Faulty Bearing Signal Analysis With Empirical Morphological Undecimated Wavelet

Morphological undecimated wavelet (MUDW) is a powerful method, which can capture features accurately through removing noise by signal decomposition. It has therefore been widely used in the fault diagnosis of rotating machine. However, the setting of some key parameters (including the selection of morphological operator (MO), the length of structure elements (SEs), and the number of decomposition level) still depends heavily on prior experience. Not only is the computational efficiency seriously affected by the time-consuming trial-and-error process, but its filtering effect is also vulnerable to the interference of external factors like random impulses. In view of that, an improved method, called empirical MUDW (EMUW), is proposed in this article. Waveform trend (WT) is utilized first to make up for the deficiency of the MOs, which eliminates the interferences brought in by random impulses. Based on the similarity evaluation between the WT signals of adjacent level, the number of decomposition level of EMUW can therefore be determined. Finally, the signal can be reconstructed with the subsignals obtained by decomposition. Compared with traditional MUDW, EMUW shortens the calculation time to less than 2 s.