VIBRATION SIGNAL ANALYSIS AND FEATURE EXTRACTION BASED ON REASSIGNED WAVELET SCALOGRAM

The wavelet scalogram has been widely used for vibration signal analysis, but it has low frequency concentration at small scales and low time concentration at large scales owing to the limitation of Heisenberg–Gabor inequality. In addition, misleading interference terms would appear in the scalogram at some conditions. All of these would influence scalogram analysis seriously. In this paper, a reassignment method is used, which can improve the concentration of the scalogram and weaken the interference terms to a certain extent; thus, the readability of the scalogram can be improved. The experimental data for three kinds of faults in rotating machinery: rotor-to-stator rub, oil whirl, coupling misalignment and the actual measurement data from a Pumped Storage Power Generator Unit are analyzed by the scalogram and the reassigned scalogram, and the results by FFT analysis are also given. A comparison among three kinds of analysis results is carried out, and the comparison results indicate that the scalogram can not only reveal the time–frequency features of vibration signal but also highlight the components of low energy. Therefore, it is suitable to extract features of faults at an early stage.