ML21286A682
| ML21286A682 | |
| Person / Time | |
|---|---|
| Site: | Callaway  |
| Issue date: | 10/13/2021 |
| From: | Ameren Missouri |
| To: | Office of Nuclear Reactor Regulation |
| Shared Package | |
| ML21286A680 | List: |
| References | |
| ULNRC-06689 | |
| Download: ML21286A682 (18) | |
Text
Enclosure to ULNRC-06689
Enclosure to
ULNRC- 06689
Ameren Missouri Response to NRC RAIs
17 pages
Enclosure to ULNRC-06689 Page 1of 17
Ameren Missouri Response to NRC RAIs
On October 30, 2020, Union Electric Company, dba Ameren Missouri, submitted a license amendment request (LAR) for Callaway Plant, Unit 1 (Agencywide Documents Access and Management System (ADAMS) Accession No. ML20304A454.) The proposed amendment would modify the Callaway licensing basis by the addition of a license condition (i.e., License Condition2.C.(19)) to allow for the implementation of the provisions of Title 10 of the Code of Federal Regulations (10 CFR),
Section 50.69, Risk-informed categorization and treatment of structures, systems and components for nuclear power reactors. On February 18, 2021, the U.S. Nuclear Regulatory Commission (NRC) staff issued an audit plan(ADAMS Accession No. ML21039A222) that conveyed the intent to conduct a regulatory audit to support its review of the subject license amendment. NRC staff personnel and representatives of Ameren Missouri participated in the regulatory audit, conducted from May 11, 2021 through May13, 2021.The NRC issued an audit summary in a letter dated June 9, 2021 (ADAMS Accession No.ML21139A022). Ameren Missouri provided additional/revised information in a supplement dated July29, 2021 (ADAMS Accession No.ML21210A025).The NRC staff has reviewed the supplemental information provided after completion of the regulatory audit, and determined that the following additional information is required in order to complete the review of the subject LAR.
REQUEST FOR ADDITIONAL INFORMATION
RAI 01 -Use of Mean Core Damage Frequency and Large Early Release Frequency Values and Consideration of the State of Knowledge Correlation
=
Background===
RegulatoryGuide (RG) 1.174, Revision3, " An Approach for Using Probabilistic Risk Assessment in Risk-Informed Decisions on Plant-Specific Changes to the Licensing Basis," dated January 2018 (ADAMS Accession No ML17317A256) and Section6.4 of NUREG -1855, Revision1, " Guidance on the Treatment of Uncertainties Associated with PRAs [Probabilistic Risk Assessments] in Risk-Informed Decision making," dated March 2017 (ADAMS Accession No.ML17062A466), for a Capability Category II risk evaluation, indicate that the mean values of the risk metrics (total and incremental values) need to be compared against the risk acceptance guidelines. The mean values referred to are the means of the probability distributions that result from the propagation of the uncertainties on the PRA input parameters and model uncertainties explicitly reflected in the PRA models. In general, the point estimate core damage frequency (CDF) and large early release frequency (LERF) obtained by quantification of the cut set probabilities using mean values for each basic event probability does not produce a true mean of the CDF and LERF. Under certain circumstances, a formal propagation of uncertainty may not be required if it can be demonstrated that the state of knowledge correlation (SOKC) is unimportant (i.e., the risk results are well below the acceptance guidelines).
Enclosure to ULNRC-06689 Page 2of 17 of the updated LAR, i.e.,the response to Audit Question APLA/APLC 05 Part (c),
provides updated quantification results based on PRA update 9.01 using mean values which account for the SOKC showing the total CDF and LERF meet the RG 1.174 risk acceptance guidelines.
However, the updated LAR did not state whether fire, seismic and wind PRA parameters that should be correlated to account for SOKC were included, such as circuit failure probabilities, suppression probabilities, and ignition frequencies for the fire PRA. The response also states that the mean values are conservative because of the inability to full y post-process with ACUBE [advanced cut set upper bound estimator]. It is not clear how this conservatism (i.e., that the mean values are 19 and 27 percent higher than the point estimate values for CDF and LERF, respectively) will be addressed for the 10CFR 50.69 program, given t hat it would appear to impact the 10CFR 50.69 risk-informed categorization.
Request
Address the following:
a) Identify the parameters derived from the same data (other than for same type code data) that were correlated for the fire, seismic, and high wind PRAs, and justify that these parameters are sufficient to estimate the SOKC.
b) Discuss how the SOKC will be treated for the 10 CFR 50.69 program consistent with NUREG -
1855, Revision1, when the risk increase associated with SOKC is considered.
Response
Part A The potential for the state of knowledge correlation (SOKC), as described in NUREG-1855, to impact the seismic PRA (SPRA) is very limited due to the nature of the component grouping supporting the fragility development. Three areas of the SPRA are considered: hazard frequency, component fragility, and events propagated from the internal events model.
