Classification of Corrosion in Underground Power Cables Using CNNs and Microscopic Imaging

This paper presents a convolutional neural network (CNN) model designed to classify the corrosion in the conductor layer of underground power cables as either aqueous corrosion or sulfide corrosion using microscopic images (MIs) obtained from scanning electron microscopy (SEM) analyses. The classification of corrosion types is typically determined through SEM-based elemental concentration analysis in combination with compound identification through X-ray photoelectron spectroscopy (XPS), Fourier transform infrared spectroscopy (FTIR), or electrochemical tests. However, the proposed CNN model eliminates the need for these additional laboratory analyses by classifying corrosion types solely based on SEM images. The dataset used for model training consists of samples collected from failed industrial underground cables, where corrosion labels are assigned based on industrial reports, SEM analysis, and XPS results. To address the limited dataset size, data augmentation techniques are employed to expand the number of training samples. The processed dataset is then used to train and test the CNN model, which is shown to achieve an accuracy of $95 \%$ when classifying the corrosion type using only SEM MIs.