Exploring Machine Learning for the Stability Analysis of Rotary‐straightened Steel Members with Multiple Stiffness Reduction Models

Abstract

AbstractRotary straightened structural steel cross‐sections have a different residual stress pattern compared to those without rotary straightening, which affects structural behavior, especially in members subjected to inelastic buckling. A stiffness reduction material model for rotary‐straightened hot rolled sections was previously developed and validated for a specific set of cross‐section sizes and load conditions. Previous research has indicated that the current stiffness reduction model specified in Chapter C of AISC 360 for stability design does not accurately capture the stiffness reduction of rotary straightened W‐shapes. Particularly, the current stiffness reduction model does not always derive conservative capacity compared with the model for rotary straightened sections. This paper explores machine learning for predicting the beam‐column capacity considering different stiffness reduction models and estimating their limit load capacity. Two extreme gradient boosting models were developed for beam‐columns under major or minor axis bending. The dataset contained 1,296 finite element simulations for beam‐columns with a range of different cross‐section geometries. Excellent prediction accuracy of the developed machine learning models showed the potential of using machine learning for the stability analysis of rotary‐straightened members.