Surrogate-assisted model methods for maximum compression capacity predictions of concrete-filled stainless-steel tubular columns

Abstract

This work explores the prediction of axial strength of circular concrete-filled stainless-steel tubular (CFSST) columns, employing advanced machine learning techniques, including Gaussian Process Regression (GPR) and Extreme Gradient Boosting (XGBoost). The dataset comprises over 100 columns from experimental tests, with only a few of them being long or slender, limiting prediction accuracy. To address this, our study introduces a robust numerical modeling approach using Finite Element Method (FEM) to generate additional data points for long columns. The results are then benchmarked against established standards such as American Institute of Steel Construction (AISC) and the Eurocode 4, illustrating the potential of machine learning algorithms to supplant the conventional specifications.