Multiple partial discharge source discrimination in a high voltage transformer winding

Partial discharge (PD) analysis is an important technique for diagnosis and online monitoring of transformer insulation systems. PD within transformer windings may be due to several causes such as manufacturing defects, degradation of the primary insulation or contamination of the oil. The degradation processes occurring in dielectric insulation components can lead to development of different types of PD source [1], [2]. Multiple PD sources induced by different defects can be simultaneously present within the transformer winding. Therefore, it is necessary to develop tools to separate measurement data from multiple PD sources in order to facilitate separate PD source identification and location to allow accurate condition assessment of the winding. Conventional techniques for multiple PD source separation are based on analysis of the phase resolved PD (PRPD) pattern [3], [4]. However, in the presence of multiple PD sources, PRPD patterns generated by different PD sources can significantly overlap each other [5], [6]. By using PRPD pattern-based techniques, multiple PD sources cannot easily be separated as one PD source can be entirely immersed in the patterns of the other sources. Recently, there have been a number of techniques reported for separating multiple PD sources such as through the use of support vector machines (SVMs) or artificial neural networks (ANNs) [3], [7].This paper reports on a technique that relies on an initial assumption that different PD sources generate different waveform characteristics [8]. The measured discharge current signals from both ends of a transformer winding were decomposed using the Mathematical Morphology (MM) decomposition technique. Then, the energy distribution for each signal for each structure element (SE) length of MM was determined. The OPTICS algorithm was then used as a density based clustering technique to reveal different clusters of signals with similar calculated energy distributions.