Bruce commented on VG4 with respect to poor US safety culture that didn’t see a problem until starting seeing circumferential cracks; Pete commented that it was an economic issue since US utilities charge 7 to 10 cents per KW-hr while overseas may charge 40 cents per KW-hr

Karen Gott pointed out that her experience was that the first time a plant experiences a problem it is a big problem so she would agree with second major bullet on VG#4; on second sub-bullet she thought a better word is experience instead of sensitive.

On VG#5 it was pointed out that while the NRC may have only one vote on code changes, it has the ultimate veto vote.

Bruce felt day of single plant utility is numbered. Helmut Schulz and Karen disagreed with the last sub­bullet on VG6; utilities are willing to invest in older plants since they are already paid for (less capital investment); may be a difference in international experience and US practice.

Helmut commented with regards to VG9 on decommissioning that they have not seen any increase in LER events in the last few years before decommissioning for those plants that were decommissioned.

With regard to VG10 and negative bullets related to risk-informed regulation. Bill Galyean commented that utilities have limited resources and risk informed process helps prioritize; Vic Chapman cautioned against turning crank and getting an answer without thinking of why.

Bruce commented that end of plant license renewal may be 80 years not 60; based on comment made at NRC recent meeting.

Helmut pointed out that boric acid corrosion of manway bolts of 15 to 20 years ago was more serious from a LOCA perspective that Davis Besse head problem of today.

Bottom line is no effect of safety culture on LOCA frequencies. There will be no adjustments to frequencies; no major discussion on part of panel with regards to this bottom line conclusion.

The failure rate development consists of determining an industry generic pipe failure rates for RR and FW piping, respectively. Next a Bayesian update is performed to generate failure rates that best represent the design and operating conditions of Plant B. Rather than taking an apriori failure from some published source, the approach in this study is to derive apriori failure information from the PIPExp database. The information that is summarized in Section D.3 provides some insights into the time-dependency of failure rates. These insights are explored further below.

D.4.2.1 RR Pipe Failure Data — Programs to mitigate the effects of certain degradation mechanisms strongly influence the achieved piping reliability. As an example, all BWR plants commissioned prior to the early to mid-1980s have experienced IGSCC. Industry initiatives to mitigate or eliminate the influence by IGSCC were implemented by the mid-1980s, and thereafter the rate of IGSCC has dropped sharply. The trend in the IGSCC rate is established by normalizing the data displayed in Figure D.2 (IGSCC by Years of Operation). Calculating the rate of IGSCC per weld-year for a given system and pipe size performs the normalization. Before performing this normalization the database is subjected to additional processing to exclude from further consideration any IGSCC data not directly applicable to the RR System that is representative of the Base Case. Similarly the part of the database including plant population and weld population data must be processed in such a way that an appropriate exposure term is developed commensurate with the failure data. In developing the RR-specific exposure term the following exclusion criteria were applied:

Plant Population Exclusion Criteria Applicable to RR Piping

• BWR plants without external recirculation loops;

• BWR plants in which the RR piping is fabricated from IGSCC resistant material e. g., Nuclear Grade stainless steel.

Table D.3 includes selected weld counts used to derive an exposure term according to Equation (D.2). Organized by pipe size and years of operation, Table D.4 is a summary of weld failures in RR piping. Noteworthy is the observation that there have been no reported through-wall defects in any plant beyond T = 15 years of operation. Using the information in Tables D.3 and D.4, Figure D.16 shows the calculated rate of RR pipe failure per weld-year.

The failure rates in Figure D.16 assume that all RR welds of a certain size to be equally susceptible to IGSCC. As was shown in SKI Report 98:30 [D.15], a correlation exists between weld failure rate and weld configuration. This correlation is assumed to be attributed to the piping layout, complexity of welding operation, and the associated weld residual stresses. The chart in Figure D.17 shows the weld configuration versus fraction of weld failures.

Figure D.16 Rate of IGSCC-Induced RR Pipe Failure (‘Prior State-of-Knowledge’)

Pipe-to-pump

Pipe-to-cross

Elbow-to-pump

Pipe-to-valve О

Elbow-to-valve

=5

Pipe-to-end-cap c

Pipe-to-tee ~o

Pipe-to-reducer

§

Pipe-to-sweepolet Pipe-to-safe-end Pipe-to-pipe Nozzle-to-safe-end Elbow-to-pipe

Figure D.17 RR Weld Failures as a Function Weld Configuration

In addition to weld configuration, the likelihood of failure is also a function of pipe size. For a weld of type “i” and size “j” the failure rate is expressed as follows:

Aj = Fj/(Wj x T) (D.3)

and with

S = Fj / Fj (D.4)

Aij = Wj/Wjj (D.5)

the failure rate of weld of type “i” and size “j” is expressed as

Aj = (F x Sij)/(Wj x T) (D.6)

Aj = Sij x AijX A (D.7)

Failure rate of an IGSCC-susceptible weld of type “i”, size “j”

Failure rate of an IGSCC susceptible weld of size ‘j’

Number of size “j” weld failures

Number of type “i” and size “j” weld failures

Size “j” weld count

Type “i” and size “j” weld count

The service experience shows the failure susceptibility to be correlated with the location of a weld relative to pipe fittings and other in-line components (flanges, pump casings, valve bodies). For a given pipe size and system, the susceptibility is expressed as the fraction of welds of type “ij” that failed due to a certain degradation mechanism). This fraction is established from the PIPExp database; Table D.5.