The seismic hazard frequency is taken from a single source and segmented to arrive at the hazard bin frequency based on hazard acceleration. Each hazard bin considers a unique region of the hazard curve over a specific range of accelerations. Therefore, each hazard interval is independent of the other hazard intervals and no SOKC exists.
Component fragilities can provide a source for SOKC impact when multiple component groups utilize the same fragility development. For example, if auxiliary feedwater and safety injection motor -
operated valves were based on the same fragility development,there would be a potential SOKC impact when both seismic failures occurred in the same cut set. For Callaway, this type of relationship does not occur because all component groups are independent of each other with respect to the fragility development as applied to the group. The component grouping is location specific and assumes complete correlation between grouped components, which results in single-point failures representing the failure of the component group.
Enclosure to ULNRC-06689 Page 3of 17
Elements of the model listed in the abo ve motor-operated valve example are not based on system but are location specific, considering local component anchorage and the floor spectra in the development.
Thereis no SOKC contribution required for this type of modeling since the correlation is complete between the grouped components and the fragility values are uniquely assessed based on grouped component location.
The last contribution involves relationships within the internal events model (random failures) that are propagated to the SPRA. These contributions can arise for low accelerations and introduce the potential for SOKC impacts. For example, the seismically induced loss of offsite power is an important contributor to plant CDF,and at the lower acceleration ranges,is controlled by the non-seismic loss of both onsite diesel generators. This situation is a parallel to the internal events station blackout sequences,and there is some potential for SOKC to be like that postulated for the internal events model. However, the lower acceleration events are not significant contributors to overall CDF and LERF, so the influence of this random combination is not significant.
Overall, SOKC is not a significant factor in the estimation of CDF or LERF for the SPRA,and special treatment is not required. The following numeric assessment supports this position and provides a basis for the difference in the baseline mean value (point estimate) for CDF and LERF and the mean value (sampled estimate) generated by the UNCERT application. The different results are shown in Table 1.
Table 1. Baseline and UNCERT Results
Metric Baseline Point UNCERT Point Increase UNCERT Sampled Increase Estimate Mean Estimate Mean over Mean (/y) over
(/y) (/y) Baseline Baseline CDF 4.01E-5 5.83E-5 45% 5.83E-5 45%
LERF 4.43E-6 4.56E-6 3% 5.96E-6 35%
Review of the uncertainty assessment identified two aspects of the uncertainty quantification process that result in the uncertainty mean values (UNCERT/ACUBE) exceeding the baseline point estimate mean generated by the baseline model (FTREX/ACUBE). Both aspects are related to differences in the quantification process associated with the use of the ACUBE software to factor the model results during the calculation process.
The first aspect is an UNCERT limitation associated with ACUBE and the quantification process.
UNCERT can apply the refinement performed by ACUBE, but certain ACUBE preprocessing steps of the refinement process cannot be applied,and this influences the results. Preprocessing includes applying rules to true, false, and events with a probability of 1.0 to remove them from the quantification. It also includes subsequent sorting, truncating, and compressing of cut sets.
In contrast, the baseline calculation can take advantage of the ACUBE preprocessing,and the results reflect the impact of this refinement.
Enclosure to ULNRC-06689 Page 4of 17
A sensitivity was performed that removed the preprocessing function (identified by application of the NoPrep option in ACUBE). The results were compared to the UNCERT results shown in Table 2 for CDF.
Table 2. Comparison of NoPrep Option on Baseline and UNCERT CDF
Baseline Point Baseline Point UNCERT UNCERT ACUBE Estimate Estimate Point Sampled Baseline Factored Mean (/y) Mean (/y) Estimate Estimated w/PREP Baseline MCS w/PREP NoPrep Mean (/y) Mean (/y) Delta1 NOPREP Delta2 0 6.02E-05 6.02E-05 6.02E-05 6.05E-05 0.5% 0.5%
1000 5.26E-05 6.02E-05 6.02E-05 6.01E-05 14.3% -0.2%
2000 5.00E-05 5.97E-05 5.98E-05 5.97E-05 19.4% 0.0%
3000 4.79E-05 5.83E-05 5.83E-05 5.83E-05 21.7% 0.0%
4000 4.63E-05 5.69E-05 5.68E-05 5.69E-05 22.9% 0.0%
5000 4.52E-05 5.57E-05 5.56E-05 5.56E-05 23.0% -0.2%
6000 4.43E-05 5.45E-05 5.45E-05 5.45E-05 23.0% 0.0%
7000 4.37E-05 5.35E-05 5.35E-05 5.35E-05 22.4% 0.0%
8000 4.31E-05 5.27E-05 5.26E-05 5.26E-05 22.0% -0.2%
9000 4.27E-05 5.20E-05 5.20E-05 5.20E-05 21.8% 0.0%
10000 4.23E-05 5.13E-05 5.12E-05 5.12E-05 21.0% -0.2%
- 1. Percent difference between Baseline mean value with preprocessing and UNCERT sampled mean value
- 2. Percent difference between Baseline-NoPrep mean value and UNCERT sampled mean value
When the ACUBE preprocessing is removed from the baseline quantification, the results match the UNCERT results. Therefore, the differences noted in the RAI are due to limitations in the UNCERT application that preclude a level of refinement possible for the baseline analysis. The same conclusion can be drawn for the LERF assessment in that the results are matched when the baseline does not apply the preprocessing option.
