A new approach to tune the vold-kalman estimator for order tracking

Abstract

In the purpose to diagnose rotating machines using vibration signal, engineers use order tracking method to process non-stationary signals. We deal here with order tracking when the vibration signal is represented in a state space model. Such a methodology leads to the Kalman estimator that requires knowledge about the noise statistics affecting the state and the measurement equation. These noise statistics are usually unknown and need to be estimated from operating data for the use of the Kalman estimation algorithm. Several methods to tune these parameters have been developed for time-invariant model. In this paper, we introduce a technique to estimate the noise covariances for a linear time-variant system using the innovation process. The efficiency of this new approach is evaluated using a synthetic non-stationary vibration signal. The advantage of this approach is that it converges quickly and provides a small estimation error compared to those used for the linear time-invariant model.