Data-driven shear strength prediction of RC beams strengthened with FRCM jackets using machine learning approach
Xiangsheng Liu, Grazziela P. Figueredo, George S. D. Gordon, Georgia E. Thermou
Abstract
Fabric Reinforced Cementitious Matrix (FRCM) is an effective intervention method for improving the shear strength of existing reinforced concrete (RC) beams, yet predictive analyses are scarce. This study introduces and compares nine machine learning (ML) models to estimate the shear capacity of FRCM-strengthened RC beams, utilizing a dataset characterized by 14 distinct variables. Among these, the Extreme Gradient Boosting (XGBoost) model distinguishes itself with the highest accuracy, with R² , MAE , RMSE , and CoV values of 98.98 %, 5.36 kN, 10.07 kN, and 9.75 %, surpassing traditional empirical and mechanical formulations. Additionally, robustness analysis across random dataset splits further confirms the stability of the XGBoost model's predictions. The superiority of XGBoost and other ML models over conventional approaches is further validated through Taylor diagrams, which demonstrate closer alignment with experimental outcomes. Shapley Additive Explanations (SHAP) analysis identifies key influential factors, including beam depth, concrete compressive strength , stirrup reinforcement ratio, and thickness of mortar, while Partial Dependence Plots (PDP) reveal their marginal effects on shear strength. These insights are crucial for enhancing design practices and guidelines for FRCM-strengthened beams. A practical graphical user interface (GUI) tool was also developed, facilitating direct application of findings in the study.