• Experimental characterization of six commercial lithium iron phosphate cells • Benchmarking of five equivalent circuit models of increasing physical fidelity • Novel physics-informed expressions derived for model parameters • First-time integration of a multi-particle phase-change model for LFP resistance • Parameterization and validation of the models using full cycle and pulse test data Lithium-ion batteries with lithium iron phosphate (LiFePO₄, LFP) positive electrodes and graphite negative electrodes exhibit complex voltage behavior, including flat, staged voltage curves at low C-rates, open-circuit voltage hysteresis, and asymmetric overpotentials, which challenges established modeling approaches. To address these complexities, six commercial LFP/graphite cells (nominal capacities between 3.4 Ah and 180 Ah) were experimentally characterized using constant current constant voltage cycling and pulse tests. Consistent self-similar voltage behaviors were observed across all cells. To capture these behaviors, a series of five physics-informed dual-electrode equivalent circuit models was developed, progressively increasing in complexity. These models integrate electrode-specific voltage sources, hysteresis, and stoichiometry-dependent resistances based on Butler-Volmer kinetics. For the first time, multi-particle phase-change behavior of LFP is integrated into a resistor element. Slow dynamic effects were modeled via either core-shell Fickian diffusion or additional resistor-capacitor (RC) elements. Models were parameterized and validated against experimental results over a range of C-rates (0.02-1 C) and temperatures (5-35°C). Results demonstrate that including physicochemical insights is critical to reproducing low-current behavior and asymmetric overpotentials. The most advanced model, featuring hysteresis and three RC elements, successfully captures behavior across all conditions. The physics-informed modeling introduced here allows higher fidelity at a reduced number of parameters compared to state-of-the-art equivalent circuit models.
Braun et al. (Wed,) studied this question.