Author:

Keywords:

Model order reduction, Vibroacoustics, Model-based sensing, Virtual sensing, Inverse identification

Abstract:

Sound and vibrations are all around us and constantly impact our everyday lives, both consciously and unconsciously. The way something feels and sounds contributes enormously to its perceived quality. For this reason the design of the acoustic and vibrational characteristics of products is becoming ever more important. The physical laws that govern sound and vibrations have already been known for a long time, but even for moderately complex shapes and structures the resulting equations are much too difficult to solve analytically. The advent of digital computing has allowed scientists, engineers and designers to use numerical methods to calculate an approximate solution to these equations. This has opened up a world of opportunities. For example, such numerical modelling techniques can be used to evaluate the sound and vibrations of virtual product prototypes, which significantly speeds up the design process. Unfortunately, the numerical models that arise in the context of sound and vibrations typically require a tremendous amount of computational effort to solve. Even on modern computing systems the resulting calculation times practically restrict the use of these numerical models to harmonic analysis at relatively low frequencies. The work in this dissertation investigates how model order reduction techniques can be used to drastically reduce the required computational effort, and explores how the resulting efficient yet accurate models can be exploited in the context of model-based sensing. The first contribution of the research presented in this dissertation is the conception of a flexible, accurate and efficient strategy for the numerical modelling of transient vibroacoustics. The computational efficiency of traditional finite element models is greatly improved by the use of model order reduction techniques, while making almost no concessions to the accuracy. A stability preserving model reduction framework is developed to make it possible for transient simulations to be carried out with the resulting reduced-order models. Next an error-controlled algorithm that automates the reduction procedure is designed, which makes it possible for non-experts to exploit the power of model order reduction in vibroacoustic applications. The combined use of model reduction methods and filtering techniques even enables the parallel computation of time domain simulations. The availability of such efficient yet accurate numerical models can be capitalized on in other application areas than simulation and virtual prototyping. The second research contribution is the development of model-based sensing techniques that fuse these numerical models with experimental measurements in order to get the best of both worlds. The use of such model-based sensing techniques makes it possible to generate virtual measurements of quantities that are not measured directly, such as the sound at inaccessible locations and even intrinsic system properties.