Optimal planning and inference for sequential accelerated life testing with two or more experimental factors

An important task before conducting Accelerated Life Testing (ALT) experiments is to specify a prior lifetime model, based on the historical data of similar products or expert opinions. The initial estimates of model parameters need to be reasonable so that the test plan can produce sufficient failure data. Though many methods have been developed to design test plans with unknown prior distributions, there is still active research in this area to obtain the best value of the final parameter estimates. A main drawback is that, in most cases, these ALT test plans consider only one stage of experimentation, which is often inadequate for building a reasonable prediction model. In this paper, we propose a modified version of sequential ALT planning and life quantile prediction framework involving multiple factors. The first stage of design is carried out based on the prior knowledge of various possible acceleration regression models for a limited testing time and experimenting at more than one level for at least one factor, followed by an adaptive second-stage ALT test planned under the given budget to improve the prediction accuracy obtained from the first stage. The proposed approach is validated through real accelerated life testing data of Multi-Layer Ceramic Capacitor (MLCC) data involving three factors: Temperature, humidity and voltage.