Physics-Informed Artificial Intelligence for Battery Degradation, Aging, and Lifetime Prediction

Abstract

Accurate prediction of battery degradation, aging trajectories, and lifetime is critical for electrochemical energy storage systems, with direct implications for safety, reliability, and lifecycle cost. Although artificial intelligence (AI)–based approaches have shown strong short-term performance in state-of-health and remaining useful life estimation, many remain data-driven and show limited robustness under long-term cycling, operating-regime shifts, and extrapolation beyond training data. Physics-based degradation models provide mechanistic insight, but face challenges related to computational cost, parameter uncertainty, and scalability. This review positions physics-informed artificial intelligence (PIAI) as an integrative modeling paradigm that can improve data efficiency, physical consistency, uncertainty awareness, and decision reliability when appropriately constrained and validated. Unlike reviews that examine AI algorithms, degradation mechanisms, hybrid models, digital twins, or uncertainty-aware prognostics separately, this work provides a unified multiscale synthesis linking degradation physics, PIAI model construction, uncertainty quantification, structured validation, and BMS/digital-twin deployment. We examine how physical laws, degradation constraints, and uncertainty representations are embedded into learning architectures and analyze failure modes of black-box and poorly constrained hybrid models. The review proposes a five-layer framework for trustworthy battery lifetime prediction and identifies directions for physically credible, uncertainty-aware, and deployment-ready battery intelligence.