A Variational Bayesian Approach for Order Tracking in Rotating Machines

Order tracking is important in rotating machine signal processing. It consists in tracking the complex envelope of phase varying sinusoids associated with distinct rotating mechanical elements. The resulting magnitudes and phases are usually used for vibration level control and system resonance identification. A common approach to tackle this issue is the Vold-Kalman filter. However, the performance of this filter is highly conditioned by the choice of some tuning parameters that control its selectivity. In this paper, a new order tracking method is proposed to jointly estimate the targeted envelopes and the optimal values of these parameters in an automatic manner. To do so, the order tracking problem is formulated within the Bayesian framework. In details, prior distributions are affected to the target variables (including the tuning parameters and the noise variance); posterior distributions are then derived and estimated using a bariational Bayesian approximation algorithm. Results on simulated and real helicopter vibration data demonstrate the effectiveness of the proposed approach.