Machine learning prediction and interpretability of bond strength in corroded reinforced concrete under high temperatures

The bond behavior of corroded reinforced concrete exposed to high temperatures is critical for evaluating infrastructure safety and durability. However, accurately predicting bond strength under such conditions remains challenging. Machine learning offers promising solutions to this problem by providing accurate predictions and enhancing the interpretability of influencing factors. In this study, a dataset containing 815 sets of samples is built, and the dataset includes seven input variables and one output variable. Based on this, ten machine learning models are developed to predict the bond strength of corroded reinforced concrete at high temperatures. And all models are evaluated and compared by 10-fold cross-validation and performance evaluation metrics. The results show that all machine learning models have superior performance in the prediction of the bond strength, and the extreme gradient boosting model is the optimal model. The proposed models have better performance than the reported empirical prediction models. The global sensitivity indices indicate that the corrosion ratio of the reinforcement, compressive strength of the concrete, and yield strength of the reinforcement have higher sensitivities in the prediction of the bond strength. Shapley additive explanations and partial dependence plots identify the input variable that has the greatest impact on the output of the optimal model as the corrosion ratio of the reinforcement and further quantify the impact of the input variables individually and interactively on the bond strength. The findings have practical implications for assessing the fire resistance and long-term durability of reinforced concrete structures under high-temperature conditions and widen the application of machine learning in civil engineering.