A machine learning framework for predicting the fire resistance time of cold-formed steel walls

Abstract

This study proposes a machine learning (ML) framework for accurately predicting the fire resistance time of cold-formed steel (CFS) walls under ISO 834 fire conditions. Only 32 experimental results regarding the fire resistance time of CFS walls were reported in the literature. To address this issue, a finite element heat-transfer model was developed and validated against the experimental test results. Using the validated models and the load ratio-hot flange critical temperature curve, a total of 592 data points were compiled to construct the dataset of CFS walls. The effects of various wall configurations, sheathing board types and thicknesses were investigated. Different ML models were used, comprising Artificial Neural Network, Extreme Learning Machine, Convolutional Neural Network and eXtreme Gradient Boosting (XGBoost). The results suggest that XGBoost can accurately predict the fire resistance time of CFS walls, and R² and MAE values were 0.986 and 2.287, respectively. The mean ratio of the fire resistance time obtained from the XGBoost prediction to the experimental test results was 1.004, with a corresponding standard deviation of 0.047. Finally, the XGBoost prediction was interpreted by the SHapley Additive exPlanations method to determine the significance of input features. The ML-based framework proposed in this study can offer a promising alternative for researchers and engineers to efficiently and effectively design the CFS walls under fire conditions.