Abstract
Engineers are increasingly relying on simulation to augment and, in some cases, replace large amounts of experimental work. However, current simulation capabilities are sometimes inadequate to capture phenomena of interest. In tracked vehicle analysis, for example, the interaction of the track with granular terrain has been difficult to characterize through simulation. This paper addresses these limitations by developing a simulation framework that is capable of simulating up to one million bodies interacting through contact and friction by drawing on the parallel computing power of graphics processing units (GPUs). With this simulation capability, it is possible to simulate tracked vehicles operating on terrain which is modeled as a collection of discrete granular particles rather than relying on empirical formulas. This simulation capability is achieved through the implementation of (1) the solution of the collision detection problem in parallel on the GPU, (2) the solution of the multibody dynamics problem in parallel on the GPU, and (3) a discrete granular terrain model which captures terrain granularity and profile. The collision detection problem is solved through the use of a spatial subdivision method which can quickly find collisions between a collection of spheres. The multibody dynamics problem is formulated as a differential variational inequality (DVI) problem. The solution of the DVI leads to a cone complementarity problem solved in parallel by an iterative method. These simulation tools are combined to simulate a tracked vehicle operating on granular terrain. The methods proposed have been implemented as a set of parallel algorithms leveraging NVIDIA's Compute Unified Device Architecture (CUDA) library support for multi-core stream computing. This, combined with Microsoft's HPC Server 2008, allows the proposed solution to run on a wide variety of GeForce and TESLA NVIDIA graphics cards for high performance computing.
