The connection between FRP profile and concrete is critical for structural performance. This study introduces a novel fiber-bridging interface to enhance the FRP-concrete interfacial behavior. The interface comprises an epoxy resin adhesive layer, a carbon fabric layer, a mixture of adhesive and sand layer, and U-shaped steel fibers. Central pull-out tests were conducted to investigate the mechanical performance of this novel interface. The investigated variables included bond length and fiber volume fraction. Test results indicate that all specimens failed in a brittle mode at the adhesive layer, with load plateauing after reaching the peak. The number of steel fibers had limited influence on the interfacial behavior. Based on the load-slip curves, a bond stress-slip model for the tangential behavior was developed. An interfacial expansion model was further developed by means of FE analysis and machine learning. The three most widely used machine learning models, i.e., the BP neural network model, the random forest model, and the XGBoost model were selected. Comparisons show that all three models provide reasonable predictions, with the XGBoost model demonstrating the best performance. These models for the tangential and normal behavior of FRP-concrete interface were implemented into FE models for numerical analysis. Comparisons between numerical and experimental results show that the proposed models accurately describe the interfacial behavior of the fiber-bridging interface under brittle failure mode. The innovative interface proposed in this paper can be used for connecting concrete and FRP in various scenarios, and the proposed methodology of calibrating local bond behavior parameters from global response offers a new approach for establishing interfacial bond models.
Read full abstract