Physics-informed neural networks for battery modeling and state estimation: A methodological review

Lithium-ion batteries (LIBs) are central to modern electrified transportation and energy storage systems, yet accurate battery modeling and state estimation remain challenging because of the strong nonlinearity, time-varying behavior, and multi-physics coupling of electrochemical processes. Conventional physics-based models provide mechanistic interpretability but are often computationally intensive, whereas purely data-driven methods offer flexibility at the expense of physical consistency and generalization under unseen operating conditions. In recent years, physics-informed neural networks (PINNs) have become a promising paradigm that bridges these two approaches by embedding governing physical laws into neural network training and architecture design. This review systematically examines PINN methodologies for battery modeling and state estimation from a methodological perspective, with particular emphasis on the sources of battery-domain physical constraints and the corresponding strategies for embedding such knowledge into learning frameworks. It further summarizes the major applications of PINNs in battery management, including state of charge (SOC) estimation, state of health (SOH) and remaining useful life (RUL) prediction, thermal field reconstruction, parameter identification, and reconstruction of unobservable internal electrochemical states. Key challenges for practical deployment are then discussed, including multi-objective optimization, training stability, domain generalization, uncertainty quantification, interpretability, and lightweight implementation. Finally, future directions are highlighted for developing more trustworthy, explainable, and deployable PINN-enabled battery digital models and intelligent battery management systems (BMS).