Crushing safety prediction of a battery pack system by fusion of multiple methods

Designers are tasked with creating a dependable battery pack system (BPS) to safeguard the property of consumers and the safety of passengers. However, the intricate architecture of BPS necessitates significant computational resources when employing conventional finite element model (FEM) for design purposes. The use of multibody system dynamics (MSD) and machine learning (ML) methodologies may assist developers in the efficient design of reliable BPS. In this work, by integrating FEM and MSD approaches, a two-step mechanical safety assessment model for BPS had been developed. This model was developed to assess the mechanical safety of the BPS under a range of conditions, including different crushing speeds, crushing positions, SOC, and the thickness of both the bottom shell and upper enclosure of the BPS. The application of experimental design methodologies in conjunction with ML models enables the efficient and precise prediction of the mechanical safety of BPS, even when utilizing a limited sample size. Four ML techniques — deep neural network (DNN), particle swarm optimization-radial basis function, bald eagle search-extreme learning machine, and random forest (RF) — were applied to develop predictive models for the mechanical safety of the BPS under various crush conditions. The predictive accuracy and stability of the four ML models were evaluated, and the interrelationships among different variables and outputs were analyzed. The DNN model demonstrated the highest accuracy of 0.9943, while the RF model demonstrated the lowest prediction accuracy of 0.9086. This study demonstrates a reduction in computational costs by two-thirds when compared to conventional FE methods. Furthermore, it enhances computational efficiency by over 90% in the context of mechanical safety predictions for BPSs. This methodology enhances the efficient design of BPSs and provides valuable insights for the swift prediction of mechanical safety in BPS, taking into account the influence of SOC on the mechanical behavior of the cell.