Finding a frequency signature for a cyclostationary signal with applications to wheel bearing diagnostics

In recent years, an increasing number of researches in signal processing was dedicated to frequency identification and analysis of cyclostationarity. The survey by Gardner et al. (2006) [8] have quoted over 1500 different papers recently published that are dedicated to cyclostationarity. An important application of cyclostationary signals is the analysis of mechanical signals generated by a vibrating mechanism. In this area of research the paper by Antoni (2009) [2] shows the importance of cyclostationary models to perform basic operations on signals in the time and frequency domain.The result in this paper presented a new perspective on cyclostationary signal analysis and on frequency identification for such signals. One of the fundamental problems in diagnosis of rotating mechanism is in identification of significant modulating frequencies that contribute to the cyclostationary nature of the signals. So far, the statistical methods for frequency identification in cyclostationary signals were based either on the assumption of gaussianity of the signal and/or on the assumption of some linear structure of the signal. The recent research by Lenart et al. (2008) [13] has shown that there are modern tools available for analyzing cyclostationary signals and they are based on the idea of resampling of observed signals. The aim of this paper is to show applicability of a resampling technique called subsampling in frequency identification for cyclostationary signals. The theoretical results are accompanied with applications to frequency analysis of cyclostationary signal generated by a wheel bearings, one without any damage and the another two with two different types of faults. The result showed that the normal operating conditions and abnormal operating conditions for the bearings can be identified via resampling-based frequency analysis and subsequent frequency identification based on a statistical test.

► The frequency signature for the cyclostationary signals is studied.
► The subsampling is introduced to establish significance tests for frequencies.
► Statistical significance is proven for differences in frequency signatures.
► The results allow identification of faults in wheel bearings.
► The results hold also for non-Gaussian signals.