabilistic evaluations. ,

SPECIFIC COhihfENTS Finite Element blodeh Use of the shell elements in ABAQUS is acceptable for determining the distribu-tion through the thickness of all components except for transverse shear. In the

. ABAQUS thick shell elements transverse shear is approximated by constant shear tbrough the section. This is not judged to be adequate for evaluating the possibil-ity of a shear type offailure. A better method for getting good approximations for all of the strain components would be to use several continuum elements through

- F-39

the thickness rather than the thick shell element. Use of many more elements would make the runs longer, but use of the exp!! cit version of ABAQUS would help in this regard (see below). In addition, the use of an axisymmetric Snite element model would afford the opportunity to use a much more dense mesh in the analysis with run times that are still relatively short. The structure and all of the loads that were considered are axisymmetric.

In fact, our main interest is for non-axisymmetric situations, and many of our calculations were not, axisymmetric. However, refined grid results were obtained and shown in the addendum to Chapter 3. Transverse shear effects were evaluated, as described in the addendum to Chapter 3, and found to be unimportant. Progressive failure could not be simulated by the code available to us, and in light of the margins available, and comment But14 below, we did not pursue this point for now. However,it should also be noted that the fragility already rises rather sharply, so the potential marginal error due to this effect cannot be very large (would make the rise steeper approaching more closely the step like behavior shown for a uniform load).

2. The mesh should be considerably more dense in order to resolve fine details in the strain distribution, especially those details relating to strains other than in planc strains. Referring to Figures 3.5a c, even the in plane strains rury from their maximum levels tojust half that level overjust one or two elements.

Although not stated in the report, I assume that the implicit versior, of the ADAQUS code was used for these calculations. The implicit version is always stable but may not always be cotwergent. There is no indication in the report as to whether the time step was varied to ensure a comtrgent solution. A better alternative may be to use the explicit version of ADAQUS for the short transient solutions that are required for the types of loads being considered here. The explicit version of ADAQUS also offers the opportunity to use a failure model that would give more realistic failure predictions (see below).

We used the implicit version, and considered convergence-see addendum to Chapter 3.

This is now noted in the report. The explicit version of ABAQUS is not available to us.

However, we have done now numerous calculations with refined grids and verified the convergence of the solutions. The results are presented in the addendum to Chapter 3 and they further verify our previous results.

F-40

p 2. ' The statement that the time duratian of the loads is less than the natural

() frequency of the head may not be correct. A handbook solution of the frequency of a full sphere with the same dimensions as the hemispherical head gh es a nat ural period of1.5 ms, very near to the 2 ms pulse duration used in this study, it is no wondct that, as stated, the impulse time is "non negligible."

Our point was the same (i.e., that it is not much greater), and for the same effect. To avoid confusion, we changed the "less" to "similar." Also, the statement is true for the actual loads in Chapter 7. Actually, as shown in Chapter 6, for the most energetic cases the main portion of the impulse is delivered in ~0.3 ms.

4. Load Strain Behavior:

Use of the Bodner and Symonds approximation for the dynamic yield stress is a reasonable approximaticn. However, use of the values assumed for the constants D and p should be JustHied raore thoroughly. The alues used here are for mild steels, and may not be appropriate for the alloy steel that is used in the pres-sure vessel being evaluated. Obtaining a good approximation for this relation is particularly importa ' because the maximum strain is very dependent upon it. I l used an axisymmetric model with 15 continuum elements through the thickness i

& to replicate some of the calculations in the report. The results showed that the maximum strain went from 0.52 to 0.16 with addition of the rate model for the yield stress. Considering the magnitude of this diEersace, one should certainly be very careful in the selection of the rate parameters.

That the effect is strong was discussed in the report, as was our choice of the lower yield stress also (300 as compared to the as tested 450 MPa value), as a contingency in lieu of directly applicable data. In the addendum to Chapter 3, we show results using 450 MPa, with and without rate dependence. .

5. Dexter and Chan (1990) address the eHects of strain rate and temperature on A533B steels. This alloy is close to A508 steel and may provide some useful information in developing an appropriate dynande yield stress model.

These data suggest that the rate dependence is not as strong as in our original calculations.

All we can do is bound the behavior, as described in the response to item 4 above.

.p V F-41

G. Failure Criteria and Frgility:

This is probably the most difficult aspect of modeling the response of the vessel head. The failure criterion that is used in the report is probably realistic and conservative except for one important aspect. The model, as reported here does not remove the load carrying capability of elements that have exceeded the failure criterion. Maintaining the load carrying capacity of damaged elements can ght signiBcant o 'cr-estimates of the capacity of the structure. I used the explicit Enite element model mentioned above to look at this aspect of the problem and found that, depending on the parameters used for the ABAQUS failure model, the head could fall for the loads that are reported. I strongly suggest using some sort of failure criterion embedded in the computational model for future calculations.

First, we should clarify that the statement "...the head could fall for the loads that are reported" refers to the loads used to derive our fragility here, not to the explosion loads derived in Chapter 6 (see also But14). Unfortunately, our code did not allow the removal of felled elements (see also response to Buti).

7. The subject report briefly mentions the effect of stress anisotropy on the failure strain. This is an important issue and needs to be more fully evaluated.

The work referenced in the report by Pao and Gilat was performed on Charpy bars (roughly unlaxial strain) and by Shockey et al. was performed in pure shear (no hydrostatic component). Therefore these data don't address the important effects coming from multi-dimensional stress Belds. Data summarized by Ju and Butler (1984) show that A533 alk>y steel when in equal b'i axial tension falls at an equhalent strain equal to about one third the strain for uniaxial tension. Equal biaxial tension is the stress state at the " pole" of the lower head where failure would Brst be expected. The alloy content of A533 steel is similar to that of the A508 steel considered in the subject report. Mirza, Barton, and Church (1996) reported the etlect of the stress Beld on failure strain and its effects in transitioning from ductile to brittle failure characteristics. Johnson and Cook (1985) also discuss the etfects of the stress Beld on fractute of ductile metals.

Other references that may be of help include Jones and Shen (1993) and Ferron and Zeghloul (1993).

This again refers to refinement of the fragility, which in light of comment But14 did not appear to be of an urgent nature. It will be addressed in the final report.

F-42

8. As previously mentioned, the head would have to be modeled with contin-uum elements to accurately predict transverse shear strains. In addition, a failure

() criterion for transverse shear needs to be established. It is unlikely that the failure criteria discussed in the above references are adequate. They may however ghe some guidance in establishing the appropriate criteria. It is possible that when the loading conditions are investigated more closely, the load cases that lead to the highest shear load (such as case 1+) can be climinated obviating the need for this criterion.

Transverse shear results are now available (see addendum to Chapter 3). The results indicate that the shell element modelis conservative.

9. hilscellaneous:

1) The use of the higher yield stress 450 h1Pa is justined based on actual data from Server and Oldneld (1978) where the average yield stress is approximately 440 hiPa for A508 steel (very close to the Japan Steel Mbrks Ltd, value of 450 h!Pa). This is one parameter with ample data to support the use of the actual, as-tested value.

,- This comment bringing in the Senser-Oldfield data is very helpful. The additional margin (3) gained by using 450 MPa,instead of 330 MPa,is discussed in the addendum to Chapter 3.

10. 2) The statement is made that A533B steel has a carbon content of 0.19 vs 0.16% for A508. Information from Server and OldSeld (1978) and the AShfE Code show that A533 has a carbon content of 0.25% maximum and A508, Class 3 has the same upper limit for carbon content. Actual analyses show carbon content from 0.21 to 0.25% for both steels.

11. 3) Chapter 3 in the subject seport does not mention radiation embrittle-ment oflects. If they can be dismissed, the reasons should be given.

4) For SA508 the transition from ductile to brittle behavior starts at about room temperature. The report should ght the approximate material tem-peratures during the postulated event ta show that it is well above room temper-ature.

The end-of-life RTNDT for the AP600 steel forging at the beltline region is specified as 23 F.

The lower head, less irradiated, would be even better than that. At the time of interest, CJ . F-43 i

the lower head would be between 50 and 100 *C, This information is now included in the report. h

12. 5) The presence of Baws is not addressed. I assume that in service inspec-tion will han identlBed any that are signincant in affecting ductile fracture.

Yes.

13. References Dexter, R. J. and K. S. Chan, "Viscoplastic Characterization of A533B Steel at High Strain Rates," Journal of Pressure Vessel Technology, Vol 112, (1990) 218-224.

Ferron, G. and A Zeghloul, " Stain Localization and Fracture in hietal Sheets and Thin-Walled Structures," in Struct ural Crashworthiness and Failure, N Jones and T. Wierzbicki, Eds. (Elsevier Applied Science, London and New York,1993),

Chap. 4, pp.131-163.

Johnson, G. R. and W. H. Cook, " Fracture Characteristics of Three bietals Sub-Jected to Various Strains, Strain rates, Temperatures, and Pressures," Journal of Engineering Fracture hiechanics, Vol. 21, No.1, (1985) 31-48.

Jones, N. and W. Q. Shen, " Criteria for the inelastic Rupture of Ductile bletal O

Beams Subjected to Large Dynamic Lands," in Structural Crashworthiness and Failure, N Jones and T. Wierzbicki, Eds. (Elsevier Applied Science, London and New York,1993), Chap. 3, pp.95-130.

Ja, F. D. and T. A. Butler, " Review of Proposed Ductile Failure Criteria for Duc-tile hinterials," Los Alamos National Laboratory report LA-10007-bis (NUREG/CR-3544), April 1984, hiirza, h! .1. and D. C. Barton, "The Effect of Stress 'Iriaxiality and Strain-Rate on the Fracture Characteristics t Ductile hietals," Journal of biaterials Science, Vol. 31 (1996) 453-461.

Server, W. L. and W. OldBeld, " Nuclear Pressure Vessel Steel Data Base," Elec-tric Power Research Institute report EPRI-NP 933, December 1978.

F-44

.i Letter dated January 8,1997 to' L.W. Deltrich.

.; 14

• A:$_f The purpose of this letter is to clarify the appliability of comments i made in the ; ,

attachment to my previous letter to you dated December 1996, regarding review ,

of the report entitled " Lower Head Integrity Under In-Vessel Steam Explosion

. l Loads," by T.G. Theofanous, et al..

..The comments made on that attachment were limited to Chapter 3 of the subject report and, consequently, affect the fragility curves developed in that chapter The i ,

fragility cunes are subsequently referred to in reaching conclusions in Chapters I6 and 7 of the report. It is important to make clear, houvver, that the fragility '

curves in question do not have major elfect on the conclusions reached in thosel  :

chapters. The loads developed in Chapter 6 and applied in Chapter 7 are low - .

enough that the vessel response is definitely below the lowest probability lesel used in defining the cunes (10~3),

'I should also point out that I concur that the probability levels uwd in devel-oping the fragility cunes are consenathv. The association of these levels with

. strain magnitudes through the vessel wall are acceptable. However, the calcu-lated strain lesvis used to develop the detailed cunes above the 10'3 Fevel may not be conservative. If the cunes are eser used for evaluating higher loads; they l

~ should be ret. .luated based on the review cortments that Ipreviously submitted.

Implementation of the information in these comments will affect the shape of the cunes and could shift them toward lower lesels of impulse load (to the left in .;

ligure 3.11 of the subject report).

IfI can be of further help please do not hesitate to contact me.

r a 0

e D

F .

F.5. Response to D.H. Cho (ANL)

General Comment and Highlights This reviewer finds that additional supporting work is needed before the results can be used in the licensing area. The review process, by comments and responses, has produced additional supporting work. Specific issues raised by the referee relate principally to scenario aspects such as the possibility of " secondary" explosions through a downward relocation path, premixing at a higher pressure, and reflood FCIs. These are addressed point by-point below.

Point-by-Point Responses

1. In response to the request made in your letter of June 17, 1996, I have re-vieued the report " Lower Head Integrity Under In vesselSteam Explosion Loads" by T. G. Theofanous et al. You indicated that this report and a companion doc-ument together " intend to demonstrate the effectiveness of 'in-vessel retention' as a severe accident management concept for a reactor like the AP600". You further indicated that "the parpose of this review is to assess whether ibis intent has been achieved to a sufficient degree for the results to be of use in the regu-latory/ licensing area". Based on my review of the report, I find that additional supporting work would be needed if the conclusions of the report were to be used in the regulatory / licensing area.

The nature and need for the " additional supporting work" identified are addressed, point-by-point, below.

2. On page 9-1, the authors state that "Melhodologically, the assessment imolved only a slight scenario dependence, principally on the permanence of the blockages preventing direct downward, through the lower core support plate, relocation", and that thus the assessment is of Grade B, in the ROAAM scale.

I think the scenario dependence is more than slight, so the assessment may be more of Grade C than Grade B in the ROAAM scale.

Disagree. Scenario dependence concerns uncertainties and a complex, long evolution.

Here we have a well-definad behavior and a robust assessment for it.

3. Suppose a steam explosion would take place in the downcomer region or in the lower plenum, as described in the report. The explosion may not be strong F-46

, enough to fall the louer head, but it may be energetic enough to mechanically disrupt the blockages formed at the lower end of the core.

Q)

The scenario proposed by the reviewer is not reasonable. First, there is no water in the downcomer. Second, an explosion cannot annihilate the lower blockages.

- 4. Further, the explosion would likely expel some water from the lou er plenum so that the lower core support plate may no longer be in contact with water (i.e.,

the ability to cool the core support plate would be lost).

As noted in the report, the heat sink of the core support plate is very significant. In the addendum to Chapter 4, this is quantified to a time margin of more than 30 minutes. Once the melt relocation begins, it will continue, vaporizing the rest of the water in the lower plenum, before failure of the blockages are possible. This was discussed in the report (see

p. 4-3).

5. Thus, the initial explosion, while not falling the lower head, could severely weaken the blockages mechanically as well as thermally. It would seem possible that a relatively smallinitial explosion would be followed by a massive downward relocation of core melt through the core support plate, setting the stage for a

- secondary explosion probably imviving a much larger melt mass. The lower head may well survhv such a secondary explosion, but a separate assessment of this possibility uvuld definitely be needed.

Based on the above, this kind of behavior is not physically reasonable.

6. Based on the code calculations performed, the report concludes that the saturated cooh nt condition in the lower plenum leads to highly wided premix-tures that have a dampening etlect on the resulting explosion energetics. Whlie I am not Judging the validity of the calculations, I Bnd it difficult to reconcile this conclusion with available experimental evidence. Experience tells us that trigger-ing of a steam explosion would be more difficult with saturated water than with .

highly subcooled water. However, once triggered, the explosion energetics does not seem to depend on the coolant temperature that much. Consider, for exam-ple, the results of the KROTOS tests Nos. 28, 29, and 30 [H. Hohmann et al.,

"FCI Experiments in the Aluminum Oxide / Water System," Nucl. Eng. Design 155 (1995) 391-403). In tbese tests, epproximately 1.5 kg quantities of Al2 0 3m elt at 2300-2400*C were poured into a column of water and steam explosions took n

U F-47

place. In KROTOS 28, the water was nearly saturated at 87* C while in KROTOS 29 and 30, the water was highly subcooled at 20*C. The energy comersion ratio was estimated to be 1.3%, 0.8%, and 1.25%, respectively, for KROTOS 28, 29, wad 30. It thus appears that the explosion with the nearly saturated water was at least as energetic as those with the highly subcooled water. Similar Bndings regarding the elfcct of water temperature on the explosion energetics were also made in our recent ZREX experiments. Such experimental evidence would need to be considered when discussing the explosion energetics.

There is a major misunderstanding here. The top sentence implies that with saturated water we get no energetics, which is clearly incorrect. Moreover, all KROMS tests were subcooled. Due to the peculiar mixing condition, timing, and relatively low temperature (compared to reactor materials) all three produced premixtures relatively low in void.

The distinction made here is not between 10 and 30% void fractions, but void fractions going to 60-90% Does the reviewer disagree that such large void fractions will have a dampening (not an eliminating) effect? Empirical considerations, on conversion ratio, such as those mentioned here by the reviewer, cannot get us too far one way or another.

It should be clear, however, that our methodology is consistent with the findings of the KROTOS experiments, as shown in DOE /ID-10503.

7. Perhaps additional parametric calculations in terms of the breakup and trigger timings might be useful.

See addenda to Chapters 5 and 6.

8. In all supporting calculations, the water was considered to be saturated with the primary system completely depressurized to 1 bar. Even in a large-break LOCA, the containment back pressure would remain in the range of 2-4 bars for a long period of time. It would appear that a system pressure somewhat higher than 1 bar (e.g., 3 bars) uvuld have been more realistic for the supporting calculations.

This is not correct. In AP600, at the time of interest, containmer.t pressures cannot exceed

~1.7 bar. In any case, to provide some perspectives on the effect of pressure and subcooling, we provide some new results in an addendum to Chapter 5.

9. ReBood FCis were discussed in Chapter 8. I suspect that renood FCIs in strati 6ed con &gr ations would be of secondary importance compared to the F-48

premixed explosions addressed in the rest of the report. Nevertheless, reBood

[q FCis need to be considered for completeness, particularly in view of the potential Q) for vessel wall thinning due to chemical attack by the metallic melt. The au-thors should be commended for making an enort here. I would have to say that this eHort represents a best-estimate assessment based on engineeringJudgment.

At present, there is no adequate database or computational tool for large-scale stratlBed explosions.

Diragree. The assessment is based on basic principles. Simply put, you cannot create any significant impulse with an inertia constraint of few inches of water!

10. On page 7-1, the authors state that 'Also in this chapter, we would nor-mally present a series of arbitrary parametric and sensitivity calculations, to illustrate, for cases where the base results happen to be benign, the margins to failure" and cielm that "This, in eHect, has already been done, too, by the breakup and triggering calculations, in the course of bounding the behavior". I believe additional work would be needed to make this claim fully valid, and I am con & dent that the authors will succeed in doing that.

p If, notwithstanding the above and the additional work provided in the various addenda j ( included as a part of this package of responses, the reviewer has specific suggestions for i further calculations, we can consider carrying them out.

11. Finally, the authors are to be commended for conda.ing such a detailed evaluation of a very complex issue. .

l * < * * * * * * * * * * * * * ' * * * * * '* * *

• 4 l

n F-49

F.11. Ecggse to M. Corradini RJ Wisconsin)

General Comment and Highlights The principal concern of this reviewer is about the melt release conditions. Other reviewers have gone into this type of question to a much greater extent, and our responses to them may be useful here too.

Point-by-Point Responses

1. COMMENTS and QUESTIONS for DOE Report 1] The authors do a good job in giving a context for their work. However, I am not sure if this analysis which is provided is a failure analysis for the AP600 reactor pressure vessel or a design analysis for the RPV. The former implies that it uvuld be a 'best-estimate' analysis, while the latter must account for factors of safety to assure survh'al. The authors need to clarify this.

This is not a design effort. The approach is adequately explained in Appendix A.

2. 2] In section 3 the authors define the failure criteria and the fragility curw for the reactor pressure vessel. IfI understand the approach a strain-failure limit is used and the associated analysis suggests a RPVlower wall failure probability ranging from 0 to 1 for a spectrum ofloading patterns with impulses between 200

- 400 kPa-sec. Currently, I wonder how this failure emelope compares to that of previous LWR plants analyzed for an in-vessel steam explosion; e.g., the ZIP study in the early 1980s by Sandia and Los Alamos Nat'l Labs?

What we could learn from previous works we mentioned in the Introduction. In those days the pressure wave dynamics were not available to the structural analysts.

3. 3] In section 4 the authors' major point is that the core and vessel design is sufficiently different from past LWRs, such that the core melt behavior is quite different. Two aspects are emphasized: first, the lower core support plate and the non-active fuellength above it [30cm]is large enough in size to delay the core melt progression downward; second, the core steel reflector in the radial direction is also thicker [over 15cm], also delaying and changing the details of radial core melt progression. In essence, the ' race' to the breach of the corium melt crucible, which is formed during the meltdown, downward or radially outward is governed by these boundaries. The authors use a specific 3BE core melt accident sequence F-50

to illustrate this behavior. If one accepts this premise t. bout a radically different core gmmetrical design, a few questkus arise; a) What is the sensitivity tI meltdown timing to downward boll off of water?

Hore examples are needed.

None. Tha 100 minutes is calculated after the water has reached the bottom of the active fal region (from then on it is heated only by radiation). Fr.ls is conservative. Also con-servatively, we do not account for any refluxing of vapors condensed in the upper, cooled parts of the primary system. Further perspectives on downward heat flux sensitivities are provided in the addendum to Chapter 4.

4. b) is the core melt event timing essentially independent of accident so-quance? No guidance is ghrn here.

Yes. In this passive plant there is an essential" collapse" of sequences, as discussed in the report and in DOE /ID-10460.

5. c] The corium exit flowrate seems to be set by the 'rlp'in the reRector along the radial edge of the core region at the very top of the pool. Is this a realistic esthnate, shice it is not much more than that one wo tid calculate from adiabatic heatup and meltdown of the core; e.g., as evidenced at TM12 7 The authors suggest that 200 to 400 kg/sec " appears to be a reasonable range physically to bound the behavior"; but I wonder I( we really know that much about this core

^

melt falle:e progression in a radically new geometric design that this Bowrate is a

' reasonable bound'? More justification is needed for one to ' buy' the argument.

The basis in quantifying this intangible factor has been explained. Some judgment is required here. It really has nothing to do with details of melt progression or the " radically new geometric design." It is a question of how much of the core barrel area can melt through coherently in a time frame of a few 100's of mill! seconds.

6. d) This last question really icads me to the key question of this whole analysis; 1.e., the authors lean me with the impression that there is a good deal of certainty in the melt progression and i have r:gnificant trouble accepting this premise. Specibcully, t!e whole analysis hingw on the fact that the melt crucible which forms during the melt progression has a structural integrity of enough certainty that it would release the melt radially through a pour area no larger than 0.02 to 0.04 sq. meters. This estimate alto seems to be robust enough that

,-3 k) F-51 l

it would be a bound" even with coolant reflood into the core region and any possible disruptive events that may occur. I am very dubious about this and would need to see more analysis to accept this as a ' reasonable bound' Thle melt falluto location and pour rate is the key determinant in limiting subsequent FCI energetics.

Again,3ee response to the previous question. Also, the reflood scenarios in this passive plant are pretty well defined, as addressed in Chapters 4 and 8 already. lt would be helpful to know what aspect of the analysis is considered doubtful and why.

7. 4) in section 5, the authors detsil their multidamensional premixing analy-ses. As stated previously, the melt flowrate of 200 400kg/s seems to predetermine the benign nature of the FCI energetics, but mixing 1., .lso part of the process.

A few questions arise here:

a) Why has the ef[cct of RPV pressure been neglec'ed? Premixing will occur at elevated pressures not i bar [Ilke 2-5 bars) and this will affect the mixing process.

Also, the rise in pressure locally during mixing will ca se the pool to subcool and this has been neglected. Were calculations done to 'bourd' these effects?

For the AP600, the relevant pressure range is 1 to 1.7 bar. Calculations were run for 3 bar (see addendum to Chapter 5). Any subcooling due to rise in pressure locally, and its consequences, are automatically taken into account in PM ALPHA.

8. b) The authors seem to have only considered the premixing process as the melt falls through the limited water pool from its surface to the curved RPV bottom.Would not mixing continue as the melt continues to fall along the wall.

This seems to have been neglected. Is this premixing process of no importance or is the premixing analysis with PM ALPHA not valid for these longer times?

The premixing results presented were not taken as far out due to time constraints. We now have extended these results to much longer times (see addendum to Chapter 5).

, 9. c) The biggest effect of these small pours in my mind is that they may cause local FCis which do not harm the R?V out totally change the melt pouring characteristics for subsequent melt pours; i.e., these small pours and associated FCis will dunage the core melt crucible and markedly increase the melt flowrate or change its location. The authors haw gone to girat pains to determine the fragility of the RPV wall, but totally ignore the fragility of the melt crucible and F 52

the effect of these FCis. I uvuld suggest that larger melt pours will be induced from the bottom of the crucible as well as along its radially edges with larger Q holes, all caused by early small FCis. Ilow have these ewnts been considered or conservathrly bounded for RPV survhal?

Subsequent events are very complicated, of course, but there are substantial solid barriers, even if the blockages were to structurally fall. Moreover, much of the water in the lower plenum would be expelled by the same forces (if present) that are considered to fail the crucible.

10. d) Finally, the PM. ALPHA rnodel has a parametric fuel breakup model that is mentioned briefly, but has yet to be assessed against experiments. For these small pour rates, the tuodel effect is not of great interest, but would be for larger pours in these compicx geometries.ls this model discussed in the support documents?

See PM ALPHA verification report. The model was shown to represent very well all available experiments.

11. 5) in section G, the authors use ESPROSE.m in a revised 3D version to simulate the explosions within the PPVlower plenum. Ghrn the premixed mass of fuel we have n s'ange of results ghen in Table 0.1. Only a couple of questions arise:

c) Why is the trigger time so short; i.e., much less than 1 sec7 Is it due to the time to the RPV wall?Why carmot further mixing along the RPV wall cause larger explosions?

We show that there is a peak (in time) hi the explosive " quality" of the premixtures and trigger times are used to bound this behavior. More calculations showing this bounding behavior can be found in the addenda to Chapters 5 and 6.

12. b] Why is the impulse largest for the mid range value of ' beta'? Is the impulscal 200 kPa sec near the failure limit? or am i reading thi.e prediction correctly?

Again, this relates to the explosive quality of the premixtures, as explained in the adden-dum to Chapter 5. Not at all. The 200 kPa.s case is not near failure. This is because p

U F 53

1 l

1 the impulse is highly localized, as explained already in the report. Also, as explained in Chapter 7, for the actual loading, the vessel remained within the elastic limits. h

13. c) The detailed calculational results in Appendix C abruptly stop in many  ;

cases at 1 or 2 or 3 milliseconds. Why? Is this an indication of something  !

numcrically fatalin the ESPROSE.m sitmdation or what is up?

Again, t',ne limitations caused us to curtail some calculations that were not very irnerest-Ing. More complete results can be found in the addendum to Chapter 6.

14. INITIAL COhih!ENTS and QUESTIONS for DOE /ID-19504 The overall report is quite informath e, but I do have specific comments / questions that need to be addressed.

1] The analysis of the QUEOS experiments are very interesting. For any of the expermsents [Q5, 6, 8,10,11) the visual image is compared to the code, and the leading edge, level swell, steaming rate, steam produced and pressure is compared.

hly first question is what is the criterion to determine tlw leading edge? In the Vetures for the tests, specifically Q10 and Q11 it seems to me that T'hiALPHA is predicting the movement of the front to be faster than the data indicates. Yet in the plots the opposite is represented. Either there is a contradiction or I am observing numerical ditfusion in the images and the researchers have a definition of the leading edge that " corrects or compensates" for this. I have seen the same behavior with IFCI and therefore, am sensitive to it. This needs to be sorted out before I would say that the agreement in the kinematics is acceptable. The hilXA results in Section 3 seem to indicate the same behavior to me and thus I am worried about this numerical ditTusion. There was also no study of the nodalization effects in Section 4 and this is surprising given the results in Section

2. This seems to be a logical thing to do and really should be done.

As stated in the report the predicted " front" was obtained from Lagrangian tracer particles.

Also as stated in the report, numerical diffusion can be controlled by the grid size, and results with still finer grids were promised, to improve the already quite good results. Such results were obtained in the interim and can be found in the addenda to the verification report. Now we have also the PM-ALPHA.L code which eliminates numerical diffusion altogether. The verification steps were redone for this code, and results are reported in the addenda to the verification report (please refer to the cover letter of the present package).

F 54

15. 2) The second comment about QUEOS relates to the radiathe heat trans-fer model. On page 2-16 the report states that the radiathe model had to be Q changed from what is normalin PM ALPilA to accommodate the experiments.

Later on page 219, the report states that the tests do not meet the ' fitness for purpose

• criteria, and one reason is that the temperature is too low [20000 com-pared to 3000C). I am troubled by this empirical "fix' to model the test and thus, am wondering about the " mixed" transport modelin PM ALPHA. This is knowr to be a tough problem, clearly, but to arbitrarily change it seems too rough. Also, I disagree that the tests are not " lit for purpose" They are more fit than others and thus, are very relevant. Thus, the proportion of the radiath'e transfer that goes into bulk heating versus steam production is important to consider and improve upon.

All we are saying is that at 2000 'C the absorption length is so short that basically all heat will be delivered to the interface, so there is no need to compute with the full radiation transport model. Quite the opposite occurs at 3000 'C. This is not an empirical fix. About the " relevance" of the QUEOS tests, see the addendum to Appendix B in DOE /ID-10504.

16. 3]I would also like to see a calculation of PM ALPHA /3D for QUEOS If O Indeed there is a benefit to a 3D calculation. It seems that the QUEOS tests are the largest and highest temperature simulant tests to date with solid particles; thus, it may be of use.

The QUEOS tests are only repeats of our MAGICO tests. Both reach 2000 *C and both have about the same masses. Moreover, and in contrast to MAGICO, nothing is known from QUEOS about the most important part of the interaction, which is the internal struc-ture of the mixing zone. Still, not only did ws interpret these tests, ours were the very interpretations, the next one coming by the investigators (or QUEOS) themselves some eight months later (at the CSNI FCI meeting in Japan, May 1997)! Since the QUEOS tests are axisymmetric nothing is to be gained by a 3D calculation.

4

17. 4] The report finally examines the FARO LWR test L 14 as a comparison with a large prototypical simulant melt poured into water. This seems like a reasonable comparison test, but I am surprised about what data is compared.

There is an enormous amount of data available mer the first twenty seconds of the test [the iltst 5-6 seconds is reasonable before heat loss comes significantly q

O F-55 i

1 l

Into play) and yet the data comparison is sparse at best. I would suggest the following variables be displayed and compared over the first 5-6 seconds:

a) the total pressure and pressure rise rate [ don; usw) b) the steam and water temperature at a few locatloas since its 2-D \

l c) the kinematics of melt entry and arrlval at the chamber base and settling d) the surface area generated by the breakup as a function of time c) the mean particle diameter as a function of time

() the energy now to val >or and coolant liquid and loss by fuel g] the level swell of the pool pone but not for long enough thnes)

We disagree with this comment. For items a], b], c]. and g], results and compariso.o to the experiment were already provided (Figures 3,4,6, and 9). These are shown for up to 2 to 3 seconds because the pro mixing process is over after th!s time. For items d], e], and f], there are no data to compare, but complying with the reviewers request we provide the result of computations in the addendum to Section 3.2.2.

18. Also l am concerned about the arbitrariness of the dynamic breakup model

[O value a 50], that is used and described in pages A 34/35. This whole procedure is a matching exercise for some value of beta unless the results begin with a jet of ~10cm and break up to a size that is consistent with the post-test debris data

[as well as the amount left as n

• cake'). It would seem advisable to compare the ' frozen' model to other FARO tests to prove that results can be consistently predicted for LOG, LOS, L11, L19 and L20; all of which were high pressure tests for quencidng. Also the ' mixed' transport including radiathe transport wuld have to be held constant in these comparisons to prove the match of L14 has some limited ' universality'.

See addendum to Sktion 3.2.2.

19. INITIAL COMMENTS and QUESTIONS for DOE /ID-10503 The report is very well organized and describes in sulli: lent detail the ability of ESPROSE.m to perform shor x propagation calculations for gas / water and va-

- por/ water situntions. I do not completely understand the origin of the C11AT [or CilAT-QL) code comparisons. Are these standard code models or a formulation F-56

i i

of the authors to do a code cross-comparison? I understood them to be the lat- j ter, and thus I wonder about the need to compare to actual experknental data

()

on shock propagation in single phase and multi phase systems. This is a minor point, but I think for completeness a link to data is best. bly main comments are about the comparison to the KROTOS data.

As explained, the CHAT (CHAT-QL) are special purpose codes for these numerical tests.

We have comparison to wave dynamics in the SIGMA experiments also, as shown in the report.

20. 1) The initial statement is made that the KROTOS tests are a challenge since they have imperfect characterization of the Initla' onditions. One question may be if there are any other tests which give them more insight? .After my own search, however Bas,ed these tests are, these and ather one dimensional experi-ments are the best we have. bly other comment is about the initial conditions.

The comments on page 4 20 Indicate that the fuel mass and flowrate, but there is a problem as far as I can tell. The mass is correct, but the initialJet size is not 1 cm but 3 cm and I think the Bowrate of 1 kg/sec is too low by at least 509(.

Finally, the fuel particle temperaturcs are diffe< ent for each oi the tests noted as is the location and timing of the trigger. I am not sure that the authors are aware V) of this. I can send them this information if needed, but in the case of hROTOS 38, the initial conditions are not correct; c.g., thejet size is 3 cm and the trigger time is 1.12 sec at or near K3 and not at the leading edge, with a pour time of about 0.75 sec.

The first part of this comment is inappropriate; the second part reflects a misunderstand-ing. The " imperfect" is the wrong word here. The initial conditions in KROTOS relative to its main task, which is to provide data for ID detonations, do not exist, period! To call ,

this " imperfect" is not simply an understatement, it misses the point completely (see also Fle18, Sch31, and Tur44). Our statement was "From a code verification perspective, the KROTOS experiments are difficult challenges because of their inadequate characteriza-tion of the initial premixture prior to triggering," and we stand by this statement. The renewer's comment is further inappropriate because we do use KRO1DS in our verifi-cation effort. To do so, without acknowledging the uncertainty in initial conditions would be a real omission; it is good that the reviewer's comment provided an opportunity for further emphasis. Even if we ignore uncertainties emphasized by this geometry, on the mul4 field aspects, uncertainties in breakup, widely known and accepted. are enough to V F-57 a

overwhelm this problem. KROTOS was not intended and it is not a premixing experiment.

Still, all we are given are initial conditions for premixing. Now, having understood this point, the initial conditions disputed in the second part of his comment, can be addressed as follows. The size of the fuelinlet for the calculation is 3 cm, which is consistent with the jet size.1 cm is the initiallength scale assumed for the fuel. Its effect on the premixing is unimportant since it will be compensated by the value of the breakup parameter B, which is adjusted parametrically. The trigger is applied near K3, not at the leading edge (see Figure 3). The flowrate of 1 kg/s is the approximate value quoted by many previous publications on KROTOS (Hohmann et al.,1994). Our understanding is that there is no new measurement to imprsve this estimate of mass flow rate.

21. 2) The concept of using the parametric mixing model for a A value of 30 or 50, again raises the question of what is appropriate and why. The kinematics in Figure 2 don't have any comparison to the thermocouple data for position of the melt as a function of time and give no indication what 50 is "better" or more correct than 30 for a value. Also what is the thee evolution of the particles as thejet breaks up from 3cm to what size? None of this is discussed at all.

As discussed in the text, s = 50 is the value at which "the melt penetrates to the region between pnessure transducers K2 and K3." As shown in Figure 2, the penetration is much slower when 0 = 30 and much faster when there is no breakup. The particle size distribution at the time of trigger was shown in Figure 3 of the report.

22. 3) The final point is the use of the parameter, xj = 0.5 to 1.0. Does this parameter mean that when the value is 1.0 all the fuel is quenched as it is fragmented with some fractlon of water and Stcam7 1i that is the correct interpretation then, the pressure plots do not seem to make sense to me. This is especially the case, since the predicted void in by Figure 2 and Figure 3 is very small. There is something missing in the description; since 1.5 kg of molten alumina has the energy of 6 7 hfJ and thus must be quenched by almost all the 35kg of coohmt if there is to be such a *small' pressurization with such little void.

Ilow much water is ' assumed" to be intermixed with this fuel to ghe the pressure signature we sec7 This is never discussed and it is the most crucial part of the model. The complete picture is missing and thus, I am not prone to agree this is a reasonable prediction until all the ' parameters' are specified and explained.

Also, comparisons to more than one test is needed. This has been done with ather FCI models.

F 58

--- _._ .-- --. .~ .. . - .- - _ - . - - -

This question stems from the reviewer's misunderstanding of the parameter rf and the microinteractions model. As explained in the report,"xf si the fraction of computed liquid meltparticipatedintheexplosion." ThisfactorisnecessarybecausePM ALPHAwasmade to underpredict melt freezing (it did not account for surface freezing of superheated melt).

When zf = 1.0, ESPROSE.m predicted a peak pressure of 4000 bar, which is consistent with our previous microinteractions predictions. The amount of water intermixed with the

- fuel is calculated based on the microinteractions parameters, which were given already in

- Appendix C. Several representative KROTOS tests are now interpreted in the addendum

- to Section 4.2.1.

O t

6 1

F 59 .

.,,-, ,, ,,,,,..-.r,,-- - . , , - . - - , , , . , - , , , , - - . , , - - . . , , , , , , , , , , , , - ..,,.5.,1,...,, , , . . ..,-,a,.n., ..4- m - . . - .. - - - .

j F.7. Response to H.K. Fauske and R.E. Henry (Fall General Comment and Highlights l

General and unqualified agreement with the conclusions of the work under review.

Point by-Point Responses

1. As requested in your letter dated June 17,1996, the following comments are offered in the areas ofhfeltdown/ Relocation Phenomenology and Steam Explosion Loads.

hiehdown/ Relocation Phenomenology - We agree completely that a downward relocation path of the melting core material through the core support structure (and resulting large fuel pour rates) is " physically unreasonable". Furthermore, the predicted relocation off to the side and from a fully developed melt pool leading to a nwlten fuel pour rate into the lower reactor vessel plenum of about 200 kg/s, is consistent with the Three hille Island Unit 2 Core Relocation as described by Epstein and Fauske (Nuclear Technology Vol. 89, p. 1021 1035, December 1989). Fuel pour rates of this magnitude by themscives eliminate concerns relathc to global vessel failures, even if an energetic steam explosion is postulated. As illustrated by Epstein and Fauske (1989) and Theofanous et al.

(1990), such low fuel pour rates limit the fuel that can be found in transit within the lower plenum to values at least an order of magnitude less than that required for incipient lower head failure (3 to 5 tons). Quoting Epstein and Fauske (1989)

"A key aspect of the relocation is, then, that significant quantifies of corium melt were not mixed with water at one time. The slow melt relocation phenomenon is, perh..ps, the most important piece ofinformation gained from Thil 2 studies and should figure prominently in future assessments of steam-explosion induced containment failure as util as lower reactor vessel plenum failure due to fuel

. debris overheating " This is clearly the case in the current assessment provided by Theofanous et al.

We regret having failed to mention this important reference. We will introduce this ref-crence in the final report. Still, the inference to AP600 is not automatic, because of the reflector!

2. Steam Explosion Loads - Having eliminated the potential for global vessel failure, Theofanous et al. proceed to evaluate the potential for localized damage, by considering local shock loading, with peak amplitudes in the Kbx range as F-60

l

, a result of a steam explosion occurrence. Again, the conclusion is that faHure

'Q is "plo sically umcasonable". This conclusion is further supported by noting the following observations. l The abow loadings are produced by subjecting the limiting premixt ures at atmo-spheric pressure to triggers resultIng from releasing steam at 100 bar. Quoting the authors, "our triggen are chosen suf5clent to initiate explosions, and they have no relation to what might arise spontaneously during a pour." We agree with this observation, and in fact believe that the occurrence of spontaneous steam ex.

plosions with the molten corium saturated water system at atmospheric pressure considered by Theofanous et al. Is " physically unreasonable". The enormous film balling heat Sux (~3 Mw/m') and corresponding vapor aux resulting with this system (several times the critical heat Bux of ~1 Mw/m') promotes separation and prevents physical contact between the molten corium and water, a prerequi-site for steam explosion ghen a fuel water pre mixture. Temperatures (~2000*C) which are well below the melting temperature of corium (~2700 C), would be re-quired in order to reduce the vapor flux in connection with film bolling to fall below the Buldization vapor Bux. The above considerations are consistent with the noted absence of "explosivity" for the corium water system (I. Iluhtinlemi et al., "FCI Experiments in the Corium/ Water System", NUREG/CP 0142,1712-1727, 1996). This is in sharp contrast to th noted "explosivity" with the often used molten alumina (Al20s). water system. Here the estimated B1m bouing va-por Bux (~0.5 Mw/m') is well below the Buldization vapor Bux allowing physical contact whHe the alumina is stui molten. Whue the noted eBiciencies are quite low, the super critical pressures observed with the alumina water tests in the KROTOS faculty (Hohmann, H. D. et al., Nuclear Eng. & Design, 155, 391-403, 1995), apparently encouragw" Theofanous et al. to model such events and apply them to the LWR system.

In fact, for the Al 2O 3system, at 2700 K, just the radiative flux is ~3 MW/m2 ,

3. We also reviewed the approach taken to assess the stcam explosion created impulshv loads. Certaluly the efforts performed by the authors are impresshe in the number of analyses performed and the detaned graphical presentation of the results. The alculations are based on the microinteractions model which appears to be applied i., a self consistent manner. Our question is whether this is the only way that the .mlevant experimental information can be interpreted.

U F-61

J The underlying supposition in such models are that dynamic fragmentation and intermixing of melt and water can occur on an explosh'e timescale. Certainly the available information shows that fragmentation can occur during an explosion.

llowt er, we are not convinced that the elements associated with fragmentation and intermixing on such a rapid timescale have been demonstrated. In particular, the SIGMA tests performed with molten aluminum indicate virtually no fragmen-tation for melt temperatures where numerous large scale studies have observed explosive events.

The relevant materials here are Fe and/or UO2/ZrO 2 We have data now with Fe, and the behavior is quite different from that of A(. See also addendum to Appendix C of DOE /ID 10503.

4. Furthermore, the detonatlon concept is compared ta the KROTOS mciten lin water and molten aluminum oxide-water tests. With the agreement from this comparison, the authors, in a previous DOE report (DOE /ID 10489), conclude that " low void fraction geometries can produce highly supercr.'tical, ene getic detonations." Our analyses show that there is an alternate explanation to the KROTOS experiments that requires no melt fragmentatlon.1( this is the case, the comparison of the mictointeractions model with the KROTOS experiments indicates nothing more than that ESPROSE approach is consistent with the ex-perimental observations, it does not providejustIfication for the microinteractions physical concept.

The microinteractions concept and the SIGM A derived law are basic information obtained with relevant materials under fully-simulated large-scale explosion conditions. This basic information cannot be doubted. KROTOS data and our interpretations of them with these laws indicate the consistency of these data with large-scale explosions, not the consistency of the ESPROSE.m approach with these data. Any other interpretations (we are not aware of such) need to show consistency with the SIGMA experiments also.

5. Therefore, we arc of the opinion that the approach taken in this document is conservathv in that it osvrstates the possible loads that could be created as a result of thermal explosions. If a design evaluation uses this model and concludes that the boundary would not be challenged, we believe that the conclusion is sound, llourver, if the modeling approach is used and the resulting loads exceed F-62

the capabilities of the structures, we do not believe that this represents an actual q) challenge to the system integrity.

6. Defore such a fragmentation approach can be recommended for realistically assessing the structure loads, it should be proven that a relathcly small pressure increase would be sufficient to self trigger a coarsely fragmented and intermixed system. In particular, it should be demonstrated that a coarsely mixed system could escalate from a smals triggering event into an vsvnt like that characterized in these evaluations. Assuming that a sang!c grid is tilled with steam at 100 bars as a triggering mechanism, it is far too coarse to provide such a definiths representation.

Triggering was not addressed in this report, as was made clear already. This should not be confused with assessing the energetics in a triggered explosion. Since we are not aware of definitive arguments that steam explosions are impossible with reactor materials, we provide here an assessment of energetics, assuming not only that an explosion can be triggered, but also that this occurs at the worst possible time, during premixing, p 7. In suminary, we believe the modeling of fuel relocation and quantification O of premixtures to be reasonable and consistent with experimental observations Inchtding the TMI 2 incident. On the other hand, the assessment of steam explo-slon loads appear to be very conservative. The corium-saturated water system is not likely to exhibit "explosivity". Therefore. a very strong case can and has been made Ibr the effectbraess of "in vessel retentlon" as a severe ac'c ident man-agement concept for a reactor 111: the APG00.

• * * * * * * * * * * * * * * * . * * * * +

• e in k.) F-63

i F.8. Response to D.E Fletcher (U Sydney)

General Comment and Illghlights General and unqualified agreement with the conclusions of the work under review.

Point by Point Responses

0. Please find enclosal my review of the DOE project on ' Lower llead in-tegrity Under in vessel Steam Explosion Loads" by Theofanous and co-workers.

As you will r>cc from the review Ijudge it to be an excellent study in terms of its depth, scope, technical quality and shear volume of work. I fully agree with the conclusions drawn by the authors.

1. Summary This review covers the study oflower head integrity under steam explosions per-formed at UCSD by Theofanous and co wotkers, together with the code validatlon reports for PM.ALPilA and ESPROSE.m. The study and validation reports con-tain a massive amount of very high quality work. The depth of the study and extremes to which the authors haw gone to use validated tools is second to none world wide. For exatnple, no one else is performing 3D premising and propagatlon calculatlans.

The work is of very high quality and in my view the conclusion that steam cxplo-slon induced lower head failure is unphysicalis completelyjustified. The technical arguments support this with a high degree of redundancy.

1 Introduction Firstly, I believe it is knportant to comment on both the quantity and quality of the documentation supplied for this review. The vety complete verification manuals for PM ALP 11A and ESPROSE.m are unique. A minor semantic point but they are mucn more than veillication (which implies that the code does what it should) manuals but are also alldation manuals as they examine how well the code represents real experiments.

Secondlv, I wl.;h to record that I was impressed by the scope, depth and quality of this study. It provides a wry comprehenshv basis for rejection of stcam explosion-induced failure of the lower head.

F44

l l

n The remainder of this document presents specific comments on the Study and U the two validation reports.

2. 2 The Study (DOE /ID-10541)

This section deals with the main document of the study (DOE /ID 10541) and pays particular attention to the steam explosion part of the study.

2.1 Introduction This section gives a brief summary of earlier work on lower head failure. It discusses three earlier studies by Dahl et al, Theofanous et al and 'R1rland et al, all of which highlight the need for mechanistic pressure loading calculations before the lower head issue could be addressed adequately. This is the first such study in which this approach has been possible.

3. 2.2 Problem Definition and Overall Approach This section sets out the methodology to be used. Essentially, the now established O ItOAAM procedure is used in which the owrall event is split up Irdo well-denned O plysical processes that can be modelled, combined with intangible parameters (such as triggering time). The proposed sequence of events and the spilt between physical processes that can be quantlSed using a validated model and those which must be treated in a parametric marmer seems correct to me. In particular, I believe that the Bow chart shown in Figure 2.3 ghes a correct and wellJudged progression of events. Details of the modelling will be discussed later. However, it is important to emphasize that the identlRcation of a sound methodology is very important and I believe that the authors have done a goodjob at this stage of making the process transparent.

4. 2.3 Structural Failure Criteria This section deals with quantlRcation of the likelihood of vessel failure for a transient, localized load. The materialis presented in a clear manner and there is a step by-step progression from an axisymmetric model to the examination of localized loads. The analysis presented in equation (3.10) and Figure 3.8 provides a neat means of determining the etlect oflocallud loading and the performance U F-65

of equation (3.10) in correlating the data is hupresshe Also i believe that the failure criteria ghrn in Table 3.3 are sensible and Rt the presented database.

This chapter is hnportant in that it sets up the basis for the determination of whether a particular explosion loading will or will not fall the lower head. There would appear to be significant conservatism in the analysis, as noted on page 31 and from Figure 3.8 at the high impulse end, and therefore it provides the required function for this study.

5. 2.4 Quantification of the hielt Relocation Characteristics This section presents an analysis of the melt relocation characteristics. It is kn-portant to note that the analysis does not use a system code but instead a number of highly specific models have been developed to address the physical processes deemed to be important. This was the approach followed in the Sizewell B study and seem to me ta be the correct way ta proceed. Based on my participatlon in the Sizewell B study I believe that the methodology used and the conclusions drawn are correct.

The melt Row rates and release conditions are consistent with those found in the Sizewell B study. In particular, I believe that masshc pours of many tonnes per second have been ruled out on the correct physical basis.

In the section on reflooding the authors do not consider the possibility that a steam explosion may occur as the water reBoods the molten pool. It is covered in a later section and it would perhaps be wise to.have given a forward reference here.

Forward reference to Chapter 8 is added.

G. 2.5 Quantl6 cation of the Premixture This sectlon addresses the determinatlon of the premixtute connguration. Firstly, it is important to note that the highly 3D nature of the pour has been taken into account via the extension of the Pht ALPilA code to 3D. Thus the localized, rather than smeared in 2D, characteristics of the melt-water interaction process can be simulated. Secondly, it should be noted that melt breakup has been taken into account in a parametric manner. At first sight this may swm like a urakness, as many proposed breakup models exist. Ilowever, ghrn that none of F 66

i these has been properly mildated it swms appropriate that the elket of breakup

& be addressed in a parametric manner. As pointed out in the report, in the event that the melt enters the water pool and runs along the vessel wall, there will be less mixing than calculated here and therefore the explosion energetics will be reduced.

Based on my experience of premixing experiments and modelling I have no diT-liculty in believing that only tens of kilogrammes of melt are likely to be mixed in the ghen configuration. Clearly the high voiding rate is a consequence of the water pool being saturated. I was leit uvndering whether in the event that the melt pour occurred during the reflooding processes whether there would be sul-Relent subcooling present to increase these masses signincantly? hly expectation is that the increase would be by no more than a factor of two, which would still result in small mixture zones.

The likelihood of subcooled water in the lower plenum pertains to the " fast" and " medium" scenarios discussed in Section 4.4. The effects of subcooling and of a higher pressure on premixing are discussed in addendum to Chapter 5.

7. 2.0 QuantiScation of Explosion Loads G

O Tids section deals with the determination of the magnitude of the possible explo-slons that could be generated from the pt emixtures calculated using Phi-ALPilA.

It is important to note that these calculations, performed using ESPROSE.m are fully 3D and can therefore account properly for explosion venting. The validation of the model is discussed in a separate secthn. It is sufficient here to note that the code has been subjected to a very signlBcant validation etfort which I believe shows that it is ' lit for purpose'.

I agree with the approach adopted regarding triggering. Specifically, triggering at ditferent times and looking for the maximum load is clearly conservat6e. In addition, the effect of the premixing breakup parameter f is consistent with cx. .

perimental observations and highlights the fact that the uncertainties in breakup can be taken into account in a parametric manner.

Ghtn the premixture configurations determined using Pht ALPHA I am not the least surprised that none of the explosions challenges the integrity of the lower head.

• * * * * * * * * * * * * * . t * * * * * *

• V F-67

8. 2.7 Integration and Assessment This very brief section explains that as a consequence of the methodology and results there is no nwd to continue with the probabilistic approach because of the enormous mismatch be: .tn explosion loads calculated and those required for failure. In order to show that this is not an artifact of the approxhnate structural treatment, full ABAQUS calculations showed there to be no problem.

I agree that the only way to obtain a signlReant explosion is to have extensive mixing which requires highly subcooled water. I believe the arguments against this are sound, especially if one keeps in inind that the enormous amount of heat which would be stored in the lower core support structure would be available to remove subcooling.

9. 2.8 ConsideratIsn of Renood FCis This is an important section, as the above analysis has clearly shown that pre-mixed explosions cannot cause fidlure of the lower head. I agree with the view taken that you need a very substantial overlying water pool to provide sufficient inertial constraint to generate a high pressure explosion. As in the previous scc-natio everything is against this, viz. the low water additlon rates, the case with which the melt surface freezes and the fact that as Rim balling occurs the overly-ing pool will develop volds reducing its ability to constrain. The analysis rules out to my satisfaction the possibility that stratified explosions could fall the vessel.

10. 2.9 Conclusions The conclusions contain a summary of the results presented in the earlier chapters and presents a concise summary of the important physical features of the system and the physical mechanisms which lead ta the conclusion thr.t fidlure oi the imver head by a steam explosion is unphysical. I really appreciated this carefully presented summary.

11. 3 PM ALPHA VerlReation Studies (DOE /ID-10504)

This section presents a review of the PM ALPHA verincation studics report. It is important to note up front that PM ALPHA has been the subject of continuous F-68

development and peer review (at conferences) over an 8-10 year period. It is therefore a mature piece of software.

3.1 Introduction The main point ofinterest in this sectl'n is Figure 1 which lays out the verifica-tion and valldation approach. This is very wmprehenshe and covers numerical aspects, comparison with other codes and analytical solutions and with experi-mental data. I can suggest no improvements to this validation matrix. It is also uvrth noting that this section highlights the new feature of PM.ALPilA, naively extension to 3D which is clearly needed in the Study. This clearly represents a masshe amount of work but the new insights gained are definitely uvrth the effort.

3.2 Multifield Aspects This sectlon deals with the testing of the multiphase constitutive relations and the modelling for the sedimentation of particles or clouds of particles. PM ALPilA computational results are compared with experimental data and analytical models (based on the drift flux approximation) for the sedimentation of single pas ticles and clouds. In all cases agreement is excellent. A novel featute of this presentation is that the trajectory of the solution in drift flux volume fraction phase space is presented. These results show that the solution is approached in a variety of ways and helps to explain why multiphase numerics prove to be so complex. These results confirm that the code can reproduce the correct particle fall speed, an important feature the steam explosion study.

' Numerous refereed papers haw been presented showing that PM ALPilA can simulate the MAGICO tests, where in most cases there is also phase transforma-tion. These simulations also show good agreement with local data on mixture ,

composition and vold fraction. This is important as PM.ALPilA must predict the correct mixture composition if the calculations of explosion propagation are to be irliable.

Comparisons of PM.ALPilA simulations with data from the QUEOS tests are generally good. There is evidence of numerical diffusion in, for example, Figure 6 but the authors are aware of this and are ,>lanning runs on liner grids. In the hot cases I agree with the authors that both the relatively low melt temperature (making radiation absorption a surface phenomenon) and the gravity induced A

V F-69 l

l

, . . _ _ _ m. -- , _ -- -

sul: cooling are important. If t Je explanation of the dillerence in steam productton advanced in the text is correct (namely the superl.'ating of a layer of water during the fall stage)it means that interpretation of exj>criments of this type, where there is relathvly little steam production, will always be very compilcated. Given the short thne available to the authors to analyze this data and the experimental uncertainties i feel that Phi ALPilA performed as well as could be expected.

See new results with finer grids and with PM ALPHA.L (addendum to Section 2.2.3 of DOE /ID 10504).

12. 3.3 Integral Aspects The code comparisons with CilYhfES and between the 2D and 3D versions of the code give a high degree of confidence that the basic numerical algorithm is correctly coded and that the 2D and 3D approaches are consistent. The compar-Ison with data from the hilXA06 experiment is at least as good as that achieved by the experimenters using the CilYh!ES code. The lack of melt spreading in the shnulations is very similar to that found using CHYhfES. The level swell and steam production data are well reproduced given the experimental uncertainties.

Again it is fair to say that this test is well simulated given that there are several important experimental uncertainties regarding particle breakup and the steam flow rate.

The comparisons of code calculatlons with data from the Ir14 FARO experiment are also good. In this experiment there is no local data and only global quantitles, such as vessel pressurization and level swell, are avullable for code comparison.

The choice of parameters to match these data seems very reasonable. I found the figures illustrating the non local absorption of radiation interesting and these clearly illustrated the huportance of this phenomenon for high temperature melts.

To my knowledge these are the first calculations to include this feature, which is clearly of unportance in high temperature melt applications.

3.4 Breakup Aspects I completely agree with the chosen approach to breakup. As more tests are analysed it will be possible to increase the degree of confidence in the chosen values for the parameters. Clearly, gh'en that the melt surface area transport equation is already coded it would be a simple matter to include a mechanistic model, should a validated breakup model become available. However, the analysis F-70

preserted in the study shows that the overall predictions ofloading are insensithe Q to the choice of these parameters. Therefore the lack of a detailed model does not in any way cTect the conclusions of this study.

3.5 Numerical Aspects The authors are clearly aware of the need to avoid numerical differencing errors and the presented calculations show that they are taking care to address this problem.

3.6 Concluding Remarks I think this section identlSes the correct areas for future focus. IfI were the authors I would have made more of the fact that this is the most comprehenske v.didation cifort to date and that the code has performed extremely well.

3.7 Appendices Appendix A provides a comprehenshe description of the constituthe laws and Appendix B provides a detailed paper on the MAGICO tests. The reviewer is familiar with the materialin t*:- Appendices and this has not been reviewed in detail.

r t * * * * * * * * * * * * * * * * * * * * * * *

13. 4 ESPROSE.m Validation Studies (DOE /ID-10503)

Firstly, it is important to tackle head on the ESPROSE.m formulation, which I believe it is fair to say has not been widely accepted. I Bnd it hard to understand why this is the case. Essentially, the novel feature in ESPROSE.m is the inclusion of an additional Buld (the m Buid) which represents the fragments and the Buld in intimate contact with them which is being heated. The need for such an approach seems beyond doubt to me following the very careful experimental analysis of Baines [1] and ny own attempts to analyze KROTOS-like tests using GULDESAC

[2]. The authors have provided comprehenshe experimental data for apc'opriate .

pressure loadings to show the finite mixing rate. They identify the need for an enlarged database but it should be recognized that the ESPROSE.mformulation is conservative in the sense that by mixing the fragments with only a fraction of the coolant they genemte high localpressures. This point should be kept in mind when examining the use of ESPRC5E.m results.

A U F-71

The remainder of this section contains detailed comments on the various chapters of the verification report.

4.1 Introduction The main feature of this chapter is Figure 1 which ghts the validation strategy.

This is very extensive and to my knowledge is the Brst model to be subjected to speclBc wave dynamics and explosion coupling verification studies against analyt.

Ical and experimental data, it also covers the two main experimental programmes KROTOS and ALPHA.

Refernng to the first sentence of this comment it is perhaps worth noting that with the possible exception of one reviewer (Jacobs) the microinteractions concept seems to be well received. Also, we should note that the idea is now being emulated in other codes such as TRIO MC.

14. 4.2 Wave Dynamics The 1D solutions for the shock speed and particle velocity (important in relath e svlocity fragmentation) are excellent. The same applies to wall renection studies, the effect of vold and the clicct of non-condensable gas. The venting calculations also show good agreement with the CHAT results. I was curious to know why the emculations were performed for a pressure step of 40 bars over a base pressure of 100 bars and over a space dimension of only 1.4 cm. I would have preferred to see venting on the 0.1 m scale (with a 1 cm mesh) and a pressure difference of e y 10 bars venting to atmosphere. Figure Ob shows that the ESPROSE.m results only exhibit dispersion at the Rrst few thne steps and that the numerical diffusion is modest.

Point well taken. More calculations provided in the addendum to Section 2.1.2.

15. The 2D comparisons are impressive and show that ESPROSE.m captures the wave dynamics very urli. The only point that this section raises for me is why in the type B behaviour the ESPROSE.m results have a spike at the origin (as expected from the source description) but the analytic solution does not (see Figures 7,13 and 19). Is this simply a plotting omission?

Yes, this was plotting artifact and was removed.

F-72

, 10. The experimental comparisons with data from the SIGMA facility are in-teresthig and show that ESPROSE.m is capturing the average wave behaviour well. Clearly, the pressure transducers are picking up many local reRection events which are due to the inhomogeneous nature of the ' mixture' and cannot be mod-elled via a continuum approach. I am surprised thist ESPROSE.m has done so well for this system with the only apparent systematic difference !s the tendency for a ~1 ms time lag.

Actually, the maximum time lag is only 0.2 ms (1 ms is a major division - this was clarified in all captions).

17. This section provides very solid verliication for the code algorithm and the cimice of solution parameters.

4.3 Explosion Coupling This section contains test cases in which energy is input into the gas phase via a parametric relationship in which the energy input into the gas phase is propor-tional to the Buld velocity to the power 1.5. This is done to represent the fact that in ESPROSE.m the energy is input in the m fluid. Results for calculations for both cases considered are in excellent agreement with the CHAT shnulations.

O Figure 5, for simulations on a larger space scale, examines the ellect of grid size.

The comparisons are good with differences being contined to the Interface region.

4.4 Integral Aspects The analytical tests show that ESPROSE.m can perform well at the extreme limit assumed in the Board Hall model. These calculations are hu.sesting as they show explicitly the ellect of the fragmentation rate and entrainment factor on the propagation characteristics. Figure 6 is interesting in that it sho s dispersive-like behaviour but if the grid is as described earlier these are real rather than numerical. Could the authors comment?

The dispersive like behavior is real; it is due to the slow fragmentation rate which allows the pressurization to occur gradually.

18. The confirmation that the 2D and 3D models give similar results is thor-ough and convincing.

I agree with the authors that the KROTOS tests are too poorly characterized for real validation studies and therefore 1 do not think this section is central to O

V F-73

the validation case. The point about melt frwzing is very interesting ad the fact that the code under predicts freezing times is important. Tids ellect will be compounded by the fact that that the melt is assumed to be at a uniform temperatute, whereas an outer shell will freeze Erst. Surface freezing provides the most convincing hypothesis (to me) of the non exploshe behaviour of UO2 in KROTOS.

19. 4.5 Nunnerical Aspects I agree with the conclusions drawn. The presented calculations clearly show that the authors are aware of the need for adequate spatial and temporal resolution.

In addition, the results show a good compromise between diffush e and dispersive errors.

4.0 Concluding Remarks This is a very important wetion and I believe the authors havejudged the current situation very well. I agree entirely with the conclusions they have drawn from the very comprehensive sets oicalculatlons performed ta date. There is a clear need for the high temperature SIGMA data and I am aware that plans to obtain this are well advanced.

I personally doubt that it will ever be possible to characterize the KROTOS experiments much better and my experience with the MIXA tests telis me that there will always be something left to be measured. Therefore I agree that this is a lower priority. The comments on secondasy pressure waves are interesting and clearly of a very fundamental nature. I do not believe that such effects could be addreued easily within the continuum model but I would certainly encourage their investigation.

Finally, I agree completely > .h the closing paragraph: moving to large-scale, multi dimensional experiments will only add confusion.

4.7 Appendix A I have no specinc comments here. I am generally familiar with the modelling approach taken and I believe appropriate modelling clwices have been mr.de from the available database of constitutive laws. It should be recognized that it is in the formulation stage that the ESPROSE.m model differs fundamentally from F-74

l others in that it is 3D and uses the mictointeraction concept to allow for thermal \

disequilibrium within the coolant.

4.8 Appendix B i

This section contains a description of the CllAT code used to provide analytical solutions for code comparisons. The model is formulated for the case of homoge-neous flow ofliquid and coolant (no slip but d!!Terent temperatures). Thus the i system has only real characteristles and therefore can be solved in an elegant and l accurate manner. It provides an excellent means of testing ESPROSE.m.

4.9 Appendix C This appendix is a reprint of a conference paper which describes the microinter-actl>n data and its implementation into ESPROSE.m. I am familiar with this v.ork (from the paper and visiting the faelhty) and believe it to be both unique and of a high quality. Whilst at present results from low temperature melts have to be exttapolated ta the reactor case, plans are well advanced ta produce the required data.

4.10 Appendix D This appendix also contains a reprint of a conference paper which discusses the manner in which the 'real world' dillers from the Board ilall model, it is very Interesting as it shows how the inclusion at mictointeraction physics produces propagation behaviour which is very different from the Board-Hall model and other propagation models which do not allow for micro-mixing. Essentially, it allows propagation in systems which are melt Jean because the energy from the melt is transferred to only a fraction of the water present. It provides an interest.

Ing perspective on which to end the ESPROSE.m validation report and clearly lilustrates what a signliicant advance the micrainteraction concept has been in

. propagation modelling.

References

[)) Baines, M. (1984). Preliminary measurements of steam explosion work yleids in a constrained system. Inst. Chem. Eng. Symp. Series, 86,97-103.

[2] Fletclur, D. F. (1991). An improved mathematical model of melt / water detonations-ll. A study of escalation, lat. J Heat Mass 'Iransfer, 34, 24492459.

?

b) F 75

F.9. Response to H. f acobs (FZK INR)

General Comment and Highlights This is a highly skeptical review, questioning even the non-existence of supercritical ther-mal detonations in highly volded premixtures. Until the reviewer resolves this trivial point in his own mind, we can make no progress here.

Point by-Point Responses

1. 1. Introductory remark In order to put my comments to follow into the right perspective, I must state first of all that I fully agree with the general approach to the problem taken by the authors, i.e. the ROAAM. 7b what extent probabilitics are used within this approach may depend on the purpose and the problem of the study. However, dividing the problem into its physical aspects, treating them in separate parts of the study that can be scrutinized by other experts and linking them in a well denned and verinable way dennes a clear path towards the resolution of the full problem.

Similarly I fully support the basic approach taken to treat the steam explosion problem. The material presented is based on and incorporates a lot ofpioneering and exemplary work in this field. I do not want to shed any doubt on that. The only question I'm discussing is: Is the state of denlopment sufficient to Baally answer the quection under discussion. This forces me to elaborate on potential weak points in the argumentation. If a technical field isn't developed suRiciently, even a ' peer review' cannot Rnally ensure the correctness of an enluation.

