Computer-Aided Engineering (CAE) is essential for accelerating product development, reducing costs, and enhancing quality. As optimization workflows become more common, the accumulation of simulation data opens the door for surrogate modeling, which offers faster predictions than traditional Finite Element Method (FEM) analysis. This study investigates the applicability of surrogate models through a case study on chain geometry optimization, targeting minimal mass and maximal strength. A model trained on 180 FEM samples was enhanced using region-specific stress predictions, additional features, and data augmentation. The improved model achieved a coefficient of determination (R²) of 0.763 for stress prediction. Integrating this model with a genetic algorithm enabled efficient multi-objective optimization, yielding solutions comparable in performance to those derived from full FEM analysis. Notably, computational time was reduced by approximately 84%. While constructing the surrogate model required significant effort, the results suggest its value in recurring or large-scale design tasks. Future work will explore integration with high-performance computing and open-source solvers to establish a scalable and rapid optimization framework for practical design implementation.
MORIMURA et al. (Wed,) studied this question.