Prediction of axial load capacity of rectangular concrete-filled steel tube columns using machine learning techniques

Abstract

This work aims to develop a novel and practical equation for predicting the axial load of rectangular concrete-filled steel tubular (CFST) columns based on soft computing techniques. More precisely, a dataset containing 880 experimental tests was first collected from the available literature for the development of an artificial neural network (ANN) model. An optimization strategy was conducted to obtain a final set of ANN’s architecture as well as its weight and bias parameters. The performance of the developed ANN was then compared to current codes (AS, EN, AIJ, ACI, AISC, LRFD, and DBJ) and existing empirical equations. The accuracy of the present model was found superior to the results obtained by others when predicting the axial load of rectangular CFST columns. For practical application, an explicit equation and an Excel-based Graphical User Interface were derived based on the ANN model. The graphical user interface is provided freely for all interested users, to support the design, teaching, and interpretation of the axial behavior of CFST columns.