Prediction of intermediate crack debonding failure in FRP-strengthened reinforced concrete beams based on ensemble learning

Fiber-reinforced polymers (FRP) are widely used in the reinforcement of reinforced concrete (RC) structures due to their lightweight, high strength, and corrosion resistance. However, FRP reinforcement of RC beams often suffers from delamination failure due to material and geometric differences between concrete, reinforcing steel, and FRP, especially debonding caused by intermediate cracks (IC), which has become a key issue affecting structural safety. Although various empirical models based on interface bond strength or delamination strain have been developed to predict IC debonding, such as the ACI 440.2R, CECS, and TR55, their accuracy and adaptability remain limited. In recent years, machine learning techniques have shown great potential in predicting the performance of FRP-strengthened concrete structures. This paper aims to collect experimental data to build a database, use machine learning methods to establish an IC debonding failure prediction model, and combine Shapley additive explanations for parameter sensitivity analysis to improve model accuracy and adaptability, thereby providing support for the safety assessment of FRP-strengthened RC beams.