Characterization of partial discharge signals

Abstract

The challenge to effectively and accurately determine pure partial discharge (PD) signals from the large amount of noise still remains. In this study, individual PD pulses were filtered, extracted and analyzed using digital signal processing techniques and data mining methods. The shape or distribution of the spectral frequency domain could be correlated with different PD signals. Feature extraction was explored using K-means clustering to categorize the similarities. A hard threshold method was applied to the time domain in which the critical PD pulses could be identified based on extracted features. A pre-determined threshold value was set and PD occurrences could be found and classified for fault diagnosis.

Combined Approach Using Clustering-Random Forest to Evaluate Partial Discharge Patterns in Hydro Generators

“…An overview on partial discharges in high voltage equipment using PD raw data was performed in [11] using k-means techniques to cluster signals. In another approach, neural networks were applied to classify partial discharges' signals into six types using real measurement data from high-voltage motors [12].…”

A novel approach towards the determination of the time to breakdown of electrical machine insulating materials

“…Second area is PD signal processing and characterization. This aspect of the field is taken up in [12][13][14][15][16][17][18][19] and many other papers. Third area is PD data analysis techniques for diagnostics and prognostics [20][21][22][23].…”