Prediction and interpretation of bond strength between FRP bars and fiber reinforced concrete using machine learning and Shapley Additive exPlanations Analysis

The bond strength between fiber-reinforced polymer (FRP) bars and fiber reinforced concrete (FRC) is a key parameter in the design of FRP-FRC structures. The interactions among the factors influencing the bond strength are highly complex and diverse, which limits the accuracy of traditional methods for developing predictive formulas. The machine learning methods that can handle complex problems have been difficult to apply in engineering due to the “black box” issue. In this study, a dataset comprising 281 pull-out test results of FRP bars embedded in FRC was collected to train and test seven ML models: Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN). Through a comprehensive evaluation and comparison with the five existing empirical equations, the ML models exhibited superior prediction accuracy, among which GBDT achieved the best performance in terms of both accuracy and generalization capability. Furthermore, based on the bonding mechanism of FRP–FRC, the influence of input features on bond strength was analyzed using the Shapley Additive Explanation (SHAP) method. The six most influential parameters affecting FRP–FRC bond strength were identified as FRP bar rib height, concrete compressive strength, FRP bar elastic modulus, FRP bar diameter, FRP bar surface condition, and bonded length to diameter ratio. Based on the SHAP algorithm and traditional physical modeling, this study proposes a novel, interpretable, and generalizable ML-based approach. A more accurate and physically meaningful FRP-FRC bond strength prediction equation was successfully developed.