Performance evaluation of dissolved gas analysis techniques against measurement errors

Abstract

Fault diagnosing power transformers based on dissolved gas analysis (DGA) has become an important and a significant tool. There are a number of techniques developed for DGA that are represented extensively in literature. In practical application of DGA, there is a degree of measurement errors in obtained data. These errors are produced from inaccuracy in measurement system, environmental impact, and human errors. In the present study, it is aimed to investigate the sensitivity of different DGA techniques against measurement errors. The considered DGA techniques are IEC Ratio, Duval Triangle, and Pentagon shape. A total number of 380 actual samples were obtained from the Egyptian Electricity Network as well as published reports. Measurement errors are modeled with various levels as a percentage of original data using a random function. The three different DGA techniques are applied for error data and the corresponding diagnostic accuracy is evaluated. The techniques are compared on two stages. The first stage includes the comparison for only main faults, thermal, arcing and partial discharge. The second stage compares the diagnostic accuracy for detailed faults, such as low thermal, medium thermal, high thermal, and so on. Based on the results, the most robust technique is adopted.