DESCRIPTION OF NON-PIPING DATABASE

H. 1 Background

The non-piping database has been compiled with the intention that it will serve as one source of information supporting the development of estimates of LOCA frequencies attributable to non-pipe components. The data has been obtained from two primary sources. First, a search of LERs was made to identify those instances where failures[17] of non-pipe components of the primary reactor coolant pressure boundary were reported to the NRC. The second source of information is data that has been incidentally collected on non-pipe components during the development and maintenance of the pipe-based OECD and SLAP databases. LER events compose the majority of the events in the database (see Attachment A of this appendix for a description of the LER reporting requirements).

The database is accessible in two formats, Table and Forms. The Table named “Failure Data” lists the data in a spreadsheet type of format where each line of the table contains one data record and each column contains the various fields that make-up the records. In the Forms format, only one record of the “Failure Data” is displayed at a time, but in a manner that allows all of the fields to be view at the same time. Both formats display the same data, only the presentation is different. Also, sorting and filtering of the data can be done in both views.

H. 2 Approach

A search of LERs was performed (see Attachment B for the specific search criteria) using the SCSS. This search returned 1,036 LERs. Each LER was reviewed and coded in the Non-Pipe database. The database structure is based on information generated during the elicitation meetings. In particular, the component, piece part, and degradation mechanism are all identified using the tables documented in the elicitation meeting notes. Other fields of the database were developed and defined as judged appropriate.

The initial screening of the 1,036 LERs to remove those that were judged to not be applicable reduced the total number to 213. As discussed in Attachment B, the data search simply looked for leak and crack events associated with primary coolant systems. This conservative search included LERs that identified potential and possible leak and crack events (e. g., a problem with ECCS such that the plant would not respond as designed to a loss of coolant accident. Screening out the potential failures resulted in a reduction to 213 records. A further 34 records were removed since they identified problems with pipes or seals. Then 37 records were added that had been collected previously during the development of the OECD and SLAP databases. This results in a current total of 216 records.

Submitted by Nuclear Energy Institute (NEI)

Comment: The Nuclear Energy Institute offers the following comments on the subject Federal Register notice, which solicited public comments on draft NUREG-1829. This NUREG was intended to provide technical support for the proposed rulemaking to 10 CFR 50.46 which would establish the option to revise the design basis LOCA break size. Thus, the emphasis of the expert elicitation was on estimating frequencies for large break LOCAs. Our comments are limited to the report’s estimation of small break LOCA frequencies, which, unlike large break LOCAs, are important contributors to PRA risk profiles.

PRA standards have been developed by consensus bodies and endorsed by NRC, with the expectation that plants will be expected to conform to these standards to support regulatory applications. ASME PRA standard RA-S-2002 (endorsed through NRG Regulatory Guide 1.200) contains the following requirement relative to initiating event frequency estimation:

IE-C1: Calculate the initiating event frequency from plant-specific data, if sufficient data are available. Otherwise, use generic data. Use the most recent applicable data to quantify the initiating event frequencies

Response: Contrary to the commenter’s assertion, the objective of the expert elicitation (Section 2) was to calculate LOCA frequencies for all break sizes, not just large break LOCAs. Also, in general, the objectives and results of the expert elicitation as summarized in NUREG-1829 are consistent with the above statement in ASME PRA standard RA-S-2002. The NUREG-1829 small break LOCA estimates represent generic values. Calculations based on plant-specific data, specifically for PRA, are acceptable for use provided that a sufficient technical basis supporting these alternative estimates has been established. There is also an implication by this comment (and subsequent Comment GC5) that the NUREG-1829 estimates do not compare favorably with operator experience. As documented in the newly-added Section 7.10, the differences between the NUREG-1829 estimates and operating experience are not statistically significant. Those differences that do exist are also supported by the panelists’ qualitative and quantitative responses. More details on the comparison between operating experience and the NUREG-1829 LOCA frequency estimates are provided in Section 7.10 which was added to address this and similar comments. See also the responses to GC3, GC5, GC6, GC7, and 7-8 for similar discussions.

