Shear performance prediction of RC beams shear-strengthened with FRP sheet: A machine learning driven design-oriented method

This study employs machine learning (ML) methods to predict and analyze the failure modes and shear capacity of RC beams shear-strengthened with FRP sheets. A dataset comprising 410 experimental parameters and results was compiled to train and test six supervised ML algorithms . The findings indicate that the eXtreme Gradient Boosting (XGBoost) model demonstrated exceptional performance in predicting failure modes, achieving 85 % accuracy on the test set. Additionally, the introduction of four performance indicators (RMSE, MAPE, MAE , and R²) highlights the superior accuracy of K-Nearest Neighbors (KNN) and XGBoost algorithms in predicting shear bearing capacity compared to three authoritative empirical formulas and ML models, with all empirical formulas exhibiting comparatively poorer performance. The Shapley Additive Explanations (SHAP) method was then applied to quantify the influence of different features, with dataset samples analyzed to assess the correlation between design parameters and both failure modes and shear capacity. Notably, the wrap scheme, FRP tensile strength and thickness, as well as beam height and width, were significant in determining failure modes, whereas beam height and width, shear span- depth ratio, FRP strength , wrap scheme, and stirrup reinforcement ratio exerted the most substantial influence on shear capacity. These insights offer valuable guidance for enhancing the safety and economic efficiency of RC beams shear-strengthened with FRP in terms of design parameters. • Six ML models are employed to predict shear behavior of FRP strengthened RC beam. • ML models are superior in shear capacity prediction than empirical formulas. • SHAP is applied to explain the prediction process and variable impact of ML models. • Wrapping scheme and FRP strength impact results most among strengthened features.