Analysis and prediction of mechanical response of cylindrical lithium batteries under compression conditions

Abstract To investigate the safety of electric vehicle battery packs under mechanical abuse and mitigate hazards such as short circuits, thermal runaways, fires, and explosions caused by collisions, a compression study is conducted on single lithium batteries. Flat plate compression tests and numerical simulations were performed on 18650 cylindrical lithium batteries. Both experiments and simulations captured the maximum compression failure displacement corresponding to the peak compression force. The correlation between the mechanical deformation and changes in voltage and temperature under mechanical loading was elucidated. Moreover, the verified finite element model was employed to generate compression force datasets for various lithium battery design parameters. Based on this, a hybrid prediction model, named CNN-BiLSTM-Attention, which integrates convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM), and an attention mechanism, was developed. Compared to three other relevant models, the proposed CNN-BiLSTM-Attention model demonstrated better prediction performance. It accurately predicts the load state, failure displacement, and maximum compression force of single lithium batteries, highlighting its potential for application in predicting mechanical responses and failure displacements in lithium battery systems. Furthermore, the model can support the development of early warning systems for compression-induced failures in battery packs subjected to mechanical abuse.