Robust Planning of Accelerated Destructive Degradation Tests: Model Discrimination, Parameter Estimation, and Lifetime Prediction

ABSTRACT Accelerated destructive degradation tests (ADDTs) play a cr ucial role in providing timely reliability information for long‐life products. Previous research has primarily focused on identifying optimal ADDT plans for accurately predicting specific quantiles of the failure‐time distribution under use conditions, utilizing a given degradation model. However, a common challenge during the experimental design phase is the lack of confidence in selecting a model that accurately represents the underlying data. In this paper, we introduce a novel approach that specifically addresses optimal ADDT planning for model discrimination. We propose KL‐optimal design criteria to enhance the model discrimination capability of ADDTs. Moreover, we present compound DKL‐ and CKL‐optimal design criteria to ensure that ADDT plans can simultaneously achieve effective model discrimination alongside accurate parameter estimation or precise quantile prediction. We establish equivalence theorems to validate the optimality of the proposed designs. Meanwhile, we employ the particle swarm optimization algorithm to efficiently compute optimal ADDT plans. Through a practical application and sensitivity analysis, we demonstrate the effectiveness and robustness of our optimal designs. Our proposed method offers engineers with a valuable solution for developing optimal ADDT plans that meet multiple objectives.