My name is Toby Heyn, and I am a PhD candidate in Mechanical Engineering. I have been working with Professor Dan Negrut at the University of Wisconsin Madison since May 2006. I received my Bachelor's Degree in Mechanical Engineering and Applied Mathematics in May 2008. I received my Master's Degree in Mechanical Engineering in December 2009. The title of my Master's Thesis was "Simulation of Tracked Vehicles on Granular Terrain Leveraging GPU Computing". I am currently pursuing a PhD in Mechanical Engineering.
My current work is related to the modeling, simulation, and visualization of many-body dynamics problems with friction and contact.A first objective of this project is to leverage parallel computers to investigate the dynamics of very large systems containing granular material such as sand. Current parallel algorithms allow the simulation of systems with over one million interacting particles. However, improved methods are required to investigate larger systems and perform accurate simulations of a Mars Rover operating on small-particle granular terrain, for example.
Firstly, this work focuses on improving the numerical methods used to solve the dynamics problem, formulated as a Cone Complementarity Problem, on the GPU. Specifically, Krylov methods will be leveraged to solve the associated optimization problem at each time-step. Additionally, parallel preconditioning is being pursued to further improve solution speed and accuracy. Secondly, this work investigates the use of heterogeneous distributed computing to analyze large systems which would not fit on a single piece of computational hardware. The Discrete Element Method will be implemented to solve large problem through domain discretization and heterogeneous computing on a cluster computer.
The image above shows a snapshot from a simulation of a Mars Rover operating on Granular Terrain. The terrain is composed of 50,000 particles. The picture on the left shows the domain-decomposition concept, were particles are colored based on which sub-domain they are contained in. The image on the right shows the system at the same instant, rendered uniformly. The simulation used one CPU core for each of the nine sub-domains, and too 75 CPU hours to finish.
2046 Mechanical Engineering Building
1513 University Avenue
Madison, WI 53706
Email: heyn (at) wisc (dot) edu