A comparison of cepstral editing methods as signal pre-processing techniques for vibration-based bearing fault detection. (July 2017)

• Record Type: Journal Article
• Authors: Peeters, Cédric; Guillaume, Patrick; Helsen, Jan
• Is Part Of: Mechanical systems and signal processing. Volume 91(2017)
• Page Start: 354
• Page End: 381
• Publication Date: 2017-07
• Subjects: Cepstrum -- Editing -- Whitening -- Bearing fault -- Deterministic/random separation; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2016.12.036
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Discrete/random separation is an important step in vibration signal analysis. Cepstrum editing is a commonly used tool for this separation. Two cepstrum editing approaches have been suggested in the past. Comparison of the approaches shows that both have pros and cons. Abstract: The detection and diagnosis of incipient rolling element bearing faults is not an undemanding task and signal analysis of vibration measurements therefore often incorporates the use of various complex processing techniques. One of the key steps in the processing procedure is the proper separation of the bearing signal from other influencing sources like shafts or gears. The latter sources produce deterministic signal components showing up as discrete frequencies in the amplitude spectrum, while bearing signals are assumed to be (quasi-) cyclostationary resulting in a smearing of the bearing frequencies in the spectrum. This property gave rise to the idea of using the cepstrum for the purpose of separating the deterministic signal content from the second-order cyclostationary bearing signal. The cepstrum essentially groups the deterministic multi-harmonic signal content in a cepstral peak at the corresponding quefrency, making it more suitable for easy removal of the discrete frequency peaks. Even though initially there was a tendency to only remove or 'lifter' the selected cepstral peaks, nowadays the full real cepstrum is set to zero instead of only certain quefrency bands. This technique, called cepstrum pre-whitening, is easy to implement, can be performed quickly without the need for additional input parameters or fine-tuning and would be well-suited for practical applications. However, these advantages do come at the cost of some control and insight over the editing procedure of the signal. In order to assess the performance of this cepstrum pre-whitening technique, it is compared to an automated cepstrum editing procedure . It automatically selects certain peaks in the real cepstrum and only sets them to zero instead of the full real cepstrum. Both methods perform quite well in separating deterministic signal content from more random content, but there are some differences to observe when using them for diagnosis purposes. An analysis of the methods is made for both simulated and experimental signals.