Vibration-Based High Voltage Shunt Reactor Fault Diagnosis with Incep-DenseNet

Abstract

It is of great significance to perform high voltage shunt reactor (HVSR) fault diagnosis and take appropriate action to strengthen HVSR reliability and durability. In general, vibration signal is commonly used for HVSR fault diagnosis. However, with development of HVSR faults like internal screw bolt loose, its vibration signal will show subtle difference, and only limited studies have been devoted to identify various fault degree. In this paper, a novel densely connected neural network defined as Incep-DenseNet is proposed for diagnosing various HVSR internal screw bolt loose faults, which integrates advantages of InceptionNet and DenseNet to extract more specific and robust features from HVSR vibration signal. In the analysis, the collected HVSR vibration signal is transformed into 2D image data, which is then used to train the Incep-DenseNet. Results demonstrate that with the trained Incep-DenseNet, the diagnostic accuracy for four different HVSR internal screw bolt loose faults can reach 94.7%.