Protecting the power battery pack during side-pole impact poses a critical challenge for new energy vehicle (NEV) passive safety, where traditional methods struggle to balance computational efficiency and predictive accuracy due to strong nonlinearity. This study proposes a multi-stage optimization framework integrating the equivalent static load (ESL) method for nonlinear topology optimization and machine learning-based surrogate modeling. A multi-cell sill beam configuration is first developed via ESL-based topology optimization with deformation mode correction to address localized buckling. Subsequently, independent XGBoost models for intrusion, energy absorption, and mass are constructed, with their hyperparameters efficiently tuned using the Optuna framework with the Tree-structured Parzen Estimator (TPE). Results demonstrate that the optimized XGBoost model achieves superior prediction accuracy (R2 > 0.96) compared to conventional models (SVR, RSM, GPR, and RBF), with RMSE approximately 40% lower than the second-best model (GPR). The Non-dominated Sorting Genetic Algorithm III (NSGA-III) is then employed for tri-objective optimization, and the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) selects an optimal design, which is validated by finite element simulation with high agreement (maximum prediction error < 3%). Compared to the baseline design, the optimized sill beam achieves an 11.44% reduction in maximum intrusion, a 9.86% increase in energy absorption, and a 13.45% reduction in structural mass, demonstrating the effectiveness of the proposed data-driven crashworthiness optimization framework.
Li et al. (Fri,) studied this question.