Reliability analysis of normal strength CFST rectangular columns through multiple approaches

Concrete-filled steel tube (CFST) rectangular columns have several benefits for structural engineering and buildings. The advantages of both steel and concrete are combined in these columns, creating a composite structural element with improved strength, ductility, aesthetics, seismic performance, construction efficiency, and cost efficiency. The current study aims to evaluate the mechanical performance of CFST rectangular specimens using an analytical approach that combines 3D-finite element analysis (3D-FEA) and artificial neural network (ANN) modeling. This methodology relies on a comprehensive database comprising 1045 sampling points from experimental studies. To simulate the response of CFST rectangular columns, an improved plasticity model for confined concrete was utilized in the 3D-FEA. This modeling considered various geometrical parameters and material features to accurately simulate the behavior of confined concrete and steel tubes. Comparison statistics were used to validate the 3D-FEA models against the experimental dataset of CFST specimens, confirming their accuracy. Furthermore, a parametric analysis was conducted using the 3D-FEA model to observe the potential impact of varying geometric and material-related variables on CFST specimen behavior. In addition to the 3D-FEA, a novel ANN model was developed to predict the axial strength of CFST specimens based on the experimental database. The comparison of R² values among the empirical model, ANN model, and 3D-FEA model indicated that the ANN estimates were 18% more accurate than the empirical model and 9% more accurate than the 3D-FEA model. When considering the sampling points from the established database, both the empirical technique and the 3D-FEA model showed a strong association, with an R² value of 0.842. However, the empirical and ANN-derived estimates exhibited an even higher correlation, with an R² value of 0.916, while the predictions of the 3D-FEA and ANN models showed a slightly lower but still significant correlation, with an R² value of 0.868.