Explainable machine learning-based prediction of plate-end debonding in fiber-reinforced polymer-strengthened reinforced concrete beams

This study investigates plate-end (PE) debonding in reinforced concrete beams strengthened with fiber-reinforced polymer (FRP) composites. PE debonding is a critical failure mode that compromises the effectiveness and structural performance of FRP reinforcement systems. Although FRP has been widely adopted in structural rehabilitation for its high strength and corrosion resistance, PE debonding—often initiated by shear or inclined cracks—remains a persistent challenge. Conventional computational models for predicting this failure mechanism exhibit limited accuracy due to the complex nonlinear interactions among influencing parameters. To overcome these limitations, this research applies machine learning techniques, with an emphasis on explainable artificial intelligence methods such as Shapley additive explanations, to build more reliable predictive models. A comprehensive database is developed incorporating key variables governing PE debonding, and various machine learning algorithms are trained, validated, and benchmarked against existing code-based models. Furthermore, parameter importance and sensitivity analyses are conducted to improve interpretability and provide insights for advancing the design and application of FRP-reinforced structures.