Machine learning (ML) based models for predicting the ultimate strength of rectangular concrete-filled steel tube (CFST) columns under eccentric loading

Abstract

Concrete-filled steel tubes (CFSTs) are popularly used in structural applications. The accurate prediction of their ultimate strength is a key for the safety of the structure. Extensive studies have been conducted on the strength prediction of CFSTs under concentric loading. However, in real situation CFSTs are usually subjected to eccentric loading. The combined compression and bending will result in more complex failure mechanisms at the ultimate strength. The accuracy of methods in design codes is usually limited due to their simplicity. In this study, three machine learning (ML) methods, namely, Support Vector Regression (SVR), Random Forest Regression (RFR), and Neural Networks (NN), are adopted to develop models to predict the ultimate strength of CFSTs under eccentric loading. A database consisting of information of 403 experimental tests from literature is created and statistically analyzed. The database was then split to a training set which was used to optimize and train the ML models, and a test set which was used to evaluate performance of trained ML models. Compared with the methods in two typical design codes, the ML models achieved notable improvement in prediction accuracy. The parametric study revealed that the trained ML models could generally capture the effect of each primary input feature, which was verified by the relevant experimental test results.