Don't ignore nonstationarity: use it to advantage

Most signal processing techniques were originally designed to work in the stationary context. However, the signals encountered in many fields are clearly non-stationary either by construction like man-made communication signals, or due to the time-varying geometry of the transmission path as in mechanics. The property of non-stationarity seems at first glance inconvenient since it limits the effectiveness of the classical techniques, unless it is exploited to design more advanced techniques. This paper shows how to make use of the non-stationary but cyclic (cyclostationary) nature of the signals issuing from rotating machinery. It provides the pre-treatments to be performed in presence of speed fluctuation in order to resample the vibration signals with respect to the shaft angle instead of time and thus accentuate the property of cyclostationarity. It then illustrates through examples how to exploit the extra information provided by cyclostationarity to effectively perform signal separation and diagnostics of complex industrial systems.