What if we revisit evaluation of PSA models with network algorithms ?

摘要

In this paper we propose an approach to solve probabilistic safety assessment models using graph theory algorithms. The problem of evaluating event tree consequences is reduced to the problem of finding minimal (s,t)-cutsets in directed graphs. The main idea is to use aggregation over thermohydraulic, electric and Instrumentation & Control systems to model the failure of system missions as a problem of finding (s,t)-cuts in corresponding directed graphs. We try to experiment graph theory algorithms on these PSA models with some adaptations to deal with some specific features (dependencies, common cause failures, and dynamic phenomena…). We expect that the algorithms developed for network area help to have some original views of probabilistic safety assessment and may be new safety/reliability metrics, even if there may be no satisfying solution. But this structure may help PSA analysts to have a comprehensive and visualized view of the characteristics of the PSA model, which may serve at at least for interpretation and diagnosis purposes.