Predicting axial compression capacity of CFDST columns and design optimization using advanced machine learning techniques

Abstract

Concrete-filled double-skin steel tubular (CFDST) columns are fundamental in civil engineering, known for their exceptional mechanical properties. Their load-bearing capacity, influenced by factors such as geometry, material properties, and loading conditions, is a critical aspect of CFDST column design. This study introduces an innovative approach that leverages the synergy between gradient-boosting regression (GBR) and three metaheuristic optimization algorithms: sine-cosine-based swarm optimization (SCSO), grey wolf optimization combined with whale optimization (GWO_WOA), and artificial rabbits optimization (ARO). Its primary aim is to accurately predict the axial compression capacity (ACC) of CFDST columns for optimal design. Using a comprehensive database of 153 empirical results from CFDST columns subjected to axial loads, the proposed model is rigorously compared with eight alternative machine learning models, three established design standards, and two empirical equations to validate its robustness. The results highlight the superiority of the finely tuned GBR model, optimized with the ARO algorithm, over peer models and conventional benchmarks. This framework also serves as the foundation for optimizing the CFDST column design by identifying the optimal dimensions that maximize ACC while meeting the design constraints. Clearly, this model can significantly improve CFDST column design, thereby benefiting structural engineers and designers striving to enhance the safety and efficiency of CFDST structures.