An expert system for EMI data classification based on complex Bispectrum representation and deep learning methods

Imene Mitiche*, Mark D. Jenkins, Philip Boreham, Alan Nesbitt, Gordon Morison

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

This paper presents expert system framework based on Machine Learning (ML) for High-Voltage (HV) asset condition monitoring. The work investigates the classification of insulation faults in HV environment, based on real-world time series signals labelled by condition monitoring experts. Extending on our previous work, the proposed approach exploits the Bispectrum analysis and deep learning for feature extraction and classification. The calculated Bispectrum on time series signals can be deployed as the complex-valued Bispectrum, which contains phase information, or as its real-valued magnitude. This can be approached as an image classification problem which can be implemented in various deep networks including Convolutional Neural Network (CNN), Residual Neural Network (ResNet) and their complex-valued version. The employed deep networks performance is compared in terms of their classification accuracy. High classification performance is obtained which produces comparable performance with expert diagnosis. Thus, it can be interpreted as transfer of expert system to an intelligent system.
Original languageEnglish
Article number14568
Number of pages9
JournalExpert Systems with Applications
Volume171
Early online date7 Jan 2021
DOIs
Publication statusPublished - 1 Jun 2021

Keywords

  • Complex and real-valued Bispectrum
  • Condition monitoring
  • Deep learning
  • EMI diagnostic
  • Expert system
  • Insulation faults

Fingerprint

Dive into the research topics of 'An expert system for EMI data classification based on complex Bispectrum representation and deep learning methods'. Together they form a unique fingerprint.

Cite this