Diagnosis of GIS Mechanical Faults in Noisy Environments Using Blind Source Separation and NSVDD

Recently, the mechanical defects diagnosis method based on the audible sound for GIS equipment is proposed and investigated. In practical application, the environment is relatively complex, and there may be noise such as human speech, partial discharge from other devices, etc. These environmental noises may overlap with the GIS operation sound signal, which may affect the discrimination of GIS mechanical status. Therefore, it is necessary to separate the sound signals and noise before diagnosis. This chapter first uses the FastICA algorithm based on negative entropy maximization to achieve blind separation of various sound source signals. In response to the signal disorder and amplitude uncertainty problems of the FastICA algorithm, variational mode decomposition is used to decompose each separated signal, extract entropy features from the decomposition results, and then use support vector data description with negative samples to identify the mechanical state of GIS. Finally, the proposed method is verified through simulation and field experiments data.