Automated explainable ensemble learning prediction of FRP-bar-to-concrete bond strength and failure pattern under diverse exposure scenarios

FRP bar has emerged as a strong contender for replacing conventional steel reinforcement to address costly corrosion in reinforced concrete structures. The bond between FRP bar and concrete plays a paramount role in transferring loads and sustaining structural integrity. In this study, the effect of exposure conditions on the FRP rebar-to-concrete bond strength (BS) was explored. Traditional design guidelines and analytical models for estimating such BS are often labor-intensive and may encounter significant complexity and limitations when applied to diverse exposure scenarios. To overcome this limitation, this study leverages machine learning (ML) techniques to predict the FRP bar-to-concrete BS and failure pattern under various exposure conditions. A pertinent comprehensive database on BS and failure type under diverse exposure conditions was compiled. The performance of eight ML algorithms in predicting the failure mode and BS was evaluated. Current design guidelines and existing models were used for benchmark comparison. The results show that the XGB and GBT models outperformed the other ML algorithms in failure mode identification and BS prediction. The FRP rebar bonded length was identified as the most significant feature influencing BS, followed by the concrete compressive strength, rebar diameter, FRP surface condition, and exposure duration. Finally, this study presents an automated toolbox that can be used through a free access user-friendly graphical interface for estimating the BS and failure patterns under diverse exposure conditions.