Applying High-Performance Computing to Discrete Computational Granular Dynamics

Abstract

This thesis investigates computational methods applied to the two leading discrete methods for simulating dry granular materials. For rigid contacts, the complementarity-based discrete element method is explored to assess its current limitations and its affinity for scaling. Topics explored are a formulation for reducing the solution set to a single solution and applicability to granular material simulation. For compliant contacts and high scalability, the penalty-based discrete element method is used. The model itself is largely taken from literature with the addition and study of a novel uniform spatial grid. This work introduces a parallel computing framework implemented to run on graphics processing units. This hardware choice provides scalability, which makes the problem of simulating discrete granular dynamics accessible for little monetary overhead. This C++ with CUDA software framework is freely available for use and modification as part of the ProjectChrono open-source physics engine on GitHub.