A partial discharge based defect-diagnosis system for cast-resin current transformers

The paper presents an application of an artificial neural network to partial discharge (PD) recognition in current transformers. The research contents of interest encompass the practical measurements and recognition of partial discharge signals. First, to yield four experimental models of partial discharge for testing, we make use of cast-resin current transformers tailor-made with insulating defects. Then, using a commercial partial discharge detector, practical measurements of 3D patterns for the above experimental models are performed in a magnetically shielded laboratory. The 3D patterns obtained from discharge measurements are, after appropriately preprocessing, used for the training of a backpropagation neural network (BPN), used as a partial discharge based defect-diagnosis system. Finally, with a view to exploring applicability in the field, different levels of noise are selected randomly to distort the original measurements. These distorted data sets are entered for testing. Research results show that, with 20% noise per discharge count, an 80% successful recognition rate is achieved.