Prediction of Bond Strength at the NC–UHPC Interface Utilizing a Multifeature-Driven Analysis

Concrete structures frequently experience different levels of deterioration and degradation as a result of structural stress damage, environmental erosion, and the aging process. Ultrahigh-performance concrete (UHPC) increasingly is recognized as a suitable material for the repair and reinforcement of conventional concrete structures, attributed to its enhanced properties. To assess the effectiveness of normal concrete (NC)–UHPC repair, it is essential to establish a method for the rapid and precise prediction of bond strength at the NC–UHPC interface. Traditional empirical models require significant time and effort, often demonstrating heightened sensitivity to their specific research data, which presents certain limitations. In contrast, machine learning methods leverage extensive existing experimental data, allowing for the effective establishment of regression relationships between input and output features through comprehensive data analysis. In this study, three machine learning algorithms were employed to predict the strength of the bond between NC and UHPC: random forest (RF), extreme gradient boosting (XGBoost), and category gradient boosting (CatBoost). The models underwent training and testing with 108 data sets sourced from the existing literature. Subsequently, the prediction outcomes of the XGBoost model were evaluated against the computational results derived from two empirical formulas, demonstrating that the machine learning model has considerable advantages. The study employed the SHapley Additive exPlanations (SHAP) feature dependency plot to quantify the contribution range of each input parameter to the output results. These analytical findings are instrumental in understanding the impact of individual variable action intervals and their various combinations on bonding performance, The newly developed machine learning model is anticipated to function as an innovative approach for the rapid and precise evaluation of bond strength at NC–UHPC interfaces.