An alternative method of analyzing multi-stress multi-level life and accelerated life tests

Abstract

In the study of life testing, it has been shown that multiple stresses and multiple stress levels are important features to exploit for better inference. In addition, the incorporating of design of experiments concepts in multi-level, multi-stress life and accelerated life tests is more frequently being advocated in the military-industrial community. The benefits of efficient testing by fractional factorial designs and the ability to estimate the effects of the stresses and their possible interactions, provides reason to investigate the combination of design of experiment techniques with multi-stress testing. However, the need for an analysis method that requires neither the 3 or 4 replications of each test cell nor the calculation of pseudo cell MTTFs, is both necessary and obvious. In this paper, some statistical problems that occur in the use of standard ANOVA and regression analysis technique for the analysis of these type of test results are identified. In addressing these problems, an alternative Bayesian analysis procedure is proposed.