ML25052A259
| ML25052A259 | |
| Person / Time | |
|---|---|
| Issue date: | 03/19/2025 |
| From: | Leslie Fields Licensing Processes Branch |
| To: | Jennivine Rankin Licensing Processes Branch |
| Shared Package | |
| ML25052A258 | List: |
| References | |
| EPID L-2024-NFN-0005 | |
| Download: ML25052A259 (1) | |
Text
March 19, 2025 MEMORANDUM TO: Jennie K. Rankin, Acting Chief Licensing Projects Branch Division of Operating Reactor Licensing Office of Nuclear Reactor Regulation FROM:
Leslie C. Fields, Senior Project Manager
/RA/
Licensing Projects Branch Division of Operating Reactor Licensing Office of Nuclear Reactor Regulation
SUBJECT:
SUMMARY
OF THE DECEMBER 4, 2024, CLOSED PRESUBMITTAL MEETING BETWEEN THE US NUCLEAR REGULATORY COMMISSION AND THE PRESSURIZED WATER REACTOR OWNERS GROUP TO DISCUSS A CONCEPTUAL ON-LINE MONITORING PROGRAM FOR TECHNICAL SPECIFICATION INSTRUMENTATION (EPID L-2024-NFN-0005)
On December 4, 2024, the U.S. Nuclear Regulatory Commission (NRC) staff held a closed meeting with representatives from the Pressurized Water Reactor Owners Group (PWROG) as a follow-up discussion to the June 3, 2024, preapplication audit concerning a proposed on-line monitoring (OLM) software program for its technical specification (TS) instrumentation. The meeting notice and agenda for the closed meeting are available in the NRCs Agencywide Documents Access and Management System (ADAMS) at Accession No. ML24327A054. The meeting attendance list is enclosed in this summary.
Due to the proprietary nature of the discussion, the meeting was closed to the public. The main purpose of the meeting was to obtain NRC staffs preapplication feedback prior to submitting a topical report (TR) for NRC review on a proposed OLM software program for its TS instrumentation. The PWROG representatives provided information on a proposed TR and presented a rationale for performance monitoring of the TS instrumentation to be used as the basis for establishing channel calibration frequency.
The meeting topics included an overview of a conceptual OLM program which included:
(1) summary of the PWROG OLM program for the TR, (2) key considerations for the OLM program, and (3) information regarding the use of machine learning to establish the frequency for surveillances that will be included in the TR submittal. The NRC staff engaged in several preapplication topics that should be considered when submitting a TR on this subject matter for NRC acceptance and approval. The PWROG representatives stated that the TR is expected to be submitted in the second quarter of 2025 with an application request for a standard NRC review.
I.
NRC Items to Consider for Topical Report Implementation With respect to implementation for this type of TR, during the meeting there was discussion about three possible steps to consider which could be combined (or partially combined) in an NRC safety evaluation as follows:
(1)
Approval of an On-Line Monitoring methodology This pertains to a methodology described in the PWROG OLM TR which could include both generic open items (e.g., criteria for software implementation of the methodology) and plant specific open items (e.g., criteria for plant specific implementation of the methodology).
The NRC staff indicated that in this scenario, engineering success is implementing the methodology at a plant, not only approving a TR. Therefore, a TR where the methodology cannot be implemented, is not a viable path.
If an applicant wants to develop a TR that is later implemented in software, then the applicant must specify the acceptance criteria or process methodologies that will be followed to ensure the software implementation is acceptable. At a minimum, the TR needs to establish the road map or describe the path to address the implementation of software.
In response the PWROG representatives stated that the software packages that are available to implement this program will be commercial software packages. Commercial vendors have optimized their kernels, and generally, when reviewing algorithms, it is the kernel itself that will be the trade secret or the proprietary element that vendors are offering. Therefore, the kernel element is the proprietary part of the algorithm including how the kernel implementation is selected.
During previous meetings on this topic there was discussion about quantitative uncertainty in the predictions (along the same lines as outlined and referenced in the NUREG/CRs) which will be the primary metric of the uncertainty in the predictions from that algorithm. It was mentioned that uncertainty is then used to choose the alerting limits that would identify a behavior that is outside of a normal operation.
When choosing an implementation plan, there will be uncertainties. If those uncertainties are too large, then the algorithm simply won't leave enough margin in the alerting limits, resulting in nuisance alarms and unnecessary instrument calibrations. In that event the program will lose its economic utility. However, the safety function will never be compromised, because in that event, it continues to unnecessarily calibrate safety-related instruments.
(2)
Approval of a software package This pertains to a software platform described in the PWROG OLM TR to implement the approved PWROG OLM TR methodology (which could address generic open items and plant specific open items from the PWROG OLM TR), and also including plant specific open items (criteria for plant specific implementation of the methodology and software).
