ML21242A031

Enclosure - Staff Responses to PWROG Peer-Review Comments on NRC Spar Parameter Data

ML21242A031
Person / Time
Issue date:	08/31/2021
From:	Mark Thaggard NRC/RES/DRA
To:	Nowinowski W Westinghouse
Lane, John
References
Download: ML21242A031 (36)

Text

STAFF RESPONSES TO PWROG PEER-REVIEW COMMENTS ON NRC SPAR PARAMETER DATA

This report summarizes the Nuclear Regulatory Commissions (NRCs ) response to issues raised concerning parameters developed for use in probabilistic risk assessments (PRAs). In support of their concerns, the Pressurized Water Reactor Owners Group (PWROG) requested a series of four meetings between NRC staff and members of the PWROG to elaborate on their proposed suggestions towards improving NRCs operating experience (OpE) data analysis program, as contained in their associated report, entitled, Component Reliability Data Issues for Discussion with NRC Research, PWROG -18029-NP (non-proprietary), Revision 1, Agency Documents Access and Management System (ADAMS) ML20279A597. The PWROG report represents the culmination of a significant industry assessment of the NRCs OpE Program managed by the Office of Nuclear Regulatory Research (RES).

The first meeting was held at NRC headquarters on January 7, 2020. At that meeting, the PWROG provided an overview of the issues raised in the PWROG report. The areas addressed represented a broad spectrum of issues in data reliability, common cause failure and uncertainty estimation, among other topics. Subsequent meetings on these same topics were held on March 19, 2020, April 9, 2020 and June 9, 2020 virtually using Skype. The meetings discussed a number of areas where the staff evaluates individual plant operating events, as available in the proprietary Institute for Nuclear Power Operations (INPO) Industry Reporting and Information System (IRIS) database and in licensee event reports (LERs). The operating events help form the basis for the parameters developed for use in PRAs such as the NRCs Standard ized Plant Analysis Risk (SPAR) models as well as industry PRA models.

To facilitate discussions, the PWROG report also provided a peer review-type template that organized the issues into categories to facilitate NRC staff responses to the issues and to serve as a convenient means to enable potential future engagement.

The staff recognizes the need to maintain confidence in all aspects of the OpE data analysis program owing to both industry and staff reliance on it as a key input to quantitative risk models employed by both groups. As part of the resolution pathway, the staff has documented its findings using the aforementioned template to provide NRC responses in a convenient format and to promote continued and increasing confidence in the NRC OpE data analysis program which, itself, is an integral part of NRCs risk -informed regulatory oversight strategy.

Tables F-1 through F-4, in this report are adapted from the tables contained in Appendix F,

Observations Regarding NRC Reliability Dataset PWROG-18029NP, Revision 1. The tables include the PWROG peer review comments, as summarized in the PWROG report appendix, and the addition of the column containing the NRC response to the observations.

• Table F-1: PWROG Peer Review Observations and NRC Responses Related to common cause failure (CCF) Issues (15 observations of common cause failure issues from CCF.1 to CCF.15)

• Table F-2: PWROG Peer Review Observations and NRC Responses Related to Data Quality (18 observations of data quality (DQ) issues from DQ1.1 to DQ9.1)

• Table F-3: PWROG Peer Review Observations and NRC Responses Related to Data Aging (5 observations of data aging (DA) issues from DA1.1 to DA5.1)

• Table F-4: PWROG Peer Review Observations and NRC Responses Related to Data

1 Classification (10 observations of data classification (DC) issues from DC1.1 to DC10.1)

The tables are similar to Facts and Observations as typically documented in nuclear industry peer reviews and include the following fields, all of which except the last one, were provided by the PWROG:

• ID: a code that helps to connect the observation to the original data issue

• Observation: a description of the data issue identified during the PWROG review of the NRC reliability datasets

• Basis: the criterion that the data issue is being judged against

• Potential Resolutions: PWROG proposed options for addressing and resolving the issue

• References : Appendices A through E in PWROG -18029-NP, Revision 1, provide more detail and examples to illustrate the issue. The references are:

o Appendix A - original detailed documentation of each data issue o Appendix B - presentations from January 7, 2020 meeting with NRC o Appendix C - presentation from March 19, 2020 webinar with NRC o Appendix D - presentation from April 9, 2020 webinar with NRC o Appendix E - presentation from June 9, 2020 webinar with NRC

• Priority: PWROG ranking of the importance to utility PRAs: High (H), Medium (M), or Low (L)

• NRC Response/Resolution: NRC response to PWROG observations and planned resolutions. NRC responses are made in close concert with advice provided by its primary data contractor, the Idaho National Laboratory (INL), who maintains the Industry Data Collection and Coding System (IDCCS) database and provides the underlying statistical and engineering support for the parameter development.

In addition to Tables F-1 through F-4, two additional tables are included here :

• Table DQ 1-1 documents the pairs of duplicate entries that contain different IDs as observed by PWROG. The column, NRC Response, was added to the table providing the NRC evaluation as to whether the pairs are duplicate entries. In the event of duplicate entries, the responses provide the resolution which is to remove the duplicates in the next parameter update.

• Table DQ 1-2 documents miscellaneous issues related t o the 2015 NRC dataset observed by PWROG. The column, NRC Response, was added to document the NRC evaluation results and NRC actions planned.

This report constitutes the staff response to issues identified by the PWROG. As these issues are discussed further or as new issues are raised by the PWROG or other stakeholders, future revisions to this report or issuance of new reports may be considered.

2 Table F-1 PWROG Peer Review Observations and NRC Responses Related to CCF Issues

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.1 The CCF Dataset (2015) includes a number of The common cause Assure that component NEW M AGREE: FUTURE ACTIVITY PLANNED component-types that do not have equivalent component-types and failure modes for which Both the NRC parameter estimations in entries in the NRC Dataset (2015). failure modes should be CCF parameters are component reliability (or the NRC Dataset as For example, the CCF Dataset includes heat defined consistent with calculated have called in the PWROG observation) and in CCF exchangers for PWR, containment spray, BWR the component-types and equivalent entries in the provide data needed by SPAR modelers (and RHR, BWR isolation condenser, and CCW, failure modes in the component reliability senior reactor analysts ). The reliability and while the NRC Dataset includes heat reliability dataset. dataset. If necessary, CCF parameters usually have the same exchangers for pooled, CCW, and CCW non-combine CCF component -types and failure modes, but could extreme environment. component failure vary with each other (e.g., with more, system -

modes or create specific parameters for CCF) due to the needs This distinction in component -types is not separate component from different SPAR models or specific case consistent with the basic definition of the CCF reliability failure modes. study and analysis. There should be no parameters. pro blem as long as the SPAR modelers can properly apply the right data to the corresponding basic events in the models.

Considering modeling needs, the staff will make both dataset s more consistent to each other going forward.

CCF.2 The prior distributions u sed for CC crs The priistributis Pride clear PtoOG - M AGREE: ACTIVITY COMPLETED cumentn Sectio1.oCCc CCcs are of documti 18029, NRC understands the CCF prior (2015) with vyittlplanatiorng criticmportance f CCc priistributis, Appendix documentation issue and has spent efforts to howye dev. caile modes iludiourc(Slide 5) develop INL/LTD -17 -43723 Developing where the data is pars and bes Generic Prior Distributions for Common Cause stributis Failure Alpha factor s and Causal Alpha tnive distributis factor s in November 2017. This report thateoidere reviewed the existing process to develop and rejted. generic prior distributions and developed new priors for alph a factor s and causal alpha factor s with data from 1997 to 2015.

The report has been revised and is undergoing internal review prior to public distribution. In the meantime, A PSAM 14 conference paper Developing Generic Prior Distributions for Common Cause Failure Alpha factor s was published in 2018 and publicly available for the CCF prior development.

3 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.3 The prior distributions used to generate CCF The priistributis Consider other t H AGREE: ACTIVITY COMPLETED cr s are critic f cponentaile modes CCcs are of ch Generic priors and cause-specific priors have where the data ispars, the criticmportance f generating CCc pri been developed in the past few years.

thcesreating sucis hav caile modes stributis. AGREE: FUTURE ACTIVITY PLANNED not beed. where the data ispars ame, develop Component-specific priors are planned for mtii development in 2022.

stributis,e s of cail modt have se cm attribut en CCcs csidered.

CCF.4 The prior distributions used for CC crs The prior distributions for Consider other NEW M AGREE: FUTURE ACTIVITY PLANNED umted in Section 3.oCCc CCF parameters are of approaches to NRC agrees that the prior distributions as well (2015),ncluxtreme rs of stritio critical importance for generating CC as the generic demand and rate distributions fre cmae component component failure modes factors, especially for include extreme ranges of distributions for

) s factors. For where the data is sparse. large CCCGs and for some CCCG sizes and alpha factor s.

example: component failure NRC will investigate alternative approaches to a4 for CCCG=4 has a RF = (1.5e-2/3.8e -4) = modes with sparse CCF better characterize the CCF parameter 39.5. data. uncertainties. Suggestions from the industry a6 for CCCG=6 has a RF = (2.4e-3/3.1e-7) = are welcomed.

7742.

a8 for CCCG=8 has a RF = (6.0e-4/5.4e-11) =

1.1e7 These extreme ranges illustrate the lack of knowledge in the likelihood of common cause failures in large groups. It is not clear that such distributions are meaningful and whether Bayesian updating of such distributions is meaningful.

4 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.5 The pooled distributions used for CC factors Component failure rate Revise the calculation NEW M AGREE: FUTURE ACTIVITY PLANNED umted in Sections 3.1 & of NoC entries that pool data of CCF parameters for Same response and action item in CCF.4.

