Soft Computing -- Applied to Transformer Incipient Fault Diagnosis

Transformer fault diagnostic methods based on dissolved gas analysis (DGA) using artificial neural network (ANN), fuzzy logic & a self-adaptive neuro-fuzzy inference system (ANFIS) has been presented in this paper. Firstly, a fuzzy system development technique is proposed, in which the rule base is derived from IEEE and IEC DGA standards. The second step of the work deals with the application of ANN for DGA interpretation where several data samples of different transformers have been selected very carefully in order to extract known as well as unknown diagnosis correlations implicitly and these samples are used for ANN training and validation. In third step a self ANFIS is developed for easy inference. The same data set as was used in ANN is used for ANFIS training. Finally, the results obtained from these three methods are compared with the actual results and analysis is presented stating the effectiveness of each method. Test results show that ANN can be used in combination with fuzzy logic to implement more complex diagnostic methods while maintaining a straightforward relationship between the enhanced method and the original one.