Structural Optimization with NSGA-II Using a Multi-Output Surrogate Model Trained on Large-Scale Finite Element Data
ACTA MECHANICA SOLIDA SINICA, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s10338-026-00822-w
- Dergi Adı: ACTA MECHANICA SOLIDA SINICA
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Compendex, ICONDA Bibliographic, INSPEC, The International Construction Database (ICONDA), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Gazi Üniversitesi Adresli: Evet
Özet
This study presents an integrated optimization framework that combines parametric computer-aided design (CAD) modeling, finite element analysis (FEA), machine learning-based surrogate modeling, and multi-objective evolutionary optimization to address the high computational cost of conventional design optimization. A multi-output multi-layer perceptron (MLP) surrogate model was developed using a dataset of approximately 5,000 high-fidelity parametric CAD-finite element simulations spanning a wide design space. Unlike many existing studies, the proposed MLP simultaneously predicts two conflicting engineering objectives (volume and maximum von Mises stress) within a single network architecture. The model achieved high predictive accuracy, with R2 values of 0.996 for volume and 0.992 for von Mises stress, demonstrating robust representation of complex stress behavior. The trained surrogate model was directly embedded into the NSGA-II algorithm as the evaluation function, replacing computationally expensive finite element simulations. This integration enabled rapid evaluation of candidate designs in milliseconds, allowing the optimization process to be completed within practical engineering time scales. The resulting Pareto front consistently captured the trade-off between volume and stress while satisfying safety factor constraints. Independent FEA validation of selected optimal designs confirmed the physical consistency and reliability of the proposed framework.