A graphical approach to identification of dependent failures

Field data provide a rich source of information about the dependent failures whose omission from existing models can result in underestimation of the reliability of repairable systems, A graphical technique has been developed to highlight these events. This involves comparing the observed number of failures with the expected pattern under a null model of no common cause dependence, a non-homogeneous Poisson process with Weibull rate function. The derivation of the graph is outlined and its use as a screening tool is illustrated by applications to field data. The dependent failures identified are described and their engineering implications are discussed. The statistical power of the technique is evaluated for a range of alternative models of dependence, including simple shock models.