Machine learning–driven multilattice optimization of an electric vehicle battery box side profile

Abstract Reducing the weight of structural profiles in electric vehicle battery systems is a critical design objective for improving vehicle range. This study aimed to obtain an optimal design for the battery box side profile using lightweight lattice structures that provide sufficient mechanical strength. In this context, an L-shaped battery profile was redesigned, and a dataset of design combinations was created by placing body-centered cubic (BCC), Kelvin (K), and octagon (O) lattice structures within 25 cells. The mechanical behavior of the 300 designs was analyzed under static loading using the finite element method (FEM). Meta-models were developed using the obtained results to predict maximum von Mises stress and displacement values. For this purpose, multilayer perceptron (MLP), random forest regression (RFR), and gradient boosted regression (GBR) models were compared, and the GBR model, which provided the highest prediction accuracy, was selected. The developed meta-models were integrated into the genetic algorithm (GA) to reduce analysis time. As a result of the optimization, a weight reduction of approximately 48.6 % was achieved without exceeding the material’s yield strength. The results show that the proposed approach provides high prediction accuracy and enables similar lightweight structural components to be optimized quickly and effectively using different lattice structures.