An intelligent data-driven multi-physical field coupling modeling framework integrating experimental data, multi-scale simulation and machine learning is proposed to accurately describe the electrochemical-thermal-mechanical coupling behavior of lithium battery materials under complex working conditions. Aiming at the problems of parameter drift and lack of local field information in the dynamic aging process of traditional models, a three-layer architecture of "physical core+intelligent engine+system integration" is constructed: Based on the pseudo-2D (P2D) model, a Gaussian process regression proxy model is introduced to dynamically correct the key aging parameters, and the internal state variables that are difficult to observe are inverted by combining with the physical information neural network (PINN), so as to realize the cross-scale high-fidelity simulation with two-way coupling and step-by-step iteration. Experimental verification was conducted using NMC532/graphite soft pack batteries, and HPPC, EIS, and in-situ XRD tests were carried out at 0-45 ° C and 0.2C-2C operating conditions. The results showed that the voltage prediction error of the hybrid model was as low as 8.7 mV and 12.1 mV in the 25 ° C and 0 ° C HPPC tests, respectively, which was significantly better than the traditional physical model (52.3 mV) and the pure data-driven model (15.8 mV); At the same time, the lithium concentration gradient and stress concentration on the surface of negative particles were accurately captured, and the risk of lithium evolution at the end of 2C fast charging was successfully predicted, which was consistent with the disassembly observation. The framework has both physical interpretability and data adaptability, and provides an efficient tool for virtual screening of battery materials, algorithm optimization of battery management system (BMS) and safety early warning.
Lei Qian (Sun,) studied this question.