Electronics Life Prediction based on Multiple Performance Parameters ADT using Multidimensional Time Series Model

Abstract

This paper proposes a new Accelerated Degradation Testing (ADT) life prediction method utilizing a multidimensional composite time series modeling procedure to take into account the integrated effect of system’s multiple performance parameters along with the random effect of environmental variables for equivalent damage in ADT. In this paper, system performance parameter ADT data are treated as a multidimensional composite time series model to predict system failure time. First, this paper decomposes these multiple performance parameters useful for ADT into three classes as trend, cyclical or random components, and describes them with a combined multi-dependent variable regressive model, hidden periodic model and multivariate auto-regression model. Second, according to standard practice, this paper assumes that the failure of such a system obeys a competing failure rule, that is, for an individual unit there is one primary controlling variable that will indicate failure even though others degrade they do not meet any failure criterion. Failure time at each test-stress level is predicted by using the best linear unbiased prediction of the multidimensional composite time series model. Finally, the reliability at use-stress level is estimated from a failure time distribution evaluation based on the failure time predictions at each test-stress level providing a relationship between failure time and test-stress levels.