Adaptive Accelerated Degradation Test Design for Tweedie Exponential Dispersion Process With Random Effects Based on the Predictive Remaining Useful Life

The Tweedie exponential dispersion (TED) process provides a flexible framework for degradation process modeling. To obtain degradation information quickly and realistically, we focus on the design of accelerated degradation tests (ADTs) under the TED process with random effects (TEDRE), commonly encountered in practice. Traditional ADT designs primarily rely on the historical data, which may not consider the differences between historical batches and current products. Therefore, we propose an adaptive ADT design method for TEDRE incorporating the historical and online degradation data. First, optimize major decision variables for ADT design using the historical data; subsequently, update measurement intervals and times based on real-time remaining useful life (RUL) predictions from the online degradation data. In addition, a three-stage D-optimality criterion is proposed to simplify the Fisher information matrix computation. To better predict RUL and optimize ADT design, we propose a novel approach to organically combine maximum likelihood estimation and generalized pivotal quantity for small samples and model complexity, and an improved EM algorithm utilizing analytical solutions of model parameters to resolve the slow convergence and accuracy issues; then, simulation studies provide detailed application recommendations. Finally, we illustrate the effectiveness of the proposed method through the stress relaxation and spiral springs data.