CCc2015),ncluxtremyranges from diverse component pooled data so that the distributionsromome CCCGizes an types should account for increased levels of non-actors (in comparison to the generic priors). the increased levels of homogeneity are better For example, for rate: non-homogeneity in the reflected in the a4 for CCCG=4 has a RF = (7.8e-3/ 5.4e -3) = uncertainty distributions. uncertainty 1.4 distributions.

a6 for CCCG=6 has a RF = (3.1e-3/ 1.8e -3) =

1.7 a8 for CCCG=8 has a RF = (1.0e-3/ 4.3e -4) =

2.3

These pooled distributions do not account for the uncertainty due to the increased level of non-homogeneity in this data.

CCF.6 CCF events that occurred in lower operational Weighting factors should Include a weighting PWROG-M DISAGREE: NO CHANGES PLANNED modes (e.g., Mode 3, Mode 4) are categorized be applied to each CCF factor that accounts for 18029, The NRC CCF database includes a field for the without any consideration of the lower event to properly account the reduced likelihood Appendix CCF event operational status that indicates likelihood that the event would have occurred for its applicability to that the event at lower D, CCF when the CCF event occurred or could occur.

during nominal full - power conditions. For baseline PRA models. operational modes Events For example, a CCF event could be detected example, CCF event #384. would have occurred during plant shutdown but could occur during CCF Event #384: Both EDGs unavailable due during nominal full - both power operations and shutdown to improper switch position. This condition power conditions. conditions. We do not quantify the likelihood of occurred during Modes 4 & 5 when much a shutdown event could occur during power longer time is typically available to locally start operations though. The PWROG example the EDGs compared to Mode 1. The condition does not explain how the CCF event could was recognized and corrected within 7 hours8.101852e-5 days <br />0.00194 hours <br />1.157407e-5 weeks <br />2.6635e-6 months <br />. have a lower likelihood to occur during power operations. We are currently n o t consider ing revis ing the CCF coding and parameter calculating process as proposed by the PWROG.

CCF.7 It is not clear how CCF events that occur due Weighting factors should Provide an explanation PWROG-M AGREE: ACTIVITY PLANNED to debris in the UHS are modeled to account be applied to each CCF for how CCF events 18029, NRC is developing a report of Loss of Service for the environmental condition. For example, event to properly account that involve Appendix Water Quantification 1998 to 2019. This item CCF event #515. for its applicability to environmental D, CCF will be closed upon the completion of the CCF Event #515: Circ water traveling screens baseline PRA models. conditions are expected Events report.

C and D failed due to sheared pin. The failures to be used. For occurred during a seaweed intrusion. The pins example, would the were an incorrect size. system model include the likelihood of the extreme environmental condition by modeling this as an initiating event (e.g., loss of circulating water)?

5 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.8 CCF event #520 (see below) includes use of Weighting factors should Document how the PWROG-M AGREE: ACTIVITY COMPLETED the Failure Mode Applicability factor (0.1). be applied to each CCF Failure Mode 18029, Section 5.1.16 of NUREG/CR -6268 describes Since this appears to be the only CCF event event to properly account Applicability factor is Appendix that Failure mode applicability represents the that uses this factor, it is not clear what for its applicability to defined and used in D, CCF percentage of specific failure modes for condition it is accounting for. baseline PRA models. CCF event weighting. Events multiple comp onent failures involved in the CCF Event #520: RHR pump A and C Consider expanding the CCF event. This is a weighting factor for minimum flow valves were found sealed shut use of the Failure Mode parameter estimation for a CCF event involving rather than sealed open. This was corrected in Applicability factor to multiple failure modes... For CCF event #520, a timely fashion due to the associated LCO, other CCF events. th e Failure Mode Applicability Factor of 0.1 with the entire Inoperable time lasting about 3 was coded incorrectly and was switched to 1.0 hours0 days <br />0 hours <br />0 weeks <br />0 months <br />. based on LER 2982017001.

CCF.9 CCF event #526 (see below) includes Weighting factors Provide an explanation PWROG-M DISAGREE: NO CHANGES PLANNED P_values of 0.5 for all three (3) components. should be applied tofor how the P_values 18029, The P_values were not based on the However, it is not clear what the bases for each CCF event to were selected for this Appendix potential for recovery, but rather because the these values are, except for the potential for properly account for its CCF event. If it is D, CCF event was more a degradation condition than recovery. applicability to baseline based on recovery Events a full failure.

CCF Event #526: Failed Programmable Logic PRA models. potential, consider For the suggestion to use weighting factor s Controllers affects all 3 SBO diesels. During expanding that use to for highly recoverable CCF events, NRC initial troubleshooting, it "appeared" that the other highly does not agree with introduc ing recovery SBO diesels were able to be started at the recoverable CCF actions and t heir likelihood as weighting human machine interface (HMI). The HMI did events. factor s in CCF (or reliability) parameter not indicate the ability to close any of the estimation. Instead, it may be more proper breakers on the PB bus. The physical Lower for recovery actions, including recovery Medium Voltage Sys 4.16 KV Bus breakers probabilit ies, to be considered during the have the capability of being closed via PRA model development phase.

mechanical push buttons, but there are no site For the CCF Event #526, the EPIX record procedures that provide guidance to do so. stated that The Fix It Now (FIN) team reset both SBO PLCs which cleared the alarm condition and put the SBO in a functional but degraded state. The p-value code is supposed to be used specifically to address degradation and not potential recovery. The EPIX record states that the SBO diesel was degraded and that the utility was unable to restore the diesel until suppor t from vendor was provided. Potential recovery does not appear to be the case.

6 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.10 CCF event #525 (see below) includes Weighting factors Consider expanding PWROG-M DISAGREE: NO CHANGES PLANNED P_values of 1.0 for both components. should be applied to the use of P_values to 18029, Same response as in CCF.9.

However, it is clear from the event each CCF event to account for highly Appendix description that these failures were properly account for its recoverable CCF D, CCF recoverable within 30 minutes. applicability to baseline events. Events CCF Event #525: During the start of Unit 1 B PRA models.

EDG, the air start valves failed to close. This depressurized the common air start header 1B EDG shares with Unit 2 A and B EDGs. Failure of newly-installed diodes for the 1B EDG was the cause, which exploited an original design vulnerability of having a common header.

Operators were able to quickly isolate the 1B air start control valve, allowing pressure to recover for the Unit 2 EDGs (within 30 minutes).

CCF.11 CCF event #534 (see below) is classified as Events should be Remove this event PWROG-M DISAGREE: NO CHANGES PLANNED a failure of both MFW pumps. However, the excluded from the from the CCF dataset 18029, NRC does not agree that a CCF event event is described as a failure due to a generic CCF dataset if since it should be Appendix should be removed from the CCF dataset if it common support system, instrument air. they can be accounted included as a Loss of D, CCF is accounted for in initiating event dataset.

CCF Event #534: Both MFPs tripped on low for in the generic Main Feedwater (or Events Such CCF events should be included in order suction pressure. Loss of IA pressure to the initiating event dataset. Loss of Instrument Air) to provide correct estimation for the CCF condensate demin. system due to catastrophic initiating event. parameters.

failure of the inline air filter bowl was the cause.

CCF.12 Many of the CCF events with CCCGs > 4 CCF events should be In the creation of large NEW M AGREE: NO CHANGES PLANNED involve main steam safety or relief valves. pooled only when the pooled groups (e.g., all We agree that setpoint failure is a different Most of those events are due to setpoint failure causes included demand CCFs), do failure mode from fail to open and fail to failures (e.g., see CCF event #491, below). could be generally not pool CCF events close. No changes are planned as the This failure mode appears to be unique to applicable to all when the failure setpoint CCF events are included in the NRC safety and relief valves and, thus, does not component -types causes are not CCF database but not used in CCF provide insights into the likelihood of included in the pooled generally applicable to parameter estimation.

common cause failures in large CCCGs data. the component-types involving other component - types. included in the pooled CCF Event #491: Corrosion bounding of three data. For example, do MSSVs. Results in high OOS setpoints (3 of not pool CCF events 12, with P_values of 1.0). involving main steam safety or relief valves with other component-types when calculating CCF parameters for large-pooled groups.

7 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution CCF.13 For CCF events due to setpoint failures CCF events should be Review CCF events NEW M DISAGREE: NO CHANGES PLANNED (e.g., see CCF event #491 in CCF.12), it is evaluated to determine involving safety/relief See the above response to CCF.12.

not clear that theseevents were, in all whether the event is a valves to determine cases, functional failures of the safety/relief functional failure. whether the events are valves. functional failures.

Where there is uncertainty about the impact of the event, use P_values to appropriately weight each event.

CCF.14 A number of CCF events at dual units The CCCG for each Review CCF events for NEW M DISAGREE: NO CHANGES PLANNED account only for the size of the CCCG in the CCF event should sites with more than The current NRC CCF characterization unit with the CCF. For example, CCF event include all component - one unit and revise the process does not include the identification of

1. 489 (see below) includes only the 3 EDGs types in the same CCCG size, as inter - system or multi - unit CCFs. The CCCGs in Unit 2, not the additional 3 EDGs in Unit 1. system in all units on necessary, to account are thus not determined by the size across CCF Event #489: 2/3 EDGs inoperable due to the site. for all similar units.

failed cylinder cap-screws. component -types in the same system on site.

