The use of neural network and statistical techniques for the identification of partial electrical discharges

Summary form only given. Over the past few years the availability and cost of computing power has made it possible for researchers in the field of dielectric behavior and insulation design to acquire large amounts of data and examine the results in a practical time scale. One area that has benefited from this approach is that of the detection, measurement and identification of partial electrical discharges (PDs). Their interpretation is of importance as it can be an indicator of the ageing of the insulation as well as being an aggressive ageing agent itself. A satisfactory statistical analysis technique has been developed and used, both in the laboratory and on site, to examine/interpret PD measurements and present the results in a simple form showing the most probable fault/s or defects/s responsible for them. New research is now underway to use the existing techniques and results to develop a neural network programme that can 'learn' to recognize patterns and automatically identify the cause even in electrically 'noisy' environments. This paper presents and discusses results form these two areas of work and assesses their future usefulness with particular reference to the practicing engineer.