Parallelization of DEM simulation on distributed-memory computer via three-dimensional slice grid method

Abstract

overlap, developed Abstract A fluidized bed can efficiently filter dust particles, but its performance depends significantly on the fluidization state. To further develop the fluidized-bed filtration method, it is important to understand the filtration mechanisms in detail. Numerical simulation via the discrete element method is useful for solving these problems because the motion of each bed and dust particle is demonstrated. This system has large number of particles, and bias of the particle distribution is generated owing to the fluidization and supply of dust particles. Parallel computing on a distributed-memory computer is necessary to simulate many particles. Additionally, dynamic load balancing is a key technique for solving these problems. In this study, we developed a simple implementation of three-dimensional slice grid method and periodically used this method to balance the workload while keeping contact information such as the pair of colliding particles and its overlap. The computational efficiency of our method was assessed through an ideal problem involving a packed particle system and dust filtration in a fluidized bed. The changes in the particle number and particle distribution were examined. In the packed particle system, linear speed-up was obtained at particle number of 100 million and a message passing interface-process number of 1024. Moreover, the effectiveness of the dynamic domain decomposition method was confirmed by solving through the dust filtration problem.