Estimating Rebar Corrosion Distribution in Singly Reinforced Concrete with Stirrups using RBSM and Machine Learning

Abstract

Previous studies estimated internal corrosion distribution from surface cracks using Machine Learning and the Rigid Body Spring Model (RBSM). However, these models were limited to reinforced concrete (RC) without stirrups. Since most real structures contain stirrups, it is crucial to investigate whether the internal corrosion distribution can also be estimated for RC models with stirrups. So, in this study, we have investigated whether internal corrosion estimation is possible for singly reinforced RC models containing stirrups. A Convolutional Neural Network (CNN) was adopted, replacing the multilayer perceptron (MLP) used in prior study, to better learn the spatial patterns of corrosion. The CNN was trained using a dataset generated by RBSM simulations of a model without stirrups. As expected, the initial estimation accuracy for models with stirrups was low, as the training data lacked the confinement effect provided by stirrups. However, by iteration process proposed in previous research—an error correction loop where the crack width error is fed back to the network—it was demonstrated that the internal expansive strain distribution could be estimated with high accuracy. This validates the method's high flexibility and applicability for estimating corrosion in real structures.