Modified Bayesian D-Optimality for Accelerated Degradation Test Design With Model Uncertainty

Accelerated degradation test (ADT) has become the main method to assess system reliability. In ADT, Bayesian D-optimality criterion is an effective objective function to deal with the model parameter uncertainty in designing an ADT plan. Current optimal ADT designs based on Bayesian D-optimality usually assume a single degradation model. However, this assumption sometimes may not be robust enough because the degradation trajectories of the target systems are usually unknown. Thus, besides the model parameter uncertainty, the degradation model uncertainty is also a question worth considering during designing an optimal ADT plan. Integrating the advantage of Bayesian model averaging to Bayesian D-optimality, this paper proposes a modified Bayesian D-optimality to handle the model parameter uncertainty and the model uncertainty simultaneously. Simulations are provided to analyze the robustness of the proposed objective function. The results show that the modified Bayesian D-optimality could effectively deal with the uncertainties of the degradation model and model parameter.