Machine learning prediction method for the interface bond strength between fiber reinforced polymer bars and concrete based on multi-feature driven analysis

Abstract

In severe service conditions, corrosion of reinforcement poses a significant challenge for concrete structures. Fiber reinforced polymer (FRP) materials, characterized by their low weight, high strength, and superior corrosion resistance, are effective in enhancing the durability of these structures. The interfacial bond performance between FRP bars and concrete is crucial for their synergistic action and is pivotal for the design and safety evaluation of FRP-reinforced concrete structures. This paper conducts a comprehensive analysis of factors including the material and mechanical properties of FRP bars, concrete mechanical properties, and bond length. Utilizing an exhaustive search strategy, it identifies the optimal characteristic parameters for assessing the bond strength between FRP bars and concrete. Subsequently, these parameters are integrated with four machine learning (ML) algorithms—DT, SVM, RF and XGB—to train the nonlinear mapping relationship between influencing factors and bond strength, proposing a unified and interpretable prediction method for the FRP bar-concrete interface bond strength. The model's predictive accuracy is then validated using a test dataset, and a comparative analysis is performed among the predictive models generated by the four ML algorithms, as well as against empirical calculation formulas. The findings indicate that the ML models offer superior predictive performance and accuracy over empirical formulas, with the XGB algorithm demonstrating the highest overall accuracy and best performance. Moreover, to address the black-box nature of ML algorithms, the SHAP method is employed to enhance the interpretability of the bond strength prediction process. The newly developed hybrid ML model holds promise as a novel approach for accurately assessing the bond strength at the FRP bar-concrete interface.