Comment Number: GC5

Submitted by Nuclear Energy Institute (NEI)

Comment: Draft NUREG-1829 used plant experiences to estimate the steam generator tube rupture (SGTR) frequency which amounts to greater-than 50% of the total small LOCA frequency. Estimate of the remaining 50% of Category 1 LOCA was entirely based on expert elicitation. The resulting Category 1 frequency estimates from the panel showed a significant divergence of opinions. It is recommend that Category 1 LOCA frequency estimate should continue to be related to the large number of years of plant experiences similar to the method used in NUREG-5750. The current lengths of those experiences amount to thousands of reactor-years, and are statistically significant for use in estimating the annual frequency of events at the 1E-2, 1E-3, and 1E-4 levels. Similar estimates are used in PRA models for numerous other important PRA parameters (such as SGTR).

Response: The commenter incorrectly states that plant experiences were used to estimate the SGTR frequency while the remaining contributions were based on expert elicitation. All contributions to the Category 1 LOCA frequency estimates were determined by the expert elicitation process. For the reasons documented in Section 1 of NUREG-1829, the authors believe that the expert elicitation results are more accurate than results calculated simply from operational experience.

However, the SGTR contributions provided by the panelists are nearly identical to estimates determined simply from operational experience. This consistency, however, simply reflects the panelists’ combined opinion that the operational experience data is relevant for calculating the current day SGTR contributions. The contributions of non-SGTR failures to the PWR LOCA Category 1 frequencies and the BWR LOCA Category 1 estimates are also not inconsistent with estimates based solely on operational experience (See Section 7.10). However, the panelists do expect some increase in the PWR LOCA Category 1 frequencies compared to operational experience-based estimates as a result of current PWSCC issues (Sections 6.3.2 and 7). See Section 7.10, which was added to address this and similar comments, for additional details on the comparison between the elicitation estimates and operational experience data. Additional relevant information is also provided in the responses to GC3, GC4, GC6, GC7, and 7-8.

E. 1 Summary

In this base case study, LOCA frequencies are calculated using a “top-down” approach. Specifically, a total LOCA frequency is calculated using U. S. commercial nuclear power plant (NPP) operating experience. This total frequency is then allocated to the LOCA size categories, RCS subsystems and components, and degradation mechanisms. This allocation is performed using data on primary system leaks and cracks from both U. S. and foreign PWR and BWR reactors.

E. 2 Assumptions and Observations

As with all analyses, there are a number of implicit assumptions associated with this approach. First is that past performance is representative of future performance. The common scenario for the occurrence of a LOCA starts with postulating the existence of a flaw or defect in the primary reactor coolant boundary. This flaw is then subjected to a stress that results in the catastrophic failure of the primary pressure boundary, producing a LOCA. The U. S. LWR operating experience to date consists of approximately 100 reactors with an average age of about 23 years. During this time the RCS of these plants have experience numerous transients and loads, which have produce a wide range of stresses. Whether these plants operate for 40 years (or 60 years with license extensions) this available operating experience represents a significant portion of the average plants lifetime. It is therefore reasonable to assume that the stresses that have already occurred are representative of those that will occur in the future. Similarly, various degradation mechanisms have affected RCS pipe, welds and components. However, when these degradation mechanisms have been detected, mitigation programs have subsequently been implemented (e. g., IGSCC in BWRs). Therefore, the number of flaws and defects in the RCS is likely to be cyclic over time. As the degradation mechanism manifests itself, the number of defects grows, as the degradation mechanism is addressed and mitigated, the number of defects is reduced. Again, the assumption here is that current 23 years of operating (on average, per reactor) are representative of the remaining operating life.

Another observation is the occurrence of zero LOCAs for both PWRs and BWRs. Although this does not prove that the LOCA frequencies are the same for both designs, it likewise does not support different LOCA frequencies. Therefore, for this analysis, the operating experience data (i. e., zero failures) will be pooled to generate a single LOCA frequency.

