Unified machine-learning-aided design of cold-formed steel channel section columns with different buckling modes at ambient and elevated temperatures

Abstract

Due to the ease of fabrication, cold-formed steel channel section members have gained popularity in the construction industry. However, the open and non-doubly symmetric geometries make them vulnerable to buckling, particularly under extreme loading conditions, such as fire. Current design codes for cold-formed steel channel section columns have been found to be cumbersome and lead to inaccurate failure load predictions. Therefore, an accurate and unified design method is proposed based on machine learning algorithms for cold-formed steel channel section columns with different material grades, section types, geometric dimensions and boundary conditions at both ambient and elevated temperatures. In this paper, a database of 473 cold-formed steel channel section columns with various material properties and geometric parameters was firstly collected. Regression models for failure load predictions were developed based on six machine learning algorithms – Random Forest, Extra Trees, Support Vector Machine, Extreme Gradient Boosting, Cat Boosting and Light Gradient Boosting Machine, with the key hyperparameters for each regression model tuned. The performance of regression models was assessed using a series of statistical metrics. The assessment results reveal that the regression model trained by Support Vector Machine achieves the best performance. The regression model trained by Support Vector Machine was then compared with the current design codes, indicating that the machine-learning-aided design method results in substantially improved design accuracy and consistency for cold-formed steel channel section columns failing by different buckling modes at both ambient and elevated temperatures over the current design codes.