Physics-informed neural networks for battery smart modeling and state estimation with electrochemical mechanisms and data-driven intelligence

Physics-informed neural networks (PINNs) have emerged as a promising framework that bridges physics-based battery models and data-driven learning. This review systematically summarizes recent progress in PINN-based battery modeling and state estimation. First, the theoretical foundations of PINNs are analyzed, including physical constraint embedding, loss function formulation, boundary condition treatment, and training stability. Second, battery modeling approaches that can be integrated with PINNs are reviewed, including electrochemical models, electro-thermal coupled models, parameter identification and uncertainty analysis frameworks, and cross-scale or cross-chemistry modeling strategies. Third, applications of PINNs in battery state estimation are reviewed, covering key tasks such as state of charge, state of health, degradation prediction, and the inference of internal states including temperature and ion concentration fields. Through comparative analysis, the advantages of PINNs in improving prediction accuracy, interpretability, and data efficiency are highlighted. The challenges related to model generalization, training stability, and computational cost are also discussed. Through comparative analysis, this review indicates that PINNs provide an effective bridge between physics-based battery models and data-driven learning by improving prediction accuracy, physical consistency, and data efficiency under limited measurement conditions. Nevertheless, challenges related to training stability, computational cost, and cross-condition generalization still restrict practical deployment. Future research is expected to focus on lightweight and transferable PINN frameworks, multi-physics integration, uncertainty-aware learning, and real-time battery management applications for next-generation intelligent energy storage systems.