Artificial intelligence model for transformer fault diagnosis using a constructed database

L. Wang*, T. Littler, X. Liu

*Corresponding author for this work

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

Abstract

High capital investment in critical assets, in particular power transformers requires application of incisive diagnostics to determine performance and identify potential issues which may impact long-term operation and maintenance. Asset failure and replacement represent a worst-case scenario which can be avoided. This paper proposes a model for accurate transformer fault diagnosis using artificial intelligence (AI) methods. In recent years, AI models have evolved for fault classification using Dissolved Gas Analysis (DGA) data. In common practice, most DGA data are collated but less often labelled, thus restricting the development of diagnostic models with different learning approaches. To overcome this and related limitations, this paper proposes a new hybrid DGA method for unlabelled data to extract information to build a diagnostic database. The method uses the Rogers Ratio and Duval Triangle Methods to identify early and premature fault onset. The method also supports common utility asset practices to reduce transformer management costs by minimising damage from unnecessary invasive maintenance. The paper provides a comparative study of artificial neural network (ANN), support vector machine (SVM), K-Nearest Neighbour (KNN) and decision tree models using the constructed database. Comparative results indicate a high level of diagnostic precision with approximately 99% accuracy for classification using a KNN-based model.

Original languageEnglish
Title of host publicationThe 16th IET International Conference on AC and DC Power Transmission (ACDC 2020): proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1474-1479
Number of pages6
Publication statusPublished - 15 Sept 2021
Event16th IET International Conference on AC and DC Power Transmission, ACDC 2020 - Virtual, Online
Duration: 02 Jul 202003 Jul 2020

Conference

Conference16th IET International Conference on AC and DC Power Transmission, ACDC 2020
CityVirtual, Online
Period02/07/202003/07/2020

Bibliographical note

Funding Information:
The authors would like to express their gratitude to General Electric for collecting and providing DGA data.

Publisher Copyright:
© 2020 Institution of Engineering and Technology. All rights reserved.

Keywords

  • Artificial intelligence model
  • Dissolved gas analysis
  • Duval triangle method
  • Rogers ratio

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Artificial intelligence model for transformer fault diagnosis using a constructed database'. Together they form a unique fingerprint.

Cite this