Furthermore, this analysis, just as every LOCA frequency estimate performed to date, assumes that the frequency of a LOCA decreases as pipe size increases. This might be attributable to a couple of issues. First, for small diameter pipe, some failure mechanisms exist that don’t apply to larger diameter pipe (e. g., compression fitting failures and socket welds). Second, the same flaw in both a small diameter pipe and a large diameter pipe represents a large percentage of the pipe diameter in the small diameter pipe. Third, inspection is probably more thorough in larger diameter pipe so that the chance of a defect going undetected is less in the larger diameter pipe. For all of these reasons (and probably others), the total LOCA frequency is reduced as LOCA size category increases. The scaling factor of Уг order of magnitude (assuming a lognormal probability distribution on LOCA frequency) appears to be reasonably consistent with historical LOCA frequency estimates.

This assumption of a half-order of magnitude (i. e., approximately a factor of 3) decrease in frequency for each increase in LOCA size is an assumption based on the general practice employed in estimating LOCA frequencies over the past 30 years starting with the Reactor Safety Study (Ref. E.1). This assumption is further supported by work done by Beliczey and Schulz (Ref. E.2). In this study, a combination of operating experience and fracture mechanics is used to demonstrate that the conditional probability of a rupture, given a leak, decreases as pipe diameter increases. This conclusion is reached because the size of detectable cracks and leaks remains relatively constant as a function of pipe size. Therefore, the relative crack or leak size as a function of the pipe circumference decreases, and the safety margin increases, as the pipe diameter increases.

Additionally, Beliczey and Schulz developed a quantitative conditional failure probability — which decreases by approximately У2 order of magnitude for each successively larger LOCA size — that was based on the propensity of through-wall fatigue flaws to lead to successively larger LOCA sizes.

Although the quantitative conditional failure probability is not applicable to all failure mechanisms and systems, this simple relationship has been extensively employed. This assumption was also employed in these analyses.

The final premise of this base case analysis is that the relative frequency of precursor data (i. e., leaks and cracks) is an indicator of the relative frequency of LOCA events. In the calculations that follow, the total LOCA frequency is allocated to the different RCS subsystems and components, and the different degradation mechanisms according to the relative frequency of observed leaks and cracks attributable to these subsystems and mechanism. Note that in order to determine the relative frequencies, complete crack and leak data are not needed, only consistent data that has not been biased by the over reporting of one attribute relative to another. Completeness in the data is neither required nor important, only consistency.

ELICITATION QUESTIONS

J.1 Instructions

There are four basic quantities that are the ultimate focus of this exercise: the LOCA frequencies of piping components, the LOCA frequencies of non-piping components, the LOCA probabilities of piping components after emergency faulted loading, and the LOCA probabilities of non-piping components after emergency faulted loading. The elicitation will be structured so that each of these questions can be answered using one of two question sets. The question sets are structured to decompose the underlying issues using different approaches so increase your flexibility.

The bottom-up approaches (3A, 4A, 5A, 6A) could entail significantly more work if every piping and non-piping system is evaluated. It is recommended that people choosing this approach focus on significant contributing issues in only significant piping and non-piping systems to reduce the workload. Similarly the people choosing the top-down approaches (3B, 4B, 5B, 6B) may want to ensure that their significant issues are manifested correctly within relevant systems. These strategies allow you to combine features of each methodology.

Only a few additional examples of these questions are provided in this document. Many examples will be similar to those included in the Elicitation Question Development document. Please refer to this document and the Kick-off Meeting Notes document as indicated within the notes section for the questions.

J.1.1 Specific Instructions: Minimum Requirements Prior to Your Elicitation

A1. Provide answers to the questions in the “Base Case Evaluation” area.

A2. Provide MV estimates for the question set within the “Regulatory and Utility Safety Culture” area.

A3. Provide MV estimates for at least one question set within the “LOCA frequencies of Piping Components” area.

A4. Provide MV estimates for at least one question set within the “LOCA frequencies of Non-Piping Components” area.

A5. Provide MV estimates for at least one question set within the “LOCA Probabilities of Piping Components under an Emergency Faulted Load” area.

A6. Provide MV estimates for at least one question set within the “LOCA Probabilities of Non-Piping Components under an Emergency Faulted Load” area.

