Machine learning models for predicting ultimate bond strength between UHPC and reinforcing steel bar

To improve the prediction accuracy of the ultimate bond strength between ultra-high-performance concrete (UHPC) and reinforcing bars while maintaining convenience in engineering applications, this study establishes a unified database containing 500 reinforced bar-UHPC bond-slip failure specimens based on experimental results reported in the published literature. Concrete compressive strength, bar diameter, anchorage length, concrete cover thickness, steel fiber volume fraction, steel fiber length, and steel fiber diameter are selected as input variables, while the ultimate bond strength is used as the output variable. First, representative existing empirical formulas are systematically evaluated. Second, eight machine learning models are employed for comparative prediction, and the optimal model is selected using five-fold cross-validation and hyperparameter optimization. Furthermore, Pearson correlation analysis and the SHAP method are used to interpret the influence patterns of the main parameters. Based on these results, a simplified explicit prediction formula for bond strength is proposed. Finally, 108 sets of independent external experimental data are used for validation. The results show that the existing empirical formulas generally have limited prediction accuracy, among which Chen’s formula performs relatively well. Among the machine learning models, the Extra Trees model exhibits the best overall performance, with R², RMSE, and MAE values of 0.926, 4.98 MPa, and 3.16 MPa, respectively, on the test set. The Pearson correlation analysis and SHAP results indicate that anchorage length has the most significant influence on bond strength, followed by concrete cover thickness, concrete compressive strength, and steel fiber volume fraction. The explicit formula proposed in this study achieves an R² of 0.801 on the source database and demonstrates good robustness in external validation. The findings indicate that machine learning models are suitable for high-accuracy prediction of the bond strength between reinforcing bars and UHPC, whereas the explicit formula is more appropriate for rapid estimation in engineering design. This study can provide a reference for the analysis of bar anchorage, lap splicing, and interfacial bond performance in UHPC structures.