Abstract
It is rather costly to simulate granular phenomena by using Discrete Element Method with non-spherical particles. In this article, non-spherical particles were modeled by collectives of spherical particles rigidly connected. We proposean efficient neighbor-particle searching method and an algorithm to improve the computational performance for GPU computing.We demonstrate the difference of the foot stamp in the case of 405,000 tetrahedron-shaped particles from simply spherical particles. The edge of footprint with tetrahedron-shaped particles is clearer due to the enhanced shear frictions of interlocked particles. Confirming the efficacy of GPU acceleration, we perform a large-scale agitation simulation using 300,000 cubic objects on a NVIDIA Tesla K20X and the simulation for 800,000 time steps has completed within 14 hours.
