Modeling and analysis for multiple stress-type accelerated life data

Abstract

This paper describes a model for multiple stress-type accelerated life data. In addition, the use of an algorithm, which was specifically developed for this model is illustrated. The model is based on the widely known log-linear model (1990) and is formulated for the Weibull and lognormal distributions for a variety of censoring schemes using likelihood theory. An algorithm has been developed for the solution of this model, and implemented in a recently released software package, ALTA Pro/sup TM/, specific to accelerated life data analysis. The algorithm has been specifically designed to be very flexible and has the capability of simultaneously solving for up to eight different stress-types. The advantage of this formulation is that it combines in one model most of the known life-stress relationships for one or two types of stresses (such as the temperature-nonthermal model), as well as the multivariable proportional hazards model. This yields a single general likelihood function (for a given distribution) whose solution is independent of both the chosen life-stress relationship and the number of stress-types. In addition, this model allows for simultaneous analysis of continuous, categorical and indicator variables. The solution to this model provides the engineers with an opportunity to expand their selection of types of stresses and test conditions when testing products.