Power transformers are most critical and expensive equipment in electrical power system. The onset of electrical discharges or thermal stresses in mineral oil or cellulose insulation of a power transformer can cause the degradation of insulation quality and ultimate failure of transformer. To avoid such a scenario, it is necessary to periodically monitor the health of transformer to keep them in satisfactory working condition. Dissolved Gas Analysis (DGA) has gained more popularity among the methods available to monitor the transformer’s health. It shows the status of the transformer accurately and provides information regarding the faults that are appearing inside the transformer. Dissolved gases can be extracted and identified by appropriate analytical technique. Seven gases are found to be key gases such as H2, CH4, C2H6, C2H4, C2H2, CO and CO2. The conventional ratio methods may not converge into correct decision in some cases and the rule-based expert systems could not learn from unidentified fault data. In this paper an attempt has been made to diagnose the faults of power transformer by using Artificial Neural Network (ANN). A two-scheme ANN is proposed to diagnose the condition (normal/abnormal) and major incipient faults. Gradient Descent (Gd), Scale Conjugate Gradient (SCG) and Levenberg-Marquardt (LM) training algorithm are used to train the ANN and compared their performance. LM algorithm converges much faster than other training algorithm. The proposed ANN has given reasonable success rate than conventional diagnostic methods.
