Cluster Analysis of Partial Discharge Based on Image Edge Detection

Partial discharge (PD) measurement is one of the most important diagnostic methods to detect local faults in high voltage cable. This requires effective means of PD detection and recognition. The paper describes a method of PD recognition by image processing. Three kinds of cable-joint fault models are made for PD test with high voltage. Large amounts of test data are got for drawing scatter map or spectrum that can reflect the characteristic of PD. There is a certain extent difference in various types of PD spectrum. Spectrums of ?-q and TF are converted to gray scale. The edge detection by mathematical morphology is used for cluster analysis. Different types of PD signal and interference signal are separated. Higher recognition accuracy rate is obtained when judging the separated signal.