Squirrel-Cage Induction Motor Malfunction Detection Using Computational Intelligence Methods

The squirrel-cage induction motors (commonly called just electric motors) are widely used in electromechanical devices. They usually act as a source of mechanical power for different types of industrial machines. There is a natural life cycle of such electric motors ending in malfunction caused by damage of particular electric or mechanical parts. Sudden and unforeseen engine failure may turn out to be a heavy cost for the company. Early detection of motor damage can minimize repair costs. In this work a machine-learning based methodology for early motor malfunction detection is presented. A test stand with a three-phase induction motor that can simulate various types of stator winding short-circuit faults under load controlled by a DC generator was build. This stand was equipped with multiple sensors for continuous monitoring. Readings from sensors were collected for different loads and types of damage. Multiple methods of preprocessing and classification were tested. Sensors types are evaluated for accuracy of malfunction recognition based on the results of computational experiments. The 5-fold stratified cross-validation was used for evaluation of preprocessing steps and classifiers. The best results were achieved for neutral voltage, axial flux, and torque sensors. Acquisition time of 0.16 s is sufficient for accurate classification.