Fault Diagnosis of Inverter Current Sensor Using Artificial Neural Network Considering Out-of-distribution

Jae Hoon Shim, Kahyun Lee, Jung Ik Ha

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

This paper presents an Artificial Neural Network (ANN)-based approach for effectively diagnosing current sensor faults in inverters under various operating conditions. In this approach, additional techniques to make a robust fault classifier for untrained out-of-distribution (OOD) data are considered. The ANN is trained and evaluated through the Python TensorFlow library. The fault diagnosis accuracy is measured using the confusion matrix on the in-distribution dataset (current sensor faults data), and the degree of performance improvement on the OOD dataset is evaluated through the Area Under the Receiver Operating Characteristic (AUROC). The proposed method has been verified to have a classification accuracy of 98.94% on the in-distribution dataset and improve the AUROC from 0.9170 to 0.9866.

Original languageEnglish
Title of host publicationProceedings of the Energy Conversion Congress and Exposition - Asia, ECCE Asia 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1520-1525
Number of pages6
ISBN (Electronic)9781728163444
DOIs
StatePublished - 24 May 2021
Event12th IEEE Energy Conversion Congress and Exposition - Asia, ECCE Asia 2021 - Virtual, Singapore, Singapore
Duration: 24 May 202127 May 2021

Publication series

NameProceedings of the Energy Conversion Congress and Exposition - Asia, ECCE Asia 2021

Conference

Conference12th IEEE Energy Conversion Congress and Exposition - Asia, ECCE Asia 2021
Country/TerritorySingapore
CityVirtual, Singapore
Period24/05/2127/05/21

Keywords

  • artificial neural network
  • fault detection
  • fault diagnosis
  • in-distribution data
  • out-of distribution data

Fingerprint

Dive into the research topics of 'Fault Diagnosis of Inverter Current Sensor Using Artificial Neural Network Considering Out-of-distribution'. Together they form a unique fingerprint.

Cite this