Using the UNCERT results presented in Table 2,the impact of SOKC can also be estimated. The UNCERT point estimate is generated without inclusion of SOKC,and the sampled estimate includes consideration of SOKC in the sampling process. The results are virtually identical and support the position that SOKC is not a sig nificant contributor. The observed difference between the baseline with preprocessing and the UNCERT sampled mean is due to the ACUBE attribute, and it is not a function of the state of knowledge correlation (SOKC).
The second aspect of the analysis is the number of cut sets factored by ACUBE using the binary decision diagram (BDD) approach. The UNCERT results are based on selecting the top 3,000 cut sets for ACUBE refinement. Examination of Table 2 identifies that the baseline result (4.79E -05/y),when using 3,000 cut sets for refinement,does not equal the reported CDF.
Enclosure to ULNRC-06689 Page 5of 17
The selection of a certain number of min-cut upper bound cut sets for refinement is an input parameter to ACUBE. The number of cut sets to be factored for the baseline quantification is based on a review of CDF improvement obtained,and the assessment is documented in Section 4.2.3 of PRA -SEISMIC-QUANT, Revision 1. This study identified that the use of 3,000 factored cut sets was the best choice for each seismic hazard range. When a hazard range did not generate 3,000 cut sets then all were factored. Given ten hazard ranges and between 1,600 and 3,000 cut sets per hazard range, the best estimate calculation applied ACUBE factoring (BDD) to approximately 27,000 cut sets.
The UNCERT application also utilized the ACUBE application but only applied BDD factoring to the top 3,000 cut sets. Since the refinement is designed to remove conservatism in the result, the UNCERT result is not as refined and generates a higher CDF and LERF. Table 3 provides a comparison of the baseline ACUBE results to the UNCERT results for CDF with increasing number of cut sets.
Table 3. Comparison of ACUBE Impact on CDF for Baseline and UNCERT
ACUBE Baseline Point Factored Estimate Mean UNCERT Point UNCERT Sampled MCS (/y) Estimate Mean (/y) Estimated Mean (/y) Baseline Delta1 0 6.02E-05 6.02E-05 6.05E-05 0.5%
1000 5.26E-05 6.02E-05 6.01E-05 14.3%
2000 5.00E-05 5.98E-05 5.97E-05 19.4%
3000 4.79E-05 5.83E-05 5.83E-05 21.7%
4000 4.63E-05 5.68E-05 5.69E-05 22.9%
5000 4.52E-05 5.56E-05 5.56E-05 23.0%
6000 4.43E-05 5.45E-05 5.45E-05 23.0%
7000 4.37E-05 5.35E-05 5.35E-05 22.4%
8000 4.31E-05 5.26E-05 5.26E-05 22.0%
9000 4.27E-05 5.20E-05 5.20E-05 21.8%
10000 4.23E-05 5.12E-05 5.12E-05 21.0%
15000 4.18E-05 4.87E-05 4.87E-05 16.5%
20000 4.03E-05 Not Possible Not Possible -
25000 3.98E-05 Not Possible Not Possible -
30000 3.95E-05 Not Possible Not Possible -
- 1. Percent difference between Baseline mean value and UNCERT sampled mean value
The table illustrates how increasing the number of cut sets factored by ACUBE influences the results.
The first case does not include ACUBE, and the values match between the baseline, UNCERT point estimate, and UNCERT mean.
All columns show decreasing CDF with increasing refinement. This supports the conclusion that the remaining difference between the baseline point estimate mean reported for CDF and LERF and the Enclosure to ULNRC-06689 Page 6of 17
UNCERT sampled mean for the same attributes is due to the increased number of refined cut sets and not due to SOKC considerations. It should be noted that the UNCERT assessment stops at 15,000 because of UNCERT code limitations that preclude a higher number of ACUBE factored cut sets.
In summary, the difference observed between the baseline point estimate mean values and the uncertainty mean values is due to two specific aspects related to the ability to utilize more of the ACUBE tools for the baseline model than when applying it to UNCERT. The comparison of UNCERT results for the point estimate mean and the sampled mean are identical and indicate that SOKC has a negligible impact on the SPRA results.
