Explainable Machine Learning Framework for Predicting Bond Strength of FRP Bars Embedded in UHPC

Abstract

The interfacial performance of advanced composites bars embedded in Ultra-High Performance Concrete (UHPC) is an important factor that controls load transfer and the performance of structural elements. Predicting bond strength is still difficult because it is affected by several factors, such as rebar type, bar profile, bar diameter, bonded length, cover depth, fiber content, UHPC compressive strength, and FRP tensile strength. Therefore, this study uses machine-learning models to estimate the the bonding capacity of FRP bars placed in UHPC Using a collected experimental database of 183 specimens from previous studies. Four machine-learning models were developed and compared, including Linear Regression, Random Trees, Multi-Layer Perceptron, and Locally Weighted Learning. The MLP model gave the best prediction performance, with a correlation coefficient of 0.9466, MAE of 2.3083 MPa, and RMSE of 3.0631 MPa. SHAP analysis showed that embedment length was the most influential variable, followed by bar surface condition, FRP tensile strength, and concrete cover. This confirms that FRP–UHPC bond behavior is controlled by the interaction between bonded length, surface condition, mechanical interlock, and confinement provided by UHPC. Overall, the developed explainable ML framework provides a useful tool for predicting FRP–UHPC bond strength and supporting future UHPC-specific bond models.