Time-varying metrics of cyclostationarity for bearing diagnostic. (April 2021)

• Record Type: Journal Article
• Authors: Pancaldi, Fabrizio; Rubini, Riccardo; Cocconcelli, Marco
• Is Part Of: Mechanical systems and signal processing. Volume 151(2021)
• Publication Date: 2021-04
• Subjects: Ball bearings -- Diagnostic -- Cyclostationary signal -- Time-varying variance -- Time-varying kurtosis -- Time-varying Kolmogorov-Smirnov test; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2020.107329
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: This paper presents and compare novel approaches to the diagnosis of ball bearings These are based on the statistical definition of cyclostationarity. The proposed algorithms are time-varying variance, time-varying kurtosis and time-varying Kolmogorov-Smirnov test Numerical results prove that the proposed algorithms can appreciably outperform the conventional approaches based on autocorrelation, Spectral Kurtosis and demodulation. Abstract: Ball bearings represent the most adopted solution to support rotating elements. Separated by the cage, the rolling elements are induced by the kinematics of the system to roll and accidentally slip on the rings. In working conditions the continuous contact of the elements leads to a wearing of the bearing surfaces. As a consequence, the early detection of faults represents an issue for modern diagnostic systems. The mathematical model of faulted rolling bearings has been extensively investigated in the last decades and it is widely accepted that a faulted bearing is subject to an unwanted slippery leading to a cyclostationary vibration signal. This paper presents a novel approach to the diagnosis of rolling bearings based on the statistical definition of cyclostationarity. In particular, various metrics have been devised to track the "cyclostationary signature" of the vibration signal and the performance of the proposed algorithms has been assessed through both experimental measurements and synthetic data. Numerical results have shown that the new approach to fault detection is comparable to conventional techniques based on spectral kurtosis, demodulation and spectral correlation, and it can outperform them in some cases; furthermore the simplicity of the proposed algorithms leads to an intrinsic robustness against the mechanical noise typical of practical scenarios.