Condition diagnostics of a physical breakdown mechanism in high voltage dielectrics utilising `AI´ evaluation techniques

Regions of discordant electrical withstand stress within dielectrics, produce an intermittent electrical breakdown mechanism known as partial discharge. By way of progressive deterioration and ultimately a failure of the dielectric, partial electrical discharge (pd) and the associated effects on dielectric materials have been studied extensively over several decades. As the condition of a dielectric is of extreme importance to maintain supply stability, degradation must be identified and rectified at an early stage. Discharges evident of such deterioration, are attributable to the very components of origin and thus illustrative of root. Contemporary instruments and techniques have been developed that enable more detailed studies of material degradation. Researchers are able to extract information understanding mechanisms of material degradation, not easily achieved using conventional instrumentation. With intensifying demands on already life extended transmission equipment, the likelihood of impending dielectric failures increases proportionally. By conducting early analysis of pd behaviour and collating historical behaviour patterns, an accurate assessment of the dielectric condition and a subsequent remaining life prediction may be deduced. Neural networks have been extensively exploited in investigations as trainable pattern classifiers, where there has been a need to characterise a voltage, or current signature of a physical phenomena. Presented with representative and characteristic signatures, neural networks provide potential methods by which partial discharges may be categorised against factors of source. The subject of the paper is the combination of partial discharge techniques with neural network analysis for the remote diagnosis of faults in dielectrics