Mixed Gamma Accelerated Degradation Model and Life Evaluation Considering Multi‐Subpopulation Heterogeneity

Acquiring sufficient and reliable full life‐cycle data is indispensable to enable accurate fault diagnosis and prognosis for engineering products. However, this poses a fundamental challenge for high‐reliability products. Accelerated Degradation Testing (ADT), which accelerates the product degradation process by applying elevated stress levels, provides an effective approach for degradation modeling of high‐reliability products. Nevertheless, significant discrepancies in degradation paths exist among individual units of the same product model due to inherent factors including batch‐to‐batch material variations and manufacturing process fluctuations. Such discrepancies are further amplified under accelerated stresses, making it critical to establish an accelerated degradation model that can accurately characterize the inter‐individual degradation heterogeneity. Two core limitations persist in mainstream accelerated degradation modeling research. First, the widely used single stochastic process models are constructed based on the homogeneous degradation assumption, which essentially performs an averaging treatment on the heterogeneous degradation behaviors of individual units. These models fail to capture the differences in degradation paths among individuals, directly resulting in inaccurate Remaining Useful Life (RUL) prediction and erroneous maintenance decision‐making. Second, although improved models with random effects can characterize stochastic degradation fluctuations at the individual unit level, they cannot depict the systematic degradation heterogeneity among multiple sub‐populations of products caused by material and process variations, and thus are difficult to adapt to engineering scenarios with multi‐mode degradation. To fill the above research gaps, this paper proposes an adaptive dynamic weighted mixed Gamma stochastic process model. This model integrates multiple Gamma sub‐stochastic processes with adaptive dynamic weights and adopts the Markov Chain Monte Carlo (MCMC) algorithm to estimate the parameters of the mixed model. Validated by the accelerated degradation data of carbon film resistors, the proposed mixed Gamma model outperforms traditional single stochastic models in both the analysis of individual degradation path characteristics and data fitting accuracy. This work provides a more reliable technical basis for fault prognosis and maintenance optimization of high‐reliability products with significant inter‐individual degradation heterogeneity.