Feature analysis of acoustic emission signals in Time-Frequency representation from partial discharge sources using Self-Organizing Map

Electrical systems throughout the world are experiencing problems with aging insulation. One of the major components, which degrade under stresses, is the paper insulation of a power transformer. In this research work, analysis of acoustic emission (AE) signal patterns due to the occurrence of partial discharge (PD) has been carried out. In order to characterize the different types of PD sources, seven descriptors or features were extracted from the Short-Time Fourier Transform spectrogram of the AE signals. These descriptors are used to generate the Self-Organizing Map (SOM) for three different types of PD sources. From the component planes of the descriptors produced by the SOM map, the analysis of the features was carried out and the best features to represent the AE signal patterns in the Time-Frequency representation were selected.