Very often processes associated with rotating machinery entail two general types of random processes: (i) quasi-periodic processes linked directly to the machine rotational frequency (including harmonics and sub-harmonics), and (ii) regular processes such as those associated with random vibration of machinery elements. Variability of the rotational frequency and slow-time variability can both be induced by factors such as time-varying external loading and changing thermodynamics (e.g. the influence of road grade and engine temperature on automobiles using cruise control). In this work we incorporate and extended Kalman filter (EKF) model guided by estimates of the family of AR and MV spectral estimates to arrive at a method of decomposing and tracking these two types of processes, along with related parameters (e.g. frequency jitter bandwidth/intensity and the strength, damping and natural frequency of randomly excited narrowband resonances). This work may be viewed in some ways as an extension of the work presented in [1]. Our findings include (i) for the case of constant nominal speed, the EKF estimates of the time-varying frequency, sine and colored noise processes are quite good; whereas the estimate of the time-varying amplitude is not as good, and (ii) in the case of linearly increasing speed (where the colored noise resonance behavior is only excited in that portion of the observation time where the sine frequency is in close proximity), the estimated root mean-squared errors related to both the frequency and the noise increase linearly with the nominal frequency rate of change.

This content is only available via PDF.
You do not currently have access to this content.