Machine Learning Models for Inferring the Axial Strength in Short Concrete-Filled Steel Tube Columns Infilled with Various Strength Concrete

Abstract

Concrete-filled steel tube (CFST) columns are used in the construction industry because of their high strength, ductility, stiffness, and fire resistance. This paper developed machine learning techniques for inferring the axial strength in short CFST columns infilled with various strength concrete. Additive Random Forests (ARF) and Artificial Neural Networks (ANNs) models were developed and tested using large experimental data. These data-driven models enable us to infer the axial strength in CFST columns based on the diameter, the tube thickness, the steel yield stress, concrete strength, column length, and diameter/tube thickness. The analytical results showed that the ARF obtained high accuracy with the 6.39% in mean absolute percentage error (MAPE) and 211.31 kN in mean absolute error (MAE). The ARF outperformed significantly the ANNs with an improvement rate at 84.1% in MAPE and 65.4% in MAE. In comparison with the design codes such as EC4 and AISC, the ARF improved the predictive accuracy with 36.9% in MAPE and 22.3% in MAE. The comparison results confirmed that the ARF was the most effective machine learning model among the investigated approaches. As a contribution, this study proposed a machine learning model for accurately inferring the axial strength in short CFST columns.