Diagnosis potential of different partial discharge features of diverse PD defects in N2/SF6mixtures

The automated partial discharge diagnosis is an universal method to identify critical insulation conditions of electrical equipment. The diagnosis process needs significant information about the PD source by measuring the key values of the continuous PD activity, the so called Phase Resolved PD Pulse Sequence. From the information of such a measurement plenty of features can be calculated, which should distinguish the different defect types. This contribution discusses the main question: which type of PD features can distinguish the different PD defect types better than the other PD features? Therefore three different PD feature extraction methods: the phase position analysis, the voltage gradient and the voltage difference of consecutive pulses were tested with different PD defect types in GIS and GIL. In an automated diagnosis process all three feature extraction methods are combined with an L2 classification algorithm to calculate the diagnosis result. The question is discussed if a reference data base has to be adapted. Preliminary results have shown, that the performance of the feature extraction methods is strongly depending on the type of PD defect but also influenced by the gas mixture percentage.