Parallelization of discrete element simulation

Abstract

The Discrete Element Method (DEM) is used for modeling discontinuous or non‐homogeneous materials and has been applied for various engineering applications. However, it often requires the use of a very large number of elements and very small time steps in order to ensure numerical stability. This in turn leads to considerable computational time requirements for each simulation of an actual engineering problem. To resolve this issue, this paper provides a systematic and general procedure for parallelizing a discrete element simulation system. Three strategies (the zone of interest, zoning, and interaction‐based geometric partitioning strategies) are proposed for hastening the simulation, at different simulation stages. An in‐house discrete element simulation framework, called VEDO (VErsatile Discrete Objects), is extended for parallel computation. Software utilizing this framework, named KNIGHT&IRIS 2.0, was developed to verify the feasibility of these strategies. Three numerical examples were tested using both shared memory and distributed memory platforms. These examples demonstrate that the strategies proposed in this study are effective and useful for researchers interested in developing a parallel discrete element simulation system.