Comparison between two very efficient signal processing approaches for vibration based condition monitoring of rolling element bearings
Keywords: Vibration analysis, bearing diagnosis, Cyclostationarity, Minimum entropy decomposition, Spectral kurtosis
AbstractThis paper compares two of the most efficient signal processing approaches used nowadays for vibration-based condition monitoring of rolling element
bearings. The first is based on pre-processing the vibration signal through the minimum entropy deconvolution method (MED) followed by the spectral
kurtosis (SK), before analysing the spectrum of the signal envelope. The MED aims at maximizing the signal impulsivity by deconvolving the system
transfer function through an optimization approach that maximizes the kurtosis of the output. Then, the spectral kurtosis (SK) is applied to conceive the
optimal filter to be applied before computing the envelope spectrum. The second approach is based on a cyclostationary modelling of the bearing signal. It
applies the spectral coherence to the signal with a special attention on setting the estimation parameters. The spectral coherence is a bi-variable map of the
cyclic frequency alpha, and the spectral frequency, f. The former variable describes the cyclic content of the modulations, while the former describes the
properties of the carrier. The improved envelope spectrum is then computed by simply projecting its squared-magnitude with respect to the f-variable.
These methods are evaluated according to their potentiality to detect the fault in its earliest stage. The comparison is be made on real bearing vibration
signals in run-to-failure tests.