Prediction of fire-induced steel beam deformation using machine learning algorithms

The versatility of steel in the construction industry stems from its advantages over other construction materials. Despite that, its poor resistance to fire has remained a serious concern. Several research efforts have been triggered on this subject following some unfortunate fire incidences in famous structures around the world, such as that of the World Trade Centre, U.S., and Grenfell Tower, U.K. While numerical simulation, commonly used by researchers suffers limited accuracy, its alternative, the experimental technique is uneconomical and time-consuming. In this study, four machine learning algorithms are trained based on a full-scale experimental dataset to predict the fire-induced deformation of steel beams. The Extra Trees model achieved the best performance with an R² of approximately 1 and an MAE of 0.009 mm, with 86% of prediction errors falling below 0.0125 mm. Therefore, promising a safe and economical application. In summary, the models prove that machine learning can capture nonlinear relationships and dependencies between these variables and the resulting deformations. Hence, the approach can potentially overcome the limitations of traditional methods.