In the above expressions the attribute (A) is defined as the ratio of the total number of welds of size “j” to the number of welds of type “i”. In expression (4.7), Ay is a correction factor and accounts for the fact that piping system design & layout constraints impose limits on the number of welds of a certain type. For example, in a given system there tends to be more elbow-to-pipe welds than, say, pipe-to-tee welds.

Combining the global (or averaged) failure rates in Figure D.16 with the information summarized in Figure D.17 and Table D.5 provides the apriori failure rates that are input to Equation (D.1). The results are summarized in Section D.4.3.

D.4.2.2 FW Pipe Failure Data — The estimation of failure rates for FW piping uses a non-informative prior distribution together with the weld population data in Table D.6. This approach is selected based on the available, limited service experience with ASME XI Class 1 FW piping; Table D.7. In developing the data summary in Table D.7 the following FW exclusion criteria were used to develop a point estimate of the failure rate:

Failure Data Exclusion Criteria Applicable to FW Piping

• Piping external to the drywell containment structure;

• Non-US data.

Lee Abramson reviewed the elicitation questionnaire results from Tuesday afternoon’s session. Overall, the results were good and consistent with expectations. The group tended to perform better on those questions that asked for the ratio of diseases between men in different age ranges (questions 3 and 4 in exercise). The mid value tended to reasonably close to the actual 2000 census values for these questions.

Additionally, the true value was contained with the 50% interquartile region (75% — 25% percentiles) 10 out of 12 times, or 83%, which is quite good.

The average coverage interval for these questions was 71%. The coverage interval should theoretically be 90%, so that group underestimated the uncertainty. Lee Abramson indicated that typically group uncertainty is about Уг of the true value. In other words, people tend to be more confident in their responses than they should be.

The results were not quite as good when the group was asked to provide absolute disease rates for an age category (Question 2). The true value was inside the 50% interquartile range 4/6 times, or 67%. Also, the coverage interval only captured the true value 61% of the time, 10% less than for the relative questions. This performance demonstrates the supposition relative differences between conditions tend to be more accurate than absolute measures for a given condition. This will be a guiding principal in developing the elicitation framework.

Analyses tend to get easier as go up on LOCA sizes in that less systems to worry about.

As a group the panel was uncomfortable with comments on Effect of Operating Time for Category 4 LOCAs.

Bill Galyean suggested that for presentation to public, may want to consider some other means of reporting extremely low frequencies (~1e-15); Rob wasn’t sure he could define a cut off value; also when look at median values, these low numbers don’t impact final answer; Helmut supported this approach.

Helmut suggested that we explicitly state that a major assumption was that everything was fabricated in accordance with Code standards; no counterfeit bolts, etc.

Participant J only has fatigue in his analysis, once he includes other mechanisms, then his frequencies will go up by 2 or 3 orders of magnitude, but will still be low and probably will not alter the final median results.

The emergency core cooling system (ECCS) requirements are contained in 10 CFR 50.46, Appendix K to Part 50, and GDC 35. Specifically, ECCS design, reliability, and operating requirements exist to ensure that the system can successfully mitigate postulated loss-of-coolant accidents (LOCAs). Consideration of an instantaneous break with a flow rate equivalent to a double-ended guillotine break (DEGB) of the largest primary piping system in the plant generally provides the limiting condition in the required 10 CFR Part 50, Appendix K analysis. However, the DEGB is widely recognized as an extremely unlikely event. Therefore, the staff is establishing a risk-informed revision of the design-basis break size requirements for operating commercial nuclear power plants.

A central consideration in selecting a risk-informed design basis break size is an understanding of the LOCA frequency as a function of break size. The most recent NRC-sponsored study of pipe break failure frequencies is contained in NUREG/CR-5750. Unfortunately, these estimates are not sufficient for design basis break size selection because they do not address all current passive-system degradation concerns and they do not discriminate among breaks having effective diameters greater than 6 inch. There have been two approaches traditionally used to assess LOCA frequencies and their relationship to pipe size: (i) estimates based on statistical analysis of operating experience and (ii) probabilistic fracture mechanics (PFM) analysis of specific postulated failure mechanisms. Neither approach is particularly suited to evaluate LOCA event frequencies due to the rareness of these events and the modeling complexity. In this study, LOCA frequency estimates have been calculated using an expert elicitation process to consolidate operating experience and insights from PFM studies with knowledge of plant design, operation, and material performance. This process is well-recognized for quantifying phenomenological knowledge when data or modeling approaches are insufficient.

The principal objective of this study was to develop separate boiling water reactor (BWR) and pressurized water reactor (PWR) piping and non-piping passive system LOCA frequency estimates as a function of effective break size at three distinct time periods: current-day (25 years fleet average), end-of-plant — license (40 years fleet average), and end-of-plant-license-renewal (60 years fleet average). These estimates are based on the responses from an expert panel and one aim of this study was to obtain estimates that represent a type of group consensus. Additionally, another objective was to reflect both the uncertainty in each panelist’s estimates as well as the diversity among the individual estimates.

The elicitation focused on developing generic, or average, estimates for the commercial fleet and the uncertainty bounds on these generic estimates rather than bounding values associated with one or two plants. This approach is consistent with prior studies that did not consider plant-specific differences in developing LOCA frequencies for use in probabilistic risk assessment (PRA) modeling. Consequently, the elicitation panelists were instructed to consider broad differences among plants related to important variables (i. e., plant system, material, geometry, degradation mechanism, loading, mitigation/maintenance) in determining both the generic LOCA frequencies and especially the estimated uncertainty bounds. It is the broad differences in these important variables that contribute most to passive system failure and there is generally sufficient commonality among plants to make such a generic assessment valuable.

