Accelerated life testings(ALT) are usually conducted at higher than normal stress levels, in order to induce failures. And it is often assumed that there is a close relation between certain paramaters of the life distribution and the stress level at which the experiment is conducted. However, because of the different effects on the failure mechanisms, this assumption is unreasonable in many practical situations. In this article, we relax this assumption to fit the data arising from an ALT, and we apply the Dynamic Bayesian method to inference from ALT, and then modify the method, which is proposed by Proschan and Singpurwalla(1980), of extrapolating the failure rate under the normal stress to the proportional hazards model. According to the results of simulation study and analysis of real data, the estimated failure rate, by our method, under the normal stress is more reasonable.
COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION