A Neural Network Based System for Prediction of Partial Discharge Pulse Height Distribution Parameters

Partial discharges (PDs) have been traditionally used to monitor tree growth in electrical insulation. In this work Perspex (PMMA) samples with a needle plane gap have been aged with AC voltage. The tree growth is monitored by collecting PDs at regular intervals of time and by taking microphotographs in real time without interrupting the aging voltage. The PD pulse amplitude records are clustered together into groups of class intervals. The sequence of PD pulse height records are quantified as time series of eta (shape) and sigma (scale) of a Weibull distribution. Artificial neural network approach is used for analyses and prediction of eta and sigma. This is applied for two samples A and B. The relative advantages and limitations of this approach are discussed.