For the Fire PRA, the events within the final results fall into a number of categories including house flags, house split fractions or conditional probabilities (e.g., train alignments), the plant availability factor, random component failures, operator actions and their associated recoveries, scenario specific frequencies (i.e., ignition frequencies, severity factors, and non-suppression probabilities), and hot short related events (i.e., hot short probabilities and spurious hot short clear duration probabilities). Of the events in the final results, those that will be impacted by the SOKC are the random failure probability events and the hot short events. The remaining events will not have any impact from the SOKC because they are independent failures. For example, non-suppression probabilities are calculated based on the source and target parameters and are ap plied as a single event to the scenario.
All events, as appropriate, are included with distribution information for the Monte Carlo sampling of the parametric uncertainty analysis. The hot short events are correlated utilizing the grouping of similar type codes within UNCERT. This sufficiently captures the necessary SOKC considerations for the Fire PRA in the final parameter uncertainty analysis.
For the High Winds PRA, the events within the results fall into a number of categories including house flags, house split fractions or conditional probabilities (e.g., train alignments), the plant availability factor, random component failures, operator actions and their associated recoveries, initiating event frequencies, and high winds related failure (e.g., missile or wind pressure). The results reflect random failures in a manner similar to internal events,and as such, the SOKC would be similar. The remaining events, including high winds fragilities, will not have any impact from the SOKC because they are independent failures. All events, as appropriate, are included with distribution information for the Monte Carlo sampling of parametric uncertainty analysis. The baseline High Winds PRA UNCERT results were presented in the original LAR without processing with ACUBE. Allowing UNCERT to process the cut sets with ACUBE shows stronger convergence between the point estimate and sampled mean value, showi ng the impact of the SOKC on the model results is minimal. Table 4 provides the UNCERT results with and without ACUBE processing.
Enclosure to ULNRC-06689 Page 7of 17
Table 4. Comparison of ACUBE Impact on HW CDF and LERF for Baseline and UNCERT
MCUB ACUBE CDF LERF CDF LERF Baseline Point Estimate 5.89E -06 2.55E -07 5.42E -06 2.55E -07 UNCERT Sampled 6.66E -06 5.27E -07 5.82E -06 2.57E -07 Estimated Mean Delta 13% 106% 7% 1%
The issue presented here is the same with the Seismic model. The inability of UNCERT to run with the NoPrep option disabled does not allow for closer convergence between the point estimate and sampled mean, but use of ACUBE does show a minimal impact fromthe SOKC. This sufficiently captures the necessary SOKC considerations for the High Winds PRA in the final parameter uncertainty analysis.
Part B The response to Part A identifies that SOKC is not a significant contributor to CDF or LERF results and tha t no specific treatment is necessary for the 50.69 application
RAI 02 -Treatment of Sensitive Electronics in the Fire PRA
=
Background===
RG 1.200, "Acceptability of Probabilistic Risk Assessment Results for Risk -Informed Activities,"
states, "NRC reviewers, [will] focus their review on key assumptions and areas identified by peer reviewers as being of concern and relevant to the application." The relatively extensive and detailed reviews of fire PRAs undertaken in support of LARs to transition to National Fire Protection Association (NFPA)-805, " Performance-Based Standard for Light Water Reactor Electric Generating Plants," determined that implementation of some of the complex fire PRA methods often used non-conservative and over-simplified assumptions to apply the method to specific plant configurations.
Some of these issues were not always identified in facts and observations (F&Os) by the peer review teams,but are considered potential key assumptions by the NRC staff because using more defensible and less simplified assumptions could substantively affect the fire risk and fire risk profile of the plant.
LAR Section 3.2.2 states that the numerous new or revised fire PRA guidance documents issued since Callaway was approved to implement NFPA-805 are being addressed through the PRA maintenance and update process. With regards to use of Frequently Asked Question (FAQ)13- 0004, "Clarifications on Treatment of Sensitive Electronics"(ADAMS Accession No. ML13322A085), Attachment 8 of the updated LAR states that "[w]ithout explicitly citing FAQ 13 -0004 for the treatment of sensitive electronics, the fire PRA does implement the salient conclusion that a generic screening heat flux damage threshold for thermoset cables, as observed on the outer surface of the cabinet, can be used as a conservative surrogate for assessing the potential for thermal damage to solid-state and sensitive electronics within an electrical panel (cabinet)." The LAR, however, does not address the two caveats Enclosure to ULNRC-06689 Page 8of 17
cited in FAQ 13-0004 that can invalidate this approach which are (1) sensitive electronics mounted on the surface of the cabinet where it can be exposed to the convective or radiant energy of a fire, and (2) the presence of a louver or other typical ventilation means. It appears to the NRC staff that this source of fire PRA modeling uncertainty could have the potential to impact the application.
