Rotating machine vibration signals modelling by a linear and periodically time-varying system
Résumé
Most of rotating machine vibrations contain possibly modulated periodic components, which are interesting to analyse from the point of view of diagnostic. To this end, a method to estimate sine wave components of frequency fc+m*fd (m in Z) in wide-band noise is elaborated in this article. The originality of this processing comes from the use of a linear and periodically time-varying system. Its input is a sine wave of frequency fc, and its variation is synchronised on the frequency spacing of the desired spectral components fd. Thanks to its ability to apply frequency shifts, its output contains the desired frequencies, and therefore is able to correcly estimate the searched components. A new adaptive algorithm is developed to ensure the convergence of this system to its optimal value in the mean square sense. Moreover, it makes it able to follow possible amplitude and phase variation of these sine waves. Finally, its good performance are illustrated on synchronous and induction machine vibration signals.