Affiliations: [a] Department of Electrical Engineering, University of
Constantine, Constantine, Algeria | [b] Department of Electrical Engineering, University of
Skikda, Skikda, Algeria | Université Laval, Canada | Tohoku University, Japan
Correspondence:
[*]
Corresponding author: Ilhem Bouchareb, LGEC, Department of
Electrical Engineering, University of Constantine 1, Constantine, Algeria.
E-mail: ilhembouchareb@yahoo.fr
Abstract: In the case of one phase failure, the switched reluctance motor
(SRM) will behave nearly the same, both in open circuit and in short circuit
failure. This means, that the machine will understand the two faults in the
same way which makes the SRM faults detection and diagnosis a more challenging
task. This paper presents a diagnosis method based on pattern recognition
analysis to detect and to classify automatically the electrical faults, short-
and open-circuit under any level of load of the studied system: redundant
three-phase power converter fed 6/4 SRM. The phases making a pattern
recognition diagnosis of SRM, the training and the decision. The training phase
consists in determining the pattern vector and the optimal kernels design (the
separating classes) by Time-Frequency Representation (TFR). The training data
is carried out using a set of fault scenarios, between healthy, single and
combined faults, in terms of torque measurement at different load level, in
order to deduce the fault severity. The second phase, consists in associating
an unknown pattern with one of the defined classes, according to the "k-nearest
neighbors" (knn) decision rule, associated with Kalman estimator to tracking of
various operating modes and to predict the evolution of the call out of the
knowledge database for a given operating mode in order to realize a preventive
maintenance. The experimental results prove the efficiency of pattern
recognition methods in condition monitoring of reluctance machine.