The elicitation was solely focused on determining LOCA frequencies that initiate by unisolable primary system side failures that can be exacerbated by material degradation with age. Therefore, active system failures (e. g., stuck open valve, pump seals, interfacing system LOCAs) and consequential primary pressure boundary failures due to either secondary side failures or failures of other plant structures (e. g., crane drops) were not considered. Active system frequency contributions should be combined with the passive system LOCA frequencies to estimate the total system risk. The effects of safety culture on primary side failure (i. e., LOCA) frequencies were also considered.

This study developed LOCA frequency estimates consistent with historical small break (SB), medium break (MB), and large break (LB) flow rate definitions. Additionally, three larger LOCA categories were defined within the classical LB LOCA regime to examine trends with increasing break size, up to and including, a DEGB of the largest piping system in the plant. Contrary to earlier studies, the six LOCA categories are defined in terms of cumulative thresholds rather than break intervals because this definition was more conducive to the elicitation structure. However, the differences between the interval and cumulative estimates are typically much smaller than the estimates’ uncertainties. It is therefore recommended that the cumulative threshold estimates from any set of tabular results be used (e. g., Table 1) if interval-defined LOCA frequencies are desired.

Because the LOCA frequency estimates were intended to be both generic and consistent with historical internal-event PRAs, the elicitation primarily considered normal plant operational cycles and loading histories. The loads include representative constant stresses (e. g. pressure, thermal, residual) and expected transient stresses (e. g. thermal striping, heat-up/cool-down, pressure transients) that occur over the license-renewal period. The elicitation implicitly considered all modes of operation based on the loading or operational experience associated with each piping system or non-piping component. Rare event loading from seismic, severe water hammer, and other sources was also not considered in this generic evaluation because of their strong dependency on plant-specific factors. However, a separate research study was conducted to assess the potential impact of seismic loading on the break frequency versus break size relationship. As part of that study, both unflawed and flawed seismic piping contributions were considered. The results of that seismic LOCA analysis are summarized in Section 7.2.

Several important assumptions were made to guide the elicitation process. One such assumption is that plant construction and operation comply with all applicable codes and standards required by the regulations and the technical specifications. The specific impact on the LOCA frequencies of purposefully violating these requirements was not considered. However, it was assumed that regulatory oversight policies and procedures will continue to be used to identify and mitigate risk associated with plants having deficient safety practices. While deviations from these requirements do represent some percentage of the events included in the passive-system failure data, extrapolation of this data implicitly assumes that similar future deviations will continue to occur with similar frequency.

Another important assumption is that current regulatory oversight practices will continue to evaluate aging management and mitigation strategies in order to reasonably assure that future plant operation and maintenance has equivalent or decreased risk. A related assumption inherent in this elicitation is that all future plant operating characteristics will be essentially consistent with past operating practice. The effects of operating profile changes were not considered because of the large uncertainty surrounding possible operational changes and the potentially wide-ranging ramifications of significant changes on the underlying LOCA frequencies.

The expert elicitation process employed in this study is an adaptation of the formal expert judgment processes used to evaluate reactor risk (NUREG-1150), to develop seismic hazard curves (NUREG/CR — 6372), and to assess the performance of radioactive waste repositories (NUREG/CR-5411). The process consisted of a number of steps. To begin, the project staff identified many of the issues to be evaluated through a pilot elicitation. The panel members for this pilot elicitation were all NRC staff. A group of twelve panel members was then selected for the formal elicitation. The staff gathered background material and prepared an initial formulation of the technical issues which was provided to the panel. At its initial meeting, the panel discussed the issues and, using the staff formulation as a starting point, developed a final formulation for the elicitation structure. This structure included the decomposition of the complex technical issues which impact LOCA frequencies into fundamental elements so that these important contributing factors could be more readily assessed. Piping and non-piping base cases were also defined for use in anchoring the quantitative elicitation responses. The base cases represent a set of well-defined conditions which could cause a LOCA. A subset of the panel was created to develop quantitative LOCA frequencies estimates associated with the base case conditions. At this initial meeting, the panel was also trained in subjective elicitation of numerical values through exercises and discussion of potential biases.

After this initial meeting, the staff prepared a draft elicitation questionnaire and iterated with the panel to develop the final questionnaire. The panelists quantifying the base case conditions also developed their initial estimates. A second meeting was then held with the entire panel to review the base case results, review the elicitation questions, and finalize the formulation of remaining technical issues. At their home institutions, the individual panel members performed analyses and computations to develop answers to the elicitation questionnaire.

The elicitation questionnaire required panelists to assess the following technical areas: the base case evaluation effort, utility and regulatory safety culture effects on LOCA frequencies, piping system LOCA frequencies, and non-piping system LOCA frequencies. The utility and safety culture questions required the panelists to compare future safety culture with the existing culture and predict the effect on LOCA frequencies. These effects were considered separately from passive system degradation because the panelist judged safety culture and degradation to be independent. The base case evaluation required the panelists to assess the accuracy and uncertainty in the base case analyses, and to also choose a particular base case approach for anchoring their elicitation responses. The piping and non-piping LOCA frequency questions required each panelist to first identify important LOCA contributing factors (i. e., piping systems, materials, degradation mechanisms, etc.) and select appropriate base case conditions for comparison. The panelists were then required to provide the relative ratios between their important contributing factors and the base case conditions based on their knowledge of passive system component failure. Each relative comparison required mid value, upper bound, and lower bound values. The mid value is defined such that, in the panelist’s judgment, there is a 50% chance that the unknown true answer lies above the mid value. The upper and lower bounds are defined such that there is a 5% chance that the true answer lies above the upper bound or below the lower bound, respectively. Each panelist was also required to provide their qualitative rationale supporting their quantitative values.

