Transformer incipient fault diagnosis on the basis of energy-weighted DGA using an arti cial neural network


Abstract: In this paper, a transformer incipient fault diagnosis model has been developed with the help of an artificial neural network (ANN), taking into account the difference in the energy required to produce the different fault gases. The key fault gases are indicative of the fault type prevailing in the transformer. However, in conventional studies, the energy difference in fault gas formation is not considered while adopting the key gas method for fault diagnosis. In this work, a weighting factor has been used to take into account this relative difference in energy requirement for various fault gas formations. The fault gas concentrations have been suitably weighted by their respective weighting factors before being used in the incipient fault diagnosis process. A backpropagation ANN has been appropriately trained using the weighted fault gas concentration for transformer incipient fault identification. The model has been trained to identify fault types as enlisted in the transformer fault-interpreting standard IEC-599. The developed ANN model has been tested for its diagnostic capability using a reported fault database. The comparative diagnosis results presented here show clear improvement in the diagnosis of transformer internal faults using the energy-weighted ANN model over the unweighted ANN model.

Keywords: Power transformer, fault diagnosis, dissolved gas analysis, arti cial neural networks

Full Text: PDF