Quite obviously, steam explosions are not phenomena that are well understood in the scientinc sense, especially if we are concerned with such large-scale events as are discussed in connection with reactor safety analysis. Unfortunately, such events lie far outside the parameter range that can easily be studled experimen-tally. This is true of the initial temperature and the composition of the melt as well as the masses involved (as mentioned above). This dilemma forces us to largely rely on codes for extrapolating from the accessible parameter range to that of the envimged accident situations. Ideally this extrapolation requires full knowledge and appropriate modeling of all relevant phenomena. Here again we are confronted with gaps, the relevance of which is dinicult to Judge. The concept of 'Rtness for purpose' may be helpful in areas in which the consequences F-76 e

4 l

of neglecting something can be estimated. But how about problems which have not yet been identilla. or the importance of which has not yet been percehrd?

&o In the present state of knowledge bad surprises cannot be excluded. The (only?)

way to deal with this difficulty is to account for all (known) possible traps in the analysis (take a conservative approach) and to require a large safety factor. To some extent, this principle is followed in the study discussed here. But in my judgement not to a sufficient extent.

2. From the point of view of quality assurance, a I>cer review like this one can become fully effecthc c:aly if at least the background material was pubilshed since quite some time so that a thorough discussion ofit has been possible among the experts. In the present case an important part of the background material was delhvred very late during the review process. This reduces the relevance of the present review process. .

A key point of RCAAM is that the review is not hurried through. Valid concerns are pursued for as long as it takes. In the present case all documents were supplied by the end of September 1995. This particular reviewer was informed by DOE's project manager that he could take as long as necessary to supply his review, and it was sent about two months later, by the end of November 1996. Our responses, including updated versions of the reports, are made available to the reviewers a little more than 9 months later (about September 15,1997), so that at this stage the process has been on going for about 1 year.

This was done by design, and, again, will continue for as long as technically substantive concerns exist. Meanwhile, the work was also presented at the CSNI FCI meeting that took place at Tokal, Japan, during May 19-21,1997.

3. 2. Scope of this review This review is concerned with the steam-explosion aspects of the study. The con-tribution of this part of the study to the positive final conclusion, i.e. Interacting ,

masses that are insignificant from an overall energetic standpoint and even local loads that lead to clastic strain only, can be attributed to small pouring rates, i

' a strong volding of the premixing zone and early explosions. The first of these are to some extent a consequence of the melt-water mixing scenarios chosen and l although core melt down is not my proper field of experience I must make a few l

l l G' F-77 i

l comments on this because the way in which melt and water are brought into con-tact is braic for the subsequent events. The possibility of a small steam explosion inducing a larger one as neglected altogethe:. The second point, i.e. the proposed strong voiding of the zone in which corium melt and water are intermixed prior to an explosion (the mixing zone or premixture), is instrumental in two ways: In addition to the small pouring rates it reduces the interacting snasses, At the same time, this voiding seems to be one reason for the dying away of the energetics of explosions with increasing time of triggering which is the most convincing argu-ment for considering carly explosions. Of course, this finding also depends on the third point, i.e. the way in which the steam expklon p oper is modeled. The above three aspects, i.e. scenarios and modeling of premixing and explosion are discusses one after the other below.

4. 3. Technical evaluation 3.1 hielt relocation scenarios By the scenarios it is defined how the melt resocates into the lower plenum and this gives the rates at which wrlum is fed into the lower plenum. Therefore this is an important aspect that must l>e scrutinized during the review step of ROA Ahi. I am not really an expert in this field, myself, but I must raise the question whether it is really possible to exclude with suBicient certainty a downward relocation that could lead to much higher corium Bow rates depending on the number of holes in the core support plate thwugh which corium Bows into the lower picnum. hfy doubts in this respect come from the agreement of the experts in this field that the late phase of core melt down, i.e. the melt relocation phase, is not well understood and from the virtual absence of mechanistic models for growth and especially radial expansion of molten pools. The study that is under discussion here tries to bridge this gap using simple and clear estimates of conditions innuencing the thermal stability of a metallic crust. But in these estimates, e.g. no consideration is given to the possible formation of cutectics which might drastically reduce the melting temperature and thus crust stability. One might also speculate that some hot material could drop into the water remaining below the corium pool, thus decreasing the time untilit is emporated and thus the time of crust stability. In the present study the evaporation time 'happens to bejust about equal the time it would take to melt through the reRector and core barrel.' Of course, there is in F-78

l addition the thermalinertia of the core support plate. But as soon as its top falls

() ~

dry, its surface temperatute will increase and thus reduce the ef(cet oiradlative heat transfer.

Aside from control rod materials which would lead the relocation and thus be eliminated, all other eutectics possible are well covered by our metallic blockages. Since paths for relocation are not available, by huge margins, one is not free to speculate that "some hot material would drop into the water . .." helping accelerate the evaporation. Finally, the reviewer's suppos! tion that "as soon as its top falls dry, its surface temperature will increase and thus reduce the effect of radiative heat transfer" implying blockage failure is incorrect. As shown in the addendum to Chapter 4, this provides an additional margin of at least 30 minutes.

5. Another possible uncertainty is the stability (leak-tightness) of a sideways (radially) advancing crust. This process might induce transverse forces on the supporting stubs of fuel pins which these cannot withstand in their damaged condition. So the crust could fail and the oxydic melt could Bow freely towards the core support plate and possibly through it. (Table 4.1 Indicates that the

' cold trap' is not likely to stop flowing oxydic corium.)- Here one may recall that processes of this nature occurred during the Thil-2 accident [)) although, in G that case the whole melt pool was submerged. As witnessed by several tonnes of corium that solidilled within the core support assembly, a large amount of corium has flown down through about 4 peripheral fuel elements around core position R6. Another downward relocation occurred at core position K8. The latter may have been brought to a stop above the core support assembly. But we do not know how and by what margin.

The TMI did not have the zirconium pellets at the bottom of the core. Still, the melt was trapped aoove the core support plate, preventing relocation through the downwards path.

Contrary to the reviewer's ktention TMI actually fully supports our scenario. See also FauL

6. Finally, the possibility of a large coherent steam explosion that is induced by a smaller one (e.g. one of those considered) is completely left aside. Such event might proceed in c:llierent ways. The common starting point of these would be the mechanical dest'uction of the crust keeping the melt pool. This might be caused directly by the action of the pressure of the first steam explosion or s

/\

U F-79 l

l Indirectly by the pressure of another melt-coolant interaction due to the addition ,

of some water into the upper zone or on top of the melt pool. The induced steam explosion would then occur either within the core volume (if there were still water l 1

left) or in the lower plenum after the melt released from the broken melt pool has drained through the still open holes in the lower grid plate. It is sometimes argued that such melt couldn't encounter water in the lower plenum because that j would have been driven away by the initial steam explosion. However, the first l (weak) explosion might have caused essentiali, a sloshing mowment of the water so that thie co'Jd mix very etfecthcly with the corium streams when returning.

In this context one should also keep in mind that with a large molten corium mass available and melt-water interactions occurring, large amounts of mechanical energy may become available. So it is often hard to argue that some process was unlikely.

These comments do not take into account the geometric features of an AP600 core trapped above the massive core support plate. Nor do they take into account the highly localized, in both space and time, pressure pulses predicted. Water on top of the melt is unimportant in this respect fe the same reason that we are not concerned for late FCIs, as explained in Chapter 9. Mnreover, as explained in this chapter, water on top of the molten core is not physically relevant in the AP600. h4

7. 3.2 Modeling of premixing Premixing is the process that is thought to be required to set the stage for any large scale coherent steam explosion it is, at the same' time, expected to inher-ently limit the masses participating in an explosion by the ' water depletion' effect, l.c. removal ofliquid water from the premixture by large amounts of steam that are created due to fast heat transfer. As these processes are difficult to simulate directly in experiments, recourse is taken to numerical modeling with the code PM ALPHA.3D. For the scenarios -onsidered, this code predicts strong voiding of the volumes accessed by melt, in combination with a cut-off of propagation that is effecthe at high voiding this ghts a strong limitation of the melt masses that can interact. And this is the second pillar on which the final result of the study is resting.

While there are good arguments for the concept of ' water depletion' and also some experimental observations that appear to support the idea in principle, there remains the question whether the quan.ilication given by PM ALPHA.3D F-80

is suHiciently reliable. The program predicts 'the major portion of it [i.e. the

&,m fuel) being in a highly volded region (a > 80%)' and also that the void fraction

' gradient is very steep', i.e. the void fraction increases from values around 20 %

to more than 80 % within a short distance. Such behavior, however, was not seen in the premixing experiments that are being conducted at Forschungszentrum Karlsruhe in order to study the phenomenon and to co!!ect data for code valida-tion [2), [3), [4). It is too early to draw Bnal conclusions from these experiments, but the void fractions in the sunounding of broken up ' fuel' appear to be smaller than expected.

The QUEOS experiments were run under conditions quite different from those o, G 'O.

from which the reviewer's " expectations" may have deri. W. We provided q i sve interpretations of the available QUEOS tests and see that this lower voiding sh: u fact, have been expected. More imporantly, to this day we are not aware of any pubh3. ; _

reliable void fraction maps over the premixing zone in the QUEOS experiments. We hav e such detailed maps in MAGICO (see Appendix B of DOE /ID40504 and the addenda to it), and show that even with very diiu:e pours (0.5%) se get void fractions in the 60-70% range, extended over the whole mixing zone. The QUEOS pours are too short, and too concentrated to reveal the impOmt thermal interactions that lead to extensive and O

g persisting volded premixtures.

8. One may also draw attention to data reported of the KROTOS experiments

[5). In these tests molten alumina was poured thrt ugh an orince with 3 cm diameter into a 10 cm wide tube limi with kater, it mixed with the water and strong steam explosions occurred either spontaneously or following an external trigger. The melt tems. ature was high, typically 2600'K, but the water was subcooled which, of course, tends to reduce voiding. In the KROTOS tests #28 and #29 the water was subcooled by 10 K and 80 N, respecthely, in both cases ,

the steam volume fractions within the reaction tube were 4 % only. But as these are mean values over the whole tube which may contain some regions occupied by water only, it may be more relevant to point out that the steam volume was only about half the melt volume. In test #30, subcooling was again 80 K but the melt mass was larger and its breakup was more intenshe. In this case the steam volume fraction reached 23 % but this is again only 1.3 times the melt volume. So we must check how well the above cited calculational results of PM-ALPHA.3D are founded which imply steam volume frc:tions that are larger than the melt volume fractions by well over an order of magnitude.

O V F-81

i i

Our calculations of KROTOS yield similarly low average void fractions (this is due to subcooling), so, again, there is no surprise here.

9. The original PM. ALPHA was one of the two pioneering codes that used three velocity fields for describing the separate motions of melt, liquid water and steam at the cost of adding considerable complexity to the already quite complicated two field description of two-phase flow. But this is the only way in which one can hope to develop a reasonable description of the phenomena dur-Ing a steam explosion. The fairly standard multiphase equations used provide compliance with the conservation equations only. All the controlling and very complicated physics in the three phase (and at least) three-component mixture must be described by constituthe relations. Here the difficulty arises that one of the main purposes of such codes is to extrapolate from the experiments that are possible in practice to the envisaged accident situation. This implies extrap-olation from simulation materials (sometimes even solid spheres) to the expected (but still quite uncertain) molten corium, from often quite low ' melt' tempera-tures to temperatures around 3000 K, and from the mostly very small scale of experiments to the reactor size. There are a few experiments in which one or the other of the above initial conditions is not as bad as indicated here but as the experimental difficulties grow enormously as the expected accident conditions are approached, the experimental information on the initial conditions and de-tails of the processes is often poor in these cases so that a successful comparison af calculational results with integral experimental results doesn't necessarily in-dicate correctness of the thcoretical model. Indeed, one can expect a code ta perform the required extrapolations only, if all relevant mechanisms are modeled mechanistically and with sufficient accuracy.

This is why we have a very carefully developed verification plan, covering all aspects of the calculations.

10. However, the constituths relations used in PM. ALPHA are often heuris-tic sometimes parametrical. The latter is described in the report for the melt breakup model but is true as well for one formulation of the evaporation rate. The other formulation looks more physical but still does not allow for the possibility that evaporation and condensation occur concurrently in the same integration volume (calculational mesh) due to limited subcooling of the water and intensive local (radiant) heat flux to the vapor / liquid interface where the melt drops are

" 82 l

l l

,, covered by a thin vapor Rim only, as e.g. on those parts of their surfaces that are oriented towards the direction of motion. So any extrapolation to accident

)

conditions must be afHicted with large uncertainties.

Only the breakup law is parametric, and its basis and rationale have been explained in the report. Most importantly, and counter to the reviewer's claims here, PM-ALPHA includes a correct phase change model, as well as a non-local radiation deposition model, as well as a non-local radiation deposition model (it is unique in recognizing this important physics among all such codes). Evaporation and condensation cannot occur simultaneously.

Probably the reviewer refers to the energy split between sensible, that going into the

- (subcooled) liquid bulk, and the " rest" going to evaporation. This is properly modelled in PM-ALTHA.

11. Validation of the original PM ALPHA code by comparison with experi-ments was first described in Reference [6] which is also reproduced as Appendix B in the special verincation report [7). An appeal of the general agreement reached may be obtained from the data on the leading edge advant ment. With cold spheres this agreement is mostly reasonable. With sphere temperatures of about 1600 K the data are reproduced within about a factor 2. In the ' production runs' of the present study the inillal temperature of the melt will have been beyond Q 2900 K so that the uncertainties will certainly have increased quite considerably.

Q We have data now up to 2300 K, and in all cases the code very accurately predicts the front advancement. Perhaps more importantly, the internal structure of the mixing zone is predicted quantitatively. Actually, as melt temperatures increase, the prediction task gets easier. The reviewer's is a very poor way to declare errors. When the quantity goes to zero the error, even for an excellent prediction, would go to infinity.

12. Here we are mainly interested in the high void fractions timt have been measured and predicted during the verincation process. The dats given in [6]

have been obtained with the MAGICO experiment and have been described as highly relevant ('the measurement not only provides insight into premixing, but .

represents probably the most important test for computer codes'). Hence our expectatlun to Bnd high local void fractions in our own experiments. However, the local void data presented in [6] have been measured in a position or better line or 'small region' (of unknown size) 15 cm below the initial water level. This depth is only two thirda of the equivalent diameter oithe pour. We may guess that the measurinc volume was centered with respect to the particle jet (the pour). How Cl F-83

i its width compares to the width of the pour is not known. The measurement was performed at 0.35 sec, i.e. Just after the end of (or behind) the pour, probably in order to avoid the presence of many spheres at the level of the measurement.

These circumstances appear to have produced the obsened high void fractions possibly without too much contribution of steaming. It is our obsenation from the QUEOS experiments [3], [4] in which streams of spheres are poured into a water pool in a similar way, that the particle cloud is always followed by a gas filled chimney - with cold spheres as well as with hot spheres. This is largely a consequenx oithe momentum transfer betuven the particles and the water while thermal effects are of secondary importance - they essentially inBuence the way in which the gas chimney is closed again. That this is also true in the MAGICO experiments is clearly shown by Figures 14 and 15 in Reference [6] which illustrate a ' cold' run. This means that the reported high void fractions have little to do with the so-called ' water depletion ' effect and there is no experimental support for the high void fractions calculated in the ' production' runs at positions far away from the melt entrance. One might add that corresponding to our obsenatlons in the QUEOS experiments, thermal effects just start to be detectable in an overall sense (beyond local elTects around each individual sphere) at sphere temperatures as low as 1600 K. Even at the much higher temperatures beyond 2300 K that g have been reached in QUEOS, no high void fractions could be obsenrd outside W the initial gas chimney produced by the entering clouds of spheres (essentially by momentum transfer).

The QUEOS behavior is peculiar to the experimental conditions and it is quite predictable with PM ALPHA and better yet with PM ALPHA.L (see addendum to Section 2.2.3 of DOE /ID-10504). The addendum to Appendix B of DOE /ID-10504 should be helpful to the reviewer in sorting out the differences in his own mind. Comments such as in his last sentence need to be supported by data, for otherwise such points and responses can only produce confusion.

13. In the main body of the verification report l7] global estimates of the water content within the mixing zone in QUEOS are used for further checking PM-ALPHA. Unfortunately this type c! data is hardly suited for a quantitathe comparison with code calculations. The difSculty is that the result very much depends on the choice of the outer radius of this zone because, due to the weighing with the radius squared, it is this region that dominates the integration over F-84

i

' the total volume. In the experiment this difficulty can be overcome to some

(

extent by precisely determining the shape of the mixing zone from high-quality photographs - at least to the extent that a qualitathe result can be obtained.

However, in code calculations, the calculational mesh is not able to sufficiently resahe this outer boundary. So, what is ghen in [7]is the 'PM-AIPHA result for the central region of the mixture, containing the main portion of the particle cloud.' As a consequence, the ca.iculated value is somewhat ambiguous and Figure 13 in Chapter 2 of Reference [7] unavoidably compares qvmstitles with different definitions.

We did the best we could with the data available in QUEOS. The weakness is not with the calculation (with fine grid and Lagrangian particles we can resolve the mixing zone to a very high degree) but rather with the experiments that give only a very rough estimate of a zone-average void fraction. The new interpretations of QUEOS with PM ALPHA.L should help tids reviewer understand what is going on in QUEOS (see addendum to Section 2.2.3 of DOE /ID-10504).

14. It remains that the code in this case predicts low voiding (in contrast with the production runs). But here the code appears to have gone to the other Q extreme due to its inability to describe evaporation in the presence of subcooled U water which even leads to the reported underestimation of evaporation (steam flow) rate and pressure rise. To explain there discrep'.acies by possible liquid i superheat of the water in the experiment is probably inappropriate in the presence oflarge free surfaces.

Incorrect in both respects. Our code describes evaporation in subcooled water well, and we estimate the steam flow quite well. The extra peak is indeed due to water surface layer superheated, as described in our report. The reviewer has not provided evidence to refute ,

this real phenomenon. We now have a more precise model for it, as well as of the mixing phenomenon with PM-ALPHA.L (see addendum to Section 2.2.3 of DOE /ID-10504).

15. Another uncertainty of the calculational results is due to modeling the corium breakup. The surface of a certain amount of material varies linearly with the (imerse of the) particie radius. Therefore modeling the corium as individual droplets with 2 cm diameter from the very beginning gives it already a quarter oi the surface that it would haw with drop diameters of 0.5 cm which can certainly be considered as well prefragmented (broken up). In the calculations presented,

[U' F-85

this initial diameter is combined with an entrance volume fraction of 25 % only so that there is an intensive thermalinteraction from the very beginning. However, in the PREMIX experiments being performed at Forschungszentrum Karlsruhe

[2), we have observed that a melt jet can penetrate to quite some depth !nto saturated water (e.g. 0.5 m for a jet diameter of about 4 cm) before it starts to break up and to int.ract more violently (still not explosively). In these cases the melt is molten alumina at about 2600 K the density of which is only about one third of that of corium. So this behavior is even more probable (should be more pronounced) with corium. Such dynamic breakup process with virtually no breakup in the beginning that allow the melt to penetrate deeply into the water followed by more rapid fragmentation that breaks the melt into medium-sized drops (which might be the most dangtrous con &guration) cannot be bounded by the parametric breakup model that was employed. Such bounding would require to model as well the entrance of coherent mek (melt being the continuous phase) that is not premixed with water artincially (by assumption) from the very beginning. In this context it is also important to note that breaking the melt into very small dropicts (e.g. 0.2 cm) may l>e very optimistic because these small drops produce a lot of vapor, i.e. high voiding and may already start to freeze so that they can no longer participate in an explosive interaction. The importance of freezing for the benign explosion results reported is not discussed.

There is no instrumentation in PREMIX to provide information on the breakup charac-teristics. Our approach easily spans all regimes, from a coherent jet (large length scale) to a broken-up cloud. The transition is controlled by the b'rea'<up parameter. The cases provided in the report were a selection from trial runs over a much wider variation of ini-tial size and breakup rates. With the radius changing from 2 to 0.2 cm we have one order variation in interfacial area. 2,iore importantly, the interaction (voiding) is controlled by specific interfacial area (that is, area percent volume of mixture) which, in turn, depends, in a highly complex, nc,n linear fashion, through the melt length scale, on momentum coupling between all three phases. Finally, the reviewer by focusing on voiding alone, is missing the point completely (explained repeatedly in the report) that it is the combination of voiding status and respective spedfic melt interfacial area that matter on energetics.

There is a key compensating effect here that is yet to be understood by the reviewer. Freez-ing is not important here due to the short contact times. In considering what kinds of voids can be produced with what kinds of drop sizes, the reviewer should take a look at the new MAGICO runs (see addendum to Appendix B of DOE /ID-10504).

F-86

i

16. 3.3 Modeling of explosions O The most important finding of the calculations in this area is the cutoff that occurs at higher void fractions. However, the model used to describe exploshe interactions - the mictointeraction model - has been developed on the basis of experimental observations in a situation with virtually zero voiding. The param-eters of the model have been fixed using these experiments and it has been shown that the model can be made to ght results looking reasonable (by proper parame-ter choices) by simulating a KROTOS experiment in whic he local void fraction was assumed to be between 25 and 40 4. It has been the declared purpose of the microinteraction model to explain the occurrence of strong pressure increases in the presence oflarge amounts of water (Iow fuel to water mass ratio). And as such it is highly interesting from a scientific point _ of view and may be very relevant - in this special situation. But one cannot expect this same model (with the same parameter settings) to work properly in a completely difTerent situation in which therc is very little water present. The failure of this special interaction model to predict strong steam explosions under conditions for which it wasn't designed does not necessarily say anything about the occurrence of steam explo-sions in situations as suggested by the premixing calculations should these ever occur. Especially in the case oflarger melt masses (and possibly smaller overall void fractions) the lower plenum of a pressurized water reactor might provide enough external confinement for completely different interaction mechanisms to become effecth*e. These mechanisms may need more time for their development but might in the end arrive at similarly ellecthe interactions. An important example of mechanisms that may contribute to such alternate types ofinterac-tions are the thermal fragmentation mechanisms that may not need much water and are completely left aside in the present study. This might explain why the most etlicient explosions are obtained very early (prior to 0.12 sec) followed by -

much less etlicient interactions at later times in all cases with a finite breakup parameter. ,

Here we are interested in highly supercritical multiphase thermal detonations, as only they can challenge the lower head. Such detonations cannot occur in the presence of large void fractions for many reasons that are well accepted (this is not a unique ESPROSE.m finding). It is not clear what "new type" of interaction the reviewer speculates about, but whatever it is, it is clearly of no interest here. Also, the reviewer's scenario for collecting a k I V F-87 I

lot of melt in the lower plenum without removing the water in thi ocessis not consistent with everything we know about mixing. settling, and voiding.

17. The picture is less clear in the cases in which additio:vi breakup was assumed not to occur. As outlined in the previous section these might be the most interesting cases in this study. Here no clear maximum of explosivity has been found among the cases considered und it is argued that 'slightly broken up premixtures remain very benign.' However, Table 6.1 shows that in the case C2-nb the maximum peak localimpulse is 30 kPa.s which may already be viewed as a low to intermediate value and that it occurs at the last trigger time considered, i.e.1.0 sec. Nothing in the results presented supports the idea that the value might not be larger (and maybe important) at even larger triggering times.

See further calculations provided in the addenda to Chapters 5 and 6.

18. There is a further and independent argument for early triggering. It states that early triggering is due to the interaction of melt (Jets) with structures. This widely used contention, however, does not agree with the observations from the PREMIX experiments at Forschungszentrum Karlsruhe. We have now performed il such tests and in 4 of these the melt was forced to interact with structures (vertical ' jet' on horizontal plate - in one case even equipped with compartments).

Only one of these tests (the last one performed on 21 August 1096) lead to a violent thermalinteraction (a weak steam explosion) about 0.8 sec (almost a full second!) after melt-structure contact [8]. One may also make refere~ to the KROTOS tests, in which the otherwise very explosive alumina melt settled at the bcttom of the reaction vessel copying its shape when solidifying, in cases in which the water was saturated and no external trigger was applied [5]. So, melt-structure interaction does not necessarily provide early triggering.

We did not limit our range of interest for trigger times based on such arguments. Rather, the trigger times were chosen to bound the energetics.

19. 4. Summary The alBrmathe Rnal result of the study follows from three Rodings: low corium-water mixing rates, very high void fractions in the premixture, and, partly de-pending on that, elfective explosions being possible only during a subsecond pe-riod at the beginning of premixit.g. I have serious doubts about all three of these.

F-88

With respect to the melt relocation scenarios i doubt that the present state of

.(x ) kn3wledge allows to deBnitely exclude downward relocation paths that could lesd to much larger relocation rates. Not really being an expert in this Beld i must leave the judgement to those experts, provided they can positively defeat my arguments. In addition, processes that are induced by a Brst (weak) steam explo-slon might lead to a more ellective melt-water mixing and thus to a larger steam explosion. With respect to premixing, the wry high vo d fractions predicted by the code Phi-ALPHA even outside the gas channel that immediately follows a mass plunging into water don't seem to be supported by experimental evidence.

The code itselfis not provided with sufficiently mechanistic models and is not suRiciently validated to support the high void fractions by itself. With respect to the explosions, the failure of the code ESPROSE.m, i.e. the peculiar interaction model in it (the mictointeraction model), to predict eBicient explosions in highly voided premixtures, doesn't prove that such explosions were not possible on the base of diiTerent interaction mechanisms, even if highly voided states would occur.

We strongly disagree with all points in this summary. Eace one of them has been refuted above. Apparently the reviewer cannot see the impossibility of propagating supercritical thermal detonations in highly voided premixtures. This is +he most trivial part of the

()

x subject, and until he resolves this in his own mind, we can make no progress here.

20. Literature

[1] J. bl. Broughton, Pui Kuan, D. A. Petti, and E. L. Tolman, A scenario of the Three hille Island Unit 2 accident, Nuclear Technology 87 (1989) 34 - 53

[2] F. Huber, A. Kaiser, bl. Steinbruck, and H. Will, PREhflX, Documentation of the Results of Experiments Ph101 to Ph106, Forschungszentrum Karlsruhe Report, FZKA 5756 (hiarch 1996) ,

[3] L. Afeyer and G. Scimmacher, QUEOS, a Simulation-Experiment of the Pre-mixing Phase of a Steam Explosion with Hot Spheres in Water, Base Case Ex. ,

periments Forscimngszentrum Karlsruhe Report, FZKA 5612 (April 1996)

[4] L. Meyer, The interaction of a falling mass of not spheres with water,1996 ASME/AIChe/ANS National Heat 'Iransfer Conference, Houston, TX, August 3-6, 1996; ANS Proceedings, HTC-Vol. 9, pp.105-114

[5] H. Hohmann, D. Magallon, H. Sc!dns and A Yerkess, FCI experimen*s in the aluminum oxide / water system, Proc. CSNI Specialist Meeting on Ehel-Coolant G

V F-89

\

Interactions, Santa Barbara, CA, January 5-8, 1993, U.S. Nuclear Regulatory Commission Report NUREG/CP-0127, NEA/CSNI/R(93)8 (hfarch 1994) pp.

193-201

[6] S. Angelini, T. G. Theofanous, and W. W. Yuen, The mixing of particle clouds plunging into water, Proc. 7th Int. hitg on Nuclear Reactor Thermal Hydraulics NURETH 7, Saratoga Springs, NY, September 10-15, 1995, NUREG/CP-0142, Vol. 3, pp.1 14- 1778

[7] T. G. Theofanous, W. W. Yuen, and S. Angelini, Premixing of Steam Ex-plosions: Phi ALPHA Verincation Studies, Report DOE /ID-10504 (September 1 996)

[8] H. Will, private communicatien (to be presented at the OECD/NEA/CSNI Spec. hits on Ehel-Coolant Interactions, Tokal, Japan,19-21 hiny 1997)

O p., 9 ,

. - ,, , - - . - . - ~ . - . - - - . _ - . - - ~ . - . . - . - - - .... . - . _

j

!O 3 1

4 F.10, Resoonne to F. Mayinger (U. Mnnchent

> General Comment and Highlights -

General and unqualified agmement with the conclusions of the work under mview. -

.r ,

Point-by Point Responses a L; Not being an expert in structural mechanics, I shall concentrate my re-view on the_ thermo fluiddynamic part of the report,; trying to give an overall .

assessment.

i For my review, I also took into account the report DOE /lD-10503 " Propagation of Steam Explosions: ESPROSE.m Verification Studies", a paper by S. Angelini u.a. on the Mixing ofParticle Clouds Plunging into Water /1/ and another paper by Chen u.a. on the Constituthe Description of the Microinteractions Concept

' in Steam Explosions /2/.

Problem

. There are many papers in the internationalliterature dealing with the phenom--

- ena and the effects of steam explosions. They differ widely in their statement on explosion loads depending on assumptions or predictions for premixing, heat transport between molten fuel and comersion of thermal energy into mechan-ical loads. Experiments were made with various melts, representing a variety

,y of boundary conditions (from one dimensional to multidimensional) and a wide range of scale.

1 The report under discussion here does deliberately not make the hopeless attempt

.to find an agreement or an average between the wide spreading results of the literature. It furthermore is based on carefully planed experiments, performed by ,

some of the authors and on constituthe descriptions of phenomena, invohrd in steam explosion processes.

Object of the study is the advanced pressurised water reactor AP600 or respec-

tively the integrity ofits pressure sessel against hypothetical loads of-team ex-

plosions.

Entering thejungle of phenomena and etlects connected with and resulting from

'I steam exploelons with the aim to come to a quantitathe and physically reasonable result with respect to the mechanical behaviour of a pressure sessel is a task, F-91 i

l which cannot be fulBiled in a complete, best estimate way on the basis of today's overall knowledge. Tids is the case in spite of the fact, that numerous research j wark has been performed world wide and that the authors a[ the report, being under discussion here, made excellent contributions, analysing steam explosion phenomena and effects in a theoretical and in an experimental way. There are many intangibles in steam explosion processes. Being forced to demonstrate the safety margins of a pressure vessel against steam explosion loads in a way, which is resistant against critical questions, it is quite obvious to apply conservative assumptions.

The design of the APG00 " invites" such consenative assumptions, because, be-sides the low power density, the core is not only surrounded by a pressure vessel with a rather thick wall, but also by a stainless steel reBector inside the core barrel. So the APG00 design can "tolerste" conservative assumptions. By doing this and regarding the results, one hu to be very careful with any attempts to transfer the data, obtained for the APG00, to other pressurised water reactors.

Conservatisms, assumed when calculating the thermo- and Buld-dynaruic situa-tions during steam explosions, could lead to predictions with respect to pressure vessel failures, which are far beyond the physical reality under such an hypo-

- thetical accident. Tilerefore, inspite of the Sne work presented in the report DOE /ID 10541, there is still a lot to do to obtain a still more realistic basis for safety analysis and realistic predictions. However, we must also be aware of the fact, that there always will remain many intangibles within the scenarios of hypothetical severe accidents.

2. Melt relocation characteristics Melt relocation characteristics are inBuenced by the heating up of the uncov-cred core, the transition to a molten pool, the availability or non-availability of downward relocation paths and several melt release conditions. The authors very carefully analysed all processes, preceding or being involved in melt relocation, including blockage coolabilities and the resistance of the reRector and the core barrel against melt-through. The conclusions, drawn from the calculations and physical considerations, are convincing. The two main conclusions, namely that the failure itself can be expected, that it will be local azimuthally and very near to the top of the oxidic pool and that F-92

i the release will occur within a time-period, which is within the coolabil-Q' )\ ity of the lower blockage, are presented in chapter 4 of the report (see page 4-25) and give the good feeling, that the maximum amount of melt, which can interact with the water in the lower plenum, forming a steam explosion, is limited and by this also the energy release and the mechanical load onto the pressure vessel wall would be within a reasonable frame. So the limitation of the energy scenario, by carefully study-ing melt relocation characteristics,.is a very important and very commendable contribution of this report to the state of art in steam explosion analysis.

A further, very important result in this chapter is, that "re-Bood scenarios" noul no further consideration from a steam explosion standpoint (lower head integrity).

This conclusion should and could have consequences for future planning of ac-cident management activities for existing pressurised water reactors, also, it means, that any elfort should be undertaken to add water again into the pressure vessel after a beginning core degradation, because it would be of advantage for preventing a further escalation of a severe accident.

• * * * * * * * * * * * * * * * * * *

• k *

• g Q 3. QuantUication of premixtures The authors of the seport came to the result, that for the AP600 the amount of melt, pouring into the lower plenum through the downcomer, would be in the order of a few hundred kg/s. Based on this information, they determined the range of premixtures of melt, water and steam and their distribution on the way to the bottom of the vessel. Their calculations are based on fundamental aspects of the premixing phase, which a part of the authors studied seriously in exper-iments (the hfAGICO-2000), involving well-characterised particle clouds mixing ,

with water /1/. In these experiments, they performed detailed measurements on external and internal characteristics of the mixing zones hiixing in saturated and in subcooled water was studied. The results of these measurements found entrance into the Phi-ALPHA code, which they at Erst used for interpreting the experimental results and which is the basis for the analysis of quantifying premix-tures during a hypothetical steam explosion scenario in an AP600. Interesting phenomena they found were the formation of densely packed regions and ofin-stabilities at the penetrating front (isothermal conditions) and local voiding in the mixing zone, as well as global voiding through the level swell (hot pours).

V F-93

it should be mentioned here that the original 2D PM. ALPHA code was extended to a three-dimensional version - called PM ALPHA.3D - version. The results, predicted for the APG00, showed, that premixing mainly takes place in the down-comer and at its lower end to the lower plenum. The average mixture zone and voidage zone is mostly shorter than 1 metre and the average fuel length scale varies between a few millimetres and 2 cm. It takes a few tenth of a second until enough small molten particles are formed during the mixing process.

This ghes hope, that a very Srst steam explosion will occur, before a larger amount of Encly dispe: sed molten liquid is mixed with the water and that this very first itcam explosion produces such a high voidage (steam) in the waterpool, that a further large steam explosion can be avoided. It .'s obvious, that the authors do not study this possibility, because it cannot be quantified, but it may be allowed to mention it in this review. Roughly speaking, one could perhaps say, that early, small steam explosions are the best guarantors, that large dangerous steam explosions probably won't occur in case of mixing hot melt with water.

Another fact, which limits the momentum of a steam explosion, is the high voidage in the mixing zone, extending over a large part ofit. This voidage has a strong damping eIfect on the migration afpressure pulses, becauseit offers a compressible volume.

The mixing deliberations and calculations, presented in the report, are phyJically well based and desent a high grade of credibility.

4. Quantification of explosion loads There are two key phenomena inRuencing loads of steam explosion. These are

- the mixing of particle clouds plunging into water and

- the mictointeraction beturen water and melt.

The first phenomenon was discussed in the chapter before. For describing the microinteractions between melt and water, the authors followed two ways. For describing the mictointeraction and for simulating the propagation of steam ex-plosions, they used the computer code ESPROSE.m. This code is based on a series of experiments - the second parallel way - which were performed in the so F-94

called SIGhfA-2000 facility /2/. Originally the formulations for the micrainter-action were based on the assumption, that the rate of coolant mixing between

) '

debris and water is proportional to the melt fragmentation rate. This is a reason-able assumption and by this it was possible to produce consistent comparisons from available experiments for a wide range of steam explosion loads, starting from weak propagations to supercritical detonations. The Brst formulations were mainly based on experimental results, obtained in the KROTOS facility. This Brst formulation was done for two dimensional geometries and could especially also demonstrate the mitigating effect of " venting", due to wave reRection at a free liquid surface. Supercritical d ...mations were observed in the KROTOS fa-cility with aluminium oxide melt only, pouring at very high temperatures into water.

in a next step, the constitutive equations were assessed by using experimental re-cults, obtained in the above mentioned SIGH 1A-2000 facility. These experiments were carried out with molten tin drops, having temperatures up to 1800 C, im-pinging into water. Of course one can argue, that there are scaling effects, if one wants to draw conclusions from the measured and evaluated data, gained in this small experimental set up, to the steam explosion loads to be expected during a severe accident in an AP600 reactor. According to the reviewer's opinion, these ecaling problems however are mainly with the mixing of particle clouds, plunging into water, a problem which was discussed in the chapter before and which was sohrd by the authors with the help of the computer code Ph! ALPHA.3D.

The SIGhfA-2000 facility was experimentally very well equipped and special measuring techniques, like radiography, gave very good quantitative information about the fragmentation of the drop mass and its distribution. The fragmenta-tion, measured with X-ray Bash, was reproducible within less than 20%, which ,

is a very good accuracy for such types of experiments. In addition the frag-mented melt was collected after freezing and was subjected to sieve analysis.

Very line fragmented particles were analysed via scanning electron microscope photographs. Generally speaking these experiments are a very reliable basis for assessing a computer code like ESPROSE.w-3D, according to the opinion of the reviewer.

In the SIGhfA-2000 facility, not only the fragmentation rate, but also the pres-sure signals of the steam explosions were recorded by using high speed pressure g

-t

. F-95

transducers. Due to the small scal af the facility, these pressure signals may be conservative when applied to a larg: 3cale geometry, like the downcomer or the lower plenum of the AP600. In a large volume, in which fragmentation of a hot melt starts, there are always voided areas, damping pressure propagation.

The veriBcation of the ESPROSE.m-code is very well documented in the report DOE /ID-10503. This report documents how the various etlects in steam explosion progress, like wave dynamics, explosion coupling and integral behaviour were assessed The report demonstrates how the code is handling pressure waves in single and in two-phase Buids and this not only in a one-dimensional, but in a two-dimensional geometry. Special attention was given to reBection and transmission behaviour. The comparison between predicted data (ESPROSE.m-code) and experimental results showed very good agreement for a wide variety of thermo-and Buld dynamic parameters. The local situations and the temporal behaviour are well predicted. So, the code is in a condition, that allows to predict steam explosion behaviour also beyond the experimentally veriBed area.

The extrapolation from the small scale to the large geometry of the reactor were done by using the basic equations for wave dynamics in multiphase media and constitutive laws for mictointeractions. The latter ones were renned via experi-ments in the SIGMA facility, also. The combined theoretical and experimental efiorts are a very good basis for predicting and simulating large scale conditions, also.

Finally one has to ask the question on " substance scaling" i.e., the applicability of the data, measured with modelling melts to liquid corium. The experiments were mainly performed with tin and with aluminium oxide. Especially aluminium oxide is very likely to produce supercritical steam explosions when it is mixed with water. The authors of the report DOE /ID-10541 write on page 2-1 (chapter 2

" problem deBnition and overall approach"):

"Also, it is important to note, that within the limited experience with reactor fuel material (UO2, ZrO2), we haw no evidence of explosions, but rather extensively voided premixtures (Huhtiniemi et al.,1995), nor is it known whether or under what conditions such premixtures can be triggered to explode".

With respect to " substance scaling" the data, presented in the report DOE /ID-10541, on explosion loads, originating from steam explosions are on the safe side without any doubt, becar:se a corium melt / water interaction will produce much F-96

. .~ . - . . - - -

3 softer pressure oulses than experienced in the experiments with aluminium oxide <

melt / water inte; actions.

5. Integration and assessment In the chapter 7 " integration and assessment",-- there are two very important ,

^

statements, nas 2ely that from a more global perspecthe, the only way "to potentially produce a sig-nlBcant structural challenge on the lower head, would be by having a highly >

subcooled poolin it" and

- "even a postulated rapid reBood scenario could not produce the ondition of concern..." .

After depressurising the primary system, following an hypothetical, severe acci-dent, there is always and everywhere saturated (not subcooled) water in the louer plenum of the pressure vessel. This would be true not only for the AP600, but .

also for all other pressurised water reactors.

So as long as one can guarantee, that the cooling of the lower core support Q ' structure is good enough to prevent it from failing and core melt Bows from the side to the lower plenum, steam explosions, originating from it, should not be a problem.

The second statement is as important as the first one, because it elimimstes doubts, exist!ng up to now, whether it would be advisable to try to Booo a degraded snre again after a certain escalation of a severe accident. This point l- was brieBy di: cussed already in a former chapter of this review. Therefore in future accident management planning, there should be given more effort to in- -

~ vessel cooling also after a partial core disintegration.

L- * * * +- * * * * * * * * * * * * * * * * * * *_ ,

l V

6. Conclusions

I fully agree with the conclusions presented in chapter 9 of the DOE /ID-10541 l report, to the statement of the authors, that *because of the wide margins, due to these controlling plysics, it has been possible to bound uncertainties to a

/  : suBicient degree...", I would like to add, that these " wide margins" are still on t

r( .

lA - F-97 L g 4 r ---

the conservative side and the n.echanical loads onto the pressure vessel and its g lower plenum would be lower in case of a hypothetical severe accident, than W predicted in the DOE /ID-10541 report.

Finally I would like to congratulate the authors to this fine work, attacking a very diflicult but important problem and solving it to a great extend from an engineer-Ing point of view, but based on controlling physics and on reliable constitutive laws for the fluiddynamics to be expected in steam explosion scenarios.

1 S. Angelini, T.G. Theofanous and W.W. Yuen, The Mixing of Particle clouds Plunging into Water, NURETH 7, Saratoga Springs, NY, September 10-15,1995, NUREG/CP-0142 Vol. 3,17541778 2 X. Chen, W.W. Yuen and T.G. Theofanous, On the Constitutive Description of the Microinteractions Concept in Steam Explosions, ProccMings NURETH-7, Saratoga Springs, NY, September 10-15, 1995, NUREG/CP-0142 O

F-98

d 1

' F.11. Resoonse to F.I. Moody General Comment and Highlights General and unqualified agreement with the conclusions of the work under review.

Point-by-Point Responses

0. Thank you for an opportunity to review this work. I think it is one of the m'ost significant pieces of research I have ever reviewed. It is of both current and long term importas. s to the nuclear industry. ,

Since I have spent my career in the nuclear energy business, I personally ap-preciate your long range viewpoint for energy needs, which is obvious from your support of this program.

• * * * * * * * * * * * * * * + * * * * * * *

1. The purposs for reviewing the subject report, with several other companion documents, was to assess whether "in-vessel retention" is demonstzated to be an effective severe accident management concept for a reactor like the AP600.

I have reviewed the work, and conclude that in-vessel retention has been shown to be an effective severe accident management concept for reactors with geometry, k fluid quantitles, event sequencing, and thermophysical properties similar to those pertaining to the AP-G00.

The documents provided for this review describe the steps taken to understand and predict the complex, multi faceted subject of steam explosions.. Associated phenomena have been closely simulated by experiments, and predicted with de-terministic theoretical formulations (causal relations) to a degree of accuracy that makan con & dent predictions possible for full size AP-600 systems. It appears that all controlling physical effects have been included, even without the need for a 9

complett understanding of the exact timing and conditions necessary to trigger steam explosions. Already known or conservatively estimated ranges have been ,

placed on parameter, timing, and scenario path uncertainties, and stillit has been shown that the expected range oflower head steam explosion pressure loads do not hatersect the vessel fragility curve.

I was asked specincally to review the material on steam explosion loads, as dis-cussed in g

.(/ -F-99 l

1-

" Propagation of Steam Explosions: Esprose.m VerlReation Studies" by T. G. Theofanous, W. W. Yuen, K. Freeman, & X. Chen, DOE /ID-10503, August 1996.

The documents provided for this review collectively lay an extensive foundation of information, which testines to the technical stature, competence, thorough-ness, and integrity of the imvstigators. Indeed, the overall work is monumental in its scope and achievewent, and it is communicated in a writing style which is one of the most scholarly to be found in reactor safety studies. Both ti.s au-thors and sponsors should be commended for a carefully formulated imestigative strategy (strong, in-depth, well-blended steps) resulting in the highest value ob-tained for the time and resources spent. Beyond steam explosions the progress and understanding achieved in this work are likely to exert a major benencial innuence, both methodological and technical, on other signincant and complex thermal-hydraulic issues.

SUhthfARY

1. The ROAAhi has shown that vessel loads, resulting from a comprehensive range of severe accident scenarios, melt conditions, relocation now, timing of release from the core region, and thermal-hydraulic processes between the melt and surrounding water, lead to the conclusion that vessel failure is "pbysically unreasonable" in an AP-600 type reactor. Parameters including pool geometry, melt release rate, shock explosive formation and propagation, and venting yield load distributions on the vessel wall which were compared with the fragility curve in order to arrive at this conclusion. It is my opinion that even though all the mechanisms contributing to steam explosions are not fully understood, results embrace the extent of rennements which could eventually be made by further experiments and theoretical model (causal relation) development.

2. I agree that it would be useful to obtain data from the QUEOS experiment for a fully saturated water system) although it would not change the conclusion that vessel failure in AP-600 type reactors is " physically unreasonable." The value in such a test is to Rilin a parameter range to give a more complete data base, and permit the technology to be extended to non-AP-600 type systems.

3. One potential benent of the ROAAhiprocedure is that it conceivably could be used in reverse. Suppose it was concluded that a system failure probability was F-100

larger than acceptable. The ROAAh! could be employed to display which pa-(p) rameter(s) dominate the outcome, thus pointing the way for design or procedural changes to reduce the failure probability.

2. 4. How does the ROAAhi accommodate different causal relations, such as Phi-ALPHA and ESPROSE.m, at ditferent stages in the methodology if they might be strongly coupled through common variab]cs? That is, the behavior of two systems alone may be altogether ditferent when they are coupled together (like tuo spring-mass systems). The probability distributions of the parameters imvived may combine differently when the separate systems are strongly coupled, leading to different probability ranges on the variables which determine success or failure of a system or process.

If such dependencies exist they can easily be accounted for. No such dependency can be identified here. Breakup and triggering are conservatively bounded with respect to both premixing (PM ALPA) and propagation (ESPROSE.m).

3. 5. The source term for area production in Appendix A of DOE /ID-10503 p is based on the assumption of particle number density remaining constant, while O their size changes. A bit more explanation orjustincation would help. Wouldn't it make more sense to predict interfacial area growth by the formation of more particles as the melt decelerates in water? Taylor instability was employed to obtain the Bond number criterion in interfacial area growth. Could that model be employed to obtain a fastest growing wave length and droplet formation?

During propagation the key mehanism is M'icrointeractions, and this involves fine scale fragmentation and mixing in the vicinity of all macroscopic particles. This is clearly supported by the SIGMA experiments. Both Taylor and Helmholtz instabilities have been employed in the consideration of hydrodynamic fragmentation. Ultimately it is more appropriate at this stage to rely on correlations suggested by such approaches and ,

the SIGMA data, which fully represent behavior in large scale explosions. This is our approach.

4. 6. It appears that in the heat transfer predictions of Phi-ALPHA in DOE /ID-10504, Bow regimes are identined by steady state correlations. Are these likely to be nonrepresentative for such transient events as fragmentation, and not provide a conservative characterization of the actual heat transfer?

'd F-101

i This question is not clear. Fragmentation is relevant to propagation (ESPROSE.m) not  !

i premixing (PM-ALPHA). Heat transfer of the fragmented debris is conservatively taken to be infinitely fa>t (thermodynamic equilibrium assumed in the m-fluid). If the reviewer  ;

is referring to breakup, this is a much slower process, that is not even predictable. The idea is to bound the behavior, and for this our constitutive treatment is quite adequate.

5. 7. Convective and radiative heat transfer from the fuel to the coolant is estimated in much detail, drawing from various experimental studies between coolant and heated solid surfaces. Is there a backup analysis to show that for the rapid heating associated with steam explosions, the heat transfer is not limited to how fast it can escape from the molten particles? Are there potential droplet sizes, relathv velocities, and Buld properties where internal conduction (or com'ection) might limit the heat exchange rate?

This really depends on the resulting size of fragments. For 1 m particles the time constant is 10-aps. We believe ignoring this limitation, by assuming instant equilibrium, is appropriately conservative.

6. STRATEGY The severe accident management strateg addressed invoh es the retention of core materialin the reactor vessel following a postulated severe accident in a reactor like the AP-600 design. Inability to cool the core leads to melting of core material by decay heat, and relocating it in stages to the reactor pressure vessel (RPV) lower plenum. Molten core debris, which may Bow to the bottom of the lower plenum can melt through the RPV wall and undergo release to the containment.

However, Booding the cavity to submerge the RPV bottom head is expected to be a means of arresting the dowmvard relocation of molten core debris.

Even ildownward relocation of molten debris is arrested, there is the possibility that some mass of debris could drop into water present in the lower head region, causing a steam explosion and further damage. Part of the mcrall study shows that failure of the bottom head by exceeding its structuralintegrity is " physically unreasonable".

THE RISK ORIENTED ACCIDENT ANALYSIS METHODOLOGY (ROAAM)

A primitive method of handling uncertainties in power systems came in the early 1960's (Moody F. J., " Probability Theory and Reactor Core Design," GE Re-port # GEAP 3819, US AEC Contract AT(04-3)-361, January,1962). One of F-102

, the greater concerns for a nuclear core during normal operation was reaching the 0,m " burnout" condition, where a hot spot in the fuel could exceed design limits, and cause fuel damage. The fuel temperature could be expressed as a function of several variables and parameters (causal relations), each with its own degree of uncertainty. If one chose the most pessimistic limit of each variable and pa-tameter, the " burnout" limit could be exceeded. The most optimistic limits the

" burnout" limit would not be exceeded. It was suggested that probability meth-ods could be applied to give a reasonable assessment of the likelihood of exceeding the " burnout" limit. Data from power plant operating logs was gathered to ob-tain probability distributions for certain variables and parameters. Wherever data was no. available, " expert opinion" was solicited. The results were then combined by the method proposed in an AShlE paper (Kline, S. J., and bicClin-tock, F. A., " Describing Uncertainties in Single Sample Experiments," hiechanical Dnrineerin.g, January,1957), which resulted in the expected mean and standard deviation for the hot spot temperature. Comparison with the established design limit showed that it was " physically unreasonable" to expect " burnout" in most cases.

The ROAAh!is an extensive, operational methodology which is more renned than O any of its primitive predecessors. It has the capacity for incorporating causal

%Y relations (describing equations relating the nriables and parameters), based on well-understood physics for the applicable phenomena, with specined parameter uncertainties, scenario bifurcations, and even a diversity of expert opinion. The process leads to a rationally-based prediction of those properties which determine the success or failure of a system or process.

The structure of ROAAh! embraces the current phenomenological state-of the-art, built-in activation response of safety and control systems, man-machine in-terfaces, and procedural understanding. As new information becomes available, the ROAAh! can accommodate it. Where expert opinions may be diverse, the ROAAh! provides a means of focusing further research to narrow the disagree- ,

ments. That is, when experts strongly disagree on the range of a parameter, the ROAAh! can be employed as a tool to display the sensitivity, showing if the parameter dominates the outcome, or is only a minor percentage effect on the overall result.

One qt'estion about use of the ROAAh! inmlves the causal relations for vari-aus phenomena. If the parameters in a causal relation are independent, their

~

/ 's e i V F 103

l probabilities can be combined in a certain way to obtain the expected mean and ,

standard deviations of that function. If the parameters are not independent, the combinatlon is more complicated. The question invohrs he,w the ROAAhiaccom-modates the possibility that some parameters appearing in more than one causal relation may not be independent. How uvuld results from ROAAh! compare with one deterministic mega-computation where all the parameters are treated by something like a monte-carlo process to obtain the distribution of variables which determine success or failure of a system?

One could always hard wire all the models in a ROAAM analysis to one mega-computation.

There would be no advantage (any dependencies can be handled just as easily), and there could be some important disadvantages; for example,in determining the bounding con-ditions for rate of breakup and trigger time. More importantly, sucn a mega-computation would be less scrutable, much less transparent, and much reduced in degrees of freedom practically explorable.

7. ROAAh! APPLICATION I have seen the ROA Ah! work in two separate campaigns to close severe acci-dent issues, namely the direct containment heating (DCH) issue for one series of PWR's, and the hfark iliner melt issue for one class of BWR containment. It is appropriate that this methodology should be applied to reach a conclusion on the in-vessel retention severe accident management concept.

Application to in-vessel retention embraces possible scenarios, melt conditions, coolant states, structural properties, debris mixing with water, triggering, explo-sion wave dynamics, and lower head fragility. Parameter ranges are associated with the amount of participating substances, the timing of events, event paths, and state properties of various subsystems. Several analytical tools, based on physical models, provide the causal relations employed, namely Phi-ALPHA for emcloping the etfect of melt breakup in water, ESPROSE.m for enveloping the ef-fects of fragmentation and microinteractions on steam explosions, and ABAQUS v.5.5 for emvloping the lower head failure esiteria. The computer programs used for causal relations to emelope important variables have be compared with other analyses and experimental data to a level where their predictive capability of the tested parameters does not introduce uncertainties which are significant enough to consider.

F-104

, The following comments are 08ercd to help substantiate my conclusion that in.

() vessel retention has been shown to be an c8cctive severe accident management concept for systems like the AP-600.

8. hfELT INTRODUCTION AND FRAGhiENTATION Early predictive models provide core melt scenarios and relocation rates with and without renood, which can arrest the melt progression. However, the melt state which may reach water in the RPV, and the subsequent breakup and penetration largely determine the rate of heat transfer, steam formation rate, and possible shock pressure loads. A quantity of melt arriving at the water can undergo Taylor unstable breakup or droplet formation at the leading edge and Helmholtz breakup or droplet stripping on those surfaces with parallel velocity components.

The Phi ALPHA model has been developed to incorporate the melt and coolant properties, and provide an emelope for the expected range of momentum, heat transfer, and phase change interactions associated with breakup for premixing considerations.

Single particle and particle cluster experiments have been employed to test pre-dictive capabilities of particle motion and energy transfer dynamics in water

(the hfAGICO and QUEOS experiments). Particle cloud elongation, steaming, spreading, and mixing with surrounding water are captured by the Phi-ALPHA code, which is employed as a causal relation in the ROAAhl Comparisons include particle cloud distortions associated with release door opening time, particle, and void volume fraction contours. Of particular interest is the pinching of the vapor volume behind moving particles, caused by condensation for the particle intro-I duction into subcooled water. Since the condensation acts to reduce me:' wical energy transfer, I agree that it would be useful to conduct QUEOS experiments in fully saturated water.

9. One of the most important considerations in fragmentation is the formation of new melt heat transfer area. Appendix A in DOE /ID-10503 describes the

" source term" for interfacial area production. Equation (3.69) is based on a change in size of particles for the same particle number density. It seems that before particles have reached a stable size? they would undergo the formation of new particles. Tids assumption needs more explanation.

( )

i ' o' F-105

i The source terms in interfacial areas are very different in " breakup" during premixing modelled in PM ALPHA, and in " fragmentation" during propagation modelled in ES-PROSE.m. This comment mixes up these two. The single particle approach is appropriate for fragmentation. For breakup we, in fact, have included both changes in numbers of large particles, as well as change of their size due to their shedding of very fine particles.

This is explained in Appendix A of DOE /ID-10504 (PM ALPHA verification).

10. STEAM EXPLOSION The mechanics of steam explosions are described in DOE /ID-10503, detailing melt introduction to water, interfacial breakup and premixing of debris parti-cles with water, the cfEcct of voidtag around the particles on heat transfer, the triggering of explosions, and propagation of pressure waves with reflections from rigid mechanical and gas-liquid interfaces. It was earlier found that 1.0 GJ of energy could fail the lower head. However, further understanding has led to a reexamination of the mechanics of steam explosion force generation to determine a more red!stic criterion for lower head failure.

It was determined that the AP-600 could withstand 500 bars of pressure for mil-liseconds without failure. Computations with the ESPROSE program displayed the difficulty in generating such pressure impulses with attenuating phenomena like voiding, which resists triggering, and pressure venting from the water sur-face. Extensive development of ESPROSE have been performed with both data from the SIGMA and KROTOS experimental facilities. Simpler analytical mod-els have provided assurance that ESPROSE accommodates detonations, shock propagation, and reflection.

Significant effects embraced by ESPROSE result from the physics incorporated, which are consistent with experiments. Calculations show the strong attenuatioa of shock pressure loads with distance, and time by venting from the water free surface in the AP600 systems. It is also realized that venting may not significantly reduce loads if the water depth is high in the lower head. Strong evidence is supplied that ESPROSE incorporates the appropriate physics, and can be used with confidence to provide the causal relation for emvloping the oflect of trigger time on steam explosion severity.

F-106 l

11 The physical mechardsms c msidered by ESPROSE.m include shock pres.

sure propagatlon from a trigger, which collapses volds, forces liquid onto the melt, producing fragmentation and mictointeractions, escalated heat transfer, further steam formation, and rapid expansion (explosion). A statement on page A 18 of DOE /ID 10503 needs further clarincation. Where the pressure increases rapidly ahead of an explosion front, why does the vapor become instantaneously sub-cooled? (if saturated steam is rapidly compressed, it would tend to follow an 1sentropic path off the vapor dome into the superheated region, not subcooled.)

What b meant is that the liquid becomes subcooled and steam is to rapidly condense.

12. On the same page, it is stated that behind the explosion front where pressure is <lectensing, the liquid can become superheated. (if you decompress saturated water, the path drops into the steam Wme.) It would make better sense to me (I can't speak for others) to note that the nonequilibrium states lag behind a steady state in the superheated or subcooled region.

Here, we refer to the liquid that participated thermally in the interaction (part of the

m. fluid).

13. LOWER HEAD RESPONSE Dynamic response of the lower head is based on well establishui physics of shells, modeled by the ABAQUS program. Mechanical failure of a shell depends not only on the magnitude of an applied load, but also on the fraguency content. It is stated that the shock pressure loads which lie in the steam explosion emelope have a short period relathc ta the structural response, so that the peak strain would be essentially independent of the pressure pulse time proBle.

The report has provided some " screening fragility" cunts which would be used to determine if predleted stcam explosion loads were of such a character that the failure criteria emvlope and fragility cune need to be further blended to ,

provide a failure likelihood it was concluded from the range of pressure loads and the lower head fragility curve, that for all relevant severe accident acenarins, melt conditions, and timing of release from the core region, w:th ensuing mixing and explosion wave dynamics, steam explosion induced lower head failure in an APG00-like reactor is "pirvsically unreasonable."

f \

D F-107 I

I

14. REVIEW OF STEAM EX'PLOSiON LOADS The verification of ESPROSE.m, based on stepwise experimental measurements and comparison with simpiined theoretical methods shows that reasonably con.

servaths assessments of steam explosions are possible in the present version.

The discussions of DOE /lD-10503 provide foundational support of the physi.

cal modeling and nuinetical procedures to predict steam explosion properties for given melt addition rates and states. Th; oasic physics inwive wave dynamics, including sound wave propagation and shock development and propagation in a water Biled region. Two-dimensional calculations performed by ESPROSE.m are compared with simpilBed computations using the method ofimages and solutions similar to classical waterhammer. Some comparisono are included based on char.

acteristic wlutions. 'The results form a strong basis for concluding inat the code is producing reaso.. le predictions for the expected range ofinput parameters.

Pressure propagation speeds, attenuation from wave interaction at free surfaces) and wave amplification by reBection from rigid surfaces have all played a role in the verincation.

Nutnerous tuv dimensional ESPROSE calculational surfaces are compared with solutions obtained from the method ofimages, and found to be sufficiently similar, leading to the conclusion that basic physics of explosions are included in the model. Several gwmetric parameters urre varied, as was the source velocity function. Good comparisons were consistently achieval.

The SIGMA testa invohrd a meh droplet which was triggeied at a specinc po-sition, leading in local pressure traces. Comparison of the pressure traces with ESPROSE caiculations showed reasonable tracking of pressure waves originating from the dropict region to the rigid end of the test section, and reBection back toward their origin. Additlonal evaluntion with the methad of characteristics urre provided. The wave dynamics, indeed, appear to be properly described in ESPROSE.

One piece of information Jack pointed out in the report is that the data base needs expansion for microinteractions with reactor materlais.

F 108

I l

l i

15. Tile NEXT STEP p/

\

I understand that a number of experts are providing reviews of the documents provided. Some may believe (as I do) that even without a complete understand-Ing of all the phenomena, the remaining uncertainties, processed by the ROAAh!,

still permit a stcong statement about failure likelihood being " physically unrea-oonable." Some experts tuay feel that the uncertainty of a given parameter should be broader. This is a simple exercise in ROAAh!, which uvuld then provide out-put with a te 3e that accornmodates the particular variable uncertainty. Other experts may wish to change the causal relations to reRect "stlous " bottom up" or line structure effects. This is always a possibility, but may be unnecessary, since the causal relations are based on macroscopic formulations of basic principles. If it were recommended that nonequilibrium models be employed for causal rela-tions, we uvuld be farther behind than using ROAAh!In its present structure, because nonequilibrium models would have to be verined by experiments.

When strong disagreements have resulted in physical modeling, small working groups have been formed to reach agreement on acceptable formulation, with appropriate modlBcations in ROAAh!.

Finally, it is possible that some would disagree with the ROA Ah! structure itself, suggesting that it skews results, or simply blurs our ignorance of phenomena.

I uvuld argue strongly that the ROAAh! blends (not blurs) uncertainties (not ignorance) In a way that makes it possible to reach conclusions with a known level of contidence. .

i G. OVERALL CONCLUSION As a curious person who erdoys formulating better theoretical models, based on more complete experimental understanding, I recomn end additional experiments (e.g., QUEOS experiments with fully saturated water) to help close the few re- ,

maining gaps in our understanding of steam explosion phenomena.

llowever, I believe that the studies provided for this review ght substantial, in.

depth evidence to help conclude that in vessel retention is supportable as a severe accident management strateg in AP-600 type reactors without additional work to close the issue.

• * * * * * * * * * * * + * * * * * * * *. * *

\

U F 109

E12. Response to RR. Schcal(RIT)

General Comment and High!!ghts This is a generally agreeable review. Many detailed questions are raised, but these are mostly of a clarification and reinforcing character, rather than strong objections. Also, many useful suggestions and opinions are offered, again in the same light. Perhaps the key point is that the review expressed caution with regard to the maturity of the analysis tools. If this refers to a stage in the " phases of development," table (Table A.2 in Appendix A), we certainly agree.

Point by Point Responses

1. L Review of the Overall Approach This is the fourth time i have had the opportunity to review a body of uvrk that Professor Theofanous and co-workers have produced for the resolution of a speelBc safety issue, or a specific concern. I believe, this is the most complex of all the issues (or concerns) so far and I believe, Professors Theofanous, Yuen and co workers have done their Bnest work so far. This body of work is ofgreater, and of more lasting, value, than earlier efforts, since a major part of this work is the development and verificalion af the methadology to describe the stcam explosion phenomena, and to predict the loads imposed by the postulated occurrence of a steam explosion. This methodology, and the codes develoI>ed, could be applied to other accident scenarios, than the one considered in the present application.

I believe, some comments are in order on the overall approach followed in these tbree reports, complemented, of course, with the ROAAM methad, and the pre-vious work that Professor Theofanous and his teams have performed, (e.g. for the Alpha mode failure of an IMR containment during a severe accident).

Prolessor Theofanous and co warken. with their accumulated experience in stcam explosion modeling and applications, have developed a very well focussed overall approach in the body of umrk presented in the three reports. It is cicar that an in house experimental program was structured to provide the key observations, for the ideas needed, to advance the steam explosion modeling to the point where some meaningfulpredictions can be made. The innmative experiments performed in the MAGICO facility provided the germane ideas on steam depletion, and on the difficulty of obtaining pre-mixtures, which could lead to very large steam explosions. Likewise, the experiments performed on the SIGMA facility provided F-110

i l

n the basis for the micro interactions concept for the steam explosion itself, i.e., the O concept and treatment of the m Buid. I believe, the experimental underpinning of the ideas and concepts employed, and the further verliication oi the methads used in the codes against the integral experiments, has provided great strength to the overall approach.

The overall apyroach followed, in the application report, conforms to the ROA Ah!

method and employs the Phi ALP:lA and the ESPROSE-m methodology. The extremely high values for the fragility curve made the task much simpler than the earlier appilcations of the ROAAh! methodology, but it is welljustified and credible.

Perhaps, the two points of possible short coming in the overall approach, which have also been admitted by the authors, should be stated:

First, is the question of maturity. Clearly, there is not enough separate-cliect and integral effect data to provide sufficient validation of the steam explosion methodology developed. This methodology employs many many correlation and submodels, whose individual verincation is a monumental task. Nevertheless, an experimental verification matrix should be developed, with priorization of im-p portant cliects, and executed, to provide greater verification of the methodology, Q thereby providing it greater maturity.

Second, a mechanistic treatment of the initial phe.se of the steam explosion sce-natio, i.e., the break up of the melt jet, and its sequential fragmentation, has not been included in the methodology developed so far. The authors claim that this phase of the steam explosion process can be conservathely-bounded param-eterically. Perhaps, the authors have done that successfully in this study, how-ever, a more general treatment of the break up phase, and its linking with the pre-mixture phase, should be pursued to provide greater assurance that all the initial-condition etTects have been taken inta accour '.