Request
Address the following:
a) Explain how sensitive electronics are modeled in the fire PRA for sensitive electronics cabinet configurations (i.e., wall mounted electronics or ventilation) that invalidate the FAQ 13 -0004 approach described in the LAR (i.e., using the heat flux damage threshold for thermoset cables for sensitive electronics inside a cabinet). Include confirmation that the approach is consistent with the guidance in FAQ 13- 0004.
b) If the approach cannot be confirmed to be consistent with the guidance in FAQ 13- 0004,then justify that it has an inconsequential impact on 10 CFR 50.59 risk-informed categorization.
Alternatively, explain how the uncertainty associated with this modelling treatment will be addressed in the 10CFR 50.69 program.
Response
Part A Sensitive electronics (SEs) at Callaway were modeled using the thermoset cable heat flux damage threshold. At the time of the analysis, the FAQ closure memo had not been published and the caveats were not addressed in the original assessment. A review of all fire modeled areas was performed to identify potential cabinets that may invalidate the FAQ caveats. This review identified a small number of cabinets that might invalidate the caveats, based on venting configuration. These cabinets were compared against the PRA to determine if they were credited for any functions. Only three cabinets were identified with a potential for invalidating the FAQ caveats and impacting the PRA. Of those three cabinets, two were confirmed to only contain electro-mechanical components, which per FAQ 13- 0004, do not need to be considered sensitive electronics. This left one cabinet containing sensitive electronics that did not meet the caveats. This cabinet is PN09, located in Fire Area C -9, which is a 5-kVA 120V Inverter.
Part B The fire scenario analysis for C-9 was reviewed to assess the zone-of-influence (ZOI) used for failures of the PN09 cabinet. The review concluded that surrounding scenarios should be expanded to account for the lower damage threshold of the SE cabinet (i.e., a larger ZOI). The only scenario impacted was IN IT-F-C9-3301-T1 (TS#1), which was a transient fire scenario abutting the cabinet. The floor area weighting factor assigned to this transient fire scenario did not adequately account for the expanded ZOI related to the lower SE damage threshold. The transient fire analysis in C -9 uses a grid approach such that increasing the floor area of one transient zone will decrease the area of another. The remaining floor area transient fire was represented by INIT-F-C9-3301-T9 (TS#9). The transient fire Enclosure to ULNRC-06689 Page 9of 17
scenario, TS#1, would require increasing the existing ZOI dimensions by 0.9 feet. There was no change to the damage set of the scenario due to the increased floor area. To assess the impact, the transient fire scenario floor areas were re-assessed with the larger ZOI and a new floor area calculated.
To conserve the compartment frequency, the remaining floor area for other transient fire scenarios decreased by the same amount that TS#1 increased. The f ire ignition frequency changes are as follows:
TS#1 TS#9 Baseline 8.584E -08 1.123E -05 Revised 1.337E -07 1.118E -05 Using the existing analysis, the delta CDF and LERF to account for this change is as follows:
CDF TS#1 TS#9 Baseline 2.913E -12 1.209E -09 Revised 4.538E -12 1.203E -09 Total Delta CDF Delta 1.625E -12 -6.000E -12 -4.375E-12
LERF TS#1 TS#9 Baseline 2.572E -14 6.705E -12 Revised 4.006E -14 6.676E -12 Total Delta LERF Delta 1.434E -14 -2.900E -14 -1.466E -14 Because the single instance of modeling sensitive electronics described above is inconsequential, the uncertainty associated with the modeling treatment is adequately addressed in the 10 CFR 50.69 program.
Background
RG 1.174, Revision3, states, in part, that the plant -specific PRA supporting the licensees proposals has been demonstrated to be acceptable.
Review of PRA-SEISMIC-QUANT (described in the Audit Question APLA/APLC -03 Response in, Attachment 8 to the LAR supplement dated July29, 2021), Table 4-2 indicates that the highest ACUBE conditional core damage probability (CCDP) is 0.9,and its note (2) states that the value in the column includes the plant availability factor (PAF) of 0.9. However, Table4-3 shows that ACUBE conditional large early release probability (CLERP) is close to 1. It is not clear why the value in the column does not include the PAF of 0.9, since they are very similarly calculated as a ratio of either ACUBE CDF or ACUBE LERF to seismic frequency. In addition, on page 54 it says that both CCDP and CLERP approach the PAF of 0.9 for the last hazard interval bin. However, Figures 4-3, Enclosure to ULNRC-06689 Page 10of 17
4-4, and 4-5 do not demonstrate this. Instead, Figures 4-3 and 4-5 show CCDP more than 0.9, while Figures 4-4 and 4-5 show CLERP is close to 1.
