Artificial neural network (ANN) based implementation of Duval pentagon

Duval pentagon is recently proposed interpretation technique of dissolved gas analysis (DGA) data for condition monitoring of oil-immersed transformer. This technique is the extension of well known Duval triangle method of DGA. Duval triangle method does not include Hydrogen (H2) and Ethane (C2H6) which are important gases for diagnosis of partial discharge and low thermal fault. Duval pentagon method includes all five hydrocarbon gases in one graphical representation. In this work, performance analysis of the technique is carried out using published fault database. The accuracy is found to be 87%. Further, Duval pentagon is implemented using ANN to enhance the diagnostic capability and to facilitate online monitoring. Proposed ANN is trained as per Duval pentagon method to detect transformers fault when data point belongs to some fault zone. The results presented in this paper show that the proposed ANN based model can reliably be used for transformer incipient fault diagnosis with enhanced diagnostic capability. The diagnostic accuracy of the proposed ANN based implementation is around 92%.