A fast crashworthiness assessment framework: Sectional force-based multi-stage physics informed surrogate model

• A Sectional Force-Based Multi-Stage Physics-Informed Surrogate Model is proposed. • Framework decouples local beam mechanics from global crash dynamics for efficiency. • Physics-informed ensemble (XGBoost-CNN-Transformer) enhances small-sample accuracy. • Achieves near-FEA fidelity with 0.038 s evaluation, five orders faster than simulation. • Eliminates manual re-meshing, enabling scalable real-time crashworthiness assessment. Finite-element analysis (FEA) is the benchmark for crashworthiness evaluation, yet its prohibitive computational cost and labour-intensive re-meshing make it unsuitable for iterative structural optimization. Although surrogate models offer partial relief, they still demand large simulation datasets and frequent mesh updates. Focusing on the vehicle front crash condition, proposing a Sectional Force-Based Multi-Stage Physics Informed Surrogate Model (SFB-MSPISM) that integrates sectional force features with semi-empirical physical priors (i.e., the Gérard buckling formula) in a two-stage architecture, thereby reducing the training data requirement and virtually eliminating manual re-meshing. In Stage-1, a Physics-Informed XGBoost-CNN-Transformer ensemble (PI-XCT) is proposed to predict the peak sectional force and energy absorption of five key beams; In Stage-2, a multi-output XGBoost regressor is proposed to estimate the maximum crash acceleration and the Toe-board intrusion. Trained on fewer than 130 high-fidelity simulations (100 data for Stage-1 and 26 data for Stage-2), SFB-MSPISM attains a coefficient of determination of 0.97 for peak deceleration and a mean intrusion error of 2.525 mm (≤5 %), while reducing per-design evaluation time from 6.5 h to 0.038 s. These results show a speed-up exceeding five orders of magnitude and virtually eliminate human intervention, thereby enabling millisecond-scale, physically consistent crashworthiness assessment for rapid design exploration.