Cross Spectrum Aided Surface Condition Assessment of Metal Oxide Surge Arrester Employing Convolutional Neural Network

This paper proposes an automated framework for surface condition assessment of metal oxide surge arresters (MOSA) employing cross-wavelet transform (XWT) of the leakage current and convolutional neural network (CNN). XWT is performed between different contaminated leakage currents with an uncontaminated leakage current. A set of cross-wavelet spectrum (XWS) images are obtained after XWT. The XWS images are fed to the proposed customized CNN model for classification of MOSA surface contamination levels. The performance of the proposed CNN model is also compared with two benchmark CNN architectures namely AlexNet, VGGNet16. It is observed that the proposed model delivers better performance and offers significantly reduced computational burden compared with the existing benchmark CNN models. The proposed method can be applied for surface condition monitoring of MOSA in real-life.