The last bin for seismic CDF represents 0.88g peak ground acceleration, covering a seismic hazard from 0.8 to 10g.
Request
Provide a sensitivity study to demonstrate no impact on the final seismic CDF when this bin is divided into several sub-bins (for example, 0.8-1.0g, 1.0- 1.2g, 1.2-1.4g, 1.4-1.6g, 1.6-2.0g and 2.0- 10g).
Response
The hazard ranges selected for the SPRA are based on a detailed assessment documented in Section 3.4.1 of PRA -SEISM IC-QUANT, Revision 1. As a result of this assessment, two different range selections were chosen for CDF and LERF to simplify the analysis once the plant conditional core damage probability (CCDP) and/or the conditional large early release probability ( CLERP) reached 1.0. For the SPRA CDF model, this was found to occur near the 0.88g acceleration. At that point, the CCDP would be so close to 1.0 that the results would be controlled by the seismic hazard frequency alone.
The one special consideration is that the plant availability factor (PAF) is applied to all cut sets, and this will reduce the calculated CCDP by the factor (0.897). Therefore, any CCDP greater than this value is indicative of a CCDP of 1.0.
A sensitivity study was conducted todivide the upper range of the hazard intervals (%G10) into six intervals. These represented smaller segments of the %G10 range and collectively cover the range.
The seismic hazard was apportioned between the new ranges,and the model was updated,in orde r to accommodate the new information. The updated model was solved for the six new intervals and the results are provided in Table 5.
Enclosure to ULNRC-06689 Page 11of 17
Table 5. Results of Sensitivity Study on Upper Seismic Hazard Interval
ACUBE CCDP Hazard Interval Description Representative Frequency ACUBE (CDF over Scenario and Range Acceleration (/y) CDF (/y) estimation)
Seismic Initiating Event (0.8g 0.8485 2.73E-06 2.44E-06 9.96E-01
%G10_1 to 0.9g)
Seismic Initiating Event (0.9g 0.9487 1.76E-06 1.58E-06 1.00E+00
%G10_2 to 1.0g)
Seismic Initiating Event (1.0g 1.0954 1.89E-06 1.72E-06 1.01E+00
%G10_3 to 1.2g) (2.5E-8)
Seismic Initiating Event (1.2g 1.3416 1.34E-06 1.24E-06 1.03E+00
%G10_4 to 1.5g) (3.92E-8)
Seismic Initiating Event (1.5g 1.7321 7.74E-07 7.00E-07 1.01E+00
%G10_5 to 2.0g) (5.77E-9)
Seismic Initiating Event 2.4495 3.63E-07 3.26E-07 1.00E+00
%G10_6 (>2.0g)
Total CDF 8.01E-06 7.00E-08
The result is then compared to the current baseline results for %G10. The baseline value is 7.96E-6/y.
Dividing the single interval into six intervals results in an increase in CDF for this range of 5.0E-8/y and represents an increase in the total CDF of approximately 0.6%.
This increase is completely driven by overestimation of CCDP (probability greater than 1.0) for three of the six ranges. When these three are normalized, the calculated mean is reduced by 7.0E-8/y, leading to a final value of 7.94E-6/y, whichis less than the baseline value by 0.3%. This indicates that there is no difference in the %G10 seismic hazard range results for the CDF when using a single bin or more finely discretizing the interval.
The use of a single interval for the acceleration range greater than 0.8g has an insignificant impact on the reported CDF.
RAI 04 -Seismic Fragility Analysis Uncertainty
=
Background===
Section 5 of Nuclear Energy Institute (NEI) 00 -04, 10CFR 50.69 SSC [Structure, System, and Component] Categorization Guideline, provides guidance for performing sensitivity studies for each PRA model to address the uncertainty associated with those models. Specifically, Sections 5.1, 5.2, and 5.3 provide guidance for such sensitivities for the internal events, fire and seismic PRA, respectively. The sensitivity studies are performed to ensure that as sumptions and sources of uncertainty (e.g., human error, common cause failure, and maintenance probabilities) do not mask importance of components.
