Impact analysis of prior distributions on ADT Bayesian optimization design based on DIC

Abstract

Accelerated degradation testing (ADT) optimization design means that the ADT plans are designed under some particular conditions, e.g. stress range, inspection times, testing cost, etc., to obtain accurate estimates of the reliability indexes. Currently, the ADT optimization design has been developed to be one of the most important techniques in the field of accelerated testing. Traditional design methods optimizes the plans through assuming that the values of parameters arbitrarily, which may lead to uncertainty in design results. On the contrary, Bayesian optimization design can utilize the prior information of the products, e.g. the historical information, similar products' data, etc., sufficiently. But different people may get different prior distributions from the same prior information, resulting in different ADT Bayesian optimization plans. How to choose a correct prior distribution becomes a difficult problem. Hence, this paper will do the impact analysis of prior distributions on ADT Bayesian optimization design method based on deviance information criterion (DIC). Firstly, different prior distributions are regarded as the input of the optimization design method to get the corresponding optimal testing plans. Then, robustness analysis of prior distributions on ADT Bayesian optimization design method is studied by comparing different optimization plans. Lastly, DIC is presented as prior distribution selection criteria of optimization design method and its effectiveness is verified through a simulation case. Furthermore, this research can guide the ADT optimization design when facing the selection problem of prior distributions and saving the test costs and resources.