Improving the accuracy of transformer DGA diagnosis in the presence of conflicting evidence

Jose Ignacio Aizpurua, Victoria M. Catterson, Brian G. Stewart, Stephen D. J. McArthur, Brandon Lambert, Bismark Ampofo, Gavin Pereira, James G. Cross

Research output: Chapter in Book/Report/Conference proceedingConference contribution book

4 Citations (Scopus)
64 Downloads (Pure)


Transformers are critical assets for the reliable and cost-effective operation of the power grid. Transformers may fail if condition monitoring does not identify degraded conditions in time. Dissolved Gas Analysis (DGA) focuses on the examination of the dissolved gasses in the transformer oil and there exist different methods for transformer fault diagnosis based on different analyses of the gassing levels. However, these methods can give conflicting results, and it is not always clear which model is most accurate in a given situation. This paper presents a novel evidence combination framework for DGA based on Bayesian networks. Bayesian network models embed expert knowledge along with learned data patterns and evidence combination which aids in the consistency of analysis. The effectiveness of the proposed framework is validated using the IEC TC 10 dataset with a maximum diagnosis accuracy of 88.3%.
Original languageEnglish
Title of host publication2017 IEEE Electrical Insulation Conference (EIC)
Place of PublicationPiscataway, NJ
Number of pages4
ISBN (Print)9781509039654
Publication statusPublished - 18 Aug 2017
EventIEEE Electrical Insulation Conference 2017 - Sheraton, Baltimore, United States
Duration: 11 Jun 201714 Jun 2017


ConferenceIEEE Electrical Insulation Conference 2017
Abbreviated titleEIC 2017
Country/TerritoryUnited States
Internet address


  • dissolved gas analysis
  • DGA
  • transformer diagnosis
  • condition monitoring
  • Bayesian networks
  • evidence combination
  • ensembles


Dive into the research topics of 'Improving the accuracy of transformer DGA diagnosis in the presence of conflicting evidence'. Together they form a unique fingerprint.

Cite this