An Accelerated Life Test Model Based on Reliability Kinetics

Abstract

Inferences from accelerated life tests typically involve extrapolation in stress. For this reason, it is important to use statistical models that have their basis in the physics and chemistry of the important failure mechanism(s). The standard accelerated failure time regression models (based for example on Weibull or lognormal approximations to the failure-time distribution at a given stress and linear scaling of time) are adequate for modeling some simple chemical processes that lead to failure. In this article we present the results from a humidity-accelerated life test of conductive anodic filament failures on printed circuit boards. Standard accelerated life test models are clearly inadequate for these data. Using an approximate chemical kinetic model of the failure process as a basis, we derive an alternative, more general, class of accelerated life test models. We illustrate the use of likelihood-based methods to estimate the model parameters. The new models fit the data better than the traditional accelerated life test models and provide extrapolations that are more consistent with actual field data.