8 Table F-2 PWROG Peer Review Observations and NRC Responses Related to Data Quality

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ1.1 A number of editorial errors were Documentation of the Review the 2020 dataset PWROG-M AGREE: ACTIVITY COMPLETED identified with the NRC Dataset (2015) methodology and results to assure that these (or 18029, NRC appreciates the peer review of the that have the potential to impact how the of the generic data similar) editorial errors Tables NRC Dataset by the PWROG. Please see data are used. These include duplicate analysis should be are not present. DQ.1-1, the NRC review results for each of the entries (e.g., PLF FTOP and PLL FTOP accurate and complete. DQ.1-2 identified issues in the attached Tables have exactly the same success and DQ.1 - 1 and 1.2, as well as the actions in the failure data), documentation errors in the 2020 parameter update efforts.

Data Source and Comment fields, and For 19 pairs of duplicate entries identified in entries labeled fail-to-start but identified Table DQ.1-1, 10 of them have no data as per-hour rather than per-demand. See issues.

Tables DQ.1-1 and DQ.1-2 for additional editorial errors. Nine pairs had duplication issue and were corrected in the 2020 Update.

For 39 entries of miscellaneous issues identified in Table DQ.1-2, 24 of them had no data issues.

Fifteen had data issue s which were corrected in the 2020 Update.

DQ1.2 Potential editorial errors were identified Documentation of the Review (and correct as PWROG - H AGREE: ACTIVITY COMPLETED with the NRC Dataset (2015) related to methodology and results needed) the questionable 18029, Same response as DQ 1.1.

questionable number of demands and of the generic datanumber of demands and Table DQ.1-run - hours. For example: analysis should be run - hours to determine 2 TDP-FS-NR-MFW (MFW turbine driven accurate and complete. whether these are simply pump fails to start, normally running) is editorial errors that do listed as having almost 6E+6 demands not impact the failure for 43 pumps over the 17 - year period rates, errors that may from 1998 to 2015. impact the failure rates, or technical issues that These potential errors are identified in need to be investigated the last column in Table DQ.1-2 of and resolved.

PWROG - 18029 as Questionable # of demands or # of run - hours seems to be too high.

9 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ2.1 The NRC Dataset (2015) and NRC CCF The generic reliability Assure that the NRC PWROG-H (1) AGREE: ACTIVITY COMPLETED Dataset (2015) are not consistent in the dataset and CCF dataset Dataset (2020) and NRC 18029, For the different date ranges in the NRC number of failure events for some should be consistent in CCF Dataset (2020) are Table DQ.2-Dataset (2015) and NRC CCF Dataset component failure modes. time period of the data consistent in the date 1 (2015), while staff does not think it would Also, these two datasets do not quite and the total number of range of data included. have adverse impact on the results, we match for event dates included. NRC failure events for each Verify that the count of agree that it is a good practice to use the dataset covers 1998 to 2015. NRC CCF component failuremode. failure events is same or similar date ranges for them.

Dataset (2015) covers 1997 to 2015. consistent between these The s ame date range has been used in the However, it is not clear the one additional databases or explain the 2020 parameter update for both component year of data for NRC CCF explains the reasons for any reliability and CCF analysis.

differences. differences. (2) DISAGREE: NO CHANGES PLANNED For example, EDG FTS has an annual The different number of failure events for average number of failures of 11.9 in the some component failure modes in the NRC NRC Dataset vs 13.8 in the NRC CCF Dataset, CCF Dataset, and NRC Reactor Dataset. Operating Experience Data (NROD), may be expected because these datasets have different purposes and thus have diff erent search results. For example, NROD returns all failure events meeting the search criteria (date range, component type, failure mode, etc.).

The NRC Dataset and CCF Dataset only use those NROD events for which devices are identified in the device list. The NRC Dataset uses only the complete failure events (i.e., P value equals to 1) while the CCF Dataset includes failure events with P_values of 0.1 and 0.5. For example, EDG FTS has approximately 260 events in NROD from 1998-2015, of which approx imately 240 are in the device list but only approximately 210 are complete failures.

DQ2.2 Inconsistencies in the number of failure Documentation of Resolve the PWROG-M DISAGREE: NO CHANGES PLANNED events for some component failure reliability datasets should inconsistencies in the 18029, See the response #2 in DQ2.1. The different modes were noted among the sources of be consistent with the 2020 datasets or explain Appendix A, numbers between NROD and 2015 NRC reliability data: NROD database. the reasons for potential DQ.2 datasheets/spreadsheet are typically Datasheets inconsistencies and how expected and reasonable due to their Spreadsheet the data should be used different purposes. However, in the event NROD (e.g., whether the inconsistencies between NROD and RADS spreadsheet results are identified, for instance, for results on See Table DQ.2-1 for examples of should be used when in batteries failing to operate, we will inconsistencies. disagreement with the investigate and make proper corrections, as Also see Observation DC5.4. datasheets). needed.

10 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ3.1 The component boundaries between The definitions of Verify that the PWROG-M DISAGREE: NO CHANGES PLANNED NRC Dataset (2015) (as defined in component boundaries component boundaries 18029, The NRC Dataset (2015) and the CCF NUREG/CR - 6928) and NRC CCF should be consistent are consistent between Appendix A, Dataset (2015) used the same component Dataset (2015) should be consistent. between these datasets the reliability dataset and DQ.3 failure database with the same component However, NUREG/CR-6928 and NRC since these are used to the CCF dataset for boundaries. Uncertain w here heating, CCF Dataset (2015) are not consistent in calculate the inter-related 2020. Where differences ventilation and AC are listed as being at least one important component component failure rates are identified, use the within the EDG boundary.

boundary, EDGs. In Section 5.1 of and the CCF parameters. boundary definitions from NUREG/CR -6928, room cooling is NUREG/CR -6928 in both excluded from the EDG component datasets.

boundary while in NRC CCF Dataset (2015), heating, ventilation and AC are within the EDG boundary.

DQ4.1 NRC datasets from 2007, 2010 and Documentation of the Document the basis for PWROG-M AGREE: FUTURE ACTIVITY PLANNED 2015 show significant changes in some methodology and results any significantchanges in 18029, When NRC updated the industry - average component failure rate mean values. It is of the generic data the ways specific Section 5; parameter estimates, the preliminary results not clear whether these are based on analysis should be component failure modes Appendix A, were reviewed both by the data analysts changes in average component accurate and complete. are defined or their DQ.4 and the SPAR modeler group for significant performance, changes in failure data failure rates calculated. changes compared to previous dataset as collection or data treatment, or changes well as unexpected results. A summary of in estimates of success data. the top 10 increased/decreased unreliability For example, CRD -FTOP (control rod fail estimates is provided in the Component to insert rod), the mean value changed Reliability Data Sheet for the latest update.

from 1.32e-5/d (2007) to 9.91e -8/hr. In the pending update, NRC will include the (2010) to 1.15e -7/d (2015). main changes (both methodologies and results) in its update documentation.

DQ4.2 The estimation of success data is an Documentation of the Document the bases for NEW L DISAGREE: NO CHANGES PLANNED important element in calculating failure methodology and results estimating success data Staff agrees that the success data is rates since the success data reported to of the generic datafor the calc ulation of important in data analysis, but has INPO is not always done in a consistent analysis should begeneric failure rates. It assumed that there was little to no manner by each utility (based on accurate and complete. might be sufficient to confusion as to how it should be defined PWROG investigations). However, no describe the general documentation was found that explainedmethods used, with the process for developing success data some examples to to support the NRC datasets. provide the detail.

11 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ5.1 MDP-SBY group includes pumps that Component-types should Several potential PWROG-M DISAGREE: NO CHANGES PLANNED are always in standby (e.g., containment be grouped on common resolutions are 18029, Distinguishing betweenstandby spray pumps, HPI pumps) and pumps attributes that may impact suggested: Appendix E components and normally running that are normally standby but have a estimates of component Create two (2) groups, (Slide 6) components is frequently a challenge in normal operational mode (e.g., RHR reliability. MDP-SBY and MDP-nuclear data analysis. Sections 5.4 and pumps, AFW pumps). Pumps always in SBY-NO, to model these A.1.2 of NUREG/CR-6928 explain the standby would be expected to have pumps in more process to divide components into standby much fewer run-hours per start compared homogeneous groups. versus running/alternating categories:

to the second group. Use the average number (1) the components were sorted by run (Note, the inclusion of pumps that are of run-hours per start as hours, normally standby but have a normal a metric to determine the (2) components with run hours fewer than operational mode explains the original pumps that should be 10% of calendar hours were placed in the issue identified in PWROG-18029, Issue included in each group. standby category, and DQ.5, that run -hours for Standby Motor-Group pump types based Driven Pumps appear to be excessive.) on other attributes (e.g., (3) components with more run hours than in high pressure vs low item (2) above were placed in the pressure). running/alternating category. The us e of the 10% cutoff was based on a review of run Group pump types based hours for components in system s known to on their system (e.g., have only standby components (The highest MD-AFW pumps, RHR result among such systems was about 8%

pumps) where sufficient and most system results less than 3%.)

data is available.

