Akademik Veri Yönetim Sistemi
Journal of Engineering Research (Kuwait), 2026 (SCI-Expanded, Scopus)
Surrogate modeling has emerged as an effective alternative to computationally expensive finite element analyses in nonlinear structural engineering problems. This study presents a systematic comparative evaluation of five widely used surrogate models (Kriging, Random Forest, XGBoost, Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP)) for predicting the nonlinear mechanical response of an aircraft landing gear bracket subjected to high loads. Using parametric finite element analyses, a high-dimensional dataset was generated, comprising six different training sets ranging from 500 to 5000 samples. All models were trained under identical conditions, with computational cost taken into account, and were evaluated using multiple performance metrics, including RMSE, MAE, MAPE, and the coefficient of determination (R²). The obtained results indicate that prediction accuracy improves consistently for all models as the dataset size increases. Among the evaluated methods, the MLP model achieved the highest accuracy, with an RMSE value of 12.01 MPa and an R² value of 0.9803, when trained on 3000 samples, while maintaining a reasonable training time. The Kriging model, on the other hand, demonstrated strong interpolation capability particularly for small datasets, achieving an R² value of 0.9722 with only 500 samples; however, its computational cost increased significantly for larger datasets. Tree-based models, namely XGBoost and Random Forest, exhibited balanced and stable performance across all dataset sizes, whereas the SVR model showed comparatively lower accuracy and higher sensitivity to data volume. Additional validations conducted under alternative loading conditions confirmed the robustness and generalization capability of the best-performing models. This study provides a quantitative guideline for selecting appropriate surrogate models in nonlinear structural design problems. It elucidates the trade-off between accuracy, robustness, and computational efficiency within surrogate-assisted engineering workflows.