The above two points, in no way, diminin the s ue of the overall approach, '

and the results achiemt The above two are outlins of further work to solidify tha validity of the overall approach followed hue, as, I believe, the authors have themselves identified. The present treatment of the physics is the

• State of Art."

I believe, rapid advances in understanding and modeling will follow the germ of ideas that the authors have provided here. Some of those advances will surely be accomplished by Professors Theofanous, Yuen and co-workers.

p.)]}

2. 11. heview of the Report DOE /ID-10504 (Sept.1990) PREh!!XING OF STEAh! EXPLOSLONS: Pht ALPHA VERIFICATION STUDIES" by T.G. The-ofanous, W.'W. Yuen, S. Angelln!

This report is the veriBcation document for the Code Pht ALPHA, which treats the premixing phase of the steam explosion scenario. The report has two im-portant appendices: (a) which describes the Phi-ALPHA models and (b), which describes a set of experiments in the h!AGICO-2000 facility, in which several kilograms of high temperature particles of a speciBc material, and of speclBc di-ameter, are dropped into water to obtain observations and data on the pre-mixing geometries and void fract!ons. The front parts of the report provide the compar-Isons of the predictions with the Phi ALPHA code against the data from selected experiments. In the following paragraphs, I will provide comments on the main sections of this report.

!!.1 Anvendix B. "hilXING OF PARTICLE CLOUDS PLUNGING INTO WA-TER" I am wry impressed with the h!AGICO facility. I believe the authors have per.

formed outstanding experiments using quite high temperatures and respectable masses of the hot particles. The video pictures are outstanding. I am a bit disap-pointed with the quantitative data that could be obtained. The X ray pictures (in reproductions) do not communicate any information and the void fraction data shown in Figures B.23, B.25 and B.27 is rather meager as a validation standard.

Point well taken. It was just too expensive to provide original prints of the X-rays. But we have more data now and quantitative reconstruction of the radiographs, provided in color (see addendum to Appendix B).

3. The coluparisons of the Phi ALPHA predictions to the measured data, shown in Appendix B, for the cold runs, show substantial differences in the ad-vancement of the particle front. It appears that a central part of the particle cloud tunnels through the water. This is not predicted well by the code. For the hot runs, it appears from Figures B.26, that the calculations predict that the dense particle cloud also leaves the steam region behind, if a slight subcooling (3*C) is present. There are no comparisons shown for the hot runs, as shown for the cold runs in the Figs. B14 and B.15.

F-112

s The question that should be asked is how much detail is necessary, and reasonable, to 3 (o) ,

expect in predictions? The cold runs, sometimes, exhibit a front instability, and this might well be due i6 some slight experimental variation such as, a slight delay in particle release, from cylinder to cylinder. Taken as a whole the PM ALPHA prediction of cold runs is very good, and the PM ALPHA.L predictions are excellent. At the time we had only very localized X ray data;however,now see Addendum to Appendix B.

4. The concluding remarks state that the hot tests quantified local voiding in the mixing zone and global voiding through the level swell. Figures B.25 and D.20 indicate that the voiding front is coincident with the particle front, only, for the zero subcooling case. The particle front is substantially ahead of the voiding front for a slight (3*C) subcooling of the coolant. The extensive ste:un generation, indicated by the axial void profile also may increuse the local subcooling by pressurization. I wish there was some quantitative data for the particle volume fractions, to compare in Figures B.25 and B.26. Was it not possible to obtain quantification of the spatial particle volume fractions from the X ray pictures?

Yes, we now have quantitath e information of particle distributions, and more importantly, on the relation of the particles to the void front. See new materialin addendum to Appendix

5. In this context, if the PM ALPHA predictions for the advancement of the particle front lagged behind the measurements in the cold runs (cf. Figs. B.14 and D.15), would they not do the same for the' hot runs, since same modeling is emphryed for both hot and cold runs. I do expect that the steam generatlon, caused by the radiathe heat flux on the coolant from the particle cloud, will retard the advancement of the particles. I believe, this etfect is represented in the .

code, since a radiation heat flux model is employed, however, I can not quantify its ellect, on the differences in the particle cloud distribution between the hot and the cold runs. .

In this, and the above two paragraphs, reviewer's concem is on whether the particle-void fronts and their relation are properly calculated. We have more detailed and complete data from MAGICO now that completely address this concern. These can be found in the addendum to Appendix B. Also, the PM ALPHA.L predictions should help alleviate this concern.

fh F-ll3

l l

l

6. The subcool~d coolant is important. The only data shown for the 18*C subcooling case is the lack of measured level swell. I would be interested in the axial void fraction and the particle volume fraction profiles, to understand ii there are significant phenomenological ditferences between the saturated and the subcooled cases, and if these difTerences can be predicted by the Phi ALPilA Code.

As noted in the report, there was no measurable void in the 18 'C subcooled case. This was very well predicted by PM ALPHA. Subcooling effects were key also in the interpretation of QUEOS, MIXA, and FARO tests - see respective sections DOE /ID-10504.

7. Allin all, I believe the hfAGICO experiments are relevant for the ideas, and data, on the mixing zone and the premixing conditions. I would like to connect the melt jet particulation to the particie-cloud water interaction. This may be in the next phase of authors experimentalinvestigations.

Yes, we plan some work in this area, but it is outside the present effort.

8. 11.2 Annendix A. " Phi ALPilA: A (COMPUTER CODEFOR ADDRESS-ING Tile PREh!!XING OF STEAM EXPLOSIONS" Phi ALPils is a three (melt, coolatu and vapour) field code employing separate mass, momentum and energy equations for each ficid. Thus, it is a very detailed code more detailed than the codes RELAP-5 and TRAC. It also employs two and three dimensional geometry. Thus, it has capabilities beyond those of the con <rntional CFD codes, which, generally, employ only a single field. Phi ALPilA is a very advanced and detailed computer code, indeed.' There are other codes, currently in development, in Europe, e.g. IVA (Siemens, Germany) and hfC 3 D (CEA, I+ance), which are also incorporating similar capability, in order to treat the very complex, and very dynamic, physics of melt-water interaction and steam explosions.

it is a general rule that more detailed the formulation for the description of a process, more detailed the information required to bring closure to the formula-tion; and more intulthrly intelligent approximations have to be made to obtain credible solutions from the formulation. This is quite apparent for Phi ALPHA, when a whole page (A 20) is needed, to show the dimensional groups that ap-pear in the constitutive laws for the fuel to coolant heat transfer. This can not F-ll4

c be avolded, howewr, the collecthe constituthr laws may provide reasonably-correct predictions for a particular set of pre mixing circumstances, and not for another set. I believe, that veriBeation on an even less integral level than the MAGICO experiments should be considered by thinking through, and devising, a set of separate-ellect experiments. They should be prioritised, so that the most important are performed Erst.

We agree with this perspective, and this is why we have gone to great lengths to test the individual pieces as well (see for example Figure 1 in the Introduction; we suspect the problem here is that the reviewer chose to go through this report in reverse order, from back to front). For example, the one page equations referred to here were obtained from an experiment, the MUPHIN, conceived and carried out specifically for this purpose. So, the question really is whether we have left out something important. This is addressed in responding to the reviewer's specific comments below.

9. In the following paragraphs, I will provide some detailed commer.ts.

11.2.1 P_M ALPHA Formulation The modeling approach is logical and well thought. The authors admit that the e formulation so far, emphasizes the ruultineld aspects of pre mixmg. The melt Jet Y and particle break-up are treated parameterically. ,

Two length scales are employed for the fuel Reid: one large, encompassing the original fuel drops, or fuel melt jet, which may break up but still are considered as fuel; and the other small enough to be called a debris, which mixes with water and gets quenched. The decisions about the amount of the ' fuel' and the ' debris' are made with a correlation for the fragmentation rate.

The debris particles assume the same temperature, and velocity, as the coolant, .

Instantly. They are not allowed to sediment down with gravity, as they would normally do. This assumption is justlBed for the time interval considered, if the particles are of micron size.

The " admit" carries a derogatory sense to it. In fact, we presented a rather detailed rationale of our approach, and we would prefer to have heard specific comment on it.

10. The large-length-scale fuel particles are assumed to have uniform temper-ature. There is no treatment of the heat conduction from the fuel particle to the U F-il5

}

coolant. For thu prototypic binary-oxide mixture melt, it is linportant to deter-mine the solidification front growth into the particle, since it may either prevent fragmentatlon, or reduce the rate vifragmentatlon, thereby changing the heat source to the coolant.

Another factor in the treatment of the fuel particles, is the change in physical properties that occurs, as the fuel particles cool down from aboveliquidus to below solidus temperature. The increase in viscosity and surface tension affect the fragmentatlon charactcristics, which in turn allect the terms in the debris mass equation, and in the liquid and debris momentum and energy equations.

A paper submitted by Okkonen and Schgalin the forthcoming FCI meetlug in Japan discuss the two factors mentioned aban for the behaviour of the fuel drops.

Particles that cannot breakup, certainly cannot support an explosion. Ir fact, we have an option in PM-ALPHA for crust growth, and we have used it in the analysis of some ex-vessel explosions in deep, subcooled, water pools. This was not emphasb:ed here, because the times and depths and other conditions are not conducive to significant solidification effects.

11. Recently, we at RoyalInstitute of 7bchnology (RIT), have performed some experiments on the interaction of relatively low temperature cerrobend (an alloy with density of s=9000 kg/m3 ) Jets with subcooled water. We have found that the Jet breaks up into small particles. There is a distribution to the particle size or mass, however, there were no particles oflength scale comparable to the jet diameter. In these experiments the jet breaks up' completely. The FARO experknents show a melt " cake" at the bottom, however, it is not clear whether it is the unbroken Jet or an agglomeration of melt droplets belonging to some size distribution, which, perhaps, does not contain length scales approaching the melt jet diameters.

Summarizing the above discussion, I believe, the treatment of fuel as having two length scales in the PM ALPilA formulation is valid. Ilowever, the source terms in the equations should be reviewed again. The variation of properties of the fuel drops, with temperature, should also be taken into account; and the change in the temperature of the fuel drop should be calculated employing conduction equations. Melt jet, or drop, interactions with subcookd coolant may produce atomization, with no large particles of size similar to that of the melt jet.

F 116

n As noted, the source term in the equations are varied parametrically to bound the behavior.

() No one knows what these source terms really are, and it is highly presumptuous at this stage to take an approach based on simulating breakup. This is not our approach. Solid-ification effects are not important for in vessel explosions, and even more so in saturated water pools. Moreover, ignoring this small effect is conservative. Finally, the atumization effect is accounted for by our breakup at two different length scales, but when this occurs to a very large extent it works against premixing.

12. IL2.2 Interfacial biomentum 'nansfer in Pht ALPHA The drag correlation used in I"1 ALPHA for fuel coolant interface distinguishes between the dispersed and the dense fuel reghnes. The latter is taken as that for Bow of gas through a densely-packed bed. This correlation, perhap.e. should be checked, since predicted penetration of the fuel cloud in the MAGICO experi-ments is less than the measurements. Also, comparisons could le inade with the isothermal tests in the BILLEAU and the QUEOS facilities. The logic diagrams on pages A 10 and A 17 were helpful.

The densely packed bed regime appears only in particles accumulating against a bound-ary. We have no problem in the PM ALPHA predictions of fine cloud penetration (see also (y above). Comparisons with QUEOS were provided in the body of the report. BILLEAU tests are not yet available.

13. IL2.3 Interfaelal llent 'I>nnsfer in Phi ALPilA There are many regimes of convecthe heat transfer and many correlations. The nuthors use the best that tbey can find. Then, there is the large elfcct af radiation heat transfer, which was found to be important for the comparisons to the QUEOS test data. Their synergism, and elfects of one regime on another, may need further exploration. For example, radiation absorption will produce upour which will change the comecthe flos patterns of the coolant, and, perhaps, change the heat trar.sfer regime. Some separate-effect tests could be designed to test the synergism and the elfect of different comvctive regimes on each other, in order to test the heat transfer correlations package employed.

We have explored these avenues already, but there isn't really much new or surprising.

Even for film boiling from single spheres, contrary to what one might expect, the super-position approach works very well Also,it should be noted that our efforts here were not limited to collecting what we could find. The major components are non local radiation U F 117

heat transfer, and film boiling in single- and two phase media. For tne former, we formu-lated a whole new approach, and for the latter, we conducted the MUPHIN experiments and developed theories and correlations for use in the code. So really we do not agree g

with the thrust of this comment.

14. 11.2.4 Thel Break-Un and Fraementation hfodeline in Phi ALPilf I have referred to this carlier in the comments on the PM ALPHA formulation.

The interfacial area equation (3.73) assumes spherical particles on break up and fragmentatlon. This may not be approprinte. Perbaps, data from FAftO or ather fragmentation break up experiments could be employed to develop a more proto-typic interfacial area representation, in sotne of our experiments with cerrobend in subcooled water, we do not find spherical particles. Perhaps, in saturated water, with large flows of steam the particle shapes may be spherical.

The cerrobend particles are not spherical, because the material solidifies at such low tem-peraturest The dimension in Eq. (3.7) should be interpreted as characteristic length, or effective diameter. The source at this time is parametric,because these is no reliable model.

However, this is sufficient for our purposes.

15. The model for fragmentation of fuel drops is based on the Bond number.

I believe, data on hydrodynamic and thermal fragmentation of large-size melt droplets saay be available in near future. The model could be checked against such data, when available.

The model for jet and large fuel drop-break up is parametric with an input-specified parameter, f, whose value is varied in analysis. This approach is, per-haps, adequate for the present. However, it will be desirable to have a phe-nomenological/ mechanistic model.

Such models already exist, and probably others will be created. The problem is, how you propose to adequately validate them for reactor conditions.

16. The authors distinguish between fragmentttion and break up *s two sep-arate processes la some of our melt jet water interaction experiments, we were not able to separate the two processes. The jet breaks up (or fragments) into particles having a size distribution ranging from submillimeter to 3-4 millime-tres. The process appears to be concurrent and not sequential, as assumed in the parametric models described here.

F-118

Not at all. Our formulation is for concurrent, not sequential processes. One should be O

v careful in how far to take the cerrobend data.

. 11.3 VERIFICATION OF the Phi ALPHA CODE The front part of the report DOE /ID-10504 describes the verincation pursued for the Phi ALPHA code by1>erforming analytical tests, and by comparing with the data measured in several experim*nts. This was a very large eEort, and I believe, it han largely achieved its purpose. I will comment on a few comparisons of the data with the code predictions.

11.3.1.QUEOS Exoeriment These experiments are sim!)ar to the blAGICO experiments. The comparisons shown in Figures 4 to 13 are remarkably good for such a dynamic process. The comy isons appear to be better than those for the hfAGICO tests.

It is not clear to me what the experimentalimage actually implics, in terms of the distribution of hot particles, and of vold. The pictues in Fig. 7 at 0.3 and 0.4 seconds seem to show that the experimental hot particle image may be not as advanced as the calculated contour. This also tappears to be the case in Fig 6. at l 0.3 and 0.4 seconds. The graphs in Fig. 8, however, show very good agreement l between measured and calculated front advance locations versus time.

The QUEOS tests have provided only photographic images of the interaction zone, and these are as clear to us as they are for the reviewer. As noted in the report, the front advancements in Figu.e 8 are from Lagrangian particles in the calculation. The Eulerian results show some numerical diffusion, and again, as noted, the results in Figure 7 were to be refined by calculating with finer grids. This is done now (see addendum to Section 2.2.3). .

18. 11.3.2 hilXA Exoeriments The bilXA experiments employ a Uranium-hfolybdenum thermite melt of several kilograms, at 3600K, poured into near-saturated water pools. The melt Jet was

oken into 6 mm diameter droplets. The blLXA 6, analysed here, used 3 kg melt pour into wry nearly ($1 K diference) saturated water. This, thus, is a prototypic experiment, albeit with small mass.

The comparisons are very good, I am somewhat concerned about the sensitivity l of the results to the break up-cut-off- vold fraction and particle size. The authors l

I ]C - F 119 l

l 1

)

recognise this, still, a change of only 5 % (85 % to 80 %), with the parthcle size of1 mm, decreases the calculated pressure rise from 0.4 bars ta n 0.2 bars. Increasing the particle size from 1 mm to 1.2 mm at the 85 % cut-offlevel decreases the pressure rise from n 0.4 to n 0.28 bars. Thus, the breakup and fragmentation models appear to be very influential in the very high temperature, prototypic material e:~periments.

d i

'Ihis is absolutely correct, and simply states the obvious fact that the steam pro uct on is a strong function of interfacial area and the couplings created thereof. This also pro-vides important perspectives on the whole question of breakup - what can resonably be expected imm a prediction, and how far could such possible predictions be taken!

19. 11.3.2 FARO Exnerlments These are, perhaps, the most important exper-Iments, since they use substantial quantitles (?> 100kg) of prototypic materials; and there are several experiments already performed and more are underway.

Tne comparisons shown are very good indeed. Unfortunately, FARO does not produce any data on the mixing region, thus the colour figures, presented, show only calculations and no data.

20. I did not understand why the initial particle size is chosen as 4 cm for a jet diameter of 10 cm. The f value chosen is 50, while for the hilXA test it wm chosen as 20. The minhnum particle size chosen is 1 mm, while in the hilXA test is was chosen as 1.2 mm.

It is inconceivable that the jet exited the nozzle and travelled all the way to the pool, totally undisturbed. We chose 4 cm as a large enough characteristic length scale. It does not matter really what you choose, as the process is really controlled by #, and only small scales have enough interfacial area to interact.

21. One experknental result, which FARO produces is the fraction of the Jet material deposited as a ' cake' on the bottom plate. This is not provided by the authors (wm their analysis with the Pht-ALPilA code.

As the reviewer notes in an earlier comment, it is not really clear what this ' cake' means, and there are many ways to interpret it.

F-120 u s ,

- _ - - - - . . - . _ - .~.. - . - -._ _ - - - - - - - - . -

r t

22. 11.4 Numerical Aspects  ;

The authors do not provid: a discussion on this topic. I beliew, this is an impor- -

tant topic. The ICE technique is known to have signincant numerical diEusion. i it is not clear whether any advanced space time discretization scheme was em- l played. Node sises of several centimetres are generally not Bne enough. The  !

authors, perhaps, by now, han Investigated the numerical aspects further, and I

[

would welcome a greater discussion of this topic.

There was nothing special employed, and elsewhere in the report we note that we may introduce such a special scheme at some future time. We now have PM-ALPHA.L lt was

~a lso noted that numerical diffusion can be well enough controlled for our purposes, by i choosing fine enough grids. A brief sample of results were already included. More can be found in the addenda.

23. 111. Review of the Report: Propagatlou st Stcam Explosions: ESPROSE.tn VeriScatlon Studles by T.G. Theofanons, W.W. Yuen, K. Freeman and X. Chen This report deals with the next phase in the steam explosion process, after the pre- i mixing has been acidered. The report, therefore, deals with the explosion process and develops a methodology to describe the process, and evaluate the energetics, O which are then employed to assess the damage potential of the explosion on structures, which surround the explosion. A trigger is assumed, which strzts the expicslon process, in which intimate contact of the fuel and the coolant leads to produc:lon oflarge amounts of vapour, and the supercritical explosion.

The report consists of the front part, where the.rcsults of the verincation cal.

culations are compared to the observations, and data, obtained in the SIGMA facility at U.C. Santa Barbara. The report also contains four important appen-dices in which the code models, a 1-D characteristics model, constituthe laws for micra Intersctions and thermal-detonations are discussed.

In the following paragraphs, I will comment on each of the major sections of this report.

111.1 Appendix A.- THE ESPROSE.m MOLJELS The overall approach of the model development is brilliant. Recognising that

' the dynamics of a pressure wave, generated by a trigger, coupled with fuel frag-mentation, micro (or local) mixing and heat transfer result in energetic steam F-121 x ,

, _ - . . , . . - . - . - . . . . . - - - - _ . - . ~ _...a-,.__.., ., . . , - - , - , -

explosions, the authors have concentrated on those aspects. Perhaps, the SIGhfA l

exg>eriments provided the key observatlons towards the devetppment of the micro-interaction concept and the m Buld, where the fuel-coolant heat transfer occurs.

The en rgy trausferred is then employed in the multinuid trentment to calculate the pressure helds as a function of time and space (2 D/ 3D). The damage pcu tential is, then, evaluated with the calculated dynamic loading imposed in terms of kilo Pascal seconds.

The modeling approach is similar in most respects to that employed for the Pht-ALP 11A code, i.e., solution of a set of multiSeld conservation equations, with speciRed constituthc relations. The fields chosen this time are fuel, liquid and the m Buld. There is an additional mass conservation equation for the debris, i.e. the fragmented material. The m Buld equations contain source and sink terms, which are based on a picture ofliquid entrainment and phase change.1%)

fragmentation is included, which contributes to the increase in interfacial areas.

The heat tram.fer across the fields is included in the energy equations. The system of equations appears to be complete. The constitutive relations between the 3 ficids for interfacial drag, heat transfer and phase change, again involve many correlations and dimensionless numbers.

24. I believe the comments that I had made regarding the complexity of the constitutive relations for the Phi-ALPilA code also apply here, and the possibility of checking the synergisms between the momeatum and heat transfer processes through separate-eflect experiments, should be explored. New data may have to be obtained and some prioritisation should be performed.

Actually, in this respect, ESPROSE.m is much less of a challenge than PH-ALPHA. All we have is wave dynamics, which were verified very carefully in a step-by-step approach, and Microinteractions, which is really obtained experimentally under conditions that fully simulate large-scale explosions. So, we really do not see the concern expressed.

25. The fuel fragmentation is treated as in the Phi ALPflA code and is con-trolled parameterically through f. There is another parameter which enhances the frogmentation ior thermal eflects. Both oithese parameters are user-speciRed.

The entrainment ofliquid in the m Buld is controlled through the parameter E, which is taken as a function of the fragmentation rate.

F-122 g

l n Actually, we used Of to distinguish .' rom the d used in PM ALPHA.

() 26. I believe, the parametric treatment is very intuitive, and the authors ad-mit that it is an important component of the micro-interactions concept with somewhat speculative constitutive laws. Since, the m fluid interactions are the basis of ESPitOSE-m, I hope that the authors have already obtained additional data from the SIGhfA facility to provide greater support for the experimental basis of the parametric treatment.

These paramete are fixed by the SIGMA experiment. We now have data also for iron and gallium (see addendum to Appendix C), and plan ZiO2 tests in the near future.

27. 111.2 Appendix C. Constitutive Laws of hilcro-interactions This appendix describes the experiments performed in the SIGhfA 2000 facility whh gallhtm and mohen tin, subjected to high pressure waves, in order to derhv the constituthe laws for the micro-interactions, needed for the m fluid.

The experknents are described. They are really very difficult, but precise exper-iments. Some results are shown as movie, X ray and SEh! Images for the change in pre mixing wlume, as n function of thne.

v The results of experiments are used to derhv the values for bf, ne, and fe, the entrainment factor. For example, Fig. C-13 shows f, = 7, 8 and 12 ght best fits, ress>ecthrly, it t three isothermal Gaillum tests i.e., G/204/45, G/68/45 and G/272/45. T% more conservative value f, a 7 is then used to determine the value of Of = 9. The value of as derbed from Fig C-10, while keeping f, = 7, and 0j = 9. It appears that ne varies from 1.4 at 68 bar pressure to 4.? at 204 bar pressure.

The above is a logical but highly empirical determination of 3 parameters from a small number of tests. Perhaps, more data has been obtained from SIGhfA to confirm the choices made for these key parameters. Obviously, more data is needed from SIGhfA or another shock tube. I believe, different materkds should also be tested, in particular, melt drops of binary oxides. Their fragmentation behaviour maybe difTerent, due to changes in properties they experience with a change in temperature, b)'

L F-123

For new data with iron melts see the addendum to Appendix C. As noted already, data with ZrO ,2 and if needed, UO2/ZrO ,2will be obtained in SIGMA 3000, currently nearing operation.

28. lill Appendix D. On the Existence of Thermal Detonations This is a very interesting re-examination of the Doard-Hall model for steam ex-plosions. The micro-interaction model and the concept of the m Buld is employed to show that supercritical steam explosions can be obtained with lean mixtures in highly volded regions; conditions for which the Board Hall model will pmilct only very weak explosions.

hly understanding of the microinteractions cor pt, introduced by the authors, is that they take place in the m Buld in a lir ed volume. I believe, this results from the observations made from the Gallium drop (also perhaps the tin drop) experiments conducted in the SIGhlA facillty. The previous concept was that the pressure wave will fragment a melt drop into fine droplets, which will mix with the whole coolant volume. The SIGhlA experknents chowed that this does not occur in the time frame of the pressunuwave melt drop interaction. The heat transfer to the m Buid s coolant, in the limited volume occupied by the m-Buld, generates very hlgh pressures. The shock wave then travels in the non-participating Buld around the m. fluid, increasing its pressure to sustain the propagation. This makes possible the supercritical stentn explosion with a fuel-coolant mixture, which is lean on an overall volume basis, but is not so ican on the m Buld vomme basis.

(Cj. Figures D-8 and D 9, where high pressures are obtained for the cochmt to debris mass ratio f, = 1 in the case of tin at 1500*C and for f, = 2 to 8 for the case of UO2 at 3300*C)

I believe, the authors have provided a very logical explanation and frame work.

I am, however, a bit concerned about the ndue for f,, which was chosen as 7 in Appendix C, based on the data from the gallium experiments in the SIGhfA facility. In Fig D-8, a value of f, = 7 will not produce a supercritical steam ex-plosion. Thus, the value of fa may be material dependent, and more information is needed to choose an appropriate ndue.

Yes, indeed. Figure D-8 shows results for tin at 1500 *C and we know (KROTOS, for example) that such premixtures do not give strong detonations.

F-124

l

. 29. 111.4 VERIFICATION STUDIES

\ The front pr.rt of the ESPROSE. m report describes the analytical tests, the l SIGhtA experiments, explosion coupling, integral aspects, numerical aspects, and finally, a comparison with the KROTOS tests. I will comment on these, brieBy, indhidually. 1 111.4.1. Analytical Tests These are very valuable exercises and show that the modeling in ESPROSE-m can predict pressure wave propagation. There are many Bgures. I wish there were more explanations e.g., Figures 17 and 18, both show very good comparison between the analytical and the ESPROS m pressure distributions for early times, but deviate at later times. Is there an explanation? Similarly, there is a crater in the middle of the pressure wave in Fig 19. Is there a physical explanatloa for that? This section may be unproved by the authors, through some explanatory text. It is very valuable, otherwise.

The slight derivation in Figures 17 and 18 at later times and also the " crater" in Figure 16 (we believe that the reviewer meant Figure 16 in his comment) are due to the non linear effect at high pressure which is present in ESPROSE.m (the runs are made with the full code with real properties) but not in the analytical solution.

O 30 111.4.2 SIGhiA Exocriments These experiments, specially conducted in the SIGhfA shock tube provide data for verincation of the ESPROSE-m models for pressure wave propagation. The comparisons are excellent. There are some differences for the inhomogeneous cases, which, perhaps, are dlBicult to fix. All in all, it is a splendid performance for the code for these separate-effect tests.

31. 111.4.3 Comparison.< with KROTOS Experiments KROTOS experiments provide the most appropriate data for the veriBcation of the ESPROSD m models. The KROTOS facility has performed steam explosion experiments by triggering the pre mixtures of water with several different material melts. The initial conditions, e.g. melt mass, melt temperature, melt superheat, l

pressure, water subcooling have been varied to provide a reasonably extenshe data base. The test program is continukg, and could provide the data base needed for the ESPROSE-m validation. Unfortunately. as in most of these melt water b)

U' - F-125

l

\

Interaction integral experiments, the data obtained is integral and the premixing and the stcam explusion processes are not delineated. Thus, detailed verification and validation of the ESPROSE-m (or any other steam explosion code) may not be possible.

The document provided on the analysis of the KROTOS tests speculates that the melt breakup and quick freezing may be a reasonable explanation for the non explosivity of the Uranium oxide tests. We reached similar conclusions, and also, evaluated the eifects oI the change in the surface tension and viscosity af the binary-oxide melt, as it cools down below the liquidus temperature. This has been reported in the 1995 ICONE meeting, and additional work will be reported in the forthcoming FCI meeting.

32. Coming back to the comparisons of ESPROSE-m (using Phi ALPHA pre-mixtures) predictions against the measured data, the authors admit diBiculties of representing particle freezing correctly in the Phi ALPHA formulation. The fuel-participatica factor chosen affects the result greatly. The pressure waysshapes versus time appear to be reasonable but there are differences e.g. for K5 there appears to be an earlier venting of pressure wave. I believe, revision of the Phi.

ALPHA numerical scheme and/or modeling of the heat conduction in the fuel particles (as was mentioned carlier in the comments on Phi-ALPHA modeling) may resohv this difficulty. The sensitivity to fuel participation factor is very large, indeed.

Disagree. The key ingredient here is the rate of breakup, which is not known, and will remain so. Also, as we discussed, there are intricate radiation reflection issues peculiar to the KROTOS geometry. These are not code issues; rather, they are physics issues peculiar to the test. The sensitivity to molten fuel content is real, not a code artifact. Again, this should provide important perspectives as to what is really predictable for these kinds of problems, and correspondingly what should be a viable strategy in safety assessment:.

33. HL l.4 Numerical Awects I have similar comments es I had for this topic in the Phi ALPHA document.

The authors should provide more discussion and, perhaps, comparisons of the use of the ICE tecimique for similar problems. The numerical diffusion issue is quite important when, tracking pressure waves and/or interfaces. Recently, F 126

l special numerical schemes have been devised to reduce or eliminate numerical diffusion. The ICE technique does not, compare well to such schemes, in term of its performance, and with respect to numerical diffusion. Perhaps, the authors have implemented another scheme in developing the ESPROSE-m-3D code.

Disagree. This comment is not consistent with the code performance presented in this document. l

34. IV. Review of the Report: Lower Head Integrity Under in-Vessel Steam Explosion, DOE /ID-10541 by TG ' aeofanous, W.W. Yuen, S. Angelini, J.J.

Slenicki, K. &ceman, X. Chen a- T. Salmassi This report is concerned wit. aswering the quesHon: "Will the lower head of the advanced passh e reactor Al '00 fall, under the dynamic loading imposed by an in vcssel steam explosion, ifit were to occur?" This is an important issue for the accident management strategy chosen for the AP Gl0, i.e. retention of the core melt h) the lower head, by employing external cooling of the vessel.

The methodology used to resolve this issue is the ROAAM method developed by Prof. Theofanous, employed most recently to respond to the companion question p "Is it possible to retain the molten core of the AP-600 reactor, in the lower head by cooling the vessel externally?" This question was answered in the allirmaths by employing the ROAAM method. The ROAAM method has been extended and further clarined by Prof. Theofanous in a recent pubilcation, attached as Appendix A in this report.

13csides the ROAAM philosophy and procedures described in Appendix A, the detailed premixing and explosion results are described in Appendices B and C respecthcly. Appendix D provides additional pre-mixing perspecthes from the THIRMAL code, prepared by Drs. Chu and Sienicki of Argonne National Lab-oratory. The important chapters, in the main body of the report, are concerned

. with structural failure criteria, melt relocation characteristics, quantilimtion of pre mixtures and explosion loads and Snally the assessment of the integrity of the lower head of AP-600.

In the following paragraphs, I have provided comments on the appendices, chap-ters and conclusions . the report in the order:

G ,

b F-127

Chapter 3: Structural failure criteria

- Chapter 4: Melt relocation characteristics

- Chapter 5: Quantification of pre mixtures Appendix D: Detailed pre mixing results

- Appendix D: Additional pre mixing perspectives froin the 7tHRMAL code

- Chapter 6: Quantification of explosion loads Appendix C: Detailed explosion results

- Chapter 8: Consideration of reflood FCis

- Chaptcrs 7 and 9: Integration, assessment and conclusions

35. IV.1 Chapter 3. Structural Failure Criteria This is an important chapter, since it establishes the fragility curve, giving the probability of the lower head failure for dynamic loads ofincreasing magnitudes.

The impulse loading, ofinterest, is in the range of 100 to 300 kilo Pascal seconds.

The nu ' hors have employed a commercial structural-analysis code, whose results i:v i ave compared with a simple analytical solution. ADAQUS is a 3 D tinite element code, able to model the hemisphericallower head and the dynamic load-Ings imposed. The code provides the strain as a (twetion of time for the assumed loading. These calculated results are, then, converted to a fragility curve, as-suming probabilities oflower head failure, when strains of greater than 11 % are reached over certain fractions of the lower head wall thickness.

The ADAQUS calculations are performed for various loading patterns on the lower head. The non-uniformity ofloading was found to decrease the strain for a specific impulse. The colour pictures provide very nice strain morphologies.

This chapter provides clear and transparent results. The ADAQUS results are confirmed against a simple model for uniform loading. The fragility curve makes good sense.

F 128

i

, 36. I am a bit concerned about the very local non homogeneous loadings of

) the type predicted, later, in the report. Perhaps, a few ABAQUS calculations could be performed to establish the fragility curve for such a local loading pat tern.

We do not understand this comment. Most if not all of the development in this chapter is for " local non homogeneous loadings of the type predicted later in the report."

37. IV.2 Chapter 4. Melt Relocation Characteristics Th!u chapter provides the initial conditions for the scenario of melt water interac-tion in the lower head. The chapter, therefore, deals with the melt pool formation in the originni core boundarles and, later, relocation of the melt from the in-core location of the lower head. The quantitles needed are the rate of melt addition to the water in the lower head, thejet geometry (diameter, velocity and location in the vessel), the melt composition and superheat and, finally, the timing of this event relative to the other events in the core mell progression process.

The authors, first point out the differences in the AP-600 core conBguration from that of the comentional PWR. The AP GOD has some features which are quite favourable in terms of the melt releases conditions. These are the massht 3G cm thick core support plate, the core reBector, the gap between the core barrel and Q the reRector on the Ba* sides of the reRector; and the long unheated section in the fuel elements at the bottom.

The authors have dewloped a credible scenario of melt pool formation, melt attack on the reRector and the core barrel. It is supported by enveloping models of appropriate complexity, which provide physicallusight and transparency. The authors are wise not to use one of the myriad codes, which provide user-motivated results. The analysis is brilliant and quite comprehensive. The melt releasa conditions of 200 to 400 kg/sec should be bounding values. The melt superheat of 180 K also should be a good bounding value. The location of the release, near the top of the core in the vessel downcomer, may also be credible. The jet velocity of few meters /second also appears to be sound.1, however, would like the authors ta consider the following cautlonary points:

38. (1) The timing of the melt release 76 to 91 minutes is much too close to the timing of ml00 minutes for evaporation of water in the bottom 25 Yo of the core height by the radiathy heat Bux imposed.

p V F 129 l

In addition, there is a huge margin in the heat capacity of the massive core support plate.

See also addendum to Chapter 4. g

39. (ii) The core plate is massive but it is also loaded heavily. If the core plate temperatures go beyond 70TC, the yield strength wul deteriorate.

The core plate is fully supported by the core support structure. Also note (addendum to Chapter 4) that it really takes a long time for the lower part to heat up.

40. (iii) The melt pool with u40 to 60 % unoxidized zirconium and some stainless steel, will probably form a primarily metallayer on the top. This layer is thin and wlH focus the heat Rux to the sides. ncent work at RIT has evaluated the heat transfer from the metallayer to the vessel (which is of a thickness similar in that of the reRector) with a tuv-dimensional code, and found that the highest heat Bux is stin at the corner of the oxide pool just below the metalHe layer.

Thus, the failure could be below the metallayer.

We do not agree with such a result, but we need to look at the RIT analysis mentioned.

41. (iv) While, I agree with the authors that the Bat part of the reRector being closest to .he core centre is most likely ta be attacked Bret by the pool. The oxide pool however may not be axially symmetric and there may be azimuthai regions in the core, where fresh fuel and high power are dominant. Evaluation of a possible attack on the non Bat parts of the reRector shouid be considered.

Failure at the corner would produce a more localized release. Failure on the flat is conser-vative.

42. (v) The draining and freezing of the metallic layer into the well between the Bat part of the reRector and the core barrel, without participation in any melt-water interaction, is very likely, but sounds too comenient. AdditionaHy, in the absence of water above the core plate in the well, the thermal loading imposed by the superheated metallic molt on the core plate, or on the core barrel region directly above the core plate should be evaluated.

There is no water between the com barrel and the reflector at this time in such an accident.

As we have shown, the core plate still would be cooled by water.

F-130

43. Summarizing, I believe, the authors e<timates for the range of melt-( release-characteristics is credible, however, additional evaluations may help to put these estimates on a more solid footing.

More evaluations can be found in the addendum to Chapter 4.

44. IV.3 Chapter 5. Quar' ncation of Pre mixtures / Appendix B. Detailed Pre mixing Results The chapter 5 denlops the rationale for the pre mixing that results from the release of the UO 2 . ZrO2 melt from near the top of the core, through the down-comer, into the water pool of the lower head. The water level is assumed to be a few centimetres above the top of the core support plate. hielt release rates of 200 and 400 kg/sec, reacidng the velocity of 5 m/sec at entry into water are considered. The melt superheat is assumed as 180K.

The oxide melt Jet is distributed over an effective radial width of 10 cm in the downcomer, with an initial melt volume fraction of n25 % at water impact. This would translate to a melt stream of dimensions n10 cm x 16 cm for the release rate of 200 kg/sec and nio cm x 32 cm for the release rate of 400 kg/sec.

O The expanded melt Jet is then allowed to traverse 20 mm in water, before break-up ensues. The break up rates are parameterized from no break up to very rapid break up (forming 2 mm size particles within 10 cm of travelin water.)

The above initial conditions were employed in the Phi ALPHA code to provide results ou pre mixture characteristics 1.c. the melt and the void volume fractions and the fuellength scale, as a function of time, and position. The integral quantity of interest is the number of kilogram of melt mixed with coolant, before the triggering and explosion.

The Appendix B presents a number of colour pictures and many graphs giving detailed results. The graphs of fuel length scale, fuel volume and void fractions are presented for more f values and for times up to n 1 sec. These pictures and graphs provide good back-up for the results, and arguments, presented in chapter 5.

I beliae the authors have presented a clsar method of evaluation and the results are credible. I do have the following comments.

• * * * * * * * * * * * * * *

• w * * * * *

• f3 b F-131

t

45. (1) The melt through failure of the reflector and core barrel are assamed I

to be near the top of the melt pool in the original core boundary. If the failure is lower, the starting velocity for the melt Jet would be higher, and so will be the velocity at water impact. This may be beneficial for break up.

We do not expect much difference. Twice the initial depth would increase initial velocity from 1 to 1.4 m/s, and the velocity of water impact remains essentially unchanged at 5 m/s.

46. (ii) The initirl impact i.rea on the water surface is quite large. The Jet going through the 2.acter steam region should not break up, to that extent.

There is also a splash oi the wall. More concentrated pours create higher voiding in the premixture, so we try to bound the behavior here, too.

47. (iii) Both the very fast and the no break up cases show (Cf. Figs. S.4 (a) and 5.4 (b)) that for the initial 0.1 see the fuel froat is more advanced than the void fraction front. This was also observed in the PM ALPffA verL6 cation report. Later on, the void fraction front seems ta catch up with the fuel front.

For the C 1 10 case at 0.4 seconds (Page D.3 3) a large fraction of fue: seems to be hung up in the volded zone. The same is true for C 1 - nb case (Page B.3-5).

In the C110 case, there would be a large steam flux rising, which could retard the descent of the fuel particles. For the C 1 - nb case the steam aux should he smaller, and the fuel particles of 2 cm should be ahead of the void fraction front.

48. Summarising, I believe the break up assumptions, both, in the steam during descent from the original core boundary, and during water interaction, play a crucial role and, perhaps, this part of the pre-mixing analysis could be strengtbened. The no-break up case appears to produce approximately the same results as the high break up case. This has been recognised, also, by the authors (Page 510). Perhaps a physical explanation of why these cases produce such similar results may be provided by the authors.

The results show that all cases become highly voided. This result is mally expected, given the radiative power of such melts. See also addendum to Chapter 5.

49. IV.4 Appendix D. Additional Premixing Perspecthes from the THIRMAL Code F-132 h ,

In this appendix, the THIRhfAL code has been used by C.C. Chu and J.J. Sienicke (J of Argonne National Laboratory to provide a perspective on premixing. The code had to be modined to describe the melt jet water interaction in the con &ned geometry of the down comer. The calculations were performed for melt release rates of 14 to 220 kg/sec, with correspondingjet diameters of 18 mm to 73 mm.

The 220 kg/sec case resulted in median droplet size of 2.75 mm, with a mixing zone radius and void fraction at pool surface of160 mm and 74 %, respectively.

These results are not too dillerent from what were obtained from the Ph! ALPHA Code, although the Jet entry conditions are different. THHth!AL calculates Jet entry diameter of 6 cm (l.c., no break up in the down comer steam zone). blodels for break up in THIRhiAL must be quite different from & parametric model employed in the Pht ALPHA Code.

50. IV.5 Chapter G. Quantincation of Explosion Loads / Appendix C. De-talled Explosion Results The chapter 6 and Appendix C present the results of explosion-propagation cal.

- culations performed with the ESPROSE-m code, using, as initial conditions, the pre mixture conBgurations calculated with the Pht ALPHA code. The trigger time is chosen as very short, since during the early thne the void fractions of the coolant around the fuel particles are relathrly low. Later, the void fractions increase substantially, and would inhibit fuel break up and triggerability.

The results are presented for the C-1 and C 2 scenarios with three values of f and a set of trigger thnes. For the no break up case these times vary - from 0.05 sec to 1.0 sec, while for the break up cases, they vary from 0.04 to up to 0.19 seconds.

The results on pressure, impulse and effecth*e area are shown for various locations in the lower head. Peak loadings histories are also shown as a function of trigger times. The extreme sensitivity to trigger time is e"Ident from Table G.1. If the trigger is delayed by 0.06 seconds for the C1 10 and C2-10 cases, there is only a very nak explosion. For the Cl 20 and C2 20 cases, there appears to be a time interval of only 15-30 msec for the trigger to generate a supercritleal explosion.

Thus, triggering time appears to be the deciding factor. A physical explanation for this extreme sensitivity should be provided by the authors.

,/m k F 133

w Very simply it has to do w!A the combination of voids and melt length scales. It was discussed already how these two compensate each other, while at both extremes, high voids or large length scales, we have benignbehavior, it is clear that the explosive quality of a premixture will maximize somewhere in between. A new way of plotting the results (see addenda toChapters 5 and 6) illustrate this very well. Also, new results with intennediate

1. values should help further in understanding this.

51. The Appendix C gives very nice pictures of the pressure wave traversing through the lower head. The pressure signals at various points in the lower head are shown and the peak pressures and impulse loadings are shown as graphs versus time. Theses pictures and graphs were very helpfulin the review of Chapter G.

Summarising, I can say that the authors have performed logical analyses of the loadings imposed by the steam explosion, and have provided very nice results.

I have not underst ' 4 the reasons for the extreme sensitivity of the calculated results to the trigger time. The peak loadings shown in Table G.1 are, in general, modest. The highest loading is found to be m200 k. Pa. s. Is it possible that for 0 = 30, a higher value than 200 k. Pa. s. is calculated?

See response to question above (Sch50).

52. IV.6 Chapter 8. Consideration of the teBood FCI's This chapter deals bricBy with the stratified steam explosions that may result, if the renood is eifectIve, and a layer oI water is brought on top of the melt por !

which has a metallic layer, at top.

It was found that the stable water kner may not exceed 10 cms, due to the low reflood rate and the time to freeze the upper metallayer. Any stratined explosion will be easily wnted.

I believe, the authors have a good argument. Certainly the peak pressures in such an explosion should be low and renood FCI s may not be a problem.

IV.7 Chapter 7. Integration, Assessment / Chapter 9. Conclusions These chapters combine the results achieved in the previous chapters and appen.

dices to provide an overall assessment. This work was already practically done to the results achieved, since the maximum impulse loading was below the min-imum of the fragility curve. This was also confirmed by performing ABAQUS calculations for the peak loading for the actual cases and Ending that the lower head strains were very low.

F 134

l l l

The authora conclude that for the saturated water case, the lower head integrity

()

m can not be compromised by a steam explosion. Having highly subcooled water is the only possible way to, potentially, im ohn a larger mass of melt, and produce a more energetic explosion. The authors conclude that obtaining highly subcooled water, even m re!!ood scenarios for the AP-600 is not credible.

53. V. Concluding Remarks in this section, I would like to provide a few concluding remarks after the review of the three reports.

I must congs atulate the authors for p >ducing such a fine and comprehensive body of work treating the tricky and controversial area of steam explosions. While, most of the researchers in this area are still trying to understand tLe fundamentals, the authors have leaped ahead with new concepts, advanced codes and considered judgements to provide a reasonably robust estimation of the damage potential of a steam explosion. They base combined this with structural analysis to show that AP-600 lower head can withstand the dynamic loads imposed.

The authors have, also, notad the pxuliarities of the AP-600 configuration and employed the advantages and disadantages they confer on the analyses. Some ,

f\

Y of these peculiarities (differences) provide great advantnges e.g. in the core melt progression rad the melt release characteristics. These sound a little bit too comenient and, perhaps, should be re-visited.

The work was done without any regard to the " convenience" of the results.

54. The authors have modelled the fuel break-up and fragmentation process only parametrically. This may be a weak point it, the whole development; since those processes provide the initial conditions for both the pre-mixing and the propagation phases af the steam explosion. Perhaps, the analyses are well-bounded for these processes; hourver, the sensitivity of the results to the break-up and the fragmentation modeling is very large.

We do not agree with the thrust of this conclusion. Only breakup is treated parametrically, not fragmentation. The main result is that premixtures void, and this is not too sensitive on the breakup used. We bound the behavior with respect to this parameter, and this is much more reliable than trying to assert the result of some predictive model. Such mod-els, even if eventually developed, could never be verified at the appropriate level -i.e, g

U F-135

the dynamics of breakup as it occurs under realistic conditions. Our treatment of frag-mentation, on the other hand, derives from the directly applicable and well-characterized g SIGMA experiments. The e.ddenda to Chapters 5 and 6 show more clearly the bounding W

nature of the results.

55. Then, there is the question of maturity and of validation scrsus verliica-tlov. I believe the methodology and the data presented, rob.sst as they are, are still very new. The comparisons presented against test data are not extensive, and I think, the authors recognising this, have wisely titled the reports as verifica-tion reports. Further experience with this methodology and further comparisons with separate-eiTect (e.g. SIGMA, hfAGICO, BILLEAU and QUEOS) data and integral etlect (e.g. FARO.and KROTOS) data would provide validation and maturity to this methodology. In particular, the constitutive relations, being so many for such complicated phenomena, need greater experimental back-up.1 be-lieve, the authors are already busy in achieving such experiments in the hfAGICO and SIGMA facilities.

Actually, we did not intend to make a distinction between " verification" and " validation."

In the sense tha+ the validation term is described here, we believe the two codes have been adequately validated in the " fitness for purpose" sense. Of course, work will continue, and thus maturation will gradually develop.

56. Lastly, I must say that I have erdoyed reading the reports and learned much from them. I think, I now understand the concept af micro-interactions and the m fluid. I have made constructive (hopefully) critical comments at places, to provide input to authors towards improvement of the reports. I believe, they have largely acideved the objective they had set out to achieve.

• * * * * * * * * *

• k

• F-136 g

F.13. Response to P. Shewmon (OSU) c 'N (V General Comment and Highlights General and unqualified agreement with the conclusions of the work under review.

Point-by-Point Responses

1. The analysis of head failure sets up a model of the lower head using ABAQUS (a well established finite element code) to relate the stress pulse from steam explosion to local stram. The vessel material (ferritic SA508 steel) will undergo large amounts of strain (elongations of 50 to 100%) before fracture oc-curs. Whether or not the vessel undergoes any plastic strain depends on theyield stress of the metal and the impulse from the steam explos:on.

2. For reason never explained or discussed, the authors chose 330 blPa for the yield stress of the vessel. They state that the consenathc ' Code Allowable' is 345 hiPa and the actual value (found in a comentional tensile test) is 450 hlPu.

The choice of 330 hfPa introduces a large conservatism (safety margin) since a best estimate should use 450 hiPa.

O V The real reason for using 330 rather than the actual 450 MPa value for yield stress is that we could not find the actual value until much of the work had been done with 330 MPa.

The additional margin due to this is now discussed in the addendum to Chapter 3. On the other hand, this margin could be used to compensate for the strain rate effect, if the latter is doubted (see But4).

3. The impulse applied to the steel in the lower head would have a rise time of a few milliseconds. When ferritic steels are loaded this quickly their yield stress is substantially greater than that obsenvd in a normal tensile test. The authors quote references that show the yield stress at this rtrain rate is about 40%

greater than that found in a tensile test. They take full credit for this strain-rate increment, which is justified and appropriate.

In summary, the analysis of head failure seems to be competently and conserva-tively done, and the conclusions dir,wn are appropriate. I have also looked at the ,

discussion ofloads and loading. I am less of a specialist in this area, but it also seems to be well done.

(")'s

'_ F-137

4. Though no mention of radiation effects is made in the reports, the analysis should be made for the vessel at end oflife (40 years?). The temperature of the head during the accident considered would be less than 212 F. This is beneath the RNDT for the beltline of some of the vessels now in service, i.e. such material might well behave in a brittle manner during an accident of the type considered here. I considered this, but feel such radiation etfects are not germane in ti.e case of the AP600 for at least two reasons:

1 ) The fast neutron and hard gamma flux in the lower head will be at least a couple of orders of magnitude less than that in the beltline region of the vessel, so radiation effects should be negligible.

2) The steel to be used m the AP600 vessel should be appreciably lower in the elements than have lead to radiation embrittlement (copper, and phosphorous) in the older vessels now of concern in p!snts in the U.S.A.

Wita this in mind, I believe there is every reason to believe that the material in the lower head would behave in a ductile manner and that the analysis given in the report is appropriate for (would apply to) a vesselin the AP600 after 4s years of service.

The end-of-life RTNDT values for AP600 steel forging at the beltline region is specified as 23 *F. The lower head,less irradiated, would be better still. At the time of interest, the lower head would be between 50 and 100 *C, F-138

I F.14. Response to B. Turland (AEA)

O General Comment and Highlights This review is hard to interpret at this stage. Concerns are raised about almost every aspect of the analysis and supporting documentation, yet we also obtain the impression that these are offered in the spirit of further improving the basis for the conclusions rather than in challenging them. The key point appears to be the one made in closing the first section of the review (Overall Comments), that is, "my residual concerns relate to the confidence in having low pour rates and the possibility of operator actions leading to some subcooling". In this response, as well as responses to several other related questions by other reviewers, we present additional material that hopefully will be found helpful in coming to a resolution, or if not, to better focusing any remaining concerns.

Point-by-Point Responses

1. OVERALL COMMENTS This report and its associated documents represent the culmination of several years work by Prof. Theofanous and his colleagues. They have now demonstrated that the basic framework for a steam explosion assessment in realistic geometry is in place. This is a major achievement.

O The reliance on detailed modelling codes makes the reviewer's task difficult - in the end one can look at the validation offered and consider whether the results presented look reasonable. In the supportbg documents the authors make good use of the available experimental data to benchmark their calculational models.

However, it is accepted that some of the constituthe physics used in'the premix-ing and propagation codes is uncertain, as are, to some extent, the melt pour characteristics. A review, such as this, can indicate that the codes app =ar ' lit for purpose' but cannot give a full endorsement for all the models they contain, without significantly greater elfort.

The situntion considered in the application presented, a modest pour of melt into saturated water at ambient pressure, is not .;onducive to large steam explosion loadings, and this is demonstrated by the calculations presented. Sullicient pa-tameter variatlons are imtstigated to indicate that this is likely to be a robust result for these conditions. As indicated below, my residual concerns relate to the confidence in having low pour rates and the possibility of operator actions leading to some subcooling.

i> F-139

2. SPECIFIC COhihiENTS Chapter 2: Problem Definition nd Overall Approach

1. Although the text maxes clear ' hat it was an intentional conservatism not to claim credit for lower head venting in the Sizewell B study, it is wrong to interpret this in the phrase lower headfailure cannot be dismissed as readily any longer. We found that the previous claims for lower head failure could not be substantiated as large explosions did not necessarily imply sustained high pressures.

• * * * * * * * * * * * * * + * * *

3. 2. It is arguable whether the ' essential basis for the current work'is the progress made in modelling explosion propagation and the pre-mixing phase, or in the assessment of melt progression.

Look at the pressure pulses predicted, the fragility, some of the reviewer's comments, including this one, and then let us imagine where we would be without such progress in premixing and propagation (microinteractions in particular). m

+=

4. 3. The statement that between 3 to S tons offuel must participate to pmduce a i GJ explosion, and consequently incipient lower headfailure, raises the question of whether larger explosions are possible that do not fall the lower head.

This is an old result. Probably yes, but this is not our concern here.

S. 4. I agree that the size of any breach is Indeed a tough question. I consider it to be the key question unless the mixing / propagation analysis by itself can be shown to be sufficient. I do not believe that this has been shown to be sufficient (as yet?).

See specifics below

6. 5. The higher explosivity of a prernixture with reduced voiding appears to be a conjecture that is not fully supported by the experimental evidence. Reported explosions in the JAERI ALPHA facility occurred with large voids in the mixing region.

Explosivity refers to intensity, not likelihood. In any case, there will be peripheral zones of low void fractions and nothing prevents these regions from initiating and propagating F-140

, i... plosions. The voided regior'.s simply damp the energetics, and reduce the amounts of (v) fuel exploding coherently.

7. 6. I need to be cominced that we need only to be worried about thefirst relocation event. I think that it depends on the timing of any subsequent relocation events.

Once the path opens, subsequent relocations should be essentially continuous.

8. Chapter 3: Structutal Failure Criteria

1. In principle, the loading may have compongnts both shorter and longer than the natural timescale of the vessel. Indeed the constrained expansions considered in the early studies do have a longer timescale. One needs to refer ahead to the results of the propagation modelling to justify the assumption through early venting of the explosion region.

A big part of the argument is due to the volded and higMy localized nature of the premix-tures.

9. 2. The boundary conditions on the ABAQUS model are not specified -

from later examples they appear to be symmetry conditions at the equator of a (U} sphere. As potential explosions may occu: near thejoin of the lower head to the cylindrical section, it is not clear that this provides a good choice (apart from validating the simpler model - which could have been done in 1-D).

The explosion loading occurs well below the hemispheric / cylinder juncture, so our choice is reasonable, and an economical approach to develop the necessary understanding for localized loads. In Chapter 7, we show full-vessel simulations also.

10. 3.1%om a non-expert viewpoint, the analysis presented in this Chapter appears a reasonable approach. However, i did note that Figure 3.9 was not consistent with Figure 3.4. To support the mitigative factor for local loading, more highly localised ABAQUS calculations should have been performed. To avoid eso falling to zero for finits values ofI and do/D,, Isuggest a auming that energy dissipation is proportional *o the magnitude of the effecthe impulse.

Valid point. More calculations were carried out. There was a problem with plotting figure 3.9. In the corrected figure the e34 goes to zero properly. See Figure 3.9 attached. The suggested idea gave about the same quality of representing the calculational results. The w/ F-141

new, more localized ABAQUS results support well the generalization in Figure 3.9 (see addendum to Chapter 3) up to a point. For do/D, < 0.25, Figure 3.9 becomes increasingly more conservative.

1

!=0.05 MPa s 0.8 -

!=0.35 MPa s ,

04 -

3 -

0.4 -

0.2 -

I '

0 e 0.2 0.4 0.6 0.8 1 d,,0, Figure 3.9. The mitigative effect of localized loading as a function of the impulse applied and the degree oflocalization.

O

11. Chapter 4: Quantification of Melt Relocation Characteristics

1. it would be useful to give an indication of the diameter of the cooling holes.

They are ~1 cm in diameter.

12. 2. I do not see that the heat sink associated with the core support plate plays a significant role, as water provides the major heat eir.k. In the absence of water, melt passage through the plate would depend on the diaceter of the Bow channels. Melt appears to have passed through relativelt small diameter holes in the presence of water at TMi-2. IE these holes are similat to that oi the hole in a PWR lower core plate, then they probably ofter little resistance to melt Bow.

The lower core plate would prevent large diameter pours penetrating the lower plenum, if downward relocation were to occur.

No. The heat sink is important in delaying the blockage, above, from melting, after the water has vaporized, to a level below the core support plate. Yes. The holes in the plate itself would offer no resistance toinelt penetration.

F-142

13. 3. At this stage (page 4.1) translating We expect this path to be blocked' (v ) into ' Downward relocation is physically unreasonable'still appears a large step.

Yes, but the statement gives a preamble of where we are going in this chapter.

14. 4. Reference to Thil-2. Looking again at the Thil-2 melt relocation ennt, I am struck by how far melt managed to progress downwards through the core, ginn the water imentory that is generally believed to have been present (minimum of 0.5 m above the base of the core). For instance at position K9 near the centre of the core there was evidence of previously neolten material between rods near the first spacer grid and in the spaces around the lower end fitting [Thil-2 Core Bore Acquisition Summary Report, EGG-Thil 7385, rev 1, February 1987, page B-30]. This relocation was physically reasonable, as it occurred, but I do not see how it differs substantially from the claim that the APR-like core downward relocation is ' physically unreasonable.' I am happy with the notion thut relocation into the bypass (most PWRs) or downcomer in the APR-600 is most likely - it is the degree of certainty that I question.

We really mean it is " physically unreasonable" to penetrate through the bottom of the core, and the TMI information noted does not conflict with this; indeed, it supports it. In the present case we have also the Zr end-plugs as a further cold trap.

15. 5. The low melting point control rod materials are expected to escape early (page 4-4). This seems counter to other arguments about heat sinks. However, if they do form a bloQage, this may be relatively weak, giving the potential for a later downward relocation.

Yes, but these are intermediate states in melt progression. We are interested in how far this can go.

16. 6. While the results on blockage formation appea realistic, the thermal equilibrium assumption in equation 4.1 is inconsistent with the growth of thermal boundary layers in the solid represented by equation 4.2. This may lead to an underprediction of the plugging time, particularly for cooler structures.

The error is negligible in the context and timing of this freezing.

I 7. 7. Table 4 what is the meaning and signincance of Meltfreezing capacirv as multiple of thefuel rod volume? For comparison (I think) one needs the channel

,q b F-143 l

volume divided by the fuel rod volume to ensure that there is sumcient heat capacity to form a blockage.

We think ours is an interesting measure because it includes thermal effects.

18. 8. Page 4 6: The efective thermalconductivity is taken as the volume weighted average. Here and elsewhere it would be useful to indicate what physical proper-ties were used. The use of a volume weighted average is probably reasonable for this application (but not generally so). Was any allowance made for the porosity of the blockage in this evaluation.

Yes. See addendum to Chapter 4.

19. 9. Page 4-7. While low melting point components of the core such as control rods are expected to relocate as they melt, this does not apply to the major metallic component - Zr. Best fits to experimental data indicate that relocation following clad breach occurs at temperatures in the range 2400 - 2450 K. Belocation imvives a significant fuel component.

This is still well below the oxides melting temperature, and the fuel content is limited by the time available for melting and dissolution.

20. 10. Nomenclature: Equations 4.8 and 4.9 refer to Cr,wn whlie Figure 4.6 and 4.7 have Craa,w,1 etc.

Typo corrected.

21. 11. The radial heat up calculations (Section 4.2) are qualital:vely in line with similar calculations that we performed for Sizewell. Was a radial power deposition shape factor used? We found that somewhat ditferent results were obtained when we did the same calculation using a 2 D, rather than cylindrically symmetric model, that took account of the proper core geometry and the power rating of individual assemblies. A dimculty with both your and our model is the absence of relocation, which may imalidate the model once any melting of material occurs.

Yes. Radial power factors were used as shown in the report. Clearly we do not expect to predict the details of relocation with this mode'., but this is not our purpose nor is it needed. Orce relocation begins overall energy conservation is sufficient to take the relatively smaller way up to the melt pool formation.

F-144 h i

l l

22. 12. The assumption of a fully oxidised pool (Section 4.3) may be inap-

&p propriate for a low pressure .tequence. This raises issues on the interactions of ,

the corium with a more metallic blockage (partially addressed in the hfP tests).

However, to conclude that something is ' physically unreasonable' all processes that may have an impact should be discussed.

The point is that the cold trap at the bottom forms and sustains the blockage. Any inter-actions with the oxides on the top of it can do nothing to violate.its integrity. All we need is the heat flux from above, as we have done.

22 13. The proposed melt release conditions and mechanism appear reason-able. The dimensions and the pour rate are no n ore than educated guesses (I would probably have made similar guesses). I note that to achieve the melt Bow rate ofim/s, only a 5 cm driving head is essumed, although there is no quantin-cation of how close to the top of the pool we breach might occur. It is desirable to analyse whether heat transfer from the melt stream through the breach may deepen the breach and lead to an increase in pour rate.

Certainly will, but not significantly in a matter of ~1 second. It is the time coherence here that determines the reasonableness of the " educated guess" as a conservative bound.

24. Chapter 5: Quantincation of Premixtures

1. Is there any likelihood for this plant of subcoolea water in the lower head (eg in an extended accident sequence with some injection)?

This was addressed in Chapter 4. No such likelihood could be found.

25. 2. The comment that the break up parameter f set to 10 produces very rapid break up in ~10 cm of water suggests that the modelling is somewhat more etficient at producing fragmentation than originally desired (break up in a spec-Ired fall distance taken as the smaller of the actual fall distance o. f.Dj). This also depends greatly on the assigned value of Dj - here set to the initial particle size (20 mm). If the melt was assumed to fall as a thinning sheet (quite possible) then the initial penetration of the water may be more local than represented in the Pht-ALPHA calculations. However, I am happy with the range chosen for 0 V F-145

26. 3. Please note that in Figure 5.2 and Appendix B the void is represented by shading, the fuel by contours. Explain the contours that follow the domain boundary.

Clarification made as requested.

27. 4. Specify the boundary conditions for the calculation. What pressurisa-tion is predicte.l?

Due to thelarge volume of the system, the localized nature ofinteractions, and short times, no pressurization is predicted. A constant presse.re outlet boundary condition at the top of the downcomer was imposed.

28. 5. The lemth scale increase referred to on page 5-5 is not evident in Figure 5.4. The area averaged over is not clear, it is obviously not the whole cross-section. Since writing this I found the 1% fuel volume fraction limit on the region considered in the text - for clarity add to caption of Figure 5 3.

Clarification made as requested.

29. 6. Middle of page 5-10: 'Only a very small fraction of the coolant is found to co-exist with the water'- I know what yoc mean! It is clear though that here we have the key result anticipated for the mixing codes. This hnplies that the key region to seek validation of the code is in the production of the high void fraction.

Yes, and this was done in the MAGICO-2000 tests. See also addendum to Appendix B.

30. 7. In my view the THERMAL calculations raise as many questions as they answer, because of the poor validation status of any jet break up model.

However, I do not think this is a key part of the argument.

31. Chapter 6: Quantification of Explosion Loads

1. Where is the trigger cell?

- At the bottom of the premixtures in each case.

32. 2. At what time was the effecthc area evaluated - that of peak press:ste?

3 If not, you obtain larger effective areas than ~0.1 m .

F-146 l

At the time of the main portion of the pressure pulse is delivered.

O 33. 3. The question raised by the calculations is how far is one from the danger zone? Could we get there by a modest increase in system pressure (what value was assumed?) and/or varying the value of f 7 More calculations presented in the addenda to Chapters 5 and 6, and their interpretation should help this kind of question. We used 1 bar, and calculations done since, at 3 bar, gave the same results. Further, results for in-between # values are now provided. The results are understandable about how and where the high pressure pulses are produced, so it is unlikely that we missed something that coup " unexpectedly bring us to the danger zone."

34. Chapter 7: Integration and Assessment 1 The conclusions reached are justined on the basis on the analysis presented.

Gu the basis of current knowledge I am still not comfortable with the observation that downward relocation scenarios are ' physically unreasonable'.

See above responses and other reviewers' questions and our answers in this area.

35, 2. I agree that there is a greater threat from subcooled conditions. It is not obvious, though, that a ' highly subcooled pool'is necessary. Perhaps this might be !!!ustrated by a calculation with modest subcooling (eg 1C degrees) to c show there is not threshold effect.

We now have such a calculation (see addendum). However, please note that the possi-bility of creating subcooled conditions has been addressed in Chapter 4. In this context, generating a highly subcooled condition is just as difficult as generating a 10 K subcocling.

36. Chapter 8: Consideration of ReBood FCIs

1. This chapter has not been considered in any detail. The arguments presented appear persuashe provided that are no other means of fast reflooding not con-sidered by the authors and that crust formation proceeds in the way that they envisage.