A facilitation team consisting of individuals knowledgeable about the technical issues (substantive members), a staff member with extensive experience conducting expert elicitations (normative member) and two recorders met separately with each panel member in day-long individual elicitation sessions. At these sessions, each panel member provided answers to the elicitation questionnaire along with their supporting technical rationales. The panelists were asked to self-select, based on their expertise, the questions that they addressed. Consequently, several panelists only provided responses for either BWR or PWR plants. After the elicitation sessions, the panel members returned to their home institutions where they refined their responses based on feedback obtain during their session. Upon receipt of the updated responses, the project staff compiled the panel’s responses and developed preliminary estimates of the LOCA frequencies. Along with the rationales, these preliminary estimates were presented to the panel at a wrap-up meeting. Panel members were invited to fill in gaps in their questionnaire responses and, if desired, to modify any of their responses based on group discussion of important technical issues considered during the individual elicitations1. Final individual estimates of the LOCA frequencies were then calculated and provided to the panel members for final review and quality assurance.

The qualitative insights provided by the panel are reasonably consistent. Panelists identified several advantages and disadvantages with determining the base case LOCA frequencies through operating — experience assessment or PFM analyses. However, there was a general consensus that operating experience provides the best basis for evaluating current-day, small-break LOCA estimates. Hence, many panelists used the operating-experience base cases to anchor their elicitation responses. The panel members also generally expressed the opinions that the future safety culture will not differ dramatically from the current culture, and that the utility and regulatory safety cultures are highly correlated. Many panelists do believe that safety culture can significantly affect LOCA frequencies at specific plants, but there is also an expectation that regulatory actions using existing enforcement measures will diminish both the possibility and impact of deficient safety culture at particular plants. However because it was thought that these plant-specific issues do not affect the generic averages, no specific adjustment to the LOCA frequency estimates was applied to explicitly account for safety culture effects. This decision was endorsed by the elicitation panelists.

There were several technical insights that were consistently identified. Many participants believe that the number of precursor events (e. g., cracks and leaks) is generally a good barometer of the LOCA susceptibility for the associated degradation mechanism. Welds are almost universally recognized as likely failure locations because they can have relatively high residual stress, are preferentially-attacked by many degradation mechanisms, and are most likely to have preexisting fabrication defects. Most panelists also agreed that a complete break of a smaller pipe, or non-piping component, is generally more likely than an equivalent size opening in a larger pipe, or component, because of the increased severity of fabrication or service cracking. Therefore, the biggest frequency contributors for each LOCA size tend to be systems having the smallest pipes, or component, which can lead to that size LOCA. The exception to this general rule is the BWR recirculation system, which is important at all LOCA sizes due to IGSCC. Many panelists thought that aging may have the greatest effect on intermediate diameter (i. e., 6 to 14-inch nominal diameter) piping systems due to the large number of components within this size range and the fact that this piping generally receives less attention than the larger diameter piping and is harder to replace than the more degradation-prone smaller diameter piping.

The participants generally identified thermal fatigue, stress corrosion cracking (SCC), flow accelerated corrosion (FAC), and mechanical fatigue as the degradation mechanisms that most significantly contribute to LOCA frequencies in BWR plants. Generally, the most important BWR degradation mechanism is intergranular SCC (IGSCC), although the panelist’s recognize that mitigation has greatly reduced the susceptibility of BWR plants to this mechanism over the past 20 years. With the exception of FAC, similar degradation mechanisms and concerns were also deemed to be important in PWR plants. Specifically, primary water SCC (PWSCC) is a principal concern. Many panelists believe that PWSCC will be mitigated in PWR plants over the next 15 years, but that effective mitigation has yet to be developed and implemented.

The panelists generally agreed on the important technical issues and LOCA-contributing factors.

However, the individual quantitative responses are much more uncertain and there are relatively large differences among the panelists’ responses. This is to be expected given the underlying scientific uncertainty. The analysis of these responses was structured to account for the uncertainty in the [1]

individual estimates and the diversity among the panelists. The quantitative responses were analyzed separately for each panel member to develop individual BWR and PWR total LOCA frequency estimates. A unified analysis format was developed to ensure consistency in processing the panelists’ inputs. The panelists’ mid-value, upper bound and lower bound responses were assumed to represent the median, 95th, and 5th percentiles, respectively, of their subjective uncertainty distributions for each elicitation response. The analysis structure was based on the assumption that all the responses correspond to percentiles of lognormal distributions. These distributions were then combined using a lognormal framework. The final outputs for each panelist are estimates for the means, medians and 5th and 95th percentiles of the total BWR and PWR LOCA frequencies. The panelists’ estimates were then aggregated to obtain group LOCA frequency estimates, along with measures of panel diversity.

The individual and group estimates for the means, medians, 5th and 95th percentiles of the LOCA frequency distributions were calculated using the following principal assumptions and choices:

(i) The mid-value, upper bound, and lower bound supplied by each panelist for each elicitation question were assumed to correspond to the median, 95th percentile and 5th percentile, respectively, of a split lognormal distribution, with the mean calculated assuming that the upper tail is truncated at the 99.9th percentile.

(ii) Only those panelists whose uncertainty ranges were relatively small were adjusted using an error-factor adjustment scheme to account for possible overconfidence (Section 7.6.2.2).

