Estimating Safety Valve Stochastic Failure-to-Close Probabilities for the Purpose of Nuclear Reactor Severe Accident Analysis

Abstract

Recent consequences analyses of potential station blackout (SBO) accidents at nuclear power plants have shown that an important uncertainty in accident progression and radionuclide release is the probability that a safety valve (SV) will fail-to-close after it has opened to relieve pressure [1]. The U.S. Nuclear Regulatory Commission's (NRC's) State-of-the-Art Reactor Consequence Analyses (SOARCA) and associated uncertainty analyses for SBOs at a pressurized-water reactor (PWR) indicated that SV behavior is an important determinant of whether an induced steam-generator tube rupture (an undesirable bypass event) may develop [2], and an important determinant of whether a PWR with an ice condenser containment may experience an early containment failure [3]. Given the importance of SV failure-to-close probabilities in these accidents, available information was reviewed to help develop better estimates of the probability for a SV's failure-to-close on demand. The SVs of interest in the SOARCA PWR analyses are the PWR code SVs, designated SVVs in a study of SVs published in 2007 (NUREG/CR-7037) [4]. There are two sets of failure probabilities reported in NUREG/CR-7037: failure probabilities based on behavior after reactor scrams i.e., after actual operating events, and failure probabilities based on tests. Information is included for both the secondary-side, main steam system (MSS) valves, as well as reactor coolant system (RCS) valves.The NUREG/CR-7037 failure probabilities based on actual operating events differ markedly from the failure probabilities based on tests. Further inquiries on valve testing and review of testing requirements show that the focus of testing is to demonstrate that the valves will open to relieve pressure during design-basis accidents to prevent overpressure events. The reseating or closing capability is not tested under severe accident conditions, in other words, the valve's repeated full-stroking and passing steam. As such, the testing data was not considered applicable for severe accident modeling purposes. Furthermore, the assumption was made that MSS data was representative of RCS valve failures too during severe accident scenarios, as it is judged that they are similar enough in weighing the difference between the valves against the lack of operational data on the RCS SVs (only four data points, and one of two failures having a cause of failure nowdefunct in the majority of operating PWRs in the U.S.). Lastly, recovered valve function, e.g., a previously stuck-open valve closing when pressure reduces, was not considered as a successful valve operation based on a review of licensee event reports.