Real-time high-fidelity generalizable model for temperature field reconstruction of lithium-ion battery

Accurate prediction of lithium-ion battery temperature distribution is critical for safe and reliable thermal management, but conventional methods often suffer from slow response, poor resolution, or limited generalizability. This study presents a real-time, high-fidelity, and generalizable model for temperature field reconstruction based on linear decomposition of transient multi-dimensional heat conduction. By representing the battery temperature history as a linear superposition of instantaneous impulse responses from discrete heat sources, only one single calibration test is needed to obtain whole-domain discrete Green's functions, and online reconstruction requires only one reference temperature to deduce heat generation rates. Numerical simulations demonstrate that across various charge and discharge rates, the reconstruction temperature field achieves average local errors below 0.1 K and relative errors within 1.5 %. By modifying the Green's functions based on Fourier number correlations, local reconstruction errors remain under 0.2 K for a ±15 % variation in thermal conductivity, ±30 % in heat capacity, ±15 % in density, and substantial changes in battery width or thickness, with no recalibration required. Experimental validations confirm that the proposed model achieves high reconstruction accuracy with average local errors below 0.4 K for different charge and discharge rates under various cooling conditions. The model exhibits low sensitivity to the choice of battery electrochemical heat generation models, which enhances its robustness and applicability in practical battery thermal management.