Automatic Early Broken-Rotor-Bar Detection and Classification Using Otsu Segmentation

Induction motors (IM) are susceptible to mechanical failures with severe consequences for production lines; hence, detection and classification of IM faults have been of great interest for researchers in last years. Broken rotor bars (BRB) are one of the most difficult faults to detect, since this fault does not give any indication of deterioration increasing significantly the production costs; hence, it is quite important to detect them in early states. Several methodologies have been proposed to extract information about the motor condition relying on motor-current-signature analysis (MCSA); however, they usually require high-computational-complexity algorithms to reach trustworthy result. In this work, a novel methodology for early detection and classification of BRB faults in IM is proposed. This methodology consists of obtaining two spectrograms using fixed-width windows, which are segmented through Otsu algorithm to visualize the time evolution of fault frequencies. The fault severity classification is performed through Kurtosis computation from non-stationary components. Obtained results from real experimentation validate the proposed-method high efficiency, reaching an overall 100% accuracy on detecting and classifying half, one, two BRBs, and healthy condition.