A novel visualisation technique for dissolved gas analysis datasets

Abstract

Dissolved gas analysis is the most widely used diagnostic test in power transformers. There are established methods used in industry for interpreting DGA results. Among these are the IEEE Key Gas Method, Rogers' Ratios and the Duval Triangle. However, collectively these methods can lead to conflicting results or unclassifiable measurements. This paper presents a visualization technique for interpreting DGA results to mitigate these effects, based on Kernel Principal Component Analysis. DGA measurements from more than 200 power transformers are used to validate the approach.