12 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ5.2 The justifications for separating short-Component failure modes Combine short-term-run PWROG-H DISAGREE: NO CHANGES PLANNED term and long-term run failure rates for should be separated only and long-term-run data 18029, Use of short - term and long-term run failure standby components are inadequate. For if the data supports the for standby components Appendix E rates for standby components such as example: sub-dividing. where the data does not (Slides 7, MDPs and EDGs are long-time practices in For many standby components, long-support separate failure 12-15) PRA modeling and data analysis. It was term-run failure rates are not lower than modes. deemed in NUREG/CR - 6928 as a short-term, inconsistent with what one fundamental impr ovement in SPAR model might expect. basic event parameter estimation. A review of the recent parameter update results Limited long-term run data are available shows that MDP, EDG, FAN, CTF and other for truly standby components (e.g., component types are still having larger early PDP-SBY, positive displacement pumps) run failure rates, while TDP, EDP, PDP, and since test runs are typically no longer CTG have lower early run failure rates. With than 1 hour1.157407e-5 days <br />2.777778e-4 hours <br />1.653439e-6 weeks <br />3.805e-7 months <br />. no plan to change the SPAR modeling For other components classified as practice, we will continue to provide short -

Standby, the long-term-run success data term and long-term run failure rates similar is much greater than the short-term-run to previous parameter updates.

data (e.g., ACX -FTS, Air Cooling Heat Exchanger, Normally Standby).

For some components, the data (failures, run-hours) appear to have been split evenly between short-term-run and long-term-run (e.g., CHL, Chiller Unit, Normally Standby).

It is difficult to determine whether failure events are actually short-term or long-term.

13 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ5.3 Based on a review of a sample of failure Failure events should be Define (or refine the PWROG-H DISAGREE: NO CHANGES PLANNED events included in the short-term-run classified as demand-definitions for) failure 18029, The IDCCS Coding Manual includes the failure mode (Fail to run < 1H) for standbybased or time-based modes for event Appendix E definitions for EDG FTS/FTLR/FTR and components, it appears that some failure using standard definitions classification: (Slides 8, 9) pump failure on demand/failure to run. It events should be classified as demand-for fail-to-start and fail-to-Fail to Start: failure to defines Pump failure on demand: A failure based rather than time-based events. run failure modes. reach a stable running to start and run for at least one hour is state within a few counted as failure on demand.

minutes of start demand The demand failures should be if it fails to (demand failure). run less than an hour. Sometimes timing Fail to Run: failure to isnt mentioned so we have to make an continue to run after assumption based on the verbiage used as reaching a stable running to whether it is a run failure vs a start failure.

state (run failure). If the industry has different opinions on some of the event characterization, please Review the standby provide support information and we can re-component failure events consider coding, if needed.

classified as Fail to run

< 1H in light of these definitions and reclassify as appropriate.

14 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ5.4 Based on a review of a sample of failure Failure events should be For generators, revise PWROG-H DISAGREE: NO CHANGES PLANNED events included in the load-run failure classified as demand-failure definitions to 18029, Same as DQ 5.3. Please provide supporting mode (FTLR) for generators (EDG, CTG based or time-based clarify demand-failures Appendix E information if specific classification issues and HTG), it appears that some failure using standard definitions from run-failures: (Slide 10) are identified.

events should be classified as demand-for fail-to-start, fail-to-FTS: failure to reach a based rather than time-based events. load, and fail-to-run failure stable start-run state.

modes.

A stable start-run state includes adequate starting air to roll the EDG; automatic start from an undervoltage signal or test start from the main control room; and reaching normal and stable speed.

FTL: failure to reach a stable load-run state.

A stable load-run state includes EDG output breaker closed, stable engine speed, generator field successfully flashes, stable output voltage &

frequency, carrying full capacity load, and cooling flow established.

FTR: failure to continue to run after reaching a stable running (or load-run) state.

Review the standby component failure events classified as Fail to load/run < 1H in light of these definitions and reclassify as appropriate.

15 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ5.5 Based on a review of a sample of failure Component failure modes Use the revised failure P5WROG-H DISAGREE: NO CHANGES PLANNED events included in the EDG-FTLR failure should be clearly defined definitions (provided in 18026, FTLR = Given that it has successfully mode, this is a mix of demand-failures as demand-based or DQ5.4) to clarify Appendix E started (FTLR is) a failure of the EDG output (fail to load) and run-failures. time-based failures. demand-failures from (Slide 11) breaker to close, to successfully load run-failures in FTL. sequence and to run/operate for one hour to Replace FTLR with FTL, perform its monitored functions. This failure where FTL is defined as mode is treated as a demand failure for a demand failure mode. calculational purposes. (Exclude post maintenance tests, unless the cause of Review the previous failure was independent of the maintenance FTLR failure events performed.)

using new definition of Note that this is consistent with the FTL and include only the guideline in INPO 009 and INPO 002 demand-based failures in for reporting requirements. (Also, there is no FTL. specific code for FTL it is only FTS, FTR, Move any run-failure and FTLR.) Contact NRC if specific events in the FTLR classification issues are identified.

failure mode to FTR.

Calculate FTR using all run-failure events and all run-hours.

DQ6.1 The FLOW sensor/transmitter failure Documentation of the Develop failure rates PWROG-L DISAGREE: NO CHANGES PLANNED rates (STF FTOP -D, STF FTOP -R) are methodology and results specifically for FLOW 18029, There is no EPIX/IRIS data provided by based on the LEVEL sensor/transmitter of the generic data process logic and FLOW Appendix A, industry to conduct specific analysis.

event data (STL FTOP-D, STL FTOP -R) analysis should be sensors/transmitters DQ.6 Engineering judgement was used for the in NUREG/CR -6928 (from NUREG/CR -accurate and complete. based on recent failure Level sensor/transmitter failure data and for 5500). However, no justification is event data (2006 to those of the Flow sensor/transmitter.

provided for this treatment. 2015).

If that change is not possible in the near term, at least provide a basis for the use of LEVEL data (rather than pressure or temperature) as an appropriate surrogate for FLOW sensors.

16 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ7.1 Basis for Error Factors: Data distributions Generic data distributions Where JNID/IL is used, PWROG-M AGREE: FUTURE ACTIVITY PLANNED in the NRC Dataset (2015) may not fully should account for all develop a method for 18029, NRC will coordinate with statisticians and account for the uncertainties in the failure contributors to accounting for the Appendix A, industry on how to better characterize error rate estimates. uncertainty. number of success DQ.7; factors and uncertainty associated with Specifically, distributions based on events in the calculation Appendix va rious failure/success data.

JNID/IL (Jeffreys non-informative of the error factor. B.3 (Slides distribution at the industry level) fail to 4, 5) account for the differences in the number of success events (or success-hours).

For example, the error factor for component failure modes with zero failure events is 8.4, while the number of successes range from 78 to 7.2E7.

DQ7.2 Basis for Error Factors : Data distributions Generic data distributions Develop a method of PWROG - L AGREE: FUTURE ACTIVITY PLANNED in the NRC Dataset (2015) may not fully should account for all accounting for an 18029, Same as in DQ7.1.

account for the uncertainties in the failure contributors to expanded set of Appendix A, rate estimates. uncertainty. contributors to the DQ.7; While a number of component failure uncertainty that underlie Appendix modes account for plant - to - plant the inputs into failure B.3 (Slides 4 variability by using EB/PL/KS (plant - level rates (and CCF to 10).

empirical Bayes statistical analyses with parameters).

the Kass - Steffey adjustment), other It may be important to contributions to uncertainties are not classify the types of reflected in the distributions based on the uncertainties (e.g.,

raw number of failures and successes for random, state-of -

a specific component failure mode. knowledge, fuzziness) to These could include: fully account for the Uncertainty in the number of failure underlying uncertainties events and number of demands or run in data distributions and times to better account for the state-of - knowledge Uncertainty in the type and consequence correlation.

of the failure event Uncertainty in the homogeneity of components in the group based on component attributes such as manufacturer, size, process fluid, ambient environment, etc.

Uncertainty in the homogeneity of components based on plant - to - plant variability.

17 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DQ7.3 The process of treating data using plant-Documentation of the Document the pooling NEW L DISAGREE: NO CHANGES PLANNED level empirical Bayes method generally methodology and results analysis that justified the The 2015 and beyond datasets are updates includes a pooling analysis to justify the of the generic data grouping of data for each of NUREG/CR-6928 with generally the grouping. NUREG/CR -6928 documented analysis should be component-type and same processes/methodologies as those in the original pooling analysis. However, no accurate and complete. failure mode. the original NUREG/CR report. If the update documentation of pooling analysis could reports do not have statement indicating a be identified for the 2015 dataset. change, it would mean the same processes/methodologies have been used in the updates.

DQ8.1 CCF Weighting Factors & Mapping - Up Documentation of the Create a detailed PWROG - M DISAGREE: NO CHANGES PLANNED Process : NUREG/CR - 6268 documents methodology and results methodology guideline 18029, NUREG/CR-6268, Rev. 1, provides the CCF database and analysis system of the generic CCF data that provides the details Appendix A, documentation on how the CCF weighting set up by INL for NRC. The NUREG analysis should be regarding how the DQ.8; factors and the mapping-up process are suggests various weighting factors that accurate and complete. NUREG method gets Appendix C used in CCF parameter estimations.

are incorporated via the database for implemented in the (Slide 5) events based on uncertainty in Timing, NROD/RADS PRA Component Degradation, and Shared Calculator. This should Cause. Three (3) additional factors are include specifically how used in assessing CCF failure events: weighting values and Shock Type, Failure Mode Ap plicability, mapping - up values are and Defense Mechanism. created. Example However, it is not clear how these factors ca lculations would be are used to create CCF parameters and helpful to explain the their uncertainty distributions. details.

DQ9.1 The PWROG review of the 2015 NRC Explicit criteria Develop explicit criteria NEW L Developing such criteria is unrelated to NRC reliability datasets has been based on documenting the documenting the data collection and analytics activities but the implicit criteria for what constitutes high elements of a high-quality elements of a high-industry may consider developing criteria to quality generic reliability datasets. generic reliability dataset quality generic reliability benefit future peer -reviews.