NRC staff indicated that in previous meetings, the PWROG proposes that its general approach is to consider the software to be maintenance and test equipment (M&TE) and to hold it to M&TE standards, instead of safety-related protection system standards. The NRC staff asked if this is still the plan or will the OLM software be considered safety-related software. Additionally, there was an inquiry asking if the software will be developed under the Appendix B program.
The PWROG representatives stated that it is their assumption that this is not safety-related software, and it is not Appendix B software. They stated that the software is more like M&TE because most of the applications that have been examined are based on screening data that's available on the Business Network. The software analyzes the Business Network data and advises if a human technical evaluation is required to determine whether or not physical calibration is needed.
(3)
Review of Plant-Specific Implementation of OLM This pertains to plant specific implementation of OLM which would need to address all remaining generic open items and plant specific open items.
The PWROG representatives stated that generally, there are two parts to the plant-specific application of an advanced pattern recognition associated with the way the PWROG is proposing to use it: (a) a model training portion, and (b) an OLM portion.
- a. The model training portion fundamentally comes down to using machine learning methods to evaluate a set of historical data. This is a curated set of historical data. It will typically be drawn from a fuel cycle where we know that the instruments used in this model have been in calibration. For example, when all the instruments in a model were found to be acceptable at the beginning and the end of a particular fuel cycle, that would provide confidence that the instruments were in calibration throughout that historical period.
The next step is to collect new snapshots (called observations of the data) periodically. Typically, those will be drawn at a 5-minute or 10-minute rate period.
Given that set of data, which is all the data in the model at a particular point in time or sample point in time, the licensee can then estimate or predict what the expected value of each of the inputs should be when the system is operating in a characteristically normal mode or behavior. Statistical tests are applied to the difference between the predicted values and the actual values which will identify all the failure modes of the instrument except for loss of dynamic response which will be handled separately.
- b. There was also discussion regarding the plant-specific application about the OLM portion.
II. Significant Details to Consider for the OLM Methodology/Criteria for NRC Review During the meeting, the NRC staff stated that in order to achieve the best results in the development of the TR evaluation, the OLM methodology/criteria in the TR submittal should contain the appropriate details to achieve the following:
(1)
Ensure Adequate Coverage Specify the plant states, instrument functions, and failure modes covered by OLM and the models intended to provide that coverage. For example, what combination of (multivariate) models are needed to adequately cover a single instrument channel or function for a given failure mode for a given plant mode (state)? One unknown is that relationships among variables in the multivariate model have some degree of state dependence (think of plant modes in which different plant equipment is in operation).
Building one model to represent all possible plant states is extremely difficult so the PWROG may wish to be clear on the methodology for building an adequately representative set of models: 1) one for all states, 2) one for each state, or 3) some combination that provides coverage for all states.
(2)
Mitigate False Negatives False negatives (e.g., the failure to alert for a potential instrument performance issue) can have a safety impact. False positives (e.g., nuisance alarms) are an operational concern. Each model is developed using data from past periods of plant performance and is intended to represent the current plant state. However, plant states will continuously change over time (e.g., pipe fouling, discrete changes, equipment replacement or failures). When the plant state and model state diverge sufficiently alerts are generated. Assuming that the model performance is fixed, alerts will be generated by changes in plant state. There are several possible cases related to alerts as follows:
- a. Plant state does not change (and model matches plant state), true negative (desired OLM functionality)
- b. Plant state changes resulting from instrument performance issues, true positive (desired OLM functionality)
- c. Plant state changes resulting from a non-instrument performance issue, false positive (nuisance alarm and operational issue)
- d. Plant state changes resulting from instrument performance issue, but divergence with model is masked by other non-instrument performance issue, false negative (potential safety issue)
A basis for avoiding or mitigating case (d) would be helpful.
(3)
Explicitly Document Assumptions Because many elements of this OLM approach are novel, clearly documenting assumptions used in the TR would be helpful to ensure the NRC understands the context and application of OLM. Examples include how OLM will detect only persistent plant conditions (just as past calibration approaches have) or advanced pattern recognition will be implemented with an auto associative kernel regression algorithm exclusively.
(4)
Setpoints The terminology used in the area of setpoints should align with NRC terminology.
The NRC staff indicated that the use of the terms as-found and as-left can be confusing because there are two different meanings for these terms. One meaning is found in the footnotes of the NRC traveler, TSTF-493, Revision 2, Clarify Application of Setpoint Methodology for LSSS Functions (ML071060059), which uses the terms as-found and as-left to refer to limits calculated. The second meaning for the terms is derived from how the terms are used during the calibration process; during an individual calibration, the terms as-found and as-left refer to values determined for that specific instrument for that specific calibration. The PWROG OLM TR write-up uses the same terms as found and as left referring to individual instrument calibration rather than the as found and as left limits in TSTF-493.
