Application of hybrid ANN models to the design of CFDST columns subjected to concentric compressive loading

This paper develops hybrid metaheuristic-based optimization methods for the estimation of the compression capacity of Concrete Filled Double Skin Steel Tubes (CFDSTs) columns under compression using Artificial Neural Networks (ANNs). With the proposed models, the weights, biases, and hidden layer size of the ANN are simultaneously optimized by using Artificial Bee Colony (ABC) optimization and Teaching Learning-Based Optimization (TLBO) algorithms, called ABC-ANN and TLBO-ANN models, respectively. A dataset containing 167 experiments reported in the literature is adopted to construct the models. It is found that both the ABC-ANN and TLBO-ANN methods are efficient in training predictive ANN models for the class of problems considered. The efficiency of the proposed models is demonstrated through the good comparisons with design standards and empirical formulae. A user-friendly Graphical User Interface (GUI) software based on the proposed models is built to conveniently estimate the axial compression capacity of CFDST columns. By performing reliability analyses using Monte Carlo simulations, the strength reduction factors are suggested to ensure the GUI program applicable for practical design applications. Finally, an optimization procedure is developed based on the proposed ABC-ANN model to determine the optimal design of CFDST columns.