The approach taken by Bill Galyean is based on the total operating experience of U. S. commercial nuclear power plants. This experience consists of approximately 2,650 LWR-years with zero category-1 (> 100 gpm [380 lpm]) loss of coolant accidents. The average age of these plants is approximately 25 years, with a number of plants being 30-plus years old. During this time, a number of RCS degradation issues have arisen and been addressed, for example, IGSCC in BWRs and thermal fatigue in PWRs. The operating experience therefore indicates that degradation will occur, but it will likely be detected and corrected before it can lead to a catastrophic failure. Consequently, this data is the basis for estimating an average LOCA frequency using a Bayesian update of a non-informative prior distribution. Since both PWRs and BWRs have zero LOCAs, the reasonable assumption is that the two designs share a similar LOCA frequency. The operating experience for the two designs is therefore pooled (i. e., use zero failures and the total 2,650 LWR-years of experience). Assuming the LOCA frequency has been (and will be) relatively constant over time (again, this seems reasonable given the history of degradation mechanisms being detected and subsequently mitigated), the resulting LOCA frequency of 1.9E-4/LWR-year produces a probability of one or more LOCA events in the 2,650 LWR years of experience of 39% (again not unreasonable, given there have been zero LOCA events). By contrast, separating the PWR and BWR experience and analyzing them separately produces LOCA frequencies of 2.8E-4/PWR-yr and 5.6E — 4/BWR-yr, and a probability of seeing one or more LOCA events (either PWR or BWR) in 2,650 LWR — years experience of 63%. Again, given that there have been no LOCA events, the first (pooled) estimate seems to be the more realistic.

This assumed LOCA frequency (1.9E-4/LWR-yr) was used for the category 1 LOCA (> 100 gpm [380 lpm]). Note that as defined in the elicitation effort, category 1 LOCA includes all larger size categories. So the approach followed by this panel member was to assume a Уг order of magnitude reduction in frequency for each next larger size category. This general approach (if not the precise value of the reduction) has been followed by virtually every LOCA frequency estimate ever made, and is supported by studies on precursor events documented in NUREG/CR-5750, Appendix J.

The time-independent assumption for the LOCA frequency is also based upon the historical experience, if only qualitatively. There seems to be no doubt that the LOCA frequency fluctuates over the age of the plant, but there is reason to believe it will both increase and decrease over time. The IGSCC experience seems to support the assertion that times of increasing frequency will be followed by times of decreasing frequency as degradation mechanisms are identified, understood, and mitigated. Indeed, even the recent RPV-head corrosion event at Davis Besse supports this model of a LOCA frequency increase as degradation occurs undetected, then a decrease as mitigation programs are implemented (e. g., in the case of Davis Besse, replacing RPV head).

The last issue to be addressed is the allocation of the total LOCA frequency among the systems and components that compose the RCS. This aspect again relies upon operating experience data, this time in the form of the relative frequency of crack and leak events (i. e., precursor events). Basically, these precursor data were collected from LER and foreign reactor experience, and then sorted by degradation mechanism and RCS subsystem/component. In many cases the information provided on the precursor event was somewhat unclear or incomplete. Also, there is little assurance that all precursor events have been captured. However, assuming there is no bias in the reporting of the events such that the data samples for each subsystem/component are equally representative of the all events for that subsystem/component, then the data can be used to support estimates of the relative contribution from each subsystem/component. That is, the precursor events do not have to be completely reported, just consistently reported. Further, the RCS subsystem/component boundaries have not been clearly defined. Hence, the relative contributions to the overall LOCA frequency would likely change somewhat if the precursor data were reviewed and categorized by a different analyst. Nevertheless, this aspect of the analysis was performed simply to allocate the total LOCA frequency (described above) to the general subsystems/components that make-up the RCS.

In summary, the entire U. S. LWR operating experience is used to estimate an average industry-wide total LOCA frequency. This frequency is used for both BWRs and PWRs, not because they are believed to be the same, but on the basis that the operating experience does not support different frequencies. Time — independence is assumed using the rationale that variation (both increases and decreases) in the frequency will occur as degradation mechanisms manifest themselves and are subsequently addressed by the industry. This total LOCA frequency is allocated by LOCA size categories using the argument that as pipe-size increases, the LOCA frequency decreases. This argument is supported by a number of studies on precursor data and if nothing else has been reflected in all LOCA frequency estimates since WASH — 1400 (1975). The total LOCA frequency is also allocated by RCS subsystem/component using data collected on primary system crack and leak events (although the details of this allocation are view as somewhat subjective with respect to the boundaries of the different subsystems/components).