Article ID: 2607R003
The accidental collapse of concrete structures due to rebar corrosion can be a serious problem. Although electromagnetic induction is advantageous among the non-destructive testing methods, the depth and diameter of the rebar need to be measured in advance. In our previous study, multi-axis and multi-channel magnetic measurements were effective in estimating the rebar corrosion level, which could be done without requiring information on the depth of the rebar thanks to using a convolutional neural network (CNN). However, information on the rebar diameter was still needed. In the current study, we investigate how to estimate the corrosion level without depth and diameter information when inputting data combining the magnetic amplitude at three different diameters of rebar arrangements. To achieve this, we propose using cubic function interpolation to increase the amount of training data for the CNN. Our results showed that the accuracy of corrosion level estimation without cubic function interpolation was 82%, whereas when the interpolation was implemented at depths of 20, 25, 30, 40, and 50 mm, the accuracy rose to 89%. These findings demonstrate that the proposed method is effective in estimating the rebar corrosion level without information on the rebar depth and diameter.
