Implementation of multi-level coarse-graining method on GPU using MUSEN

GPU-based discrete element method (DEM) simulations have recently gained popularity due to their ability to reduce the computational costs associated with modelling large-scale granular systems. To further enhance the efficiency of these simulations, we implement the multi-level coarse-graining (MCG) method on GPU using the open-source GPU-accelerated DEM software MUSEN. The MCG method significantly improves performance by reducing the number of simulated particles while maintaining accuracy. The key simulation parameters such as the Verlet distance and the interval of performing refinement or coarsening of particles are optimized to maximize computational efficiency. The MCG-GPU method is validated using a hopper discharge system, achieving a speedup of about 3.4× compared to the MCG-CPU simulation. The robustness of this method is further demonstrated through its application to two industrial systems: a tablet-press feeder and a twin-screw feeder, where speedups of approximately 5.5× and 8×, respectively, are achieved relative to MCG-CPU simulations. These results affirm that the MCG-GPU method is a powerful tool for efficiently conducting large-scale particle simulations with complex geometries.

Subjects

DEM

GPU

MCG

MUSEN

Tablet-press feeder

Twin-screw feeder

DDC Class

540: Chemistry

Options

Implementation of multi-level coarse-graining method on GPU using MUSEN