ISBN 9783843903554

978-3-8439-0355-4, Reihe Informatik

Simon Pabst
Advances in Physically Based Deformable Object Simulation

120 Seiten, Dissertation Eberhard-Karls-Universität Tübingen (2011), Hardcover, B5

Zusammenfassung / Abstract

Physically based simulation of thin deformable objects is a well-established, fascinating field of research in Computer Graphics. It is influenced by many different disciplines, ranging from physics, computational mathematics and computer science to material sciences and is a rewarding topic for study.

In this thesis, we present a number of new techniques that improve the state of the art in the field. Chapter 2 describes a new nonlinear and efficient bending model for cloth simulations that can make use of measured material data and is also able to include the influence of seams and interlinings on fabric drape. Simulations using the new approach are compared to real-world counterparts for quality comparison.

Physically based simulation is computationally expensive. Real-time performance is difficult to achieve due to high resolution geometry and mathematically involved computations, and keeping the highly deformable geometry intersection-free further aggravates the problem. In Chapter 3 we present techniques for parallel simulation on modern off-the-shelf multi-processor systems.

Building on the insights gained in Chapter 3, we propose a second approach to parallel collision detection. Chapter 4 shows how to exploit modern graphics processors (GPUs) in addition to multiple CPUs.

In Chapter 5 we present a generalized dry Coulomb friction model in the anisotropic domain. The method is applicable to both elastic surfaces, like cloth and shells, and volumetric solids. It can realistically model anisotropic, heterogeneous, and asymmetric friction. Only negligibly higher computational costs compared to isotropic approaches are incurred and the additional effort is generously rewarded by a variety of interesting new effects.

Finally, in Chapter 6 we propose a technique that simulates wet textiles using translational diffusion of liquids inside porous textiles. We combine this approach with our cloth simulation and additionally couple it to a particle-based fluid simulation. The fluid can interact with the textile and also wet it, thereby changing its material properties.