Simulation-based PRA for spent fuel pool: Final report

This report presents an approach to simulation-based PRA of a spent fuel pool. Simulation-based event tree models are developed to analyse loss of offsite power (LOOP) and transient scenarios of a fictive spent fuel pool. In the simulation-based event tree, event timings, such as failure times of components and durations of manual actions, are simulated to analyse time-dependencies. The time windows for probabilistic analysis, namely mission times for safety functions and available times for manual actions, are calculated based on spent fuel pool conditions affected by the timings of previous events. The model combines deterministic and probabilistic analysis; the spent fuel pool conditions are calculated by a simplified, but sufficiently realistic deterministic model. In this report, the simulation-based models are used to quantify minimal cut sets of a static PRA model more realistically. The results of dynamic and static analyses are compared. The dynamic analysis decreases the frequencies of many minimal cut sets significantly. The decrease is particularly related to more realistic definition of mission times and crediting the operation of the cooling/make-up systems before they fail, which gives more time for the following manual actions. The results also indicate that crediting repairs can greatly decrease the frequencies. As an alternative to Monte Carlo simulation, discretization of time distributions is investigated as a method to select the simulation cases in order to reduce the computation time. One sequence from both LOOP and transient models is analysed using the discretization. In the transient case, the approach produces only slightly conservative result with a small number of simulation cycles. On the other hand, in the LOOP case, the result is very conservative even with a very dense discretization, because the tails of the offsite power recovery time and diesel generator repair time distributions dominate the result. The usefulness of the time discretization approach seems to be case-specific.