Designing new hybrid artificial intelligence model for CFST beam flexural performance prediction

Abstract

A substantial number of experimental studies have reported on the flexural performance of concrete-filled steel tube (CFST) beams. Due to the problem complexity, theoretically modeling of the flexural bending capacity (Mu) and the flexural stiffness at the initial and serviceability limits ( $$K_{{\text{i}}}$$ and $$K_{{\text{s}}}$$ ) of CFST beams remains challenging mission in the structural engineering field. Hence, this research proposes new numerical models for modeling the flexural capacities (Mu, $$K_{{\text{i}}}$$ , and $$K_{{\text{s}}}$$ ) of CFST beams under static bending load. For this purpose, numerous existing experimental and numerical results of CFST beams are collected for developing a new numerical model called as hybridized artificial neural network (ANN) model with particle swarm optimization (PSO) algorithm. The results of the proposed model validated against the existing results of CFST beams tested over the literature. In addition, PSO–ANN model verified with those obtained by the existing standards and approaches (EC4, BS5400, AISC, AIJ, and others) for the same corresponding beams. The proposed PSO–ANN model confirmed its capability to be used as an alternative theoretical approach to predict the flexural strength and stiffness capacities of CFST beams. The PSO–ANN model achieved mean values of about 0.933–0.989 with a coefficient of variation ranged from 4.98 to 9.53% compared to the existing results that obtained by others.