The NRC mentioned that it would be beneficial if the TR submittal included a clearer mapping of the upper calibration limit in the PWROG OLM TR to the Regulatory Guide 1.105, Setpoints for Safety-Related Instrumentation (ML20330A329) or TSTF-493 as left tolerance interval or as found tolerance interval. There is interest in how the OLM is consistent with the limits established in setpoint control.
The PWROG representatives stated that this could easily be accomplished in the TR submittal.
III.
Discussion on the Draft Topical Report Submittal and Technical Specification Task Force Traveler The PWROG representatives mentioned that they will prepare the OLM TR, and the information discussed during this meeting will be incorporated into a draft TR which will be submitted. The PWROG representatives also suggested that they will request a future pre-submittal meeting to discuss that draft TR.
The OLM TR can be used to address the technical specification instrumentation that will be monitored by the OLM software. The OLM TR will provide justification that Common Mode drift is not an issue because all channels in a function are monitored separately as opposed to against each other. The OLM TR will not propose a backstop.
On a plant-specific basis, the functions will be identified based on plant-specific operating experience information.
The OLM software will determine if there's an issue with a channel such that it will be investigated, which could include response time degradation. Therefore, existing response time testing, elimination remains applicable, and the specific instrumentation that cannot be included in the OLM program will be determined on a plant specific basis.
A draft TSTF traveler will be submitted for NRC review after the NRC accepts the TR for review. The traveler will add the OLM program to the existing surveillance frequency control program.
At the conclusion of the meeting, all participants were informed that the feedback provided during the meeting met the goals and objectives of the meeting. No regulatory decisions or commitments were made during the meeting.
Please direct any inquiries to me at (301) 415-1186 or by email at leslie.fields@nrc.gov.
Enclosure:
List of Attendees
LIST OF ATTENDEES
SUMMARY
OF THE DECEMBER 4, 2024, CLOSED PRESUBMITTAL MEETING BETWEEN THE US NUCLEAR REGULATORY COMMISSION AND THE PRESSURIZED WATER REACTOR OWNERS GROUP TO DISCUSS A CONCEPTUAL ON-LINE MONITORING PROGRAM FOR TECHNICAL SPECIFICATION INSTRUMENTATION Enclosure Nuclear Regulatory Commission (NRC)
Name Organization Leslie Fields NRC Gerond George NRC Norbert Carte NRC Matthew Dennis NRC Ismael Garcia NRC Doug Eskins NRC John Hughey NRC Brian Green NRC Alex Prada NRC Tim Marshall NRC Mario Fernandez NRC Robert Pascarelli NRC Shivani Mehta NRC Alexander Schwab NRC Amy D'Agostino NRC Derek Halverson NRC Roy Hardin NRC Sergiu Basturescu NRC Luis Betancourt NRC Jennie Rankin NRC Nicolas Mertz NRC Anita Ghosh Naber NRC Kobe Oley NRC Cherish Alston NRC Elizabeth McAndrew-Benavides NRC Tanvir Siddiky NRC Shyrl Coker NRC Tammie Rivera NRC Julie Ezell NRC Tarico Sweat NRC Michael Brown NRC Ralph Costello NRC Eric Nour NRC Daniel Daszkiewicz NRC Kerri Kavanagh NRC Deanna Zhang NRC Kim Lawson-Jenkins NRC Rodney Fanner NRC
LIST OF ATTENDEES
SUMMARY
OF THE DECEMBER 4, 2024, CLOSED PRESUBMITTAL MEETING BETWEEN THE US NUCLEAR REGULATORY COMMISSION AND THE PRESSURIZED WATER REACTOR OWNERS GROUP TO DISCUSS A CONCEPTUAL ON-LINE MONITORING PROGRAM FOR TECHNICAL SPECIFICATION INSTRUMENTATION PRESSURIZED WATER REACTOR OWNERS GROUP (PWROG) Representatives James Andrachek Westinghouse Electric Company (Westinghouse)
Chad Holderbaum Westinghouse John Moorehead Westinghouse Jay Boardman Westinghouse Dewey Olinski Westinghouse Jeff Hudson Aep Randy Bickford Expmicrosys
ML25052A258 (Package)
ML24208A000 (Audit Report Summary, 06/03/24)
ML24327A054 (Meeting Notice)
ML25052A259 (Meeting Summary)
- eConcurrence NRR-106 OFFICE NRR/DORL/LLPB/PM NRR/DORL/LLPB/LA NRR/DEX/EICB/PM NAME LFields DHarrison NCarte DATE 02/24/2025 02/25/2025 03/10/2025 OFFICE NRR/DEX/EICB/BC NRR/DORL/LLPB/BC NAME FSacko JRankin DATE 03/17/2025 03/19/2025