Guidelines for DEM

DEM gives high resolution, but at a high cost in computational power. Guidelines for DEM mostly consist of finding ways to reduce the computational cost.

Use the following guidelines:

• Maximize the number of processors. However, do not exceed about 1000 particles per processor. If you continue to add processors, speed goes down as processors spend more time communicating with each other. (So, for three million particles, use about three thousand processors, not more.)
• In composite particles, minimize the number of spheres in a particle.
• Maximize particle size or use the Coarse Grain Model to maximize parcel size.

  One DEM particle can represent several objective particles. Using larger particles reduces particle count and increases the DEM time-step, which reduces both the time that is needed and the number of processors needed. Consider the example of a simulation with particles in a rotating drum. Enlarging the particles from 2 mm to 3 mm produced the same results in half the time, using 5 processors instead of 18.

  Using the Coarse Grain model reduces time that is spent on contact physics while maintaining full accuracy for particle-fluid interactions. Coarse Grain is best used for fluidized beds applications, where accuracy of particle-fluid interactions is more important than accuracy of particle-particle interactions.

  Note	The change in size in the example simulation also results in slightly different bulk properties (such as angle of repose) that depend on contact physics. You can often compensate for similar changes in bulk properties by slightly adjusting the parameters of the DEM Contact Model, such as friction.

  Note	To maintain the accuracy of the DEM model, scaled-up particles or coarse-grain parcels must still be smaller than the smallest significant geometrical feature of the surroundings. In the example simulation, the parcels had to be smaller than the smallest important length scale in the geometry.

• Minimize Young's modulus. This practice increases the DEM time-step, resulting in faster computation. For applications with low-confinement shear flow regimes, you can often reduce Young’s modulus by several orders of magnitude and still get accurate results. For example, in one simulation, going from 1000 MPa to 100 MPa reduced time to one third, using 5 processors instead of 18.
• Make the mesh as coarse as practical.
• When using polyhedral particles, minimize the number of faces as much as possible.
• Reduce the modeled physical time as much as practical:
  • When possible, optimize injector settings to introduce the maximum number of particles in the minimum physical time.
  • Sometimes, you can reduce time by increasing physical speeds of moving geometrical boundaries, such as using higher rpm on a rotating boundary.
  • If you only need to establish the direction of a trend, you can use a short physical time.
• Use double precision in simulations that integrate quantities over a large number of time-steps or that deal in a large number of small energy exchanges. For example:
  • Simulations where particles recirculate or settle within the simulation domain
  • Predicted particle temperature increases in each sub-step that are negligible compared to the ambient temperature
  • Chemical reactions
  • Mass transfers between particles
• To speed memory access, use the DEM solver property Reordering Frequency. On reordering, Simcenter STAR-CCM+ resets the sequence in which DEM particles are accessed in memory. For quasi-stationary cases, it is enough to reorder once, after the particles are injected and settle. For rapidly moving particles, reorder at least once every 20 iterations. The benefit of reordering decreases with repetition.
• To reduce time that is spent running contact detection, use the DEM solver property Skin. The Skin property defines a volume around each DEM particle, a “skin.” The solver runs contact detection only when the particle moves through a distance equal to the skin thickness. As a rule of thumb, set skin thickness to $10v_{p}t$ where $v_p$ is a characteristic particle velocity and $t$ is length of DEM time-step.