Model Reduction of Muscle-Driven Tissue Models

Abstract

Biomechanical simulations are a necessary tool for a proper understanding of biomechanics and hence are subject to intense research. One field that relies on this research is articulatory speech synthesis as it attempts to simulate the physics of the speech production process. Out of the many aspects involved, muscle driven tissue is one of the most important as it is required to simulate the deformable structures of the vocal tract. Modelling of muscle driven tissue requires continuum models of high complexity for the purpose of accuracy. On the other hand, time-efficient models are desirable in order to provide fast simulations which enable the user to test input parameters interactively. These requirements impose limitations on each other as the time-efficiency of a model is reduced with increasing complexity, hence techniques that can bridge the gap between these requirements are needed.

This thesis attempts to bridge this gap through two major contributions. Model reduction techniques, that up until now have only been applied to inactive materials, have been implemented and tested for muscle driven tissue models. The implementation has been made in a general way to ensure that it can be used for biomechanical simulations in other fields than articulatory speech synthesis. In addition, the implementation has been made such that it can handle more advanced simulations than those investigated in this thesis. The simulations show acceptable but not ideal accuracy in both dynamic simulations and in measurements of equilibrium configurations. In addition, the reduced simulations using hyperreduction show good speedup for the more complex models investigated.