Mechanical Fault Diagnosis Method for GIS Based on Convolution Neural Network and Enhanced Gramian Angular Field

Abstract

To reliably identify the operating state of Gas Insulated Switchgear (GIS), a mechanical fault diagnosis method for GIS, based on a convolutional neural network and enhanced gramian angular field is proposed. Firstly, the vibration sensor is used to obtain the original time domain signals of GIS equipment in different states. Afterwards, the gramian angular field is employed to encode the one-dimensional time domain signals into two-dimensional maps. Finally, the convolution neural network is utilized to identify the different mechanical defects of GIS. Moreover, the typical mechanical defects of three kinds of GIS equipment are simulated. The calculation results show that the proposed method could effectively represent the different operating states of GIS, and the identification accuracy reached 98%, which provides a reliable basis for the state evaluation of GIS.