Using machine learning to predict fire resistance of FRP strengthened concrete beams

Abstract

In recent advancements in structural engineering, fiber-reinforced polymer (FRP) reinforcement has emerged as a notable innovation, for enhancing the strength of reinforced concrete (RC) structures. However, ensuring sufficient fire resilience remains a critical challenge, particularly in structures where fire safety is paramount. The conventional approach to assessing fire resistance through experimental and numerical analysis impairs resource constraints. This study investigates the potential of Machine Learning (ML) in predicting the fire resistance of FRP-strengthened RC beams. By using a comprehensive dataset of over 21,000 numerical and experimental data, with diverse geometric, insulation, loading, and material variables, six ML methodologies were employed, including both ensemble and traditional techniques. Remarkably, the Light Gradient Boosting (LGB) model demonstrated exceptional accuracy, exceeding 92%. To refine model accuracy, Grid Search optimization and k-fold crossvalidation were applied. Additionally, the SHapley Additive exPlanation (SHAP) method was utilized to interpret the results. The findings highlight that parameters such as loading ratio, insulation depth, tensile reinforcement area, concrete cover, and FRP area significantly influence the fire resistance prediction. This study underscores the efficacy of ensemble ML techniques in accurately forecasting fire resistance, providing valuable insights for optimizing structural design and ensuring enhanced structural fire safety.