Corrosion effect on bond behavior between rebar and concrete using Bayesian regularized feed-forward neural network

Abstract

Corrosion of reinforcement in concrete structures is essentially detrimental to the lifelong service of the reinforced concrete structures. The diminishing of the bond strength is the most influential factor. This study gains insight into the mechanisms of bond strength reduction due to corrosion by adopting an advanced machine-learning model. At first, the tension pull-out tests are conducted to reveal the mechanism and create the dataset. Based on the experimental results, the dataset is further enriched by collecting more samples in the literature, thus an effective soft computing approach i.e., Bayesian Regularized Feed-Forward Neural Network (BR-FFNN), is proposed to evaluate the bond behavior of corroded reinforcement. The newly developed BR-FFNN model demonstrates its robustness and efficacy by providing the highest performance indicators hitherto. Furthermore, the importance-based sensitivity analysis is also carried out to capture the effect of each input on the ultimate bond strength. To facilitate the engineering practice, mathematical modeling is put in place to formulate an explicit function that the ultimate bond strength could directly estimate thereof.