Unified artificial neural network, linear and nonlinear statistics methods for forecasting the capacity of axially loaded concrete-filled steel tube (CFST) members

Abstract

Due to the fast development of technology that facilitates human work, there is an urgent need to use artificial intelligence as a new and accurate technology in Structural Engineering to design structural members. This paper presents unified statistical linear regression (LR) and nonlinear regression (NLR) analysis using the statistical package for the social sciences (SPSS) and unified advanced design methods of artificial neural network (ANN) using MATLAB for forecasting the capacity of the composite concrete-filled steel tube (CFST) columns with different cross-sections under axial loads. Overall, test data of about 390 specimens were collected, including 13 specimens used to validate the mentioned methods. The unified ANN method gives the best conservative and save results with an average value of the test to ANN analysis result of 1.01, followed by the statistical NLR analysis method with an average value of the test to NLR analysis result of 0.992, and then the statistical LR analysis method with an average value of the test to LR analysis result of 0.971. The results of the proposed method based on statistical LR analysis have shown negative values for the predicted ultimate capacity, which means this method does not capture the ultimate capacity behavior well. Validating the accuracy of the predicted analysis results with four available design codes has been done. The ACI 318–19 and DBJ 13–51-2010 gave conservative and save predictions with average test values to calculate results equal to 1.115 and 1.126, respectively; the unified ANN model provides the most accurate assessment predictions compared to available design codes.