Online plant modelling methods for active control of sound and vibration

  • Laugesen, Søren (Project Manager)

    Project Details

    Description

    Active control systems that use the popular filtered-x adaptive algorithm require so-called plant models to describe the relation between the secondary source outputs and the error sensor inputs in order to ensure convergence. For most laboratory experiments reported in the literature the plant models have been identified a priori and have remained fixed during the control experiment itself. This approach is viable as long as the physics of the plant does not change drastically during the experiments. However, for long-term practical applications an online plant identification scheme may be required, which continuously adapts the plant model to the possibly changing environment.
    Especially two of the methods suggested in the literature seem appropriate for this purpose. One method uses a low level probe noise and a standard LMS adaptive filter to model the plant, whereas the other method models both the primary and secondary plant responses using a projection algorithm. The objective of this work has been to investigate these two approaches of online plant modelling in multichannel active control systems, and to compare their performance. This has been done by means of computer simulations, which readily allow for reference results to be established.
    For a broadband random disturbance the first approach was found to give results that were very close to the theoretically optimal ones, whereas the results obtained for sinusoidal disturbances were less impressive. In the latter case the choice of plant model filter length and spectral shape of the probe signals turned out to be critical. The other approach, which envolves 'complete' system identification, gave seemingly convincing results, as convergence was always obtained. However, in some cases the steady state performance could be quite far from optimal, because the secondary plant models provided by the estimation algorithm were erroneous.
    StatusFinished
    Effective start/end date01/01/199631/07/1996