Data-driven buckling capacity prediction of normal- and high-strength steel hollow structural section columns

The use of high-strength steel has been increasing in the steel construction industry due to its high strength-to-weight ratio. Currently, high-strength steel columns are designed by equations that were developed for normal-strength steel, which has shown to result in conservative predictions for high-strength steel columns. Moreover, the current version of AISC 360 is limited to hollow structural sections (HSS) with steel grades up to 485 MPa. This study establishes a data-driven design approach for HSS columns that addresses a range of steel grades up to 960 MPa, residual stresses, geometric imperfections, and geometry including member and element slenderness. An extensive database for square and rectangular HSS columns is constructed from the existing literature containing both experimental and numerical research. Conventional interpolation and newer machine learning methods were adopted to examine the buckling strength estimation. The predictions derived from the proposed data-driven model were compared with the capacities stimated by AISC 360 to examine the accuracy of current design rules for HSS columns. The data was divided into groups to study the effect of factors on the design prediction accuracy, including (1) normalvs. high-strength grades and (2) hot-rolled vs. cold-formed sections. In addition, the database and all analysis methods are coded in a user-friendly computational notebook format for ease of use and saved in the open-source repository so future researchers and users can update and modify the analysis as future data is generated.

View Content