Enclosure to ULNRC-06689 Page 12of 17
The high seismic CDF and LERF values relative to the CDF and LERF values from the other hazards including internal events suggest that the uncertainty in seismic PRA modeling involving the level of detail used to model fragility could potentially impact 10 CFR 50.69 risk-informed categorization. It is not clear from the discussion provided in Attachment 8 of the updated LAR in response to Audit Question APLC 04 on the level of fragility analyses performed for four dominant CDF and LERF importance contributors (i.e., seismically induced loss of offsite power, failure of service water, failure of steam gener ator supports and soil failure) whether the Callaway seismic CDF and LERF values could be further reduced by further refining the fragility analyses for other SSCs. NRC staff observes that since [subsequent to] the point estimate seismic CDF of 5.59 E-05 per year [being] presented in Section 6 of the Callaway seismic PRA report dated July10, 2020 (ADAMS Accession No. ML20192A244) the calculated point estimate seismic CDF value has been reduced to 4.01E-05 based on refinements according to Attachment 1 of the updated 10CFR 50.69 LAR. An overly conservative seismic PRA model could skew the integrated importance measures calculated for 10CFR 50.69 categorization. Importance measures provide a relative measure of risk importance and, therefore, if the importance of certain SSCs is significantly overestimated, then the importance of other SSCs will be underestimated.
Request
In light of these observations, address the following:
a) Justify that not using more refined fragility analysis for certain important SSCs will not have a consequential impact on 10CFR 50.69 risk-informed categorization. Include in the discussion how the SSC fragility analysis for important SSCs beyond just the top four contributors is performed.
b) If it cannot be justified that using more refined fragility analysis for certain important SSCs will not have a consequential impact on 10CFR 50.69 categorization, then explain how this modeling uncertainty will be addressed for the 10 CFR 50.69 program.
Enclosure to ULNRC-06689 Page 13of 17
Response
Part A The significant CDF and LERF contributing fragilities are listed in Table 6.
Table 6. Fragilities with Fussell-Vesely Importance above 0.01
Metric Fragility Failure Mode HCLPF Am Bc Level of Assessment Generic value based on ALWR Requirements Value (PRA-Offsite Functional SEISMIC-FRAGILITY APP40, CDF Power Failure 0.083 0.3 0.55 Revision 0)
Plant Walkdown, Experience mapping (PRA-SEISMIC-FRAGILITY APP40, Revision 0);
Normal Non-safety class I components, not Service Functional available after loss of non-safety CDF Water Failure 0.067 0.24 0.45 power Essential Service Plant-specific and location-specific Water Bldg assessment (PRA-SEISMIC-MCCs FRAGILITY APP40, Revision 0);
NG05E and Functional EPRI 3002002933, Section 4.3 source CDF NG06E Failure 0.3 0.76 0.4 of capacity data MOVs in Detailed evaluation based on site-Auxiliary specific realistic in structure response Building at (PRA-SEISMIC-FRAGILITY 2047' or Functional APP24, Revision 0); EPRI CDF below Failure 0.35 0.79 0.35 3002002933 source of capacity data
Steam Generator Structural Detailed evaluation (PRA-SEISMIC-LERF Supports Failure 0.8 1.86 0.43 FRAGILITY APP14, Revision 0 )
Soil Plant Specific Assessment (PRA-Structure SEISMIC-FRAGILITY APP40, LERF Interaction Soil Failure 0.74 1.67 0.35 Revision 0)
Plant-specific and location-specific calculation for Equipment Hatch RB (PRA-SEISMIC-FRAGILITY Penetration Structural APP15, Revision 0/ PRA -SEISMIC-LERF Failure Shear Failure 0.78 1.78 0.35 FRAGILITY APP32, Revision 0 )
Plant-specific and location-specific RB Functional calculation (PRA-SEISMIC-LERF MiniPurge Failure 0.55 1.56 0.45 FRAGILITY APP40, Revision 0 )
The listed fragilities control the contribution to CDF and LERF for the SPRA. As indicated, all listed fragility values, except for offsite power, are based on plant -specific evaluations that address location-specific considerations such as floor response. Given the current level of assessment, significant changes in these component capacities would not be expected.
Enclosure to ULNRC-06689 Page 14of 17
The offsite power fragility is based on generic information contained in the EPRI ALWR Requirement Document [Advanced Light Water Reactor Utility Requirements Document (ALWR URD) Volume III, Chapter 1 Appendix A: PRA Key Assumptions and Ground Rules, Revision 7, December 1995].
The value is based on an evaluation of industry fragility data contained in report UCID-20571
[Lawrence Livermore National Laboratory, Compilation of Fragility Information from Available Probabilistic Risk Assessments, UCID -20571, September 1985].
The evaluation is documented in a supporting report [Generic Component Fragilities for the GE Advanced BWR Seismic Analysis, Advanced Reactor Severe Accident Program, September 1988].
This document defines the characteristics related to the fragility development.
The offsite fragility is an aggregate of possible failure modes within the switchyard and transmission grid. Examining the collected data identified that the most limiting factors were centered around the main transformers. These transformers are not typically designed for seismic loading and exhibited a fairly low capacity.
Most of the information was basedon historical experience for conventional power plants subjected to seismic events. The governing failure mode was found to be a failure of the transformer ceramic insulators due to cracking. Other potential failures were associated with other electrical malfunctions such as an oil circuit breaker control.
