A novel demagnetization fault detection of brushless DC motors based on current time-series features

This paper proposes a robust time-series feature extraction of the currents for permanent magnet (PM) defect classification in brushless DC (bLDC) motors. Thanks to the six-step operation of the BLDC motors, current waveform can be precisely represented by its extrema without loss of too much information, resulting in more computational benefits and less required memory. Effect of PM demagnetization on these extrema analytically investigated here. In addition to the extrema features, autoregressive coefficients and K-means distances are used as auxiliary features. Then, these patterns are utilized for classification of windowed current time-series in order to detect PM defects in the machine. Each window of the current waveform is classified; then final decision is based on majority voting between the decisions for each segment. Segmentation along with majority voting makes this fault detection scheme more robust to noise and external disturbances such as load oscillations due to required short-time for capturing waveforms.