Neural Network Modeling of Corrosion Level of Rebar in Steel Fiber Reinforced Self-Compacting Concrete

Abstract

Corrosion is one of the biggest problems of reinforced concrete structures prone to high chloride environments such as ports and harbors. Due to a lack of studies that can support the use of steel fiber reinforced self-compacting concrete, researchers are still in dispute regarding the effect of using steel fibers in chloride-rich environments. This paper explores the use of neural network modeling to precisely predict and further analyze this problem. Twenty-six different mixtures of steel fiber reinforced self-compacting concrete with varying amounts of cement, water-cement ratio, superplasticizer, and steel fiber were used to derive the feed forward back propagation neural network and compared to a derived non-linear model. The derived neural network model with fourteen hidden nodes and tansig as transfer function has an R-squared of 0.949 for the training. The comparison shows that ANN has superior predicting capability compared to non-linear modeling even with a limited number of data. Parametric analysis was performed and found that steel fiber shows improvement in the corrosion resistance of concrete for mixtures with low to moderate water-cement ratio and an opposite behavior for high water-cement ratio. This is due to the presence of voids formed around the surface of the steel fiber due to capillary action. These voids serve as highways for chloride ions.