A hybrid model for predicting the axial compression capacity of square concrete-filled steel tubular columns

Abstract

Accurate prediction of axial compression capacity (ACC) of concrete-filled steel tubular (CFST) columns is an important issue to maintain the safety levels of related structures and avoiding failure consequences. This paper aims to develop a new framework for accurate estimation of the ACC for square concrete-filled steel tubular (SCFST) columns based on a novel hybrid artificial intelligence technique. Therefore, the radial basis function neural network (RBFNN) was used as a predictive model to solve this problem, whereas for optimum generalization and accurate prediction, a new optimization algorithm inspired by the firefly movement was proposed, namely the firefly algorithm (FFA). Besides that, other well-known optimization algorithms were used to compare the accuracy of the new-developed predictive model, namely Differential Evolution (DE) and Genetic algorithm (GA). In addition, a large database of 300 experimental tests was collected from the open published literature to train the new hybrid proposed models in terms of RBFNN-GA, RBFNN-DE, and RBFNN-FFA. Several comparative criteria were used to evaluate the robustness and accuracy of the new proposed model. The obtained performances were compared with the ones given from the artificial neural network (ANN) method based on the trial and error method. Results showed that the novel predictive model based on the hybrid RBFNN with FFA provides the highest efficiency and accuracy in terms of predictive estimations of the ACC for SCFST columns compared to ANN, whereas the novel RBFNN-FFA model enhances the prediction results by 28%, 37%, and 52% compared to RBFNN-GA, RBFNN-DE, and ANN, respectively.