Automatic Classification of Winding Asymmetries in Wound Rotor Induction Motors Based on Bicoherence and Fuzzy C-Means Algorithms of Stray Flux Signals

• Published in: IEEE transactions on industry applications Vol. 57; no. 6; pp. 5876 - 5886
• Main Authors: Iglesias Martinez, Miguel Enrique, Antonino-Daviu, Jose A., de Cordoba, Pedro Fernandez, Conejero, J. Alberto, Dunai, Larisa
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.11.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Anomalies; Asymmetries; Asymmetry; Automatic classification; bicoherence; Classification; Failure detection; Fault diagnosis; Fourier transforms; Fuzzy C-Means; Induction motors; Industrial applications; Machine learning; Machinery condition monitoring; Position sensing; Resistance; Rotors; Time-frequency analysis; Winding; Windings
• ISSN: 0093-9994; 1939-9367; 1939-9367
• DOI: 10.1109/TIA.2021.3108413

Wound rotor induction motors are used in a certain number of industrial applications due to their interesting advantages, such as the possibility of inserting external rheostats in series with the rotor winding to enhance the torque characteristics under starting and to decrease the high inrush currents. However, the more complex structure of the rotor winding, compared to cage induction motors, is a source for potential maintenance problems. In this regard, several anomalies can lead to the occurrence of asymmetries in the rotor winding that may yield terrible repercussions for the machine's integrity. Therefore, monitoring the levels of asymmetry in the rotor winding is of paramount importance to ensure the correct operation of the motor. This work proposes the use of bicoherence of the stray flux signal, as an indicator to obtain an automatic classification of the rotor winding condition. For this, the Fuzzy C-Means machine learning algorithm is used, which starts with the bicoherence calculation and generates the different clusters for grouping and classification, according to the level of winding asymmetry. In addition, an analysis regarding the influence of the flux sensor position on the automatic classification and the failure detection is carried out. The results are highly satisfactory and prove the potential of the method for its future incorporation in autonomous condition monitoring systems that can be satisfactorily applied to determine the health of these machines.