An Improved Nonparametric Method for Fault Detection of Induction Motors Based on the Statistics of the Fractional Moments

Accurate detection of faults of induction motors is an important challenge for industry. Often only stator currents/voltages are measurable and, hence, can be analyzed for fault detection. Among current analysis methods, the most attractive ones have low computational cost and little memory requirement for treatment of the measured data. Motor current signature analysis based on Fast Fourier Transform (FFT) provides fast computation, however it requires a long acquisition time interval for accurate fault detection. In fact, a short acquisition time causes the available frequency resolution of FFT spectrum to decrease and, hence, the spectra of the healthy and broken motors become indistinguishable. Recently, a statistical method, the so-called Statistics of Fractional Moments (SFM) allowed to distinguish signals with small differences. The statistically “close” signals are clustered and separated from “anomalous” signals. In this paper, the FFT and SFM are combined to reduce the acquisition time interval and hence the required memory and computation, while the fault detection capability is preserved. The effectiveness of the proposed approach is tested both on a healthy induction motor and on the same motor with broken rotor bars, for different acquisition time intervals.