Data-driven prediction and cost-effective design optimization for concrete-filled double-skin steel tubes

Concrete-filled double-skin steel tubes (CFDST) exhibit superior mechanical properties, fire resistance, and seismic ductility. However, current design codes and empirical models often fail to provide accurate predictions of the ultimate load capacity (P_u). While existing prediction methods provide relatively accurate predictions, there is still room for improvement in precision. Moreover, research optimizing both P_u and cost remains limited. To bridge this research gap, this study introduces an ultra-high-precision prediction method (R² = 0.996) that integrates the Bat Algorithm (BA) and CatBoost model. Moreover, as a complementary approach, a prediction formula was developed based on prior research and further optimized using the Genetic Algorithm (GA), providing a rapid estimation of P_u (R² = 0.941). Both models exhibit higher predictive accuracy than three design codes and four empirical formulas. For design optimization, the BA is employed to design the lowest-cost configurations that meet specific P_u thresholds. Additionally, the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) is applied to balance the conflicting objectives of maximizing P_u and minimizing cost, effectively generating the Pareto front. Lastly, web applications were developed to enable real-time CFDST performance prediction and optimized low-cost design solutions.