Importance measure evaluation based on sensitivity coefficient for probabilistic risk assessment

Abstract

The failure probability of the target event might not be linear to that of the component in the probabilistic risk assessment model if the component is correlated with another one or it has redundancies. If the failure probability of the component is uncertain in the non-linear model, the failure probability of the target event cannot be accurately obtained only from the expectation value of the component failure probability. Therefore, generally, a lot of Monte Carlo simulations are required for obtaining the importance measure such as FV, RRW, RAW, and BI. Since the calculation cost of the simulation is large in the analysis of complicated systems, the present paper proposes the estimation technique based on the sensitivity coefficient for importance measures. In the proposed method, the sensitivity coefficient is estimated from one Monte Carlo simulation result, thus the calculation cost is much less than the direct Monte Carlo simulations. As the non-linear model, the fault tree model with correlated components and that with the β factor method were analyzed. The results show that the proposed method based on the sensitivity coefficient can accurately estimate the importance measure for these models.