Base Case Report 2 develops BWR and PWR LOCA frequency distributions using a ‘bottom-up approach.’ Statistical analysis of relevant service experience data is used to quantify the weld failure rate and rupture frequency of individual welds. Next the failure rate and rupture frequency (= LOCA frequency) for an entire system is calculated by concatenating the individual weld failure rates and rupture frequencies. Markov model theory is used to evaluate the influence of alternate strategies for in-service inspection and leak detection on the frequency of leaks and ruptures.

D.2.1 Overview of Analysis Steps

Different approaches have been applied to estimating pipe failure rates and rupture frequencies; from probabilistic fracture mechanics, via direct statistical estimation to expert judgment. The most straightforward approach is to obtain statistical estimates of piping component failure rates based on data collected from field experience. A variation of this approach is to augment statistical estimates of pipe failure parameters with simple correlations that express the problem in terms of a failure rate and a conditional probability for each failure mode of interest such as the approach used in NUREG/CR-5750 [D.17].

A limitation of the statistical analysis approach is that attempts to segregate the service data to isolate the impact of key design parameters and properties of various degradation and damage mechanisms often leads to subdividing a database into very sparse data sets. If not optimized properly, this approach may introduce large uncertainties in the failure rate estimates. In addition, historical data may reflect the influence of no longer relevant inspection programs. If changes to these programs have been implemented, such changes may render the failure rate estimates no longer relevant. In risk-informed applications, the failure data and analysis methods need to provide future predictions of piping system reliability that can account for changes in the inspection strategy or improvement in the NDE technology.

An objective of the work documented in this report is to demonstrate the utility of a pipe failure data collection. Time-dependent LOCA frequencies are calculated by making full use of the PIPExp database in combination with Markov model theory [D.14]. The LOCA frequency calculation in this report is structured to support the Expert Elicitation and consists of four steps; each step is addressed in a separate report section:

• Section D.3. The service experience that is applicable to the five bases cases is summarized in this section. The data summaries correspond to queries in the PIPExp database.

• Section D.4. The approach to calculating time-dependent LOCA frequencies is presented. A Bayesian update process is used to derive failure parameters that reflect the attributes of respective base case definition. The results of this analysis step are in the form of generic weld failure rate distributions. These distributions represent the industry-wide service experience prior to the implementation of the specific pipe failure mitigation programs that are currently in place.

• Section D.5. In this section current state-of-knowledge (or base case specific) weld failure rate distributions are develop. The chosen estimation approach includes a formal uncertainty analysis that accounts for uncertainty in the failure data and exposure data. Engineering judgment and insights from the review of service data are used to address the conditional probability of pipe failure given presence of through-wall flaws.

• Section D.6. An Excel spreadsheet format is used to develop LOCA frequency models corresponding to each of the five base cases. These models generate LOCA frequency distributions at T = 25 years. A Markov model is used to investigate the time-dependency of LOCA frequencies. The output of this model consists of LOCA frequencies at T = 40 years and T = 60 years.