Modelling the interface bond strength of corroded reinforced concrete using hybrid machine learning algorithms

The mechanical properties of corroded reinforced concrete are intricately associated with bond strength between corroded reinforcement and concrete. This strength must be predicted with ease and precision in practical engineering settings. This research proposes a prediction model for the bond strength of corroded reinforced concrete based on a hybrid machine learning algorithm using SHAP values. The model is based on ML algorithms, such as SVM, RF, and XGB, and is constructed based on 166 sets of experimental research data on the interface bond strength of corroded reinforced concrete in the past. The relationship between the factors affecting bond performance and bond strength is explored, and the predictive effect of SVM, RF, and XGB algorithms on bonding strength is discussed. In addition, this model innovatively uses the SHAP method to achieve the interpretability of ML models and overcome the "black box" problem of ML methods. Empirical formulas and ML models are compared as well. The results indicate that the predictive performance of the model far exceeds that of the empirical formula, and the RF model has the best predictive effect. Furthermore, the SHAP method quantitatively confirms that the corrosion rate has the most significant impact on bond strength, which is consistent with experimental observations. Based on the experimental results, the model provides a new choice for addressing the problem of bond strength in corroded reinforced concrete.