UNCERTAINTY PROPAGATION IN PROBABILISTIC SAFETY ANALYSIS USING LOGNORMAL DISTRIBUTION

Abstract

ncertainty analysis, which is one of the major elements in Probabilistic Safety Assessments (PSA) of Nuclear Power Plants (NPP's), involves quantifying the uncertainties of the occurrence of accident scenarios. The traditional approximations used in current PSA models are limited, and normally conservative, based on not fully accounting for the dependence between Minimal Cut Sets (MCS). In this work, a mathematical development of an approximation method to propagate the uncertainty of lognormal distributions is carried out by modifying the approach suggested by Fenton and Wilkinson. When the uncertainties of basic events are modelled with lognormal random variables, the top event frequency or probability is well approximated as the sum of the correlated lognormal random variables. This study focuses on how to minimize or eliminate the adverse effect of Rare Event Approximation (RAE) that induces the overestimated top event mean value. The probability distribution of the case study top event presented in this work is compared with analyses available in the literature using different approaches, such as Monte Carlo and Fenton-Wilkinson (FW) method. The application of this method to propagate uncertainty of lognormal distributions results in a better estimation of the top event probability distribution. It is shown how the cut set information for a model can be used together with the analytic expression to give closed-form approximation for the top event uncertainty distribution. This approach appears attractive and can provide a reasonable approximation, without incurring the computational expense.