Font Size: 

This paper presents the machine learning-based methods that construct the accuratemodel prediction for the maximum compression capacity responses of concrete filled doubleskin steel tube (CFDST) columns under uniaxial compression forces. The so-called surrogateassistedmodel is generated from the set of training data collected from available experimentalresults. The dataset is classified into two classes, namely the behaviors of short (section failure)and long (member failure) CFDST columns. Two machine learning, including gaussian processregression (GPR) and extreme gradient boosting (XGBoost), methods are encoded in this study.The input data considers geometry (i.e., external and internal diameters/thicknesses of steeltubes, and column length) and material properties (concrete compressive strength, and yieldstrengths of external and internal steel tubes) of the columns. The output data is the maximumcompression capacity of the CFDST columns. The total training datasets comprise of 122 datafrom the short column tests and 146 data from the long column tests. The surrogate-assistedmodels determine the accurate uniaxial compression strengths for both short and long CFDSTcolumns, where the good comparisons with relevant standard design specifications areevidenced. The long column responses given by the GPR model are more accurate than thoseperformed by the XGBoost approach, but the short column responses given by the XGBoostapproach are more accurate than those performed by the GPR.

Composite columns, Machine Learning, Gaussian Process Regression, Extreme Gradient Boosting, CFDST, Compression Capacity