Accurate estimation of the State of Health (SOH) of lithium-ion batteries is essential for the safe operation and lifecycle management of battery systems in electric vehicles and energy storage applications. Existing approaches face significant limitations: physics-based models are computationally expensive and require chemistry-specific parameterization; data-driven methods offer high accuracy but tend to be dataset-specific and lack physical consistency, leading to poor generalization across different operating conditions. This study develops an Enhanced Physics-Informed Neural Network (PINN) framework that integrates data-driven learning with multiple physical degradation constraints to achieve accurate, stable, and generalizable battery SOH estimation. Building upon a dual-network architecture comprising a Solution Network and a Dynamic Network, the proposed framework introduces a multi-physics enhancement strategy incorporating three physically meaningful degradation factors—temperature effects modeled via Arrhenius relationships, capacity degradation laws, and charging efficiency dynamics—through dedicated expert subnets. Experiments on the NASA PCoE battery aging dataset demonstrate that the Temperature-Enhanced PINN reduces MSE by 36.68% relative to the baseline PINN, while the Capacity-Enhanced PINN achieves a 42.13% MSE reduction. These results, validated through 30 independent training runs, demonstrate that embedding physically meaningful degradation mechanisms significantly improves both predictive accuracy and training stability. The optimized framework achieves RMSE of approximately 1.64% and MAE of approximately 1.32%, suitable for practical battery management system deployment.

Keywords

Physics-Informed Neural Network, Battery State of Health, Lithium-Ion Battery, Machine Learning and Degradation Prognosis.