Cascade based methods in detecting rotating faults using vibration measurements

摘要

In the paper, the pursued objective is to take advantage of two main relevant cascade methods, namely Ensemble Empirical Mode Decomposition (EEMD) and Discrete Wavelet Transform (DWT), for the improvement of the sensitivity of scalar indicators such as Kurtosis (Kurt) and Crest Factor (CF) within the application of condition monitoring by vibration analysis on electric machines. The measurements were possible thanks to the piezoelectric sensors, where the signals were recorded from the machine's critical and judiciously chosen points. The paper demonstrates that when the motor runs under faulty conditions, it is possible to notice the appearance of spallings, which cause the signal to be disturbed and consequently modify the distribution (which is of Gaussian kind in a flawless situation). Nevertheless, those impulse excitations can have a tremendous effect on the values of time-domain indicators. The paper proposes two powerful denoising methods, discussed in-depth the effectiveness of each technique. The conclusion drawn from the analysis shows that the approach improves the sensitivity of selected indicators and therefore increases their reliability for fault presence detection.