Prediction of interlaminar shear strength retention of FRP bars in marine concrete environments using XGBoost model

Abstract

The degradation of interlaminar shear strength (ILSS) of fiber-reinforced polymer (FRP) bars exhibits highly nonlinear characteristics when exposed to marine concrete environments. To address this phenomenon, a novel machine learning approach utilizing XGBoost algorithm was developed to predict ILSS retention (ILSSR). A comprehensive dataset was compiled from experimental results and existing literature. The constructed XGBoost model successfully captured the nonlinear degradation of ILSS and demonstrated excellent performance in terms of accuracy and generalization, achieving a testing R² value of 0.93 and a mean absolute error (MAE) of 3.78. The model represents a reliable and efficient tool for forecasting ILSSR of FRP bars in marine concrete environments. The coupling effects of FRP bar properties, exposure conditions and test parameters on ILSS degradation was further investigated through SHapley Additive exPlanations (SHAP), which offered a data-driven perspective that corroborated previous experimental findings. Utilizing the developed model, environmental reduction factors for ILSS under seawater sea-sand concrete (SWSSC) environment were quantified, taking into account the influence of fiber and matrix composition. The analysis revealed distinct environmental reduction factors: 0.71 for epoxy-based GFRP, 0.62 for epoxy-based BFRP, and 0.87 for vinyl-based GFRP bars. These findings underscore the substantial impact of matrix and fiber types on ILSSR, with the matrix type exerting a more dominant influence. The XGBoost-based predictive model developed in this study offers an effective and efficient method for assessing the long-term durability of FRP bars in marine concrete environments.