Enabling intelligent HVSR loose fault diagnosis based on vibration perception

Abstract

The significance of high-voltage shunt reactors (HVSRs) in ensuring the reliability of power grids is widely acknowledged. Nevertheless, the persistent mechanical vibrations experienced by HVSR can loosen vital components such as windings, iron cores, and bolt fasteners. If the mechanical loose fault cannot be detected and mitigated timely, the vibration will be further aggravated, which tends to cause additional detects such as overheating and discharge, even fire disasters. Therefore, it is essential to diagnose loose faults in high-voltage shunt reactors. This study presents a unique data-driven method that utilizes optimal vibration test location selection and a densely connected neural network (DenseNet) for diagnosing loose faults in shunt reactors. Firstly, we integrate the least square method and R-square to explore the fault sensitivity of the vibration signal collected in different locations on the external tank surface, based on which the optimal vibration test points can be determined. Then the vibration data acquired in the selected points are input into the DenseNet model to achieve high-precise loose fault detection and diagnosis for high-voltage shunt reactor. Furthermore, with test data collected from a 35 kV high-voltage shunt reactor, the diagnostic performance of the proposed method is verified and highlighted.