A7. Categorize the uncertainty ranges (90% coverage intervals) associated with your MV estimates in A2 — A6 as low, medium, or high.

J.1.2 Specific Instructions: Additional Questions During Your Elicitation

We will be asking for your response within the following general areas.

B1. Provide rationale and discuss those important issues that you identified and quantified in questions A2 — A6.

B2. Quantify the uncertainty ranges (90% coverage intervals) associated with estimates provided for A2 — A6. This will quantify the initial responses in A7.

B3. Provide MV estimates for the question sets that you did not initially answer in A2 — A6.

B4. Quantify uncertainties associated with answers in B3.

B5. Ensure that the critical issues for LOCA frequencies are captured.

The earlier sections of this document contain many sets of results for each base case component. The multiple cases were generated primarily as a series of sensitivity studies. For these results are to be useful in the LOCA elicitation, a reference case must be selected for each component as being representative for that component. This section briefly discusses which case for each component is suggested as the reference case, and system leak frequencies are presented for each reference case.

The joint frequency is calculated from the cumulative results reported above by use of Equation F.2. The system frequencies are then obtained by multiplying by the number of highly stressed joints in the system (this approximation works because the failure probabilities are generally small).

Each component is discussed, with a summary table provided after all components are discussed.

F.4.1 Hot Leg Pressure Vessel

As shown in Tables F.5 — F.7, the large leak (> 380 lpm [100 gpm] and larger) probabilities for this component varied considerably, depending on the crack growth mechanism (cycle-dependent fatigue or time-dependent stress corrosion cracking [PWSCC]), and whether crack initiation or growth from pre­existing defects was considered. The fatigue crack growth results (Table F.5) were very low (~10-18), and the PWSCC crack initiation results (Table F.6) were quite large (~10-5). Since it is expected that this component will totally dominate the very large (> 380,000 lpm [100,000 gpm]) leak category, the selection of the reference case is critical for very large leak estimates. The PWSCC with fabrication defects has intermediate failure probability results (~10-10), and is recommended as the reference case.

The case without residual stresses is selected. Table F.6 shows that residual stresses do not have a large influence. The time dependency of the large leak cumulative probability is very small, which suggests that the leak frequency is very small. For estimation purposes, the leak frequencies are estimated by taking the value of the cumulative at 60 years, dividing it by 60, and assuming the value to be applicable independently of time. This will overestimate the leak frequency at long time and underestimate it at short time.

For extension to system failure frequency, it is assumed that there are three comparably stressed joints in the large main coolant piping.

The general approach and philosophy used follows the approached taken by GRS to estimate frequency of LOCA initiating events at passive systems for German PSAs.

The major steps and assumptions of this approach are as follows:

— In principle, wall penetration of pipes which would result into a leak follows in their geometries either

• a slit type penetration originating from cracks caused by fatigue or corrosion or

• a bulging type penetration caused by wall thinning.

Beyond critical dimensions wall penetrating stable defects turn into a full break. This means in practice that for each pipe size there are two or more leak sizes which are of a distinct different probability of occurrence governed by the likelihood of the respective active failure mechanism and the reliability to detect leaks and to take actions to avoid aggravation of the situation e. g. isolation of the leak, stop operation.

The maximum leak size related to a wall penetrating stable defect (undercritical crack, bulge, pit) depends on the actual load specifically the relationship between membrane and bending stresses. The majority of systems being considered in the safety analysis of NPP’s fall either into the category of high pressure or low pressure systems. For reasons of simplicity UB values can be taken to describe maximum leak sizes connected to wall penetrating stable defects for each pipe size. Based on experimental evidence as well as fracture mechanics calculations the maximum leak size resulting from an undercritical crack is rather limited, expressed in terms of fractions of the pipe cross section it is only a few percent. This approach uses 2% of the cross section as a rule of the thumb for high pressure systems. Through wall corrosion pits are generally very small. Bulge-type wall penetrations caused by wall thinning have a potential for stable leaks of a considerable size.

