Prediction of rotor slot size variations in induction motor using multimodal sensor signal and machine learning

Abstract

newline Condition monitoring of the rotor lamination surface and rotor slots of an Induction Motor (IM) is performed through the proposed Average Rotor Slot Size Variation (ARSSV) prediction methods such as DWT-MLR, NDWTMLR, TQWT-MLR,PCT-PR, RaDWT-LR and ORaDWT-LR during the runtime of the IM. The expansion of the rotor slot size of IM is due to high magnetic flux intensity and magnetic stress created by thermal stress. The Stretching and Curving (SC) magnetic flux on the rotor lamination sheet leads to high flux intensity. The thermal stress is generated due to overcurrent and produces heat on the rotor winding, parallel to the rotor lamination, which then expands the rotor slots on the rotor lamination. Until now, the magnetic stress based ARSSV prediction has never been measured or monitored. ARSSV due to magnetic and thermal stress on the rotor is predicted using multimodal sensor signals from the Giant Magneto Resistance (GMR), temperature, current, and vibration sensors. The multimodal sensor signals are processed with different transforms such as Discrete Wavelet Transform (DWT), Non-Decimated Wavelet Transform (NDWT), Tunable Q Wavelet Transform (TQWT), Polynomial Chirplet Transform (PCT), Rational Dilation Wavelet Transform (RaDWT), Over Complete Rational Dilation Wavelet Transform (ORaDWT) and obtained mean/energy values from sensor signals. The mean/energy band of multimodal sensor signal values and Machine Learning (ML) algorithms are used for ARSSV measurement. Manual measurement of ARSSV on the rotor lamination surface is performed through microscopic camera images. The automatic prediction of ARSSV on the rotor lamination surface is computed using the energy band values of the sensor signal and manually measured rotor slot values. From experimental analysis, ARSSV has more than 2% of the rotor lamination surface produces serious damage to the rotor through vibration, harmonics, and sparking. The ARSSV prediction accuracy is about 95.6% compared to manual measurements from microscopic camera i