Insipient Insulation Fault Detection Using Phase-Resolved Partial Discharge Pattern Matching

The phenomenon of partial discharge in highvoltage equipment poses a significant threat to system reliability, emphasising the importance of early fault detection and detailed analysis. This paper presents CosNet, a deep learning model developed to identify pertinent features in 2D images of PhaseResolved Partial Discharge (PRPD) patterns and established templates. Subsequently, these features are evaluated using a cosine similarity function to assess the similarity between specific PRPD pulse patterns and these templates. Initial experiments of limited scope have shown that the developed method can successfully identify PRPD patterns resulting from stator slot defects in motors which frequently manifest themselves in a triangular shape. This initial success is very promising and more extensive testing is being planned to fully validate the methodology.