— The frequency of leaks is estimated based on the operating experience of the national population of nuclear power plants and in addition the worldwide experience is considered as far as applicable and available. In general, the operating experience give indications of leaks in a sense of precursors for most classes of piping or give indications of zero failures statistics only.

— To estimate the probability of a break (which is connected by the diameter of the piping to the flow rate) a relationship is used to describe the frequency of breaks in relation to the frequency of leaks as the function of the diameter of piping systems being designed to the same design parameters. For the small bore piping (less than 2 inch) the relationship between leak and break is arrived from operating experience. For the largest pipe (main primary pipe) the relationship between leak and break is based on a number of technical arguments and PFM analyses. For the pipe sizes in between a linear relationship is used between the UB and LB as described before.

— For reasons of simplicity and in accordance with technical experience it is assumed that within the piping systems only so called leak relevant elements are contributing to the frequencies. These leak relevant elements are essentially welds which are adjacent to changes in geometry (nozzles, branches, reducers etc.) which in itself would introduce enhanced loads and to some extend represent more difficult areas for manufacturing and inspection.

— The whole population of pipes, nozzles and penetrations connected to the main components are divided into subpopulations taking pipe diameters as orientation values, using e. g. 5 or 6 subpopulations to represent the difference pipe sizes, materials and operating conditions. For each subpopulation the frequency of leaks is based according to the procedure described before (operating experience, zero failure statistics), the frequency of leaks is adjusted to the size of the relevant population each time and the frequency of breaks within the subpopulation is estimated using the described relationship.

— The frequency of the different subpopulation which could contribute in a different way (leak or break) to the specified LOCA classes is then summed up. In view of the limitation regarding the verification of very low values of estimated frequencies a cut-off value is used.

For respective Base Case Plant, the LOCA frequencies are determined for three time periods: T= = 25 years after plant startup (corresponding to today’s state-of-knowledge), T = 40 years after plant startup (corresponding to original design life), and T = 60 years after plant startup (corresponding to end-of-life extension). The time-dependent analysis is performed in two different ways. First a ‘prospective analysis’ is performed based on a Markov model of piping reliability (Figure D.15). Second, a ‘retrospective analysis’ is performed by using Bayesian statistics.

D.6.4.1 Use of Markov Model to Determine Time-Dependency — According to the Markov model diagram in Figure D.15, a piping component can be in four mutually exclusive states: S (= Success), C (= Cracked), F (= Leaking, non-active leakage, or active leakage with leak rate within Technical Specification Limit) or L (= Leaking, with leak rate in excess of Technical Specification Limit). The time-dependent probability that a piping component is in each state S, C, F, or L is described by a differential equation. Under the assumption that all the state transition rates are constant the Markov model equations will consist of a set of coupled linear differential equations with constant coefficients. The reliability term needed to represent LOCA frequency is the system failure rate or hazard rate h{t}, which is time-dependent. The hazard rate is defined as:

h{t} = (1/(1- L{t})) x dL{t}/dt (D.9)

Where:

1 — L{t} = S{t} + C{t} + F{t} (D. 10)

The hazard rate is a function of time and the parameters of the Markov model; h{t} is the time-dependent frequency of pipe rupture. Reference [D.14] provides solutions to the Markov model and derives an expression for h{t} as a function of the six parameters associated with the 4-state Markov model: An occurrence rate for detectable flaws (ф), a failure rate for leaks given the existence of a flaw (Д) two rupture frequencies including one from the initial state of a flaw (pF) and one from the initial state of a leak (pL), a repair rate for detectable flaws (ft), and a repair rate for leaks (p). The latter two parameters dealing with repair are further developed by the following simple models.

P p

ft = —FI FD (D.11)

(TFI + TR)

Where:

PFI = probability that a piping element with a flaw will be inspected per inspection interval. This parameter has a value of 0 if it is not in the inspection program and 1 if it is in the inspection program. For the inspected elements, a value of 1 is used for any ISI inspection case and 0 for the case of no ISI. The element may be selected for inspection directly by being included in the sections sampled for ISI inspection, or indirectly by having a rule such that if degradation is detected anywhere in the system, the search will be expanded to include examination of that element.

