Bond strength prediction of FRP bars to seawater sea sand concrete based on ensemble learning models

The bond strength between fiber-reinforced polymer (FRP) bars and seawater sea sand concrete (SWSSC) is a critical factor in structural design and applications. To address the inherent nonlinear problem, where bond strength is closely tied to constitutes of concrete and properties of FRP bar, this paper presented a novel approach to predict the bond strength using ensemble learning models and investigated the effects of input parameters on bond strength. For this purpose, a dataset of tests of pullout failures was determined consisting of 13 input parameters. Based on the determined dataset, two prediction models were developed using ensemble learning algorithms Random Forest (RF) and XGBoost. Simultaneously, a bond strength equation for pullout failures was derived from the ACI 440.1R-15 database since the existing equation in the code was originally developed to address splitting failures. The developed models demonstrated commendable accuracy based on statistical evaluations. Given its superior prediction and comparable generalization performances compared to the XGBoost model, the RF model was chosen for bond strength prediction compared to the developed equation. Comparative results favored the RF model, which significantly outperformed the developed equation in bond strength pre-diction. Furthermore, the study investigated the effects of input parameters by analyzing SHAP (SHapley Additive exPlanations) values. The analysis unveiled the influential mechanisms of mechanical property and composition of SWSSC, along with the mechanical, geometric, and surface properties of FRP bar on bond strength. In conclusion, the developed models represent an effective approach for predicting the bond strength of pullout failures between FRP bars and SWSSC.