Transformer Incipient Fault Detection Using Energy-Weighted DGA data and ANN Based Binary Classifier

The power transformer is the crucial asset in any electrical system. The health and condition monitoring of power transformers therefore becomes very important for the early detection of incipient faults. The conventional methods based on Dissolved Gas Analysis (DGA) overlook the amount of formation energy required for the production of each gas under faulty conditions and solely rely on the concentration of gases for the identification of a faults in a power transformer. This can affect the diagnostic performance of the fault detection methods leading to inaccurate results. In this paper, the concept of multistage binary classification of faults is utilized with the help of Artificial Neural Network (ANN). To enhance the detection performance, the ANN is trained using chemically weighted DGA data (IEC 60599) which is obtained with the help of weighing factor of each faulty gas. ANN binary classifier trained with weighted DGA data exhibited superior performance in terms of detection capability over ANN trained with non-weighted DGA data.