Subspecies analysis of partial discharge data from a generator stator conductor

Partial discharge (PD) pattern recognition systems are often based on analysis of the data structure as a whole. This approach may not explain the underlying discharge processes. An alternative concept is proposed which is based on the examination and analysis of the subspecies identifiable in the data. PD data obtained from a generator stator conductor are used to demonstrate this approach. Using well-behaved data it is possible to identify two distinctive types of discharges originating from a generator stator conductor. One type has properties ascribable to surface discharges, whereas the other type exhibits properties akin to those from internal discharges. The approach allows ill-behaved PD data to be segmented in a supervised manner using knowledge about the likely behaviour of different types of PD. This permits individual PD generating mechanisms to be monitored, which may be difficult to achieve through analysis of the overall magnitude and phase distributions. Using the method, it is demonstrated that the surface discharges have a magnitude distribution whose shape is influenced by a multivalued relationship with the phase distribution. These relationships are obscured if magnitude distribution only is used for the analysis but are observable using subspecies analysis.