However, no explicit criteria exist that would help support a dataset to support future would support systematic reviews in the systematic r eview. updates and peer future (as well as future updates to these Criteria could provide revi ews.

datasets). For example, supporting structure to the review as requirement DA - C1 from the PRA well as identifying Standard (Addendum B) states, USE important elements that generic parameter estimates from could be included.

recognized sources and includes NUREG/CR - 6928 as an example of recognized source for component failure rates. There are no criteria specified except that these be recognized sources.

18 Table F-3 PWROG Peer Review Observations and NRC Responses Related to Data Aging

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DA1.1 Time Trends of Component Failure Events: Component Use the data from the most PWROG-H AGREE: FUTURE ACTIVITY PLANNED The current NRC Dataset (2015) includes reliability over the recent time period to 18029, NRC has evaluated whether using a shorter, data for a 20- year period, 1996 to 2015. recent past should calculate generic failure Appendix more recent time period for parameter estimate However, the average performance of be the best rates. The length of this A, DA.1; is appropr iate and could better reflect the current components industry -wide is significantly estimator of future time period should be short Appendix industry performance. The NRC determined that better in the most recent 10- year period performance. The enough to reflect recent C (Slide updates will use a 15-yr rolling period to estimate compared to earlier time periods. The most time period used component performance 9) component reliability and CCF parameters.

recent decade would be more indicative of for component (e.g., 2009 to 2019) but may near-term future component performance. reliability data need to be extended for needs to balance some components where the value of recent failure events are rare.

past data with the The time periods used need for adequate should be consistently used amounts of data. in the generic CCF parameter calculations (see Observation DA2.1).

DA2.1 Time Trends of CCF Events : The current Component Use the data from the most PWROG - H AGREE: FUTURE ACTIVITY PLANNED NRC CCF Dataset (2015) includes data for reliability over the recent time period to 18029, See the response in DA1.1.

a 19-year period, 1997 to 2015. The NROD recent past should calculate generic CCF Appendix Database includes CCF events from 1996 be the best parameters. The length of A, DA.2; to 2019. estimator of future this time period should be Appendix However, the average performance of performance. The short enough to reflect C (Slide components industry - wide as measured by time period used recent component 6)

CCF data is significantly better in the most for component performance (e.g., 2009 recent 10 - year period compared to ea rlier reliability datathrough 2019) but may need time periods. needs to balanceto be extended for some the value of recent components where common past data with the cause failure events are need for adequate rare. The time periods used amounts of data. should be consistently used The time periods in the generic component for generic failure rate calculatio ns (see reliability data and Observation DA1.1).

CCF data should be consistent for each component failure mode.

19 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DA3.1 Failure Rates Based on Old Data: Of the Generic component Delete external leakage and PWROG-M DISAGREE: NO CHANGES PLANNED 332 component failure rates in the NRC failure rates should pipe leakage failure modes 18029, The results from NUREG/CR - 6928 and Dataset (2015), 70 are based on old data: be based on recent from the 2020 NRC Dataset Appendix subsequent updates have been used as inputs to Large internal/external leaks (38) - from reliability data for since those failure modes A, DA.3 the SPAR models. Unless the associated EPIX, 1997 to 2004. While the small each component are addressed in internal component failure mode data are no longer leakage events have been updated to failure mode. flood datasets (see needed by SPAR models, such parameter 2015, the large leakage rates are still Observations DC2.1 and estimates will, of necessity, continue to be based on the factors developed from the DC4.1). provided based on the best available data. It is 1997 to 2004 EPIX data. Eliminate WSRV-based true that other techniques such as expert Pipe leaks (4) - from EPIX, 1997 to 2004. data for the five (5) elicitation could be used to replace the results PIPE SWS-ELS and PIPE OTHER -ELS component failure modes from old data, but the resources needed for are based on data. The ELL failure rates that use this source based these efforts must be commensurate with the risk are based on the ELS failure rates times on its age and lack of significance of the associated component failure factor (0.2 for SWS, 0.1 for Other). availability of the source modes.

Control circuit failures (3) - from WSRV, data. If failure rates for 1980s to 1990. The NRC Dataset (2015) these component failure labels the source of these component modes are required, failure modes (ICC-FA, ICC-FC, ACT-FC) consider creating failure as NUCLARR but the Datasheets identify rates from similar the source as WSRC. equipment (per requirement Air dryer unit (1) - from WSRV, 1980s to DA-D2 in the PRA 1990. The NRC Dataset (2015) labels the Standard).

source of this component failure mode, In general, where recent ADU-FTOP, as NUREG/CR -6928 but the reliability data are not Datasheets identify the source as WSRC. available, use other Orifice plugging (1) - from WSRV, 1980s to techniques (e.g., expert 1990. The failure rate for this component elicitation) to develop failure mode, ORF-PG, comes from the appropriate failure rate Westinghouse Savannah River Co mpany distributions.

(WSRC) database.

Sensors, bistable, manual switch, process logic, RPS breaker (19) - from NUREG/CR -5500, 1984-1995. See Observation DA5.1 PORV (4) - special calc, from NUREG/CR-7037, 1987 to 2007. See Observation DA4.1.

20

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DA4.1 Failure Rates for Safety Valve Fail to Generic component Update the component PWROG-M AGREE: FUTURE ACTIVITY TO BE Reclose Based on Judgment : NUREG/CR -failure rates should failure mode PORV-Liquid18029, CONSIDERED 6928 included Safety Valve and Safety be based on recent with recent failure event Appendix The inclusion of aged data (1980s - early 1990s)

Relief Valve failure to reclose passing reliability data for data since the basis for this A, DA.4; or older studies (such as RPS system study or liquid (SVV FTCL, SVR FTCL) with a mean each component is old and very limited data Appendix Westinghouse Savana River Complex) in value of 0.1, based on judgment with very failure mode. with zero failures. C (Slide NUREG/CR-6928 and its updates is due to one limited basis. NRC Dataset (2015) Where recent Consider use of expert 18) of the following reasons:

eliminated these failure modes. It includes reliability data are elicitation for this (1) the data is not available from EPIX or LER a new failure mode, PORV-Liquid, with not available, use component failure rate if review or mean value of 6.25E-2, given the evidence other techniques data continues to be sparse.

of zero failures and seven demands from (e.g., expert (2) the data is available but sparse.

1987 to 2007 (from NUREG/CR -7037). elicitation) to As PWROG suggested, there are two ways to develop address the data aging issue: either collect appropriate failure recent failure data or use other approaches such rate distributions. as expert elicitation.

The collection of more and better failure data will require effort on the part of the industry and the NRC. The industry could report such data directly to INPO (voluntarily or required by INPO reporting requirements) or it could include the demand and success/failure information in LERs or other forms of reporting. An example of this is PWR RCS safety valves open and reclose status. If done, NRC c ould better characterize and classify such events/data for analysis.

Expert elicitation could be another approach to address sparse data issue, but it would need even more effort from the industry and NRC. This approach should be reserved for the most important component failure mode parameters (which could be determined by a joint industry and NRC working group).

Interested industry participants may consider forming a working group to develop a list of component failure modes having data aging or sparse data issue, to determine which approaches should be used to address issues such as:

Increase data collection efforts?

Use expert elicitation for the most important and emergency items?

(3) Keep the current status for less important items?

21

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DA5.1 Sensor Failure Rates Based on Old Data: Generic component Update the data that PWROG-L AGREE: FUTURE ACTIVITY PLANNED The sensor/transmitter failure rates are failure rates should supports the 18029, See responses in DA4.1.

based on data from 1984 through 1995 be based on recent sensor/transmitter failure Appendix (documented in the NUREG/CR -5500 reliability data for rates. It is expected that A, DA.5 series). A total of 5.5 Level sensor failures, each component failure event reports have 37.5 pressure sensor failures, and 46.1 failure mode. improved in level of detail temperature sensor failures were used in Where recent since the data was collected the failure rate calculations. The failures reliability data are to support the 2015 dataset.

were divided into Demand failures and not available, use Consider collecting recent Time-related failures. A number of the other techniques failure data for one sensor-failure events are described as uncertain (e.g., expert type (e.g., pressure) and based on limited failure event reports. elicitation) to calculate new failure rates These uncertain events are assigned a develop (dem and, time) with this weighting factor (less than 1.0). appropriate failure data. Compare this to the rate distributions. failure rates from the1984 to 1995 -time frame currently in the NRC Dataset (2015).

If the failure rates are significantly different, expand the data analysis effort to address the other sensor-types.

22

Table F-4 PWROG Peer Review Observations and NRC Responses Related to Data Classification

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC1.1 NRC introduces new component Component-types Review current component PWROG-L DISAGREE: NO CHANGES PLANNED failure modes in NRC datasets from should be modeled types with limited data to 18029, Over time, NRC has improve d the data analysis 2007, 2010, and 2015. However, it is as separate see if grouping similar Appendix process including upgrades to failure mode not clear the justification for including reliability groups to component types would be A, DC.1 evaluations to help meet stakeholder needs. These some specific failure modes. In other support the a more appropriate changes are reflected in the most recent parameter cases, there may not be adequate homogeneity of the balance of homogeneity of estimate updates, but t hey will probably continue to data to justify separating out some group but only to the group with the amount change in the future. In the past, the trend was similar failure modes (e.g., where the extent sufficient of data available. For towards ever increasing detail (and so more system -

limited component reliability data has reliability data is example, consider motor-specific parameters), but t he more recent trend has been divided by system). available. driven compressors (PCA-involved pooling some failure modes. NRC will See related observations for MDC, IAS-MDC, MDC) continue its current process to make proper decision component leakage (DC2.1, DC3.1, & and engine -driven pumps on how to better meet various needs.

