Partial discharge diagnosis using statistical optimization on a PC-based system

Personal computer (PC)-aided partial discharge (PD) evaluation needs high-speed electronic devices for on-line measurement and digital conversion of PD signals. The authors include quantities to be measured to evaluate PD signals, which are characterized by a statistical scatter of magnitude and duration and are additionally influenced by noise and a complex time behavior. Using the correct algorithms and parameters, the PC gains some intelligence to discriminate unknown defects. Artificial sample defects, representing sources of PD, have been implanted into a GIS system. By performing a statistical analysis of charge, energy and phase angle on the measured signals, it is possible to solve the diagnosis problem by a noise resistant software solution. The final diagnosis is carried out by pattern recognition, using a specially calculated identification data set, which is compared to reference patterns measured earlier. The strategy takes into account the fundamental differences in the physics of discernible defects