Hybrid Condition Monitoring System for Power Transformer Fault Diagnosis

Abstract

The important parts of a transformer, such as the core, windings, and insulation materials, are in the oil-filled tank. It is difficult to detect faults in these materials in a closed area. Dissolved Gas Analysis (DGA)-based fault diagnosis methods predict a fault that may occur in the transformer and take the necessary precautions before the fault grows. Although these fault diagnosis methods have an accuracy of over 95%, their validity is controversial since limited data are used in the studies. The success rates and reliability of fault diagnosis methods in transformers, one of the most important pieces of power systems equipment, should be increased. In this study, a hybrid fault diagnosis system is designed using DGA-based methods and Fuzzy Logic. A mathematical approach and support vector machines (SVMs) were used as decision-making methods in the hybrid fault diagnosis systems. The results of tests performed with 317 real fault data sets relating to transformers showed accuracy of 95.58% using a mathematical approach and 96.23% using SVMs.