Enhanced bond strength prediction in corroded reinforced concrete using optimized ML models

Abstract

In reinforced concrete (RC) structures, the reduction of bond strength (BS) between reinforcing bars and concrete is significantly influenced by corrosion, resulting in a notable impact on their remaining service life and load-bearing capacity. The investigation of corrosion’s effect on BS has encountered challenges due to its resource-intensive nature. To overcome these limitations, the CatBoost (CAT) machine learning model is introduced in this study to accurately predict BS in RC. A comprehensive database comprising 378 bond tests was compiled to train and evaluate the CAT model. Input parameters include bar specifications, stirrup properties, concrete characteristics, corrosion factors, and test types. The optimal CAT hyperparameters and the superior model were determined using cross-validation in conjunction with Monte Carlo simulations. The results demonstrate that the CAT model achieves an impressive correlation value of 0.944, surpassing the performance of the five empirical equations found in the literature. Partial Dependence Plots (PDP) analysis was conducted to assess the impact of individual input variables and the interdependence between factors on the BS of corroded RC. Finally, a graphical user interface (GUI) was developed to facilitate user-friendly BS predictions, enhancing the model’s practicality.