Abstract:
This study benchmarks three surrogate‐based metamodeling techniques—Successive Linear Response Surface Methodology, feed‑forward Neural Networks, and Kriging—within a parametric LS‑OPT/LS‑DYNA framework for crashworthiness shape optimization. Parametrized crash‐box and full‑vehicle geometries are explored under nonlinear dynamic loading, comparing convergence speed, predictive accuracy, and robustness against a sequential random search baseline. Results show that despite greater initial sampling for Neural Networks and Kriging, all three metamodels attain comparable optimization efficiency and enable rapid design‐space exploration, making them powerful tools for parametric shape optimization in complex crashworthiness design.
