Abstract
Accurate remaining useful life (RUL) prediction of electronic components is of paramount importance for maintaining the long-term reliability and safety of complex systems exposed to highly variable and demanding operational environments. However, most existing physics-informed machine learning (PIML) approaches for electronics reliability still suffer from a key limitation: they typically incorporate fixed physical priors while assigning static importance to multivariate degradation indicators throughout the component’s lifetime. Consequently, the predicted and actual degradation trajectories tend to become misaligned, and the models often lack adequate uncertainty estimates required for robust decision-making in the presence of distribution shifts. To address these limitations, this paper proposes the Self-Attention-Guided Quantum Assimilation with Bayesian Updated Nonlinear Wiener process (SAQ-BUNW), a physics-inspired quantum-Bayesian Wiener framework for RUL prediction. Rather than embedding a closed-form degradation equation, the framework injects physical knowledge through an energy-landscape prior built from a quantum-mechanical analogy, while a nonlinear Wiener process models the stochastic degradation dynamics and quantifies predictive uncertainty. SAQ-BUNW integrates quantum state-based feature assimilation with a self-attention mechanism that emphasizes degradation-sensitive variables, while Bayesian updating within a nonlinear Wiener process enables uncertainty estimation. We validate the method on SF158MD operational amplifiers under accelerated degradation. Tested by leave-one-out cross-validation (LOOCV) against a broad range of baselines, SAQ-BUNW improves point-prediction accuracy while also producing better-calibrated uncertainty estimates.