Axial Strength Prediction of Square CFST Columns Based on The ANN Model

Abstract

Due to numerous advantages, concrete-filled steel tubular (CFST) columns have an increasingly important role in the civil engineering industry. Because of the expensive experimental testing of these members, it is beneficial to provide prognostic models. In this study, an artificial neural network (ANN) model for predicting the axial compressive strength of square CFST columns has been developed. A dataset of 1022 samples (685 stub columns and 337 slender columns) was collected from available literature in order to compare the accuracy of the fast predictive Levenberg-Marquardt algorithm (LM) and Eurocode 4 (EC4) design code. Analyses showed that the ANN model has better accuracy than EC4. Over a whole domain, the ANN model has higher coefficient of determination (R2), and lower root mean squared error (RMSE). The same conclusion is valid when two separate datasets are considered: one for stub columns and the other for slender columns. The benefit of the ANN model is its applicability in a broader range of column parameters. At the same time, EC4 puts several limitations on its use and gives satisfactory results only in limited circumstances. Empirical equations have also been proposed from the best ANN model, which is useful for engineering practice.