(iii) Split lognormal distributions were summed by assuming perfect rank correlation among the individual terms.

(iv) The individual estimates are the total LOCA frequency parameters (i. e., mean, median, 5 th percentile, and 95th percentile) determined for each panelist.

(v) The group estimates of the total LOCA frequency parameters were determined using the geometric means of the individual estimates.

(vi) Panel diversity was characterized with two-sided 95% confidence intervals based on an assumed lognormal model for the individual estimates.

In the report, these six assumptions and choices define what are termed the summary estimates. The report also calculates and discusses baseline estimates. These baseline estimates are calculated using all the above assumptions, except no overconfidence adjustment (as described in (ii) above) is applied. Because it is well-established that experts tend to be overconfident, the overconfidence adjustment is deemed to result in improved LOCA frequency estimates

The resultant individual and group summary estimates are consistent with the elicitation objectives and structure and are reasonably representative of the panelists’ quantitative judgments. In particular, they are not dominated by extreme results, either on the high or low end, and the geometric means of the individual estimates approximate the medians of these estimates. The median is often used to represent group opinion in elicitations, especially when the individual estimates differ by several orders of magnitude, as they do in this study.

The LOCA frequency summary estimates for the current-day (25 years) and end-of-plant-license (40 years) periods are provided in Table 1 for both BWR and PWR plant types. The aggregated group estimates for the median, mean, 5th and 95th percentiles are presented. Frequency estimates are not expected to change dramatically over the next 15 years for any size LOCA, or even over the next 35 years for BWR LOCA Categories 1- 5 and PWR LOCA Categories 1 — 2 (see results in Appendix L). While order of magnitude increases in BWR LOCA Category 6 and PWR LOCA Categories 3 — 6 over the next 35 years are expected, these increases are largely due to uncertainty about the future and the concern that new degradation mechanisms could arise in the operating fleet. However, while aging will continue, the panelists’ consensus is that mitigation procedures are in place, or will be implemented in a timely manner,

to alleviate significant increases in future LOCA frequencies for existing degradation mechanisms. Because of the predicted stability in these estimates over the near-term, the current-day (25 year) results can be used to represent the LOCA frequencies over the next 15 years of fleet operation.

The current-day median, mean, and 95th percentile estimates are graphically presented in Figure 1. The 95% confidence intervals calculated for these parameters are also illustrated in this figure. The LOCA frequencies as a function of threshold break diameter were estimated only for the six specified LOCA categories in the elicitation. The plotted points are connected with straight lines in the figure for visual clarity, but this should not be construed as a recommended interpolation scheme. Interpolation of frequencies between category sizes can be done at the user’s discretion depending on the conservatism required by the application. Some common interpolation schemes are linear, multi-point nonlinear and cubic spline. A step-wise or stair-step interpolation between two categories where the frequency for the lower category size is used for all flow rates or corresponding break sizes between the two categories provides the most conservative interpolation scheme. Note that any interpolation scheme does not reflect the uncertainty in the interpolated frequencies.

A measure of the individual uncertainties in Table 1 and Figure 1 is given by the differences between the medians and the corresponding 5th or 95th percentile estimates. Panel diversity is reflected in the confidence bounds in Figure 1. The large widths of these confidence bounds (as much as 3 orders of magnitude) reflects the significant diversity of the individual estimates in this study As the LOCA size increased, the panel members generally expressed greater uncertainty in their predictions, and the variability among individual panelists’ estimates increased. This is to be expected because of the increased extrapolation required from available passive-system failure data for larger LOCA sizes.

While it is acknowledged that operating experience-based estimates do not necessarily reflect the current state, it is informative to compare such estimates with the elicitation frequencies for the smallest LOCAs (Category 1). This comparison requires the least extrapolation of passive-system failure data since no larger LOCAs have occurred. The BWR and PWR Category 1 LOCA frequencies (including the steam generator tube rupture (SGTR) frequency for PWRs) were estimated up through December 2006. For BWR plants, the average SB LOCA frequency based solely on the number of reported events was estimated to be 5.5E-04 per calendar year. The mean elicitation BWR SB LOCA estimate is 6.5E-04 per calendar year. The BWR elicitation estimate is less than 20 percent higher than the operating-experience estimate which, given the uncertainty of these estimates, is not statistically significant.

For PWR plants, the average SB LOCA frequency estimated analogously is 3.6E-03 per calendar year. The mean elicitation PWR SB LOCA estimate is 7.3E-03 per calendar year. The PWR elicitation estimate is about 100 percent (or a factor of two) higher than the operating-experience-based estimate. Additional insight into this difference can be gained by partitioning the PWR passive-system failure data into frequencies for SGTRs and for all other passive-system SB LOCAs.

Based on reported failures, the mean SGTR LOCA frequency is 3.2E-03 per calendar year. This result is almost identical to the current-day elicitation estimate of 3.7E-03 per calendar year. The frequency of all other Category 1 PWR passive-system failures is 4.0E-04 per calendar year based on operating experience. The corresponding elicitation estimate is 1.9E-03 per calendar year. While this value is 5 times greater than the operating-experience-based estimate, this difference is explained by the elicitation panelists’ estimation of the effect of PWSCC on small diameter component failures.