F-147

37. Chapter 9: Conclusions

1. I have indicated above that my principal reservations lie in the areas of the downward blockage and in ensuring that there are na oper'ator actions that may prejudice the assumptions made in the analysis. I agree with the authors that consideration of additional pathways is unlikely to change the conclusion.

38. 2. For this application, the supporting analysis ought to concentrate on the melt relocation scenario. This would include obtaining a better understanding of melt relocation in TMI 2 (eq why did it occur after reBooding the vessel?), to demonstrate that the processes are indeed understood.

See Epstein and Fauske (1989) in Faul, and addendum to Chapter 4.

39. 3. It would have been useful to have an indication of the effects of uncer-talnty in the constitutive laws (eg mictointeractions) to determine where conBT-matory studies are required.

Presently we have chosen very conservadve parameters for microinteractions. Results for parametric variations, guided by the most recent SIGMA experiments, are now provided.

40. Comments on DOE /ID-10503: Propagation of Steam Explosions: ES-PROSE.m Verincation Studies Only a limited time was anllable to review this supporting document.

Much of the document is concerned with the ability of the ESPROSE.m code to represent the wave dynamics correctly for single and two phase regions in one and two dimensions. The information presented, along with the comparisons with the SIGMA experiments with a voided expansion region, indicate that this part of the code is doing its job correctly, even when relatively coarse (~0.01 m) meshes are used. This does not surprise me. Numericalstudies we performed when extending CULDESAC from one to two dimensions indicated good capabilities to capture the wave dynamics with relatively simple numerical schemes (the numerics of propagation are simplcr than those of premixing). I am therefore satisned with the code's capabilities in this area and would expect that the 3-D version of the code would also perform satisfactorily in this respect.

F-148

l

41. . Winie Chapter 2.1 uses a homogeneous model for the two phase behaviour (by forcing large drag between the phases), it is unclear whether the calculations

(~f reported in Chapter 2.2 still use this model. If not, it would be interesting to compare how much better the full model perfortm against the experimental data, compared with the homogeneou's model.

The full two-fluid model is used for the calculations presented in Chapter 2.2.

42. The authors of ESPROSE.m have implemented an, at the time, novel ap-proach to cover lack of thermal equilib:ium in the coolant during the propagation.

This approach is physically based and can be considered to be well-justined.

43. The application of the ESPROSE.m code to steam explosions depends on the assumed cm.et:tutive physics. As Appendix D (particularly Bgures D8 and D9) illustrates, the assumed parameters of the microinteraction model can have a major impact on the prediction (eq changing the parameter for coolant entrainment can change the C-J pressures by two orders of magnitude). Ap-pendix C provides results from a series of experiments with one high temperature simulant, that has been used to modify a hydrodynamic fragmentation model to take accouat of thermal effects. This approach is acceptable, but the range of G uncertainty in the model parameters needs to bt allowed for in any assessment.

The authors note that 'the main need identiBed is for constitutive laws for mi-crointeractions with reactor materia 3' [ Abstract) - I agree. They also claim that

' reasonably conservath e assessments are possible'- however the main report does not indicate what parameters were used to obtain' a sulficiently conservative as-sessment.

See Appendix C and the addendum to it. Also, see new reactor calculation .%r different microinteractions parameters (see addendum to Chapter 6 of DOE /ID-10541).

44. I would have expected to see mo: 3 discussion of the comparison with KROTOS experiments in the report as originally supplied, rather than a reference back to the study. Although there are somelimitations on knowledge of theinitial conditions and, : - of the tests with explosions, some loss of data, these provide the greatest .onBdence in the application of any model to the steam explosion propagation phase. The calculations for KROTOS-38 provided as a supplement are useful. With curre.1t knowledge it is more important to be able to p

d F-149

demonstrate conservatism in the calculations rather than good agreement through parameter adjustment. Recently i saw calculations with TEXAS-IV for this test, '

where a very different melt distribution was calculated that led to very good agreement with the observed pressurisation following the trigger. Until there is a visual record of such tests it is not possible to determine which simulation is closer to reality.

45. In reading the material, I noted a number of exampics where detail was not clear to me. These are listed here for cotwenience, but have no impact on my overall assessment of the methodology:

1. In chapter 2.1 what value is used for Ps? Figure 6 a implies 100 bar, but elsewhere finite results are given when P2 is only 10 bar.

For results in Figures 1 through 5, Pi = 1 bar. For results in Figures 6 through 9, P2 = 100 bar.

46. 2. In figs 7 and 8 of Chapter 2.1 a is shown as varying. I assume a is a void fraction - of which region' a is the void fraction of the region ahead of the shock. g

47. 3. Chapter 2.1 presents results with an without phase change of the gas. It would have been instructhm to see a direct comparison to illustrate the importance, or otherwise, of the phase change on wa- , propagation.

Results presented in Figures 4 and 5 are for a 10% steam / water mixture with phase change.

48. 4. I had difficulty understanding the location of the pre-voided region discussed in Chapter 2.2. Note that Fig 7 is incorrectly referred to as Fig 8 in the text.1(for Fig 3 the pre wided region stretches to the base of the tube, I do not understand the respective difference in timings of (1) the time between the shock arriving at PT3 after PT1 and (2) the time between the shock arriving at PT3 and its reflection from the base arriving at PT3. Note that you have of(set the pressures in the figures for ease of presentation.

The typo on Figure 7 is corrected. The pre-void region is the region between L = 100 cm to L ~185 cm, as shown in Figure 3. The bottom of the tank is at L = 300 cm. The region between L = 0 and L = 100 cm is the driver section,. as shown in Figure 1 F-150

_ _ _ _ _ .. _ . - . _ . . ._ _. . _ _ ... _ _ ._. _ ___ . ~. . .

f L

A9. - Additional Conunents on DOE /1D-10504: PM. ALPHA-Verincation Stud-les by T G Theofanous, W W Yuen and S Angelini INTRODUCTION This docunient represents the culmination of a substantial piece of work to'de-. o

~

velop a mixhig code fo~ steam explosion studies and to validate it against the experimental data. The report makes good use of the (still rather limited) ex .

perimental data'available for this purpose. The report concentrates on the pre- -

sentation of results rather ti'an their evaluation. It would benent from a leading chapter on the philosophy of the verincation/ validation process, accompanied by a matrix' indicating which of the code's models are tested, and to what extent, by_ the comparisons reported. it would further beneht from a longer concluding chapter that draws together the results in the context of this matrix.

The philosophy of our approach is explained in Appendix A. Our verification / validation

- plan is shown in Figure 1 in the Introduction. Additional discussions as the or'es re-quested here would draw us into judging the quality of the results, which we scrupu-lously avoided. We left this judging for the reviewers. As far as testing of the individual models, this is a good suggestion, and we will include such a table in the final report.

50. It is noticeable that efforts are made to compare isothermal particle-water

. predictions with accepted correlations. There ought to be scope to include similar material on two. phase Bow in the absence of particles; this is probably more a important in establishing the reliability of the code to predict voiding behaviour.

Such tests can be meaningfully be done for the dispersed regimes (bubbly, droplet), and -

the particle cases considered are already quite sufficient for this purpose. The drag laws used for fluid or gas " particles" are slight modifications of these and they are supported

- by wide data bases.

51. . While there are many detailed comments below, these should not detract from the achievement of the authors. The comparisons performed indicate that the code has the ability to make reasonable predictions for reactor conditions.

However, the results should still be used cautiously, as the data currently do not exist to provide full validation of the model.

• * * * =* * * * * * * *** ** * - * * * * *

• O e.isi m-'f M ++m-- ig .-e n. m , +4 w w

52. Specific Comments Chapter 2 Sineel narticle settline While tracking a representative particle in a Lagrangian fashion gives the expected analytic result, melt mass is usually tracked through the volume fraction. This can be much more diffusive.

These were simple tests made to begin with. For numerical diffusion see below.

53. Settline rf narticle clouds I have tried to check the consistency between the drag law for particles given by equations 3.14,3.21 and 3.22 of Appendix A with the drift Bux formulation, but have been unsuccessful. There appear to be inconsistencies between equation 2.4 and Figures 2 and 3. Taking % = 0.487 m/s, ghes the liquid superncial velocity for a = 0.5 as 0.093 m/s. Figure 2 shows this as 0.12 m/s, while Figure 3 indicates 0.10 m/s. This suggests that it is not the superncial velocity that is being pictted in Figure 3 but the Bow velocity, which would be 0.186 m/s from equation 2.3. hiy evaluations of the drag coefficient given in Appendix A for this case give a relative velocity of 0.286 m/s, or a superficial velocity of 0.143 m/s. Howent, Phi-ALPHA has produced, according to Figure 3, a value close to 0.2 m/s. h!y hand calculations indicate that the Phi-ALPHA modelis not as close to the drift Bux model as implied by Figure 3.

The closeness in Figure 3 is correct. The confusion was generated because there was a mislabeling of the vertical axis in Figures 2,3, and 4. It is the velocity, not the superficial velocity plotted. Also,0.186 is in good agreement with 0.2, isn't it? Finally, we found also a slight plotting error in Figure 2 regarding the drift flux line. The correct figure is attached.

54. Settline of narticle clouds The comparison with the drift flux model is clearly important as it goes some way to establishing the reliability of the drag coenicient modelling in Phi ALPHA (although it should be noted that the par-tscle volume fraction is unlikely to exceed 20%, where the enhanced drag due to particle-particle effects is not that significant). It is less clear what one is ex-pected to learn from the material presented on transient analysis regarding the validity of the code's models. It would have been usefulinstead/in addition to perform the same comparison with the drift Bux model for gas-water interactions a where the form of the drag coeRicient is rather different.

F-152

.- - . . - - - . . . _ - . . . . - -_ ~. _ . . - . - . . . -.. - - . . . .

l b  : 0.6 , , ,. , , ,

drift flux model x- x PM ALPHA initial condit;ons x' x ,

X X x x y- x

• x-- *'

0.4 -

)[ -x x X x. x

. 0.3 -

x x -

• x x x x x x x 0.2 - x x _

1 x Y

x x -

x x '

' 0.1 -

x ,x -

x x x x x x

0'

• s 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Particle Volume Faction Figure 2. Initial Conditions in PM-ALPHA simulations.

Several reviewers found these results very interesting!

O 55.- ' Section 2.2.2: MAGICO exoeriments It would have been useful to have .

a 'short synopeis of the conclasions drawn about the model from the analysis of the MAGICO tests. Besides the qualitsthe agreement (and general quantitative agreement) on the nature of the interaction. I think the most signincant Bnding

, is the prediction and measurement oflarge void fractions (greater than 70%)

illustrated in Figure B23). It would be useful to provide a statement on the specinc code models that these observations are believed to validate (eq water-steam drag, Rhn boiling, radiative heat transfer??).

All play a role. These are integral comparisons carried out for a wide range of conditions (see also addendum to Appendix B), and the results speak for themselves. We don't think it is very fruitful or even appropriate to assign significance. Note that each model stands .

- on its own merits, and these comparisons show that when put together the result is very

, consi stent wiht reality. This should not be too surprising.

O V F-153

, - - , - . , , ., ,, - , . - . . - ---w . - + , -

• 5G. The OUEOS Experiments: This looks a very interesting analysis of these tests. The presentation of reaults in Figure 4 etc gives an excellent way of qualita-tively comparing code results and experimental observations. Perhaps some com-ment should be made about the apparently coherent release of large gas / steam volumes, seen eq at 0.41 s in Figure 4; also on the water spout e8ect predicted at this time (this seems to provide the mass difference between hieyer's interpre-tation of the water fraction in the mixing region and the Phi ALPHA values).

The acceptability, or otherwise of numerical diEusion, is a complicated matter, because of non-linear feedbacks thrJUgh the drag laws; it is very easy to Under-predict the peak particle volume fraction. Figure 5 does not give units fr 'e liquid flux. Condensation in Pht ALPHA looks too eEective at later times in Figure 6 compared with the experimentalimage.

The liquid flux units are cm/s. We have done calculations of QUEOS runs with I cm grid and obtained very good agreement with the results presented previously. Also see Addendum 2 to Appendix B with PM-ALPHA.L simulations of MAGICO tests and the addendum to Section 2.2.3 with simulations of QUEOS tests.

57. Chapter 3 Comonrison with CHYhfES: It is only fair to note that this comparison was only possible by turning off sub cooling in Pht ALPHA. hiuch of the detail of the Phi-ALPHA predictions depend on the modelling of sub-cooled boiling. The observation that Ph! ALPHA often only produces any void somewhat behirad the particle front, whereas other codes tend to produce some miding wherever there are hot particles can have signliicant implications on the initialilow of water. For instance, we did not reproduce the so-called ETHICCA eWect with CHYhlES. In addition, CHYhlES drag laws were modified for the comparison. However, the main result - water depletion is predicted by both codes (at least for low pressure systems close to saturation temperature) - is robust.

58. The hilXA Exoeriments: I will try to clarify the question of time origins for the data. The experimental report, which I have, has unequivocal timings, with an origin starting at the ignition of the pyrofuse for the thermitic reaction.

On this timing the melt Iirst contacted the water at 3140 ms, the peak (measured) steaming rate was at 3810 ms and the peak pressure occurred at 4215 ms. The F-154

authors have adopted a timescale'(their Bgure 4) where the time of Brst melt

(, ,) contact is taken to be zero. This is the same timescale used in Figure 1 of Fletcher u,

and Denham for the mesured pressure in the gas space - so the comparison given for pressure in the top free c! Figure C (page 3-?1) is correct. However, the transient steaming rate Bgure (Rgure 8 of Fletcher and Denham) does not use this time basis - this is because it was derived from the CHYhfES calculations with the experimental data over-plotted). There is a signincant outBow of gas before the melt reaches the water surface as shown in this Bgure. This may be due to (i) preheating and expansion of the gas in the test vessel; (ii) evaporation of a water Blm on the test vessel wall (the favoured explanation for similar observations in FARO), and/or (iii) evaporation from the water surface. The experimental data on the middle and lower frames of Figure G should therefore be shifted to the leit by about 0.32 s (error on this is only from my reading of the graph in Fletcher and Denham - it is no more than 0.02 s). The effect of this is to move the measured peak steaming rate ahead of the measured peak pressure. However, I now believe that the measured steaming rates become increasingly smreliable (as quoted) due to carry-over of a two-phase mixture; similar behaviour has been observed in PREhHX. Unfortunately, while the experimenters noted water carry-over post r test, and observed a reduction of water height in the vessel post-test of 25 mm (the measured steam would produce a reduction of only about 4 mm), there is no information to determine how much of this occurred because of evaporation during the heating of the water. The same comments apply to Figures 7 to 10.

While the Phi ALPHA calculations are as good as or better than any I have seen for hHXA-06, I am not convinced that the real behaviour in the test is being captured. The most noticeable features are the radial expansion of the melt as it enters the water and the apparent lack of any visual record of droplet break-up.

Both of these effects seem to be connected with sudden expansions of the melt region, due to enhaked steam generation, giving much more coupling between melt and steam that accounted for in CHYhfES, and, by the look ofit, in Phi-ALPHA. I conjecture that droplet fragmentation is occurring during these rapid events. The formation of smaller particulate then encourages another process of melt spreading. Smaller particles are carried upwards by the central steam flow, move outwards, and fallin the periphery, thus extending the melt emelope outwards.

. A}

i d F-155

There is no visual evidence of the predicted extensive voiding atound the melt region - the leading droplets appear to be falling through water - the steam generating region is large because of the spread of the melt droplets.

I agree on the sensitivity of calculations to assumptions on break up. Has the predicted mean particle size been compared with the experimental value of about 3 mm?

This section should .'ontain discussion / conclusions on implications of the com-parison for model validation.

With so many uncertainties in the test how can we reasonably draw conclusions? In any case the additional information and insights provided by the reviewer are very helpful in our further interpretations of MIXA with PM-ALPHA.L (see addendum to Section 3.2.1).

59. The FARO Exocriments: Clearly the initial melt droplet site is very un-certain, as is the spread of the melt. L-14 appears to be the test in which the melt stre.sm was best collimated, but one cannot tell whether the stream contracted as it poured through the gas space, or underwent a mild expansion (in L-11 the melt stream appeared to undergo a major expansion). Ifit is believed that the melt jet contracted (note typo: steam for stream 4 lines from end of page 3-25),

then the radial meshing with ar = 5 cm is too small. The choice of break-up parameters appears arbitrary - presumably these were selected to give reason-able agreement with the experimental data. More detailed modelling of the melt release vessel indicates that the melt exit velocity was close to 3 m/s for most of the pour; this will not be replicated by the model shown in Figure 2. I am surprised that a Weber number criterion did not limit the droplet size; with the CHYMES implementation of this criterion we almost always get mean particles close to those ol> served in experiments (typically 3 - 5 mm). The comment on the absence of signliicantly superheated steam in the experimental data seems to me to be special pleading - it might be right, or the steam flow might be much

ess concent,ated on axis than predicted by PM-ALPHA, giving steam closer to saturation conditions. It is difficult to relate the scales on the coloured contour plots in Figures 10 and 11 to the colour-scale, particularly because of interpo-lation cifects. Is break-up still occurring after the particles have settled (unless they have solidified)?

F 156

i l

No, actually at this stage reagglomeration of any inadequately solidified particles begins.

In relation to the other comments see further interpretations with PM-ALPHA.L (adden-(v) dum to Section 3.2.2).

60. Again, this section should be supplemented by an evaluation of the im-plications for the reality of PM-ALPHA predictions. I think a word of caution is necessary, as although PM-ALPHA, with the assumption used, performs well egainst experimental data, it predicts a highly tuv-dimensional configuration.

Alternatively, good comparisons against the data have also been produced with the one-dimensional code, TEXAS-IV. Until we see the natute of the interaction zone (I expect it to be between these two computational extremes) then it is not possible to say that one simulation is better than the other.

We do not think it would be appropriate for us to shed vague doubts on the validity of the PM-ALPHA comparisons because the ID TEXAS-IV was made to produce good comparisons too. The last sentence is very puzzling. What is intermediate between ID and 2D?, or is it 1-1/27 When a behavior is not ID, it can only be 2D (or 3D, of course, but this is not the issue here). When a behavior is not ID, a 2D computational framework is a necessary starting point, before one begins to examine any further the degree of rN. " simulation" obtained.

(  !

%/

61. Chapter 4 I agree with the general comments on break-up modelling. As implied in my comments above, backing out break up behaviour from the experimental data may compensate for other errors in the modelling. As I also noted, it is unclear, even with the visualisation, what break up processes were occurring in MIXA; I suspect the processes are much more dynamic than are currently embodied in the models, and coupled strongly with events of enhanced steam generation (coolant trapping?). New FARO tests with visualisation should provide information on the coherency of the initial pour, besides evidence of any subsequent break-up.

62. Numerical aspects Our experience is not as comforting as that presented by the authors. I think that numerical diffusion is probably not an important issue for large-scale mixing calculations. However, it becomes important in comparisons with smaller-scale 3

(V F-157

l experiments, which are often dominated by leading edge eifects. Humerical com-parisons that we have performed (external to CHYhfES) show that upwinding schemes run below the material Courant condition lead to very poor predictions of peak particle fraction, and thus drag. Higher order schemes have to cater for possible discontinuities at the leading edge. Lagrangian approaches, as used by the authors for their front tracking, provides much better accuracy, both for ve-locity and peak volume fractions. I believe that current schemes in the mixing codes can be improved substantially using physically based Lagrangian limiters, rather than mathematicallimiters. Ehlly Lagrangian approaches have the greater benefit of handling a spectrum ofparticle s zes. This may be the best way to treat jet break up and is necessary if one is going to capture the role of the smaller droplets in spreading the melt, as observed in hilXA-06.

The current presence of numerical ditfusion makes the code results dinicult to imerpret (eq how far back is the predicted peak concentration from the melt leading edge in the hilXA-06 calculations?). Our experience with more refined meshes is that numerical diffusion is indeed reduced, but the calculations are much more prone to instability of the resulting interface; this numericalinstability probably reflects the actual instability ofinterfaces observed in experiments.

This whole discussion reflects the reviewer's own experiences with CHYMES, and it should not be confused as being applicable to what we have presented. Our results should be judged on their own merits, and there are ample comparisons to allow one to comment directly on the present experience. We find no instabilities in fine meshed calculations, and they agree with our new PM-ALPHA.L results.

63. Concluding Remarks I would expect a more detailed technical eduation of the calculations presented.

I am surprised that questions related to the radiation transport modelling, which was clearly important in the FARO simulation, have not been highlighted. I would have liked to see more explicit bounds on models emerge form the work.

We wanted to keep our concluding remarks brief and to the point. If we were to provide highlights in this section, we would have to include much more than the non-local radi-ation model (actually it is even more important in MIXA, as discussed in the respective section). The last sentence indicates a degree of dissctisfaction, we think, with how far, quantitatively, we could go based on the comparisons. This is a question of judgment, F-158

T

and one thing for sure is that we do not wish to oversell what we have been able' to

!show. We believe the comparisons, and the whole array of situations considereVspeak for themselves, and adequately loud to be understood, both in their success, as well as .

their limitation by those involved in this kind of work, and at this state we are satDfied with that.-

64. Phi ALPHA hiodels -

The details of the correlations embodied in Phi-ALPHA will not be reviewer in r

detail.

. I believe the modelling approach is sound. Inote that reactor geometries may im - ,

pose strongly three-dimensional Row regions, so a 3 D code is needed for detailed appilcations (if found to be necessary). Iget the impression that the modelling ,

philosopl;y falls between two stools. At places it is admitted that the model nec-essarily contains many simpilBcations and constituthe physics that is uncertain, but only in the Reid ofJet breakup is a parametric approach used. I would prefer

!" a broader approach to treatment of uncertainties.

_ This comment forgets that we have indeed a 3D code. We did our best to catch and present a correct picture regarding uncertainties.- If the reviewer has particular aspects

-- beyond those presented that he would like to see, we would be happy to undertake the ,

computations needed.

65. With sub cooling implemented in CHYhiES, it is closer in concept -to TRIO-hfC rather than Phi-ALPHA. (EVA should be sylled IVA).

h,

66. Elscwhere we have queried the use of the drag coeflicients for droplet and bubbly Row. These are derhed for bubbles rising at terminal velocity in a

+

gravitational Beld. It is found that the shape factor for the bubble causes the

, drag per unit mass of gas to be independent oflength-scale. It is noticeable that no effect of melt droplet shape-appears in the corresponding formula for drag coeRicient'for the melt phase (equation 3.21). A completely different form for the

. liquid-vapour drag is used for intermediate values of void fraction; this may give .

large changes in drag when the transition void fractions'are crossed. It is not evident that there are such sudden changes in flow regime in plenum geometry.

L,(

A F 159

. . . - -  :-. - . = - . - a. . .- - -  := - -

The intermediate regime is very important in gas-liquid flows, because it indicates an initial " break" in the liquid continuous regime. It is well known that this leads to a suddenly much greater slip, and this leads to churning. For drag coefficients we used the best available correlations. In all our experience with these models we find no reason to raise significant questions to them.

67. I have not had the time ta consider the radiation treatment in detall; also the relevant appendices are not included in the exerpt. For dense clouds of particles, the self absorption effect will be very important. I would like assurance that this does not allow the region to emit more radiation externally than that of a black-body covering its surface at the same temperature.

As explained in the text, self absorption is included in a consistent manner. The material in the appendices referenced in this section pertain to detailed numerical procedures in evaluating integrals and is really not essential for such a review.

O F-160

... - - _. . . _ - __ _ . - - - . - . - .. ~ . - - ._

y - E15. Resoonse to M.E Young (SND General Comment 'and Highlights i L

. General and unqualified agreement with the conclusions of the work under review.

Point-by-Point Responses

- 1. - This is a masshe piece of uvrk which includes, in addition to steam explo-

~ sion loads, some areas with which I am not particularly familiar, such as proba-bility methodologies and plugging behavior of molten materials. I will therefore comment mostly on the steam explosion loading. .

The approach taken in this report to determine steam explosion loads is essen-tially the one that has been recommended by most if not all steam explosion researchers; use of computational models validated ag'ainst experiments to deter-mine bounding envelopes for reactor accident scenarios. Prof. Theofanous has taken an additional step here in simplifying the probabilistic framework with his ,

ROAAM method; I think this is entirely in keeping with the use of these type of calculations in risk assessment and rulemaking. I believe that the work described-in this report has successfully accomplished the goal of enveloping the steam ex-plosion loading. The usefulness of the results in rulemaking, however, therefore depends on the conBdence placed in the initial conditions of the accident scenario

- and in the analytic tools used.

la regard to the initial conditions, it is very important to the conclusions reached in the report that the melt be introduced through the side of the reRector and that the lower core structure and support plate be plugged. As I mentioned before,-

plugging is not my area of expertise, so I will not comnaent further other than to point out again that conRdence in the initial conditions is very important.

In regard to the analytic tools used for calculating steam explosion loads, I have some comments concerning possible gaps in the cases considered and in verinca-tion of certain parameters used in the models.

2. First, I see that trigger timing was varied parametrically but not trigger -

. location; I assume that the cases were triggered near the bottom of the mixture

' region next to the wall; although I suspect that this is probably the most severe case, I am wondering about the consequences of other trigger locations.

F-161

We did not do extensive variations onloation of trigger, but what we have seen agrees with what we expect: it is the premixture composition rather than the location or magnitude of trigger that controls the energetics.

3. Second, in ESPROSE.m, there are tbree parameters, that must be set from experiment: an entrainment factor, a fragmentation constant, and a thermal en-hancement factor. There appears to be some dependence on the melt material for tbese factors, so the 1ack of data with reactor materlais ta set these param-eters cancerns me. I believe this was also pointed out by Theofanous et al. In the " Concluding Remt:ks" section of the ESPROSE.m veri & cation report, in re-gard to expanding the microinteractions database to reactor materlais. Useof parameters that had been set from experiments using reactor materials would enhance confidence in this aspect of the calculations. In regard to the microint-eraction model itself, I believe that this model is sufficiently close to reality that experimental results can be extrapolated to reactor scale.

4. Third, the lack of a stratified mixing case bothers me, or rather, the lack of data to properly nodel this case. I do not doubt that what the authors say is correct: if ESPROSE.m were run with a stratified case, it would probably produce a very nonenergetic steam explosion for the reasons stated. However, if memory serves, a stratified explosion in a foundry imolving water dripping into a " car" containing molten iron produced an explosion strong enough to take out some of the walls of the plant. This incident seems in contrast to what would be predicted with ESPROSE.m, although the melt is ditTerent (iron versus reactor material) and the water was undoubtedly subcooled. The incident mentioned probably imohrs mixing of the stratified material caused by the province of PM-ALPHA? Maybe the authors could comment on this, or maybe it indicates the need for some stratified experiments.

The incident mentioned occurred at the Farthingham Foundry. It involved water pouring in a steel " torpedo," already pretty full of molten steel, through a small opening on top. The explosion occurred as the torpedo was set in motion, to transport the melt (apparently not realizing tne accidental presence of water). The water became trapped in a highly confined geometry and pressures developed made the torpedo explode. This type of phenomenon is not relevant to the stratified interaction considered here.

F 162

i

,- 5. Fourth, on page G 4, tl'ere is a conclusion that the size of the impulse

( does not depend strongly en the size of the mixture region. I think that thic is in contrast to first principles, which would suggest that it would be directly proportional, ignoring other effects like the varying wtd fraction, and to the ,,

results in Table 6.1, which indicate a strong vnslation, ignoring the time, between cases C1 and C2: 90 vs 120 for f a 10, and 120 vs. 200 for 0 = 20. Or did I misread the sentencc7 Also, how do the results compare in magnitude to the impulse of the initial trigger itself?

The resulta do not depend on the magnitude of the trigger. The realts in Table 6.1 show that, indeed, the impulse does not depend on the size of the premixture (note that bigger premixtures are more volded). There is only one case, the 200 kPa s one, that stands out.

This case has been reexamined and discussed in an addendum to Chapter 6.

G. FI[th, in thr section on stratified layer of molten steel and tellood, Chapter 8, there appear to be one piece missing to make the case that the scenario is impossible: the thickness of the crust is nos mentioned. Specifically, is this a

" thick" crust that is stab c, or is it a " thin" skir. that could be broken?

n ihe thickness of the crust wil! depend on the time after addition of water, it will get thicke-V with time, and hence less likely to allow contact, as more water accumulates with time.

7. Allin all, this appears to be a very complete piece of work.

8. Fcllowing are minor points and typos.

1. In the discussion of the ABAQUS model of the lower head, :t is referred to ns a shell model; this is somewhat confusing at first, as shell models are normally thought of as meaning. thin shells, i.e., no bending moments are supported. This could be cla:llied by calling the model a thick shell, for instance.

We have mostly used thin shel!. elements. We ran a comparison for loading pattern 1+,

using thick element and found a slight decrease of plastic strain from 21.3 to 20.6%.

9. 2. On p. 4-4, there is a comment about approximately 25% of the fuel remaining uncovered. Is there a reference for this?

lypical result of computations such as with MAAP.

/3

..) F 163

10. 3. On pp. 5 G through 5 9, it is hard to compare the graphs chosen because of varying i axis scales and varying times for the plotted lines. For instance, the C1 nb plots start at 0.4 s whereas the Rci nb plots end at 0.12 s. I see that there are other plots with overlapping times in the Appendix, so maybe one oItbese would be better.

The point is well taken. See better representation in the addendum to Chapter 5.

11. 4. On p. 510, it says "only a very small fruction of the coolant is found ta coexist with the water"; should this be mdt?

Yes; correction made.

12. 5. In the graph for C210 a on p. B2 7, the last time is sken as 215 s instead of 0.215 s.

Typo corrected.

O

e. . e

APPENDIX G ,

to)

EXPERT COMMENTS AND AUTHORS' RESPONSES CLASSIFIED BY TOPIC TABLE OF CONTENTS 0 Overall Comments and liighlights ...................G-3 1 Overall Comments, Conclusions ....................G-6 2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . G 32 3 Structural, Fragility . . . . . . . . . . . . . . . . . . . . . . . . . . G-37 4 Melt Release Scenarios . . . . . . . . . . . . . . . . . . . . . . . . G-50 5 Premixing and Triggering (Reactor) . . . . . . . . . . . . . . . . . . . G-73 6 PM ALPHA Code and Verification . . . . . . . . . . . . . . . . . . . G -86 7 Propagation (Reactor) . . . . . . . . . . . . . . . . . . . . . . . . G 119 8 EGPROSE.m Code and Verification . . . . . . . . . . . . . . . . . . G 133 9 Typographical Errors ........................ G 155 10 Stratified, Reflood Explosions .................... G-157 11 "Other" (Subsequent) Explosion Scenarios . . . . . . . . . . . . . . . G 160 Each question or comment is identified by the first three letters of the reviewers'last names and a number that corresponds to the index used, by us, in Appendix F. That is, May3 is point #3 in our breakdown of Prof. Mayinger's review comments, as done in Appendix E The three letter code used for each revlewer is as follows:

Bankoff (Ban) Berthoud (Ber) Burger (Bllr)

Butler (But) Cho(Cho) Corradini(Cor)

Fauske/ Henry (Fau) Fletcher (Fle) Jecobs (Jac)

Mayinger (May) Moody (Moo) Sehgal(Sch)

Shewmon (She) Turland (Tur) Young (You) >

Nate11 the mark * * * *

• in the authors' responses indicates agreement, or no comment.

Note 2: Because here comments and response; are collated by topic, sometimes the terms "above" or "below" may appear to be out of context. The appropriate context for these is that in Appendix E (3

G-1

l l

rx Overall Comments and Highlights b On S.G. Bankoff's Review General and unqur.lified agreement with the conclusions of the work under review.

On G. Berthoud's Review The review is generally agreeable, but reluctant to fully accept that a bottom relocation path is physically unreasonable. This is a key point of our evaluation, and we provide responses to the reviewer's specifie questions. We expect these, plus our responses to similar questions from other reviewers, to be helpful in further focusing the issues towards resolution.

On M. B!1rger's Review Many specific issues are raised in this review, about almost every aspect of the analysis and supporting documentation. However,it is also stated that in an overall way the analysis is convincing and that the spirit of the criticism is to help provide further supporting evidence. To the extent possible, this is done in the point by-point iesponses below.

On T. Butler's Review Two reports were provided. The first addresses the fragility and inacates that the portions of the curves above the threshold failure (10-8 probability) level may rise faster than given in the report, because we neglected progressive failure effects (through the wall thickness).

The second report examines the loads in relation to the fragility and concludes that, sf ace the two do not intersect, the above criticism on the fragility does not affect the conclusions of the report. Based n this, the present response does not address this' criticism. We plan to carry out the kinds of calculations suggested and will include the results in an addendum to Chapter 3,in the final report. Also, this reviewer provides evidence that the appropriate yield stress is 450 MPa (rather than the 330 value utilized), and that the strain rate effect may not be as strong as taken in the calculations. These variations are mutually compensating as intended, to begin with.

On D.H. Cho's Review This reviewer finds that additional supporting work is needed before the results can be used in the licensing area. The review process, by comments and responses, has produced additional supporting work. Specific issues raised ty the referee relate principally to scenario aspects such as the possibility of " secondary" explosions through a downward n

o3

relocatim p .th, premixing at a higher pressure, and reflood FCis. These are addressed point by point below. g On M.L. Corradini's Review The principalconcen of this revieweris about the melt release conditions. Other reviewers have gone into this type of question to a much greater extent, and our zesponses to them may be useful here too.

On H. Fauske's and R E. Henry's Review General and unqualified agreement with the conclusions of the work under reviev.

On D.F. Fletcher's Review General and unqualified agreement with the conclusions of the work under review.

On H. Jacob's Review This is a highly skeptical review, questioning even the non-existence of supercritical ther-mal detonations in highly voided premixtures. Until the myiewer resolves this trivial point in his own mind, we can make no progress here.

On F. Mayinger's Review General and unqualified agreement with the conclusions of the work under review.

On F.J. Moody's Review General and unqualified agreement with the conclusions of the work under review.

On B.R. Schgal's Review This is a generally agreeable review, Many detailed question's are raised, but these are mostly of a clarification and reinforcing character, rather than strong objections. Also, many useful suggestions and opinions are offered, again in the same hght. Perhaps the key point is that the review expressed caution with regard to the maturity of the analysis tools. If this refers to a stage in the " phases of development," table (Table A.2 in Appendix A), we certainly agree.

On P. Shewmon's Review General and unqualified agreement with the conclusions cf the work under review.

On B.D. Turland's Review This review is hard to interpret at this stage. Concems are raised about almost every aspect of the analysis and supportir.g documentation, yet we siso obtain the impression G-4

1 that these are offered in the spirit of further improving the basis for the conclusions rather than in challenging them. The key point appears to be the one made in closing the first sectime of the review (Overall Comments), that is, "my residual concerns relate to the confidence in having low pour rr.tes and the possibility of operator actions leading to some subcooling". In this response, as well as responses to several other related questions by other reviewers, we present additional material that hopefully will be found helpful in coming to a resolution, or if not, to better focusing aay remaining concerns.

On M.F. Young's Review General and unqualified agreement with the conclusions of the work under review.

O o v G.5

l Overall Comments, Conclusions Dano. I enclose herewith my revien of DOE /1D 10541. I had to read the supporting documents as well in order to get the necessary perspective. In the process I spent 32 hours3.703704e-4 days <br />0.00889 hours <br />5.291005e-5 weeks <br />1.2176e-5 months <br />, which convinced me that in-vessel retention is a valid concept for the AP-GOO.

Dan 1. The principal documents which were reed by the outbor were DOE /ID.

10541 (June 1996), DOE-10460 Vols.1 and 2 (July 1995), and DOE-10849 (Jan.

1995), as weil as various papers published and/or presented by Prof. Thwfanaus and one or more of his co authors. My generalimpression is that this is a massim piece of work, which attacks all sspects of the steam explosion probicm in the Westinghouse AP-G00 reactor, and conclushrly demonstrates that failure of the vessel, to say nothing of the contal.nment, is physically unreasonable.1[no failure occurs in the reactor vessel, essentially no release can occur to the inside of the containment building, and hence the threat to the public health and safety is climinated. In the process of developing the evidence in terms of focused experi-ments, development of new and improved codes, tying in work done around the world, and developing a methodology for assessing the safety goals and margins for rate, but high consequence, hazards, a set of tools has been developed which represents a huge step forward in examining severe accidents in new types of advanced nuclear reactors and in existing nuclear reactors.

In other words, in execution, scope and potential consequences, the total of this work represents a very important achievement.

Dan 4. All the codes (ESPROSE.m, ADAQUS and PM ALPHA) appear to be well-tested against available data, and have h~n conservat!vely adjusted.

Dan 11. Convincing arguments have been addressed, backed up by a huge volume of high-quality experimental, analytical and computational work, that the AP-G00 reactor will not fallin the course of a severe accident. This impiles that all later scenarios of containment-building pressurization and heat up are no longer necessary. In my opinion, this closes the severe accident scenario for G6

m ,

l the AP-600, and leads to consideration for licensing. The consideration of other

& reactor types, on the ather hand, does not appear ta be so sttalghtforw.std, and further work needs to be done.

Der 1. This document presents an analysis of the potentiality oflower head failure of the AP600 resulting from a Steam Explosion. The conclusion that the risk is negligible (< physically unreasonable :>) is quite convincing, and is based on:

1. the fact that water will be raturated and at 1 bar due to complete depres-sutisation to the containment pressure and that these conditions willlead to large and rapid voiding which is not favorable for large S. E.

2. the fact that we have permanent blockage: at the bottom of the core that willimpeed any coherent relocation through the core support plate

3. the fact that relocation will occur sideways through the reRector and core barrel and so that the Steam Explosion will occur in a 3D geometry without any lar e constraint allowing large sustained pressure

4. the fact that - even if reflooding is taken into account - when the melt will be ejected sideways, we will have enough time to heat the added water up to saturation and so to prevent good mixing.

The validity of the conclusion is then linked to the validity of the above four arguments.

As for the first argument, there is no doubt that water will be sa'urated as far as reBooding is not taken into account, the fact that the pressure will be atmospheric cannot be discussed here as this is justlBed in another report (IVR Report table 7.3) however, i think that this has to be justined as the voiding will be less important at pressures a little bit higher, around some bars. At these pressures, we can also recall that it was found it was easier to trigger an explosion in the single iron oxide droplet experiments of Nelson in Sandia.

The fact what we have up to now no evidence of explosion in experiments using reactor like materials (Krotos) (and that this is due to the non occurrence ofgood mixing) is stressed by the authors. But once again these Krotos experiments are performed at 1 bar pressure while in Faro experiments at pressures of 50 and 20 U G4

bars, with saturated water some mixing was obtalnni. In a near future a Fara experiment usins; initially saturated water at 5 bar will be performed and we will then 1. ave an indication of the quality of mixing at small pressure.

WATER SUSC00UNO (K) 70 87ft 130 146 180163 11 i I "

( ,

1.6 -

1.4 ,

e -

g 1.2 .

1.0 -

exetossons 3 I aa -l.l  : . /

e 0.0 .8-------*--**0----

/

l -

0.4 -e , l S . / NO REPt.0840NS 0.2 g- -~ ~~g -- . ko o

~~

e i o on 1 0e i

• 0.0 0.2 0.4 0.4 0.8 1.0 1.2 1.4 AMBENT PRESSURE (MPa)

Figure 8. Relation entre le " trigger" decessaire au declenchement necienchement d'une interaction et la pression ambiante (cas de gouttes d'oxyde de fer de 2, 9 mm de diambtre b 2230'K tombant dans de l' eau i 298*K).

Because of the passive design of the AP600 the pressures applicable to the severe accident management window are in the range 1.1 to 1.7 bar. This is too narrow a range to bring in significant pressure effects, and for completeness, we have nm some calculations and satisfied ourselves that indeed this is so (see addendum to Chapter 5).

Der 22. Apart from this problem, if I accept the figures mentionned in the text, I agree when coming back figure 3.9 that there is no risk for i ~ 0.2 and

-ffg ~ 0.15 as L}G ~ 0 This is confirmed by the ABAQUS calculations of the tsvo most energetic explosicn calculations obtained by Phf ALPHA plus ESPROSE-m.

O-8

,, Ber23. As a conclusion, I can say that if we accept the scenasio which is

( retained by the authors, I think that whatever my remarks about premixing quantifications - the AP C00 RPV cannot be challenged by a steam explosion.

110 wever, I uvuld like to have more established conBrmations of this scenario by mechanistic calculations when possible or parametric calculations when it is not.

The main thing to be conSrmed is the impossibility to have a downward relocation i.e

- the possibility to have a break down of the lower fission gas plenum rather than a continuous draining. This will ght a sudden access of the core support plate holes to the melt

- the inBuence of the already relocated metallic pool on the oxidic release. It may take a longer time to break through and the blockage integrity may then be challer.ged. The inBuence of an interfacial resistance between the oxidic solid crust and the wall- specially at the top of the pool- will also participate to an inerease o[the time oibreak tbrough and oithe evaluation oimelt superheat.

Other branches for tl.a scenario should also be e. luated:

- the possibility of the metallic melt to rapidly go through the core barrelleading

'G/ to metal water steam explosions

- the possibility to have steam explosion at later times (in the oxidic case) when water is sloshing back after a first small scale (i.e low energetic)estat As for reflooding scenerlo, the fact that water will be closed to saturation should also be evaluated.

This is a summary of points made and responded to point-by point.

Dur1. 1. Purpose, Procedure and Main Conclusions of the Study The purpose of the uvrk is to show that the lower head of a reactor like the APG00 withstands the load of steam explosions. According to the ROAAM phl-losophy, all physically meaningful causal paths that could lead to failure have to be imestigated. The decomposition yleids the following central areas of analysis:

1, Since pressures in the kilobar range have been obtained in the KROTOS experiments (although not yet with corium and in one dimension), a direct exclusion oflower head failure cannot be done. Thus, detailed calculations O

V G9

of possible explosi.n loads are required, taking into account the speelBc ge-ometry with respect to venting effects This is done by use of a 3D version of ESPitOSE.m.

2. The possible spectrum of melt / coolant mixturcs developing in the lower head due to an assumed core melting must be determined. This is done by use of a 3D version of PM ALPilA.

3. Possible thnings and strengths of triggers have to be considered. Due to the uncertaintles, an enveloping approach is pursued here, concerning the timing as well as the strength.

4. Since close to the wall pressures in the kilobar range are obtained in the calculations, speclRc investigations on failure criteria are required. This is done by a si:nple estimate and also by means of the ABAQUS code. Con-siderations on the possible interaction with thermalloads are also required.

5. Considerations on the melting and relocation process in the core and the release to the lower plenum have been considered as necessary for restricting the possible rpectrum oimelt/ coolant mixtures. This is done by separate estimates.

The midn arguments in the report are:

it is assumed that a pool of ceramic melt surrounded by crusts forms in the core due to the cooling capabilities of remaining water in the lower plenum (at the beginning of melt motions with level at ~25% of acthc core height) and the large heat capacity of the lower part of fuel bundle with lower Zr plugs and the lowermost spacer grid. The key points are then that a downward relocation path of melt through the core suppcrt plate is excluded and meltthrough of reRector and core barrel is assumed yleiding Snally a siderards relocation through the downcomer.

This sidewards relocation is restricted in extent assuming failure at the upper end of the pool based on the analysis of heat transfer from the pool and assuming plausible failure sizes. Estther, it is argued that only one failure location is available within relevant times for pretnixing of the relocated melt in the water and triggering. Strong voiding of the mixtures under the expected conditions of saturated water is expected and calculated by 3D-PM ALPHA. Thus, only small amounts of melt in the lower tens of kg are considered to be potentially explosive.

G 10

n This is taken to directly exclude large break possibuities for the lower head.

() Various calculations with ESPROSE.m assuming sufficiently strong triggers are additionally t.aken to exclude also local threats to the RPV. This is Bnally done by comparison with fallare criteria for the RPV wan yielding directly (without application of the probabilistic framework) the conclusion that failure is physically unreasonable.

Ehrther, renood scenarios are evaluated to even mitigate the possibility of vapor explosion threats, due to cooling and preventing melt outBow. Mixing with the melt in the poolis not considered as effective (smallyleid ofstratined explosions).

In addition, preventing outBow by reflood would also mean to exclude mixing of melt with highly subcooled water, which is considered as the only case with a potential to challenge the lower head due to increased penetration depths without excessive voids.

Cases with thermally weakened RPV wn:Is are restricted to later phases of melt outflow. Then the water and mlxturcs are already assumed as strongly volded.

ReBood FCis are considered in the report in stratHied configuration, i.e. water above a metallic layer. A threat is excluded due to rapid spontaneous interactions p with the subcooled water and rapid freezing of the metal surface before a thick V ater layer establishes which cculd yield sufficient constraint for strong psessure buildup.

Based on these considerations, the major conclusion of the study is that steam explosion induced lower head faHure in an APG00 like reactor is " physically un-reasonable".

DGr2. 2. General Comments The procedure as well as the general arguments are convincing. This concerns especinHy:

l

• The argument that a strong cold trap at the core support plate, especially if stilI connected with water, can prevent the downwards release path to occur l before sidewards release at the upper regicn of the melt pool. This yields a signincant reduction in melt Bow rates to the water, especially to a possible downwards release in multiple stteams. I( the cold ttap at the bottom L

/ \

l 4

\d G-l l

strong enough, no downwards release will occur until all melt is released sidewards due to a continuous failure progression.

• Then, the saturated coolant condition prevents larger premixtures without high volds, since larget premixtures could only develop within longer times.

The geometrical conditions of the Bow through the downcomer also favors this, o The strong voiding of mixt urcs calculated with PM ALPilA is thus plausible, o With the small mixtures (small melt masses) of rot extensive void plausible from the above statements, the ESPROSE results are also plausible (the obtained pressures even appear astonishingly high - probably due to the restricted venting).

e Thus, also the conclusions on the tinents are plausible.

• The high number of calculations with PM-ALPilA and ESPROSE covering a wide range of conditions can also be taken as svoporting.

In spite of this agreement in principle, therr main problems in the details of the argumentation and performance of the analysis. Improvements may be performed to even better confirm the statements and ( .nclusions as a basis for use in licensing actions. This will be discussed as follows in some detail. Since, in my opinion, the Statements on the relocation path are a most critical point, I Brstly concentrate on this, then considering the subsequent analyses on premixing and explosion. I will not consider the aspects of structurel failure criteria which appear to be well established. I will only give (cw arguments on possible further scenario aspects concerning reBood.

Chol. In response to the request made in your letter of June 17, 1996, I have reviewed the report " Lower licad integrity Under In-vessel Steam Explo-sion Loads" by T. G. Theofanous et al. You indicated that this report and a companion document together " intend to demonstrate the effectiveness of 'in-vessel retention' as a severe accident management concept for a reactor like the A!YOO". You kr*her indicated that "the purpose of this review is to assess whether this intent has been achieved to a suBicient degree for the results to be of use in the regtdatory/acensing arca". Based on my review of the report, I Bnd G 12

1 1

that additional supporting work would be nwded if the conclusions of the report were to be used in the regulatory / licensing area.

&m The nature and need for the " additional supporting work" identified are addressed, point-by point,below.

Cho10. On page 71, the authors state that "Also in this chapter, we would normally present a series of arbitrary parametric and sensitivity calculations, to illustrate, for cases where the base results happen to be benign, the margins to failure" and claim that "This, in effect, has already been done, too, by the breakup and triggering calculations, in the course of bounding the behavior". I believe additional work would be needed to make this claim fully valid, and I am confident that the authors will succeed in doing that.

If, notwithstanding the above and the additional work provided in the various addenda included as a part of this package of responses, the reviewer has specific suggestions for further calculations, we can consider carrying them out.

Cho11. Finally, the authors are to be commended for conducting such a detailed evaluation of a very complex issue.

O(~N Fau5. Therefore, we are of the opinion that the apprcach taken in ;his docu.

ment is conscrutive in that it overstates the possible loads that could be created as a result of thermal explosions. If a design evaluation uses this model and con-cludes that the boundary would not be challenged, we believe that the conclusion is sound. However, if the modeling approach is used and the resulting loads ex-cced the capabilities of the structures, we do ncat believe that this represents an actual challenge to the system integrity.

Pau'f. In summary, we believe the modeling of fuel relocation and quantili-cation of premixtures to be reasonable and consistent with experimental observa-tions including the TMI 2 incident. On the other hand, the assessment of steam explosion loads appear to be very conservaths. The corium saturated water sys-tem is not likely to exhibit "explosivity". Therefore, a very strong case can and has been made for the elTectiveness of "in wssel retention" as a severe accident management concept for a reactor like the AP600.

?

! )

V G 13

Fico. Phase find enclosed my review of the DOE project on " Lower Head Integrity Under in vesselSteam Explosion Loads" by Theofanous and co-uvrkers.

As you will see from the review ljudge it to be an excellent study in tenns ofits depth, sctipe, technical quality and shear volume of work. I fully agree with the conclusions drawn by the authors.

Fiel. Summary This review covers the study oflower head integrity under steam explosions per-fortned at UCSD by Theofanous and co.uvrkers, together with the code validation reports for PM ALPHA and ESPROSE.m. The study and validation reports cot:-

tain a massive arnount of very high quality work. The depth of the study and extren es to which the authors have gone to use validated tools is second to none world wide. For example, no one else is performing 3D premixing and propagation calculations.

The work is of vezy high quallt.' ' In my view the conclusion that steam explo-sion induced lower head failure is unphysicalis completelyJmtiBed. The technical arguments uupport this with a high degree of redundancy.

1 Introduction Firstly, I believe it is important to comment on both the quantity and quality of the documentation supplied for this review. The very complete verification manuals for PM-ALPilA and ESPROSE.m are unique. A minor semantic point but they are much more than verification (which implies that the code does wipat it should) manuals but are also validation manuals as they examine how well the code represents real experiments.

Secondly, I wish to record that I was impressed by the scope, depth and quality of this st udy. It provides a very comprehensive basis for rejection ofsteam explosion-induced failure of the lower head.

The remainder of this document presents specific comments or: the Study and the two validation reports.

G 14

l l

i Fle2. 2 The Study (DOE /ID-10541) p>

This section deals with the main document of the study (DOE /ID-10541) and

[.uys partlcular attention to the steam explosion part of the study.

2.1 Introduction This section gives a brief summary of earlier work on lower head failure. It discusses three earlier studies by Bohl et al, Theofanous et al and Thrland et al, all of which highlight the need for mechanistic pressure loading calculations before the lower head issue could be addressed adequately. This is the Brst such study in which this approach has been possible.

Flc8. 2.7 Integratlon and Assessment This very brief section explains that as a consequence of the methodology and results there is no need to continue with the probabilistic apptcach because of the enormous mismatch betmn explosion loads calculated and those required for failure. In order to show that this is not an artifact of the approxhnate structural treatment, full ABAQUS calculations showed there to be no problem.

O G 1 agree that the only way to obtain a signincant explosion is to have extenshe mixing which requires highly subcooled water. I believe the arguments assinst this are sound, especially if one keeps in mind that the enormous amount of heat which would be stored in the lower core support structure would bc available to remove subcooling.

Fle10 2.9 Conclusions The conclusions contain a suramary of the results presented in the earlier chapters und presents a concise summary of the important physical features of the system and the physical mn:hanisms which lead to the conclusion that failure of the lower head by a steam explosion is unphysical. I re, :ly appreciated this carefully presented summary.

bi b G 15

Fle12 (in part) 3.6 Concluding Remarks I think this section identines the correct areas for futute focus. If I were the authors i uould have made mors of the fact that this is the most comprehensive validation effort to date and that the code has performed extremely well.

3ac1. 1. Introductory temark in order to put my comments to follow into the right perspective, I must state first of all that i fully agtw with the general approach to the problem taken by the authors, i.e. the ROAAM. To what extent probabilities are used within this approach may depend on the purpose and the problem of the study. However, dividing the problem into its physical aspects, treating them in separate parts of the study that can be scrutinized by other experts and linking them in a well denned and verinable way dennes a clear path towards the sesolution of the full problem.

Similarly I fully support the basic apptcach taken to treat the steam explosion problem. The material presented is based on and incorporates a lot of pioneering and exemplary work in this Beld. I do not want to shed any doubt on that. The only question I'm discussing is: Is the state of development sufficient to Rnally answer the question under discussion. This forces me to elaborate on potential weak points in the argumentation. If a tecimical Beld isn't developed sufficiently, even a ' peer review' cannot Snally ensure the correctness of an evaluation.

Quite obviously, steam explosions are not phenomena that are well understood in the scientiSc sense, especially i( we are concerned with such large-scale events as are discussed in connection with reactor safety analysis. Unfortunately, such ewnts lie far outside the parameter range that can easily be studied experimen-tally. This is true of the initial temperature and the composition of the melt as well as the masses imohrd (as mentioned above). This dilemma forces us to largely rely on codes for extrnpolating from the accessible parameter range to that of the envisaged accident situations. Ideally this extrapolation requires full knowledge and appropriate modeling of all relemnt phenomena. Here again we are ccafronted with gaps, the relevance of which is dlBicult to judge. The concept of'Stnest for purpose' may be helpfulin areas in which the consequences of neglecting something can be estimated. But how about problems which have G 16

, not yet been identined or the importance of which has not yet been percehed?

&, In the present state of knowledge bad surprises cannot be excluded. The (only?)

way to deal with this diBiculty is to account for all (known) possible traps in the analysis (take a conservathe approach) and to require a large safety factor. 'lls some extent, this principle is followed in the study discussed here. But in my judgement not to a suf&cient extent.

Jac2. From the point of view of quality assurance, a peer review like this one ca'i bcLame fully effecth'e only if at least the background material was published since quite some time so that a thorough discussion ofit has been possible among the experts. In the present case an important part of the background material was dethcred very late during the review process. This reduces the relevance of the present review process.

A key point of ROAAM is that the review is not hurried through. Valid concerns are pursued for as long as it takes. In the present case, all documents were supplied by the end of September 1996. This particular reviewer was informed by DOE's project manager that he could take as long as necessary to supply his review, and it was sent about two

/9 months later, by the end of November 1996. Our t:sponses, including updated versions V of the reports, are made available to the reviewers a little more than 9 months later (about September 15,1997), so that at this stage the piocess has been on going for about 13 ear.

This was done by design, and, again, will continue for as long as technically substantive concerns exist. Meanwhile, the work was also presented at the CSNI FCI meeting that took place at Tokai, Japan, during May 19-21,1997.

Jac3. 2. Scope of this review This review is concerned with the steam-explosion aspects of the study. The con-tribution of this part of the study to the posithe Snal conclusion, i.e. Interacting masses that are insignincant from an overall eneigetic standpoint and even local loth that lead to clastic strain only, can be attributed to small pouring rates, a strong voiding of the premixing zone and early explcsions. The Brst of these are to some extent a consequence of the melt water mixing scenarlos chosen and although core melt down is not my proper Scid of experience i must make a few comments on this because the wpy in which melt and water are brought into con-tact is basic for the subsequent events. The possibility of a small steam explosion G 17

Induclag a larger one is neglected altogether. The second pot , i.e. the proposed strong voiding of the zone in which corium incit and water are intermixed prior to an explosion (the mixing zone or premixture), is lustrumental in two ways: In addition to the small pouring rates it reduces the interacting masses, At the same time, this voiding seems to be one reason for the dying away of the energetics of explosions with increasing time of triggering which is the most convincing argu-ment for considering early explosions. Of course, this linding also depends on the third point, i.e. the way in which the steam explosion p:vper is modeled. 7'he above three aspects, i.e. scenarios and modeling of premixing and explosion are discusses one after the other below.

Jacto. 4. Summary The cllltmative final result of the study fol!ows from three findings: low corium.

water mixing rates, very high vold fractions in the premixture, and, partly de-pending on that, effecthe explosions being possible only during a subsecond pe-riod at the beginning of premixing. I have serious doubts about all three of these.

With respect to the melt relocation scenarios I doubt that the present state of knowledge allows to definitely exclude downward relocation paths that could lead to much larger relocation rates. Not really being an expert in this field I must leave the Judgement ta those experts. provided they can posithvly defeat my arguments, in addition, processes that are induced by a first (weak) steam exph>.

slon might lead to a more etfective melt water mixing and thus to a larger steam explosion. With respect to premixing, the very high vold fractions predicted by the code Phi-ALPHA even outside the gas clumnel that immediately follows a mass plunging into weter don't seem to be supported by experimental evidence.

The code itselfis not provided with sufficiently mechanistic models and is not sutliciently validated to support the high void fractions by itself. With respect to the explosions, the failure of the code ESPitOSE.m, i.e. the prculiar interaction modelin it (the micromteraction model), to predict efficient explosions in highly volded premixtures, doesn't prove that such explosions were not possible on the base of different interaction mechanisms, even i(highly voided states would occur.

We strongly disagree with all points in this summary. Each one of them has been refuted above. Apparently the reviewer cannot see the impossibility of propagating supercritical G 18 h

p thermal detonations in highly voided premixtures. This is the most trivial part of the Q subject, and until he resolves this in his own mind, we can make no progress here.

May1. Not being an expert in structural mechanics, I shall concentrate my review on the thermo-fluiddynamic part of the report, trying to ght an overall assessment.

For my review, I also took into account the report DOE /lD 10503 " Propagation of Steam Explosions: ESPROSE.m Verification Studies", a paper by S. Angelici u.a. on the Mixing of Particle Clouds Plunging into Water /1/ and another paper by Chen u.a. on the Constitutive Description of the Microinteractions Concept in Steam Explosions /2/.

1. Problem There are many papers in the Internationalliterature dealing with the phenom-enn and the effects of steam explosions. They differ widely in their statement on explosion loads depending on assumptions or predictions for premixing, heat transport between molten fuel and comvrsion of thermal energy into mechan-leal loads. Experiments were made with various melts, representing a variety of boundary conditions (from one dimensional to multidimensional) and a wide J range of scale.

The report under discussion here does deliberately not snake the hopeless attempt to find an agreement or an average between the wide spreading results of the literature. It furthermore is based on carefully planed experiments, performed by some of the authors and on constitutive descriptions of phenomena, involved in steam explosion processes.

Object of the study is the advanced pressurised water reactor APG00 or respec-thrly the integrity ofits pressure vessel against hypotheticalloads of steam ex-plosions.

Entering theJungle of phenomena and etlects connected with and resulting from steam explosions with the aim to come to a quantitath e and physically reasonable result with respect to the mechanical behaviour of a pressure vessel is a task, which casmot be fulfilled in a complete, best estimate way on the basis of today's mvrall knowledge. This is the case in spite of the fact, that numerous research work has been performed world wide and that the authors of the report, being undct discussion here; made excellent contributions, analysing steam explosion f 0 19 l

phenomena and ellects in a thwretical and in an experimental way. There are many intangibles in steam explosion processes. Deing forced to demonstrate the safety margins of a pressure vessel against steam explosion loads in a way, which is resistant against critical questions, it is quite obvious to apply conservathe assumptions. .

The design of the APG00 " invites" such conservative usumptions, because, be-sides the low power d.~nsity, the core is not only surrounded by a pressure vessel with a rather thick wall, but also by a stainless steel teBector inside the core barrel. So the APG00 design can " tolerate" conservative assumptions. By doing this and regarding the results, one har to be very careful with any attempts to transfer the data, obtamed for the APG00, to other pressurised water reactors.

Conservatisms, assumed when calculating the thermo- and fluid-dynamic situa-tions during steam explosions, could lead to predictions with respect to pressure vessel failures, which nu far beyond the physical reality under such an hypo-thetical accident. Therefore, inspite of the Bnc work presented in the report DOE /ID-10541, there is still a lot to do to obtain a still more realistic basis for safety analysis and realistic predictions. However, we must also be aware of the fact, that there always will remain many intangibles within the scenarios of hypothetical severe accidents.

May5. 5. Integrat!on and assessment in the chapter 7 " integration and assessment", there are two very important statements, namely that

- from a more global perspecthr, the only H.y "to potentially produce a sig-nificant structural challenge on the lower head, would be by having a highly subcooled pc 31 in it" and

- "tven a postulated rapid renood scenario could not produce the condition of concern..." .

After depressurising the primary system, following an hypothetical, severe acci-dent, there is always and everywhere saturated (not subcooled) water in the lower plenum of the pressure vessel. This would be true not only for the AP600, but also for all other pressurised water reactors.

G 20

So as long as one can guarantee, that the cooling of the lower core support structure is good enough to prevent it from falling and core melt Bows from the i side to the lower plenum, steam explosions, originating from it, should not be a ,

problem.-

The second statement is as important as the Brst one, because it eliminates doubts, existing up to now, whether it would be advisable to try to Bood a  ;

degraded core again after a certain escalation of a severe accident. Tids point '

l was brieBy discussed already in a former chapter of this review. Therefore in '

future accident management planning, there should be ghen more effort to in-vessel cooling also after a partial core disintegration. -i i

Mayo. 6. Conclusions I ful.*y agree with the conclusions presented in chapter 9 of the DOE /ID-10541 report, to the statement of the authors, that "because of the wide margins, due to these controlling physics, it has been possible to bound uncertaintles to a ,

suBicient degree...", I would like to add, that these " wide margins" are still on the conservative side and the mechanicalloads onto the pressure vessel and its lower plenum would be lower in case of a hypothetical severe accident, than predicted in the DOE /ID-10541 report. j Finally I would like to congratulate the authors to this line work, attacking a very l dlBicult but important problem and solving it to a great extend from an engineer-  ;

ing point of view, but based on controlling physics and on reliable constituthe laws for the Bulddynamics to be expected in steam explosion scenarios. .

I 1 S. Angelini, T.G. Theofanous and W.W. Yuen, The Mixing of Particle clouds Plunging into Water, NURETH 7, Saratoga Springs, NY, September 10-15,1995, 1

NUREG/CP-0142 Vol. 3,17541778 2 X. Chen, W.W. Yuen and T.G. Theofanous, On the Constituthe Description of the Microhiteractions Concept in Steam Explosions, Proceedings NURETH-7, Saratoga Springs, NY, September 1015,1995, NUREG/CP-0142  ;

Moo 0. . Thank you for an opportunity to review this work. I think it is one of the most signliicant pieces of research I have ever reviesvd. It is of both current .

l and long-term importsace to the nuclear industry.

0 21 1

. + - . - .. , - . -- - - - . - . . . , - - , -. .-- -- - - - . . - ,, , . . + - . .

I Since I have spent my career in the nuclear energy business, I personally ap-preciate your long range viewpoint for energy needs, which is obvious from your support of this program.

M001. The purpose for reviewing the subject report, with several other com-panion documents, was to assess whether 'in vessel retention" h demonstrated to be an cflective severe accident management concept for a reactor like the AP600.

I have reviewed the work, and conclude that in vessel retention has been shown to be an cffcctive severe accident management wncept for reactors with geometry, fluid 'uantitles, event sequencing, and thermophysical properties similar to those pertaining to the AP 600.

The documents provided for this review describe the steps taken to understand and predict the complex, multi faceted subject of a, team explosions. Associated phenomena have been closely simulated by experiments, and predicted with de-terministic theoretical formulations (causal relations) to a degree of accuracy that makes confident pre &tions possible for full size AP-600 systems. It appears that all controlling physical ellects have been included, even without the need for a complete understanding of the exact timing and conditions necessary to trigger steam explocions. Already known or conservalhely estimated ranges have been placed on parameter, timing, and scenario path uncertainties, and stillit has been shown that the expected range oflower head steam explosion pressure loads do not intersect the vessel fragility curve.

I was asked specifically to review the material on steam explosion loads, as dis-cussed in

" Propagation of Steam Explosions: Esprose.m Verification Studies" by T. G. Theofanous. W. W. Yuen, K. Freeman, & X. Chen, DOE /ID 10503, August 1990.

The documents provided for this review collecthely lay an extensive foundation of information, which testifies to the tecimical stature, competence, thorough-ness, and integrity of the investigators. Indeed, the overall work is monumental in its scope and acidevement, and it is communicated in a writing style which is one of the most scholarly to be found in reactor safety studies. Both the au-thors and sponsors should be commended for a carefully formulated investigathe G 22

strategy (strong, in depth, well biended steps) resulting in the highest value ob-tained for the time and resources spent. Beyond steam explosions the progress and understanding achieved in this work are likely to exert a major benencial influence, both methodological and technical, on other significant and complex thermal hydraulic issues.

SUMhiARY

1. The ROAAh! has shown that vessel loads, resulting from a comprehensive range of severe accident scenarios, melt conditions, relocation flow, timing of release from the core region, and thermal hydraulic processes between the inelt and surrounding water, lead to the conclusion that vessel failure is " physically unreasonabW' in an AP GOO type reactor. Parameters inclading pool gwmetry, melt release rate, shock explosive formation and propagation, and venting yleid Icad distributions on the vessel wall which were compared with the fragility curve in order to arrhe at this conclusion. It is my opinion that even though all the mechanisms contributing to stesm explosions are not fully understood, results embrace the extent of refinements which could eventually be made by further experiments and theoretical model (causal relation) development.

