Design of cold-formed stainless steel circular hollow section columns using machine learning methods

Most existing design methods for the bearing capacity of stainless steel circular hollow section (CHS) columns were developed for a specific grade considering merely the global buckling failure mode. However, a variety of stainless steel grades with significant differences in material properties exist, and CHS columns may undergo local buckling, global buckling and global–local interactive buckling. To develop a unified design method suitable for various stainless steel grades and failure modes, this study adopted a machine learning based framework. First, 39 tests were conducted on cold-formed stainless steel CHS columns. Material properties, imperfections, load-deformation curves, and failure modes were reported in detail. Then, test data on stainless steel CHS columns in literature were collected and formed a database with 280 columns. Afterwards, two machine learning algorithms, Random Forest and Extreme Gradient Boosting, were used to predict the bearing capacity of column based on four types of input parameters. The Random Forest algorithm obtained the highest prediction accuracy when using all the design parameters as input. The accuracy of Random Forest algorithm based on the Comprehensive parameters (i.e., the non-dimensional slenderness of cross-section and member) is improved considerably when including the ratio of yield strength over the Young’s modulus as the input parameter. Finally, the prediction of machine learning method was compared with that of the design method in Design manual for structural stainless steel, and the proposed method shows a better accuracy.