Optimal design of constant stress accelerated degradation test plan with multiple stresses and multiple degradation measures

Abstract

High-reliability and long-lifetime products are often exposed to a variety of stresses in use condition, and there are usually multiple performance indicators of the products which gradually degrade over time. The technology of constant stress accelerated degradation test with multiple stresses and multiple degradation measures provides an efficient and feasible way for lifetime prediction of such products. It is one of the core issues in accelerated degradation test research how to design the plan of constant stress accelerated degradation test with multiple stresses and multiple degradation measures to result in the most accurate lifetime prediction for such products under the constraint of test budget. Existing methods of optimal design for accelerated degradation test are only for cases of a single stress or a single degradation measure which cannot be applied to solve the complex problem of optimal design for test plans under an occasion of multiple stresses and multiple degradation measures. Aimed at an occasion that analytical form of optimization objective function is difficult to derive or even does not exist, a novelty method of optimal design for constant stress accelerated degradation test with multiple stresses and multiple degradation measures is presented by incorporating Monte Carlo simulation into dealing with this problem, setting up a optimization model including design variables, optimization objective function and constraints, presenting optimization algorithm based on Monte Carlo simulation and surface fitting. Optimal design for constant stress accelerated degradation test with multiple stresses and multiple degradation measures of rubber sealing O-ring is taken as an illustrative example to demonstrate validity of the presented method. The results of sensitivity analysis of the optimal test plan show that the plan is robust when the deviation of parameters of the model is within 10%.