2. I agree that it would be useful to obtain data from the QUEOS experiment for O a fully saturated water system) although it would not change the conclusion that vessel failure in AP-GOO type reactors is " physically unreasonable " The value in such a test is to fill in a parameter range to give a more complete data base, and permit the technology to be extended to non-AP GOO type systems.

3. One potential benefit of the ROAAh! procedure ~is that it conceivably could be used in reverse. Suppose it was concluded that a system failure probability was larger than acceptable. The ROAAh! could be employed to display which pa-rameter(s) dominate the outcome, thus pointing the way for design or procedural changes to reduce the failure probability.

• • • • • • • • • • • • • • • • • • • • +**

Moo 16. OVERALL CONCLUSION As a curious person who cidoys formulating better theoretical models, based on more complete experimental understanding, I recommend additional experiments (e.g., QUEOS experiments with fully saturated water) to help close the few re-maining gaps in our understanding of steam explosion phenomena.

G 23

However, I believe that the studies provided for this review give substantial, in-depth evidence ta help conclude that in vessel retention is supportable as n sev*re accident management strategy in AP-600 type reactors without additional work to close the issue.

Sch1. 1. Review of the Ovesall Approach This is the fourth time I have had the opportunity to review a body of work that Professor Theofanous and co workers have produced for the resolution of a specific safety issue, or a specific concern. I believe, this is the most complex of all the issues (or concerns) :o far and I believe, Professors Theofanous, Yuen and co workers have done their fimst work so far. This body of work is of greater, and of more lasting, value, than earlier efforts, since a major part of this work is the development and verification of the methodology to describe the steam explosion phenamena, and to predict the loads imposed by the postulated occurrence of a steam explosion. This methodology, and the codes developed, could be applied to athet eccident scenarios, than the one considered in the present application.

I believe, some comments are in order on the overall approach followed in these tbree reports, complemented, oicourse, with the ROAAM methad, and the pre-vious work that Professor Theofanous and his teams have performed, (e.g. for the Alpha mode failure of an LWR containment during a severe accident).

Professor Theofanous and co workers, with their accumulated experience in steam explosion modeling and applications, have developed a very well focussed overall approach in the body of work presented in the three reports. it is clear that an in house experimental program was structured to provide the key observations, for the ideas needed, to advance the stcam explosion modeling to the point where some meaningful predictions can be made. The innovative experiments performed in the MAGICO facility provided the germane ideas on steam depletion, and on the difficulty of obtaining pre mixtures, which could lead to very large steam explosions. Likewise, the experiments performed on the SIGMA facility provided the basis for the micro-interactions concept for the steam explosion itself, i.e., the concept and treatment of the m fluid. I believe, the experimental underpinning of the ideas and concepts employed, and the further verification of the methods used in the codes against the integral experiments, has provided great strength to the overall approach.

G 24

The overall approach followed, in the application report, conforms to the ROAAM method and employs the PM ALPHA and the ESPROSE-m methodology. The extremely high values for the fragility curve made the task much simpler than the earlier applications of the ROAAM methodology, but it is wellJustined and credibic.

Perhaps, the two points of possible short coming in the overall approach, which have also been admitted by the authors, should be stated:

First, is the question of maturity. Clearly, there is not enough separate effect and integral-effect data to provide sufficient validation of the steam explosion methodology developed. This methodology employs many many correlation and submodels, whose individual verification is a monumental task. Nevertheless, an experimental verincation matrix should be developed, with priorizasion of im-portant c!fects, and executed, to provide greater veri & cation of the methodology, thereby providing it greater maturity.

Second, a mechanistic treatment af the initial phase of the steam explosion sce-navio, i.e., the break up of the melt jet, and its sequential fragmentation, has not been included in the methodology developed so far. The authors claim that this phase oi the stcam explosion process can be conservathcly-bounded param-eterically. Perhaps, the authors hav: done that successfully in this study, how-ever, a more general treatment of the break up phase, and its linking with tne pre-mixture phase, should be pursued to provide greater assurance that all the initial-condition-effects have been taken into account.

The above two points, in no way, diminish the value of the overall approach, and the results achieved. The above two are outlines of further work to solidify the validity of the overall approach followed here, as, I believe, the authors have themschts identified. The present treatment of the physics is the " State of Art."

I beliew, rapid advances in understanding and modeling will follow the germ of ideas that the authors haw provided here. Some of those advances will surely be accomplished by Professors Theofanous, Yuen and co-uvrkers.

Sch34. IV. Review of the Report: Lower Head Integrity Under In-Vessel Steam Explosion, DOE /ID-10541 by TG Theofanous, W.W. Yuen, S. Angelini, J.J. Sienicki, K. Dreman, X. Chen and T. Salmassi O G 25

This report is concerned with answering the question: "Will the lower head of the advanced passive reactor AP 600 fall, under the dynamic loading imposed by an in vessel steam explosion, if it were to occur?" This is an important issue for the accident management strategy chosen for the AP-600, i.e. retention of the care melt in the lower head, by employing external cooling of the vessel.

The methodology used to resohe this issue is the ROAAh! method developed by Prof. Theofanous, employed most recently to respond to the companion question "Is it possible to retain the molten core of the AP-G00 reactor, in the lower head by cooling the vessel externally?" This question was answered in the affirmathe by employing the ROAAh! method. The ROAAh! method has been extended and further clari?ed by Prof. Theofanous in a recent publication, attached as Appendix A in this report.

Besides the ROAAh! philosophy and procedures described in Appendix A, the detailed premixing and explosion results are described in Appendices B and C respectively. Appendix D provides additional pre mixing perspectives from the THIRhfAL code, prepared by Drs. Chu and Sienick! of Argonne National Lab-oratory. The important chapters, in the main body of the report, are concerned witn structural failure criteria, melt relocation characteristics, quantification of pre mixtures and explosion loads and finally the assessment of the integrity of the lower head of AP 600.

In the following paragraphe. I have provided comments on the appendices, chap-ters and conclusions of the report in the order:

- Chapter 3: Structural failure criteria

- Chapter .t: hielt relocation characteristics

- Chapter 5: Quantification of pre-mixtures

- Appendix B: Detailed pre mixing results

- Appendix D: Additional pre mixing perspectives from the THIRh!AL code

- Chapter 6: Quantification of explosion loads

- Appendix C: Detailed explosion results

- Chapter 8: Consideration of reflood FCis

- Chapters 7 and 9: Integration, assessment and conclusions G 26

4 g Sch52. (in part) IV.7 Chapter 7. Integration, Assessment / Chapter 9. Con-V clusions Thece chapters combine the results achieved in the previous chapters and appen-dices to provide an overall assessment. This work was already practically done by the results achieved, since the maximum impulse loading was below the min-imum of the fragility curve. This was also confirmed by performing ABAQUS calculations for the peak loading for the actual cases and finding that the lower head strains were very low.

The authors conclude that for the saturated water case, the lower head integrity can not be compmmised by a steam explosion. Having highly subcooled water is the only possible way to, potentially, involve a larger mass of melt, and produce a more energetic explosion. The authors conclude that obtaining highly subcooled water, even in reflood scenarios for the AP-600 is not credible.

Sch53. V. Concluding Remarks In this section, I would like to provide a few concluding remarks after the review of the three reports.

A O I must congratulate the authors for producing such a fine and comprehensive body of work treating the tricky and controversial area of steam explosions. While, most of the researchers in this area are still trying to understand the fundamentals, the authors have leaped ahead with new concepts, advanced codes and considered judgements to provide a tensanably robust estimation of the damage potential of a steam explosion. They have combined this with structural analysis to show that AP-GOO lower head can withstand the dynamic loads imposed.

The authors have, also, noted the peculiarities of the AP-600 configuration and emplo>vd the advantages and disadvantages they confer on the analyses. Some of these peculiarities (ditferences) provide great advantages e.g. In the core melt progression and the melt release characteristics. These sound a little bit too comvnient and, perhaps, should be re visited.

The work was done without any regard to the " convenience" of the results.

Sch54. The authors have modelled the fuel break-up and fragmentation pro-cess only parametrically L'his may be a weak point in the whole development; since those processes provide the initial . anditions for both the pre mixing and p

b G 27

}

the propagation phases of the steam explosion. Perhaps, the analyses are well-bounded for these processes; howevez, the sensitivity of the results to the break up and t.%e fragmentation modeling is very large.

We do not agree with the thrust of this conclu sion. Only breakup is treated parametrically, not fragmentation. The main result is that premixtures void, and this is not too sensitive on the breakup used. We bound the behavior with respect to this parameter, and this 8.s much more reliable than trying to assert the result of some predictive model. Such mod-y els, evan if eventually developed, could never be verified at the appropriate level-i.e, the dynamics of breakup as it occurs under realistic conditions. Our treatment of frag-mentation, on the other hand, derives from the directly appliable and well-characterized SIGMA experiments. The andenda to Chapters 5 and 6 show rnore clearly the bounding i

nature of the results.

Sch5E. Then, there is the question o. 'naturity and of validation versus ver-IIicatlon. I believe the methadology at d the data presented, robust a they are,

are still very new. The comparisons presented against test data are not exten-sive. and I think, the authors recognising this, have wisely titled the reports as '

veriEcation reports. Ehrther experience with this methodology and further com-

[ parisons with separate-etlect (e.g. SIGhfA, hiAGICO, BILLEAU and QUEOS) data and integral-effect (e.g. FARO and KROTOS) data would provide valida-tion and maturity to this methodology. in particular, the constituth'e relations, being so many for such complicated phenomena, aced greater experimental back-up. I believe, the av:. hors are already busy i.n achieving such experiments in the h!AGICO and SIGhiA facilities. ,

- Actually, we did not intand v make a distinction between " verification" and " validation."

" In the sense that the validation term is described here, we believe the two codes have been adequately validated in the " fitness for purpose" sense. Of course, work will continue, and thus maturativa will gradually develop.

Sch56. Lastly, I must say that I1 ave enjoyed reading the reports and learned much from them. I think, I now understand the cncept of micro-interactions and the m Ruid. I have made constructive (hopefully) critical comments at places, to provide input to authors towards improvement of the rep ~ts. I believe, they have largely acideved the objective they had set out to achieve.

L G-28

~- 3 39

Tur1. OVERALL COMMENTS This report and its associated documents represent the culmination of several years work by Prof. Theofanous and his colleagues. They have now demonstrated that the basic framework for a steam explosion as-sessment in realistic geometry is in place. This is a major achievement.

The reliance on detailed modelling codes makes the reviewer's tssk dif.

, licult - ha che end one can look at the validation offered and consider whether the results presented look reasonable. In the supporting doc-uments the authors make good use of the available experimental data to benchmark their calculational models. However, it is accepted that some of the constitutive physics used in the premixing and propagation codes is uncertain, as are, to some extent, the melt pour characteristics.

A review, such as this, can indicate that the codes appear ' lit for pur-pose' but cannot ght a full endorsement for all the models they contain, without signliicantly greater effort.

The situation considered in the application presented, a modest pour of <

melt into saturated water at ambient pressure, is not conducive to large steam explosion loadings, and this is demonstrated by the calculations presented. Suricient perameter variations are investigated to indicate that this is likc.'y to be a robust result for these couditions. As indicated below, my residual concerns relate to the confidence in having low pour rates and the possibility of operator actions leading to some subcooling.

• * * * * * * * * * * * * *. * *

• w * * * *

• Tur2. SPECIFIC COMMENTS Chapter 2: Problem Definition and Overall Approach

1. Although the text makes clear that it was an intentional conservatism not to claim ciedit for lower head venting in the Sizewell B study, it is wrong tc interpret this in the phrase lowerheadfailure cannot be dismissed as readily any longer. We found that the previous claims for lower head failure condd not be substantiated as large explosions did not necessari!y imply sustained high pressures.

A i'u) G-29

i Tur3. 2. It is arguable whether the ' essential basis for the current work'is the progress made in modelling explosion propagatim and the pre-mixing phase, or in the assessment of melt progression.

Look at the pressure pulses predicted, the fragility, some of the reviewer's com-ments, including this one, and then let us imagine where we would be without auch progress in premixing and propagation (mictointeractions in particular).

Tur4. 3. The statement that between 3 to S tons offuelmustparticipate to pmduce a 1 G) explosion, and consequently incipient lower headfailure, raises the question of whether larger explosions are possible that do not fall the lower head.

This is an old result. Probably yes, but tnu Is not our concern here.

Tur34. Chapter 7: Integration and Assessment

1. The conclusions reached are justified on the basis on the analysis presented. On the basis of current knowledge I am still not comfortaNe with the observation that downward relocation scenarios are ' physically unreasonable'.

See above responses and other reviewers' questions and our answers in this area.

Tur37. Chapter 9: Conclusions

1. I have indicated above that my principal reservations lie in the areas of the downward blockage and in ensuring that there are no operator actions that may prejudice the assumptions made in the analysis. I agres with the authors that consideration of additional pathways is unlikely to change the conclusion.

Youl. This is a massive piece of work which includes, c addition to steam explosion loads, some areas with which I am not particularly familiar, such as probability methodclogies and plugging behavior of molten materials. I will therefore comment mostly on the steam explo-sion loading.

The approach taken in this report to determine steam explosion loads is essentially the one that has been recommended by most if not all steam G-30 1

I l

explosion researchers: use of computational models validated against p& experiments to determine bounding emclopes for reactor accident sce-narlos. Prof. Theofanous 'sas taken an additional step here in simplify-ing the probabilistic framework with his ROAAM method; I think this is entirely in keeping with the use of these type of calculations in risk assessment and rulemaking. I believe that the work described in this report has successfully accomplished the goal of enveloping the steam explosion locding. The usefulness of the results in rulemaking, however, therefore depends on the confidence placed in the it.!tial conditions of the accident scenario and in the analytic tools used.

In regard to the initial conditions, it is very important to the conclusions rec.ched in the report that the melt be introduced tbrough the side of the reflector and that the lower core structure and support plate be plugged As I mentioned before, plugging is not my area of expertise, so I will not comment further other than to point out again that confidence in the initial conditions is very important.

in regard to the analytic tools used for calculating steam explosion loads, I have some comments concernhag possibic gaps in the cases considered

' and in verification of certain parameters used in the models.

You7. Allin all, this appears to be a very complete piece of work.

/ \

\C) G-31 1

l

t Methodology O

Ban 2. I think that the ROAhthi approach makes very good sense for rare, but high. consequence, events. I believe that a similar approach has been used before, but never so explicitly and clearly stated. In particular, the recognition that there are " intangibles" which will never be known in advance, conserm!vely bounding them at each stage. and then enveloping the pdf passed on to the next stage, makes the uncertainties clear.

Cho2. On page 9-1, the authors state that "hiethodologically, the assessment involved only a slight scenario depcodence, principally on the permanence of the blockages preventing dl rect downward, through the lower core support plate, relocation", and that thus the assessment is of Grade B, in the ROAAh! scale.

! think the scenario dependence is more than slight, so the asset,sment may be more of Grade C than Grade B in the ROAAh! scale.

Disagree. Scenario dependence concerns uncertainties and a complex, long evolution.

Here we have a well-defined behavior and a robust assessment for it.

Cor1. COhthiENTS and QUESTIONS for DOE Report

1) The authors do a good job in giving a context for their work. However, I am not sure if this analysis which is provided is a failure analysis for the APF^O reactor pressure vessel or a design analysis for the RPV. The former implies that it would be a 'best-estimate' analysis, while the latter m t account for factors of safety to assure survival. The authors need to clarify ans.

This is not a design effort. The approach is adequately explained in Appendix A.

Fle3. 2.2 Problem Definition and Overall Approach This section sets out the methodology to be used. Essentially, the now established ROA Ah! procedure is used in which the overall event is split up into well-defined physical processes that can be modelled, combined with intangible parameters (such as triggering time). The proposed sequence of events and the split between physical processes that can be quantifled using a validated model and those which must be treated in a pa: . metric manner seems correct to me. In particular, I G-32

believe that the Bow chart shown in Figure 2.3 gives a correct and well-Judged progression of events. Details of the modelling will be discussed later. However,

\}

it is important to emphasize that the identlBcation of a sound methodology is very Important and i believe that thc authors have done a good job et this Stage of making the process transparent.

Moo 2. 4. How does the ROAAM accommodate different causal relations, such as PM-ALPHA and ESPROSE.m, at diEerent stages in the methodology if they might be strongly coupled through common variables'? That is, the behavior of two systems alone may be altogether diEerent when they are coupled together (like two spring-mass systems). The probability distributions of the parameters involved may combine diEerently when the separate systems are strongly coupled, leading to dlEerent probability ranges on the variables which determine success or failure of a system or process.

If such dependencies exist they can easily be accounted for. No such dependency can be identified here. Breakup and '.riggering are conservatively bounded with respect to both premixing (PM-ALPA) and propagation (ESPROSE.m).

p V Moo 6. STRATEGY The severe accident management strategy addressed invoh s the retention of core material in the reactor vessel following a postulated severe accident in a reactor like the AP-600 design. Inability to cool the core leads to melting of core material by decay heat, and relocating it in stages to the reactor pressure vessel (RPV) lower plenum. Molten core debris, which may Bow to the bottom of the lower plenutm can melt through the RPV wall and undergo release to the containment.

However, Booding the cavity to submerge the RPV bottom head is expected to be a meana of arresting the downward relocation of molten core debris.

Even if down+vd relocation ol molten debris is arrested, there is the possibility that some mass of debris could drop into water present in the lower head region, causing a steam explosion and further damage. Part of the overall study shows that failure of the bottom head by exceeding its structuralintegrity is " physically unreasonable".

bl G-33 l

w . - - - _ - - _ _ _ _ _ _ _ _ - _ - _ _ _ _ - _ _ - - _ _ - - _ _ _ _ _ _ - _ - _ - _ - _ - _ - _ - _ - _ _ _ _ - - _ - _ _ _ _ _ _ _ - _ _

THE RISK ORIENTED ACCIDENT ANA hYSIS hiETHODOLOGY (ROAAM)

A primitive method of handling uncertainties in power systems came in the early 1960's (Moody F. J., " Probability Theory and Reactor Core Design," GE Re-port 1) GEAP 3819, US AEC Contract AT(Ob3)-361, January,1962). One of the greater concerns for a nuclear core during normal operation was reaching the

" burnout" condition, where a hot spot in the fuel could exceed design limits, and cause fuel damage. The fuel temperature could be expressed as a function of several variables and parameters (causal relations), each with its own degree of uncertainty. If one chose the most ;>essimistic limit of each variable and pa-rameter, the " burnout" limit could be exceeded. The most optimistic limits the

" burnout" lhnit would not be exceeded. It was suggested that probability meth-ods could be apphed to ghr a reasonable assessment of the likelihood of exceeding the " burnout" limit. Data from power plant operating logs was gathered to ob-tain prabability distributions for certain variables and parameters. Wherever data was not availabic, " expert opinion" was rolicited. The results were then combined by the method proposed in an ASME paper (Kline, S. J., and McClin-tock, F. A., " Describing Uncert .tinties in Single %nple Experiments," Mechanical Encineerine, January,1957), which resulted in the expected mean and standard deviation for the hot spot temperature. Comparison with the established design limit showed that it was " physically unreasonable" to expect " burnout" in most cases.

The ROAAM is an extenshr, operational methodology which is more refined than any of its primitive predecessors. It has the capacity for incorporating causal relations (describing equations relati: g the variables and parameters), based on well understood physics for the applicable phenomena, with specilled parameter uncertainties, scene'o bifurcations, and even a diversity of expert opinion. The process leads to a rationally-based prediction of those properties which determine the success or failure of a system or v > cess.

The structure of ROAAM embraces the current phenomenological state-of-the-art, built-!n acthution response of safety and control systems, man-machine in-terfaces, and procedural understanding. As new information becomes available, the ROAAM can accommodate it. Where expert opinions may be diverse, the ROAAM provides a means of focusing further research to narrow the disagree-ments. That is, when experts strongly disagree on the range of a parametes, G-34

I a the ROAAM can be employed as a tool to display the sensitivity, showing if the

) parameter dominates the outcome, or is only a minor percentage eEect on the overall result.

One question about use of the ROAAM imolves the causal relations for vari-ous phenomena. If the parameters in e causal relation are independent, their probabilities can be combined in a certain way to obtain the expected mean and standard deviations of that function. If the parameters are not independent, the combinatlon is more complicated. The question im'oh'es how the ROAAM accom-modates the possil>ility that some parameters appearing in more than one causal relation may not be independent. How would results from ROAAM compare with one deterministic mega-computation where all the parameters are treated by something like a monte-cerlo process to obtain the distribution of variables which determine success or failure of a system?

One could always hard -wire all the mod els in a ROAAM analysis to one mega-computation.

There would be no advantage (any dependencies can be handled just as easily), and there could be some important disadvantages; for example, in detern:ining the bounding con-ditions for rate of breakup and trigger time. More importantly, such a mega-computation G would be less scrutable, much less transparent, and much reduced in degrees of freedom U practically explorable.

Moo 7. ROAAM APPLICATION I have seen the ROAAM work in two separate campaigns to close severe acci-dent issues, namely the direct containment heating (DCH) issue for one series of PWR's, and the Mark Iliner melt issue for one class of BWR containment. It is appropriate that this methodology should be applied to reach a conclusion ca the in-vessel retention severe accident management concept.

Application to in-vessel retention embraces possible scenarios, melt conditions, coolant states, structural properti3s, debt 3 mixing with water, triggering, explo-sion wave dynamics, and lower head fragility. Parameter ranges are associated with the amount of participating substances, the timing of events, event paths, and state properties of various subsystems. Several analytical tools, based on physical models, provide the causal relations employed, namely PM-ALPHA for emvloping the eEcct of melt breakup in water, ESPROSE.m for enveloping the ef-fects of fragmentation and mictointeractions on steam explosions, and ABAQUS

!D U G 35

v.5.5 for erweloping the louvr head failure criteria. The computer programs used for causal relations to emelope important variables have be compared with other analyses and experimental data to a level where their predictive capability of the tested parameters does not introduce uncertainties which are significant enough to consider.

The following comments are offered to help substantiate my conclusion that in-vessel retention has been shown to be an elfective severe accident management concept for systems like the AP-600.

• 6-

• Moo 15. THE NEXT STEP I understand that a number of experts are providing reviews of the documents provided. Some may believe (as I do) that even without a complete understand-ing of all the phenomena, the remaining uncertainties, processed by the ROAAh!,

still pctmit a strong statement about failure likelihood being " physically unrea-sonable." Some experts may feel that the uncertainty of a given parameter should be broader. This is a simple exercise in ROAAhi a hich would then provide out-put with a range that accommodates the particular variable uncertainty. Other experts may wish to change the causal relations to reflect various " bottom up" or fine structure effects. This is always a possibility, but may be unnecessary, since the causal relations are based on macroscopic formulations of basic principles. If it were recommended that nonequilibrium models be employed for causal rela-tions, we would be farther behind than using ROAAh!in its present structure,

%cause nonequilibrium models would have to be verilled by experiments.

When strong disagreements have resulted in physical modeling, small uorking groups have been formed to reach agreement on acceptable formulation, with appropriate modifications in ROAAhi.

Finally, it - possible that some would disagree with the ROAAhi structure itself, suggesting that it skews results, or simply blurs our ignorance of phenomena.

I would argue strongly that the ROAAh! blends (not blurs) uncertainties (not ignorance) in a way that makes it possible to reach conclusions with a known level of confidence.

G-36 l

f Structural, Fragility a b,

Dan 3. There is no esthnate of the conservatisms induced by ignoring Buld-structure interactioni; assuming plastic Bow with no strain hardening, and using a much lower yield stress (330 MPA) compared to the measured yield stress (450 MP.I.). Such estimates would be helpful. ,

The scaling of the fragility with the yield stress of the RPV wall is now discussed in an addendum to Chapter 3. The other two aspects are harder to estimate, and we did not manage to include them in our priorities list for this response.

Dan 5. I would think the result that the greater the 'md localization, the smaner the effective impulse, has practical limits. For a periectly plastic mate-rial, a delta-function load (point source) will always panetrate conti. - ly. The introduction of bending moments to spread the load, of cource, represents the real situation. However, how far does the use of the etfective impulse go for failure criterion with a very concentrated load?

Both in terms of amplitudes and areas we have well covered the conditions of interest (Q here. But this is an interesting question, so we carried out additional calculations with V still more concentrated loads. The results are shown in the addendum to Chapter 3, and they indicate that the scaling effected by Eq. (3.10) is still appropriate, ahhough it becomes increasingly more conservative for load application areas with do/D, < zzzz. -

Dan 6. The net result, from the axisymmetric and non axisymme'tric calcula-tions, that the details of the loading pattern are not particularly important is, in itself, very important. This result uvuld seem to hold not only for the AP-600 reactor, but for all reactors, and greatly reduces the probability oilocal failure.

• * * * * * * * * * * * * * * + * * * * * *

• Ber4. Chapter 3: Structural failure criteria in this chapter, it is stated that < the time duratien of the loads ofinterest hereis less than the structural frequency >, so it is expected that < peak strain would be basically independ of the details of the pressure pulse shape >. However, nothing is said about the estimated value of the natural frequency of the R.P.V.

which seems to me to be of the order of magnitude of some msec so not so far from the load duration.

. p.,

f i V G-37 l

1 Howeser, all the analysis is made with the analytical rolution of Dufley and hiltchell which assumes < short pressure pulse > and allows to evaluate the plastic equivalent strain with incorporation of strain rate effect by formula (3.2).

But the comparison of the analytical results to ABAQUS calculations shows that the analytical relation gives conservative results for the plastic strain evaluation (fig. 3.2).

No, all the analysis is not made with the analytical solution; it is made with ABAQUS.

The analytical solution is used to obtain insights helpful to the task of coming up with the lower head fragility.

Ber5. Another mitigating factor is imutigated: the elfect ofload localization which shows that for a gh en impulse, the equhalent plastic strain is sma .et when the loading is smaller. Use of thev results is then made by assuming that a fraction f of the impulse is ased for bending energy so that only e.n KefEccthe impulsc> is applied for the evaluation of the equh'alent plastic strain. We are told that f is a material and geometric < constant > but I have not found any indication ofits evaluation i.e how the results shown on Eg. 3.8 and fig. 3.9 are obtained. As Rg. 3.9 is used to evaluate the cliect ofloads calculated in Chaptc.

6, I think that it should be a little more explained.

The s is chosen so as to fit the " data" (i.e., the results of ABAQUS simulations). The value O

is 0.05. This was added on Figure 3.8.

Ber6. It also seems to me that the localized loadings are applied on the axis of the hemisphere (see tabic 3.2). Does the fact that these localized loedings will occur on the side of the hemVphere with singularity where the sphere is hnked to the cylinder will modify the conclusions we can draw from ng. 3.9.

TI,e cylinder juncture to the lower head is quite a bit higher than where the water level is, and the further below location where the premixture develops. Moreover, in Chapter 6, we have full vessel simulations (see Eq. (6.2)).

But1. INTRODUCTION AND

SUMMARY

The comments contalned in this review are restricted only to a review o[Section 3 of the subject report. The report's authors have done a good job of scoping the possibilill3s of falling the lower vessel head under the assumed loading conditions.

Well established analytical cyproximations were used to establish the vahdity of G-38 h

_7 the Rnite element model that was developed to study local failure of the head.

A more detailed model needs to be developed to include transveise shear effects and to simulate it.Ilute of damaged elements during the course of the calculation.

. This lack 6f simulating progic sive failure is the weakest point of the analysis.

Appropriate simulation ofprogressive failure has to bv included in order to obtain '

defensible resultt that can be included in probab:listic cvaluations.

SPECIFIC COMMENTS

- Finite Element Modeh Use of the sheD elements in ABAQUS is acceptable for determining the distribu-tion through the thickness oN) components except for transverse shear. In the ABAQ US thick shell elements transverse shear is approximated by constant shear through the section. This is not Judged to be adequate for evaluating the possibil- -

ity of a shear type of failure. A better method for getting good approximations for all of the strain components would be to use several continuum elements through the thickness rather than the thick she'l element. Use of many more elements wo sid make the runs longer, but use of the explicit version of ABAQUS would help in this regard (see below). In addition, the use of an ansymmetric Bnite D- element model would af!ord the opportunity to use a much more dense mesh in the analysis with run times that are stui relatively short. The structure and all of the loads that were considered are axisymmetric.

In fact, our main interest is for non axisymmetric situations, and many of our calculations were not axisymmetric. However, refined grid results were obtained and shown in the addendum to Chapter 3. Transverse shea: effects were evaluated, as described in the addendum to Chapter 3, and found to be unimportant. Progressive failure could not be simulated by the code available to us, and in light of the margins available, and comment But14 below, we did not pursue this point for now. However,it should also be noted that the fragility already rises rather sharply, so the potential marginal error due to this effect cannot be very large (would make the rise steeper approaching more closely the step-like -

behavior shown for a uniform load).

But2. The mesh should be considerably more dense in order to resolve Bne details in the strain distribution, especially those details relating to strains other

. than in-plane strains. Referring to Figures 3.5a-c, even the in-plane strains vary from their maximum levels to just half that level over just one or two elements.

.?

? -

G-39 ,

l

\ Although not stated in the report, I assume that the implicit version of the l ABAQUS code was used for these calculations. The implicit version is always stable but may not always be com'ergent. There is no indication in the report at to whether the time step was varied to ensure a comergent solution. A better alternative may be to use the explicit version of ABAQUS for the short transient solutions that are required for the types of loads being considered here. The expilcit version of ABAQUS also offers the opportunity to use a failure model that would give more realistic failure predictions (see below).

We used the implicit version, and considered convergence - see addendum to Chapter 3.

This is now noted in the report. The explicit version of ABAQUS is not available to us.

However, we have done now numerous calculations with refined grids and verified the convergence of the solutions. The results are presented in the addendum to Chapter 3 and they further verify our previous results.

But3. The statement that the time duration of the loads is less than the natural frequency of the head may not be correct. A handbook solution of the frequency of a full sphere with the same dimensions as the hemispherical head ghts a natural period of 1.5 ms, very near to the 2 ms pulse duration used in this study. It is no wonder that, as rtated, the impulse time is "non-negligible."

O Our point was the same (i.e., that it is not much greater), and for the same effect. To avoid confusion, we changed the "less" to "similar." Also, the statement is true for the actual loads in Chapter 7. Actually, as shown in Chapter 6, for the most energetic cases the main portion of the impulse is delivered in ~0.3 ms.

Bute. Load-Strain Behavior:

Use oi the Eodner and Symonds approximation for the dynamic yield stress is a reasonable approximation. However, use of the values assumed for the constants D and p should be justliied more thoroughly. The values used here are for mild steels, and may not be appropriate for the alloy steel that is used in the pres-sure vessel being evaluate 1. Obtaining a good approximation for this relation is particuh:iy important because the maximum strain is very dependent upon 1;. I used an axisymmetric model with 15 continuum elements through the thickness to replicate some of the calculations in the report. The results showed that the maximum strain urnt from 0.52 to 0.16 with addition of the rate model for the G 40

=. - . -. . .- -

yield stress. - Considering the magnitule of this diEerence, one should certainly p(). be very carefulin the selection of the rate parameters.

That the effect is strong was discussed in the report, as was our choice of the lower yield stress also (300 as compared to the as tested 450 MPa value), as a contingency in lieu of directly applicable data. In the addendum to Chapter 3, we show results using 450 MPa, with and without rate dcoendence.

But5. Dexter and Chan (1990) address the etlects of strain rate and temper-ature on A533B steels. This alloy is close to A508 steel and may provide some useful information in developing an appropriate dynamic yield stress model S

These data suggest that the rate dependence is not as strong as in our original calculations.

All we can do is bound the behavior, as described in the response to item 4 above.

But6. Failure Criteria and Fragility:

This is probably the most diBicult aspect of modeling the response of the vessel head. 'lhe failure criterion that is used in the report is probably realistic and conservathe except for one important aspect. The model, as reported here does p not remove the load carrying capability of elements that have exceeded the failure Q) criterion. Maintaining the load carrying capacity of damaged elements can give signiScant over-estimates of the capacity of the structure. I used the explicit Bnite element model mentioned above to look at this aspect of the problem and found that, depending on the parameters used for the ABAQUS failure model, the head could fah for the loads t&t are reported. I strongly suggest using some sort of failure criterion embedded in the computational model for future calculations.

First, we should clarify that the statement "...the head could fall for the loads tnat are reported" refers to the loads used to derive our fragility here, not to the explosion loads derived in Chapter 6 (see also But14). Unfortunately, our code did not allow the removal of failed elements (see also response to Butl).

L Buti. The subject report briery mentions the eEect of s'ress anisotropy on the failure strain. This is an important issue and needs to be more fully evaluated.

The work referenced in the report by Pao and Gilst was performed on Charpy bars (roughly unlaxial strain) and by Shockey et al. was performed in pure shear (no !;ydrostatic component). Therefore these data dcn't address the important etlects coming from multi-dimensional stress Belds. Data summarized by Ju and.

O' U G-41

Butler (1984) show that A533 alloy steel when in equal biaxial tension falls at an equhalent strain equal to about one third the strain for uniaxial tension. Equal blaxial tension is the stress state at the " pole" of the lower head where failure would Brst be expected. The alloy content of A533 steel is similar to that of the A508 steel considered in the subject report. hiirza, Barton, and Church (1996) reported the effect of the stress Beld on failure strain and its effects in transitioning from ductile to brittle failure characteristics. Johnson and Cook (1985) also discuss the effects of the stress Beld on fracture of ductile metals.

Other references that may be of help include Jones and Shen (1993) and Ferron and Zeghloul (1993).

This again refers to refinement of the fragility. which in light of comment But14 did not appear to be of an urgent nature. It will be addressed in the final report.

B ut8. As previously mentioned, the herd would have ta be modeled with continuum elements to accurately predict transverse shear strains. In addition.

a failure criterion for transverse shear needs to be established. It is unlikely that the failure criteria discussed in the above references are adequate. They may however give some guidance in establishing the appropriate criteria. It is possible that when the loading conditions are investiguted note closely, the load cases that lead to the highest shear load (such as case 1+) can be eliminated obviating the need for this criterion.

Transverse shear results are now available (see addendum to Chapter 3). The results indicate that the shell element model is conservative.

But9. hiiscellaneous:

1) The use of the higher yield stress 450 hi% is justined based on actual data from Server and Oldneld (1978) where the average yield stress is approximately 440 h1Pa for A508 steel (very close to the Japan Steel Works Ltd, value of 450 hfPa). This is one parameter with ample data to support the use of the actual, as tested value.

This comment bringing in the Senser-Oldfield data is very helpful. The additional margin gained by using 450 MPa,instead of 330 MPa,is discussed in the addendum to Chapter 3.

But10. 2) The statement is made that A533B steel has a carbon content of 0.19 vs 0.16% for A508. Information from Server and OldReid (1978) and the G-42

i ASME Code show that A533 has a carbon content of 0.25% maximum and A508, m) Class 3 has the same upper limit for carbon content. Actual analyses show carbon content from 0.21 to 0.25% for both steels.

Dut11. 3) Chapter 3 in the subject report does not mention radiation em-brittlement effects. I( they can be dismissed, the reasons should be given.

4) For SA508 the transition from ductile to brittle behavior starts at about room temperature. The report should give the approximate material tem-perttures during the postulated event to show that it is well above room temper-ature.

The end-of-life RTNDT for the AP600 steel forging at the beltline region is specified as 23 'F.

The lower head, less irradiated, would be even better than that. At the time of interest, the lower head would be between 50 and 100 *C. This information is now included in the report.

But12. 5) The presence of flaws is not addressed. I assume that in service inspection will have identHied any that are significant in affecting ductile fracture.

O b Yes.

But14 Letter dated January 8,1997 to L.W. Deitrich.

The purpose of this letter is to clarify the eppliability of comments I made in the attachment to my previous letter to you dated December 1996, regarding review of the report entitled " Lower Head Integrity Under In-Vessel Steam Explosion Loads," by T.G. Theofanous, et al.

The comments made on that attachment were limited to Chapter 3 of the subject report and, consequently, affect the fragility curves developed in that chapter The fragility curves are subsequently referred to in reaching conclusions in Chapters 6 and 7 of the report. It is important to make clear, however, that the fragility ,

curves in question do not have major etfect on the conclusions reached in those chapters. The loads developed in Chapter 6 and applied in Chapter i are low enough that the vessel response is definitely below the lowest probability level used in defining the curves (10~ ).

/'~T

( 13 -4 3 1

1 should also point out that I concur that the probability levels used in devel-oping the fragility curves are conservative. The association of these levels w.'th strain magnitudes through the vessel wall are acceptable. However, the calcu-lated strain levels used ta develop the detailed curves above the 10~3 level may not be conservative. If the curves are ever used for evaluating higher loads; they should be teevaluated based on the review comments that I previously submitted.

Ernplementation of the information in these comments will affect the shape of the curves and could sh![t them toward lower levels of impulse load (to the left in figure 3.11 of the subject report).

IfI can be of further help please do not hesitate to contact me.

Cor2. 2) In section 3 the authors denne the failure criteria and the fragility cune for the reactor pressure vessel. If I understand the approach a strain-failure limit is used and the associated analysis suggests a RPV lower wall failure probability ranging from 0 to 1 for a spectrum ofloading patterns with impulses betwe=n 200 - 400 kPa-sec. Currently, I wonder how this failure envelope compares to that of previous LWR plants analyzed for av in-vessel steam explosion; e.g.,

the ZIP study in the early 1980s by Sandia and Los Alamos Nat'l Labs?

What we could learn from previous works we mentioned in the Introduction. In those O

days the pressure wave dynamics were not available to the structural analysts.

Ele 4. 2.3 Structural Failure Criteria This.section deals with quantincation of the likelihood of vessel failure for a transient, localized load. The material is presented in a clear manner and there is a step by-step progression from an axisymmetric model to the examination of localized loads. The analysis presented in equation (3.10) and Figure 3.8 provides a neat means of determining the cfTect oflocalized loading and the performance of equation (3.10) in correlating the data is impressive. Also I believe that the failure criteria given in Table 3.3 are sensible and fit the presented database.

This chapter is important in that it sets up the basis for the determination of

. whether a particular explosion loading will or will not fail the lower head. There would appear to be signiRcant conservatism in the analysis, as noted on page G-44

I 3-1 and from Figure 3.8 at the high impulse end, and therefore it provides the (Ji ~

required function for this study.

Moo 13. LOWER HEAD RESPONSE Dynamh respo.1su of the lower head is based on well established physics of shells, modeled by the ABAQUS program. Mechanical failure of a shell depends not only on the magnitude of an applied load, but also on the frequency content. It is stated that the shock pressure loads uhich lie in the steam explosion emelope have a short period relative to the structural response, so that the peak strain would be essentially independent of the pressure pulse time profile.

The report has provided some " screening fragility" curves which would be used to determine ilpredicted steam exolosion loads were of such a character that the failure criteria envelope and fragility cune need to be further blended to '

provide a failure likelihood. It was concluded from the range of pressure loads and the lower head fragility curve, that for all relevant severe accident scenarios, melt conditions, and timing of release from the core region, with ensuing mixing and explosion wave dynamics, steam explosion induced lower head failure in an O AP600 like reactor is " physically unreasonable."

Sch35. IV.1 Chapter 3. Structural Failure Criteria This is an important chapter, since it establishes the fragility curve, giving the probability of the lower head failure for dynamic-loads ofincreasing magnitudes.

The impulse loading, cfinterest, is in the range of100 to 300 kilo Pascal-seconds.

The authors have employed a commercial structural-analysis code, whose results they have compared with a simple analytical solution. ABAQUS is a 3-D linite element code, able to model the hemisphericallower head and the dynamic load-Ings imposed. The code provides the strain as a function of time for the assumed loading. These calculated results are, then, cometted to a fragility curve, as-suming probabilities oflower head failure, when strains of greater than 11 % are reached over certain fractions of the lower head u.,J1 thickness.

The ABAQUS calculations are performed for various loading patterns on the lower head. The non-uniformity ofloading was found to decrease the stra!n for a specific impulse. The colour pictures provide very nice strain morphologies.

.nq x ,' G-45 l

l This chapter provides clear and transparent results. Tbc ABAQUS results are conBrmed against a simple model for uniform loading. The fragility curvt nakes good sense.

Sch30. I am a bit concerned about the very local non-homogeneous loadings of the type predicted, later, in the report. Perhaps, a few ABAQUS calculations could be performed to establish the fragility curve for such a local-loading pattern.

We do not understand this comment. Most if not all of the deve:n ment in this chapter is for " local non homogeneous loadings of the type predicted later in the report."

Shel. The analysis of head failure sets up a model of the lower head using ABAQUS (a well established Bnite element code) to relate the stress pulse from steam explosion to local strain. The vessel material (ferritic SA508 steel) will undergo large amounts ofstrain (elongations of 50 to 100%) before fractute occurs.

Whether or not the vessel undergoes any plastic strain depends on the yield stress af the metal and the impulse from the steam explosion.

Sbc2. For reason never explained or discussed, the authors chose 330 b'Pa for the yield stress of the vessel. They state that the corservathe ' Code Ahowable'is 345 hlPa and the actual value (found in a conventional tensile test) is 450 hiPa.

The choice 01330 MPa introduces a large conservatism (safety margin) since a best estimate should use 450 MPa.

The real reason for using 330 rather than the actual 450 MPa value for yield stress is that we could not find the actual value until much of the work had been done with 330 MPa.

The additional margin due to this is now dbcussed in the addendum to Chapter 3. On the other hand, this margin could be used to compensate for the strain rate effect, if the latter is doubted (see But4).

She3. The impulse applied to the steel in the lower head would have a rise time cf a few milliseconds. When ferritic stecis are loaded this quickly their yield stress is substantially greater than that observed in a normal tensile test. The authors quote references that show the yield W.ress at this strain rate is about 40%

greater than that found in a tensile test. They take full credit for tids strain-rat:

increment, which is justined and appropriate.

G-46

_ . _ _ _ _ _ _ . .._ _._ _._ ___-..___.__m-_

2 A

in summary, the analysis of head failure seems to be competently and conarva.

. thtly done, and the conclusions drawn are appropriate. I have also looked at the

discussion ofloads and loading. I am less of a specialist in this area, out it also . ]

e;

. 'seems to be well done.

^

. * :s :n s . * -s

• s* * * .* *

• s. s - s * *

• s: , s-She4. Though no mention of radiation eHects is made in the reports, the

- ' analysis should be made for the vessel at end o[]Ife (40 years?). The temperatc'e p of the head during the accident considered would be less than 212 F. This is l beneath the RNDT for the beltline of some of the vessels now la service, i.e. such material might well behave in a brittle menner during an accident of the type considered here. I considered this, but feel such radiation eHects are not germane -

la she case of tha AP600 for at least two reasons-

, ' 1 ) The fast neutron and hard gamma flux in the lower head will be at least a couple of orders of magnitude less than that in the beltline region of the vessel, so radiation eHects should be negligible.

2) The steel to be used in the AP600 vessel should be appreciably lower in the

< elements than have lead to radiation embrittlernent (copper, and phosphorous) in the older vessels now of concern in plaats in the U.S.A.

i- With this in mind, I believe there is every reason to believe that the material

-lu the lower head would behave in a ductile manner and that the analysis given c in the report is appicpriate for (wo t]d apply to) a vessel in the AP600 after 40 years of service.

The end-of-life RTelDT values for AP600 steel forging at the teltline region is specified as 23 *E The lower head, less irradiated, would be better still. At the time of interest, the -

lower head would be between 50 and 100 'C.

Tur8. Chapter 3: Structural Failure Criteria

1. In principle, the loading may have components both shorter and longer than the nat ural timescale of the vessel. Indeed the constrained expansions considered in the early studies do have a longer timescale. One needs to refer ahead to

the results of the propagation modelling to justify the assumption through early

- venting of;he explosion region.

G-47 af ,

. -ty = ~, _ r- ,' _b_ -+ _ - _#_ _ -

A big part of the arbument is due to the voided and highlylocalized nature of the premix-tures. g Tur9. 2. The boundary conditions on the ABAQUS model are not speclSed

- from later examples they appear to be symmetry conditions at the equator of a sphere. As potential explosions may occur near the join of the lower head to the cylindrical section, it is not clear that this provides a good choice (apart from validating the simpler model - which could have been done in 1.D).

The explosion loading occurs well below the hemispheric / cylinder juncture, so our choice is reasonable, and an economical approach to develop the necessary understanding for localized loads, in Chapter 7, we show full-vessel simulations also.

Tur10. 3. From a non-expert viewpoint, the analysis presented in this Chap-ter appears a reasonable approach. However, I did note that Figure 3.9 was not consistent with Figure 3.4. To support the mitigathc factor for local loading, more highly localised ABAQUS calculations should have been performed. To avoid es o falling to zero for finite values ofI and da/D,, I suggest assuming that energy dissipation is proportiomil to the magnitude of the effective impulse.

Valid point More calculations were carried out. There was a problem with plotting figure 3.9. In the corrected figure the eaa goes to zero properly. See Figure 3.9 attacaed. The suggested idea gave about the same quality of representing the calculational results. The new, more localized ABAQUS results support well the generalization in Figure 3.9 (see addendum to Chapter 3) up to a point. For do/D, < 0.25, Figure 3.9 becomes increasingly more conservative.

You8. Following are minor points and typos.

1. In the discussion of the ABAQUS model of the lower head, it is referred to as a shell model; this is somewhat confusing at first, as shell mod =1s are normally thought of as meaning thin shells, i.e., no bending moments are supported. This could be clarined by calling the model a thick shell, for instance.

We have mostly used thin shell elements. We ran a comparison for loading pattern 1+,

using thick element and found a slight decrease of plastic strain from 21.3 to 20.6%.

G-48

. - . - . ~ . . .. - - - - . . - - =

6 e e e e e e = = s g g

.!=0.05 MPa s f' ,

0.s -

..  !=0.35 MPa s .

0.6 -

w.

- 0.4 -

0.2 -

\ -

1 0

0 0.2 Os4 0.6 0.8 1 d,/D, Figure 3.9. The mitigative effect of localized loading as a function of tne impulse applied and the degree of localization.

O n

U: G-49

m. - . - _.:_- . _ _ _

Melt Release Scenarios

' O Ban 8. The simple models for blockage formation and blockage coolability, leading to non-availability of downwards relocation paths, and transition to a molten pool, are made credible because of the relathrly Bat radial power dis-tribution in the AP-600 design. This lumped approximation would have to be re-examined for other reactor designs.

Der 8. Chapter 4: Quanti & cation of melt relocation characteristics This is a very important task as most of the boundary conditions for premixing and explosion calculations are obtained from such an evaluation.

e The downward relocation path (arguments 2) is not envisaged: < we expect this path to be blocked by molten cladding and the blockage be robust >. This expert judgment is supported by the large heat sink associated with the large amount of < cold > materials in the lower part of the core. As it is said that due to be big stainless steel reRector, the Erst relocation will be delayed compared to what occurred in TIhi and that at this time, we will have a la:ge oxide pool, it is important to know if this molten pool will reach the region of the lower Ession gas plenum where the heat sink is not very large and where we can have a breakdown of the supporting material. However, in the paper, the blockage is said to occur in the region of the 7 cm < lower Zr plug and lowermost spacer grid >.

Some calculations are presented to show that the plugging time of this region by melt with negligible superheat is of the order of seconds. For this calculation I have some trouble with formula 4.2 where, as for me, X is not the same as in the Carslaw and Jaegger text book but I did not try to perform the calculation. We can also make another remark: if we have some breaking down of the Ession gas plenum region, when the molten pool arrives we may have superheated molten material from this pool that with Bow in the lower blockage region for some tiir '.

before plugging It would be interesting to know what amount of molten materials can be transfered in the lower plenum through the holes in the core support plate before plugging of the passages in the blockage region. As for this plugging time

- uhlch is crucial to support argument 2 - it would be interesting to see more realistic calculations including the inBu;nce of the interface thermal resistance G-50 l

/

g gS* O k*& k IMAGE EVALUATION TEST TARGET (MT-3)

((*f 4[b 4 4 ,-

4)4y,;e,%+ ,y 4,,g;[33p'#

r, 1.0  !!2E sljEla 1.1 E.* E l.8

' ' ~ ~

1.25 1.4 1.6

< 150mm >

< 6" >

• $ t k k ,,D //

/* '4 op,,,

- ,; = , - $ 3 ,S ,a WEBSTER hE YORK 14580 (716) 265 1600

s h

4 s$*

k* IMAGE EVALUATION ///jg/p k4

/// 't, ,  % f/' TEST TARGET (MT-3) */s/, p#

'g

+ kks%y j*4 y,,,, ,,,, ,

+ 4 0

1.0 l#En BE

.. yll EE l'l E m HE I

I =l.8

~ ' ~ '

l.25 1.4 l 1.6 i__

< 150mm >

< 6" >

4 *'/4 4*' 'b

+%K%zzzzzA, PHOTOGRAPHIC SCIE CES CORPORATION

<Q.p.4

'W wessieR Ne YORK 14580

, (716) 265-1600

a*[*&

+ 48 IMAGE EVALUATION TEST TARGET (MT-3)

//[f jg 0($ '(,#

4 k//77 gy,>y!t[#

< g.,y+;g(@

1.0 'd 2 Bla 5!!!E i*i E L:

m HM l.8

!=

~

I.25 1.4 1.6 4 150mm >

< 6" >

1. 1. A ,, 7777 f4<

+ Af4 PHOTOGRAPHIC SCIENCES CORPORATION h k/A h 770 BASKET ROAD [1  ;

P.O. BOX 338 4<4(t,  !

WEBSTER, NEW YORK 14580 >

(716) 265-1600

q 9 &>  % o

+g>M;p', IMAGE EVALUATION d'b,

/j///'/

e ' Q3 p4 N9  % '/

\\\//,/ % TEST TARGET (MT-3) k ypM' /7 1.0 dmna yl[lias

~

I.I [* E l

'lr.8:

l.25 1.4 1.6 4 150mm >

< 6" >

1. ' % ,, <$'A4%

4;;&y%f777

1. ,# __==__ +y<>;;,4

_ rea = _

(716) 265-1600

Y

$ 49 0

%S IMAGE EVAL.UATION [jgf f/ $

k//77g@ sf/# TEST TARGET (MT-3) Qi j e$[k,p 4'

h'g

, $Y/ bhs '1

+ +

1.0 l#82 En

llIL4 1.1 [" E h

'~ '

l.25 1.4 1.6 4 150mm >

4 6" >

1. % if '/b t '

• ? 7k ,,3,,b//// _ _ = ,,,__ to,,W. ,O t

WEBSTER hE YORK 14580 (716) 265-1600

I k l

between the crust and the sohd wall that will slow down the freezing process and p)

( then increase the plugging time.

See addendum to Chapter 4.

Ber9. e As for the blockage coo! ability:

- the stable blockage thickness should be sensible to the radiation factor f,. which is set to 0.7 without any explanations

- the cooling of this blockrge is ensured for about 100 min which is the time re-quired to vaporize the water which Bils a!! the volume between F.,ttom of active core and bottom of core support plate. It is later estimated that meltthrough of the reRector by the molten oxidic pool will occur between 76 and 91 min accord-ing to the amount of oxidation (r* to 95 min in the calculation < without 's>

preheating). If un add the time require to melt through the core barrel, we get timing af the release of the rame order oimagnitude than the insurrance of block-age coolability. As all these calculations are order of magnitude ones, I think that argument 2 (no downward relocation path) raay be questioned.

All relevant factors were evaluated with some conservative bias, and these are not order n

Q of magnitude calculations. Rather, we would characterize theim as basic-principles based conservative estimates, clear and supportable in every respect. So, to dispute the quanti-tative result, one would need to be specific about which input or aspect of the calculation is being questioned. In the report, we also emphasized that beyond the 100 min, the heat capacity of the core support plate provides in ther margins to failure-this is a key point.

In any case, some further elaboration on uncertainties is provided in an addendum to Chapter 4. In this addendum, an estimation of the additional margin, due to the thermal inertia of the core support plate,is also included.

Ber10. e biolten pool formation

- In the initial heat up calculation, are the reRector and core barrel in contact everywhere as it is shown Eg. 4.8 and 4.97 In that case the cooling effect will be owrestimated and the melt superheat underestimated.

No. The melt superheat is only controlled by the pool convection processes. The crust thickness adjusts itself only to the losses, and this is not important.

O) i kl G-51

Ser11. . - During the .ransient heat up calculation, what happens to the molten cladding and how the calculation with the eiTective thermal conductiv-ity is eventually modified?

The effective conductivity is not modified. The anoxidized cladding is assumed to drain.

Ber12.

• hiolten pool calculation Such a calculation is performed for the oxidic and metallic pool, and there is a crucial hypothesis which is the presence of a stable oxidic crust at the upper surface of the oxidic' pool. In the document, it is mentioned that it is assumed that the clad drains but is it fully true? Cannot we have some metals included in the moving down oxidic pool? What happen to the part of Assion products which are released at fuel mciting? Will they modify the molten pool behaviour for the stability of the upper crust and the evaluation of the differents Buxes?

The stability of the upper crust is not important here. If it is distorted and broken by some vapaization from below,it will form again right away. We know that the metallayer will

" cat thiough" the reflector rather quickly, and the exact timing of it does not make any difference. Certainly it makes no significant difference to the treatment of the oxidic pool either.

Ber13.

• hielt through and melt release calculations O

it is said that rapidly the metallic pool will melt through the reRector but it is assumed that the metal < will be gradually draining :p into the space between the Bats of reacctor and the core barrel. Cannot we have some kind of metallic jet impacting on the core barrel with some rapid meltthrough leading to a steam explosion between metals and water?

fW(NT

/ f l f mebike y/ \"*!

i -,a . -

e, x. x.

'*" J' Y nfIeu,e G 52 Q

No. Because of the highly erosive property of the metalit could not accumulate to signifi-cant depths to produce the stream needed to penetrate the " cold" core barrel. The process willbe more like a gradual overflow.

Berle.

• From the above analysis it is concluded that when the oxidic melts tbrough the reRector, there is no metal on it and that fauure oi the core barrel }

occurs soon after. First, it would be interesting to evaluate the time required for core barrel meltthrough (if there is an open space between the two of them). l But there is another problem if the space between the Bats of the reBector and the l co:e barrel is already filled with the meta? from the metaHic layer, how the cxidic pool can rapidly go tbrough the core barrel. This situation may be a promoter for .

downward relocation if this added metal may increase the time for meltthrough.

As noted on pA-23, the total capacity of the spaces between the reflector and the core barrel is ~10 tons of steel, and this corresponds to 50% of the reflector corresponding to the pool he!ght. The oxidic pool will fail the reflector and the core barrel well before such an extensive melting of the reflector; that is, before the spaces are completely filled with metals. If this were not the case, the failure of the core barrel would occur at one of the comers of the reflector, making the failure more localized and hence producing a more O 8e is ee r 1 r e e.* sere e ee me is i<ic *e e 8s aei r i - e><i 8 *we eere barrel as compared to the times necessary to lose lower blockage integrity.

Der 15.

• As for the location and size of the failure, most of the information is obtained from expert judgment and should be'furtherjustined:

- the failure < is expected :p to be local azimuthany and very near'the top of the oxidic pool. I would agree with that statement as even, h the calculation is 2D cylindrical once a flat wiu fail, the rapid relocation wiu impede failure on other Rats. But I would not be t'; M give any probability for 2 quasi simultacecus failure, or 3....

- for the size of the breach, it is said that 0.4 m < would appear geometrically a good upper bound on the Brst breach width and that a < 10 cm axial gap is believed to be conservative :>.

- there is no mention of the rapid increase of the size of the breach during the melt

. release as it has been observed in experiments. However, as only short duration premixing-triggering scenario are taken to be ofinterest, this enlargment uvuld bV G-53

l not be important. But, if we take into account steam explosion occuring when water is sloshing back after a Brst event, this has to be taken into account.

See response to previous question about the postulated " series" of steam explosions.

Bur 3. 3. Comments On Melting and Relocation Firstly, it is shown by estimath2g plugging times from a freezing model based on semi-innnite heat conduction that the plugging takes place in a range of seconds (for the lowest initial rod temper.1tures 0.6 s with Zr and 2.4 s with UO2 ). How-ever, the melt now is r..it taken into account in tbese considerations. This means, that the freez!ng zone may extend over quite a distance. E.g., ~0.6 m would result for 1 s with 0.6 m/s as a typical velocity from CORA experiments. iiigher niocities would result with thicker melt B1ms. Thus, the Snal blockage formation should require some more time and distance (need of additional melting and melt Row or compaction to a crust by remelting and relocation of upper parts of a partly blocked region). Further, this process may yield local incoherencies of the crust formation, i.e. also weaker regions, although the cold traps at the bottom give certainly a unifying trend. Thus, in order to further verify the statement of rapid blockage formation calculations with a core melt code would be desirable.

These could also yield a more detailed perspective on related (subsequent) im-portant questions, especially the heatup of the cold trap regions and the water level development.

To our knowledge there is no mechanistic code calculation out there that shows the pos-sibility of direct melt relocation through the bottom of the core. Even for BWRs, with' much more open geometry and a much larger metallic component in the core, there are significant doubts about such a direct relocation scenario. Given the results of our further evaluations reported in the addendum to Chapter 4, we do not think such calculations are necessary.

Bur 4. Secondly, assuming an existing blockage with an overlying melt pool, it is cluded whether a steady state with a stable crust below melting point (~2800 K for axidic or 2100 K for metallic material respectively) can exist. Metallic and ceramic crusts are considered t.lternatively, with a heat Bux from the molten 3

ceramic pool above of S 0.02 MW/m , a volumetric heating of ~0.5 MW/m in the ceramic crust and cooling from below via radiation. Here, it appears not clear to me why the fraction of fuci volume is only taken as ~30% (p. 4-6). This seems G 54

to be a value for intact structures. However, if the metallic parts are all relocated p

Q ,) during establishment of the ceramic crust, then this inay consist essentially of UO2/ZrO 2(~80/20 wt ratio). The local shape factor should also not be decishv due to the crust formation from upper material. Taking a value of decay heat of 300 W per kg it.el (p. 4-18) this would yield ~2 MW per m3 of the UO 2ZrO2 crust. j Existing and relocated fuel volumes and relocated metallic melt volumes were consistently j taken into account. The relocated material is taken at the radially local, axially averaged core power, while the fuel stubs that support the blockage is taken at the local peaking factor - but without releasing its volatile fission products as it is found in a cold, solid state. See additional results in the addendum to Chapter 4.

D Gr5. For the downward heat flux from the molten corium pool above, a maximum value of 0.02 MW/m3 (fully developed) is assumed. This is derived from Eq. (4.14) based on the Steinberner-Reinecke correlations for a res. angular geometry (typing error in (4.14): exponent 0.095 instead of 0.049). However, this correlation is only confirmed for Ra' < 5 10*. For the conditions considered here, I obtain a value of Ra' ~ 100 , assuming H = 1.8 m and Q = 2 MW/m .

(3 The correlation is also derived for non-isothermal lateral boundary couu';;!ans,

\v )

in contrast to the present assumptions. The influence of the lateral boundary conditions appears to be sitall, however. For a case with vertical cylinder and melting point temperature ut all boundaries THEKAR calculations [1] also yield a rather similar correlation (Nu, = 0.935 Ra'010), but only for Ra' numbers below 109.

The value of 0.02 MW/m 2is absolutely insignificant in comparisen to the heat flux gener-ated due to the heating in the blockage itself (see also the addendum to Chapter 4), which is in the 0.1 to 0.2 MW/m 2range. Thus the 0.02 value could be varied by +100% or more without changing the results - such uncertainty is not expected in the kinds of questions asked here.

DurG. Bat, the main quest:on is to me whether -in view of the above argu-ments and at least some lateral cooling potential- the assumption of a rectangular pool geometry is a too strong idealization and other geometries closer to hemi-spherical shapes can really be excluded. Such geometrical variations would yield significant variations in the heat transfer to the lower boundary. The influence of

)

w/ o-55

l the lateral boundary would increase (natural comection inBuence versus stable stratification). For a hemisphere (certainly an extreme under the ghrn flat radial power shape) even a mean heat flux of 1.05 blW/m3 would result at the curved lower boundary according to (5.28) from the IVR report and at the center still 0.1 hfW/m 3according to (5.30a) from IVR. With a thermalload from the melt pool of 0.1 hlW/m3and Q = 2hiW/m in3 the crust only 3 cm of stable ceramic crust would result from equations (4.3) and (4.4), with 0.02 hlW/m' about 5 cm.

The scenario described in the report does not lead directly to a rectangular pool. It is rather the final state, evolving through intermediate shapes such as that in TMI. Exactly because the fluxes on the vertical and slanted boundaries are greater, and includind the non-coolable nature of these intermediate blockages there will be a gradual expansion of the pool, all the way to the reflector radially, and eventually to the final cold trap at the core bottom, where the blockage can be stabilized by radiation cooling. Neither the 0.1 8

MW/m nor 2 the 2 MW/m values are appropriate. Nevertheless, even a 3 cm blockage at the lowermost extremity c,f the core would be more than sufficient to contain the melt pool.

Dur7. Further, the first blockage should be metallic and a ceiamic crust should settle above. Then, the combined system of ceramic and metallic crusts should be considered. This yields a lower bottom temperature, thus lower ra-dlathe heat remom!. Therefore, the crusts should become even thinner. If, due to hcatup of the lower structures, the lower region of the metallic crust remelts and relocates, this yicids a further decrease of downwards heat removal from the ceramic crust region, thus inducing further remelting.

This sequence of events describes one part of the phenomena associated with the pool expansion phase (as discussed in the answer to the previous question). The fact remains that this expansion will be stopped radially by the reflector, t.nd axially by the cold traps at the lowermost ends of the bundles. There, the blockages will be coolable and we have shown them to be so both for metallic as well as oxidic compositions. Because of the tight geometry even a few centimeters of blockage would be adequate to support the molten pool.

. Bur 8. Finally, the downwards relocation path appears not yet as surely ex-cluded as stated. It is also to be mentioned here that local melt streams into the melt pool could simngly enhance the local heat transfer to the bottom crust as G-56 h

e i

shown in [)). Thus, together whh the uncertainties of the process of crust forma-() tion considered above and the smaller crust thicknesses of the above estimates, local inhomogeneities of the crust may become important and may induce local failure at the bottom.

This mechanism is not appropriate here. Here the pool forms from the middle and top with a downwards progression. By the time of interest, when the blockages have reached the lower extremities, there can be no supply of long-duration cold streams of melt. Also, the pool is rather deep (compared to the experiments mentioned) and would mix, and stop well before the cold plume could reach the bottom crusts.

Dur9. However, the basic idea that signiScant cooling potentialis provided from %e remaining water in the lower head and the massive core support plate is pomising. Perhaps, some further calculations related to the above objections could yield further support, But, the steady state consideration for the crust may not be suf5cient in general. Calculations on the time development of melting and crust development with available codes may be in ary for better conBrmation.

Following the whole melt progression process, as suggested here, is a very complicated matter, and subject to much greater uncertainty than the basic-principles approach taken Q in the report. Such calculations may be useful in providing another perspective, but we don't believe we could defend them as the basis of our case. We are aware that this reviewer may have such capability, and would like to see related results be added here for such further support.

BGr10. The main statement is that sidewards melt-through occum signin-cantly before possible dowmvards relocation, within the time of ~ 100 minutes during which etlecthe cooling from remaining water above the core support plate is available. The basic statement is that sidewards cooling is much less etlecthe than downwards cooling. The evaluations in the report take the sidewards bound-ary condition as adiabatic, i.e. no lateral heat remmal is assumed. This appears ta be a too strong restrictlon. On one hand, heat removal by the produced steam should be taken into account. On the other hand, heatup of the RPV wall bv radi-ation from the barrel and outside vessel cooling by Booding should be considered.

Taking a temperature ditference of ~ 500 K over the barrel and reRector and an outer barrel temperature of ~ 1000 K as ghen from the calculat:ons (Figs. 4.8 -

()

V G 57 l

l

, 4.12) nearly half of the heat Bux through barrel and reflector could be radiated to the RPV wall (if taken at saturation).

The radiative heat sink to the vessel wall and through it to the outside water was taken into account. Because of the low pressure, steam cooling (through the reflector holes) was found to be negligible.

Bur 11. The calcult.tions on core heatup and melting essentially yield the timing for melt pool formation (~ 42 -57 minutes from core uncovery to 20%)

to be related to the times for evaporation of water above the lower core support plate and for heatup of reflector and barrel. The further heatup of the ceramic pool and the overlying metallic melt layer resulting from reRector melting as well as the heatup and melting of reRector and t>arrelis calculated by means of equa-tions (4.10) - (4.15), of which (4.14) has been questioned above (questioning the assumption of rectangular pool shape). With the lateral heat Bux in the ceramic pool an additional time of 34-38 minutes is calculated for reRector melting. This is taken to verify lateral melt release at a time with still elfective cooling from below (water above lower core plate). But, it has to be remarked again that lateral heat removal is neglected. At me' ting temperature, at least half of the lateral heat flux could be radiated to the RPV wall (if this is not taken to be superheated sulficiently). Further, heatup and melting of the sidewards ceramic crust as well as of the barrel h not considered.