There are several prior studies that also estimated LOCA frequencies. Some care is needed when comparing the elicitation LOCA frequency estimates with these earlier studies because the LOCA categories are defined differently. Specifically, the current-day LOCA Category 1 and 2 estimates in Table 1 are comparable to total system SB, MB, and LB LOCA frequencies, respectively, reported in

NUREG/CR-5750. Additionally, current-day SGTR frequencies (Table 7.18) and PWR LOCA frequencies for all other passive system failures (i. e., frequencies for breaks greater than 100 gpm (380 lpm) in Table 7.19) are comparable to NUREG/CR-5750 SGTR and PWR SB LOCA frequencies when these failure modes are analyzed separately. The NUREG/CR-5750 LB LOCA frequency estimates are best compared to the elicitation LOCA Category 4 frequency estimates because the pipe break sizes are most similar.

After accounting for these differences, the elicitation LOCA frequency estimates are generally much lower than the WASH-1400 estimates and more consistent with the NUREG/CR-5750 estimates. The SB LOCA PWR elicitation estimates after subtracting the SGTR frequencies are approximately 3 times greater than the NUREG/CR-5750 estimates, due to the aforementioned PWSCC concerns. However, the total BWR and PWR SB LOCA frequency estimates are similar once the SGTR frequencies are added to the NUREG/CR-5750 PWR results. The elicitation MB LOCA estimates are higher than the NUREG/CR-5750 estimates by factors of approximately 4 and 20 for BWR and PWR plant types, respectively. These increases are partly due to concerns about PWSCC of piping and non-piping (e. g., CRDM) components as well as general aging concerns with piping in this size range. The NUREG/CR — 5750 LB LOCA frequency estimates are slightly higher (less than a factor of 3) than the current elicitation results for both PWR and BWR plants. The generally good agreement between the NUREG/CR-5750 and current elicitation estimates is somewhat surprising given the markedly different methodologies used to arrive at these results.

Sensitivity analyses were also conducted to examine the robustness of the quantitative results to the analysis procedure used to develop the summary estimates. These sensitivity analyses investigated the effect of distribution shape on the means as well as the effects of correlation structure, panelist overconfidence, panel diversity measure, and aggregation method on the estimated parameters. The mean calculation in the analysis procedure used a split lognormal distribution truncated at the 99.9th percentile to obtain reasonably conservative values compared with other possible choices. The correlation structure in the analysis procedure assumed maximal correlation, which is reasonably representative of the elicitation structure. The structure provides conservative 95th percentile estimates. However, based on selected Monte Carlo simulations, an independent correlation structure leads to larger median and 5th percentile estimates. The means are unaffected by the choice of the correlation structure. The analysis procedure also used confidence intervals for the aggregated estimates to measure panel diversity. An alternative approach used quartiles of the individual estimates, leading to comparable, but narrower intervals.

The analysis procedure also adjusted those panelists’ responses that had relatively narrow uncertainty bands using an error-factor scheme to account for a known tendency for people, including experts, to be overconfident when making subjective judgments. Sensitivity analyses examined the effects of other overconfidence adjustments of the nominal subjective confidence levels supplied by the panelists. No overconfidence adjustment was also investigated. While blanket overconfidence adjustments can result in large, unsupportable increases in the mean and 95th percentile frequency estimates, no adjustment results in modest decreases in these estimates. Therefore, the error-factor scheme, which adjusts only those panelists who are most overconfident, is deemed to the most appropriate.

Finally, the largest sensitivity is associated with the method used to aggregate the individual panelist estimates to obtain group estimates. The baseline and summary estimates were developed using geometric-mean aggregation. In this study, the geometric-mean aggregation produces frequency estimates that approximate the median of the panelists’ estimates and therefore effectively leads to consensus-type results. Therefore, the summary estimates in Table 1 are believed to be a reasonable representation of the expert panel’s current state of knowledge regarding LOCA frequencies.

The sensitivity analyses evaluated the effect of using alternative aggregation methods to calculate the group estimates. Specifically, mixture-distribution and arithmetic-mean aggregation were evaluated. For the panelists’ responses in this study, these alternative aggregation methods can lead to significantly higher mean and 95th percentile estimates than those obtained using geometric-mean aggregation. Alternative LOCA frequency estimates that are higher than the summary estimates (Table 1) can be derived by using either the summary estimates with 95% confidence bounds (Tables 7.8), the arithmetic — mean aggregated results (Table 7.13), or the mixture-distribution results (Table 7.16). These estimates also incorporate the same overconfidence adjustment as the summary estimates.

Because alternative aggregation methods can lead to significantly different results, a particular set of LOCA frequency estimates is not generically recommended for all risk-informed applications. The purposes and context of the application must be considered when determining the appropriateness of any set of elicitation results. In particular, during the selection of the BWR and PWR transition break sizes for the proposed 10CFR50.46a rule making, the NRC staff considered the totality of the results from the sensitivity studies rather than only the summary estimates from this study The NRC anticipates that a similar approach will be used in selecting appropriate replacement frequencies for NUREG/CR-5750 estimates and for other applications where frequencies for break sizes other than those in NUREG/CR — 5750 are required. While the lack of clear application guidance places an additional burden on the users of the study results, those users are in the best position to judge which study results are most appropriate to consider for their particular applications.

BWR: Error Factor Adjustment Threshold Break Diameter (in)

PWR: Error Factor Adjustment

Threshold Break Diameter (in) Figure 1 BWR and PWR Error-Factor Adjusted LOCA Frequency Estimates

Bengt, like Bill Galyean, used passive-system failure data in his analysis, but in a much different manner. Whereas Bill Galyean followed a “Top Down” approach, Bengt followed a “Bottoms Up” approach. Bengt’s presentation assumed that all BWR welds were category D & E welds. For BWRs, Bengt indicated that Category D & E welds specify inspection criteria based on Generic Letter 88-01. Category D welds have been subjected to weld overlay repairs and Category E welds are subject to IGSCC.

