Scholarlink

<p>A combined dataset of 216 cases was employed in this study to predict the ultimate axial strength (Pu) of concrete-filled corrugated steel columns (CFCSCs), The dataset integrates 130 finite element (FE) simulations, 25 newly developed ABAQUS models, and 61 experimental tests from the literature. It covers trapezoidal, sinusoidal, and helical corrugation profiles, with trapezoidal columns consistently exhibiting the highest strength under comparable conditions. The FE framework was validated against the experimental results, which included under both monotonic and cyclic loading scenarios. Data preprocessing involved scaling of continuous variables, checking multivariate outliers with Mahalanobis distance, and controlling collinearity through the Variance Inflation Factor. Nine regression algorithms were trained and compared under stratified cross-validation with hyperparameter tuning via GridSearchCV. CatBoost, k-Nearest Neighbors, and Gradient Boosting delivered the strongest predictions, and Analytic Hierarchy Process (AHP) ranked CatBoost as the preferred model. SHAP interpretation identified corrugation fold angle (θ) as the dominant factor influencing Pu, with robustness confirmed through bootstrap resampling and noise injection. From CatBoost outputs, a symbolic polynomial surrogate model (SPSM) was derived, achieving R² ≈ 0.9936. When applied to the experimental subset yielded R² = 0.9897 with RMSE = 47.87, closely matching the ML predictions (RMSE = 37.34). The resulting closed-form expression provides a transparent and design-compatible alternative to black-box ML models, offering engineers with a reliable predictive tool for CFCSCs under both monotonic and cyclic loading.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance behaviour of concrete-filled corrugated steel columns predicted using machine learning on FEM and experimental data

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Saket Rusia,
  • S. S. Mishra

摘要

A combined dataset of 216 cases was employed in this study to predict the ultimate axial strength (Pu) of concrete-filled corrugated steel columns (CFCSCs), The dataset integrates 130 finite element (FE) simulations, 25 newly developed ABAQUS models, and 61 experimental tests from the literature. It covers trapezoidal, sinusoidal, and helical corrugation profiles, with trapezoidal columns consistently exhibiting the highest strength under comparable conditions. The FE framework was validated against the experimental results, which included under both monotonic and cyclic loading scenarios. Data preprocessing involved scaling of continuous variables, checking multivariate outliers with Mahalanobis distance, and controlling collinearity through the Variance Inflation Factor. Nine regression algorithms were trained and compared under stratified cross-validation with hyperparameter tuning via GridSearchCV. CatBoost, k-Nearest Neighbors, and Gradient Boosting delivered the strongest predictions, and Analytic Hierarchy Process (AHP) ranked CatBoost as the preferred model. SHAP interpretation identified corrugation fold angle (θ) as the dominant factor influencing Pu, with robustness confirmed through bootstrap resampling and noise injection. From CatBoost outputs, a symbolic polynomial surrogate model (SPSM) was derived, achieving R² ≈ 0.9936. When applied to the experimental subset yielded R² = 0.9897 with RMSE = 47.87, closely matching the ML predictions (RMSE = 37.34). The resulting closed-form expression provides a transparent and design-compatible alternative to black-box ML models, offering engineers with a reliable predictive tool for CFCSCs under both monotonic and cyclic loading.