Both questions (on rectangular pool and lateral heat removal by radiation and conduction) have been responded to above.

Dur12. The considerations on the overlying metal layer resulting from the melting reRector seem not to yield important effects with respect to the final melt release. Although earlier melt through of the reflector can be expected in this tange, this only means that the reRector r.elt is essentially relocated into the gaps between the reflector and the barrel. But then, the refrozen material must melt again to get break-through.

Not so, because the axial path now is of large dimension, and the relocated metal goes to the bottom of these spaces and builds up from there.

Dur13. Certainly, freezing heats up the still solid reRector and barrel mate-rial. But, the material and energy redistribution by these processes may yield G-58

,I p some azimuthal homogenization: Thus, the assumption of a local azimuthal fall-0 .ure may not bejustined by the considerations on geometrical inhomogeneities in the report. In general, the assumptions on failure locations and size are problem-stic, nlthough the bounding assumptions appear to be reasonable. In rny view, the main objections could be, on one hand, those of above, questioning the exclu-

. sion of downwards failure, and, on the ocher hand,~ the exclusion of several failure locations within a certainly short time frame.

Note that once a relocation begins, by local meltthrough, the melt height drops and there is less opportunity of other melt breeakthroughs azimuthally. Rather, we think the path, once opened, will continue to enlarge and melt will be released basically from the same location. This, however, will have to be very gradual. Also, regarding time coherence, as we understand from the mixing explosion dynamics, has to be seen in the context of a few 100s of milliseconds, while we would have a hard time visualizing melting coherence even on a time scale of a few seconds Bur 14. The latter point indicates a further dencit: the further cou:se of melt release is not considered suBiciently. Even with an outBow rate of 400 kg/s

-(see below) the time of outBow of the whole corium melt pool would last some minutes. During this time, failure could occur at multiple locations, overlapping in time. Ebrthes, local melt /ctnlant interactions could yield additional and also larger failures. A question is whether failures of the bottom of the crust by such interactions can be excluded. On the other hand, the strong voiding with the resulting necessity of early triggering to get explosions restricts the possibilities for coinciding events. Enhanced evaporation of the water pool also acts in this direction.

Subsequent events are obviously impossible to predict in exact sequence, but the train of though explained here by the reviewer is similar to ours. That is, as melt relocation will continue the water pool will remain with a lot of voids, and the water in it will deplete

, rather quickly. We would take issue only with the statement that failure will occur at multiple locations. As explained above, it is the nature of the process, with the upper 30 to 50% of the core barrel thinned out by melting, once a relocation begins it would tend to remain focused on the same path, enlarging it downwards.

Bur 15. Scenarios of ex-vessel reBood are considered with respect to the time

^

of vessel Booding but not concerning the establishment af efTective lateral cooling G-59

- , - .,~n

l l

as discussed above. The considerstions on vessel flooding before orJust about the time of reRector melt-through consider only the cooling aspects and thermal ef-(ccts of focusing by thin inetal layers. But, embrittlement due to rapid quenching may favor failure of the pool surroundings at any location. Further, especially en-trapment explosions in the gaps may yleid such failures and thus more ext-nsive melt release.

Disagree. We consider all three types of reflood scenarios, and show that the two that are relevant (in terms of their thaing) to the melt progression process, actually could lead to melt arrest within the core barrel boundaries. Embrittlement cannot lead to failure under these high temperature conditions. Moreover, as shown in Figure 4.15,in the " fast" scenario, the core barrel and reflector would be cooled well before they could heat up by the melt, and in the " medium" scenario, one would require a totally singular coincidence between core barrel meltthrough and water level traversing the downcomer length.

Dur18. The exclusion may also be better based by considering additionally the cold trap properties of the lower spacer grid and the Zr plugs quantitatively for conditions after boil-offin this region and with melt relocation to this region before lateral melting of reRector and barrel occurs (or: improved considerations on the timing of the events, with respect to the above discussions).

The explicit consideration of blockages in the Zr plug region is now provided in the addendum to Chapter 4.

B Gr27. 6. Comments on Renood Scenarios As already remarked under 3. of my comments, the ex-vessel renood should also be taken into account with respect to the considerations on the cooling aspects determining the conclusions on the relocation path. In the context of vessel reBood also the possibility o(embrittlement and thermal stresses favouring failure should be considered. It should be shown that also under these conditions larger melt release and in this case contact with subcooled water is avoided or not threatening. Entrapment explosions, e.g. in the gaps between reflector and barrel, should also be addressed concerning a possible increase in failure and melt release.

There is no mechanism for entrapment explosions, and all other items noted here have been discussed already above.

G-60

,7 Cho4. Asrther, the explosion would likely expel some water from the lower

) plenum so that the lower core support plate may no longer be in contact with water (i.e., the ability to cool the core support plate would be lost).

As noted in the report, the heat sink of the core support plate is very significant. In the addendum to Chapter 4, this is quantified to a time margin of more than 30 minutes. Once the melt relocation begins, it will continue, vaporizing the rest of the water in the lower plenum, before failure of the blockages are possible. This was discussed in the report (see

p. 4 3).

Cho8. In all supporting calculations, the water was considered to be satu-rated with the primary system completely depressurized to 1 bar. Even in a large break LOCA, the containment back pressure uvuld remain in the range of 2-4 bars for a long period of time. It would appear that a system pressure somewhat higher than 1 bar (e.g., 3 bars) would have been more realistic for the supporting calculations.

This is not correct. In AP600, at the time of interest, containment pressures cannot exceed

~1.7 bar. In any case, to provide some perspectives on the effect of pressure and subcooling, we provide some new results in an addendum to Chapter 5.

'd Cor3. 3] In section 4 the authors' major point is that the core and vessel design is sufficiently different from past LWRs, such that the core melt behavior is quite ditferent. Two aspects are emphasized: first, the lower core support plate and the non-active fuel length above it [30cm] is large enough in size to delay the core melt progression downward; second, the core steel reflector in the radial direction is also thicker [over 15cm], also delaving and changing the detal s of radial core melt progression. In essence, the ' race' to the breach of the corium melt cruciole, which is formed during the meltdown, downward or radially outward is governed by these boundaries. The authors use a specific 3BE core melt accident sequence to illustrate this behavior. If one accepts this premise about a radically different core geometrical design, a few questions arise:

a) What is the sensitivity of meltdown timing to downward boil off of water?

More examples are needed.

None. The 100 minutes is calculated after the water has reached the bottom of the active fuel region (from then on it is heated only by radiation). This is conservative. Also con-servatively, we do not account for any refluxing of vapors condensed in the upper, cooled C/ G-61

parts of the primary system. Further perspectives on downward heat flux sensitivities are provided in the addendum to Chapter 4.

Cor4. b] is the core melt event timing essentially independent of accident sequence? No guidance is given here.

Yes. In this passive plant there is an essential " collapse" of sequences, as discussed in the report and in DOE /ID-10460.

Cor5. c) The corium exit Bowrate seems to be set by the 'rlp'in the rewor along the radlal edge of the core region at the very top of the pool. Is this a realistic estimate, since it is not much more than that one would calculate from adiabatic heatup and meltdown of the core; e.g., as evidenced at TMI2 ?

The authors suggest that 200 to 400 kg/sec " appears to be a reasonable range physically to bound the behavior"; but I wonder if we really know that much about this core melt failure progression in a radically new geometric design that this Bowrate is a ' reasonable bound'? MoreJustiScation is needed for one to ' buy' the argument.

The basis in quantifying this intangible factor has been explained. Some judgment is required here. It really has nothing to do with details of melt progression or the " radically new geometric design." It is a question of how much of the core barrel area can melt through coherently in a time frame of a few 100's of milliseconds.

Cor6. d] This last question really leads me to the key question of this whole analysis; i.e., the authors leave me with the impression 'that there is a good deal of certainty in the melt progression and I have s'gnlRcant trouble accepting this premise. Specincally, the whole analysis hinges on the fact that the melt crucible which forms during the melt progression has a structural integrity of enough certainty that it would release the melt radially through a pour area no larger than 0.02 to 0.04 sq. meters. This estimate also seems to be robust enough that it would be a " bound" even with coolant reRood into the core region and any possible disruptive events that may occur. I am very dubious about this and would need to see more analysis to accept this as a ' reasonable bound'. This melt failure location and pour rate is the key determinant in limiting subsequent FCI energetics.

G-62

8 Again, see response to the previous question. Also, the reflood scenarios in this passive 7

i plant are pretty well defined, as addressed in Chapters 4 and 8 already. It would be helpful to know what aspect of the analysis is considered doubtful and why.

Cor7. . 4] In section 5, the authors detail their multidimensional premixing analyses. As stated previously, the melt flowrate of 200-400kg/s seems to prede.

termine' the benign nature of the FCI energetics, but mixing is also part of the process. A few questions arise here:

a] Why has the elfcct of RPV pressure been neglected? Premixing will occur at elevated pressures not 1 bar [like 2-5 bars] and.'this will affect the mixing process.

Also, the rise in pressure locally during mixing will cause the pool to subcool and this has been neglected. Were calculations done to ' bound' these ellects?

For the AP600, the relevant pressure range is 1 to 1.7 bar. Calculations were run for 3 bar (see addendum to Chapter 5). Any subcooling due to rise in pressure locally, and its consequences, are automatically taken into account in PM-ALPHA.

Faut. As requested in your letter dated June 17,1996, the following com-ments are offered in the areas of hieltdown/ Relocation Phenomenology and Steam Explosion Loads, hieltdown/ Relocation Phenomenology - We agree completely that a downward relocation path of the melting core materia! through the core support structure (and resulting large fuel pour rates) is " physically unreasonable". Furthermore, the predicted relocation off to the side and from a fully developed melt pool leading to a molten fuel pour rate into the lower reactor vessel plenuni of about 200 kg/s, is consistent with the Three hiile Island Unit 2 Core Relocation as described by Epstein and Fauske (Nuclear Technology Vol. 89, p.1021-1035, December 1989). Fuel pour rates of this magnitude by themselves eliminate concerns relative to global vessel failures, even if an energetic steam explosion is

,>ostulated. As illustrated by Epstein and Fauske (1989) and Theofanous et al.

. (1996), such low fuel pour rates limit the fuel that can be found in transit within the lower plenum to values at least an order of magnitude less than that required for incipient lower head failure (3 to 5 tons). Quoting Epstein and Fauske (1989)

"A key aspect of the relocation is, then, that signincant quantines of corium melt were not mixed with water at one time. The slow melt relocation phenomenon is, perhaps, the most important piece ofinfortnation gained from Thil-2 studies n-Q G-63

~L ' .

, . . , , . - , . . . - - - , . ~ .

and should figure prominently in future assessments of steam-explosion-induced containment failure as well as lower reactor sessel plenum failure due to fuel debris overheating." This is clearly the case in the current assessment provided

)

by Theofanous et al.

We regret having failed to mention this important reference. We will introduce this ref-crence in the final report. Still, the inference to AP600 is not automatic, because of the reflector!

Fle5. 2.4 Quantification of the Melt Relocation Characteristics This section presents an analysis of the melt relocation characteristics. It is im-portant to note that the analysis does not use a system code but instead a number of highly speciBc models have been developed to address the pin'sical processes deemed to be important. This was the approach followed in the Sizewell B study and seem to me to be the correct way to proceed. Based on nn participation in the Sizewell B study I believe that the methodology used and the conclusions drawn are ectrect.

The melt Row rates and release conditions are consistent with those found in the Sizenvil B study. In particular, I believe that massive pours of many tonnes per second have been ruled out on the correct physical basis.

In the section on renooding the authors do not consider the possibility that a steam explosion may occur as the water renoods the molten pool. It is cotcred in a later section and it would perhaps be wise to have ghen a forward reference here.

Forward reference to Chapter 8 is added.

Jac4. 3.Tecimical evaluation 3.1 Melt relocation scenarios By the scenarios h is denned how the melt relocates into the lower plenum and this gha the rates at which corium is fed into the lower plenum. Therefore this is an importunt aspect that must be scrutinized during the review step of ROAAM. I am not really an expert in this Reid, my3clf, but 1 must raise the question whether it is really possible to exclude with sufficient certainty a downward relocation that could lead to much higher corium flow rates dep:nding on the number of holes G-64

, in the core support plate through which corium Bows into the lower plenum. hiy doubts in this respect come from the agreement of the experts in this Beld that the

()

late phase of core melt down, i.e. the melt relocation phase, is not well understood and imm the vinual absence of mechanistic models for growth and especially radial expansion of molten pools. The study that is under discussion here tries to bridge th!s gap using simple and clear estimates of conditions influenchig the thermal stability of a metallic crust. But in these estimates, e.g. no consideration is given to the possible formation of cutectics which might drastically reduce the melting temperature and thus crust stability. One might also speculate that some hot material could drop into the water remaining below the corium pool, thus decreasing the time untilit is evaporated and thus the time of crust stability. In the present study the evaporation time 'happent to bejust abont equal the time it would take to melt through the reflector and core barrel.' Of ourse, there is in addition the thermalinertia of the core su,) port plate. But as soon as its top falls dry, its surface temperature willincrease and thus reduce the etlect of radiative heat transfer.

Aside from control rod materials which would lead the relocation and thus be eliminated, all other eutectics possible are well covered by our metallic blockages. Since paths for Q relocation are not available, by huge margins, one is not free to speculato that "some hot material would drop into the water .. ." helping accelerate the evaporation. Fhtally, the reviewer's supposition that "as soon as its top falls dry, its surface temperature will increase and thus reduce the effect of radiative heat transfer" implying blockage failure is incorrect. As shown in the addendum to Chapter 4, this provides an additional margin of at least 50 minutes.

Jac5. Another possible uncertainty is the stability (leak-tightness) of a side-ways (radially) advancing crust. This process might induce transverse forces on the supporting stubs of fuel pins which these cannot withstand in their damaged condition. So the crust could fall and the oxydic melt could Bow freely towards the core support plate and possibly through it. (Table 4.1 Indicates that the ' cold trap' is not likely to stop Bowing oxydic corium.) Here one muy recall that pro-cesses of this nature occurred during the ThH 2 accident [l] although, in that case the whole melt pool was submerged. As witnessed by several tonnes of corium that solidined within the core support assembly, a large amount of corium has Bown down through about 4 peripheral fuel elements around core position R6.

g U G-65

Another downward relocation occurred at core position KS. The latter may haw been brought to a stop ahave the core support assembly, But we do not know how and by what margin.

The TMI did not have the zirconium pellets at the bottom of the core. Still, the melt was trapped above the core support plate, preventing relocation through the downwards path.

Contrary to the reviewer's intention TM1 actually fully supports our scenario. See also Faul.

May2. 2. Melt relocation characteristics Melt relocation characteristics are inBuenced by the heating up of the uncov-cred core, the transition to a molten pool, the cr.:!! ability or non-availability of downward relocation paths and several melt release conditions. The authors re y carefully analysed all processes, preceding or being involved in melt relocation, including blockage coolabilities and the resistance of the tcBector and the core barrel against melt through. The cunclusions, drawn from the calculations and physical considerations, are convincing. The two main conclusions, namely that the failure itself can be expected, that it will be local azimuthally and wsy near to the top of the oxidic pool and that the release will occur within a time period, which is within the coolabil-ity of the lower blockage, are presented in chapter 4 of the report (see page 4-25) and ght the good feeling, that the maximum amount of melt, which can interact with the water in the lower plenum, forming a steam explosion, is limited and by this also the energy release and the mechanical load onto the pressure vessel wall would be within a reasonable frame. So the limitation of the energy scenario, by carefully study-ing melt relocation characteristics, is a wry nmportant and very commendable contribution of this report to the state of att in steam explosion analysis.

A further, very important result in this chapter is, that "re-Bood scenarios" need no further consideration from a steam explosion standpoint (lower head integrity).

This conclusion should and could have consequences for future planning of ac-cident management activities for existing pressurised water reactors, also. It means, that any effort should be undertaken to add water again into the pressure G-66

~ - . . -- .. , - _- - - - _-. -. - - -. .-

I vessel after a beginning core degradation, because it would be of advantage for

&n preventing a further escalation of a severe accident.

Seh37. IV.2 Chapter 4. Melt Relocation Characteristics This chapter provides the faltlal conditions for the scenario of melt-water interac-tion in the lower head. The ch' apter, therefore, deals with the melt pool format.'on -

in the origin.d core boundaries and, later, relocation of the melt from the in-core location of the lower head. The quantitles needed are the rate of melt addition to the water in the lower head, theJet geometry (diameter, srlocity and location in the vessel), the melt composition and superheat and, Snally, the timing of this event relative to the other events in the core melt-pwgression process.

The authors, Brst point out the differences in the AP-600 core con &guration from that of the conventional PWR. The AP-600 has some features which are quite favourable in terms of the melt releases conditions. These are the massive 36 cm thick core support plate, the core reflector, the gap beturen the core barrel and the reRector on the Bat sides of the tvBector; and the long unheated section in the fuel elements at the bottom.

The authors have developed a credible scenario of melt pool formation, melt attack on the reRector and the core barrel. It is supported by emvloping models of appropriate s>mplexity, which provide physicalinsight and transparency. The authors are wise not to use one of the myriad codes, s hich provide user motivated results. The analysis is brilliant and quite comprehensive. The melt release conditions of 200 to 400 kg/sec should be bounding values. The melt superheat of 180 K also should be a good bounding value. The location of the release, near the top of the core in the vessel downcomer, may also be credible. The jet velocity of few meters /second also appears to be sound.1, however, would like -

the authors to consider the following cautionary points:

• * * *- * * * * * * * * * * * * + * * * * *

• Seh38.- (i) The timing of the melt reicase 76 to 91 minutes is much too close to the timing of m100 minutes for evaporation of wuter in the bottom 25 % of the com height by the radiative heat Bux imposed.

In addition, the re is a huge margin in the heat capacity of the massive core support plate.

See also addendum to Chapter 4.

A L) G-67

Sch39. (ii) The core plate is masshe but it is also loaded heavuy. If the core plate temperatures go beyond 700*C, the yield strength will deteriorate.

The core plate is fully supported by the core support structure. Also note (addendum to Chapter 4) that it really takes a long time for the lower part to heat up.

Sch40. (ill) The melt pool with m40 so 60 % unoxidized zirconium and some stainless steel, will probably form a primarily metallayer on the top. This layer is thin and will focus the heat Bux to the sides. Recent uvrk at RIT has evaluated the heat transfer from the metallayer to the vessel (which is of a thickness simuar to that of the reBector) with a tw~ dimensional codn, and found that the highest heat Bux is stb 1 at the corner of the oxide pool just below the metallic layer.

Thus, the failure could be below the .netallayer.

We do not agree with such a result, but we need to look at the RIT analysis mentioned.

Sch41. (iv) While, I agree with the authors that the Bat part of the reRector being closest to the core centre is most likely to be attacked Brst by the pool. The oxide pool however may not be axially symmetric and there may be azimuthal regions in the core, where fresh fuel and high power are dominant. Evaluaion of a possible attack on the non-flat parts of the reRector should be conside.ed.

Failure at the comer would produce a more localized release. Failure on the flat is conser-vative.

Sch42. (v) The draining and freezing of the metallic layer into the weu be-tween the Bat part of the reBector and the core barrel, without participation in any melt-water interaction, is very likely, but sounds too comvnient. Addition-ally, in the absence of water abow the core plate in the well, the thermalloading imposed by the superheated metallic melt on the core plate, or on the core barrel region directly above the core plate should be evaluated.

There is no water between the core barrel and the reflector at this time in such an accident.

As we have shown, the core plate still would be cooled by water.

Sch43. Summarizing, I believe, the authors estimates for the range of melt-release-characterist!:s is credible, however; additional evaluations may help to put

' these estimates on a more solid footing.

More evaluations can be found in the addendum to Chapter 4.

G-68

-. - . . ~ . -

, Tur5. : 4. ;I agree that the size of any breach is indeed a tough question.f I Q)3 '

consider it to be the key question unless the mixing / propagation analysis by itself ,

.can be shown to be sum: lent. I do not believe that this has been shown to be suMclent (as yet?). .

See specifics below-Tur12. 2. I do not see that the heat sink associated with the core suppart plate plays a sigmlicant role, as water provides the major heat sink. In the absence of water, melt passage through ti.e plate would depend on the diameter of the Bow channels. - Melt appears to have passed through relativelt small diameter holea in the presence of water at Thfi-2. If these holes are similar to that of the hole in a PWR lower core plau- (hen they probably oEer little resistance to melt flow. The lower core plate se .1 prevent large diameter pours penetrating the lower plenum,11 downward tv ation were to occur.

No. The heat sink is important in delaying the blockage, above, from melting, after the water has vaporized, to a level below the core support plate. Yes. The holes in the plate it.wlf would offer no resistance to melt penetration.

Tur13. 3. At this stage (page 4.1) translating W:expectthispath to be blocked' into ' Downward relocation is physically unreasonable'still appears a large step.

Yes, but the statement gives a preamble of where we are going in this chapter.

Tur14. 4. Reference to TMI-2. Looking again at the TMI 2 melt relocation event, I am struck by how far melt managed to progress downwards through the core, ghrn the water inventory that is generally believed to have been present (minimum of 0.5 m above the base of the core). For instance at position K9 near the centre of the core there was evidence of previously molten material beturen rods near the first spacer grid and in the spaces around the lower end Etting [TMI-2 Core Bore Acquisition Summary Report, EGG-TMI-7385, rev 1, February 1987, page B-30). This relocation was physically reasonable, as it occurred, but I do not see how it diEers substantially &om the claim that the APR-like core downward relocation is 'plc sically unreasonable.' Iam happy with the notion that relocation into the bypass (most PWRs) or downcomer in the APR-600 ic must likely - it is the degree of certainty that I question.

I V G-69

We really mean it is " physically unreasonable" to penetrate through the bottom of the core, and the TMI information noted does not conflict with this; indeed, it supports it. In the present cw we have also the Zr end-plugs as a further cold trap.

Tur%. 5. The low meldng point control rod materials are expected to escape early (page 4-4). This seems counter to other arguments about heat sinks. How-crer, if they do form a blockage, this may be relathely weak, giving the potential for a later downward relocation.

Yes, but these are intermediate states in melt progression. We are interested in how far this can go.

Tur10. 6. While the results on blockage formation appear realistic, the ther-mal equilibrium assumption in equation 4.1 is inconsistent with the growth of thermal boundary layers in the solid represented by equation 4.2. This may lead to an underpsediction of the plugging time, particularly for cooler structures.

The error is negligible in the context and timing of this freezing.

Tur17. 7. Table 41 - what is the meaning and signincance of Meltfree:ing capacity as multiple of thefuel md volume? For comparison (I think) one ceeas the channel volume dhided by the fuel rod volume to ensure that there is sutlicient heat capacity to form a blockage.

We think ours is en interesting measure because it includes thermal effects.

Tur18. 8. Page 4-6: The efective thermal conductivity is taken as the volume-weightedaverage. Here and elsewhere it would be useful to indicate what physical properties were used. The use of a volume weighted average is probably reason-able for this application (but not generally so). Was any allowance made for the porosity of the blockage in this evaluation.

Yes. See addendum to Chapter 4.

Tur19. 9. Page 4-7. While low melting point components of the core such as control rods are expected to relocate as they melt, this does not apply to the major metallic component - Zr. Best Sts to experimental data indicate that relocation iallowing clad breach occurs at temperatures in the range 2400 - 2450 K Relocation imvhes a signincant fuel component.

G-70

I c This is still well below the oxides melting temperature, and the fuel content is limited by

- the time available for melting and dissolution.

Tur21. . 11. The radial heat up calculations (Section 4.2) are qualitathely in line with similar calculations that we performed for Sizewell. Was a radial power deposition shape factor used? We found that somewhat different results were ,

obtained when we did the same calculation using a 2-D, rather than cylindrically ,

symmetric model, that took account of the proper core geometry and the power rating of individual assemblics. A difliculty with both your and our model is the absence o's relocation, which may imalidate the model once any melting of material occurs.

Yes. . Radial power factors were used as shown in the report. Clearly we do not expect to predict the details of relocation with this model, but this is not our purpose nor is it needed. Once relocation begins overall energy conservation is sufficient to take the relatively smaller way up to the melt pool formation.

Tur22. 12. The assumption of a fully axidised pool (Section 4.3) may be Inappropriate for a low pressure sequence. This raises issues on the interactions

. of the corium with a more metallic blockage (partially addressed in the MP tests).

Ov However, to conclude that something is ' physically unreasonable' all proccu.es that may have an impact should be discussed.

The point is that the cold trap at the bottom forms and sustains the blockage. Any inter-actions with the oxides on the top of it can do nothing to violate its integrity. All we need

, is the heat flux from above, as we have done.

Tur23. 13. The proposed melt release conditions and mechanism appear reasonable. The dimensions and the pour rate are no more than educated guesses (I would probably have made similar guesses). I note that to achieve the melt flow rate of im/s, only a 5 cm driving head is assumed, although there is no quantification of how close to the top of the pool the breach might occur. It is desirable to analyse whether heat transfer from the melt stteam tbrough the breach may deepen the breach and lead to an increase in pour rate.

Certainly will, but not significantly in a matter of ~1 second. It is the time coherence here that determines the reasonableness of the " educated guess" as a conservative bound.

,q U G-71

Tur38. 2. For this application, the supporting analysis ought to concentrate on the melt telocation scenario. This would include obtaining a better under-standing of melt :clocation in TMI 2 (en why did it occur after reBooding the vessel?), to demonstrate that the processes are indeed understood.

See Epstein and Fauske (1989) in Faul, and addendum to Chapter 4.

You9. 2. On p. 4-4, there is a comment about approximately 25?l of the fuel remaining uncovewd. Is there a reference for this?

Typical result of computations such as with MAAP.

O G 72

]

Premixing and Triggering (Reactor)

- Ber16. . Chapter 5: QuantiScation of Premixturee Ghen the melt release conditicus (Bow rate, location, temperature and compo-sition), the premixtures are calculated with an improved version of PM-ALPHA which is now 3D and includes a melt ' fragmentation law (which was lacking up to - ,

.now) as it is recognized that it is interesting < to know the distributions of the.

melt length scale >.J However, this fragmentation law is not described and this should be' done and justined as fragmentation is responsible for voiding (< the..

rate of voiding increasing rapidly with the rate of breakup >). I svuld also like to know why the < breaking law is operathe only for as long as the coolant has

.'s void fraction ofless than 50% >. If the fragmentation was always operathe, i voiding uvuld be larger so there must be a good reason for doing so but I do not ; ,

see why.

The breakup (not fragmentation) law is described in the PM ALPHA verification study,-

which was also supplied to the myiewers. There is a void fraction limit (conservatively set to 50%) to express the fact that, breakup occurs significantly due to the presence of water-inertia effects and melt-water thermal interaction effects. See, for example, our interpretation of MIXA and FARO data.

Ber17. . The melt entrance conditions into water are also speclSed and not calculated: ,

. - entrance velocity whose enluation is correct -

- distribution of the melt < over an ellecthe radial width of10 cm > with a melt solume fraction evaluated to get the correct mass Bow rate. This distribution is -p crucial in determining the amount pf vapour which is produced as the larger the .

>- entrance area, the longer Svu are in the B1m boiling regime in which the steam production is at maximum. This behaviour was observed in MC3D recalculations of FARO tests, where a doubling of the pressure increase (linked with vapour pro-duction) was obtained with a doubling of the diameter of the melt flow. Recently^

- CHYMES 2 recalculations showed the same trend.

Steaming rate shoul'd not be confused with voiding of a premixture. Here we are prin-cipally interested in the extent'of voiding, not steaming rates. Moreover, independently G-73 p

of the specified radial width, the melt will quickly disperse to cover the relatively narrow downcomer width. h Ber18. - initial droplet diameter which is set to 20 mm (a large enough value to represent a minimally broken up melt stream). This parameter is also impor-tant for vapour production. It would be interesting to see sensitivity calculaticns with diameters varying from 10 to 1 cm.

The initial melt length scale is absolutely unimportant, as long as it is large enough to represent a minimally broken melt stream as we expect to be the case here in the travel through the vapor space. We have demonstrated that by the 2 cm no-breakup case, which is less volded, but still rather benign, a larger length scale will still be more benign. Now, due to the breakup, in the advanced portions of the jet, the vapor, like a chimney, also would cover the less broken portions and give still more benign explosions.

Ber19, I am not so sure that the melt will be transformed in a droplet pop-ulation before entering the water. We may have a large melt stream or the wall with subsequent fragmentation into the water but with a different law than droplet fragmentation. Would it lead to a < benign e volution > as it is men-tionned. This is again an expert judgment.

This concem would be applicable if we used a mechanistic law for droplet breakup, but for this and other reasons we don't. Rather, we vary parametrically the law, and envelop the behavior regarding extent of voiding as related to the law (and extent) of breakup.

Ber20. As forjet fragmentation calculations with THIRMAL,Icannot trust thens if the fragmentation is still governed by Kelvin Helmholtz type calculations.

Moreover, in FARO experiments with 10 cm melt jet, it took more than 2 m of water to break theJets in a 50 bar atmosphere for which voiding is smaller.

A t 50 bar the voiding is smaller and the steam velocities generated are also smaller (than at 1 bar), and this is why in FARO we expect less breakup. Still, however, the premixture was highly volded, and the melt extensively fragmented according to our interpretations of the tests. Ours are fully consistent calculations using the proper non-local radiation absorptio:

law. The THIRMAL calculations are offered as another perspective, still parametric, not rrxhanistic. We do not believe anyone can do mechanistic calculations that can be trusted; moreover, we do not believe such would be a good approach even for the future (it would run into serious verification problems due to inherent problems with directly quantifying G-74

g this aspect in experiments). We do not know where the reviewer's assertion about the

() FARO tests derives from.

BGr16. 4. Comments on Breakup and Premixing From the assumed failure location and size, melt Bow rates of 200 to 400 kg/s are estimated, yielding ~5m/s entrance velocity into the saturated water poo! at a level in the range of the cor support plate. Then, the next main point is in my view the fragmentation process. It is stated that adequately bounding the cflect of various degrees of breakup leads to extensive voiding developing rapidly in all cases. This voiding of prenti xtures in calculated with a 3D version of Phi-ALPHA. The melt stream is assumed to be broken up initially into drops of diameter 20 mm ("large enough value to represent a minimally broken-up melt stream"). However, as compared to a coherent stream of ~11 cm diameter (with 400 kg/s and Sm/s), this yields a surface of factor 8.4 higher and correspondingly a higher heat transfer and steam production. Transient breakup could thus yield signincantly less steam production. On the other hand, the breakup may then not be sunicient for explosive premixtures. A factor of 6 still results for two jets of melt with correspondingly smaller diameters which facilitates breakup again

,p to some extent. Thus, mixtures with smaller void may result from transient G breakup and assuming severaljets with smaller diameters. On the other hand, there remains certainly a limitation to breakup due to the time consuming process of breakup.

In my opinion, these contrary elfects with respect to getting an effective mirture, i.e. too less breakup or too strong voiding with stronger breakup, should be ex-piored more for getting the inherent limitations to explosive mixtures. Although the statement of strong steaming appears to be plausible, it may not be possible to demonstrate it for all possible cases, as indicated above. A smaller window for explosive mixtures may become plausible taking into account the above effects of time requirement for breakup and too coarse breakup combined with weaker voiding. Perhaps, some additional variations with Pht-ALPHA could be done to show this, e.g. by considering plausible time laws for breakup of coherent jets together with varying breakup length scales.

Indeed, the THHthfAL calculations give some perspective on this, showing the extreme cases oflittle stripping of small fragments for a thickjet (7.3 cm diame-ter) and conne breakup for the smallerjets (2.9 cm and 1.8 cm) due to long-wave o

l i V G-75

instabilities. However, certainly cases in between these extremes of breakup be-haviour should be considered. Further, the present state ofJet breakup modeling cannot be taken as veriBed. This is also indicated by the signincant ditferences between results based on Kelvin Helmholtz instabilities and on the theory taking into account velocity proBles [hilles) which have been obtained with IKEJET, e.g. [2].

Since inultiple Jets may occur from one hole by some separation effects (e.g.

connected with the fallere iade) or from several failure locations, the restriction of mass assumed in Ap?cndix D, p. D-6 for the case with thinnerjets appears not to bejustined. Also, the concluding statement of p. 5-12, "that both length scales and void fractions are well encompassed by the Phi-ALPHA calculations" appears to me as too rough, in view of the variations of cases indicated with the THIRhiAL calculations r.nd consio*ered at>ove. On the other hand, I agree in principle to the expectation of strong voiding based on the situation considered, with melt into saturated water and with the necessity of breakup for explosive mixtures. Further variations may even better demonstrate this, as indicated above.

The key point is that one cannot have simultaneously enough interfaJal area for a strong explosion and low enough void fractions to produce energetics. This was amply demonstrated by the calculations made already. We have made additional calculations, and we have provided further interpretations on this compensating mechanism in the addenda to Chapters 5 and 6.

Cho7. Perhaps additional parametric calculations in terms.of the breakup and trigger timings might be useful.

See addenda to Chapters 5 and 6.

Cor8. b] The authors seem to have only considered the premixing process as the melt falis tbrough the limited water pool from its surface ta the curved RPV bottom.Would not mixing continue as the melt continues to fall along the wall.

This seems to have been neglected. Is this premixing process of no importance or is the premixing analysis with Phl-ALPHA not valid for these longer times?

The premixing results presented were not taken as far out due to time constraints. We now have extended these results to much longer times (see addendum to Chapter 5).

G-76

1 f ~. ; Cor11: [5)in section 6, the authora use ESPROSE.m in a revised 3D version - ' ,'

$ ' to simulkte the explosions within the RPVloner plenum. Ghen the premixed Y mass of fuel we have a range of results given in Table 6.L Only a couple of =

- questions arise:

a) Why is the trigger time so short; i.e., much less the 1 sect is it due to the thne to the RPV wall?Why cannot further mixing along the RPV wall cause larger.

explonicas?.-

We show that there is a peak (in time) in the explosive " quality" of the premixtures and.

trigger times are used to bound this behavior. More calculations showing this bounding -

behwlor can be found in the addenda to Chapters 5 and 6.  ;

i

. Fle6. 2.5 QuantlBcation of the Premixture This section addresses the determination cf the premixture connguration. Firstly,  ;

it is important to note that the highly 3D nature of the pour has been taken into account via the extension oi the Phi ALPHA code to 3D. Thus the localized, rather than smeared in 2D, characteristics of the melt-water interaction process can be simulated. Secondly, it should be noted that melt breakup has been ,

p) taken into account in a parametric manner. At Brst sight this may seem like a N weakness, as many proposed breakup models exist. However, ginn that none of.

these has been properly niidated it seems appropriate that the effect of breakup be addressed in a parametric manner. As pointed out in the report, in the event that the melt enters the water pool and runs along the nssel wall, t,here will be less mixing than calculated here and therefore the' explosion energetics will be reduced.

i. Based on my experience of premixing experiments and modelling I han no dif.

Sculty in believing that only tens of kilogrammes of melt are likely to be mixed

- in the ghtn con &guration. Clearly the high voiding rare 4 a consequence of the-y water pool being saturated. I was left wondering whethu in the event that the melt pour occurred during the reRooding processes shether there would be suf-Bcient subcooling present to increase these masses signincantly? hly expectation

. is that the increase would be by no more thun a factor of two, which svuld still i- reuult in small mixture zones.

r JL u -

~ -

.G :

4 w-y u a-'s

. -e e-.4 e-,,ie , ,hh w s =y y - . . ,a yy - y w- .,- , , y -

y- -g e-- g- g- y -r

The likelihood of subcooled water in the lower plenum pertains to the " fast" and " medium" scenarios discussed in Section 4.4. The effects of subcooling and of a higher pressure on g premixing are discussed in addendum to Chapter 5.

Jac7. 3.2 Modeling of premixing Premixing is the process that is thought to be required to set the stage for any large scale coherent steam explosion. It is, at the same time, expected to inher-ently limit the masses participating in an explosion by the ' water depletion' effect, i.e. removal ofliquid water from the premixture by large anwunts of steam that are createa due to fast heat transfer. As these processes are diRicult to simulate directly in experiments, recourse is taken to numerical modeling with the code PM ALPHA.3D. For the scenarios considered, this code predicts strong voiding of the volumes accessed by melt. In combination with a cut-off of propagation that is effecth'e at high voiding this gh*es a strong limitation of the melt masses that can interact. And this is the second pillar on which the final result of the study is resting.

While there are good asguments for the concept of ' water depletion' and also some experimental observations that appear to support the idea in principle, there remains the question whether the quantification ghen by PM-ALPHA.3D is sufficiently reliable. The program predicts 'the major portion of it [i.e. the fuel] being in a highly voided region (a > 80%)' and also that the void fraction

' gradient is very steep', i.e. the void fraction increases from values around 20 %

to more than 80 % within a short distance. Such behav'ar, however, was not seen in the premixing experiments that are being conducted at Forscimngszentrum Karlsruhe in order to study the phenomenon and to coUect data for code valida-tion [2), [3), [4]. It is too early to draw final conclusions from these experiments, but the void fractions in the surrounding of broken up ' fuel' appear to be smaller than expected.

The QUEOS experiments werc run under conditions quite different from those of M AGICO, from which the reviewer's " expectations" may have derived. We provided quantitative interpretations of the available QUEOS tests and see that this lower voiding should,in fact, have been expected. More imporantly, to this day we are not aware of any published, reliable void fraction maps over the premixing zone in the QUEOS experiments. We have such detailed maps in MAGICO (see Appendix B of DOE /ID-10504 and the addenda to G-78

l 1

l I

g it), and show that even with very dilute pours (0.5%) we get void fractions in the 60-() 70% range, extended over the whole mixing zone. The QUEOS pours are too short, and too concentrated to reveal the important thermal interactions that lead to extensive and l

persisting volded premixtures. l 3ac8. One may also draw attention to data reported oithe KROTOS experi-ments [5). In Gnese tests mc.iten alumina was poured tbrough an orifice with 3 cm diameter into a 10 cm wide tube filled with water, it mixed with the vater and strong steam explosions occurred either spor.taneously or following an external trigger. The me!t temperature was high, typically 2600 K, but the water was subcooled which, of course, tends to reduce voiding. In the KROTOS tests #28 and #29 the water was subcooled by 10 K and 80 K, respectively. In both cases the steam volume fractions within the reaction tube were 4 % only. But as these are mean values over the whole tube which may contain some regions occupied by water only, it may be more relevant to point out that the steam volume was only about half the melt volume. In test #30, subcooling was again 80 K but the melt mass was larger and its breakup was more intensive. in this case the steam volume fraction reached 23 % but this is again only 1.3 times the melt volume. So

'e must check how well tlm above cited calculational results of PM ALPilA.3D 0 are founded which imply steam volume fractions that are larger than the melt volume fractions by well over an order of magnitude.

Our calculations of KROTOS yield similarly low average void fractions (this is due to subec,oiing), so, again, there is no surprise here.

Jac18. There is a further and independent argument for early triggering.

It statw that early trig 3ering is due to the interaction of melt (jets) with struc-tutes. This widely used contention, however, does not agree with the observations from the PREMIX experiments at Forschungszentrum Karlsruhe. We have now performed il such tests and in 4 of these the melt was forced to interact with structures (vertical *)et' on horizontal plate - in one case even equipped with compartments). Only one of these tests (the : st one periormed on 21 August 1996) lead to a violent thermalinteraction (a weak stee.m explosion) al>out 0.8 sec (almost a full second!) after melt structure contact [8). One may also make reference to the KROTOS tests, in which the otherwise very explashv alumina melt settled at the bottom of the reaction vessel copying its shape when solidify.

Ing, in cases in which the water was saturated and no external trigger was applied

[5]. So, melt structure interaction does not necessarily provide early triggering.

t \

U 0 79

We did not limit our range of interest for trigger times based on such arguments. Rather, the trigger times were chosen to bound the energetics. h May3, 3. Quantification of premixtures The authors of the report came to the result, that for the APG00 the amount of melt, pouring into the lower plenum through the downcomer, would be in the order of a few hundred kg/s. Based on this information., they determined the range of premixtures of melt, water and steam and their distribution on the way to the bottom of the vessel. Their calculations are based on fundamental aspects of the premixing phase, which a part of the authors studied seriously in exper-iments (the blAGICO 2000), involving well-characterised particle clouds mixing with water /1/. In these experiments, they performed detailed measurements on external and internal characteristics of the mixing zones. hilxing in saturated and in subcooled water was studied. The results of these measurements found entrance into the Phi ALPHA code, which they at first used for interpreting the experimental results and which is the basis for the antlysis of quantifying premix-tures during a hypothetical steam exphelon scenario in an APG00. Interesting phenomena they found were the formation of densely packed regions and ofin-stabilities at the penetrating front (isothermal conditions) and local voiding in the mixing zone, as well as global voiding through the level swell (hot pours),

it shou!d be mentioned here that the original 2D Phi ALPHA code was extended to a three. dimensional version called Phi ALPHA.3D - version. The results, predicted for the APG00, showed, that parmixing mainly takes place in the down-comer and at its lower end to the lower plenum. The average mixture zone and voidage zone is mostly shorter than 1 metre and the average fuel length scale varies between a few mi!Ilmetres and 2 cm. It takes a (cw tenth of a second until enough small molten particles are formed during the mixing process.

This ghrs hope, that a very first steam explosion will occur, before a larger amotmt of finely dispersed molten liquid is mixed with the water and that this stry first steam explosion produces such a high voidage (steam) in the waterpool, that a further large steam explosion can be avolded. It is obvious, that the authors do not study this possibility, because it cannot be quantihed, but it may be alloned to mention it in this review. Roughly speaking, one could perhaps say, that early, small steam explosions are the best guarantors, that large dangerous steam explosions probably won't occur in case of mixing hot melt with water.

G-80

Another fact, which limits the momentum of a steam explosion, is thv high voidage 0,a In the mixing zone, extending over a large part ofit. This voidage has a strong damping effect on the migration ofpressure pulses, because it offers a compressible volume.

The mixing deliberations and calculations, presented in the report, are physically uvil based and desene a high grade of credibility.

Sch44. IV.3 Chapter 5. Quantification of Pre mixtures / Appendix B. De.

talled Pre mixing Results The chapter 5 develops the rationale for the pre mixing that results from the tchase oIthe UOz - Zr02 melt from near the top oithe core, tbrough the down.

comer, inta the water pool oi the lower head. The water level is assumed ta be a few centimetres above the top of the core support plate. Melt release rates of 200 and 400 kg/sec, reacidng the velocity of 5 m/sec at entry into water are considered. The melt saperheat is assumed as 180K.

The oxide melt Jet is distributed over an effecthe radial width of 10 cm in the q downcomer, with an initial melt volume fraction of x25 % at water impact. This N) would translate to a melt stream of dimensions ul0 cm x 16 cm for the release rate of 200 kg/sec and ul0 cm x 32 cm for the release rate of 400 kg/sec.

The expandad melt Jet is then allowed to traverse 20 mm in water, before break-up ensues. The break-up rates are parameterized from no break up tu very rapid break up (forming 2 mm size particles within 10 cm of travelin water.)

The aborn initial conditions were emplo>vd in the PM-ALPHA code to provide results on pre mixtare characteristles i.e. the melt and the vold volume fractlons and the fuellength scale, as a function of time, and position. The integral quantity of interest is the number of kilogram of melt mixed with coolant, before the triggering and explosion.

The Appendix D presents a number of colour pictures and many graphs giving detailed results. The graphs of fuellength sc.de, fuel volume and void fractions are presented for more f values and for timee up to n 1 sec. These pictures and graphs provide good back-up for the results, :wd e:guments, presented in chapter 5.

(3

() 0 81

I bellew the authors have presented a clear method of evaluation and the results are credible. I do have the following conunents.

Sch45. (1) The melt through fallure of the tcBector and cose barrel are as-sumed to be near the top of the melt poolin the original core boundary. If the failure is lower, the starting velocity for the melt Jet would be higher, and so will be the velocity at water impact. This may be benencial for break up.

We do not expect much difference. Twice the initial depth would increase initial velocity from 1 to 1.4 m/s, and the velocity of water impact remains essentially unchanged at 5 m/s.

Sch40. (ii) The initialimpact area on the water surface is quite large. The jet going through the 2 meter steam region should not break up, to that extent.

There is also a splash off the wall. More concentrated pours create higher voiding in the premixture, so we try to bound the behavior here, too.

Sch47. (lil) Both the very fast and the no break up cases show (Cf. Figs. 5.4 (a) and 5.4 (b)) that for the initial 0.1 see the fuel front is more advanced than the vold fraction front. This was also obsenvd in the PM ALP 11A veillication report. Later on, the void fre: tion front seems to catch up with the fuel front.

For the C 1 10 case at 0.4 seconds (Page B.3-3) a large fraction of fuel seems to be hung up in the volded zone. The same is true for C 1 - nb case (Page D.3-5).

In the C110 case, there would be a large steam Bux rising, which could retard the descent of the fuel particles. For the C 1 - nb case the steam Rux should he smaller, and the fuel particles of 2 cm should be ahead of the void fraction front.

Sch48. Suminarising, I believe the break up assumptions, both, in the steam during descent from the original core boundary, and during water interaction, play a crucial role nad, perhaps, this part of the pre mixing analysis could be strengthened. The no-break up case appears to produce approximately the same results as the high break up case. This has been recognised, also, by the authors (Page 510). Perhaps a physical explanation of why these cases produce such similar results may be provided by the authors.

G-82

The results show that all craes become highly volded. This result is really expected, given the radiative powet of such melts. See also addendum to Chapter 5.

Seh49. IV.4 Appendix D. Additional Premixing Ferspecthrs from the THIR.

hlAL Code i

in this appendix, the THIRh!AL code has been used by C.C. Chu and J.J. Slenicke of Argonne National Laboratory to' provide a perspecthe on premixing. The code had to be modlSed to describe the melt jet water interaction in the conSned ~!

geometry of the down comer. The calculations were perfortned for melt release ,

rates of14 to 220 kg/sec, with correspondingjet diameters of18 mm to 73 inm.

The 220 kg/sec case resulted in median droplet size of 2.75 mm, with a mixing zone radius and vold fraction at pool surface of 160 mm and 74 %, respectively.

These results are not too diflerent from what were obtained from the Phi ALPHA .

Code, although the jet entry conditions are different. THIRhlAL calculates Jet entry diameter of 6 cm (i.e., no break up in the down comer steam zone). blodels for break up in THIRhfAL must be quite different from the parametric model i emplo>vd in the Pht ALPHA Code.

O Tur11. Chapter 4: QuantlBcation of hielt Relocation Characteristics

1. It would be useful to ght an Indication of the diameter of the cooling holes.

t They are ~1 cm in diameter.

' Tur24. Chapter 5: QuantlReation of Premixtures .

1. Is there any likelihood for this plant of subcooled water in the lower head (eg in an extended accident sequence with some it&ction)?

This was addressed in Ch pter 4. No such likelihood could be found.

Tur25. 2. The comment that the break up parameter f set to 10 produces ,

very rapid break up in ~10 ctn of water suggests that the modelling is somewhat more efBclent at producing fragmentation than originally desired (break up in a speciRed fall distance taken as the smaber of the actual fall distance or f.Dj).

This also depends greatly on the assigned value of Dj - here set to the initial particle size (20 mm). If the melt was assumed to fall as a thinning sheet (quite O

V G 83

J possible) then the initial penetration of the water may be more local than sep-p b,

l resented in the PM ALPilA calculations. Ilowever, I am happy with the range chosen for 0 Tur20, 3. Please note that in Figure 5.2 and Appendix D the vold is rep-O?A resented by shading, the fuel 01 contours. Explain the contours that follow the damaln boundary.

Clarification made as requested.

Tur27. 4. Specify the boundary conditions for the calculation. What pres-surisation is predicted?

Due to the large volume of the system, the localized nature of interactions, and short times, no pressurization is predicted. A constant pressure outlet boundary condition at the top of the downcomer was imposed.

Tur28. 5. The length scale increase referred to on page 5 5 is not evident in Figure 5.4. The aren averaged over is not clear, it is obviously not the whole cross section. Since writing this i found the 1% fuel volume fraction limit on the region considered in the text for clarity add to caption of Figure 5.3.

Clarification made as requested.

Tur29. 6. Middle of page 510: 'Only a very small fraction of the coohnnt is found to co-exist with the water'-I know what you mean! It is clear though that here we have the key tesult anticipated for the mixing codes. This impiles that the key region to seek validation of the code is in the production of the high vold fraction.

Yes, and this was done in the MAGICO-2000 tests. See also addendum to Appendix B.

Tur30. 7. In un view the TillitMAL calculations raise as many questions as they answer, because of the poor ndidation status of anyJet break up model.

Ilo vever, I do not think this is a key part of the argument.

You2. First, I see that trigger timing was utried parametrically but not trig-ger locatlon; I assume that the cases were triggered near the bottom of the mixture G 84

~!

nglon next to the wall; although I suspect that this is probably the most severe -!

cane, I am wondering about the consequences of other trigger locations.

We did not do extensive variations on loation of trigger, but what we have seen agrees with - _ ]

what we expect: it is the premixture composition rather than the location or magnitude l t

of trl 5ger that controls the energetics. j i

.Youl0. 3. On pp. 5 6 through 5 9, it is hard to compare the graphs chosen I i.

because of varying z axis scales and varying times for the plotted lines. For  ;

instance, the C1 nb plots start at 0.4 s whenas the Rc1 nb plots end at 0.12 s. I }

see that there are other plots with overlapping times in the Appendix, so maybe .,

one of these would be better.

The point is well taken. See better representation in the addendum to Chapter 5. .  ;

-r I

t 1

i .

O .

I 3

t L

9 k

O .

e.,s a

.l

PM.ALPilA Code and Verification Dan 17. 2. DOE /ID 10504

• Premixing of Steam Explosions: PM ALP 11A Verincation Studies", Sept.1996 This is, once again, a thorough, and high professional, document on the verin-cation of the PM ALPilA code against available experimental data and known physics. The agreement with a wide range of data, from single particles settling in water, to particle swarms, both cold and hot (up to 2000'C), to integral tests with prototypic materials at high pressure, 's rather remarkable. The breakup constant B, has been chosen to Rt the integral data from several tests, but it is used consistently. The Richardson Zaki exponent for a monodisperse system of spherical particles has been used without modlRcation for thermal effects, with excellent results. The FARO experiments, which gave very little usable data on the Jet breakup and dispersion, has been well ayproximated for the measured steam Bow and pressure. The overall result is that the code seems to be well suited for licensing purposes.

Dan 18. However, some speelRc commentu may be made and questions raised:

1. Eq. 2.3. This equation is incorporated into the code, but no independent check on the R Z exponent is made. On the other hand, the R Z exponent was chosen by comparison with a large body of data on systems of spherical particles.

2. Figs 9 and 10 (p. 210'. are interes:Ing in showing decaying oscillations, and an at tractor above, but clore to, th= steady drift flux / particle volume fraction cune. This physics appears to bo new, and should be further imestigated.

3. p. 2-16. The careful trentmen'. oI the radiation boundary condition with slight subcooling is noteworthy.

4. Figs. 8 and 9 (pp. 2 30/2.41). The comparisons between the predictions for front position and the data (c.r the QS - Q11 experiments in the QEOS series is remarkable. The lewl sur11 is not well predicted in the Rrst 0.1 s, but this may be due to experimental uncertainty. For the important range t > 0.2s the agreement is excellent.

G 86

m 5. The explanation for the absence of a pressure hump at early times compared

) to measurements for Qli, as being due to the extra radiant heating before impact, seems to me to be reasonable.

6. Turning to the hilXA experiments, there is reasonable agrwment with the pressure data, and excellent agreement for the cumulathy steam flow.

7. Similarly, there is remarkable agreement with FARO L 14 water level swell, pressure, and pressurization rate, especially considering the complex geom-etry of the equipment. The additionalinformstlan from the code on local wid fraction, saelt temperature, melt volumetric fraction and melt location seems to me to be very useful, in view of the inherent limitations of the ex-periment. hly own view is that the cost-benefit ratio for further experiments of this sort is large.

• * * * * * * * *

• c: * * * * * * * * * * *

• Dur22. It remains to formulate some general questioning concerning the ver.

ification state of Phi ALPHA and ESPROSE. Although a lot of work has been performed on this, i think that severe questions remain. Even ifnumerical aspects may be considered as well established, also with respect to 3D, there remain open areas concerning the physical formulations. These are e.g.:

o Chech with h!AGICO were to my knowledge restricted to relatively small volume parts of spheres.

Quite obviously this is the most interesting ecndition. See also addendum to Appendix B of DOE /ID-10504.

Dur23. e in general, correlations for exchange processes in three phases are uncertain and need further clarification.

We have an extensive data base on film boiling in steam water flows, and a rather elaborate non local radiation transport model. Also, our drag laws are well based on existing fundamental knowledge. Furth=r, a wide range of tests on the multifield aspects show that the code performs well. Of course, there is always room for further developments, but the issue here is whether the reviewer sees any specific limitations or concerns.

Dur24.

• The uncertainties on jet breakup have aircady been mentioned above.

o 0 87

Our approach is specifically based on explicitly recognizing, and bounding, these uncer-tainties. h Carlo, d] Finally, the Phl.ALPilA model has a parametric fuel breakup

' model that is mentioned bricBy, but has yet to be assessed against experiments.

For tbese small pour rates, the model ellect is not af great intetest, but would be for larger pours in these complex gwmetries.ls this model discussed in the support documents?

See PM ALPHA verification report. The model was shown to represent very well all available experiments.

Cnr14, INITIAL COhlblENTS and QUESTIONS for DOE /ID-10504 The overall report is quite informative, but I do have specific comments / questions that need to be addressed.

1) The analysis of the QUEOS experiments are very interesting. For any of the experknents lQS, G, 8,10,11] the visual image is compared to the code, and the leading edge, level swell, stcaming rate, stcam produced and pressure is compared.

bly first question is what is the criterion to determine the leading edge? In the pictures for the tests, specifically Q10 and Q11 it seems to me that Phi ALPilA is predicting the movement of the front to be faster than the data indicates. Yet in the plots the opposite is represented. Either there is a contradiction or I am observing numerical diffusion in the images and the researchers have a definition of the leading edge that " corrects or compensates" for this. I have seen the same behavior with IFCI and therefore, am sensitive to it. This needs to be sorted out before I would say that the agreement in the kinematics is acceptable. The h!!XA results in Section 3 seem to indicate the same behavior to me and thus 1 am worried about this numerical diffusion. There was also no study of the nodalization cliects in Section 4 and this is surprising ghen the results in Section

2. This seems to be a logical thing to do and really should be done.

As stated in the report the predicted " front" was obtained from Lagrangian tracer particles.

Also as stated in the report, numerical diffusion can be controlled by the grid size, and results with still finer grids were pmmised, to improve the already quite good results. Such results were obtained in the interim and can be found in the addenda to the verification report. Now we have also the PM ALPHA.L code which eliminates numerical diffusion G-88

g altogether. The verification steps were redone for this code, and results are reported in the addenda to the verification report (please refer to the cover letter of the present package).

Cor15. 2) The second comment about QUEOS relates to the radiathe heat transfer model. On page 216 the report states that the radiathe model had to be changed from what is normalin PM ALPHA to accommodate the experiments.

Later on page 219, the report states that the tests do not meet the 'Rtness for purpose' criteria, and one reason is that the temperature is too low [2000C compared to 3000C). I am troubled by this empirical "Rx" to model the test and thus, am wondering about the " mixed" transport model in PM ALPHA. This is known to be a tough problem, clearly, but to arbitrarily change it seems too rough. Also, I disagree that the tests are not "Bt for purpose". They are more Rt than others and thus, are very relevant. Thus, the proportion of the radiathv transfer that goes into bulk heating versus steam production is important to consider and improve upon.

All we are saying is that at 2000 *C the absorption length is so short that basically all heat will be delivered to the interface, so there is no need to compute with the full radiation transport model. Quite the opposite occurs at 3000 *C, This is not an empirical fix. About p) the " relevance" of the QUEOS tests, see the addendum to Appendix B in DOE /ID-10504.

Cor16. 3) I would also like to see a calculation of PM.AL PHA/3D for Q UEOS ifindeed thereis a benent to a 3D calculation. It seems that the QUEOS tests are the largest and highest temperature shnulant tests to date with solid particles; thus, it may be of use.

The QUEOS tests are only repeats of our MAGICO tests. Both reach 2000 *C and both have about the same masses. Moreover, and in contrast to MAGICO, nothing is known from QUEOS about the most important part of the interaction, which is the internal struc-ture of the mixing zone. Still, not only did we interpret these tests, ours were the very interpretations, the next one coming by the investigators (of QUEOS) themselves some eight months later (at the CSNI FCI meeting in Japan, May 1997)! Since the QUEOS tests are axisymmetric nothing is to be gained by a 3D calculation.

Cor17. 4) The report Snally examines the FARO-LWR test L-14 as a com-parison with a large prototypical simulant melt poured into water. This seems like a reasonable comparison test, but I am surprised about what data is compared.

There is an enormous amount of data available over the Brst twenty seconds of

() G 89

the test [the first 5 6 seconds is reasonable before heat loss comes significantly into play) and yet the data comparison is sparse at best. I would suggest the following wrlables be displayed and compared mer the first 5 6 seconds:

a] the total pressure and pressure rise rate [done now) b) the steau, and water temperature at a few locations since its 2-D c) the kinematics of melt entry and arrhal at the chamber base and settling d) the surface area generated by the breakup as a function of time e) the mean particle diameter as a function of time

[] the energy flow to vapor and coolant liquid and loss by fuel g] the level swc!) of the pool [done but not for long enough times)

We disagree with this comment. For items a], b), c), and g], results and comparisons to the experiment were already provided (Figures 3,4,6, and 9). These are shown for up to 2 to 3 seconds because the premixing process is over after this time. For items d], c), and f], there are no data to compare, but complying with the reviewer's request we provide the result of computations in the addendum to Section 3.2.2.

Cor18. A!so I am concerned about the arbitrariness of the dynamic breakup modell0 value = 50), that is used and described in pages A 34/35. This whole proccdure is a matching exercise for some value of beta unless the results begin with a jet of ~10cn and break up to a size that is consistent with the post test debris data lns well as the amount left as a ' cake'). It would seem advisable to compare the ' frozen' model to other FARO tests to prove that results can l

be consistently predicted for LOG, LOS, L11, L19 and L20; all of which were high pressure tests for quenching. Also the ' mixed' transport including radiative transport would haw to be held constant in these comparisons to prove the match of L14 has some limited 'unhersality'.

See addendum to Section 3.2.2.

Fle7. 2.6 Quantification of Explosion Loads This section deals with the determination of the magnitude of the possible explo-sions that could be generated from the premixtures calculated using PM ALPHA.

It is important to note that tbese calculations, performed using ESPROSE.m are G 90

l fully 3D and can therefore account prope-ly for explosion venting. The validation

(") of the modelis discussed in a separate oection. It is sutlicient here to note that the code has been subjected to a very signiBeant validatson elfort which I believe shows that it is 'St for purpose'.

I agree with the approach adopted regarding triggering. SpeciScally, triggering i

at different times and looking for the maximum load is clearly conservative. In addition, the effect of the premixiq breakup parameter 6 is consistent with ex.

pe:imental observations and highlights the fact that the uncertainties in breakup can be taken into account in a parametric manner.

Given the premixture conngurations determined using Phi ALPHA I am not the least surprised that none of the explosions challenges the integrity of the lower head.

Fle11. 3 Phi ALPHA Verincation Studies (DOE /ID-10504)

This section presents a review of the Phi ALPHA verincatlan studies report. It is important to note up front that Phi ALPHA has been the subject of continuous

,a development and peer review (at conferences) over an 8-10 year period. It is

() therefore a mature piece of software.

3.1 Introduction The main point ofinterest in this section is Figure 1 which lays out the verinca.

tion and validation approach. This is very comprehenshv and covers numerical aspects, comparison with other codes and analytical solutions and with experi-mental data. I can suggest no improvements to this validation matrix. It is also worth noting that this section highlights the new feature of Pht-ALPHA, namely extension to 3D which is clearly needed in the Study. Th!s clearly represents a :nassive anwunt of uvrk but the new insights gained are deBuitely worth the effort.

3,2 hiultineld Aspects This section deals with the testing of the multiphase constituthe relations and the modelling for the sedimentation of partic!es or clouds of particles. Phi-ALPHA compu tational results are compared with experimental data and analytical models (based on the drift Bux approximation) for the sedimentation of single particles O

() G 91

l i

l i

and clouds. In allcases agreetnent is excellent. A novel feature cf this presentation is that the trajectory of the solution in drift flux volume fraction phase space is presented. These results show that the solution is approached in a variety of ways and helps to explain why multiphase numerics prove to l>e so complex. These results confirm that the code can reproduce the conect particle fall speed, an important feature the steam explosion study.

Numerous refereed papers have been presented showing that Pht ALP 11A can simulate the hfAGICO tests, where in most cues there is also phase transforma-tion. These sihnulations also show good agreement with local data on mixture composition and vold fraction. This is important as Phi ALPflA must predict the correct mixture composition if the calculations of explosion propagation are to be reliable.

Cornparisons of Phl ALPilA simulations with data from the QUEOS tests are generally good. There is evidence of numerical diffusion in, for example, Figure 6 but the authors are aware of this and are planning runs on finct grids. In the hot cases I agree with the authors that both the relathrlylow melt temperature (making radiation absorption a surface phenomenon) and the gravity-induced subcooling are important. If the explanation of the difference in steam production advanced in the text !s correct (namely the superheating of a layer of water during the fallstage)it means that interpretation ofexperiments of this type, where there is reinthrly little steam production, will always be very compilcated. Ghrn the short time available to the authors to analyze this data and the experitnental uncertaltaties i feel that Ph! ALP 11A performed as well'as could be expected.

See new results with finer grids and with PM ALPHA.L (addendum to Section 2.2.3 of DOE /ID 10504).

Fle12. 3.3 Integral Aspects The code comparisons with CilYhlES and between the 2D and 3D w.,wns of the code ghv a high degree of confidence that the basic numerical algorithm is correctly coded and that the 2D and 3D approaches are consistent. The compar-Ison with data from the bilXA06 experiment is at least as good as that acideved by the experimenters using the CilYb!ES code. The lack of melt spreading in the simulations is very similar to that found using CilYh!ES. The level swell and steam production data are well reproduced ghtn the experimental uncertainties.

G 92

f Again it is fair to say that this test is well simulated glwn that there are several important experimental uncertainties regarding particle breakup and the steam Row rate.

The comparisons of code calculations with data from the IA4 FARO experiment are also good, in this experiment there is no local data and only global quatitles, such as vessel pressurization and level swell, are asallable for code comparison.

- The choke of parameters to match these data seems very reasonabic. I found the Bguren illustrating the non local absorption of radiation lateresting and these clearly !!!ustrated the importance of this phenomenon for high temperature melts.

To my knowledge these are the Brst calculations to include this feature, which is

' clearly ofimportance in high temperature melt applications.

3.4 Breakup .tspects

- ! completely agree with the chosen approach to breakup. As more tests are analysed it will be possible to increase the degree of con &dence in the chosen values for the parameters. Clearly, given that the melt surface area transport equation is already caded it would be a simple matter to include a mechanistic model, should a validated breakup model become asullable. However, the analysis presented in the study shows that the overall predictions ofloading are Insensitin O ta the choice of tbese parameters. Therefore the lack of a detalled model does not in any way effect the conclusions of this study.

3.5 Numerical Aspects The authors are clearly awate of the need to avoid numerical differencing ett'ors and the presented calculations show that they are taking care to address this problem.

3.6 Concluding Remarks I think this section identlSes the correct areas for future focus. If I were the authors I would haw made more of the fact that this is the most comprehensive validation effort to date and that the code has performed extwmely well.

3.7 Appendices Appendix- A provides a comprehensive description of the constituthe laws and -

- Appendix B provides a detalled paper on the MAGICO tests. The revienvr is 0 93 h

familiar with the materialin the Appendices and this has not been revleurd in detail.

Jac9. The original PM.ALPilA was one of the two plancering codes that used tbree velocity fields for describing the separate motlons of melt, liquid water and steam at the cost of adding considerable complexity to the already quite complicated two field description of two-phase flow. But this is the only way in which one can hope to develop a reasonable description of the phenomena dur-Ing a steam explosion. The fairly standard multiphue equations used provide compliance with the conservation equations only. All the controlling and very complicated physics in the three phase (and at least) three-component inixture must be described by constitutive relations. Ilere the difficulty arises that one of the main purposes of such codes is to extrapolate from the experiments that are possible in practice to the envisaged accident situation. This implies extrap-olation from simulation materials (sometimer enn solid spheres) to the expected (but still quite uncertain) molten corium, from often quite low 'mcit' tempera-tures to temperatures around 3000 K, and from the mostly very small scale of

• cxperitr.cnts tv 'he reactor size. There are a few experiments in which one or the other of the abow initial conditions is not as bad as indicated here but as the experimental difficulties sio., i.uvrmously as the expected accident conditlons are approached, the experknental information on the initial conditions and de.

talls of the processes is often poor in these cases so that a successful comparison of calculatiotud results with Integral experknental results doesn't necessarily in-dicate correctness of the theoretical model. Indeed, one can ex;>ect a code' to perfortn the required extrapolations only, I(all relevant mechardsms are modeled mechanistically and with sullicient accuracy.

