Intelligent diagnosis of mechanical fault of on-load tap-changer based on tensor decomposition in phase space

Abstract

Vibration signals associated with on-load tap-changer (OLTC) gear switching is closely related to its mechanical state. Based on the high-dimensional phase point spatial distribution of the vibration signal of OLTC, the vibration signals at multiple positions of OLTC are represented by tensor quantization to capture as rich as possible the mechanical status information of OLTC. Then, the third order tensor in the phase space is decomposed into Tucker tensor to obtain the core tensor, and a discriminative model of OLTC mechanical fault based on convolutional neural network is established. Taking the vibration signal of a certain CM type OLTC as an example for analysis, the results show that the phase space core tensor information of the vibration signal of OLTC is comprehensive and less redundant when the OLTC acts. The mechanical fault diagnosis model based on the convolutional neural network has good performance, with an accuracy rate of more than 95%, which can provide a reference for fault identification and condition maintenance of OLTC.

Keywords

• on-load tap-changer
• vibration signal
• fault diagnosis
• phase space reconstruction
• tensor decomposition
• convolutional neural network