Prediction and Model Validation of UHPC-NSC Interface Bonding Performance Based on Machine Learning Algorithms

Ultrahigh performance concrete (UHPC) is a highly durable material characterized by its exceptional properties, such as high compressive strength, dense microstructure, and enhanced surface texture, which contribute to improved bond strength, making it an ideal choice for a variety of applications, including the repair of normal strength concrete (NSC). This study focuses on exploring the interfacial bond properties between UHPC and NSC, which are crucial for ensuring long-term performance and durability of repaired structures. The research specifically investigates the effects of curing temperature and substrate water saturation on the interfacial bond strength, utilizing split tensile and oblique shear tests to measure bond properties. The experimental results show that higher curing temperatures significantly enhance the bond strength, with the maximum strengths reaching 18.19 and 17.71 at 20°C and 50°C, respectively, while only 16.41 at −20°C. Additionally, as the substrate water saturation increases, the failure strength also gradually increases, with the maximum strength growth rates reaching 68.61% and 65.13%, both occurring at a substrate water saturation of 75%. These findings suggest that curing conditions and water saturation play a crucial role in optimizing bond strength, providing valuable insights for improving material performance in practical applications. Due to the nonlinear relationship between input and output and through model comparison, the machine learning model based on support vector regression, which is suitable for nonlinear relationships and demonstrates better predictive capabilities, is selected to predict interface bond strength. The model is validated using experimental data and optimized through correlation analysis. The validation results indicate that the model can reliably predict the interface bond strength between UHPC and NSC, demonstrating high accuracy in handling complex nonlinear relationships. Further, it offers a promising solution for optimizing concrete repair strategies, highlighting its potential for widespread application in future concrete repair projects.