Small CouplingInterval blows up CFDEM simulation

Submitted by AndresMM on Wed, 11/08/2017 - 08:55

Hi everyone,

I am simulating the pressure drop caused by a compressed bed of packed particles. On the DEM-side I mechanically compressed particles with a moving mesh. I then used this state to create the restart data where I changed the STL for a wall. On the CFDEM-side I apply a velocity of 0.001 in the inlet. I am using cfdemSolverPiso.

I have noticed a couple things:
- The simulations blows up if the mesh is too small. This is to be expected as my particle packing is quite dense. However what is surprising is that if I set the coupling interval to a higher value (1000) I can get away much much finer meshes. If I try the same mesh with an interval of 1, 10 or 100, the simulation will blow up (I maintain Timestep_CFD = Timestep_DEM * CouplingInterval in all cases).
- Even if I reduce all timesteps one magnitude I still observe this behaviour, i.e. Timestep_DEM = 2e-6 down to 2e-7.
- I have also read some topics in the forum where a similar behavior is encountered, i.e. a smaller couplingInterval leads to unstable or failed simulations, for example:
https://www.cfdem.com/forums/erratic-and-unexpected-particle-velocities
https://www.cfdem.com/forums/optimal-coupling-interval-time

So my doubts are:
- Are there guidelines when setting the couplingInterval?
- What can be causing the behaviour of being able to run simulations with big CouplingIntervals and not small ones when it should be the exact opposite? Is it just "luck" that it happens to work with large couplingIntervals and the simulation itself is flawed?

Regards,
Andrés

Forums:

AndresMM | Wed, 11/08/2017 - 10:25

I decided to keep reducing the DEM_Timestep and CouplingInterval. At DEM_DT = 2e-8 and CouplingInterval =10, the simulation runs smoothly (albeit forever) even with the finer meshes.

I guess the problem was simply too large timesteps!

medvedeg | Wed, 11/08/2017 - 10:32

Hallo Andres,

the main assumption of unresolved CFD-DEM is that CFD cell size is bigger than particles size. Otherwise, stability of the solver is not guaranteed. I would suggest to check the cell size in order to be sure that it is at least twice bigger than particle diameter.

Alexander Podlozhnyuk

AndresMM | Wed, 11/08/2017 - 10:42

Hi Alexander,

Thanks for your answer. The mesh fulfills this condition. It was the timestep that was just too large.

Regarding the mesh/particle relation, is there a rule of thumb regarding how small can the void-fraction in a mesh cell be before it affects the CFD-solver? I currently have a gaussian distribution of closely packed particles, so our void fraction currently reaches XX.XX.

I would like to keep refining the mesh, is there a guideline-value at which the voidfraction is just too small (too many particles occupying a cell)?

Regards,
Andrés

alice's picture

alice | Mon, 11/13/2017 - 14:20

Hi Andrés,
currently we implmeneted the equations in a way that 0.1 is the limit (can be changed in the couplingProperties file). Generally I cannot recommnd to run simulations with 0.1 everywhere since this is somwhow numerically still possible but might not make much sense in a physical way (the equations are simply not designed for such cases). I think ~0.2 should be fine. Please also not that if particles cover entire cells this would lead to a "cutoff" of the voidfraction (i.e. particle volume of up to 10% of the cell volume's worth) is just neglected in the calculations. As always such number are just rules of thumb and depend on the very cases (also depending on the mesh one case might in some regions permit a non-idal setup while another case would react to it very sensitively).
Cheers,
Alice