Periodic autoregressive time-frequency analysis for monitoring of rotating machinery with variable period

Abstract

The development of a periodic autoregressive spectral estimator which is insensitive to variations in period is considered. The extended Kalman filter is investigated for estimation of the period. This estimate is then incorporated into a rotational framework to produce a time-varying spectral estimate for a nominally periodic system. It is expected that the method will be useful for monitoring the intracycle behaviour of rotating machinery such as diesel engines, compressors, gears and bearings. Use of the short time Fourier transform for variable speed envelope analysis is also shown to be a necessary tool for detection of faults in low, variable speed rotating machinery.