This article covers the considerations to be made when looking to speed up a DEM simulation.
There are several different tactics to use in attempting to speed up a DEM calculation:
Use the latest version of Simcenter STAR-CCM+; there is a continuous effort to improve the turnaround time for DEM simulations with each version. See the Spotlight On... Discrete Element Method (DEM) article for the latest version information.
The number of processors should be limited by the number derived from dividing the (target) number of particles by 1500. With the number of particles per processor less than 1500, the benefits of parallel runs are reduced.
For parallel runs, enable Load Balancing feature of the Lagrangian Multiphase solver. For parallel runs using multiple regions, always set the ‘Partitioning method’ of ‘Partitioning’ solver to ‘Per-Continuum’.
The DEM particle time step depends on the particle density, the particle Young’s modulus and the particle size. Decreasing Young’s modulus or increasing particle size will increase the particle time step and reduce the simulation time. This does alter the problem physics but will help speed things up. Young modulus should not be set to value lower than 100 MPa (Oh no! My DEM particles are going out of the geometry!), for small particles, changing Young’s modulus will not have a significant effect in all likelihood.
Activate the Lagrangian and DEM solver verbosities, and look for how many substeps are taken by the DEM/L2P solver. If this number is large and/or is maxing out (default max is 20000 substeps) then stop the run and look at the parcel DEM time step. The code time step divided by the parcel DEM time step is the substep number. If large, lowering the code time step can enable the DEM solver to run faster.