Model-based and wavelet approaches to induction motor on-line fault detection

This paper compares model-based and signal-based approaches to induction motor fault detection. A Park model was used to describe the electrical variables. System faults such as stator winding failures or broken rotor bars are investigated. Parameter variations due to heating are considered. The identification of a single parameter, modelling the temperature effect, has allowed good detection within a classical analytical redundancy scheme. The wavelet transform is applied to the residuals rather than to the raw stator currents, in order to filter variations due to variable speed. An experimental benchmark (Leroy-Sommer, 1.1 kW) has allowed a practical comparison of the methods. (C) 2002 Elsevier Science Ltd. All rights reserved.