This is why we have a very carefully developed verification plan, covering all aspects of the calculations.

Jac10. Ilowever, the cons *itutive relations used in PM ALPilA are often heuristic, sometimes parametrical. The latter is described in th report for the melt breakup model but is true as well for one formulation of the evaporation rate. The other formulation looks more physical but still does not allow for the possibility that evaporation and condensation occur concurrently in the same integration wlume (calculational mesh) due to limited subcooling of the water 0 94

and intenshe local (radiant) beat Bux to the vapor / liquid interface where the

&o melt drops are covered by a thin vapor Rim only, as e.g. on those parts of their surfaces that are oriented towards the direction of motion. So any extrapolation to accident conditions must be aRlicted with large uncertainties.

Only the breakup law is parametric, and its basis and rationale have been explained in

• the report. Most importantly, and counter to the reviewer's clairns here, PM ALPHA includes a correct phase change model, as well as a non lecal radiation deposition model (it is unique in mcognizing this important physics among all such codes). Evaporation and condensation cannot occur simultaneously. Probably the reviewer refers to the energy split between sensible, that going into the (subcooled) liquid bulk, and the " rest" going to evaporation. This is properly modelled in PM ALPHA.

Jac11. Validation of the original PM. ALPHA code by comparison with ex-periments was Brst described in Reference [6] which is also reproduced as Ap-pendix B in the special verincation report [7). An appeal of the general agree-ment .mached may be obtained from the data on the leading edge advancement.

With cold spheres this agreement is mostly reasonable. With sphere tempera-tures of about 1600 K the data are reproduced within about a factor 2. In the

' production runs' of the present study the laulal temperature of the melt will have been beyond 2900 K so that the uncertainties will certainly have increased

(~)

G quite considerably.

We have data now up to 2300 K, and in all cases the code very accurately predicts the front advancement. Perhaps more importantly, the Internal structure of the mixing zone is predicted quantitatively. Actually, as .aelt temperatures increase, the prediction task gets easier. The reviewer's is a very poor way to declare errors. When the quantity goes to zero the error, even for an excellent prediction, would go to infinity.

Jac12. Here we are mtJnly interested in the high void fractions that have been measured and predicted during the wrincation process. The data given in

[6] have been obtained with the MAGICO experiment and have been described as highly relevant ('the measurement not only provides insight into premixing, but represents probably the most important test for computer codes'). Hence our expectation to find high local void fractions in our own experiments. However, the local void dsta presented la [6] have been measured in a position or bet ter line or 'small region' (of unknown size) 15 cm below the initia! water level. This depth

'~

is only two thirds of the equhalent diameter of the pour. We may guce

measuring wlume was centered with respect to the particlejet (the ]

(O V G 95

its width compares to the width of the pour is not known. The measurement was performed at 0.35 sec, i.e. Just after the end of (or behind) the pour, probably in order to avoid the presence of many spheres at the level of the measurement.

These circumstances appear to have produced the observed high vold fractions passibly without too much contribution of steaming. It is our observation from the QUEOS experiments l3], [4] in which streams of spheres are poured into a water poolin a similar way, that the particle cloud is always followed by a gas filled chimney - with cold spheres as well as with hot spheres. This is largely a consequence of the momentum transfer between the particles and the water while thermal effects are of secondary importance - they essentially innuence the way in which the gas chimney is closed again. That this is also true in the MAGICO experiments is clearly shown by Figures 14 and 15 in Reference [6] which illustrate a ' cold' run. This means that the reported high vold fractions have little to do with the so-called ' water depletion' effect and there is no experimental support for the high void fractions calculated in the ' production' runs at positions far away from the melt entrance. One might add that corresponding to our observations in the QUEOS experiments, thermal effectsjust start to be detectable in an overall sense (beyond local effects around each individual sphere) at sphere temperat ures as low as 1600 K. Even at the much higher temperatures beyond 2300 K that have been reached in QUEOS, no high void fractions could hc observed outside the initial gas chimney produced by the entering clouds of spheres (essentially by momentum transfer).

The QUEOS behavior is peculiar to the experimental conditions and it is quite predictable with PM ALPHA and better yet with PM ALPHA.L (we addendum to Section 2.2.3 of DOE /ID-10504). The addendum to Appendix B of DOE /ID-10504 should be helpful to the reviewer in sorting out the diffemnces in his own mind. Comments such as in his last sentence need to be supported by data, for otherwise such points and responses can only produce confusion.

Jac13. In the main body of the verification report [7] global estimates of the water content within the mixing zone in QUEOS are used for further checking PM ALPHA. Unfortunately this type of data is hardly suited for a quantitaths comparison with code calculations. The diBiculty is that the result very much depends on the choice of the outer radius of this zone because, due to the weighing with the radius squared, it is this region that dominates the integration over the total volume. In the experiment this difficulty can be overcome to some G 96

n extent by precisely determining the shape of the mixing zone from high quality photographs - at least to the extent that a qualitative result can be obtained.

Ifowever, in code calculation *, the calculational mesh is not able to sufficiently i resohe this outer boundary. So, what is ghen in [7]is the 'PM-ALPHA result for the central region of the mixture, containing the main portion of the particle i cloud.' As a consequence, the calculated valut is somewhat ambiguous and Figure 13 in Chapter 2 of Reference [7] unavoidably compares quantitles with different definitions.

We did the best we could with the data available in QUEOS. The weakness is not with the calculation (with fine grid and Lagranglan particles we can resolve tne mixing zone to a very high degree) but rather with the experiments that give only a very rough estimate of a zone average void fraction. The new interpretations of QUEOS with PM ALPHA.L should help this reviewer understand what is going on in QUEOS (see addendum to Sectlon 2.2.3 of DOE /1D-10504).

Jac14. It remains that the code in this case predicts low voiding (in contrast with the product 2on runn). But here the code appears to have gone to the other extreme due to its inability to describe evaporation in the presence of subcooled water which even hads to the reported underestimation of caporation (steam a flow) rate and pressure rise. Tb explain these discrepancies by possible liquid

\ cuperheat of the water in the experiment is probably inappropriate in the presence oflarge free surfaces.

Incorrect in both respects. Our code describes evaporation in subcooled water well, and we estimate the steam flow quite well. The extra peak is indeed due to water surface layer superheated, as described in our report. The reviewer has not provided evidence to refute this real phenomenon. We now have a more precise model for it, as well as of the mixing phenomenon with PM ALPHA.L (see addendum to Section 2.2.3 of DOE /ID 10504).

Jac15. Another uncertainty of the calculational results is due to modeling the corium breakup. The surface of a certain amount of material varies linearly w"'

the (inverse of the) particle radius. Therefore modeling the corium as individual droplets with 2 cm diameter (wm the very beginning ghes it aircady a quarter of the carface that it uvuld have with drop diameters of 0.5 cm wh!ch can certainly be considered as well prefragmented (broken up). In the calculations presented, this Inillal diameter is combined with an entrance volume fraction of 25 % only so that there is an intensive thermallateraction from the very beginning. However, in the PREMIX experiments being performed at Forschungszentrum Karlsruhe (2), we have observed that a melt jet can penetrate to quite some depth lato V G 97

__ ~

saturated water (e.g. 0.5 m for a jet diameter of about 4 cm) before lt starts to break up and to interact more violently (still not exploshrly). In these cases the melt is molten alumina at about 2000 K the density of wh!ch is only about one third of that of corium. So this behavior is esen more probable (should be n.Jre pronounced) with corium. Such dynamic breakup process with virtually no breakup in the beginning that al cw the melt to penetrate deeply into the water followed by more rap;d fragmentation that breaks the melt into medium-sized drops (which might be the most dangerous conGguration) cannot be bounded by tbo parametric breakup model that was employed. Such bounding would require to model as well the entrance of coherent melt (melt being the continuous phase) that is not prem!xed with water artlBclally (by uumption) from the very beginning. La this context it is also important to note that breaking the melt into ven s.vd1 droplets (e.g. 0.2 cm) may be very optimistic because these small drops produce a lot of vapor, b.e. high tviding and may already start to freeze so that %y can so longer participate in an explashe Interaction. The importance of knuq im the benign explosion results reported is not discussed.

There is no instrumentitlen in PREMIX to provide 6 formation on the breakup charac-teristics. Our approach easily spans all regimes, from a coherent jet (large length scale) to a broken up cloed. The transition is controlled by the breakup parameter. The cases provided in the report were a selection from trial runs over a much wider variation of ini-tial size and breakup rates. With the radius changing from 2 to 0.2 cm we have one order g

variation la interfacla' stea. More importantly, the interaction (voiding) is controlled by specific interfa .al a .a (that is, area percent volume of mixture) which,in turn, depends, in a highly comp.ex, notritnear fashion, through the melt length scale, on momentum coupling between all thrre phases. Finally, the reviewer by focusing on voiding alone, is missing the point completely (explained repeatedly in the report) that it is the combination sf voiding status and respective specific melt interfacial area that matter on energetics.

There is a key compensating effect here that is yet to be understood by the reviewer. Freez-ing is not important here due to the short contact times. In considering what kinds of voids can be produced with what kinds of drop sizes, the revie" <r should take a look at the new M AGICO runs (see addendum to Appendix B of DOE /ID-10504).

Moo 4. 6. It appears that in the heat transfer predictious of PM ALPilA in DOS /ID-10504, flow regimes are identified by steady : tate corrtlations. Are these likely to be nonrepresentathe for such transient events as fragmentation, and not provide a conservathv characterization of the actual heat transfer?

G 98 i

l

This question is not clear. Fragmentation is relevant to propagation (ESPROSE.m) not p) premixing 'PM ALPHA). Heat transfer of tlie fragmented debris is conservatively taken to be infinitely fast (thermodynamic equilibrium assumed in the m fluid). If the reviewer is referring to breakup, this is a much sk,wer process, that is not even predictable. The idea is to bound the behavior, and for this our constitutive treatment is quite adequate.

Moo 8. biELT INTRODUCTION AND FRAGh!ENTATION Early predictive models provide core melt scvnarlos and relocation rates with and without reBood, whic.*. ran arrest the melt progression. flowever, the melt state which may reach water in the RPV, and the subsequent breakup and penetration latgely determine the raie oIheat transfer, eteam formatlon rate, and possible shock pressure loads. A quantity of melt arriving at the water can undergo Taylor unstable breakup or droplet formation at the leading edge and fielmholtz breakup or droplet stripping on those surfaces with parallel velocity components.

. The PM ALPilA model has been developed to incorporate the melt and coolant properties, and provide an envelope for the expected range of momentum, heat transfer, and phase change interactions associated with breakup for premixing considerations.

Single particle and particle cluster experiments have been employed to test pre-(~) dicthc capabilities of particle motion and energy transfer dynamics in water G' (the MAGICO and QUEOS experiments). Particle cloud clongation, steaming, spreading, and mhing with surrounding water are captured by the PM. ALP 11A code, which is employed as a causal relation in the ROAAM. Comparisons include

> particle cloud distortions associated with release door opening time, particle, and vold wlume fraction contours. Of particular interest is the pinching of the vapor volume behind moving particles, caused by condensation for the particle intro-duction into subcooled water. Since the condensation acts to reduce mechanical energy transfer, I agree that it would be useful to conduct QUEOS experiments in fully saturated water.

Moo 9. One of the saast important considerations in fragmentation is the for-mation of new melt heat transfer area. Appendix A in DOE /ID 10503 describes the " source term" for interiscial area production. Equation (3,69) is based on a change in size of particles for the same particle number density. It seems that before particles haw reached a stable size? they would undergo the formation of new particles. This assumption needs more explanation.

/D U G 99

The source terms in interfacial areas are very different in " breakup" during premixing modelled in PM ALPHA, and in " fragmentation" during propagation modelled in ES-PROSE.m. This comment mixes up these two. The single particle approach is appropriate h

for fragmentation. For breakup we, in fact, have included both changes in numbers of large particles, as well as change of their size due to their shedding of very fine particles.

This is explained in Appendix A of DOE /ID-10504 (PM ALPHA verification).

Sch2. 11. Review of the Report DOE /ID 10504 (Sept.1990) PREMIXING OF STEAM EXPLOSIONS: PM ALPHA VERIFICATION STUDIES" by T.G.

Theofanous, W.W. Yuen, S. Angelln!

This report is the verification document for the Code PM ALPHA, which treats the premixing phase of the steam explosion scenario. The report has two im.

portant appendices: (a) which describes the PM ALPHA models and (b), which describes a set of experiments in the MAGICO 2000 facility, in which several kilograms of high temperature particles of a spec'ific material, and of specific di-ameter, are dropped into water to obtain observations and data on the pre-mixing geometries and void fractions. The front parts of the report provide the compar-Isons of the predictions with the PM ALPHA code against the data from selected experiments. In the following paragraphs, I will provide connments on the main sections of this report.

11.1 Avvendix B. " MIXING OF PARTICLE CLOUDS PLUNGING INTO WA.

TER" I am very impressed with the MAGICO facility. I bellew the authors have per-formed outstanding experiments using quite high temperatures and respectable masses of the hot partichm. The video pictnres are outstanding. Iam a bit disap-pointed with the quantitative data that could be obtained. The X ray pictures (in reproductions) do not communicate any inforniation and the void fraction data shown in Figures B.23, B.26 and B.27 is rather raeager as a validation standard.

Point well taken. It was just too expensive to provide original prints of the X rays. But we have more data now and quantitative reconstruction of the radiographs, provided in color (see addendum to Appendix B).

Sch3. The comparisons of the PM. ALPHA predictions to the measured data, shown in Appendix B, for the cold runs, show substantial differences in the ad.

vancement of the particle front. It appears that a central part of the particle cloud tunnels tbrough the water. This is not predicted well by the code. For the hot runs, it appears from Figures B.26, that the calculations predict that the G 100

dense particle cloud also leaves the steam region behind, if a slight subcooling

(')

(TC) is present. There are no comparisons shown for the hot runs, as shown for the cold runs in the Figs. B14 and B.15.

The question that should be asked is how much detail is necessary, and reasonable, to expect in predictions? The cold runs, sometimes, exhibit a front instability, and this might well be due to some slight experimental variation such as, a slight delay in particle release, from cylinder to cylinder. Taken as a whole the PM ALPHA prediction of cold runs is very good, and the PM-ALPHA.L predictions are excellent. At the time we had only very localized X ray data; however, now see addendum to Appendix B.

Sch4. The concluding remarks state thet the hot tests quantifled local volding in the mixing zona and global volding through the level swell. Figures B.25 and B.26 Indicate that the voiding hont is coincident with the particle front, only, for the zero subcooling case. The particle front is substantially ahead of the volding front (c. a slight (TC) subcooling of the coolant. The extensive steam generation, Indicated by the axial vold proflie also may increase the local subcooling by pressurization. I wish there was some quantitative data for the particle volume fractions, to compare in Figures B.25 and B.2G. Was it not possible to obtain quantification of the spatial particle wlume fractions from the p X ray pictures?

V Yes, we now have quantitative information of particle distributions, and more importantly, on the relation of the particles to the void front. See nu materialin addendum to Appendix B.

Sch6.In this context, if the PM ALPHA predic'lons t for the advancement of the particle front lagged behind the measurements in the cold runs (cf. Figs. B.14 and B.15), sould they not do the same for the hot runs, since :nsme modeling is employed for both hot and cold runs. I do expect that the sv^c pneration, caused by the radiative heat Bux :. the coolant from the parUcia cloud, will retard the advancement oithe partleks. I belleve, this etkct n. 4. resented in the code, since a radiation heat Bux model as employed, however, I can not quantify its effect, on the differences in the particle cloud distribution between the hot and the cold runs.

In this, and the above two paragraphs, reviewer's concern is on whether the particle-void fronts and their relation are properly calculated. We have more detailed and complete data from MAGICO now that completely addresa this concern. These can be found in the f3 U 0 101

addendum to Appendix B. Also, the PM ALPHA.L predictions should help alleviate this concern.

Sch6. The subcooled coolant is important. The only data shown for the 18'C subcooling case is the lack of measured level swell. I would be Interested in the axial void fraction and the particle volume fraction profiles, to understand ii there are signiticant phenomenological differences between the saturated and the subcooled cases, and if these diiferences can be predicted by the PM ALP 11A Code.

As noted in the report, there was no measurable void in the 18 'C subcooled case. This was very well predicted by PM. ALPHA. Subcooling effects were key also in the interpretation of QUF.OS, MIXA, and FARO tests - see respective sections DOE /ID-10504.

Sch7. All in all, I believe the MAGICO experiments are relevant for the ideas, and data, on the mixing zone and the premixing conditions. I would like to cNnect the melt jet particulation to the particle-cloud water interaction. This may be in the next phase of authors experimentalimystigations.

Yes, we plan some work in this area, but it is outside the present effort.

Sch8. IL2 Annendix A. *PM ALP 11A: A (COMPUTER CODE FOR AD.

DRESSING Tile PREMIXING OF STEAM EXPLOSIONS" PM ALPilA is a three (melt, coolant and vapour) field code employing separate mass, momentum and energy equations for each field. Thus, it is a very detailed code more detailed than the codes RELAP-5 and TRAC. It also employs two and three dimensional geometry. Thus, it has capabilities beyond those of the conventisnal CFD codes, which, generally, employ only a single field. PM-ALP 11A is a very advanced and detallcd computer code, indeed. There are other codes, "urrently in development, in Europe, e.g. IVA (Siemens, Germany) and MC 3 D (CEA, hance), which are also incorporating similar capability, in order to treat the very complex, and very dynamic, physics of melt water interaction and steam explosions.

It is a general rule that more detailed the formulation for the description of a process, more detailed the information required to bring closure to the formula-tion; and more intuithely intelligent approximations have to be made to obtain credible solutions from the formulation. This is quite apparent for PM ALPilA, when a whole page (A 20) is needed, to show the dimensional groups that ap-pear in the constitutive laws for the fuel to coolant heat transfer. This can not G 102

p be avoided, howewr, the collective constituthe laws rnay provide reasonably.

G' correct predictions for a particular set of pre mixing circumstances, and not for another set. I believe, that verlRcation on an even less integral level than the h!AGICO experiments should be considered by thinking-through, and devising, a set of separate ellect experiments. They should be prioritised, so that the most important are performed Brst.

We agree with this perspective, and this is why we have gone to great lengths to test the individual pieces as well (see, for example, Figure 1 in the introduction; we suspect the problem here is that the reviewer chose to go through this report in reverse order, from back to front). For example, the one-page equations referred to here were obtained from an experiment, the MUPHIN, conceived and carried out specifically for this purpose. So, the question really is whether we have left out something important. This is addressed in responding to the reviewer's specific comments below.

Seh9. In the following paragraphs, I will provide some detailed comments.

!!.2.1 Phi ALPHA Formulsign The modeling appronch is logical and well thought. The authors admit that the formulation so far, emphasizes the multineld aspects of pre-mixing. The melt Jet and particle break up are trented parametcrically.

Two length scales are employed for the fuel Beld: one large, encompassing the original fuel drops, or fuel melt jet, which may creak up but still are considered as fuel; and the other small enough to be called a debris, which mixes with water and gets quenched. The decisions about the amount af the ' fuel' and the ' debris' are made with a correlation for the fragmentation rate.

The debris particles assume the same temperature, and velocity, as the coolant, instantly. They are not allowed to sediment down with gravity, as they would normally do. This assumption is justined for the time interval considered, if the particles are of micron size.

The " admit" carries a derogatory sense to it. In fact, we presented a rather detailed rationale of our approach, and we would prefer to have heard specific comment on it.

Seh10. The large length scale fuelparticles are assumed to have untform tem-paraiare. There is no treatment of the heat conduction from the fuel particle to the coolant. For the pwtotypic binary-oxide mixture melt, it is important to determine the solidlScation front growth into the partic!c, since it may either (j G 103

1 I

prewnt fragmentation, or reduce the rate of fragmentation, thereby changing the heat source to the coolant.

Another factor in the treatment of the fuel particles, is the change in physical properties that occurs, as the fuel particles cool down from above liquidus to below solidus temperature. The increase in viscosity and surface tension aIfect the fragmentation characteristics, which in turn affect the terms in the debris mass equation, and in the liquid and debris momentum and energy equations.

A paper submitted by Okkonen and Sehgal in the forthcoming FCI meeting in Japan discuss the two factors mentioned above for the behaviour of the fuel drops.

Particles that cannot brokup, certainly cannot support an explosion. In fact, we have an option in PM-ALPHA for crust growth, and we nave used it in the analysis of some ex-vessel explosions in deep, subcooled, water pools. This was not emphasized here, bxause the times and depths and other conditions are not conducive to significant solidification elfccts.

Sch11. Recently, we at RoyalInstitute oiTecimology (RIT), have performed some experiments on the interaction of relatively low temperature cerrobend (an alloy with density of r<9000 kg/m3) Jets with subcooled water. We have found that theJet breaks up into small particles. There is a distribution to the particle size or mass, however, there were no particles oflength scale comparable to the Jet diameter. In these experiments the Jet breaks-up completely. The FARO experiments show a melt " cake" at the bottom, however. It is not clear whether it is the unbroken jet or an agglomeration of melt droplets belonging to some size distribution, which, perhaps, does not contain length scales approacidng the melt Jet diameters.

Summarizing the above discussion, i l>elleve, the treatment of fuel as having two length scales in the PM-ALPHA formulation is valid. However, the source terms in the equations should be reviewed again. The wrintion of properties of the fuel drops, with temperature, should also be taken into account; and the change in the temperatute oi the fuel drop should be calculated employing conduction equations. Melt jet, or drop, interactions with subcooled coolant may produce atomization, with no large particles of size similar to that of the melt jet.

As noted, the source term in the equations are varied parametrically to bound the behavior.

No one knows what these source terms really are, and it is highly presumptuous at this stage to take an approach based on simulating breakup. This is not our approach. Solid-ification effects are not important for in vessel explosions, and even more so in saturated G 104

water pools. Moreover, ignoring this small effect is conservative. Finally, the atomization effect is accounted for by our breakup at two different length scales, but when this occurs to a very large extent it works against premixing.

Seh13.  !!.2.2 Int *rtwini Mnmentum 'h n=ler in PM. ALPHA The drag correlation used 'n PM-ALPHA for fuel-coolant interface distinguishes-between the dispersed and the dense fuel regimes. The latter is taken as that for Bow of gas through a densely packed t *. This correlation, perhaps, should be checked, since predicted penetration of we fuel cloud in the MAGICO experi-ments is less than the inessurements. Also, comparisons co'dd be made with the isothermal tests in the BILLEAU an ' the QUEOS fac111tles. The logic clagrams on pages A 16 and A 17 were helpful.

The densely packed bed regime appears only in particles accumulating against a bound-ary. We have no problem in the PM ALPHA predictions of fuel cloud penetration (see also above). Comparisons w)th QUEOS were provided in the body of the report. BILLEAU tests are not yet available.

Seh13.  !!.2.3 Interfacial Heat hnnater in PM. ALPHA There are many regimes of comecthe heat transfer and many correlations. The authors use the best that they can End. Then, there is the large eMect of radlatlon heat transfer, which was found to be important for the comparisons to the QUEOS test dats ' Their synergism, and eWects ofone regime on another, may need further exploration. For example, radiation-absorption will produce vapour which will change the comvetive Bow siterna of the coolant, and, perhaps, change the heat transfer regime. Some aparate-eWect tests could be designed to test the synergism and the eMect of diferent comecthe regimes on each other, in order to test the heat transfer correlations package employed.

We have explored these avenues already, but there isn't really much new or surprising.

Even for film boiling from single spheres, contrary to what one might expect, the super-position approach works very well. Also, it should be noted that our efforts here were not limited to collecting what we could Tnd. The major components are non local radiation heat transfer, and film boiling in single and two-phase media. For the former, we formu-

' lated a whole new approach, and for v .e latter, we conducted the MUPHIN experiments and developed theories and correlations for use in the code. So really we do not agree with the thrust of this comment.

O 105 e

Sch14. 11.2.4 hel Dreak Un and 1%sementation Modeline in Phi-ALPilA l

I have referred to this earlier in the comments on the PM. ALPHA formulation.

The laterfacial area equation (3.73) assumes spherical particles on break up and l fragmentation. This may not be appropriate. Perhaps, data from FARO or other fragmentation break up experiments could be employed to develop a more proto-typic interfacial area representation. In some of our experiments with cerrobend in subcooled water, we do not find spherical particles. Perhaps, in saturated water, with large flows of steam the particle shapes may be spherical.

The cerrobend particles are not spherical, because the material solidifies at such low tem-peratures! The dimension in Eq. (3.7) should be inte preted as characteristic length, or effective diameter. The source at this time is parametric, because there is no reliable model.

However, this is sufficient for our purposes.

15. The model for fragmentation of fuel drops is based on the Bond number.

I believe, data on hydrodynamic and thermal fragmentation of large-size melt droplets may be available in near future. The model could be ci ecked against such data, when available.

The model for jet and large fuel-drop-break up is parametric with an input-specllied parameter, f, whose value is varied in analysis. This approach is, per-haps, adequate for the present. However, it will be desirable to have a phe-nomenological/ mech ~nistic model.

Such models already exist, and probably others will be created. The problem is, how you propose to adequately validate them for reactor conditions.

Sch16. The authors distinguish between fragmentation and break-up as two separate processe?. In some of our melt jet-water interaction experiments, we were not able to separate the two processes. Thejet breaks-up (or fragments) into particles having a size distribution ranging from submillimeter to 3-4 millimetres.

The process appears to be concurrent and not sequential, as assumed in the parametric models described here.

Not at all. Our formulation is for concurrent, not sequential processes. One should be carefulin how far to take the cerrobend data.

G 106

+ .-

. r Seh17.' 11.3 VERIFICATION OF the PM. ALPHA CODE The Erkmt part of the teport DOE /ID 10504 describes the writication pursued for; the PM-ALPHA code by performing analytical tests, and by comparing with the _

)

data messured la sental experiments. This was' a very large etfort, and I believe, : [

lt has lugely achlewd its purpose. 'I will comment on a few comparisons of the ['

. data with the code predictions. )t 11.3.1 QUEOS Exoeriment These experiments are similar to the MAGICO experiments. ~ The comparisons  ;

shown in Figures 4 to 13 are remarkably good for such a dynamic process. The +

comparisons appear to be better than those for the MAGICO tests.

It is not clear to me what'the experimentalimage actually implies, in terms of ,

. the distribution of hot particles, and of void. The pictures in Fig. 7 at 0.3 and 0.4 seconds seem to show that the experimental hot particle image may be not as. ,

advanced as the calculated contour. This also appears to be the case in Fig 6. at 0.3 'and' O.4 seconds. The graphs in Fig. 8, howewr, show vesy good agreement between measured and calculated front-advance locations versus time.

. The QUEOS tests have provided only photographic images of the interaction zone, and

' ? (*

these are as clear to us as they are for the reviewer. As noted in the mport,' the front advancements in Figure 8 are from Lagrangian particles in the calculation. The Eulerian results show some numerical diffusion, and again, as noted, the results in Figure 7 were to be refined by calculating with finer grids. This is done now (see addendum to Section 2.2.3). ,

Seh18. 11.3.2 MIXA Exoeriments The MIXA experiments employ a Uranium Molybdenum thermite melt of several kilograms, at 3600K, poured into near-saturated water pools. The melt jet was

- broken into 6 mm diameter droplets. The MIXA-6, analysed here, used 3 kg melt pour into very nearly ($1 K difference) saturated water. This, thus, is a-1 prototypic experiment, albeit with small mass.

The comparisons are very good. I am somewhat concerned about the sensitivity of the results to the break up-cut off- void-fraction and particle size. The authors recognise this, still, a change ofonly 5 % (85 % to 80 %), with the particle size of1 :

mm, decreases the calculated pressure rise from 0.4 bars to n 0.2 bars. Increasing o.107 E .

., _ = - _w- . . ; -, u- a.

the particle size from 1 sam to 1.2 mm at the 85 Yi cut-off level decreases the pressure rise from u 0.4 to n 0.28 bars. Thus, the breakup and fragmentation w models appear to be very influential in the very high temperature, prototypic material experiments.

This is absolutely correct, and simply states the obvious fact that the steam production is a strong function of interfacial area and the couplings created thereof. This also pro-vides impor+ ant perspectives on the whole question of breakup - what can resonably be expected from a prediction, and how far could such possible predictions be taken!

Sch19. IL3.2 FARO Exneriments These are, perhaps, the most important experiments, since they use substantial quantitles (? 100kg) of prototypic materi-als; and there are several experiments already performed and more are underway.

The comparisons shown are very good indeed. Unfortunately, FARO does not produce any data on the mixing region, thus the colour figures, presented, st.ow only calculations and no data.

• * * * * * * * + * * * * * * * * * * * * *

• Sch20. I did not understand why the initial particle size is chosen as 4 cm for a jet diameter of 10 cm. The 0 value chosen is 50, while for the MIXA test it was chosen as 20. The minimum particle size chosen is 1 mm, while in the MIXA test is was chosen as 1.2 mm.

It is inconceivable that the jet exited the nozzle and travelled all the way to the pool, totally undisturbed. We chose 4 cm as a large enough characteristic length scale. It does not matter really what you choose, as the process is really controlled by #, and only small scales have enough interfacial area to interact.

Sch21. One experimental result, which FARO produces is the fraction of the jet material deposited as a ' cake' on the bottom plate. This is not provided by the authors from their analysis with the PM ALPHA code.

As the reviewer notes in an earlier comment, it is not really clear what this ' cake' means, and there are many ways to interpret it.

Sch22. IL4 Numerical Aspects The authors do not provide a discussion on this topic. I believe, this is an impor-tant topic. The ICE tecimique is kncwn to have significant numerical diffusion.

G 108

It is not clear whether any advanced space-time discretization scheme was em-Q( played. Node sizes of several centimetres are generally not Sne enough. The authots, perhaps, by now, have imestigated the numerical aspects further, and I would welcome a greater discumion of this topic.

There was nothing special employed, and elsewhere in the report we note that we may introduce such a special scheme at some future time. We now have PM-ALPHA.L. It was also noted that numerical diffusion can be well enough controlled for our purposes, by choosing fine enough grids. A brief sample of results were already included. More can be found in the addenda.

Tur49. Additional Comments on DOE /ID-10504: Phf-ALPHA Verincation Studies by T G Theofanous, W W Yuen and S Angelini INTRODUCTION This document represents the culmination of a substantial piece of work to de-velop a mixing code for steam explosion studies and to validate it against the experimental data. The report makes good use of the (still rather limited) ex-perimental data available for this purpose. The report concentrates on the pre-sentation of results rather than their evaluation, it would benent from a leading O,i- chapter on the philosophy of the verincation/ validation proceos, accompanied by a matrix indicating which of the code's models are tested, and to what extent, by the comparisons reported, it would further beneBt from a longer concluding chapter that draws together the results in the context of this matrix.

The philosophy of our approach is explained in Appendix A. Our verification / validation plan is shown in Figure 1 in the Introduction. Additional discussions as the ones re-quested here would draw us into judging the quality of the results, which we scrupu-lously avoided. We left this judging for the reviewers. As far as testing of the individual models, this is a good suggestion, and we will include such a table in the final report.

Tur50. It is noticeable that etforts are made to compare isothermal particle-water predictions with accepted correlations. There ought to be scope to include similar material on two-phase Bow in the absence of particles; this is probably more important in est.ablishing the reliability of the code to predict voiding be-haviour.

e

I Such tests can be meaningfully be done for the dispersed regimes (bubbly, droplet), and the particle cases considered are already quite sufficient for this purpose. The drag laws used for fluid or gas " panicles" are slight modifications of these and they are supported by wide data bases.

Tur51. While there are many detailed comments below, these should not detract from the achievement o[the outhors. The comparisons performed indicate \

that the code has the ability to make reasonable predictions for reactor conditions. j However, the results should still be used cautiously, as the data currently do not exist to provide full validation of the model.

Tur52. Specinc Comments Chapter 2 Sineel particle settline While tracking a representative particle in a Lagrangian fashion ghts the expected analytic result, melt mass is usually tracked through the volume fraction. This can be much more diffusive.

These were simple tests maue to begin with. Fc numerical diffusion see below.

Tur53. Settline of oarticle clouds I have tried tc check the consistency be-O tween the drag law for particles given by equations 3.14, 3.21 and 3.22 of Ap-pendix A with the drift Bux formulation, but have been unsuccessful. There appear to be inconsistencies be: ween equation 2.4 and Figures 2 and 3. Taking

% = 0.487 m/s, gives the liquid superficial velocity for a = 0.5 as 0.093 m/s.

Figure 2 shows this as 0.12 m/s, while Figure 3 indicates 0.19 m/s. This suggests that it is not the superficial velocity that is being plotted in Figure 3 but the now velocity, which would be 0.186 m/s from equation 2.3. hiy evaluations of the drag coeRicient given in Appendix A for this case give a relathe velocity of 0.286 m/s, or a superficial velocity of 0.143 m/s. However, Phi-ALPHA has produced, according to Figure 3, a value close to 0.2 m/s. hiy hand calculations indicate that the Pht-ALPHA modelis not as close to the drift Bux model as implied by Figure 3.

The closeness in Figure 3 is correct. The conf tsion was generated because there was a mislabeling of the vertical axis in Figures 2,3, and 4. It is the velocity, not the superficial velocity plotted. Also,0.186 is in good agreement with 0.2, isn't it? Finally, we found G 110

~t also a slight plotting error in Figure 2 regarding the drift flux line. The correct figure is .

t attached. 4 0.. . . . . . .

drift fluu model x< - x ~ PM. ALPHA trutial conditions 0.5 -

x -

x- x

~

0 .4 - x g x- -

}k x. ,

x x * ,

- * *

• x -

0.3 x .

x x * '

, x *

• x O.2 -

x x ,

x

• x x x x x 0.1 -

x * ~

x x x x x x

• x x 0

O 0.1 0.2 0.3 0.4 0.5 - 0.6 0.7 Particle Volume Faction Figure 2. Initial Conditions in PM ALPHA simulations.

Tur54. Settlins of oarticle clouds The comparison witt. he drift flux model is clearly important as it goes some way to establishing the reliability of the

- drag coeBicient modelling in PM. ALPHA (although it should be noted that the particle volume fraction is unlikely to exceed 20%, where the enhanced drag due to particle particle effects is not that significant).- It is less clear what one is expected to learn from the material presented on transient analysis regarding the validity of the code's models. It uould have been usefullastead/in addition to perform the same comparison with the drift flux model for gas-water interactions where the form of the drag coefficient is rather different.

Several reviewers found these results very interesting!

Tur55. Section 2.2.2. MAGICO exoeriments It would have been useful to

. han a short synopsis of the conclusions drawn about the model from the analysis

. of the MAGICO tests. Besides the qualitative agreement (and general quanti-tative agreement) on the nature of the interaction. I think the most significant l linding is the prediction and measurement oflarge vold fractions (greater than p.

)

' U. .

o.111

l 709E) lilustrated in Figure B23). It would be useful to provide a statement on the speclRc code models that these observations'are believed to validate (eq water-steam drag, Rim boiling, radiative heat transfer??).

All play a role. Thece are integral comparisons carried out for a wide range of conditions (see also addendum to Appendix B), and the results speak for themselves, We don't think it is very fruitful or even appropriate to assign significance. Note that each model stands on its own merits, and these comparisons show that when put together the result is very consistent with reality. This should not be too surprising.

Tur50. The OUEOS Experimerds: This looks a very interesting analysis of these tests. The presentation of results in Figure 4 etc gives an excellent way of qualitathely comparing code results and experimental obsen'ations. Perhaps some comment should be made about the apparently coherent release oflarge gas / steam volumes, seen eq at 0.41 s in Figure 4; also on the water spout effect predicted at this time (this seems to provide the mass difference between hieyer's interpretation oI the water fraction in the mixing region and the Phi ALPHA values). The acceptability, or otherwise of numerical ditfusion, is a complicated matter, because of non linear udbacks through the drag laws; it is very easy to underpredict the peak particle volume fraction. Figure 5 does not give units for the liquid Bux. Condensation in Phi ALPHA looks too effective at later times in Figure 6 compared with the experimentalimage.

The liquid flux units are cm/s. We have done calculations of QUEOS runs with I cm grid and obtained very good agreement with the results presented previously. Also see Addendum 2 to Appendix B with PM ALPHA.L simulation of MAGICO tests and the Addendum to Section 2.2.3 with simulation of QUEOS tests.

Tur57. Chapter 3 Comparison with CHnfES: It is only fair to note that this comparison was only possible by turning off sub-cooling in Phi-ALPHA. hiuch of the detail of the Pht-ALPHA predictions depend on the modelling of sub-cooled boiling. The observation that Phi-ALPHA often only produces any void somewhat behind the particle front, whereas other codes tend to produce some voiding wherever there are hot particles can have signincant implications on the initial Bow of water. For instance, we did not reproduce the so-called ETHICCA etlect with CHniES. In addition, CHYhfES drag laws were modined for the comparison. However, the G-il2

4 main result a water depletion is predicted by both codes (at least for low pressure

( systems close to saturation temperatu.m)- is robust.-

Tur58. The MIXA Exocringag I will try to clarify the question of time origins for the data. The experimental report, which I have, has unequivocal timings, with an origin starting at the ignition of the pyrofuse for the thermitic reaction. On this timing the melt Brst contacted the water at 3140 ms, the peak (measured) steaming rate was at 3810 ms and the peak pressure occurred at 4215 ms. The authors have adopted a timescale (their Rgure 4) where the time of Brst melt contact is taken to be zero. This is the same timescale used

. in Figure 1 of Fletcher and Denham for the measured pressure in the gas space

- so the comparison given for pressure in the top frame of Figure 6 (page 3-

21) is correct. However, the transient steaming rate Bgure (Bgure 8 of Fletcher and Denham) does not use this time basis - this is because it was derived from the CHYhfES calculations with the experimental data over-plotted). There is a significant outBow of gas before the melt reaches the water surface as shown in this Bgure. This may be due to (1) preheating and expansion of the gas in the test vessel; (ii) evaporation of a water Blm on the test vessel wall (the favoured explanation for similar observations in FARO), and/or (ill) evaporation from the water surface. The experimental data on the middle and lower frames of Figure 6 should therefore be shifted to the left by about 0.32 s (error on this is only from nty reading of the graph in Fletcher and Denham - it is no more than 0.02 s). The effect of this is to move the measured peak steaming rate ahead of the measured peak pressure. However, I now believe that the measured steaming rates become increasingly unreliable (as quoted) due to carry-over of a tuwphase mixture; similar behaviour has been observed in PREhflX. Unfortunately, while the experimenters noted water carry-over post test, and observed a reduction of water height in the vessel post-test of 25 mm (the measured steam svuld produce a reduction of only about 4 mm), there is no information to determine how much

^

of this occurred because of eraporation during the heating 0f the water. The same comments apply to Figures 7 to 10.

While the Phi-ALPHA calculations are as good as or better than any I have seen for hHXA-06, I am not convinced that the real behaviour in the test is being captured. The most noticeable features are the radial expansion of the melt as it G-113

enters the water and the apparent lack of any visual record of droplet break up.

Both of these effects seem to be connected with sudden expansions of the melt region, due to enhanced steam generation, giving much more coupling between melt and steam that accounted for in CHYh1ES, and, by the look ofit, in Phi.

ALPHA. I conjecture that droplet fragmentation is occurring during these rapid events. The formation of smaller particulate then encourages another process of melt spreading. Smaller particles are carried upwards by the central steam 1

flow, move outwards, and fall in the periphery, thus extending the melt envelope outwards. j There is no visual evidence of the predicted extenske voiding atound the melt region - the leading droplets appear to be falling through water - the steam ,

generating region is large because of the spread of the melt droplets.

I agree on the sensitivity of calculations to assumptions on break-up. Has the predicted mean particle size been compared with the experimental value of about 3mm7 This section should contain discussion / conclusions on implications of the com-parison for model validation.

With so many uncertainties in the test how can we reasonably draw cenclusions? In any case the additional information and insights provided by the reviewer are very helpful in our further interpretations of MIXA with PM ALPHA.L (see addendum to Section 3.2.1).

~

Tur59. .De FARO Exr>eriments: Clearly the initial melt droplet size is very uncertain, as is the spread of the melt. L-14 appears to be the test in which the melt stream was best collimated, but one cannot tell whether the stream contracted as it poured through the gas space, or underwent a mild expansion (in L-11 the melt stream appeared to undergo a major expansion), Ifit is believed that the melt jet contracted (note typo: steam for stream 4 lines from end of page 3-25), then the radial meshing with ar = 5 cm is too small. The choice of break-up parameters appears arbitrary - presumably these were selected to give reasonable agreement with the experimental data. blore detailed modelling of the melt release vesselindicates that the melt exit velocity was close to 3 m/s for most of the pour; this will not be replicated by the model shown in Figure 2. I am surprised that a Weber number criterion did not limit the droplet size; with the CHYhlES implementation of this criterion we almost always get mean particles G-114

close to those obsened in experiments (typically 3 - 5 mm). The comment on

) the absence of significantly superheated steam in the experimental data seems to me to be special pleading - it might be right, or the steam Bow might be much less concentrated on axis than predicted by Pht-ALPHA, giving steam closer to saturation conditions. It is diBicult to relate the scales on the coloured contour plots in Figures 10 and 11 ta the colour scale, particularly because ofinterpalation effects. Is break-up still occurring after the particles have settied (unless they have solidilled)?

No, actually at this stage reagglomeration of any inadequately solidified particles begins.

In relation to the other comments see further interpretations with PM-ALPHA.L (adden-dum to Section 3.2.2). ,

TurGO. Again, this section should be supplemented by an evaluation of the implications for the reality of Phi ALPHA predictions. I think a word of caution is necessary, as although Pht-ALPHA, with the assumption used, performs well against experimental data, it predicts a highly two-dimensional configuration.

Alternathely, good comparisons against the data have also been produced with the one-dimensional code, TEXAS-IV. Until we see the nature of the interaction t' zone (I expect it to be between these two computational extremes) then it is not O possible to say that one simulation is better than the other.

We do not think it would be appropriate for us to shed vague doubts on the validity of the PM-ALPHA comparisons because the ID TEXAS-IV was made to produce good comparisons too. The last sentence is very puzzling. What is intermediate between ID and 2D?, or is it 1-1/27 When a behavior is not ID, it can only be 2D (or 3D, of course, but this is not the issue here). When a behavior is not ID, a 2D computational framework is a necessary starting point, before one begins to examine any further the degree of

" simulation" obtained.

Tur61. Chapter 4 I agree with the general comments on break up modelling. As implied in ny comments abow, backing out break up behaviour from the experimental data may compensate for other errors in the modelling. As I also noted, it is unclear, enn with the visualisation, what break up processes were occurring in hilXA; I suspect the processes are much more dynamic than are currently embodied in the .

models, and coupled strongly with events of enhanced steam generation (coolant 77 mi G-115

(

trapping?). New FARO tests with visualisation should provide information on the coherency of the inillal pour, besides evidence of any subsequent break-up.

• w * * * * * * * *

• Tur02. Numerical aspects Our experience is not as comforting as that presented by the authors. I think that numerical diffusion is ptobably not an important issue for large-scale mixing calculations. However, it becomes important in comparisons with smaller-scale experiments, which are often dominated by leading edge effects. Numerical com-parisons that we have performed (external to CHYMES) show that upwinding achemes run below the material Courant condition lead to very poor predictions of peak particle fraction, and thus drag. Higher order schemes have to cater for possible discontinuilles at the leading edge. Lagrangian approaches, as used by the authors for their front tracking, provides much better accuracy, both for re-locity and peak volume fractions. I believe that current schemes in the mixing codes can be improved substantially using physically based Lagrangian limiters, rather than mathematicallimiters. Fully Lagrangian approaches have she greater benefit of handling a spectrum of particle sizes. This may be the best way to treat Jet break up and is necessary if one is going to capture the role of the smaller droplets in spreading the melt, as observed in MIXA-06.

The cur'ent presence of numerical diffusion makes the code results difficult to interpret (eq how far back is the predicted peak concentration from the melt leading edge in the MIXA 06 calculations?). Our experience with more refined meshes is that numerical diffusion is indeed reduced, but the calculations are much more prone to instability of the resulting interface; this numerical instability probably reflects the actualinstability ofinterfaces obsened in experiments.

This whole discussion reflects the reviewer's own experiences with CHYMES, and it should not be confused as being applicable to what we have presented. Our results should be judged on their own merits, and there are ample comparisons to allow one to comment directly on the present experience. We find no instabilities in fine-meshed calculations, and they agree with our new PM ALPHA.L results.

Tura3. Concluding Remarks I would expect a more detailed technical evaluation of the calc 11atiores presented.

I am surprised that questions related to the radiation transport modelling, which G-116 1

was clearly important in the FARO simulation, have not been highlighted. I-Q uvuld have liked to see more explicit bounds on models emerge form the work.

- We wanted to keep our concluding remarks brief and to the point. If we were to provide highlights in this section, we would have to include much more than the non-local radi--

ation model (actually it is even more important in MIXA, as discussed in the respective section). The last sentence indicates a degree of dissatisfaction, we think, with how far, quantitatively, we could go based on the comparisons. This is a question of judgmet, and one thing for sure is that we do not wish to oversell what we have been able to show. We believe the comparisons, and the whole array of situations considered, speak for themselves, and adequately loud to be understood, both in their success, as well as their limitation by those involved in this kind of work, and at this state we are satisfied with that.

Tur64. Phi ALPHA hiodels The details of the correlations embodied in Phi ALPHA will not be reviewer in detail.

I believe the modelling approach is sound. Inote that reactorgeometries may im-p pose strongly three-dimensional Bow regions, so a 3-D code is needed for detailed applications (if found to be necessary). Iget the impression that the modelling philosopig falls between two stools. At places it is admitted that the model nec-essarily contains many simplincations and constitutive physics that is uncertain, but only in the Beld ofJet breakup is a parametric approach used. I would prefer a broader approach to treatment of uncertainties.

This comment forgets that we have indeed a 3D code. We did our best to catch and present a correct picture regarding uncertamties. If the reviewer has particular aspects beyond those presented that he would like to see, we would be happy to undertake the computations needed.

Tur65. With sub cooling implemented in CHYhfES, it is closer in concept to TRIO-hic rather than Ph:-ALPHA. (EVA should be spelled IVA).

Tur66. Elsewhere we have queried the use of the drag coenicients for droplet and bubbly Bow. These are derived for bubbles rising at terminal velocity in a gravitational Beld. It is found that the shape factor for the bubble causes the

,3 b G-117

drag per unit mass of gas to be independent oflength-scale. It is noticeable that no effect of melt droplet shape appears in the corresponding formula for drag '

coefficient for the melt phase (equation 3.21). A completely dilferent form for the liquid vapour drag is used for intermediate values of void fraction; this may ghe large changes in drag uhen the transition void fractions are crossed. It is not evident that there are such sudden changes in flow regime in plenum geometry.

The intermediate regime is very important in gas-liquid flows, because it indicates an initial " break" in the liquid-continuous regime. It is well known that this leads to a suddenly much greater slip, and this leads to churning. For drag coefficients we used the best available correlations. In all our experience with these models we find no reason to raise significant questions to them.

Tur6*l. I have not had the time to consider the radiation treatment in detail; also the relevant appendices are not included in thc exerpt. For dense clouds of particles, the sel( absorption etlect will be very important. I would like assurance that this does not allow the region to emit more radiation externally than that of a black body covering its surface at the same temperature.

As explained in the text, self-absorption is included in a consistent manner. The material in the appendices referenced in this section pertain to detailed numcrical procedures in evaluating integrals and is really not essential for such a review.

g G-118

n Propagation (Reactor)

Dan 9. The calculations of melt length scales and local void fractions lead to quantitative results which are more realistic and detailed than previously avull-able. As expected, liquid water is rapidly depleted hom regions of high fuel con-centration, and the boundaries of such regions can be quite sharp. Board-Hall thermodynamics theory for steady plane shocks, previous multiphase calculations of the flow Belds behind the shock front (Sharon end Bankoff) agree that regions oflarge void content cannot sustain shock propagation at supercritical pressures.

This is the principal reason that the SERG-2 panel felt that the a-mode failure was not physically reasonable. Precisely the same results are obtained by the PM Alpha and ESPROSE.m calculations.

Dan 10. The residual uncertainties proposed in NUREG-1524 wereJet breakup triggering, 2D vs. 3D codes, and chemical augmentation. For the AP-600, Jet breakup is no longer a major concern, as discussed above. The 2D vs. 3D contro-versy is no longer relemnt, since validated 3D codes are now available. Chemical augmentation with the real corium produced in the reactor will have no impor-(~)

\' tant elfects. 'IYiggering is the sole intangible which will never be known for a real accident. However, it is irrelemnt ifit is assumed that triggering always occurs at the worst time and place, and the result is enluated by energetics, which is to say the validated codes. The appmach taken of triggering by setting one mesh to a high initial pressure seems to me to be a perfectly valid procedure.

Der 2. I will now go through the different chapters trying to analyze the justl6 cations which are presented to support the crucial arguments mentioned previously.

Chapter 2: Problem de&nition and over all approach e It is mentioned that it is only recently that pressures in the kbar range were obsenrd experimentally in constrained one dimensional geometry. Houever, I think that a pressu s peak of the order of the kbar amplitude and millisecond duration was measured in the Sandia FITS-RC2 experiment which was well vented V G il9

(initially open at the top and later vented at the bottom as the vessel left the ground). But this was obtained using iron alumina thermite and subcooled water.

We are aware of *Se experiment, and its energetic natures (from movies), but we do not have indications in the literature of reliable pressure measurements.

Der 21. Chapter 6: Quantificatian of explosion loads Nothing is said about the parameters used in ESPROSE-m 3D but as the trigger uses a 100 bar steam release, we may think that the hydrodynamics fragmenta-tion law will be correct Due to the small amount of melt imolved in explosion calculations, there is no problem with the energetics of the explosion and we are only interested in dynamicalloadings of the RPV. This is done by the estimation of the impulse and of the local area of loadings from ESPROSE-m results. I have some problems to understand how fag si estimated page 6.3 from the aren evolution as shown on lig. 6.5.a.

In the text, it is said that peak impulses are around 0.1 and 0.2 MPa.s with cffecthe area around 0.1 m' (which gives f9 ~ 0.15 ) and from lig 6.5 c where I find a 0.2 Mpa.s impulse, I do not understand how I get Ao ~ 0.1 m' from the area evolution which is shown.

We take the area over which the main hnpulse is delivered. For example, in Figure 6.5(a),

O this occurs in the time interval from 03 to ~0.5 ms. Over this time interval the area is then seen to be ~0.1 m2 . Then using the ecuation on p. 6-3 we find do/D, ~ 0.16. These estimates of areas and effective times are presented in a more refined manner in the new calculations.

Dur19. 5. Comments on Explosion, together with Premixing A trigger of sufficient strength is applied to the mixtures in ESPROSE calcula-tions to quantify explosions. The chosen trigger appears to be sulliciently strong to produce strong escalations as in the KROTOS experiments, but its strength is not assessed with respect to possible trigger strengths. I agree that with a sulliciently strong trigger the escalation dynamics may no longer depend on the trigger strength (if owrdriven cases are excluded as unrealistic). Thus, the re-suits of the numerical tests may indicate such a limiting strength and no need for further variation. In view of the effects in the KROTOS experiments, the chosen trigger can also be considered as strong enough to yield major effects. Looking G 120

p at early rather than later times for bounding the cffect of trigger timing appears

() also appropriate in view of the strong voiding (excluding other possibilities of melt release as discussed in the previous chapters).

Bur 20. The results given in the report show sigr.lScant differences in the maximum local pres. cures, the maximum impulses as well as loaded areas and times ofloading depending on the chosen melt mass Bow, the breakup behaviour and the time.o[ triggering. E.g., with the higher melt now a maximum pressure

> 5000 bar, impulse 100-120 kPa s and maxhaum area of 5 m' results choosing 0 = 10 for fragmentation and an instant of triggering at 0.05 s (case C2-10(0.05)),

as compared to nearly 104 bar,100-200 kPa s and 3.5 m* for 0 = 20 and 0.12 s (case C2-20(0.12)). On the other hand, there seem also to be similarities or bounding trends. E.g., for case C1-10(0.05) with the lower melt Bow, f = 10 and 0 05 :, the results are rather similar to C2-10(0.05), with somewhat smaller impulses and areas in C1-10(0.05). But small shifts in trigger time give also strong diIferences, e.g. pages C.3- 16 to C.3-20 in the report for case C2 20. The same is valid for the comparison of C2-10 and C2-20 with similar trigger times r (pages C.3-13 and C.3-16). This is certainly due to the relation between time development of breakup and voiding, producing optimum mixture conBgurations at different tima. This is one cause of uncertainties in getting exploshe events or not (together with triggering time).

Concerning this problem of sensitivity, the large number of calculations performed is convincing. They yield maximum events but in a limited range and not as singular cases. Some questioning Istill have with this respect concerns the choice of f for premixing breakup and perhaps the underlying time law of breakup (not given in the report). Since the maximum loads appear to be obtained with case C2-20(0.12) as compared to the lower D, it is not quite convincing to jump tc nb and not to consider cases between. Other time dependences may yield further variations. This concerns the questioning of above concerning the premixing process as well as the melt Bows.

Point is well taken. See addenda to Chapters 5 and 6 for further insights into relating premixture characteristics to resulting energetics and for consideration of intermediate cases. Through these interpretations a more coherent picture regarding the origins and implications of the relatively narrow " explosion-sensitive" region can be gained.

,,)

(

- c.-i 2 i

Dllr21. Concerning the latter point, it is ta be remarked that the main eEcct of multiple melt streams into the water - if taken as saturated would be that a larger region is loaded by the explosion (perhaps also some further escalations in more extended premixtures may be possible, this could be checked by ESPROSE calculations) and that thus the venting will be further limited. Also the pressure relief in the vessel wall will then be limited. Thus, it is important to further confirm the exclusion of such multiple events (small windows for this!) or to check the coincidence eHects.

Actually, our pour characteristics correspond to multiple melt jet streams, rather closely spaced within the lateral space dimension assumed in the meltthrough. To widely separate these jets would be tantamount to assuming a much larger azimuthal coherence, which as discussed above, we do not consider appropnate.

Cho6. Based on the code calculations performed, the report concludes that the saturat4 coolant condition in the lower plenum leads to highly voided premix-tures that have a dampening eEect on the resulting explosion energetics. While I am not judging the validity of the calculations, I find it difficult to reconcile thh conclusion with availabic cxperimental evidence. Experience tells us that trigger-Ing of a steam expicsion would be more difficult with saturated water than with highly subcooled water. However, once triggered, the explosion energetics does not seem to depend on the coolant semperature that much. Consider, for exam-pie, the results of the KROTOS tests Nos. 28, 29, and 30 [H. Hohmann et al.,

"FCI Experiments in the Aluminum Oxide / Water System," Nucl. Eng. Design 155 (1995) 391 403). In these tests, approximately 1.5 kg quantities of Al2 0s melt at 2300-2400*C were pou.~ed into a column of water and steam explosions took place in KROTOS 28, the water was nearly saturated at 87* C while in KROTOS 29 and 30, the water was highly subcooled at 20*C. The energy comtrsion 2atio was estimated to be 1.3%, 0.8%, and 1.25%, respectively, for KROTOS 28, 29, and 30. It thus appears that the explosion with the nearly saturated water was at least as energetic as those with the highly subcooled water. Similar findings regarding the cEcct of water temperature on the explosion energetics were also made in our recent ZREX experiments. Such experimental evidence would need to be considered when discussing the explosion energetics.

There is a major misunderstanding here. The top sentence implies that with saturated water we get no energetics, which is clearly incorrect. Moreover, all KROTOS tests were G-122

a a

subcooled. Due to the peculiar mixing condition, timing, rnd relatively low temperature

. (compared to reactor materials) all three produced premixtures relatively low in void.:

The distinction made here is not between 10 and 30% void fractions, but void fractions-

-going'to 50-90% Does the reviewer disagree that such large void fractions will have a dampening (not an eliminating) effect? _ Empirical considerations, on conversion ratio, .;

such as those mentioned here by the reviewer, cannot get us too far one way or another. j

- It should be clear, however, that our me'hodology t is consistent ,with the findings of the }

. KROTOS experiments, as shown in DOE /ID-10503.- I

. Cor12. b] Why is the impulse largest for the mid-range value of ' beta'? Is the' knpulseof 200 kPa-sec near the failure limit 7 or am I reading this prediction correctly?-

Again, this relates to the explosive quality.of tlie premixtures, as explained in the adden-dum to Chapter 5. Not at all. The 200 kPa.s case is not near failure. This is'because the impulse is highly localized, as explained alreae/ in the report. Also, as explained in Chapter 7, for the actual loading, the vessel remained within the elastic limits. .

. Cor13. c) The detailed calculational results in Appendix C abruptly stop in

~

many cases at 1 or 2 or 3 t. Illisecunds. Why? Is this an indication of something

• ;h 2

numerically fatal in the ESPROSE.m simulation or what is up?

Again, time limitations caused us to curtail some calculations that were not very interest- j ing. More complete results can be found in the addendum to Chapter 6.

Fau2. Steam Explosion Loads - Having eliminated the potential for global vessel failure, Theofanous et al. proceed to evaluate the potential for localized damage, by considering local shock loading, with peak amplitudes in the Kbar range as a result of a steam explosion occurrence. Again, the conclusion is that fa3ure is " physically unreasonable". This conclusion is further supported by .

1 noting the following observations.

The above loadings are produced by subjecting the ihniting premixtures at atmo- .

sphnric pressure to triggers resulting from releasing steam at 100 bar. Quoting the authors,I"our triggers are chosen sufficient to initiate explosions, and they have

.no relation to what might arise spontaneously during a pour." We agree with

[ this observation, and in fact believe that the occurrence ofspontaneous steam ex-

' plosions with the molten corium-saturated water system at atmospheric pressure considered by Theofanous et al. is " physically unreasonable". The enormous liim G-123

. u . ;-- . -__ .u . _ __- _ , _ _

.. a : -, _ _ - .

-. a. , . _ _ , ,

1 I

boiling heat flux (~3 hfw/m') and corresponding vapor Bux resulting with this

. system (several times the critical heat flux of ~1 h1w/m') promotes separation and prevents plo sical contact between the molten corium and water, a prerequi-site for steam explosion ghen a fuel-water pre mixture. Temperatures (~200T C) which are well below the melting temperature of corium (~2700*C), would be re-t qtdred in order to reduce the vapor Bux in connection with B1m boiling to fall below the Buldization vapor Bux. The above considerations are consistent with the noted absence of "explosivity" for the corium-water system (I. IInhtiniemi et al., "FCI Experiments in the Corium/ Water System", NUREG/CP 0142,1712-1727, 1996). This is in sharp contrast to the noted "explosivity" with the often used molten alumina (Al 2O3)-water system. Here the estimated B1m boiling va-por ilux (~0.5 biw/m') is well below the Buldization vapor Bux allowing physical contact while the alumina is still molten. While the noted efficiencies are quite low, the super critical pressures observed with the alumina-water tests in the KROTOS facility (Hohmann, H. D. et al., Nuclear Eng. & Design, 155, 391-403, 1993), apparently encouraged Theofanous et al. to model such events and apply them to the LWR system.

2 In fact, for the A(2 0 3system, at 2700 K, just the radiative flux is ~3 MW/m ,

Fau3. We also reviewed the approach taken to assess the steam explosion cre-ated impulsive loads. Certainly the ellotts performed by the authors are impres-sive in the number of analyses performed and the detailed graphical presentation of the results. The calculations are based on the microinteractions model which appears to be applied in a self consistent manner. Our question is whether this is the only way that the relevant experimental information can be interpreted.

The underlying supposition in such models are that dynamic fragmentation and intermixing of melt and water can occur on an explosive timescale. Certainly the available information shows that fragmentation can occur during an explosion.

However, we are not convinced that the elements associated with fragmentation and intermixing on such a rapid timescale have been demonstrated. In particular, the SIGhiA tests performed with molten aluminum indicate virtually no fragmen-tation for melt temperatures where numerous large scale studies have observed exploshv events.

The relevant materials here are Fe and/or UO 2 /ZrO2 . We have data now with Fe, and the behavior is quite different from that of At. See also addendum to Appendix C of DOE /ID-10503.

G-124

Fau4. Furthermon the detonation concept is compared to the KROTOS

() molten tin water and molten aluminum oxide water tests. With the agreement from this comparison, the authors, in a previous DOE report (DOE /ID-10489),

conclude that " low void fraction geometries can produce highly supercritical, en-crgetic detonations." Our analyses show that therc is an alternate explanation to the KROTOS experiments that requires no melt fragmentation. If this 's the case, the comparison of the mictointeractions model with the KROTOS experi-ments indicates nothing more than that ESPROSE approach is consistent with the experimental observations. It does not providejustincation for the micrcint-eractions physical concept, The microinteractions concept and the SIGMA derived law are basic information obtained with relevant materials under fully-simulated large-scale explosion conditions. This basic information cannot be doubted. KROTOS data and our interpretations of them with these laws indicate the consistency of these data with large-scale explosions, not the consistency of the ESPROSE.m approach with these data. Any other interpretations (we are not aware of such) need to show consistency with the SIGMA experiments also.

FauG. Before such a fragmentation approach can be recommended for real-istically assessing the strur*ure loads, it should be proven that a relatively small (V~} pressure increase would be suBicient to sel(-trigger a coarsely fragmented and in-termixed system. In particular, it should be demonstrated that a coarsely mLsed system could escalate from a small triggering event into an event like that char-actenzed in these evaluations. Assuming that a single grid is Biled with steam at 100 bars as a triggering mechanism, it is far too coarse to provide such a deRnithe representation.

Triggering was not addressed in this report, as was made clear already. This should not be confused with assessing the energetics in a triggered explosion. Since we are not aware of definitive arguments that steam explosions are impossible with reactor materials, we provide here an assessment of energetics, assuming not only that an explosion can be triggered, but also that this occurs at the worst possible time, during premixing.

Jac16. 3.3 Modeling of explosions Tim most important Roding of the calculations in this area is the cutoff that occurs at higher void fractions. Hmvever, the model used to describe explosive interactions - the c.ictointeraction model - has been developed ~ 1 the basis of G-125

experimental observations in a situation with virtually zero wl ding. The param-eters of the model have been fixed using these experiments and it has bwn shown that the model can be made to give results looking reasonable (by proper partane-ter choices) by simulating a KROTOS experiment in which the local void fraction was assumed to be betwwn 25 and 40 %. h has been the dedared purpose of the mictointeraction model ta explain the occurrence of strong pressure lucreases in the presence of large amounts of water (Iow fuel to water mass ratio). And l I

ns such it is highly interesting from a scicatlBc wint of view and may be very i

relevant - in this special situation. But or.e cannot expect this same model (wit %

the same parameter acttings) ta work properly in a completely difTerent situatio 2 in whids there is very little water present. The failure of this specialinteracuon model to predict strong steam explosions under conditions for which it wasn't designed does not necessarily say anything about the occurrence of steam exph>-

sions in situations as suggested by the premixing calculations should these ever occur. Especially in the case oflarger melt masses (and possibly smaller overall void fractione) the lower plenum of a pressurized v:ater reactor might provide enough external confinement for completely different interaction mechanisms to become effective. These mechanisms may need more tim = for their development but might in the end arrhc at similarly effective interactions, An impartant example of mechanisms that may contribute to such alternate types of interac-tions are the thermal fragmentation mechanisms that may not need much water and are completely left aside in the present study. This might exphdn why the most efficient explosions are obtained very early (prior to 0.12 sec) followed by much less etlicient interactions at later times in all cases with a Enite breakup parameter.

Here we are interested in highly supercritical multiphase thermal detonations, as only they can challenge the lower head. Such detonations cannot occur in the presence of large void fractions for many reasons that are well accepted (this is not a unique ESPROSE.m finding). It is not clear what "new type" of interact:an the reviewer speculates about, but whatever it is, it is clearly of no interest here. Also, the reviewer's scenario for collecting a lot of melt in the lower plenum without removing the water in the process is not consistent with everything we know about mixing, settling, and voiding.

Jac17. The picture is less clear in the cases in which additional breakup was assumed not to occur. As outlined in the previous section these might be the G-126

p most interesting casse in this study. Here no clear maximum of exploshity has

() bee.1 found among the cases considered and it is argued that 'slightly broken up premixtures remain very benign.' However, Table 0.1 shows that in the case C2-nb the, maximum peak localimpulse is 30 kPa.s which may already be viewed as a low to intermediate value and that it occure at the last trigger time considered, La. ? O sic. Nothing In the results presented supports the idea that the value might not be larger (and maybe important) at even larger triggering times.

