Application of the Gradient Boosting Algorithm for Bond Strength Prediction of GFRP bars with Concrete

Abstract

In marine environment, chloride-induced rebar corrosion is the principal cause of reinforced concrete degradation, resulting in material loss, cracking, and bond failure, all of which dramatically degrade structural safety, load-bearing capacity, and service life. As a result, using non-corrosive reinforcement, such as Glass Fibre Reinforced Polymer (GFRP) bars, is critical for increasing the service life of structures exposed to marine environments.  It has a variety of other benefits over steel rebar as well, including, higher tensile strength, cost-effectiveness, reduced density, and non-magnetic property.  The bond behaviour of GFRP bars with concrete differs from that of steel, as it exhibits linearly elastic behaviour with a distinct surface deformation pattern. While several empirical models exist to estimate bond strength, their predictive accuracy is limited; therefore, it is imperative to develop a more robust, data-driven prediction model. This study demonstrates the effectiveness of Gradient Boosting-based Machine Learning model, optimized via Optuna with K-fold cross-validation, on estimating the bond strength of GFRP bars with concrete. Analysis of the results indicates that the model yields superior performance metrics across both the training and testing phases, characterized by a higher coefficient of determination (R²) and a reduced Root Mean Square Error (RMSE). As a result, it may be inferred that the model has a higher predictive capacity than empirical equations.