BOND STRENGTH PREDICTION MODEL OF CORRODED REINFORCEMENT IN CONCRETE USING NEURAL NETWORK

Abstract

The expansion of corrosion products in the steel-concrete interface offers radial tensile stress
resulting in the development of cracks in reinforced concrete structures. This corrosion-induced crack promotes
bond reduction involving intricate non-linear interactions. To deeply understand the underlying mechanisms in the
bond strength of corroded rebars in concrete, a novel bond prediction model using artificial neural network (ANN)
was developed. Accelerated corrosion was performed to 108 cube samples using 500 µA/cm2 current density. Steel
bond strength after 35 and 70 days impressed corrosion exposure of concrete cube samples was measured using a
single pull out test. The compressive strength, tensile strength, rebar diameter, embedment length, concrete cover,
ultrasonic pulse velocity (UPV), crack severity, and corrosion level were the predictors in the ANN bond model.
Among all the bond strength models considered in this study, the proposed neural network model provided the
most desirable bond estimates in good agreement with experimental results. The ANN model further showed
superior prediction performance against the derived regression model.