Ensemble machine learning models for estimating mechanical curves of concrete-timber-filled steel tubes

Abstract

The use of concrete-timber-filled steel tubes (CTFSTs) as composite structural elements in buildings is gaining attraction among researchers due to their positive structural behavior and high load-bearing capacity. The combination of steel, concrete and timber materials improves energy absorption and ductility making CTFSTs a promising choice for modern construction. However, finding the mechanical properties of CTFSTs is a challenge during the design process due to the complexity of predicting behavior. Reliable modeling of these interactions is essential for an optimal design, which requires extensive experimental data and advanced computational methodologies. Therefore, this study proposed ensemble machine learning (ML) models for estimating load-displacement and stress-strain curves as well as the maximum axial capacity and elastic stiffness of CTFSTs. The results confirm the reliability of ensemble ML models for predicting the elastic stiffness and the maximum axial capacity of CTFST specimens with error percentages of 0.57 and 0.72, respectively. In addition, proposed ensemble ML models were used to estimate axial load-displacement and stress-strain curves of CTFSTs having different shapes of timber, which their curve fitting ability were superior compared to other ML models (i.e., accuracy of 97.6 %). Having ensemble ML models validated by experimental tests, a graphical user interface (GUI) tool is prepared for the preliminary evaluation of CTFST specimens, which can ease the way for reducing the experimental costs.