Dirichlet binomial attribute testing model: a Bayesian approach to estimating reliability decay

A model capable of estimating the current reliability and offer predictions of future reliability of products or materials that are in a prolonged nonoperating state, such as long-term storage, would be of benefit to both commercial and government organizations alike. Many items are subject to nonoperating or dormancy reliability decay with classic examples including missiles and munitions. This paper presents a Bayesian model that incorporates an ordered Dirichlet prior with a binomial attribute testing model to estimate reliability decay. Reliability decay predictions can also be generated. Classical approaches to this problem assume an exponential based reliability aging model. The strength of the proposed model is that it makes no assumption of the underlying failure distribution, but is based purely on pass/fail parameters of an attribute test.