Signal pre-processing using cepstral editing for vibration-based bearing fault detection

A key step towards proper detection and diagnosis of bearing faults through vibration analysis is the separation
of bearing fault signal components from other masking deterministic components. Bearing fault signals
can generally be classified as (quasi-)cyclostationary which makes the separation possible from deterministic
content, e.g. from gears or shafts. The preprocessing techniques investigated in this paper are the automated
cepstral editing method (ACEP) and cepstrum pre-whitening (CPW). In the past these techniques already
showed great potential in removing discrete signal components and they often outperform alternative techniques.
The cepstral editing procedure utilizes the fact that deterministic components appear in the cepstrum
as narrow peaks and removal of these cepstral peaks reduces the amplitude of the corresponding discrete
components. A comparison of the findings for both cepstrum methods is discussed in this paper for virtual
signals.