Machine learning prediction of interfacial bond strength of FRP bars with different surface characteristics to concrete

Fiber-reinforced polymer (FRP) bars have been implemented in civil infrastructures as internal reinforcement. The bond strength of FRP bars to concrete depends on the surface characteristics considerably. This study used machine learning (ML) techniques to explore the influence of bar surface types on the bond properties quantitatively. A database including 158 FRP bars-concrete pull-out testing results was compiled. The geometric factors, including rib spacing (w_c), rib width (w_f) and rib height (r_h), were proposed to quantify the surface configurations of FRP bars. Twelve ML models were trained to predict the interfacial bond strength. The ML models demonstrated higher accuracy in predicting the bond strength compared to eight existing equations from the literature. Among these models, CatBoost exhibited the highest accuracy, with an RMSE 58.3 % lower than the most accurate existing equation. CatBoost was utilized in parametric research on the influencing factors, and demonstrated that w_f had the highest weight contribution to interfacial bond strength. Additionally, this study effectively combines ML models with physical meaning-driven analysis methods, resulting in a practical and interpretable equation for calculating FRP bars-concrete interfacial bond strength.