Stress-coupled parameter learning in stochastic processes for reliability-oriented accelerated degradation modeling

Accelerated degradation testing is widely used for reliability assessment, yet stochastic processes cannot be directly utilized to model accelerated degradation data because they lack inherent stress covariate. To address this problem, the stress-coupled parameters, whose values vary with stress levels, must first be determined. Then, the stress-coupled parameters are replaced by acceleration models to construct a degradation model incorporating stress covariate. However, the assumed stress-coupled parameters have commonly been adopted, resulting in poor explicability of accelerated degradation modeling and unreliable life extrapolation results, particularly when multiple assumptions on stress-coupled parameters exist for a given degradation model. This study proposes an explicable framework for determining stress-coupled parameters in stochastic processes. The framework consists of two approaches: an analytical derivation method for linear degradation models and a numerical computation method for nonlinear models. These approaches are applied to Wiener, Gamma, and Inverse Gaussian processes to derive stress-coupled parameters, validating the equivalence of them for linear degradation models and revealing that different time functions fundamentally determine the stress-dependency patterns of model parameters. The numerical computation method is applicable to complex degradation models, but it can not discover the potential proportional relationship among multiple stress-coupled parameters. To address this issue, the framework offers a workflow combining the two approaches. A case study on electrical connectors further verifies the practical effectiveness of the proposed framework. The results show that the framework works well in deriving stress-coupled parameters and benefits strengthening the theoretical basis of accelerate degradation modeling.