Application of the convolutional neural network in partial discharge spectrum recognition of power apparatus

Abstract

Partial discharge (PD) detection is used to evaluate the insulation status of high-voltage equipment. The most challenging aspect of traditional PD recognition is extracting features from the discharge signal. Accordingly, this study applied the visual geometry group-19 (VGG-19) model to gas-insulated switchgear (GIS) PD image recognition. A high frequency current transformer and an LDP-5 inductive sensor measured PD electrical signals emitted by 15-kV GIS. Next, the Hilbert energy spectrum was obtained by Hilbert transform in the time and frequency domains. Compared with a phase-resolved PD pattern, the Hilbert spectrum can represent the energy and instantaneous frequency with the time variable. Finally, the VGG-19 model was applied for PD pattern recognition. For validation, its recognition performance was compared with that of a fractal theory by using a neural network method. The VGG-19 method is straightforward and has a high PD recognition rate, thereby enabling equipment manufacturers to quickly verify the insulation of GIS during assembly or operation.

Gearbox Fault Diagnosis Based on Gramian Angular Field and CSKD-ResNeXt

“…Because of its powerful feature extraction and classification ability in the face of complex data, CNN has been widely used in the field of fault identification [42]. However, gradient dispersion/explosion will occur in some networks when the depth of the network is increased, such as AlexNet [43] and VGG [44]. The proposal of BatchNorm can alleviate the gradient problem to a certain extent [45], but there is network degradation.…”