ANN-Based Axial Strength Prediction of Short Columns with Double and Bar-Reinforced Concrete-Filled Steel Tubes Subjected to Concentric and Eccentric Loading

This study is aimed to predict axial compressive strength of circular concrete-filled double steel tube (CFDST) and reinforced concrete-filled steel tubular (RCFST) column subjected to concentric and eccentric loading using artificial neural network (ANN). In this context, initially an extensive literature review was carried out to compile a comprehensive database related to CFDST and RCFST column. Three separate databases were compiled related to (1) CFDST column under concentric loading, (2) CFDST column under eccentric loading, and (3) RCFST column under concentric loading. Neural network toolbox in MATLAB was used to develop and train ANN models. Key ANN features like network type, activation function, number of hidden layers, epochs and number of neurons in each layer were finalized after optimizing regression value. Additionally, sensitivity analysis was performed to understand the influence of various geometric and material parameters on the design strength. It was observed that geometric configurations like diameter of outer and inner steel tubes in case of CFDST and thickness, height, diameter in case of RCFST significantly influenced the axial strength as compared to material grade of concrete and steel tubes. The mean for the ratio of predictions and experimental observations were 0.962 (for CFDST) and 0.981 (for RCFST). Also, standard deviation for such ratio were 0.082 for both specimen type. Predicted results by ANN depicted that axial strength obtained using present formulation are coherent and even sometimes closer to experimental evidences compared to empirical type predictions from several design codes.

Automated summary

This study predicts the axial compressive strength of CFDST and RCFST columns under concentric and eccentric loading using ANN. Three separate databases were compiled based on extensive literature review. ANN models were developed and trained, and sensitivity analysis was performed to understand the influence of various parameters on design strength. The predicted results show that the present formulation has good agreement with the experimental evidence.