MACHINE LEARNING-BASED INVERSE DESIGN OF ECCENTRICALLY LOADED RECTANGULAR CONCRETE-FILLED STEEL TUBE (R-CFST) COLUMNS

Abstract

Concrete-filled steel tubes (CFST) have been used in structural engineering due to the advantages of the ductility of steel and strength of concrete. While a large number of research studies are available for estimating the ultimate load capacity of the CFST cross-sections, the inverse problem of designing the optimum cross-section for a given load, particularly for rectangular sections, has not yet gained much attention in the literature. In this study, based on the experiments, two regression models of Random Forest (RF) and XGBoost are developed for estimating the optimal width, depth and thickness of R-CFST short columns when subjected to an eccentric load. A total of 377 specimens of R-CFST short columns with various steel grades, concrete grades and eccentricity ratios are used in the dataset for developing the regression models. The direct geometric synthesis method is proposed as an alternative to the conventional trial-and-error design check methods in codes of practice. The ensemble machine learning regression models are preferred in this study as they have an edge over conventional design formulas provided in codes, as they can account for the non-linear relationships between load effects and member buckling. The structural engineers can utilize the proposed method to determine the required cross-section of R-CFST short columns quickly, with minimal amount of materials and to reduce the time involved in manual calculations. Therefore, the proposed method will assist in saving time, enhance the structural design processes and contribute towards sustainable constructions.