DC4.1) and spurious oper ation (DC5.1 (AFW-EDP, EDP).

to 5.4).

23 ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC2.1 Leakage Probability: Section A.1.2 of Failure modes in Redefine Leakage to PWROG-M AGREE: FUTURE ACTIVITY TO BE CONSIDERED NUREG/CR -6928 discusses the the generic include only Internal 18029, Items DC2.1, DC3.2, and DC4.1 present criteria for leakage events from EPIX component Leakage in the range 10 to Appendix observations over the leakage probability, large (1997 to 2004): reliability dataset 50 gpm. A, DC.2 internal leakage probability, and large external External leakage events were should include only External leakage should be leakage probability. The following potential reviewed to identify small leaks (1 to those associated classified as an internal resolutions have been offered by the PWROG :

50 gpm), large leaks (> 50 gpm), and with internal events flood event and be Define leakage to include only internal leakage leaks too small to be of interest in this hazards. They excluded from this in the range 10 to 50 gpm, study (< 1 gpm). should not overlap Database (see Eliminate large internal leakage as a failure with unique failure Observation DC4.1). mode, or Internal leakage events were reviewed modes considered Eliminate all external leakage as a failure to identify small leaks (leaks in other hazards. Leakage any larger than mode.

exceeding the local leak rate test 50 gpm internal to the allowable limits or 1 to 50 gpm), large device could be For the second and third suggestions, NRC does not leaks (typically resulting from considered a component believe that eliminating the leakage failure mode component internal degradations failure, e.g., manual valve from the data analysis is appropriate. The data greater than just pitting or wearing or > leakage larger than 50 analysis results from NUREG/CR-6928 and 50 gpm), and negligible leaks (less gpm should be counted as subsequent updates are used as inputs to the SPAR than the local leak rate test limits or < 1 Manual Valve Fails to models so that it is the PRA modelers decision on gpm). Remain Closed. (See how to utilize the data results and whether or not to From this description, it would appear Observation DC3.1.) include large internal leakage or all external leakage that external leaks may have specified Internal leakage less than in PRA model.

a leakage rate while internal leaks may 10 gpm should be For the first suggestion, while we can check whether have only identified leak exceeds considered nuisance changing the threshold value for small leakage from local leak rate test or internal leakage which should not 1 gpm to 10 gpm is proper, we wonder whether or degradation rather than a specific leak impact system not the benefits from such a redefini tion and rate. performance. Leakage in reclassification of the leakage events are worth the The definition of Small Leakage (1 to the range from 10 to 50 effort as the leakage probabilities are already very 50 gpm) is not helpful because it gpm is still a challen ge to small (mostly in the range of E-8 range) compared to covers the range from nuisance model but should help to the failure probabilities/rates for other failure modes.

leakage which should not impact focus the concern on A joint industry/NRC working group c ould consider system performance to the point leakage events that are items such as those above to address.

where leakage may not be insignificant more likely to challenge (depending system function over a on the system). period of time.

Note, the probabilities may be low enough that any internal leakage events could be included in existing component f ailure modes.

24

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC3.1 Large Internal Leakage Probability: Failure modes in Eliminate large internal PWROG-M AGREE: FUTURE ACTIVITY TO BE CONSIDERED The value 0.02 is used as the ratio of the generic leakage as a failure mode 18029, See responses in DC2.1.

ILL/ILS (large internal leak failure rate / component in the 2020 reliability Appendix small) for valves (SRV, SVV, PORV, reliability dataset dataset. A, DC.3 RVL), tanks (TNK, ACC, STR), andshould not be Map these failure modes to heat exchangers (HTX). This value is expanded where an existing component failure documented in Table A.1.2-1 of existing failure modes that account for the NUREG/CR -6928 and is based on 3 mode can same impact (e.g., MOV large internal leaks vs 185.5 small adequately account large internal leakage may internal leaks (with the ratio rounded for the impact of the have the same impact as off). This data is from the 8-year period failure event. failure to close). If the 2015 from 1997 to 2004. This same ratio probability estimates of (0.02) is used in the 2015 dataset large internal leakage are although the data used to calculate ILL significant contributors to failure rate appears to be updated. the mapped component Thus, this value is based on 3 large failure modes, update the leaks from data that is 15 to 20 years event data for large old. internal leakage.

DC4.1 Large External Leakage Probability : Failure modes in Eliminate all external PWROG - M AGREE: FUTURE ACTIVITY TO BE CONSIDERED The value 0.07 is used as the ratio of the generic leakage as a failure mode 18029, See responses in DC2.1.

ELL/ELS (large external leak failure component in the 2020 reliability Appendix rate/small) for valves (SRV, SVV, reliability dataset dataset since this data is A, DC.3 PORV, RVL) and tanks (TNK, ACC, should include only (should be) accounted for STR). This value is documented in those associated in the internal flood Table A.1.2-1 of NUREG/CR-6928 and with internal events frequency calculations.

is based on 2.0 large external leaks vs hazards. They 35.0 small external leaks (with the should not overlap ratio ro unded off). This data is from the w ith unique failure 8-year period from 1997 to 2004. This modes considered same ratio (0.07) is used in the 2015 in other hazards.

dataset although the data used to calculate ILL failure rate appears to be updated. Thus, this value is based on 2 large leaks from data that is 15 to 20 years old.

Also, this data overlaps with the data used to calculate internal flood frequencies.

25

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC5.1 NRC Dataset (2015) identifies events The generic Remove events from the PWROG-M DISAGREE: NO CHANGES PLANNED classified as spurious operation for component component reliability 18029, The results from NUREG/CR - 6928 and the following several component types: reliability dataset dataset where the spurious Appendix updates are used as the inputs to the SPAR models.

PORVs should not include operation causes a plant A, DC.5, Unless the associated component failure mode data Safety/Relief valves events that cause a transient. These events DC.6, are no longer needed by SPAR models, such Breakers plant transient. should be included only in DC.8. parameter estimates will continue to be provided Those events the initiating event dataset. Appendix based on available data.

However, failure events that caused a should be account See Appendix E (Slide 18) E (Slides plant transient are initiating events or for in the initiating for a logical classification 17, 18, precursor events rather than (simply) event analysis. of spurious operation 21, 22, component failures. failure modes and their 23) impacts.

DC5.2 NRC Dataset (2015) identifies events Failure modes in Model failure events PWROG-M DISAGREE: NO CHANGES PLANNED classified as spurious operation for the generic included in spurious 18029, Same as DC 5.1.

several component types: component operation failure modes Appendix AOVs, reliability dataset using other existing failure A, DC.7, MOVs, should not be modes. Eliminate spurious DC.8.

SOVs expanded where an operation as a component Appendix Breakers existing failure failure mode. E (Slides mode can See Appendix E (Slide 18) 17, 18, The use of the spurious operation adequately account for a logical classification 19, 23) failure mode creates unnecessary for the impact of the of spurious operation complication in system modeling failure event. failure modes and their where the impact could be modeled impacts.

with an existing failure mode (fail to open, fail to close).

DC5.3 The term spurious operation is used Failure modes in Eliminate spurious PWROG - M AGREE: FUTURE ACTIVITY PLANNED in Fire PRA for a specific fire - induced the generic operation as a component 18029, Spurious operation has long been used as a failure failure mode. This term should not be component failure mode (see DC5.1). Appendix mode in nuclear data analysis. As such, t here are a used for hardware failures. reliability dataset If not eliminated, then A, DC.7; number of IDs that use the term, some for historical Spurious operation failure mode is should not be clarify the definition and Appendix purposes.

defined using a number of IDs (SOP, expanded where an use consistent IDs and E (Slides The 2020 Update will clarify the various usages of SO, SC, OC) and descriptions existing failure descriptions for this failure 17, 19) these IDs to be more consistent.

(spurious operation, spurious opening, mode can mode.

spuriously transfers, transfers open, adequately account fails to remain open) without any clear for the impact of the dist inction on these failure modes. failure event.

26

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC5.4 The NRC Dataset (2015) identifies a Documentation of Resolve the PWROG-M AGREE: FUTURE ACTIVITY PLANNED spurious-operation failure mode for a reliability datasets inconsistencies in the 2020 18029, NRC confirmed that there are some inconsistencies number of valve types, including AOV-should be datasets or explain the Appendix between the NROD and RADS. NRC will investigate OC/SOP, MOV-OC/SOP, and SOV - consistent with the reasons for potential A, DC.7; such discrepancies and make corrections as SOP. The count of events in the NROD database. inconsistencies and how Appendix needed.

NROD Database was significantly the data should be used E (Slides lower than in the NRC Dataset: (e.g., whether the 17, 20)

MOV_SOP events: 63 in NRC Dataset spreadsheet results should (2015) and 48 in the NROD Database. be used when in disagreement with the AOV_SOP events: 132 in NRC datasheets).

Dataset (2015) and 67 in the NROD Database.

Spurious operation of SOVs, check valves, and manual valves had counts of 9, 2, and 6 (respectively) in NRC Dataset (2015) but zero events in the NROD Database.

Also see Observation DQ2.2.

