Physical information mining from three-representative voltage curves for high-fidelity digital modeling of lithium-ion batteries

Physics-based digital modeling is increasingly becoming a vital tool for accelerating battery technology innovation. Yet, existing methods often require extensive datasets and still struggle to achieve rapid and accurate decoupling of geometric, thermodynamic and kinetic parameters, limiting model generalization across diverse scenarios. Here, we propose a physics-informed parameter identification framework that extracts the complex electrochemical–thermal coupling relationships of lithium-ion batteries (LIBs) from three-representative voltage curves, achieving both efficiency and fidelity. First, electrode balancing information is obtained by inverse identification from OCV curves, laying the foundation for the next modeling. Second, a multi-scenario sensitivity analysis reveals that alternating external stimuli trigger distinct parameter sensitivities, indicating an effective pathway for partially decoupling key parameters. Inspired by these findings, a hierarchical identification framework is established to progressively extract parameters, effectively mitigating overfitting, local minima, and slow convergence. Finally, the extracted parameter set is validated against experimental data under 10 diverse operating conditions. The calibrated model achieves an average MAE of 27.5 mV for voltage and 0.65 °C for temperature rise, exhibiting exceptional accuracy and robustness. Moreover, it successfully captures the nonlinear transition in long-term cyclic aging from linear capacity fade to accelerated degradation and exhibits strong cross-dimensional transferability. This work develops an efficient offline parameter extracted framework based on data–physics fusion, enabling high-fidelity digital modeling of LIBs and offering good scalability to next-generation chemistries.