Data‐driven moment‐rotation formulation of slender cold‐formed steel beams subjected to local buckling

Abstract

Accurately predicting the structural response of Cold‐Formed Steel (CFS) members under extreme load events is essential for reliable seismic performance assessment and design of multi‐storey CFS structures. However, this remains challenging due to the complex behaviour of thin‐walled CFS elements influenced by interacting buckling modes. Current seismic performance‐based design guidelines, such as ASCE 41‐23, primarily rely on local slenderness and do not account for other influential parameters. To address this limitation, the present study proposes a novel data‐driven formulation to predict the moment–rotation behaviour of CFS beams. A comprehensive series of finite element (FE) models was developed and validated against experimental data to generate a robust dataset representing the moment–rotation response of CFS beams governed by local buckling. Following a comparative analysis with the ASCE 41‐23 model, Multi‐Gene Genetic Programming (MGGP) was employed as a machine learning (ML) technique to predict the moment–rotation response of CFS beams. The input parameters include key geometric and material properties, such as local slenderness, web‐to‐thickness ratio, flange‐to‐thickness ratio, and yield and ultimate steel strengths for C‐ and Z‐shaped cross‐sections. The results demonstrate that the proposed MGGP model achieves significantly higher predictive accuracy than ASCE 41‐23, reducing average errors by 70% for peak moment, 78% for rotation at peak moment, and 87% for ultimate rotation. Finally, a web‐based application was developed to enable rapid ML‐based prediction of the output parameters.