This article has an erratum: [https://doi.org/10.1051/epjap:2008169]
Eur. Phys. J. Appl. Phys.
Volume 43, Number 2, August 2008
Reliability in Electromagnetic Systems (IET)
|Page(s)||245 - 251|
|Section||Instrumentation and Metrology|
|Published online||19 July 2008|
Application of classification methods in fault detection and diagnosis of inverter fed induction machine drive: a trend towards reliability*
L2S, CNRS UMR 8506, Supélec, Université Paris Sud 11, Paris, France
2 LGEP/Spee Labs, CNRS UMR 8507, Supélec, Université Paris Sud 11, Université Pierre et Marie Curie–Paris 6, France
3 LBMS, IUT de Brest, Electrical Engineering Department, University of Western Brittany, Rue de Kergoat, BP 93169, 29231 Brest Cedex 3, France
Corresponding author: Claude.Delpha@lss.supelec.fr
Revised: 1 April 2008
Accepted: 30 May 2008
Published online: 19 July 2008
The aim of this paper is to present a method of detection and isolation of intermittent misfiring in power switches of a three phase inverter feeding an induction machine drive. The detection and diagnosis procedure is based solely on the output currents of the inverter flowing into the machine windings. The measured currents are transformed in the two dimensional frame obtained with the Concordia transform. The data are then treated by a time-average method. The results even promising lack of accuracy mainly in the fault isolation step. To enhance the fault detection and diagnosis by the use of the information enclosed in the data, a Principal Component Analysis classifier is applied. The detection of a fault occurrence is made by a two-class classifier. The isolation is a two-step approach which uses the Linear Discriminant Analysis; the first is to identify the faulty leg with a three-class classifier and the second one discriminates the faulty power switch. Both methods are evaluated with experimental data and pattern recognition method proves its effectiveness and accuracy in the faulty leg detection and isolation.
PACS: 02.50.Sk – Multivariate analysis
© EDP Sciences, 2008
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.