Bengt’s used a different database than Bill Galyean did is his analysis. Bengt’s database is proprietary (PIPEex) and the panel will not have access to this during the elicitation. The SLAP database that Bill used is available to the panel members on the website. The cut-off date for PIPEex events is the end of 2002 while the cut-off date for the SLAP database is the end of 1998. PIPEex has about twice the number of data entries as does SLAP and includes international experience.

Bengt only looked at welds in his analysis. He did not consider base metals. His database did not show any occurrences of base metal indications. However, he invoked a wide definition for what was encompassed by the term “weld”. He included the heat-affected-zone (HAZ) and counter bore region into his definition of what a weld was. Bengt also did not consider degradation of non-piping passive components in his analysis.

For PWR systems, he assumed that the V. C. Summer and Ringhals cracks were circumferentially oriented cracks, not axially oriented. Whereas Bill did a “top down” analysis, Bengt did a “Bottoms up” analysis. The failure rates were derived for individual welds, and then an integration system level model was formed by combining the contributions from each individual weld failure to an overall pipe system failure frequency.

The term “prior” has very specific meaning in this analysis. It means “before mitigation/remedial action in response to a significant pipe failure”. Hence, failure rates that are input to LOCA frequency calculations explicitly account for reliability improvements (mitigation methods) made in response to past pipe degradation histories. Failure in Bengt’s analysis is defined as a “through-wall flaw resulting in leakage”. Bengt did not include surface cracks found during ISI in his data reduction process.

Karen Gott has a report in Swedish that may be valuable in this effort. This report gives the number of leaks and the number of ISI detected surface cracks. Gery Wilkowski thought that was important information since leaking through-wall cracks are readily detected, but the surface cracks that would grow to be long in length are more of a LOCA threat. The number of records shown on slide 10 from Bengt’s presentation includes both leaks and cracks.

There was much discussion among the group in an effort to understand slides 11 through 14. In slide 13, Bengt did not eliminate welds if mitigation was performed prior to 15 years of operation in developing his prior distribution. There were no leaks in BWRs after 15 years. From years 10 to 15, it is possible that there may have been some plants that used mitigation, but those mitigated plant weld numbers were still included in the weld failure rate analysis, i. e., that may account for why the weld failure rate was not accelerating as the number of years increase. Bengt used a Monte Carlo simulation to create this plot, where he needed to estimate the number of welds for the number of plants that were at a certain age. Results in slide 14 include results from US, Spanish, Swedish, and Japanese plants. The results were adjusted by the number of welds and type of welds.

These preliminary results are used to determine the “prior” LOCA frequencies (weld failure rate). The next step is to determine the “posterior” frequencies based on “prior” distributions. Bengt accounts for uncertainty in the knowledge base, which is fundamental difference between his analysis and Bill’s.

Dave Harris thought the “posterior” frequencies should be equal to the Prior LOCA frequencies times the Likelihood Function. Lee Abramson explained that the likelihood function was built in.

It is noted in slide 19 that the Bayesian update strategies are different for each base case. It was also noted again that the weld failures represent leaks only and not cracks. Weld failure was defined as a through-wall flaw with leakage less than or equal to the tech spec limit for undefined leakage.

Gery Wilkowski noted that in January 2003, PWSCC was found in a surge line bimetallic weld at the pressurizer in a Belgium plant. It was noted that some transient event is needed to cause a tech spec limit flaw to propagate to a higher category (well beyond 100 gpm [380 lpm]) LOCA. Slide 23 shows that the conditional probability of failure (PL/F) is a function of the nominal pipe diameter (DN), i. e., PL/F = a x DNb, much like what is in NUREG/CR-5750 (Beliczey and Schulz, i. e., PR/TWC = 2.5/DN). While the exact formulations are different, the conditional failure probability in all cases is an inverse function of pipe size.

The form of the conditional leak probability given a failure is inconsistent with the original development of the Beliczey and Schulz correlation which was developed to relate leaks to breaks as a function of pipe size. This use (see slide 24) may not be physically realistic because it assumes that larger diameter pipes are less likely to result in a category 0 leak. The implication is that larger diameter pipes are less likely to reach a Category 0 leak, and then progress to higher leak rates. Slide 26 presents information on the aspect ratios of IGSCC cracks. It was asked how the deep, full circumference cracks (a/t = 0.5, 2/B = 1.0) formed. The expectation is that these are likely crevice cracks. It was suggested that it would be nice to break down this data by pipe size in order to assess the relevance.

Bengt assumes that a through-wall crack can only propagate into a large leak if there is a large transient event. Slide 27 documents the loading categories assumed to drive the crack among various LOCA categories. Category 0 to Category 1 LOCA progression can occur assuming moderate loading, while to go from a Category 0 or Category 1 LOCA to a Category 6 LOCA, would require an extreme loading transient. The general consensus of the panel members was that this was a very subjective approach. It was noted that there were about 400 water hammer events reported, but Bruce Bishop and Guy DeBoo said that if there were this many water hammers, then the plant piping system was probably redesigned.

The extrapolation of results from the “Current Estimate” to 40 and 60 years is based on posterior analysis of prior results assuming no additional failures. If one assumes no additional failures, the failure rates will go down with time. Bengt will examine possibly extrapolating his base case results out to 60 years using another assumption.

