NUMERICAL ANALYSIS OF FLEXURAL BEHAVIOUR OF CONCRETE FILLED STEEL TUBES(CFST) WITH TWO DIFFERENT CONCRETE INFILL

Abstract

The main goal of this research is set out to study the flexural behaviour of concrete filled steel tubes (CFST) giving insight to two different concrete infills and their behaviour in different parametric combinations. A numerical study is done on high strength concrete filled CFST and geopolymer filled CFSTs. Separate models were used for both high strength concrete and geopolymer concrete. The numerical investigation is done using ANSYS software. The Drucker Prager (DP) model is used for modelling the concrete. An extensive parametric study was performed to investigate the influences of depth-to-thickness ratio (28.57−200), effect of D/B ratio (24-42.8) and steel yield strengths (235–420 MPa) on the fundamental behavior of CFST beams under flexure load. When the D/B ratio increased from 0.54 to 2, the ultimate capacity of members is increased by more than 50%. In the analysis, the geopolymer filled CFST was able to take more load as compared with the high strength concrete. It can be seen that by increasing D/t ratio from 28.57 to 200 (7 times), the ultimate bending moment capacity reduced from 207.8 to 21 kNm (almost 8 times) for square CFST beams. The ultimate flexure strength of square CFST beams was found to increase significantly with an increase in the yield strength of steel. By increasing the yield strength of steel from 235 MPa to 420 MPa, the ultimate flexure load of the square CFST beam was found to increase. The developed models were also analyzed under cyclic loading condition. An artificial neural network prediction model was developed to predict the flexural strength of the CFST models. This study aims to establish a comparison between Multilayer Feed forward - a Multilayer Perceptron network (MLP) with feed forward learning - and a Radial Basis Function Network (RBF). From the results, it was observed that the two types of networks (Multilayer Perceptron and Radial Basis Function) were able to predict the results satisfactorily. However, the Multilayer Perceptron network showed higher accuracy than the radial basis function. The R² value of MLP analysis was found to be better than RBF