DC9.1 The NRC CCF Dataset (2015) CCF parameters Eliminate the spurious PWROG - M AGREE: FUTURE ACTIVITY PLANNED provides extremely sparse evidence of should not be operation failure modes 18029, NRC agrees that CCF parameters might not be common cause failures for spurious developed for from the CCF Dataset Appendix needed for component failure modes where both the operation failure modes: zero events failure modes based on the limited data A, DC.9 independent and the common cause failures are for check valves and DC circuit where both for both independent and rare.

breakers; one event each for MOVs, independent and common cause spurious The 2020 CCF parameter estimate will remove the AOVs, and 480VAC circuit breakers; common cause operation. component failure modes that have no CCF and and three events for 4160VAC circuit failures are rare independent failures.

breakers. As discussed in events.

Observations DC5.1, DC5.2 and On the other hand, note that the 2015 CCF DC5.3, the evidence for independent Parameter Report has a statement in Section 1.2 spurious operation events is limited that asks user s to decide what is the appropriate and, in many cases, would be better approach for their PRA.

characterized as precursor events. The report is meant to be a reference and analysts Despite this limited data, CC F should decide whether or not to model specific CCF parameters are calculated and events. Should they decide to do so, the results are reported in the NRC CCF Dataset there for use.

(2015) for spurious operation modes for valves and circuit breakers.

27

ID Observation Basis Potential Resolutions Ref Priority NRC Response/Resolution DC10.1 The failure events recorded in the Failure events that Consider revising the PWROG-M DISAGREE: NO CHANGES PLANNED NROD Database include an are highly P_value parameter to 18029, While t he thought of revis ing P_value with weighting assessment of whether the failure was recoverable (e.g., include a weighting factor Appendix factor s based on the recoverability of the failure recoverable and an estimate of the recoverable within based on the recoverability A, DC.9; event appears interesting, NRC does not believe it is recovery duration. However, these 60 minutes) should of the failure event. One Appendix a good idea to mix or combine failure probability and failures are treated the same as other be treated as possible treatment: C (Slide recovery probability. NRC believes that failures that may be highly non-weighted failures to Failure events recoverable 17) r ecoverability should be treated separately, at least recoverable (i.e., with a much longer acknowledge that from the control room that has been common PRA practice.

recovery time). such events have within a few minutes Implementation of such an approach also has the Also see Observations CCF.9 and much less risk without any significant drawback that it would require wholesale CCF.10. importance than troubleshooting (e.g., reclassification of thousands of failure records for other failure events. control switch in pull-to-new P_values.

lock): P_value = 0.1 Failure events recoverable within 15 minutes without any significant troubleshooting(e.g.,

resetting the turbine-driven AFW pump trip/throttle valve): P_value = 0.2 Failure events recoverable within 60 minutes without any significant troubleshooting (e.g.,

resetting a pump breaker):

P_value = 0.5

28

Table DQ 1-1 Pairs of Duplicate Entries with Different IDs with NRC Review Results

ID Failures Demands/Hours Components NRC Response TBV-FTO 8 2,725 73 Not duplicates: these are two different failure modes, fail-to-open and fail-to-open/close, both having 8 failures.

TBV-FTOC 8 2,725 73 SVV-FTC-PWR-RCS 1 2,907 155 Not duplicates: there was 1 fail-to-close event (Point Beach 1, 2004) and 1 fail-to-close event (Indian Point 2, SVV-FTO-PWR-RCS 1 2,907 155 2006) of RCS SVVs.

PLF FTOP 3 6,075

• Not duplicates: the 2015 Update has a comment The PLL data was used to estimate the flow process logic PLL FTOP 3 6,075 * (PLF) reliability.

PORV-FTO-PPR 16 6,130 114 Duplicates corrected: these two templates are redundant. Redundant entry was removed from the 2020 PRV-CC-PZR 16 6,130 114 Update.

STF FTOP-D 5 6,750

• Not duplicates: the 2015 Update has a comment The STL data was used to estimate the flow STL FTOP-D 5 6,750

• sensor/transmitter (STF) reliability.

ARV-OO-MSS 19 10,401 111 Duplicates corrected: these two templates are redundant. ARV-is used in SPAR models while PORV-is used PORV-FTC-MSS 19 10,401 111 in data collection and analysis. Redundant entry was removed from the 2020 Update.

ARV-CC-MSS 42 10,401 111 Duplicates corrected: similar with the above, Redundant entry (ARV-CC-MSS) was removed from the 2020 PORV-FTO-MSS 42 10,401 111 Update.

ACX-FTS 45 17,336 139 Duplicates corrected: these two templates are redundant. ACX-is used in SPAR models while AHU-was used AHU-NR-FTS 45 17,336 139 in NUREG/CR-6928 and the 2015 Update. Redundant entry (AHU-NR-FTS as well as AHU -NR-FTR) were removed from the 2020 Update.

IAS-FLT-FC 0 122,688 2 Not duplicates but removed: these are for two different failure modes, fail-to-operate (FC) and Plug (PG), both IAS-FLT-PG 0 122,688 2 having 0 failure. But since IAS-FLT -PG is the only one used in SPAR models, IAS -FLT -FC was removed from the 2020 Update spreadsheet.

CHL-FTR<1H 61 279,348 63 Not duplicates: chiller has only FTR failure mode in RADS. So FTR<1H and FTR>1H use the same data to CHL-FTR>1H 61 279,348 63 achieve the same failure rate. A comment was added to the 2020 Update to explain the situation.

MDC-FTR<1H 22 1,683,943 58 Duplicates corrected: the MDC-FTR<1H values in the 2015 Update were incorrect. The 2020 Update will show MDC-FTR>1H 22 1,683,943 58 correct values.

SWS-TSA-PG 0 2,331,600 15 Not duplicates but removed: these are two different templates with the -NE one for non-environment failure TSA-PG-NE-SWS 0 2,331,600 15 causes. But since TSA-PG -NE -SWS is the only one used in SPAR models, SWS -TSA-PG was removed from the 2020 Update spreadsheet.

TDP-FR-NR-MFW 62 5,984,882 43 Data error corrected:

TDP-FS-NR-MFW 62 5,984,882 43 (1) The TDP -FS -NR-MFW values in the 2015 Update were incorrect.

TDP-NR-FTR 62 5,984,882 43 (2) TDP-NR-FTR has the same values as TDP -FR -NR-MFW since normally running TDPs are from MFW only.

Since TDP-FR-NR-MFW is the one used in SPAR models, TDP-NR-FTR was removed from the 2020 Update.

STF FTOP-R 0 9,831,970

• Not duplicates: the 2015 Update has a comment The STL data was used to estimate the flow STL FTOP-R 0 9,831,970

• sensor/transmitter (STF) reliability.

TSA-FR 97 30,417,290 212 Data error corrected: TSA-FS and TSA-FR were added to the 2015 Update using the same data of TSA-TSA-FS 97 30,417,290 212 FTOP, which may not be appropriate. TSA -FS and TSA -FR were removed from the 2020 Update.

TSA-FTOP 97 30,417,290 212 HTX-CCW-LOHT 17 34,265,020 227 Duplicates corrected: these two templates are redundant. -PG is used in SPAR models while -LOHT is used in HTX-PG-CCW 17 34,265,020 227 data collection and analysis. Redundant entry w as removed from the 2019 Update.

PORV-FC 13 49,398,360 317 Duplicates corrected: PORV-FC is a redundant template in the 2015 Update. It was removed from the 2020

29 ID Failures Demands/Hours Components NRC Response PORV-FC-MSS 13 49,398,360 317 Update.

SRV-ELS 0 72,220,220 558 Not duplicates: there was 0 SRV external leakage small (ELS) event and 0 fails-to-control event (FC).

SRV-FC 0 72,220,220 558 ROD-FTOP 20 132,832,800 846 Not duplicates: these are for two different failure modes, control rod fails-to-operate (FTOP) and control rod ROD-SOP 20 132,832,800 846 spurious operation (SOP). It was a coincidence that there were 20 FTOP events (5 from Columbia, 3 from SLC2, 2 from Oconee 1, etc.) and 20 SOP events (7 from S LC1, 8 from SLC2, and 1 each for other plants).

Notes:

• These entries are blank in the NRC Dataset (2015).

30

Table DQ 1-2 Miscellaneous Issues on 2015 NRC Dataset with NRC Review Results

Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range AFW-EDP-FTR<1H AFW Engine-EPIX/RADS 4 739 h 5 6.09E-3 2.0 1998-2015 Limited Data The data and estimation are driven pump Fails (<1000 hrs) appropriate to Run <1H AFW-EDP-FTR>1H AFW Engine-EPIX/RADS 2 262 h 5 9.53E-3 2.5 1998-2015 Limited run-driven pump Fails hours do not to Run >1H justify separate data parameters for Early Term and Late Term.

AOV-FTC Air Operated Valve EPIX/RADS 53 201,147 d 1767 3.63E-4 7.6 1998-2015 Questionable # Not an error even though Fails to Close of demands FTC, FTO, and FTOC have (same as AOV-the same demands.

FTOC)

AOV-FTO Air Operated Valve EPIX/RADS 78 201,147 d 1767 3.91E-4 5.2 1998-2015 Questionable #

Fails to Open of demands (same as AOV-FTOC)

CRB-CO-480 Low Voltage EPIX/RADS 39 396,295,000 h 2629 9.97E-8 1.3 1998-2015 ID should be CRB-was changed to CBK-in (480V) Circuit CBK-xxx to be the 2020 Update.

