Axial Capacity of Rectangular Concrete-Filled Steel Tube Columns Using Artificial Neural Network

Abstract

In this study, artificial neural network analysis is introduced to aid in estimating the axial capacity of concrete-filled steel tube stub columns. A wide range of experimental database of rectangular concrete-filled steel tube columns available in the literature is used here in order to construct and verify the model. The Levenberg-Marquardt algorithm is employed in this back-propagation training procedure. The present artificial neural network model has been created using six input neurons that have different geometric and material properties of the column including the length of the column, the width and the depth of the section, the thickness of the steel tube, the compressive strength of the concrete core, and the yield strength of the steel tube. Six hidden neurons have been employed, in addition to one output neuron that gives the ultimate capacity of the column. The verification process has showed that the present model is very accurate in predicting the ultimate capacities of the columns. This model is tested against other methods found in codes and standards and a method that has been previously proposed by the authors. With the limitations set by the codes and standards, the present model can be used safely and with sufficient accuracy. A parametric study using the artificial neural network model has been conducted to explore the effect of different parameters on axial capacities of the composite columns. The results have showed that careful optimizing techniques should be followed in order to get the most efficient sections.