Shear strength modeling for reinforced concrete beams strengthened with externally bonded fiber-reinforced polymer using machine learning
Ali Benzaamia, Mohamed Ghrici, Redouane Rebouh, Tryfon Sivenas, Panagiotis G. Asteris
Abstract
Accurately predicting the shear strength of externally bonded fiber-reinforced polymer (EB-FRP) composites in reinforced concrete (RC) beams remains challenging in structural engineering. Traditional models often fail to account for complex interactions between concrete, steel, and FRP, resulting in inaccurate predictions. This study applies advanced machine learning (ML) techniques, developing and evaluating six soft computing models, including eXtreme Gradient Boosting (XGB), Deep Neural Networks (DNN), and their monotonic counterparts, on a dataset of 394 FRP-strengthened RC beam specimens. Results show that incorporating monotonicity constraints improves predictive performance, with the Monotonic XGBoost and Constrained Monotonic Neural Networks (CMNN) models performing best. CMNN demonstrates strong alignment with FRP shear mechanics and outperforms existing design guidelines. This study highlights ML's potential in optimizing FRP shear strengthening design, enhancing the safety, reliability, and cost-effectiveness of RC structures.