Ultimate axial capacity prediction of CCFST columns using hybrid intelligence models - a new approach

Abstract

This study aims to propose a new intelligence technique of predicting the ultimate capacity of axially loaded circular concrete-filled steel tube (CCFST) columns. A hybrid system based on one of the evolution algorithm - Genetic Algorithm (GA), fused with a well-known data-driven model of multivariate adaptive regression splines (MARS), namely G-MARS, was proposed and applied. To construct the MARS model, a database of 504 experimental cases was collected from the available literature. The GA was utilized to determine an optimal set of MARS' hyperparameters, to improve the prediction accuracy. The compiled database covered five input variables, including the diameter of the circular cross section-section (D), the wall thickness of the steel tube (t), the length of the column (L), the compressive strength of the concrete (f(c)'), and the yield strength of the steel tube (f(y)). A new explicit formulation was derived from MARS in further analysis, and its estimation accuracy was validated against a benchmark model, G-ANN, an artificial neural network (ANN) optimized using the same metaheuristic algorithm. The simulation results in terms of error range and statistical indices indicated that the derived formula had a superior capability in predicting the ultimate capacity of CCFST columns, relative to the G-ANN model and the other existing empirical methods.