Breaker Spurious consistent with Operation other circuit breakers CRB-FTOC-480 Low Voltage EPIX/RADS 56 55,060 d 1776 1.03E-3 1.2 1998-2015 ID should be (480V) Circuit CBK-xxx to be Breaker Fails to consistent with Open/Close other circuit breakers CSW-MDP-FR CSW Motor Driven EPIX/RADS 25 3,654,539 h 32 6.98E-6 1.4 1998-2015 Data variable Data Correction: CSW was Pump Fails to Run and description changed to CWS in the 2020 should be CWS Update.

(circ water system) rather than CSW per Table 2-2 of Datasheets.

CTG-FTR Gas Turbine EPIX/RADS 5 648 h 2 8.49E-3 1.9 1998-2015 Limited run-The data and estimation are Generator Fails to hours do not appropriate.

Run, Late Term justify separate data parameters for Early Term and Late Term.

31 Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range HCU-FTI Hydraulic Control EPIX/RADS 0 d 0 1998-2015 No Failure Rate Data Correction: The data Unit Fails to Insert source should be the RPS System Study with a mean value of 1.10E-7; corrected in the 2020 Update.

MCC-FTOP Motor Control EPIX/RADS 6 34,080,880 h 217 1.91E-7 1.8 1998-2015 The end of the Data Correction: Typo in the Center Fail to Date Range 2015 Update; should be 2015 Operate (2018) is not 2018.

inconsistent with the rest of the data (2015) and appears to be in error.

MDC-FTR>1H Motor Driven EPIX/RADS 22 1,683,943 h 58 1.34E-5 1.4 1998-2015 It appears that Data Correction: The MDC-Compressor Fail to the data FTR<1H values in the 2015 Run (> 1 Hour) (successes, Update were incorrect. The failures) for 2020 Update will show correct MDC-FTR<1H values.

were used here.

MDP-CCW-FTS CCW Motor-driven EPIX/RADS 69 88,693 h 291 8.78E-4 4.0 1998-2015 Labeled FTS, Data Correction: Typo in the pump Fail to Start but "h" (per hour) 2015 Update; 2020 Update will show correct values.

MDP-SBY-FTR>1H Motor-Driven EPIX/RADS 143 20,062,180 h 1311 1.15E-5 6.5 1998-2015 # of run-hours No issue: There were 482,286 Pump Fails to Run, seems to be too demands over the time. The Late Term high for standby ratio of run-hrs/demands is pump (based on about 40 hrs/demand, which is RunHrs per in the normal area for a pump per yr) standby pump.

MOD-ILS Motor Operated EPIX/RADS 1 17,147,900 d 111 8.75E-8 3.3 1998-2015 Questionable # No issue: The run-hours for Damper Internal of demands leakage is based on calendar Leakage (Small) year (d should be h). 17.1 million hours for 111 components in 18 years (average 8583 run-hours/component/yr) look good.

MOV-FTC Motor Operated EPIX/RADS 234 740,890 d 6902 3.35E-4 3.0 1998-2015 Questionable # No issue that FTC, FTO, and Valve Fail to Close of demands FTOC have the same (same as MOV-demands.

FTOC)

MOV-FTC-BFV Butterfly Valve Fail EPIX/RADS 34 109,522 d 961 3.38E-4 4.0 1998-2015 Questionable #

to Close of demands (same as MOV-FTOC)

32

Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range MOV-FTO Motor Operated EPIX/RADS 293 740,890 d 6902 4.21E-4 2.3 1998-2015 Questionable #

Valve Fail to Open of demands (same as MOV-FTOC)

MOV-FTO-BFV Butterfly Valve Fail EPIX/RADS 27 109,522 d 961 2.51E-4 1.3 1998-2015 Questionable #

to Open of demands (same as MOV-FTOC)

NSW-MDP-FR Nuclear Service EPIX/RADS 64 10,256,170 h 104 6.72E-6 4.0 1998-2015 This variable is NSW-MDP-FR and MDP-Water MDP Fails named SWN SWS-FTR have different to Run (normally criteria with NSW for normally operating SW) in operating SW only and SWS the data sheets for both normally operating (Table 2-2). SW and standby SW.

Also, this data However, both use failure set has 104 mode of Fail to Run, which is components rarely used in standby compared to systems (which mostly use MDP-SWS-FTR FTR<1H and FTR>1H). The which has 106. 2020 Update addressed the It is not clear issue by clarifying whether a what the template is for normally difference is operating SW only or for all between these SW.

two groups.

PDP-SBY-FTR>1H Positive EPIX/RADS 2 1,710 h 72 1.46E-3 2.5 1998-2015 Limited run-The data and estimation are Displacement hours do not appropriate.

Pump Fails to Run, justify separate Late Term data parameters for Early Term and Late Term.

PLDTFTOP Process Logic RPS 24 4,887 d 5.07E-3 8.4 Documentation Data Correction: the comment (Delta SSs error. The should be removed.

Temperature) Fail Comment says, to Operate The PLL data was used to estimate the flow process logic reliability. It should not refer to flow devices.

33

Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range PMP-Volute Pump Volute Fails EPIX/RADS 25 158,885 h 207 1.60E-4 1.4 1998-2015 # of run-hours There were about 54,000 to Run (Driver seems to be too demands over the time. The Independent high compared ratio of run-hrs/demands is Centrifugal Pumps) to other MDP-about 3 hrs/demand, which is SBY based on actually on the lower side of RunHrs per normal for standby pumps.

pump per yr)

PORV-FC Main Steam Power EPIX/RADS 13 49,398,360 d 317 2.57E-7 10. 1998-2015 Questionable # Data Correction: per demand Operated Relief 0 of demands should be per hour Fails to Control (Cooldown)

PORV-FC-MSS Main Steam Power EPIX/RADS 13 49,398,360 d 317 2.57E-7 10. 1998-2015 Questionable #

Operated Relief 0 of demands Fails to Control (Cooldown)

PORV-L PORVs/SRVs Special d 0 1.48E-1 Data Source The special calculation refers Open During Calculation should use to NUREG/CR-7037, Relief LOOP reference: Valve Study NUREG/CR-7037 Table 13 PORV-Liquid PORVs Fail to Special 0 7 d 0 6.25E-2 10. Data Source Close After Calculation 3 should use Passing Liquid reference:

NUREG/CR-7037 Table 30 PORV-P1 PWR One Special 18 13,897 d 0 1.46E-3 2.8 Data Source The special calculation refers PORV/SRV Sticks Calculation should use to NUREG/CR-7037.

Open reference:

NUREG/CR-7037 Table 30 PORV-T PORVs/SRVs Special d 0 3.55E-2 Data Source Open During Calculation should use Transient reference:

NUREG/CR-7037 Table 13 ROD-FTOP Control Rod Fails EPIX/RADS 20 132,832,800 d 846 1.54E-7 1.4 1998-2015 Questionable EF Parameter estimate to Operate/Insert (b) uncertainty would be a special Rod issue to address.

34

Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range STTFTOP-D Sensor/Transmitter RPS 17 40,759 d --- 4.32E-4 8.4 Documentation Yes, this was an error. The (Temperature) Fail SSs error. The comment should be removed.

to Operate on Comment says, Demand The STL data was used to estimate the flow sensor/transmitt er reliability. It should not refer to flow devices.

SWS-MDP-FS-NE SWS Pump Non-EPIX/RADS 149 249,957 h 446 7.55E-4 3.8 1998-2015 Labeled FTS, Yes, it should be d Enviro-FTS but "h" (per hour)

TBV-FTC Turbine Bypass EPIX/RADS 0 2,725 d 73 1.83E-4 8.4 1998-2015 Questionable # No issue that FTC, FTO, and Valve Fails to of demands FTOC have the same Close (same as TBV-demands.

FTOC)

TBV-FTO Turbine Bypass EPIX/RADS 8 2,725 d 73 3.12E-3 1.7 1998-2015 Questionable #

Valve Fail to Open of demands (same as TBV-FTOC)

TDP-FR-L-HCI-RCI HCI-RCI Turbine EPIX/RADS 10 1,922 h 59 5.52E-3 2.8 1998-2015 Limited run-The data and estimation are Driven Pump Fails hours do not fine.

to Run, Late Term justify separate data parameters for Early Term and Late Term.

TDP-FS-NR-MFW MFW Turbine EPIX/RADS 62 5,984,882 d 43 1.09E-5 3.2 1998-2015 # of demands This was a copy and paste Driven Pump Fails seems to be too error (using the FTR data) in to Start, Normally high (based on the spreadsheet. The 2015 Running demands per Data Sheet shows the correct pump per yr) number of 12 failures in 1,395 demands.

TNK-FC Tank Rupture EPIX/RADS 15 59,350,270 h 379 2.61E-7 1.5 1998-2015 The end of the Yes, this is a typo in the 2015 Date Range Update.

(2018) is inconsistent with the rest of the data (2015) and appears to be in error.

TSA-FS Traveling Screen EPIX/RADS 97 30,417,290 d 212 3.67E-6 6.9 1998-2015 Questionable # TSA-FS was added to the Fails to Start of demands 2015 Update using the same data of TSA-FTOP, which may not be appropriate. TSA -FS was removed from the 2020 Update.

35

Component Failure Description Data Failures Demands/ d/h Component Mean EF Date Data Issue NRC Response Mode Source Hours s Range VBV-FTC Vacuum breaker EPIX/RADS 6 27,842 d 167 2.15E-4 5.7 1998-2015 Questionable # No issue that FTC, FTO, and fails to close of demands FTOC have the same (same as VBV-demands.

FTOC)

VBV-FTO Vacuum Breaker EPIX/RADS 2 27,842 d 167 8.98E-5.5 1998-2015 Questionable #

Valve Fail to Open of demands (same as VBV-FTOC)

36