Home > Research > Publications & Outputs > A Machine Learning Approach for Fault Detection...

Electronic data

  • IET-SMT.2018.5523

    Rights statement: This paper is a postprint of a paper submitted to and accepted for publication in IET Science Measurement and Technology and is subject to Institution of Engineering and Technology Copyright. The copy of record is available at the IET Digital Library.

    Accepted author manuscript, 2 MB, PDF document

    Available under license: CC BY-NC: Creative Commons Attribution-NonCommercial 4.0 International License

Links

Text available via DOI:

View graph of relations

A Machine Learning Approach for Fault Detection in Brushless Synchronous Generator Using Vibration Signals

Research output: Contribution to journalJournal article

Published
Close
<mark>Journal publication date</mark>1/08/2019
<mark>Journal</mark> IET Science, Measurement & Technology
Issue number6
Volume13
Number of pages10
Pages (from-to)852-861
Publication statusPublished
Early online date15/04/19
Original languageEnglish

Abstract

In order to maintain continuous production and to avoid the maintenance cost increment in power plants, it is important to monitor the condition of equipment, especially the generator. Regarding the impossibility of direct access to rotating diodes in brushless synchronous generators, the condition monitoring of these elements is very important. In this paper, a novel fault detection method is proposed for the diode rectifier of brushless synchronous generator. At the first stage of this method, the vibration signals are recorded and feature extraction is performed by calculating the relative energy of discrete wavelet transform components. Multiclass support vector machine (MSVM) is used for classification, and the best mother wavelet and number of decomposition level are chosen based on classification performance. To enhance the performance of the classification, a modified sequential forward subset selection approach is included by which the best statistical features are selected. In this approach, besides selecting the best subset of statistical features, the classification parameter is tuned according to the selected subset to achieve the best performance. The result of the proposed method is eventually compared with those results of classification performance using conventional subset selection. Experimental results show that the proposed method can detect rectifier faults effectively.

Bibliographic note

This paper is a postprint of a paper submitted to and accepted for publication in IET Science Measurement and Technology and is subject to Institution of Engineering and Technology Copyright. The copy of record is available at the IET Digital Library.