Another consideration was that the transformers were located at site level and there was very limited impact by amplification factors to adjust the base performance. The range of offsite median capacity is presented in Figure 1.
Enclosure to ULNRC-06689 Page 15 of 17
Figure 1. Seismic Capacity Range for Offsite Power (reproduced from Generic Component Fragilities for the GE Advanced BWR Seismic Analysis, Advanced Reactor Severe Accident Program, September 1988)
The fragility spread of experience data is small and the median value was 0.29g. This was increased to 0.3g for ALWR performance.
The characteristics identified in the ALWR evaluation as being pertinent to the developed fragility were compared to the Callaway facility to determine if the designs were applicable and the predicted performance was a realistic estimate for the Callaway site. The comparison included items such as major equipment spatial information, the applicability of failure modes to major equipment, and soil structures. The evaluation concluded that the characteristics matched between the ALWR evaluation and the Callaway facility. The reported fragility is appropriate for use in the Callaway SPRA.
Given the importance to CDF, a review of several more recent offsite power seismic fragility assessments was conducted to determine the potential for increasing the current offsite power fragility in terms of the current state of practice. The findings of the review are summarized below.
Diablo Canyon Power Plant (DCPP) was requested by the NRC to reassess the fragility value (for the 230-kV switchyard) used in the plants Long Term Seismic Program (LTSP) considering 1989 Loma Prieta earthquake operational experience data. The reassessment showed that the DCPP switchyard itself did not suffer any damage for an estimated horizontal ground acceleration of 0.35g.
Enclosure to ULNRC-06689 Page 16of 17
However, damages to transformers and switches were reported at other non-nuclear power stations operated by PG&E during other earthquakes with estimated ground acceleration in the range of 0.20 to 0.29g [Reassessment of the HCLPF Value of the Diablo Canyon 230 kV Switchyard Following the 1989 Loma Prieta Earthquake, Pacific Gas and Electric Co., Diablo Canyon Long Term Seismic Program, August 1990.].
The current state of practice in SPRA is to use the fragility values recommended in EPRI SPRAIG
[Seismic Probabilistic Risk Assessment Implementation Guidelines, EPRI, Final Report 3002000709, December 2013.]. Table H-1 of the SPRAIG provides a compilation of fragility values used in past SPRAs and recommends representative values for use in SPRAs going forward.
In the case of offsite power fragility, it is seen that past SPRAs have used fragility values (Am) ranging from 0.2g to 0.3g as was observed in the ALWR evaluation. SPRAIG recommended a value of 0.3g,which corresponds to the higher end of the values used in the past which is also consistent with the ALWR recommendations. In doing so, the SPRAIG notes that representative fr agility values are not intended to be conservatively low values.
The SPRAIG recommended fragility values were used in the more recent SPRAs submitted (2018-19) to the NRC in response to the NTTF 2.1. A few plants (e.g., VC Summer, Diablo Canyon) were unsuccessful in attempts to improve the existing fragility values due to the complexity of the problem when the complete station switchyard and other aspects such as incoming lines, etc. are considered.
As a result of these activities, the consensus of industry experts is to use the EPRI SPRAIG value of 0.3g (Am),and this value represents the current state of practice.
More recently,EPRI started a study on the subject and issued a guidance report [ Loss of Offsite Power Fragility Guidance, EPRI, Technical Report 3002015993, August 2019]. That report reviewed available earthquake experience and operational data to date and alternative fragility characterizations for offsite power. It was recognized that the offsite power fragility is dependent on several components in the electrical distribution system serving the offsite power to a nuclear power plant.
The components include generation stations, transmission lines, switchyard, substations, etc. To derive a more realistic fragility value, a comprehensive response analysis of the entire offsite power system would be needed, with varying soil profiles, seismic hazards and structural response analysis.
Given the current state of knowledge and practice, such a task is just not considered practical and not commensurate with the improvement in offsite power capacity.
EPRIs most recent assessment [ Loss of Offsite Power Fragility Guidance, EPRI, Technical Report 3002015993, August 2019] identified that the use of the SPRAIG value (Am = 0.3g) fits well with the earthquake experience data (site independent) for midrange and highseismic ground motions (pga >
0.3g).
Given the current state of knowledge,as documented in EPRI report 3002015993, use of the ALWR Requirements Document value for offsite power fragility can still be considered the state of practice and the fragility value (Am = 0.3g) for the Callaway Plant offsite power is deemed to be appropriate and as site-specific as can be reasonably developed.
Enclosure to ULNRC-06689 Page 17of 17
Part B
The response to Part A justifies the current fragilities as realistic and that additional refinement w ould not provide an improvement sufficient to have any consequential impact on the equipment categorization for 10 CFR 50.69.