Shape Optimization of an Automotive Rubber Bumper Using a Surrogate Model-Based Parameter Tuning of a Genetic Algorithm

Abstract

This thesis focuses on the shape optimization of an automotive rubber bumper using a computational method based on genetic algorithms. A canonical Genetic Algorithm was implemented in MATLAB and tuned through benchmark tests and hyperparameter analysis. To improve accuracy and reduce computational cost, the GA was hybridized with the Nelder–Mead algorithm and supported by a surrogate model created using Support Vector Regression. The surrogate model approximated the objective function derived from finite element simulations of the rubber bumper, modeled in ANSYS with a Mooney–Rivlin hyperelastic material. A semi-automatic MATLAB–ANSYS interface was developed due to software limitations, enabling real FEA-based objective evaluations. The hybrid method successfully identified the optimal geometric parameters with high accuracy. The results show that the developed optimization framework is effective for nonlinear, simulation-driven engineering design problems.