Problem when running CFDEM3.5.0

Submitted by NTT1508 on Tue, 01/17/2017 - 07:44

Hi all,

I found a strange behaviour in my simulation run on CFDEM 3.5 (LIGGGHTS 3.5 and OpenFoam 3.0.1). I generate particles having the diameter varying from 0.1 mm to 10 mm. I apply a flow through this medium and observe the behaviour of particles. It is weird that the simulation cannot go for long and stop because it warns that " time step too large or skin too small - particles may relatively travel a distance of 0.000223 per time-step, but skin is 0.000100". OK it sounds like a simple problem, I reduced time step to quite small number (i.e., dt=1e-6 (s) in CFD and 1e-8 in DEM). However the problem is still existing. Then I try to reduce the input velocity in CFD to a very small value, but the problem cannot be solved. I check the drag force and found it is increasing dramatically over time.

However I bring the same simulation and run on CFDEM 2.9 (LIGGGHTS 3.2.0 and OpenFoam 2.3.0), it works fine with reasonable results. I noted something has changed in the code of CFDEM3.5 compared to CFDEM2.9; for example dragforce using DiFelice model. TurbulenceProperties is also modified. I have not had enough time to check what else has changed and how much they can effect on result.

Any ideas for this problem ? Thanks you,

Regards,

Nathan,

j-kerbl's picture

j-kerbl | Tue, 01/17/2017 - 18:08

Hi Nathan,

this is a LIGGGHTS specific warning. This has to do with the neighbour lists, which can be a bit confusing, but I'll try to stick to the point here. The skin size defines the size of a virtual sphere, in which the particles are checked for overlap (-> DEM force calculation) and if a particle moves further, it could collide with a particle that hasn't been checked before.
The reason is your high size difference in radii.
Please have a look at
http://www.cfdem.com/media/DEM/docu/neigh_modify.html
and
http://www.cfdem.com/media/DEM/docu/neighbor.html

I'd suggest to use something like

neighbor    1e-3 bin
neigh_modify    delay 0
neigh_modify    binsize 1.5e-3

Usually a small parameter study with 10 timesteps in LIGGGHTS can increase simulation performance.

Cheers,
Josef

NTT1508 | Wed, 01/18/2017 - 09:10

HI Josef,

Many thanks for your quick response. Yes I can see the wide variation in particle radii can cause the problem, that why I have reduced the bin size and time step many times. What I am confusing here is why the same case with exactly similar parameters (the same time step and bin size) in both DEM and CFD domains worked well in CFDEM2.9.0 but not work in the version 3.5.0. The drag force computed by CFD is exceptionally high. Do you mean because I have used improper bin size, resulting in the wrong calculation of drag force (-> DEM force calculation) and making particles moving very fast ?

Regards,

j-kerbl's picture

j-kerbl | Fri, 01/20/2017 - 10:12

Hi Nathan,

no the neighbour settings shouldn't affect the drag-forces. But if the particles are much faster, of course the forces would go up due to higher relative velocity. The question is why the particles should become faster. Have you changed the LIGGGHTS version accordingly? What is the Youngs Modulus of your particles?
How does the case behave with DEM only, is it stable?

Btw. a new release of CFDEMcoupling and LIGGGHTS is available now, you might want to check that out too.

Cheers,
Josef