Impact of translation approach for modelling correlated non-normal variables on parallel system reliability

Abstract

The adequacy of two approximate methods based on incomplete information, namely method P and method S, for constructing multivariate distributions with given marginal distributions and covariance has not been studied systematically. This article aims to study the errors of the method P and method S. First, the method P and method S as well as the exact method are presented. Second, the performance of the two approximate methods is evaluated based on their abilities to match exact solutions for system probabilities of failure. Finally, an illustrative example of a parallel system is investigated to demonstrate the errors associated with the two methods. The results indicate that the errors in system probabilities of failure for the two methods highly depend on the level of system probability of failure, the performance function underlying the system, and the degree of correlation. Such errors increase greatly with decreasing system probabilities of failure. When the target system probability of failure is larger than 1.0E-03, the system probabilities of failure obtained from the two methods and the exact method are of the same order of magnitude. The maximum error in the system probability of failure may not be associated with a large correlation. It can happen at an intermediate correlation. © 2013 Copyright Taylor and Francis Group, LLC.