Comparative Study of Recursive Parameter Estimation Algorithms with Application to Active Vibration Isolation

In this paper, adaptive filtering is adopted for active automotive engine vibration isolation where both transient and stationary engine internal excitations as well as structure flexibility are considered. The adaptive filtering problem is formulated using a linear regression model representation. This allows for an application of a general family of state-of-the-art recursive parameter estimation algorithms. The performance of two specific members of this family has been compared. Those are the well-known normalised least mean square (NLMS) algorithm and a recently suggested Kalman filter based algorithm originally proposed as a method to avoid covariance windup, here referred to as Stenlund-Gustafsson (SG). A virtual non-linear 43 degrees of freedom engine and subframe suspension model and measurement based engine excitation are used in evaluation of algorithm performance. With respect to trade-off between convergence and steady-state variance, the difference Riccati equation included in SG implies superior performance of SG compared to NLMS. However, none of the proposed algorithms provide sufficient tracking performance to deal with transient engine excitation corresponding, for instance, to rapid acceleration of the car. In this case, the adaptive filtering strategy is found to be inadequate.