Prediction of mechanical-electrical-thermal behavior in lithium-ion batteries using a neural network surrogate model

Lithium-ion batteries (LIBs) are highly vulnerable to mechanical abuse, which can induce structural damage and internal short circuit (ISC), accompanied by a pronounced temperature rise and increased thermal runaway risk. To enhance predictive safety and early warning capabilities, a neural network surrogate model (NNSM) is developed for LIBs in this work. A coupled mechanical-electrical-thermal model was developed under quasi-static loading. Combined experimental and simulated data were used to train a deep neural network (DNN). The trained model efficiently predicts the mechanical, electrical, and thermal behaviors of LIBs. The results show that the proposed approach can accurately reproduce the coupled responses under compression loading. The deformation evolves from casing support to internal densification, and an ISC occurs at a critical displacement, leading to a rapid voltage drop and the rise of temperature. The peak temperature reaches about 79°C at high state of charge (SOC) versus 23°C at low SOC, while the maximum stress increases to 82 MPa at 0°C. Higher SOC and elevated temperature intensify electro-thermal coupling and increase thermal runaway risk, whereas lower temperatures enhance stress concentration and promote brittle fracture. This study will provide an effective approach for safety prediction of LIBs.