Critical temperature prediction in cold-formed steel columns exposed to local fire

Abstract

The structural performance of cold-formed steel (CFS) compression members can be significantly affected when exposed to non-uniform thermal distributions, particularly under localized fire. A nonlinear finite element model has been developed and is found to be in good agreement with previous experimental studies. Based on the developed model, extensive parametric investigation encompassing a total of 1728 FE simulations is carried out for heat transfer analysis (320), and column strength computation (128) and thermo-mechanical simulations (1280) considering a broad range of geometric, thermal, and load combinations. The findings indicate that the lowest critical temperature (θ_cr) predominantly occur in the mid-height heating zone due to a combination of localized heating and buckling instability. The study also reveals that the initial applied load level (LL) is the most influential factor on failure temperature, with a strong inverse correlation (ρ_θ,LL = −0.76). At 70 % LL, the θ_cr varies between 335 °C (for L = 1600 mm) and 432 °C (for L = 2800 mm), whereas at 30 % LL, it ranges between 171 °C and 219 °C. Eight supervised machine learning (ML) regression algorithms are evaluated for their critical temperature (θ_cr) predictive capabilities using various statistical metrics. Among these the support vector regression (SVR) is found to outperforms other ML algorithms, providing consistent prediction accuracy. Additionally, the Shapley additive explanation (SHAP) approach is used to understand the ML models, offering insights into how various factors influence the θ_cr. This study shows the potential of machine learning techniques in predicting θ_cr for structural fire design and the behaviour of CFS components in localized fire environments.