The system descriptions in this section are extracted from design information supplied by members of the Expert Elicitation Panel. The PWR-specific system information is included in the following documents and drawings:

• Document No. EPRI-156-330: Degradation Mechanism Evaluation for Class 1 Piping Welds at Plant A. a [D.6]. This document summarizes the degradation mechanisms applicable to a Westinghouse 3- loop PWR.

• References [D.7-D.9] include design information as well as degradation mechanism information applicable to the HPI/NMU system of Plant A. b.

• Excel file entitled “PlantAWelds.” This Excel file includes weld lists for the RC system of Plant A. a. The lists are organized by weld identification numbers (as they appear on the isometric drawings identified below), nominal pipe size and pipe schedule. This Excel-file forms one of the bases for the PWR LOCA frequency model used to derive the LOCA frequency distributions.

• Isometric drawing numbers 1MS-22-2262 and CGE-1-4100A (RC Hot Leg), C-314-601 and CGE-1- 4500A (pressurizer surge line), and 17-MU-23 (HPI/NMU piping).

D. 1.4.1 Reactor Coolant (RC) System (Plant A. a) — The RC System evaluated in this study consists of three similar heat transfer loops connected in parallel to the RPV. Each loop contains a reactor coolant pump (RCP), steam generator, and associated piping and valves. In addition, the system includes a pressurizer, a pressurizer relief tank, interconnecting piping, and instrumentation necessary for operational control. The analysis in this report is concerned with a portion of one of the three RC loops; the portion from the RCP to the RPV (this is one of the hot legs). The pressurizer surge line connects the pressurizer to the RC cold leg Loop A. In summary, the piping sections that are subject to evaluation in this study consist of.

RC-HL: The analysis is concerned with 1-of-3 hot legs. The Loop A HL starts at the RPV, includes

an RCP and connects to the ‘A’ steam generator (S/G). The HL piping is fabricated from stainless steel piping. The section of the HL from the RPV to the RCP is of 31 inch inside diameter, while the section from the RCP to S/G is of 27.5 inch inside diameter piping.

Surge Line: The single surge line is fabricated form NPS14 stainless steel piping and connects to the 29-

inch RC cold leg.

D. 1.4.2 High Pressure Injection (HPI)/Normal Makeup (NMU) Line (Plant A. b) — In Plant A. b, each of the four RCS cold legs is equipped with high-pressure injection piping. Two of these 2 Уг inch (ID, or approximately NPS3-%) stainless steel piping lines provide the normal makeup flow to the RCS and they connect to the cold leg via nozzle assemblies. Each of the nozzles is comprised of a base nozzle and a safe — end. To prevent thermal cycling of the base metal each nozzle is equipped with a 1.5-inch thermal sleeve. The analysis is concerned with one of the two HPI/NMU lines.

Mr. Schulz has over 35 years of experience in nuclear engineering, structural and fracture mechanics, materials, and nuclear safety. At GRS he is Head of the Department of Components Integrity and was/is a member of various national and international advisory bodies regarding nuclear safety, component integrity, and codes and standards. In this role, he is responsible for the safety assessment of nuclear components and structures, as well as related research and verification of fracture mechanics codes for safety applications.

Prior to joining GRS, he worked for Gesellschaft fur Reaktorsicherheit where he held various staff positions and project management responsibilities for PWR safety assessment work. Prior to that he was with United Nuclear Corporation where he did work with fuel element design and inspection of reference elements in US Nuclear Power Plants. He was also on staff with AEG Research Center for which he worked in the area of fuel element design, qualification of fabrication processes, testing programs on fuel elements and inspection of reference elements in nuclear pilot plants. Mr. Schulz holds a B. S. /M. S. in mechanical and nuclear engineering and has served on the engineering faculty at Essen.

Bengt used a bottoms up evaluation based on operating experience.

Markov is standard approach common to any advanced reliability approach; this was the technical basis used by Bengt to develop time dependency.

Bengt’s model allows for imperfect repairs or inspections.

Can go from S (unflawed condition) to R (rupture condition) if have some extraordinary event such as gas accumulation at Hamaoka in Japan.

VG13 results are per “weld year”; some of earlier plots are “per reactor year”.

As summarized in Section D.3.3, there have been only a few through-wall defects in Class 1 PWR piping. For the RC-HL, the rate of PWSCC per weld-year is established using the normalization process discussed in Section D.4.2.1 and with the following specializations:

RC-HL Apriori Failure Rate

• The apriori failure rate is derived using the PIPExp for the time-period 1970 through 2000 to include the consideration of the through-wall defect at V. C. Summer.

• Table D.8 includes the weld population data used to calculate ^NPS3o = 8.12E-05 per weld-year.

• Failure rate “post-processing” to account for different weld configuration susceptibilities to PWSCC is done consistent with Section D.4.2 and with the S and Aj assumed values shown in Table D.9.

In contrast to the BWR weld susceptibility factors in Table D.5, the weld susceptibility factors in Table D.10 are assumed values that reflect the applicable service experience. As an example, for the RC Hot Leg the nozzle-to-safe-end weld is assigned the highest value in view of the available service experience; i. e., the Ringhals and V. C. Summer hot leg cracking as described in Section D.3.3.1. As another example, for the RC Surge Line, relatively high weld susceptibility factors are assigned the safe-end welds and Hot Leg branch connection. In view of the recent experience at TMI-1, the nozzle-to-safe-end weld is given a greater weight than other weld configurations. The uncertainty in the PWR weld susceptibility factors is not evaluated

further in this study, however. EPRI TR-111880[10] is used for characterizing the prior knowledge about pipe failure due to thermal fatigue.