Electro-Thermo-Mechanical Simulation and Machine Learning for Temperature Prediction of Lithium-Ion Batteries under Mechanical Abuse

Lithium-ion battery safety under mechanical abuse has become a critical challenge with the widespread adoption of electric vehicles. This study proposes a predictive framework combining multi-physics finite element simulation and machine learning to estimate the temperature rise of lithium-ion cells under impact conditions. An Electro-Thermo-Mechanical (ETM) coupled model was established in LS-DYNA to simulate the effects of impactor radius, velocity, and ambient temperature on internal heat generation. Using a full factorial sampling design, 125 simulation scenarios were generated to extract maximum temperature data. These data were used to train and compare several regression models, including Support Vector Machines (SVM), Decision Trees (DT), Back Propagation Neural Networks (BPNN), and Random Forests (RF). A Stacking ensemble model integrating these base learners achieved the highest prediction accuracy, with an R² of 0.996 and RMSE below 0.5. Performance remained robust even outside the original design domain, with prediction errors under 5% in 93.1% of test cases. The results demonstrate the effectiveness of integrating machine learning with physics-based modeling for reliable, data-efficient prediction of battery behavior under abusive conditions, offering new insights into battery safety design and real-time risk assessment.