A Bayesian approach to modeling two-phase degradation using change-point regression

Abstract

Influenced by defects or contaminants remaining after a series of manufacturing processes, the degradation paths of some products exhibit two-phase patterns over the testing period. This paper proposes a hierarchical Bayesian change-point regression model to fit the two-phase degradation patterns, and derives the failure-time distribution of a unit that is randomly selected from its population. A Gibbs sampling algorithm is developed for the inference of the parameters in the change-point degradation model, as well as for the prediction of the failure-time distribution of the randomly selected unit. The proposed approach is applied to the degradation paths of plasma display panels (PDPs) presenting the two-phase pattern.