Abstract
The trend toward increasing energy density in lithium-ion batteries (LIBs) makes thermal safety more critical. However, traditional adiabatic calorimetry fails to simulate flame exposure in real fire incidents. Herein, a framework of the thermal runaway (TR) temperature prediction of LIBs is proposed using experimental and virtual data for extreme high-temperature shock conditions. Specifically, high-temperature shock wave-induced TR tests are conducted on LiNi(0.5)Co(0.2)Mn(0.3)O(2) (NCM523) and LiFePO(4) (LFP) batteries to compare combustion behavior and TR characteristics, while obtaining realistic temperature data. A 3D conjugate heat transfer and TR coupling model is established, and characteristic temperature parameters of TR under this condition are extracted for comparison and validation of accuracy with experimental data. The simulation model further enriches TR data under different states of charge (SOC) and distances from the heat source. The virtual data generated by the simulation model compensates for insufficient TR experimental data, enabling the establishment of a data-driven prediction model. Three different deep learning models are compared to predict the trend of TR temperature variations under different heat source distances and SOC conditions. The results indicate that the proposed framework, which combines experimental and virtual data, achieves high-fidelity, fast-response TR temperature predictions. The optimal framework maintains a mean absolute percentage error (MAPE) below 5% across all studied conditions.