PFD = probability that a flaw will be detected given this element is inspected. This is the reliability of the inspection program and is equivalent to the term used by NDE experts, “Probability of detection (POD).” This probability is conditioned on the occurrence of one or more detectable flaws in the segment according to the assumptions of the model. Also note that

TFI = mean time between inspections for flaws, (inspection interval).

TR = mean time to repair once detected. Depending on the location of the weld to be repaired, the actual weld repair could take on the order of several days to much more than a week. Accounting for time to prepare for repair, NDE, root cause evaluation, etc., the total outage time attributed to the repair of a Class 1 weld is on the order of 1 month or more. However, since this term is always combined with TFI, and TFI could be 10 years, in practice the results are insensitive to assumptions regarding TR

Similarly, estimates of the repair rate for leaks can be estimated according to:

Where:

PLD = probability that the leak in the element will be detected per leak inspection or detection period

TLI = mean time between inspections for leaks. For RCPB piping the time interval between leaks can be essentially instantaneous if the leak is picked up by radiation alarms, to as long as the time period between leak tests performed on the system.

TR = as defined above but for full power applications, this time should be the minimum of the actual repair time and the time associated with cooldown to enable repair and any waiting time for replacement piping.

A summary of the root input parameters of the Markov model and the general strategy for estimation of each parameter is presented in Table D.18.

Table D.18 Four-State Markov Model Root Input Parameters

In addition to generating a time-dependent LOCA frequency, the Markov model provides a basis for investigating the sensitivity of LOCA frequency to different in-service inspection and leak detection strategies. The Markov model determines the inspection effectiveness factor, I, which is the ratio of the LOCA frequency with credit for inspections to that given no credit for inspections:

h25{inspprog’ j’}
h25{noinsp}

Where:

h25{inspprog ‘j’} = hazard rate at T = 25 given inspection strategy ‘j.’ h25{noinspj = hazard rate given no inspections.

The solutions to the Markov model for time dependent hazard rates are developed in terms of closed form analytical solutions using an Excel spreadsheet. In this study the time-dependent LOCA frequencies are determined for twelve cases that are defined by varying the following parameters (Table D.19):

• Whether or not the piping segment is subjected to any ISI program;

• The extent of the ISI program (‘Caused-Based’ vs. ‘Extensive’, all encompassing);

• The inspection interval of the ISI program;

• Type and frequency of leak detection. The different leak detection methods include primary system mass balance calculations, visual observation (through video monitor), (PWR) containment sump level and flow rate monitors, airborne particulate radioactivity and gaseous radioactivity monitors, and different main control room monitors for primary system temperature, pressure, etc.

Table D.19 Inspection Cases Evaluated for Selected Pipe Segments

The time-dependent LOCA frequencies associated with the five Base Cases are summarized in Figures D.26 through D.40. Figure D.26 is assumed to be representative of Base Case 1; the LOCA frequency at T = 25 years is equal to the calculated point estimate of 8.24E-06 per reactor-year under an assumption of “caused — based” ISI with POD = 0.5 and leak detection (e. g., hydrostatic pressure testing prior to exiting a refueling outage). This assumption is applied to the other base cases as well (Figures D.29, D.32, D.35, and D.38).

It is noted that the service data input to the calculation is associated with piping that has been subjected to different inspection strategies. In some cases flaws have been detected fortuitously and in other cases the flaw detection has resulted from augmented IGSCC inspection programs. The results in Figure D.26 are based on an assumed ‘cause-based’ inspection strategy whereby the inspection sample is determined by an initial discovery of a flaw. If a flaw is found, the inspection is immediately expanded to cover other similar locations. The combination of inspection sample and rules for expanded search for flaws are sufficient to result in an average probability of detection (POD) of 0.50. The analysis also considers what in this study is

termed “comprehensive” ISI, which implies 100% ISI coverage using state-of-the-art NDE technology. Such a program is assumed to result in an average probability of detection of 0.90.