See further calculations provided in the addenda to Chapters 5 and 6.

May4. 4. QuantlBcation of explosion loads There arc two key phenomena inBuencing loads of steam explosion. These are

- the mixing of particle clouds plunging into water and

- the mictointeraction between water and melt.

The Brst phen >menon was diccnssed in the chapter before. For describing the microinteractions between melt and water, the authors followed two ways. For describing the microinteraction and for simulating the propagation of steam ex-

. plosions, they used the computer code ESPROSE.m. This code is based on a O series of experknents - the second parallel way - which were performed in the so Q

called SIGMA 2000 facility /2/. Originally the formulations for the microinter-action were based on the assumption, that the rate of coolant mixing between debris and water is proportional to the melt fragmentation rate. This is a reason-able assumption and by this it was possible to produce consistent comparisons from available experiments for a wide range of steam explosion loads, starting from weak propagations to supercritical detonations. The Brst formulations were mainly based on experimental results, obtained in the KROTOS facility. This Brst formulation was done for two-dimensional geometries and could especially also demonstrate the mitigating effect of " venting", due to wave reBection at a free liquid surface. Supercritical detonations were observec the KROTOS fa-cility with aluminium oxide melt only, pouring at very high emperatures into water, in a next step, the constitutive equations were, assessed by using experimental re-suits, obtained in the above mentioned SIGH!A 2000 facility. These experiments were carried out with molten tin drops, having temperatures up to 1800 C, im-pinging into water. Of course one can argue, that there are scaling effects, if one

( )-

V G-127

s:

I I

i wants to draw conclusions from the measured and evaluated data, gained in this small experimental set-up, to the steam explosion loads to be expected during a sewre accident in an APG00 reactor. According to the reviewer's opinion, these scaling problems however are mainly with the mixing of particle clouds, plunging into water, a problem which was discussed in the chapter before and which was solved by the authors with the help of the computer code PhiALPHA.3D.

The SIGMA 2000 facility was experimentally very well equipped and special measuring techniques, like radiography, gave very good quantitathe information about the fragmentation of the drop mass and its distribution. The fragmenta-tion, measured with X-ray flash, was reproducible wtthin less than 207c, which is a very good accuracy for such types of experiments. In addition the frag-mented melt was collected after freezing and was subjected to sieve analysis.

Very fine fragmented particles were analysed via scanning electron microscope photograp'as. Generally speaking these experiments are a very reliable basis for assd:.g a computer code like ESPROSE.m 3D, according to the opinion of the ret iewer.

In the SIGMA-2000 facility, not only the fragmentation rate, but also the pres-sure signals of the steam explosions were recorded by using high speed pressure transducers. Due to the small scale of the facility, these pressure signals may be conservalhe when applied to a large scale geometry, like the downcomer or the lower plenum of the AP600. In a large volume, in which fragmentation of a hot melt starts, there are always voided areas, damping pressure propagation.

The verification of the ESPROSE.m-code is very well documented in the report DOE /ID-10503. This report documents how the various cfTects in steam explosion progress, like wave dynamics, explosion coupling and integral behaviour utre assessed. The report demonstrates how the code is handling pressure waves in single and in tso-phase fluids and this not only in a one-dimensional, but in a two-dimensional geometry. Special attention was given to reflection and transmission behaviour. The comparison between predicted data (ESPROSE.m-code) and experimental results showed very good agreement for a wide variety of thermo-and fluid-dynamic parameters. The local situations and the temporal behaviour are well predicted. So, the code is in a condition, that allows to predict steam explosion behaviour also beyond the experimentally willied area.

The extrapolation from the small scale to the large geometry of the reactor were done by using the basic equations for wave dynamics la multiphase media and G 128

J~ y a <

constituthe laws for microinteractions. The latter ones were tenned via experi.

p ments in the SIGMA facility, also. The combined theoretical and experimental

-Q) efforts are a very good basis for predicting and simulating large scale conditions, also.

Finally one has to ask the question on " substance scaling" i.e., the applicability.

of the data, measured with modelling melts to liquid corium. The experiments were mainly performed with tin and with aluminium oxide: Especially aluminium oxide is verylikely to produce supercritical steam explosions when it is mixed with water. The authors of the report DOE /ID-10541 write on page 2-1 (chapter 2

" problem deRnition and overall approach"):

"Also, it is important to note, that within the limited experience with reactor fuel material (UO2, ZrO2), we have no evidence of explosions, but rather extensively voided premixtures (Huhtiniemi et al.,1995), nor is it known whether or under what conditions such premixtures can be triggered to explode".

With respect to " substance scaling" the data, presented in the report DOE /ID-10541, on explosion loads, originating from steam explosions are on the safe side without any doubt, because a corium melt / water interaction will produce much -

softer pressure pulscs than experienced in the experiments with aluminium oxide

melt / water interactions.

Seh50. 1V.5 Chapter 6. QuantlRcation of Explosion Loads / Appendix C.

Detailed Explosion Results The chapter 6 and Appendix C present.the results of explosion-propagation cal-culations performed with the ESPROSE-m code, using, as initial conditions, the pre mixture conRgurations calculated with the PM-ALPHA code. The trigger '

time is chosen as very short, since during the early time the void fractions of the coolant around the fuel particles are relatively low. Later, the void fractions increase substantially, and would inhibit fuel break up and triggerability.

The results are presented for the C-1 and C-2 scenarios with three values of f and a set of trigger times. For the no break-up case these times vary - from 0.05 sec to 1.0 sec, while for the break-up cases, they vary from 0.04 to up to 0.19 seconds.

3 (V G-129

l l

The results on pressure, impulse and effective area are shown for various locations In the lower head. Peak loadings histories are also shown as a function of trigger times. The extreme sensitivity to trigger time is evident from Table 6.1. If the trigger is delayed by 0.06 seconds for the C1-10 and C2-10 cases, there is only a very weak explosion. For the C1 -20 and C2-20 cases, there appears 'o be a time interval of only 15-30 msec for the trigger to generate a supercritical explosion.

Thus, triggering time appears to be the deciding factor. A physical explanation for this extreme sensitivity should be provided by the authors.

Very simply it has to do with the combination of volds and melt length scales. It was discussed already how these two compensate each other, while at both extremes, high voids or large length scales, we have benign behavior. It is clear that the explosive quality of a prcmixture will maximize somewhere in between. A new way of plotting the results (see addenda to Chapters 5 and 6) illustrate this very well. Also, new results with intermediate

! values should help further in understanding this.

Sch51. The Appendix Cghes very nice pictures of the pressure wave travers-ing through the lower head. The pressure signals at various points in the lower hecd are shown and the peak pressures and impulse loadings are shown as graphs versus time. Theses pictures and graphs were very helpfulin the review of Chap-l ter G.

Summarising, I can say that the authors have performed logical analyses of the loadings imposed by the steam explosion, and have provided very nice results.

I have not understood the reasons for the extreme sensitivity of the calculated results to the trigger time. The peak loadings shown in Table 6.1 are, in general, modest. The highest loading is found to be m200 k. Pa. s. Is it possible that for 6 = 30, a higher value than 200 k. Pa. s. is calculated?

See response to question above (Sch50).

Tur6. 5. The higherexplosivity ofa premixture with reduced voiding appears to be a cordecture that is not fully supported by the experimental evidence. Reported explosions in the JAERI ALPHA facility occurred with large voids in the mixing region.

Explosivity refers to intensity, not likelihood. In any case, there will be peripheral zones of low void fractions and nothing prevents these regions from initiating and propagating G-130 l

l

1 explosions. The voided regions simply damp the energetics, and reduce the amounts of

( ) fuel exploding coherently.

Tur31. Chapter 6: QuantiScation of Explosion Loads \

i

1. Where is the trigger cell?

At the bottom of the premixtures in each case.

i Tur32. 2. At what time was the etfective area evaluated - that of peak pessure? If not, you obtain larger effecthe areas than ~0.1 m'.

~

At the time of the main portion of the pressure pulsels delivered.

Tur33. 3. The question raised by the calculations is how far is one from the danger zone? Could we get there by a modest increase in system pressure (what value was assumed?) and/or varying the value of f?

More calculations presented in tlie addenda to Chapters 5 and 6, and their interpretation should help this kind of question. We used 1 bar, and calculations done since, at 3 bar, gave the same results. Further. results for in-between S values are now provided. The p results are understandable about inw and where the high pressure pulses are produced,

(  :,o it is unlikely that we missed somethbg that could " unexpectedly bring us to the danger zone."

Tur39. 3. It would have been useful to have av indication of the elfects of uncertainty in the constituth'e laws (eg mictoihteractions) to determine where confirmatory studies are required.

Presently w'e have chosen very conservative parameters fr r microinteractions. Results for parametric variations, guided by the most recent SIGMA experiments, are now provided.

You3. Second, in ESPROSE.m, there are three parameters, that must be set from experiment: an entrainment factor, a fragmentation constant, and a thermal enhancement factor. There appears to be .come dependence on the melt material for these factors, so the lack of data with reactor materials to set these parameters concerns me. I believe this was also pointed out by Theofanous et al. In the " Concluding Remarks" section of the ESPROSE.m verification report, in regard to expanding the microintemctions database to reactor materials. Use of parameters that had been set from experiments using reactor materials would p

V G-131

enhance confidence in this aspect of the calculations. In regard to the mictoint-eraction model itself, I believe that this model is suMciently close to reality that experimental results can be extrapolated to reactor scale.

YouS. Fourth, on page 6-4, there is a conclusion that the size of the impulse does not depend strongly on the size of the mixture region. I think that this is in contrast to first principles, which would suggest that it would be directly proportional, ignoring other elTects like the varying void fraction, and to the results in Table 6.1, which indicate a strong variation, ignoring the %e, between cases C1 and C2: 90 vs 120 for 0 = 10, and 120 vs. 200 for f = 20. Or did I misread the sentence? Also, how do the results compare in magnitude to the impulse of the initial trigger itself?

The results do not depend on the magnitude of the trigger. The results in Table 6.1 show that, indeed, the impulse does not depend on the size of the premixture (note that bigger premixtures are more voided). There is only one case, the 200 kPa-s one, that stands out.

This case has been reexamined and discussed in an addendum to Chapter 6.

G-132

O ESPROSE.m Code and Verification

)

L ,i Dan 12. 1. DOE /1D-10503 " Propagation of Steam Explosion: ESPROSE.m Verincation Studies", by T. G. Theofanons, et al. Aug.1996, and update of Sect.

4.2.1 This is a convincing document, laying out the evidence that ESPROSE.m has the capability successfully to predict wrious shock and vapor explosion scenar-ios. These range from simple steady, one-dimensional shocks propagating ihvough single-phase liquid and homogeneous gas-liquid mixtures, for which exact (or nearly exact) solutions can be found, to experimental shock and explosion data in the SIGMA and KROTOS facilities. The 1-D waw dynamics were tested for shock speed, fluid velocity, reRection at a rigid wall and reRection at a free interface with venting for single-phase, liquid-air and liquid vapor cases, using ESPROSE.m and CHAT. The small deviations between the analytical solutions and the codes can be attributed, at least in part, to the fact that the analytl-cal solutions used an assum4 constant sound speed in the liquid, taken as 1500 m/s and 2000 m/s, while ESPROSE.m used the real properties of water. In some n cases, as in the deviation in reRected shock speed, the differences in the analytical i >

G predictions u. the two assumed sound speeds is considerable, but ESPROSE.m gh es a smooth function ofshock pressure which interpolates between the two lim-its, and is hence more credible than either of the two analytical shock reRected shock speeds. As a check on the wave dynamics with reRection/ transmission at interfaces between tuo materials of diB'erent acoustic impedance, widch governs the unloading-explosion coupling near a free interface, the CHAT code, using the method of characteristics, was written. For large amplitudes the quasi-linear code CHAT-QL cvaluates the coeBicients in terms of the local Buid properties. These codes were then compared with ESPROSE.m for pressure and velocity distri-bution for 1 D single phase venting and for shock speed, Buid velocity, reRected shock speed and shock amplitude for a 10% void non-condensible steam / water mixture, with excellent results. This sort of independent cross-checking lends considerable conndence in the basic structure of the ESPROSE.m code, which is a Snite-difference code in laboratory coordinates. Exact solutions in 2D geome-tries are then given for innnite pool, cylindrical open pool and cylindrical closed pool geometries, by st.perposition of an inRnite array of sources and sinks in or-der to obtain reBection and transmi;sion behavior at rigid and free boundaries.

't )

'd G-133

These solutions are in themselves impressh e, and likewise the general agreement between the code and analytical predictions for pressure as a function of time ,

over the two-dimensional region. In fact, the agreement be*xecn the two predic-tIons for the centerline pressure distribution as a funnn of time and distance is remarkable.

There is also good agreement between the ESPROSE.m prediction and the exper-imental data in the SIGMA facility with allliquid, and with a liquid-air mixture in the expansion section. The key paratneters of time of arrhal and amplitude of pressure waves at several locations are well predicted, particularly in view of some high frequency ringing in the experimental transducers. The same is true for multi region runs with high pressure (6P = 68 and 136 bars) with diEerent initial void fractions.

Dan 13. One aspect of vapor explosions which is difficu!t to model properly is the presence of strong energy sources, especially near a free boundary, which are caused by local explosions (rapid mixing) of fuel drops produced in the course ofjet breakup. These sources can distort venting and reflection phenomena near free boundaries. This was modeled first by characteristics solutions with single internal heat sources, with the energy assumed to be going only into the vapor, with heating rate increasing with wlocity:

Qg = Cu" (3.2) with C being an empirical constant. More explanation is i.eeded for the assumed form of this equation and the magnitude of the exponent, based on energy dissi-pation. Howem, excellent agreement is obtained for various assumed values of C between CHAT and ESPROSE.m. However, comparisons with data for single exploding drops are lacking.

Here we addressed a potential numericalissue, and the form of Eq. (3.2)is immaterial. All we wanted was to create a source which increases in intensity with velocity, so we pick a rather strong dependence on u to the 1.5 power. Through the coefficient c, we can further adjust the " base level" of the intensity of such a source. We then show how one can have different degrees of interaction between such a source and venting, and we show atso that G-134

o ESPROSE.m performs quite well against this rather ma}or computational challenge. We h do not see any need for special experiments in this area.

Dan 14. At the ather extreme is a plane shock wave moving into a fuel /stcam/ water mixture at low pressure. This is the scenario envisaged by Board and 11all, fol-lowing the one-dimensional theories of Lifshitz and Zeldovich. In the B 11 theory the average speclBc volume of the mixture is plotted against pressure, starting with adiabatic compression to a peak pressure (von Neumann point) at which the steady mass continuity condition for the Bow behind the shock front is satisfied with sninimum entropy generation. This results in the Chapman Jouguet (C J) condition of tangency to the reaction adiabat, or ilugonlot curve, leading to a minimum of shock sj>ced with pressure. The ESPROSE.m calculation shows the development of the shock into a steady state detonation, using a fragmentation rate given by Fr a p}-f; (4.1) a where p} ls the local macroscopic density, C is the expected" shock speed (1500 rn/s), and bz is the grid slie. Why the fragmentation rate should be a function of ax is not clear, nor is the form of this equation. More explana-O tion is needed liowever, detailed comparisons between 2D and 3D versions of 0 ESPROSE.m are encouraging, and the ESPROSE.m P-V lines are close to the expected Rayleigh line, and bounded by the shoch adiabat.

There is nothing special in Eq. (4.1), except that we wanted to be sure the fragmentation rate is fast enough, so as to be comparable with the " infinite" rate assumed in the steady detonation theory. The form of Eq. (4.1) assures that fragmentation occurs within the time it takes the wave to traverse one computational cell. We have since that time done calculations with very high, constant, fragmenta tion rates niso. By improving our steady state theory (see Appendix D) to account for finite compression trajectory effects on the Hugoniot, the agreement now is even better than befwe.

Dan 15. These results are excellent back up for the interpretations of the KROTOS experiments gNen in DOE /ID-10489. All in ell, even at this stage of limited comparison with Integsel explosion tests, one has coalidence m the prodletion of pressure n time at various locations on the pool boundaries, and consequently of the initial kinedc energy oflarge masses impacting on reactor structure,

• * * * * *

• c * * *************

O C/ G 135

l 1

Dur25. e The microinteractions formulation for hydrodynamic fragmentation in thermal detonation waves needs further clarification and development, for sin-gle drop a well as finally for drop assemblics. This concerns the ccnclusions based on theory and on single effect experknents as well as on NROTOS experiments.

This was already recognized in the report, leading us, therefore, to take a censervative approach in the microinteractions parameiers.

Dur20. With respect to FARO and KROTOS analyses, dinerent premixing and explosion codes have shown the capabilities to calculate the experimental behaviour. But, the underlying physical formulations and thus the physical in-terptctations differ strongly. Further, even no convincing comprehensive under-standing has up te now been gained on the differences especially in premixing behavior between UO 2 /ZrO2 and Al 203 In KROTOS. Thus, further work is nec-essary to get approved understanding, models and codes. In general, the results would be more convincing if supported also by other codes based on a cominon physical understanding and corresponding formulations.

Clearly, this is so, but we cannot be responsible for the codes of others. KROTOS was not designed and is not appropriate for understanding premixing behavior. What is needed is better diagnostics on the premixtures triggered (not predictions) so that predictions of g energetics can be made on a more secure basis.

Cor10. INITIAL COMMENTS and QUES".'lONS for DOE /ID-10503 The report is very well organized and describes in sufficient detall the ability of ESPROSE.m to perform shock propagation calculations for gas / water and va-por/ water situations. I do not completely understand the origin of the CHAT [or CHAT-QL) code compt. 3ans. Are these standard code models or a formulation of the authors to do a code cross-comparison? I understood them to be the lat.

ter, and thus I wonder about the need to compare to actual experimental data on shock propagation in single phase and multi phase systems. This is a minor point, but ,' think for completeness a link to data is best. My main comments are about the comparison to the KROTOS data.

As explained, the CHAT (CHAT-QL) are special purpose codes for these numerical tests.

We have comparison to wave dynamics in the SIGMA experiments also, as shown in the report.

G 136

• i Cor20, 1) The lultlal statement is made that the KROTOS tests are a chal.

lenge since they have Imperfect characterization of the initial conditions. One ,

question my be if there are any other tests which she them more insight? After my own search, however Bawed these tests are, these and other one-dimensional experitnents are the best we have. My other wmment is about the initial con.

ditions. The comments on page 4 20 indicate that the fuel mass and Bowrate, i

but there is a problem as far as I can tell. The mass is wrrect, but the initial Jet size is not 1 cm but 3 ctn and I think the Bowrate of 1 kg/sec is too low by at least 50%. Finally, the fuel particle temperatures are different for each of the  ;

tests noted as is the location and timing of the trigger. I am not sure that the 7 authors are aware of this. I can send them this infortnation if needed, but in the l case of KROTOS 38, the bitial conditions are not correct; e.g., the Jet size is 3  :

cm and the trigger time is 1..*2 sec at or near K3 and not at the leading edge, with a pour titne of about 0.7b sec.

The first part of this comment is inappropriate; the second part reflects a misunderstand-ing. The " imperfect" is the wrong word here. The initial conditions in KROTOS relative to its main task, which is to provide data for ID detonations, do not exist, periodt To call this " imperfect" is not simply an understatement, it misses the point completely (see also >

Fle18, Seh31, and Tur44). Our statement was "From a code verification perspective, the  ;

KROTOS experiments are difficult challenges because of their inadequate characteriza.

tion of the initial premixture prior to triggering," and we stand by this statement. The ,

reviewer's comment is further inappropriate because we do use KROTOS in our verifi.

cation effort. To do so, without acknowledging the uncertainty in initial conditions would be a real omission; it is good that the reviewer's comment provided an opportunity for further emphasis. Even if we ignore uncertainties emphasized by this geometry, on the  ;

multifield aspects, uncertainties in breakup, widely known and accepted, are enough to overwhelm this problem. KROTOS was not intended and it is not a premixing experiment.

Still, all we are given are initial conditions for premixing. Now, having understood this .

point, the initial conditions disputed in the second part of his comment, can be addressed -

as follows. The size of the fuel inlet for the calculation is 3 cm, which is consistent with the jet size 1 cm is the initiallength scale assumed for the fuel. Its effect on the premixing is unimportant since it will be compensated by the value of the breakup parameter #,

~

which is adjusted parametrically. The trigger is applied near K3, not at the leading edge (see Figure 3). The flowrate of 1 kg/s is the approximate value quoted by many previous O o.,n

.h r - , ,# -

.g.c, -

- - + . . . -

-. ..u,, , -w,,,-. , - . . , - ,. . ,~ n~.e_,5-- . ~ . - ~ , ,.., --,. ,-..

publications on KROTOS (Hohmann et al.,1994). Our understanding is that there is no new measurement to improve this estimate of mass flow rate, g Cor21. 2] The concept of using the parametric mixing model for a f value of 30 or 50, again ralsw the questlon of what is appropriate and why. The kinematics in Figure 2 don't have any comparison to the thermocouple data for position of the melt as a function of time and give no indication what 50 is *better" or more correct than 30 for a value. Also what is the time evolution of the particles as thejet breaks up from 3cm to what size? None of this is discussed at all.

As discussed in the text, # = 50 is the value at which "the melt penetrates to the region between pressure transducers K2 and K3." As shown in Figure 2, the penetration is much slower when 0 = 30 and much faster when there is no breakup. The particle size distribution at the time of triper was shown in Figure 3 of the report.

Cor22. 3) The final point is the use of the parameter, xj = 0.5 to 1.0. Does this parameter mean that when the ndue is 1.0 all the fue, as quenched as it is fragmented with some fractlon oI water and stcam? 1( that is the correct interpretatlon then, the pressure plots do not seem to make sense to me. This is especially the case, since the predicted vold in by Figure 2 and Figure 3 is very small. There is something missing in the description; since 1.5 kg of molten alumina has the energy of 6-7 MJ and thus must be gaenched by almost all the 35kg of coolant if there is to be such a 'small' pressurization with such little vold.

How much water is assumed" to be intermixed with this fuel to give the pressure signature we see? This is never discussed and it is the most crucial part of the model. The complete picture is missing and thus,1 am not prone to agree this is a reasonable prediction until all the ' parameters' are specified and explained.

Also, comparisons to more than one test is needed. This has been done with ather FCI models.

This question stems from the reviewer's misunderstanding of the parameter rf and the microinteractions model. As explained in the report,"xf si the fraction of computed liquid melt participated in the explosion." This factor is necessary because PM ALPHA was made to underpredict melt freezing (it did not account for surface freezing of superheated melt).

When rf = 1.0, ESPROSE.m predicted a peak pressure of 4000 bar, which is consistent with our previous microinteractions predictions. The amount of water intermixed with the fuel is calculated based on the microinteractions parameters, which were given already in G 138

-, Appendix C. Several representative KROTOS tests are now interpreted in the addendum

( to Section 4.2.1.

Fle13. 4 ESPROSE.m Validation Studies (DOE /ID-10503)

Firstly, it is important ta tackle head on the ESPROSE.m formulatlon, which 1

. believe it is fair to say has not been widely accepted. I Bnd it hard to understand why this is the case. Essentially, the novel feature in ESPROSE.m is the inclusion of an additional fluid (the m Buld) which represents the fragments and the Buld in intimate contact with them which is being heated. The need for such an approach seems beyond doubt to me following the very careful experimental analysis of Baines [1] and my own at tempts to analyze KROTOS-like tests using CULDESAC l2). The authors have provided comprehenske experimental data for appropriate pressure loadings to show the Rnite mixing rate. They identify the need for an enlarged database but it should be recognized that the ESPROSErnfonnulation is conservative in the sense that by rnixing thefragtnents with only afraction of the coolant they generate high localpressures. This point should be kept in mind when examining the use of ESPROSE.m results.

The remainder of this section contains detailed comments on the various chapters of the verification report.

(

4.1 Introduction The main feature of this chapter is Figure 1 which ghts the validation strategy.

This is very extensive and to my knowkwige is the first model to be subjected to spectBc wave dynamics and explosion coupilng wilBeation studies against analyt-leal and experimental data. It also covers the tuo main experimental programmes KROTOS and ALPHA.

Referring to the first sentence of this comment it is perhaps worth noting that with the possible exception of one reviewer Oacobs) the microinteractions concept seems to be well received. Also, we should note that the idea is now being emulated in other codes such as TRIO-MC.

Fle14. 4.2 Wave Dynamics The 1D solutions for the shock speed and particle velocity (important in relative velocity fragmentation) are excellent. The same applies to wall reRection studies, the eHect of vold and the eHect of non-condensable gas. The venting calculations .

> )

v O 139

also show good agreement with the CHAT results. I was curious to know why the calculatians were performed for a pressure step of 40 bars over a base pressure of 100 bars and over a space dimension of only 1.4 cm. I would have preferred to see venting on the 0.1 m scale (with a 1 cm mesh) and a pressure difTerence of say 10 l

bars venting to atmosphere. Figure 9b shows that the ESPROSE.m results only exhibit disperslun at the first few time steps and that the numerical diffusion is modest.

Point well taken. More calculations provided in the addendum to Section 2.1.2.

Flo15. The 2D comparisons are impressive and show that ESPROSE.m cap-tures the wave dynamics very well. The only point that this section raises for me is why in the type B behaviour the ESPROSE.m results have a spike at the origin (as expected from the source description) but the analytic solution does not (see Figures 7,13 and 19). Is this simply a plotting omission?

Yes, this was plotting artifact and was removed.

Fle10. The experimental comparisons with data from the SIGMA facility are interesting and show that ESPROSE.m is capturing the average wave behaviour well. Clearly, the pressure transducers are picking up many local reflection events which are due to the inhomogeneous nature of the ' mixture' and cannot be mod.

clied via a continuum approach. I am surprised that ESPROSE.m has done so well for this system with the only apparent systematic difference is the tendency for a ~1 ms time lag.

Actually, the maximum time lag is only 0.2 ms (1 ms is a mafor division - this was cla rified in all captions).

Flc17. This section provides very solid verification for the code algorithm and the choice of solution parameters.

4.3 Explosion Coupling This section contains test cases in which energy is input into the gas phase via a parametric relationship in which the energy input into the gas phase is propor-tional to the fluid velocity to the power 1.5. This is done to represent the fact that in ESPROSE.m the energy is input in the m fluid. Results for calculations for both cases considered are in excellent agreement with the CHAT simulations.

G 140

Figure 5, for simulations on a larger space scale, examines the effect of grid size.

( ) The comparisons are good with dllferences being confined to the interface region.

4.4 Integral Aspects The analytical tests show that ESPROSE.m can perform well at the extreme limit assun ed in the Board Hall rnodel. These calculations are interesting as they show explichly the effect of *he fragmentation rate and entrainment factor on the propagation characteristics. Figure 6 is interesting in that it shows dispersive-like behaviour but if the grid is as described earlier these are real rather than numerical. Could the authors comment 7

'6; dispersive-like behavior is real; it is due to the slow fragmentation rate which allows the pressurization to occur gradually.

Flc18. The confirmation that the 2D and 3D models ghe similar results is thorongh and convincing.

I agree with the authors that the KROTOS tests are too poorly characterized for real validation studies and therefore I do not think this section is central to the validation case. The ralnt about melt frwzing is very interesting and the fact that the code under predicts freezing times is important. This effect will be compounded by the fact that that the melt is assumed to be at a uniform temperature, whereas an outer shell will freeze first. Surface freezing provides the most convincing hypothesis (to me) of the non-explosive behaviour of UO 2 in KROTOS.

Flc19. 4.5 Numerical Aspects I agree with the conclusions drawn. The presented calculations clearly show that the authors are aware of the need for adequate spatial and temporal resolution.

In addition, the results show a good compromise between diffushv and disymhv errors.

4.6 Concluding Remarks This is a very important sectlon and I believe the authors haveJudged the current situation very well. I agree entirely with the conclusions they have drawn from the very comprehenshe sets of calculations performed to date. There is a clear

.(l V G 141

nwd for the high temperature SIGMA data and I am aware that plans to obtain this are well advanced.

I personally doubt that it will ever be possible to characterize the KROTOS experiments much better and my experience with the MIXA tests tells me that there will always be something leit ta be measured. Therefore I asree that this is a lower priority. The comments on secondary pressure waves are interesting and clearly of a very fundamental nature. I do not believe that such cffects could l>e addressed easily within the continuum model but I would certainly encourage their investigation.

Finally, I agree completely with the closing paragraph: moving to large scale, multi dimensional experiments will only add confusion.

4.7 Appendix A 1 have no specific comments here. I am generally familiar with the modelling approach taken and I believe appropriate modelling choices have been made from the available database of constitutive laws, it should be recognized that it is in the formulation stage that the ESPROSE.m model differs fundamentally from others in that it is 3D and uses the microinteraction concept to allow for thermal disequilibrium within the coolant.

4.8 Appendix B This section contains a description of the CilAT code used to provide analytical solutions for code comparisons. The model is formulated for the case of homogo-neous flow ofliquid and cookmt (no slip but ditTerent temperatures). Thus the system has only real characteristics and themfore can be solved in an elegant and accurate manner. It provides an excellent means of testing ESPROSE.m.

4.9 Appendix C This appendix is a reprint of a conference paper which describes the microinter-action data and its implementation into ESPROSE.m. I am familiar with this work (from the paper and visiting the facility) and believe it to be both unique and of a high quality. Whilst at present results from low temperature melts have ta be exttapolated ta the reactor case, plans are well advanced ta produce the required data.

G-142

t 4.10 Appendix D .

This appendix also contains a reprint of a conference paper which discusses the manner in which the *real uvrid' differs from the Board Hall model. It is very interesting as it shows how the inclusion of microinteraction physics produces propagation behaviour which is very different from the Board Hall mode) and other propagation models which do not allow for micro mixing. Essentially, it allows propagation in systems which are melt lean because the energy from the melt is transferred to only a fraction of the water present. It provides an interest.

Ing perspective on which to end the ESPROSE.m valida%n report and clearly I tilustrates what a significant advance the mictointeraction concept has been In propagation modelling. -

References

[1] Baines, M. (1984). Preliminary measurements of steam explosion uvik yleids '

in a constralned ry., tem. Inst. Chem. Eng. Symp. Series, 80,97-108.

[2] Fletcher, D. F. (1991) An improved mathematical model of melt / water detonations-H. A study of escalation. Int. J Heat Mass hansfer,34, 24492459. <

O

\U Moo 3. 5. The source term for area production in Appendix A of DOE /ID-10503 is based on the assumption of particle number density remaining constant, while their size changes. A bit more explanation or justification would help.

Wouldn't it make more sense to predict Interfacial area growth by the formation ofmoreparticles as the melt decelerates in water? Taylorinstability was employed i to obtain the Bond number criterion in interfacial area growth. Could that model

, be employed to obtain a fastest growing wave length and droplet formation?

During pmpagation the key mehanism is Microinteractions, and this involves fine scale fragmentation and mixing in the vicinity of all macroscopic particles. This is clearly supported by the SIGMA experiments. Both Taylor and Helmholtz instabilities have been employed in the consideration of hydrodynamic fragmentation. Ultimately it is ,

more appropriate at this stage to rely on correlations suggested by such appmaches and the SIGMA data, which fully represent behavior in large scale explosions. This is our appmach.

G-143

M005, 7. C<nnvethv and radiathe heat transfer from the fuel to the coolant is estimated in much detail, drawing from various experimental studies between coolant and heated solid surfaces. Is there a backup analysis t a show that for the rapid heating associated with steam explosions, the heat transfer is not limited to how fast it can escape from the molten particles? Are there potential droplet sizes, relative velocities, and fluid properties where internal conduction (or convection) might limit the heat exchange rate?

This really depends on the resulting size of fragments. For 1 p m particles the time constant is 10-8ps. We believe ignoring this limit don, by assuming instant equilibrium, is appropriately conservative.

Moo 10. STEAM EXPLOSION The mechanics of steam explosions are described lu DOE /ID-10503, detailing melt introduction to water, interfacial breakup and premb:ing of debris parti-cles with water, the cITect of voiding around the particles con heat transfer, the triggering of explosions, and propagation of pressure waves with reBections from rigid mechanical and gas-liquid interfaces. It was earlier hund that 1.0 GJ of energy could fall the lower head. Ilowev-r, further unders:anding has led to a reexamination of the mechanics of steam explosion force getcration to determine a more realistic criterion for lower head failure.

It was determined that the AP G00 could withstand 500 bars of pressure for mil-liseconds without failure. Computations with the ESPROSE program displayed the difficulty in generating such pressure impulses with attenuating phenomena like widing, which resists triggering, and pressure venting from the water sur-face. Extenshv development of ESPROSE have been performed with both data from the SIGMA and KROTOS experimental facilities. Simpler analytical mod-cis have provided assurance that ESPROSE accommodates detonations, shock propagation, and reBection.

Significant etlects embraced by ESPROSE result from the physics incorporated, which are consistent with experiments. Calculations show the strong attenuation of shock pressure loads with distance, and time by venting from the water free surface in the APG00 systems. It is also realized that venting may not signincantly reduce loads if the water depth is high in the lower head. Strong evidence is supplied that ESPROSE incorporates the appropriate physics, and can be used G 144

with con &dence to provide the causal relation for emeloping the eMect of trigger time on steam explosion severity.

Moo 11. The physical mechanisms considered by ESPROSE.m include shock pressure propagation from a trigger, which collapses volds, forces liquid onto the melt, producing fragmentation and mictointeractions, escalated heat transfer, further steam formation, and rapid expansion (explosion). A statement on page A 18 of DOE /ID 10503 needs farther clarlBcation. Where the pressure increases rapidly ahead of an explosion front, why does the vapor become instantaneously subcooled? (if saturated steam is rapidly compressed, it would tend to follow an Isentropic path on the vapor dome into the superheated region, not subcooled.)

What is meant is that the liquid becomes subcooled and steam is to rapidly condense.

Moo 12. On the same page, it is stated that behind the explosion front where pressure is decreasing, the liquid can become superheated. (if you decompress saturated water, the path drops into the steam dome.) It would make better sense to me (I can't speak for others) to note that the nonequilibrium states lag behind a steady state in the superheated or subcooled region.

Here, we refer to the liquid that participated thermally in the interaction (part of the m fluid).

Moo 14. REVIEW OF STEAM EXPLOSION LOADS The verlBcation of ESPitOSE.m, based on stepwise experimental measurements and comparison with simpilBed theoretical methods shows that reasonably con-servative assessments of steam explosions are possible in the present version.

The discussions of DOE /ID 10503 provide foundational support of the physi-cal modeling and numerical procedures to predict steam explosion properties for given melt addition rates and states. The basic physics invohr wave dynamics, including sound wave propagation and shock development and propagation in a water Biled wglon. Two-dimensional calculations performed by ESPROSE.m are compared with siinpilBed computations using the method ofimages and solutions similar to classical waterhammer. Some comparisons are included based on char-acteristic solutions. The results form a strong basis for concluding that the code is producing reasonable predictions for the expected range ofinput parameters.

G 145

i Pressure propagation speeds, attenuation from wave interaction at free surfaces) and wave amplification by reBection from rigid surfaces have all played a role in the verification.

Numerous two-dimensional ESPROSE calculational surfaces are compared with solutlons obtained from the methad ofimages, and found to be suBiciently similar, leading to the conclusion that basic physics of explosions are included in the model. "Several geometric parameters were varied, as was the source velocity function. Good comparisons were consistently achieved.

The SIGhiA tests involved a melt droplet which was triggered at a speclBc po-sition, leading to local pressure traces. Comparison of the pressure traces with ESPROSE calculations showed reasonable tracking of pressure waves originating from the droplet region to the rigid end of the test section, and reRection back toward their origin. Additional evaluation with the method of characteristics were provided. The wave dynamics, indeed, appear to be properly described in ESPROSE.

One piece of informatIon lack pointed out in the report is that the data base needs expansion for mictointeractions with reactor materials.

g Sch23. Ill. Revien of the Report: Propagation of Steam Explosions: ES-PROSE.m Verification Studies by T.G. Theofanous, W.W. Yuen, K. Freeman and X. Chen This report deals with the next phase in the steam explosion process, after the pre-mixing has been achieved. The report, therefore, deals with the explosion process and develops a methodology to describe the process, and evaluate the energetics, which are then employed to assess the damage potential of the explosion on structures, which surround the explosion. A trigger la assumed, which starts the explosion process, in which intimate contact of the fuel and the coolant leads to production oflarge amounts of vapour, and the supercritical explosion.

The report consists of the front part, where the results of the verification cal-culations are compared to the observations, and data, obtained ia the SIGhfA facility at U.C. Santa Barbara. The report also contains four important appen-dices in which the code models, a 1 D characteristics model, constitutive laws far micro-interactions and thermal-detonations are discussed.

G 146

,~

In the following paragraphs, I will comment on each of the major sections of this

/

report.

)\

Uli Appendix A. THE E3 PROSE.m h!ODELS

, The overall approach of the model development is brilliant. Recogalsing that the dynamics of a pressure wave, generated by a trigger, coupled with fuel frag- ,

montation, micro (or local) mixing and heat transfer result in energetic steam explosions, the authora have concentrated on those aspects. Perheps, the SIGhIA experiments provided the key observations towards the development of the micro-Interaction concept and the m fluid, where the fuel-coolant heat transfer occurs.

The energy transferred is then employed in the multif1uld treatment ta calculare the pressure Belds as a function of time and space (2 D/ 3D). The damage po-ternial is, then, evaluated with the calculated dynamic loading knposed in terms of klio Pascal seconds.

The modeling approach is similar in most respects to that employed for the Phi-ALPHA code, i.e., solution of a set of multincid conservation equations, with specined constitutive relations. The Belds chosen this time are fuel, liquid and the m fiuld. There is an additional mass conservation equation for the debris, p

A I.e. the fragmented material. The m Buld equations contain source and sink terms, which are based on a picture ofliquid entrainment and phan change. h2el fragmentation is included, whhh contributes to the increase in interfacial areas.

The heat transfer across the Belds is included in the energy equations. The system of equations appears to be complete. The constitutive relations between the 3 Belds for interfacial drag, heat transfer and phase change, again invoke many correlations and dimensionless numbers.

Sch24.  : believe the comments that I had made regarding the complexity of the constitutive relations for the Pht ALPHA code also apply here, and the possibility of checking the synergisms between the momentum and heat transfer processes through separate-eflect experiments, should be expiored. New data may have to be obtained and some prioritisation should be performed.

Actually, in this respect, ESPROSE.m is much less of a challenge than PH-ALPHA. All we have is wave dynnmics, which were verified very carefully in a step-by-step approach, O

O G 147

and Micminteractions, which is really o'otained experimentally under conditions that fully simulate large-scale explosions. So, we really do not see the concern expressed. h Sch25. The fuel fragmentation is treated as in the Phi ALPHA code and is controlled parameterically tbrough 0. Thete is another parameter which enhances the fragmentation for thermal elfects. Both oIthese parameters are user specifled.

The entrainment ofliquid in the m Buld is controlled through the parameter E, which is taken as a function of the fragmentatioh rate.

Actually, we used Bf to distinguish from the # used in PM ALPHA.

Seh20. I believe, the parametric trentment is very intuitive, and Ihe authors admit that it is an important component of the micro-interactions concept with somewhat speculatsve constitutive laws. Since, the m Buid interactions are the basis of ESPROSE m, I hope that the authors have already obtalt.xl additional data from the SIGH!A facility to provide greater support for the experimental basis of the parametric treatment.

l These parameters t re fixed by the SIGMA experiment. We now have data also for iron and gallium (see addendum to Appendix C), and plan ZrO2 tests in the near future.

Sch27. IIL2 Appendix C. Constituthv Laws of bilcro interactions This appendix describes the experiments performed in the SIGH!A 2000 facility with gallium and molten tin, subjected ta high pressure waves, in order ta derive the constituthc laws for the micro-interactions, needed for the m Buld.

The experknents are des ribed. They are really very dlBicult, but precise exper-iments. Some results are shown as movie. X ray and SEh! Images for the change in pre mixing volume, as n function of time.

The results af experiments are used to derive the values for bf, ns, and fe, the entrainment factor. For example, Fig. C-13 shows f, = 7, 8 and 12 give best fits, respectively, for three isothermal Gallium tests i.e., G/204/45, G/G8/45 and G/272/45. The mox conservative value f, = 7 is then used to determine the value of 6j = 9. The value of as derhed from Fig C 10, while keeping f, a 7, and Of = 9. It appears that as varies from 1.4 at 68 bar pressure to 4.2 at 204 bar pressure.

The above is a logical but highly empirical determination of 3 parameters from a small number of tests. Perhaps, more data has been obtained from SIGH!A G 148 l

l 1

4 to confirm the choices made for these key parameters. Obviously, more data is needed from SIGhfA or another shock tube. I believe, dlKerent materials should also be tested, in particular, rnelt drope of binary oxides. Thelt fragmentation behaviour maybe different, due to changes in properties they experience with a chaage in temperature.

For new data with iron melts see the addendum to Appendix C. As noted already, data with ZrO , and if needed, UO /ZrO , will be obtained in SIGMA 3000, currently nearing operation.

Seh38. Ill.3 Appendix D. On the Existence of Thermal Detonations This is a wry interesting re examination of the Board Hall model for steam ex.

. plosions. The micro interaction model and the concept of the m Buld is employed to show that supercritical steam explosions can be obtained with Jean rnixtures in highly volded regions; conditions for which the Board Hall muJe! will predict only very weak explosions.

hly understanding of the mic:v Interactions concept, Introduced by the authors, is that they take place in the m Buld in a limited volume. I believe, this results from the observations made from the Gallium drop (also perhaps the tin drop)

J experiments conducted in the SIGhfA facility. The previous concept was that the pressure wave will fragment a melt drop into Sne droplets, which will mix with the whole coolant volume. The SIGhfA experiments showed that this does not occur in the time frame of the pressure wave melt drop Interaction. The heat transfer to the m fluid s coolant, in the limited volume occupied by the m Buld, generates

' very hlgh pressurcs. The shock wave then travels into the non participating 2, tid around the m Buld, increasing its pressure to sustain the pwpagation. This ms!:c< .

possible the supercritical steam explosion with a fuel-coolant mixture, which is lean on an overall volume basis, but is not so lean on the m Buld volume basis.

(Cf. Figures D-8 and D 9, where high pressures are obtained for the coolant to debris mass ratio f, a 1 in the case of tin at 1500*C and for f, = 2 to 8 for the case of UO at 3300*C)

I believe, the authors have provided a very logical explanation and frame work.

I am, however, a bit concerned about the value for fa, which was chosen as 7 in Appendix C, based on the data from the gallium experiments in the SIGhfA O

k.) 0 149 l

1

i facility. In Fig D 8, a value of f, a 7 will not produce a supercritscal steam ex.

pimion. Thus, the value of fa may be material dependent, and more information is needed to choose an appropriate value.

Yes, indeed. Figure D-8 shows results for tin at 1500 'C and we know (KROTOS, for example) that such premixtures do not give strong detonations.

Sch20. 111.4 VERIFICATION STUDIES The front part of the ESPROSE. m report describes the analytical tests, the SIGMA experiments, explosion coupling, integral aspects, numerical aspects, and finally, a comparison with the KROTOS tests. I will comment on these, briefly, individually.

!!!.4.1. Analytical fests These are very valuable exercises and show that the modeling in ESPROSE-m can predict pressure wave propagation. There are u.any gutes. I wish there were more explanations e.g., Figures 17 and 18, both show very good comparison between the analytleal and the ESPROS m pressure distributions for early times, but deviate at later times. Is there an explanation? Similarly, there is a crater in the iniddle of the pressure wave in Fig 19. Is there u physical explanation for that? This wetion may be improved by the authors, through some explanatory text. It is very valuable, otherwise.

The slight derivation in Figure.s 17 and 18 at later times and also the " crater" in Figure 16 (we believe that the reviewer meant Figure 16 in his comment) are due to the non linear effect at high pressure which is present in ESPROSE.m (the runs are made with the full code with real properties) but not in the analytical solution.

Sch30 111.4.2 SIGMA Experiments These experiments, spechdly conducted in the SIGMA shock tube provide data for verification of the ESPROSE m models for pressure wave propagation. The comparisons are excellent. There are some differences for the inhomogeneous cases, which, perhaps, are difficult to fix. Allin all, it is a splendid performance for the code for these separate-ellect tests.

G 150

, Sch31.  !!b4.3 Comparisons with KROTOS Experiments KROTOS experiments provide the rnost appropriate data for the verliication of the ESPROSE-m models. The KROTOS facility has performed steam explosion experiments by triggering the pre mixt ures of water with several ditferent material melts. The initial conditions, e.g. melt mas.t, melt temperature, melt superheat, pressure, water subcooling have been varini to provide a reasonably extenshv

' data base. The test pwgram is continuing, and could provide the data base needed for the ESPROSE-m validation. Unfortunately, as in most of these melt water interaction Integral experiments, the data obtained is integral and the premixing and tiw steam explosion processes are not delineated. Thus, detailed verification and validation of the ESPROSE-m (or any other steam explosion code) may not be possible.

The document provided on the analysls oi the KROTOS tests speculates that the melt breakup and quick freezing may be a reasonable explanation for the non explosivity of the Uranium oxide tests. We reached similar conclusions, and also, evaluated the c![ccts of the change in the surface tension and viscosity of the binary-oxide melt, as it cools down below the liquidus temperature. This has been reported in the 1995 ICONE meeting, and additional work will be reported t%

Q

• in the forthcoming FCI meeting.

Sch32. Coming back to the comparisons of ESPROSE m (using Phi ALPHA pre-mixtures) predictions against the measured data, the authors admit difficul-ties of representing particle freezing correctly in the Pht ALPHA formulation.

The fuel participation factor chosen affects the result greatly. The pressure wave shapes versus thne appear to be reasonable but there are differences e.g.

for K5 there appears to be an earlier venting of pressure wave. I believe, revision of the Phi ALPHA numerical scheme and/or modeling of the heat conduction in the fuel particles (as was mentioned earlier in the comments on Phi ALPHA modeling) may resohn this difficulty. The sensitivity to fuel participation factor is very large, indeed.

Disagree. The key ingredient here is the rate of breakup, which is not known, and will remain so. Also, as we discussed, there are intricate radiation reflection issues peculiar to the KROTOS geometry. These are not code issues; rather, they are physics issues peculiar to the test. The sensitivity to molten fuel content is real, not a code artifact. Again, this

/O V O 151

should provide important perspectives as to what is really predictable for these kinds of problems, and correspondingly what should be a viable strategy in safety assessments.

Sch33. IlL4.4 NutneticsLAstwets I have similar comments as I had for this topic in the Phf ALPHA document.

The authors should provide more discussion and, perhaps, comparisons of the use of the ICE technique for similar problems. The numerical diffusion issue is quite important when, ttacking pnssure waves andfor interfaces. Recently, special numerictd schemes have been devised to reduce or eliminate numerical diffusion. The ICE technique does not, compare well to such schemes, in term of its performauce, ar,d with respect to numerical di((usion. Perbaps, the nuthors have implemented another scheme in developing the ESPROSE-m 3D code.

Disagree. This comment is not consistent with the code perfo mance presented in this document.

Tur40. Comments on DOE /ID 10503: Propagation of Steam Explosions:

ESPROSE.m \*erification Studies Only a limited thne was available to seriew this supporting document.

hiuch of the document is concerned with the ability of the ESPROSE.m code to represent the unve dynamics correctly fo~ single and tuo phase regions in one and two dimensions. The information presented, along wath the comparisons with the SIGhlA experknents with a voided expansion region, indicate that this part of the code is Joing its job correctly, even when relatively coarse (~0.01 m) meshes oue used. This does not surprlse me. Numericalstudies we performed when extending CULDESAC from one to tav dimensions indicated good capabilities to capture the wave dynamhs with relathcly simple numerical schemes (the numerics of propegatIon are simpler than those of premixing). I am therefore satisfied w!th the code's capabilith~s in this area and would expect that the 3 D version of the code would also perform satisfactorily in this respect.

Tur41. While Chapter 2,1 uses a homogeneous model for the tuv-phase be-haviour (by forcing large drag between the phases), it is unclear whether the calculations reported in Chapter 2.2 still use this model. If not, it would be G 152

,, interesting to cornpare how inuch better the full modd performs against the ex-

) perimental data, compared with the homogeneous model.

The full two fluid modelis used for the calculations presented in Chapter 2.2.

Tur42. The authors of ESPROSE.m have implemented an, at the tirne, novel u;>proach to cover lack of thermal equilibrium in the coolant during the propaga-tion. This approach is ptysically based and can be considered to be wellJustified. i Tur43. The applicction of the ESPROSE.m code to steam explosions de-pends on the assumed constituthe physics. As Appendix D (particularly Sgures D8 and D9) lilustrates, the assumed parameters of the microinteraction model can have a major knpact on the prediction (eq changing the parameter for coolant entraintnent can change the C J pressures by two ordens of magnhude). Ap-pendix C provides results from a series of experiments with one high temperature simulant, that has been used to modify a hydrodynamic fragmentation model to take account of thermal effeus. This approach is acceptable, but the range of uncertainty in the model parameters needs to be allowed for in any assessment.

(O The authors note that 'the main need identified is for constitutive laws for ml.

crainteractions with reactor materials'[ Abstract) I agree. They also claim that

' reasonably conservathe assessments are possible'- however the ma!n report does not indicate what parameters were used to obtain a sufficiently conservathv as.

sesstnent.

See Appendix C and the addendum to it. Also, see new reactor calculations for different microinteractions parameters (see addendum to Chapter 6 of DOE /ID 10541).

Tur44. I would have expected to see more discussion of the comparison with KROTOS experknents in the report as originally supplied, rather than a reference back to the study. Although there are some limitations on knowledge of the inittal conditions and, in most of the tests with explosions, some loss of data, these provide the greatest confidence in the application of any model to the steam explosion propagation phase. The calculations for KROTOS 38 provided as a supplement are useful. With current knowledge it is more important to be able to demonstrate conservatism in the calculations rather than good agreement through parameter adjustment. Recently I saw calculations with TEXAS-IV for this test, n

I V) O 153

where a very different melt distribution was calculated that led ta very good agreement with the obscrted pressurisation folkaving the trigger. Until there is a visual record of such tests it is not possible to determine which simulation is closer to reality.

Tur45. In reading the material, i noted a number of examples where detail was not clear to me. These are listed here for convenience, but hasr no impact on my overall assessment of the methodology:

1. In chapter 2.1 what value is used for Pi ? Figure 6 a impiles 100 bar, but elsewhere finite results are ghrn when P2 is only 10 bar.

For results in Figures 1 through 5, Pi = 1 bar. For results in Figures 6 through 9, P2 = 100 bar.

Tur46, 2. In ligs 7 and 8 of Chapter 2.1 a is shown as varying, I assume a is a svid fraction - of which region 7 o is the void fraction of the region ahead of the shock.

Tur47. 3. Chapter 2.1 presents results with an without phase change of the gas. It uould have been instructhe to see a direct comparison to illustrate the knportance, or otherwise, of the phase change on wave propagation.

Results presented in Figures 4 and 5 are for a 10% steam / water mixture with phase change.

Tur48. 4. I had dilliculty understanding thelocation of the pre voided region discussed in Chapter 2.2. Note that Fig 7 is incorrectly refeired to as Fig 8 in the text.11for Fig 3 the pre volded region stretches to the ba oc of the tube, I do not understand the respecthe difference in timings of(A) th s time between the shock arriving at PT3 after PT1 and (2) the thne between the shock arriving at PT3 and its reflection from the base arriving at PT3. Note that you have oilset the pressures in the figures for case of presentation.

The typo on Figure 7 is corrected. The pre void region is the region between L = 100 cm to L ~185 cm, as shown in Figure 3. The bottom of the tank is at L = 300 cm. The region between L = 0 and L = 100 cm is the driver section, as shown in Figure 1 G 154

Typographical Errors

)

L,/

Dan 10. Typos:

p. 1 1 veriBed Fig. 1 viscosity 4-3 and IT Hugoniot von Neumann C-J point Typos corrected.

Der 7. N.B. There are some errors in table 3.2 as for the value of Ao which does not correspond to NE as written in the caption.

Typo corrected.

Dur20. 7. Some Comments for Formal Improvements Some typing errors of elevance are gh en below (I had no time for further detailed checidng), together with some need for detailed descriptions:

) e Ra' in Nomenclature: factor g is missing, g also missing in Nomenclature.

• P. 4 6, second line from below: 0.2 MW/m3 instead of m 3.

• P. 4 18: effecthe power density .. of 0.26 W/g: does this mean of core steriniin contrast to fuel?

e P.4 21, eq.(4.14): Ra'* **.

• P. 5-2: riving the breakup law wculd be helpful.

s P. 5-4: so.ne further informatloas on this and on the ather color Bgures would be helpful, e.g. length scales, quantitles of melt volume fraction.

• P. 5 5: z directed to top, but in subsequent results inversely, o P. 510,13th line from above: " melt" instead " coolant".

• P. 510, second line from below: "two sionvr pours" - seem to be better characterized by pours witr smaller diameter, m

xj 0 155

• P. G 1, second !!ne from below: " propagation intensity is basically indepen-dent af the magnitude of the trigger." it is not quite clear to me what is meant, e.g. with " propagation intensity".

• P. G 1, end of second paragraph: (Theofanous et al,1996a)?.

• P. 6-4: 6th line from below: 0.1m' e Fig. 6.2: should be better characterized: length scales, pressure scales e Fig. G.4: locations are identified in Fig. G.2?

e Fig. G.5: location of peak loading: where?, not included in 6.4!

• P. B.11 etc.: color characterization is not quite clear to me: fuel void, red lines, blue isolines in Jet.

• Appendix C: color pictures: case? characterizationst e choice of parameters in the ESPROSE calculations, especially concerning micro interactions?

e P. C.313 etc.: locations?

e P. D4,6th line from above: Run, instead Rk, S R 1 All typos have been corrected and requested information was supplied to the text. " Prop-agation intensity" refers to the strength of the explosion. We appreciate the reviewer's attention to detail.

Tur20, 10. Nom nclature: Equations 4.8 and 4.9 refer to Cr,wn while Figure 4.0 and 4.7 have Cras,w,1 etc.

Typo corrected.

Youl:. 4. On p. 5-10, it says "only a very small fraction of the coolant is found to coexist with the water"; should this be mdt?

Yes; correction made.

You12. 5. In the graph for C2-10 a on p. B2-7, the last time is given as 215 s instead of 0.215 s.

Typo corrected.

G-156

i stwng enough to fail the lower head, but it may be energetic enough to mechan.

Ically disrupt the blockages formed at the lower end of the core.

\}

The scenario proposed by the reviewer is not reasonable. First, there is no water in the downcomer. Second, an explosion cannot annihilate the lower blockages.

Cho5. Thus, the initial explosion, while not falling the lower head, could l severely weaken the blockages mechanically as well as thermally. It would seem possible that a relathely smallinitial explosion would be fo)) owed by a masshe downward relocation oicore melt tbrough the core support plate, setting the stage for a secondary explosion probably involving a much intger melt mass. The lower head may well survive such a secondary explosion, but a separate assessment of this possibility would definitely be needed.

Based on the above, thia kind of behavior is not physically reasonable.

Cor9. c) The biggest effect of these small pours in my mind is that they may cause local FCis which do not harm the RPV but totally change the melt pouring characteristics for subsequent melt pours; 1.c., thev small pours and associated FCis w'll damage the core melt crucible and markedly increase the melt flowrate f or chance its location. The authors have gone to great pains to determine the

• fragility of the RPV wall, but totally ignore the fragility of the melt crucibic and the effect of these FCis. I would suggest that larger melt pours will be induced from the bottom of the crucible as well as along its radially edges with larger holes, all caused by early small FCis. Ilow heve these events been considered or conservathcly bounded for RPV survival?

Subsequent events are very complicated, of course, but there are substantial solid barriers, even if the blockages were to structurally fail. Moreover, much of the water in the lower plenum would be expelled by the same forces (if present) that are considered to fall the crucible.

Jac6. Finally, the possibility of a large coherent steam explosion that is in-duced by a smaller one (e.g. one of those considered) is completely left aside.

Such event might proceed in different ways. The common starting point of these would be the mechanical destruction of the crust keeping the melt pool. This might be caused directly by the action of the pressure of the first steam explosion g

V G 161

or indirectly by the pressure of anot:.sr melt coolant interaction due to the addl. l tion of some water into the upper zone or on top of the melt pool. The induced steam explosion would then occur either within the core volume (if there were still water left) or in the lower plenutn after the melt released from the broken melt pool has dralned through the still open holes in the lower grid plate. It is sometimes argued that such melt couldn't encounter water in the lower plenum because that would have been drhrn away by the initial steam explosion. How-ever, the first (weak) explosion might have caused essentfally a sloshing movement of the water so that this could mix very etlectively with the corium streams when returning. In this context one should also keep in mind that with a large molten corium mass available and melt water interactions occurring, large amounts of mechanical energ may become available. So it is often hard to argue that some process was unlikely.

These comments do not take into account the geometric features of an AP600 core trapped above the massive core support plate. Nor do they take into account the highly localized, in both space and time, pressure pulses predicted. Water on top of the melt is .mimportant in this respect for the same reason that we are not concerned for lute iC!s, as explained in Chapter 9. Moreover, as explained in this chapter, water on top of the molten core is not physically relevant in the AP600.

Tur7. G. I need to be convinced that we need only to be worried about thefirst rvloca: ion event. I think that it depends on the timing of any subsequent relocation events.

Once the path opens, subsequent relocatir,ns should be essentially continuous.

Tur35. 2. I agree that there is a greater threat from subcooled conditions. It is not obvious, though, that a 'hlghly subcooled pool'is necessary. Perhaps this might be illustrated by a calculation with modest subcooling (eg 10 degrees) to show there is not threshold effect.

We now have such a calculation (see addendum). However, please note that the possi-bliity of creating subcooled conditions has been addressed in Chapter 4. In this context, generating a highly subcooled c sndition is just as difficult as generating a 10 K subcooling.

G 162

Youe, Third, the lack of a stratlSed mixing case bothers me, or rather, the Q) fack of data to properly model this case. I do not doubt that what the authors say is correct: if ESPROSE.m were run with a stratlSed case, it would probably produce a very nonenergetic steam explosion for the reasons stated. However, if memory serves, a stratlSed explosion in a foundry imviving water dripping into a " car" containing molten fron produced an explosion strong enough to take out some of the walls of the plant. This incident seems in contrast to what would be predicted with ESPROSE.m, although the melt is different (hon versus reactor material) and the water was undoubtedly subcooled. The incident mentioned probably im.oh cs mixing of the stratlSed material caused by the province of PM-ALPflA7 Maybe the authors could comment on this, or maybe it Indicates the need for ane stnalBed experiments.

The incident mentioned occurred at the Farthingham Foundry. It involved water pouring in a steel " torpedo," already pretty full of molten steel, through a small opening on top. The explosion occurred as the torpedo was set in motion, to transport the melt (apparently not realizing the accidental presence of water). The water became trapped in a highly confined geometry and pressures developed made the torpedo explode. This type of phenomenon is not relevant to the stratified interaction considered here.

O O You0, Fifth, in the section on stratlSed layer of molten stel and renood, Chapter 8, there appears to be one piece missing to make the case that the scenarlo is hnpossible: the thickness of the crust is not mentbned. SpeciScally, is this a " thick" crust that is stable, or is it a " thin" skin that could be broken; The thickness of the crust will depend on the time after addition of water. It will get thicker with time, and hence less likely to allow contact, as more water accumulates with time.

,a 0 159

"Other" (Subsequent) Explosion Scenarios (Subsequent)

Dan 7. All the premixing unrk that I am familiar with (including my own) assumes that a corlum Jet falls into the lower pool by melt through of the LSP

' (louer support plate). A great deal of uvrk has bwn done (and is contemplated for the future) onjet breakup, with various materials, Jet diameters, temperatures and velocities, that is predicated on this assumption. The unexpected result that for the AP 600 melt tbrough will occur through the side, rather than the bottom, of the melt pool, is therefore of first importance. Of course, this happened at TMI, but in AP-600 there is a thick vtainleas-steel reBector around the core. The code capability allows calculation of the subsequent premixing and melt / water dynamics. There can be no large melt Jet in transit; stratlEcation appears very likely even in transit; and the whole scenario of demaging multiple explosions disappears.

Der 3.

• Another important argument is that < because of extensive vold-ing, we need only be concerned about the first relocation event, and only for early trigger in it ;>. This seems to se justlRed by the premixing and explosion calculations presented later but I wonder why, after a first event, when water is slashing hek a second event cannot occur at about the same location where the structure will has been already dynamically loaded and eventually alr=ady defonned by the first event.

As seen in Chapter 7, the structure response remains within the dastic limits. Subsequent relocations in the scenario proposed by the reviewer would be too gradual, and water would be quickly depleted from the lower plenum.

Bilr17. I think, this could alco be done for the situation of bottom failures of the melt pool, excluded in the report.

We also believe that bottom failures could not jeopardize lower head integrity, but prefer to not open up this direction of thinking arbitrarily; that is, without a viable blockage failure mechanism.

Cho3. Suppose a steam explosion would take place in the downcomer region or in the lower plenum, as described in the report. The explosion may n't be G 160

I Stratined. Reflood Explosions

- Bur 38. Concerning the Interaction of rcBood water with melt pools, I agree to the argument of rapid small-scale interactions, rapid solidlBeation and in ten-eral small eHectivity of stratlSed explosions. However, it should be addressed

. whether relevant cWects of mixing due to the water impact and due to small scale interactions (also taking into account the falling back of expelkd water) can be excluded. The situation of renood under conditions of still existing melt / water  ;

mixtures in the lower head may oc even more important than the extreme of the melt pools in the loner head, if the reBood water could enhance mixing again, collapse steam, favor further melt release 19 the above processes, and this under  ;

the conditions of already settled melt, i.e. thermal load at the bottom.

Reflood scenarios in this pressure reactor are not arbitrary in their timing, but rather well-defined, as explained at the end of Chapter 4 and in Chapter 8. On this basis we have  ;

considered all that needs to be considered.

Cho9. Renood FCis urre discussed in Chapter 8. I suspect that renood FCis in stratined conBgurations would be of secondary importance compared to the premixed exploslens addressed in the rest of the report. Nevertheless, renood FCis need to be considered for completeness, particularly in view of the potential for vessel wall thinning due to chemical attack by the metallic melt.

The authors should be commended for making an e80rt here. I nvuld have to say that this edart represents a best estimate assasment based on engineering Judgment. At present, them is no adequate database or computational tool for  ;

large scale stratlBed explosions.

Disagree. The assessment is based on basic principles. Simply put, you cannot create any significant impulse with an inertia constraint of a few inches of water!

Fle9. 2.8 Consideration of Renood FCis This is run important section, as the above analysis has clearly shown that pre-mixed explosions cannot cause failure of the lower head. I agree with the view taken that you need a very substantial overlying water pool to provide suBicient inertial constraint to generate a high pressure explosion. As in the previous sco-natio everything is against this, viz. the low water addition rates, the case with 0 157

I which the snelt surface freezes and the fact that as Blm boiling occurs the overly- <

ing pool will develop volds reducing its ability to a>nstrain. The analysis rules out  :

to rny satisfact'on the possibility that stratified explosions could fail the vessel.

Sch52. IV.6 Chapter 8. Consideration of the reBood FC1's This chapter deals bricBy with the stratine' steam explosions that ruay usult, if the renood is eEectin, and a layer of water is I: .%t on top of the snelt pool, which has a snetallic layer, at top.

It was found that the stable water layer may not exceed 10 cms, due to the low renood rate and the t1.ne to freeze the upper maallayer. Any stratified explosion will be easily vented.

I believe, the authors have a good argument. Certainly the peak pressures in such an explosion should be low and reBood FCI s may not be a proba.

IV.7 Chapter 7. Integrat su, Assessment / Chapter 9. Conclusions These chapters combine the results achieved in the previous chapters and appen-dices to provide an overall assessment. This work was already practically done l>y the resulta acidered, slncs the maximum impulse loading was below the min-knum of the fragih*y cune. This was also confirmed by performing ABAQUS calculations for the peak loading for the actual cases and Bnding that the lower head strains were very low.

The nuthors conclude that for the saturated water case, the lower head integrity can not be compromised by a steam explosion. Having highly subcooled water is the only possible way to, potent'. Jly, involve a larger mass of melt, and produce a more energetic explosion. The authors conclude that obtaining highly-subcooled water, even in renood scenarios for the AP-600 is not credible.

• * * * * * * . * + * * * * * * * * * * * .

• Tur30. Chapter 8: Consideration of Renood FCis

1. This chapter has not been considered in any detail. The arguments presented appear persunshe provided that are no other means of fast renooding not con-sidered by the authors and that crust format!on proceeds in the way that they envisage.

G 158