Improvement of Transformer Fault Diagnosis using Fuzzy Rule and Decision Tree Based on Dissolved Gas Analysis

Abstract

Early and correct diagnosis of faults in power transformers (PTs) are important aspects of electrical system maintenance. In addition to insulation and cooling functions, insulating oil contains the by-products of degradation and ageing reactions of the insulation system and related components inside the PT. In addition to sludge, water and acids, gaseous products are also generated within the transformer. Dissolved gas analysis (DGA) based on the identity and quantity of the generated gases is the most widely used technique for the early detection of faults in the active parts of PTs. In this paper, fuzzy rule (FR) and the decision tree (DT) algorithms are used for PT fault diagnosis. The ratios of Roger’s four ratios and IEC 60599 methods were used as input feature vectors. The proposed methods were carried out using 168 data samples and tested on 72 data samples. The performance of the proposed diagnostic methods was evaluated and compared to the IEC 60599 and Roger’s four ratios methods. From the results obtained, with a diagnostic accuracy of 95.83%, the best performance is obtained with the FR classifier using the Log of Rogers ratios as input vector.