Using counterpropagation neural networks for partial discharge diagnosis

In high voltage engineering, various methods of non-destructive fault diagn osis are applied for investigating the quality of insulating materials and systems. The methods are aimed at classifying patterns derived from the mea sured characteristics of the electrical signals typically resulting from in sulation defects. In this paper, variants of the counterpropagation neural network architecture are wed to classify patterns representing various prop erties of partial discharges. It is shown that the classification quality c an be improved considerably when an extended counterpropagation network wit h a dynamically changing network topology, and an additional vigilance unit for monitoring the behaviour of the network during the learning phase is a pplied. The extended network has significant advantages over the standard c ounterpropagation network in cases where outliers in the training data seri ously degrade the approximation quality of the standard network. When using the proposed network in conjunction with physically motivated discharge da ta, input patterns from defect categories not considered during training ca n be rejected more reliably This rejection problem is particularly importan t for practical applications where misclassifications cannot be tolerated.