People Education Research Industrial Agenda  
 
Current Research Archive Publications PhD theses Software      

 
next up previous contents


Robust active control of noise and vibrations

Project members: P.R. Fraanje, M. Verhaegen

Our research on robust active control of noise and vibrations focuses on the suppression of broadband stochastic disturbances (see e.g. Figure 5). Until now active noise/vibration control (ANVC) has been successfully applied to cancel harmonic disturbances, e.g. in propeller aircrafts and air conditioning systems. In most of these ANVC systems, the Filtered-X LMS algorithm is applied, because of its simplicity and adaptivity.

Figure 5: Artist impression of a smart panel system, in which disturbing vibrations are suppressed by controlling piezo-electric actuators.
\includegraphics[width=0.8\linewidth]{pics/smart_panel}

However, the control of broadband disturbances, e.g. in jet aircrafts and road-tire/aerodynamic noise in cars, suffers to some important problems, which has hold back from practical application:

  • contrary to the case of harmonic disturbances, in the case of broadband disturbances usually no reference signal which is highly correlated with the disturbance is available;
  • adaptive LMS algorithms, like Filtered-X LMS converges very slowly due to correlation in the broadband regression signal;
  • the system to be controlled is usually an infinite dimensional system, which yields high model orders in its finite order approximation;
  • the system, especially actuators and sensors, often suffer to nonlinear behavior.

To solve these problems, we have proposed two solutions:

  1. an off-line probabilistic control design method, which optimizes the average performance over a stochastic model uncertainty set;
  2. an robust fast adaptive control design method, which is basically a robust preconditioned version of the Filtered-X LMS algorithm.
Our research to improve the performance of the first method is on control-relevant (closed-loop) identification of the model and iterative identification and controller design. In cooperation with the University of California, Los Angels, we investigate how to improve the second method, such that performance bounds can be guaranteed.

For large scale applications, e.g. in smart materials, with many sensors and actuators, decentralized and distributed control methods needs to be developed to reduce the complexity of the real-time controller. Therefore a second main research goal is to design and practically validate decentralized ANVC algorithms for broadband disturbance suppression.

This project is done in cooperation with TNO-TPD.


next up previous contents
Next: On weight adjustment in H Up: Controller design Previous: Robust model predictive control

Back to top

Last modified: 24 March 2005, 10:16 UTC
Search   Site map