Figure D.26 Time-Dependent BWR-1 Cat 1 LOCA Frequency Given ‘Cause-Based’ ISI

Figure D.27 Time-Dependent BWR-1 Cat 1 LOCA Frequency Assuming no ISI

Figure D.33 Time-Dependent PWR-1 Cat 1 LOCA Frequency Assuming no ISI

Figure D.39 Time-Dependent PWR-3 Cat 1 LOCA Frequency Assuming no ISI

This process results in 216 data records that document crack (both partial and full) and leak events associated with non-pipe primary coolant system components. This dataset can be considered complete for U. S. NPP operation from 1990 through 2002, inclusive, in as much as the LER reporting requirements (Attachment A) can be relied upon to generate complete reporting. Additionally, the dataset does include a limited amount of data from outside this time frame and from non-U. S. plants. Nevertheless, the dataset can be considered to be internally consistent, that is, the various components, failures and degradation

mechanisms are believed to be represented equally such that relative ratios (if not the absolute frequencies) can be assumed to be reasonably accurate. Several of the database records represent multiple cases of degradation or failure. Attachment C includes a sample of multiple event records, including a discussion on how to estimate flaw frequencies from the observed events as recorded in the database.

The figure below (Forms view of the database) identifies the various fields maintained by the database. For additional detail on those records based on LERs, the LER hyperlink can be clicked to retrieve the full LER (internet access is require for this).

Submitted by Nuclear Energy Institute (NEI)

Comment: The report should provide a discussion on statistical validation of small LOCA frequency. By using the method of Jeffrey’s non-informative prior (over the past 2,500 reactor years with zero events excluding steam generator tube ruptures), the expected small LOCA frequency is at or below the 1E-04 level. This frequency is over one order of magnitude lower than the frequency reported in the draft NUREG. Plant operational experience of over 2,500 reactor-years should be considered as a valid predictor of small LOCAs. That consideration is further strengthened by improved methods and increased requirements for in-service-inspections and leak detections.

Response: The use of Jeffrey’s non-informative prior to estimate a frequency if no events have been observed leads to a significant underestimate of the mean frequency. Using this method, NEI states that the expected frequency is no more than 1E-04, based on a denominator of 2,500 reactor years. With zero events, a reasonable estimate of the mean is based on using a 50% confidence bound or a value of 0.7 in the numerator. This yields 0.7/2,500 = 2.8E-04 which is a factor of 3 larger than the estimate using Jeffrey’s non-informative prior. Furthermore, the use of 2,500 reactor years as the denominator clearly combines both PWR and BWR operating experience. More rigorous evaluation of operational experience data (See Section 7.10) has demonstrated that the operational experience-based and elicitation-based SB LOCA estimates are not inconsistent. Furthermore, differences are supported by the panelists’ responses and rationale. Section 7.10, which was added to address this and similar comments, provides additional details on the comparison between the elicitation estimates and operational experience data. Also, see the responses to GC3, GC4, GC5, GC7, and 7-8 for related information.

The total LOCA frequency is calculated using U. S. NPP experience of zero Category-1 LOCAs (i. e., greater than 380 lpm [100 gpm]) in 2,647 LWR-years of operation (as of 4/24/2003). A Bayesian update of a non-informative prior-distribution was performed to produce a total LOCA frequency of 1.9E-4 per LWR-year.

Table E.1 Total LOCA Frequency (per LWR-Year) Including Uncertainty, Using a Non-
Informative Prior and U. S. LWR Operating Experience

5% 50% mean 95%

7.4E-07 8.6E-05 1.9E-04 7.3E-04

E. 4 LOCA Frequency Allocation by RCS Pipe and Non-Pipe

The total LOCA frequency calculated above is first allocated between pipe and non-pipe passive components using data on primary system leaks and cracks collected from licensee event reports (LERs). These data records were collected, reviewed and categorized specifically for this effort. Since these data will only be used to ascribe a relative frequency between pipe and passive non-pipe components, complete data are not necessary, only data that have been reported consistently. These data and the resultant allocation are summarized in the table below. Steam generator tube ruptures are being assessed separately, and are therefore removed from this allocation.