Uncertainty analysis in PSA with correlated input parameters

Abstract

The probability of occurrence of top event of a fault tree in probabilistic safety assessment (PSA) is estimated from the probabilities of the basic events which constitute the fault tree. However the failure probabilities of basic events are subjected to statistical uncertainty. While analyzing the uncertainty of top event, the basic events are assumed uncorrelated or independent in most of the situation, but are not the case every time. Such statistical correlations are introduced into failure data by many causes. To handle the propagation of uncertainties in the presence of correlations, there are three methods: (i) Method of moments, (ii) P-box approach and (iii) Monte Carlo simulations. However the first two methods are difficult for implementation in large scale problems when partial correlations are present instead of fully correlated data. The paper presents a methodology based Monte Carlo simulation with Nataf Transformation of generating correlated random variables for uncertainty analysis in PSA. Computer code has been developed to implement the proposed methodology and a case study from NPP has been carried out.