Abstract
Lithium-ion batteries (LIBs) dominate the electric vehicle (EV) market, but highly transient operations, like extreme fast charging and regenerative braking, accelerate coupled degradation mechanisms. The most critical of these, solid-electrolyte interphase (SEI) growth and lithium plating, drive capacity fade and elevate internal resistance. Predicting these non-linear trajectories is essential for onboard Battery Management Systems (BMS). Currently, high-fidelity Doyle-Fuller-Newman (DFN) models capture these multiphysics pathways but are computationally expensive for extended simulations. Conversely, computationally efficient reduced-order lumped models rely on static parameters, unable to mechanistically distinguish between aging mechanisms. To bridge this gap, this study proposes a physics-informed, multi-scale reduced-order lumped aging modeling framework. Using a DFN model as a virtual lab, micro-scale SEI evolution, localized lithium plating and cell resistance growth are mapped onto a computationally lightweight lumped model. Validated on 21700 NMC/Gr prototype cells under extensive dynamic driving profiles (WLTP, US06, DST, and an extended Mixed profile), the proposed model achieved state-of-the-art voltage precision. Benchmarked directly against the high-fidelity DFN model, the framework maintained an RMSE below 0.50% across all tested cycles, while operating ~49 times faster and successfully completing long-term simulations where the DFN model became computationally unfeasible. This bottom-up methodology ensures high-fidelity, physics-informed state tracking without compromising the computational efficiency required for next-generation EV applications.