Integrating electrochemical-thermal-mechanical physical informed neural network framework for lithium-ion batteries state of health estimation

• A first-principles electrochemical-thermal-mechanical-aging coupled model is developed to generate internal aging states. • Internal degradation parameters are directly predicted using a physics-informed neural network framework. • Physical consistency is enforced by embedding the governing differential equations into the loss function of neural network as soft constraints. • The fusion of simulated internal states with measurable multiphysics external features enables accurate SOH estimation under sparse data conditions. Accurate estimation of the State of Health (SOH) for lithium-ion batteries remains challenging due to the complex interplay of electrochemical, thermal, and mechanical degradation mechanisms. This study proposes a Multihysics-Informed Neural Network (MPINN) framework that integrates physical laws with data-driven modeling to enable electrode-level SOH diagnostics. A high-fidelity electrochemical-thermal-mechanical (ETM) coupling model is developed to simulate key aging processes, including solid electrolyte interphase (SEI) growth, lithium plating, and particle fracture. The MPINN utilizes a total of six features, comprising three external signals namely voltage, cycle duration and initial cycle strain, and three internal physical parameters including SEI concentration, lithium deposition and active material loss. Notably, the internal parameters are derived from multi-condition cycling tests. A hybrid CNN-GRU/BiLSTM-Attention architecture captures both short-term dynamics and long-term dependencies in aging behavior. Physical consistency is enforced through custom loss functions derived from governing differential equations. Results demonstrate that the proposed MPINN achieves high-precision SOH estimation under unseen operating profiles, with mean absolute errors below 0.5% in capacity fade prediction, even with sparse measurement data. This approach not only improves estimation accuracy but also provides interpretable insights into degradation mechanisms, offering a powerful tool for battery management and lifetime prediction.