Abstract
Explicitly addressing the effects of correlated component failures is a challenge associated with reliability modeling. Several recent studies utilize an approach based on common cause groups (CCG). While this method enables modeling correlation, the number of possible common cause groups is an exponential function of the number of components. The large number of parameters makes it difficult to estimate the probability of such common cause failures with any certainty, especially when testing data is limited. This paper presents a technique to determine the joint reliability distribution of a set of components subject to correlated failures. It is then possible to perform reliability analysis for systems built from these components. The only inputs required are the components' expected reliabilities and their correlation matrix. Thus, one need only consider a quadratic number of pairwise component correlations. Several applications of this technique are compared with the traditional approach; which ignores correlation. In some cases, a system with common mode failures exhibits a higher reliability than one with statistically independent components. Finally, the danger of assuming independent components is demonstrated, showing that the simplified approach produces suboptimal solutions.