Advice for the simulation of roller compaction (powders)?

Submitted by neven.marticnevic on Fri, 07/28/2017 - 14:43

Hi all,

I am doing my master thesis (simulation of roller compaction-pharmaceutical powder).

I have verry big problems with high number of particles (mean diameter is 120 µm).

For 1 g of powder I need 700 000 particles. That is a lot because for filling volume of roller compactor I need "rough estimation" 20 milion particles.

So for that many particles I don't have proper hardware.

In roller compactor I must meausre density in specific regions (picture "Density measure").

Any advice how to do that with 4 cores for now??? I hope that my professor will acomplished install and compile liggghts on hpc workstations with 40 or 200 cores. But this is stil to large number of particles???

And one more question about size of mesh. Two pictures is in attachment where You can se size of particle and mesh size. Is mesh size good???

Best regards,Neven.

arnom's picture

arnom | Thu, 08/03/2017 - 09:57

Mesh size looks good. What we generally do is make the particles larger than they really are. We then use these larger particles to replicate calibration experiments (e.g. inclined wall) to figure out what parameters we need to chose so that the large particles behave as the bulk of the small particles. This allows you to significantly reduce the number of particles required. Calibration for DEM is its own science, so I suggest you read up on the issues associated with it.

DCS team member & LIGGGHTS(R) core developer

neven.marticnevic | Mon, 08/21/2017 - 15:03

Dear arnom,

I have problem with making compacted ribbon of specifed compacted density 0,93 g/ml during roll compaction.

First I made experiment where I measured angle of repose (37° I get).
Bulk density (0,69 g/ml) , true density (2,203 g/ml) and compacted density (0,93 g/ml) I get from pharmaceutical company which send me 2 kg of powder. Mean value of powder PSD is 60μm.

So I scaled up mean valueo of PSD six times (Rparticle=360 μm) and make DEM calibration of angle of repose. Some coefficients was taken from literature where AVICEL PH 102 and Di-Cafos was used. I understand that particle size have verry big influence on bulk density because small particles have high affinity to create tree structures whit big pores in it. So I reduced true density of particles (1,4 g/ml) to get bulk density (0,69 g/ml).

I used Hertz-Mindlin (no slip) with rolling friction cdt and sjkr model in calibration.
Following material parameteres and coefficients gave me acceptable angle of repose:
PP- particle-particle, PG- particle-geometry
Constant parameters and coefficients:
Young modulus: particle- 1.8e+7 ,geometry- 2.1e+8
Poisson ratio: particle- 0.3. geometry- 0.3
Restitution coefficient: PP- 0.83,PG-0.8
Coefficient of friction (COF): PP- 0.8, PG- 0.35

Coefficient which was changed:
Calibration 1
Coefficient of rolling friction (CORF): PP- 0.3, PG-0.15
Cohesion energy density: PP- 160000, PG - 70 000

Calibration 2
Coefficient of rolling friction (CORF): PP- 0.052, PG-0.02
Cohesion energy density: PP- 191250, PG - 70 000

After first calibration I made calibration with these coefficients on roller compactor where I want to get ribbon of compacted density (0,93g/ml).
But every time ribbon fall apart, in attachment You can see pictures.

What I am doing wrong?
Do I need to use Hooke model maybe? Because Hooke model have discontinuous contact force which ion the end of contact have attractive properties.

Best regards, Neven.

arnom's picture

arnom | Mon, 08/21/2017 - 16:02

Your problem is that the sjkr model might not be suitable for this case. The reason for this is that it does not include any history effect (i.e. you want that the particle remembers that it was already compressed). We have models in our premium repository that would deal with this and other physics that you might be seeing.
One possibility for you would be to implement the model by Obermayr et al. 2014 (A discrete element model and its experimental validation
for the prediction of draft forces in cohesive soil). The cohesion model that is part of this paper would probably do what you want.

Another warning: Setting the particle density to the bulk density is not fully correct as your particles will not cover the whole particle continuum (volume fraction: ~0.6).

DCS team member & LIGGGHTS(R) core developer

neven.marticnevic | Mon, 08/21/2017 - 18:18

First of all thanks for advice!

About bulk density:
For true density 2,203 g/ml I can't get in DEM bulk density 0,69 g/ml. So I lowered true density to 1,4 g/ml and then I get bulk density 0,69 g/ml.

Is there a different way to get so high ratio betwen true and bulk density with scaled (coarse-grain) particles in DEM?

arnom's picture

arnom | Tue, 08/22/2017 - 08:59

Sorry I misread your comment above. What you do to choose the density is fine :)

DCS team member & LIGGGHTS(R) core developer

neven.marticnevic | Fri, 08/25/2017 - 21:15

Hi arnom, sorry for bothering!!!

But I have two questions.
In followed link there is a video of animation (compaction). Link: https://drive.google.com/file/d/0Bwbhv1GHC2pZZzZPRnEwNmJoMUE/view?usp=sh...
And in attachment there are two pictures "region of density measuring" and "Density graph".

In picture "Density graph" there is line chart which is showing measured densities in regions which are showed in picture "region of density measuring".

Is it possible to get with current Hertz/tangential history/ sjkr/cdt rolling model bigger difference between densities?

I mean betwen densities on begining (0,3s) 710 kg/m^3 (blue line) and on the end of simulation (now for blue line 770 kg/m^3).

Why I get different values for densities (blue and red line)? That is a little bit strange.

In video, Sila (N)=Force (N), Hitrost(m/s)=velocity(m/s).

Best regards, Neven

neven.marticnevic | Thu, 08/03/2017 - 16:00

Thank You on advice,

I readed one article about scale up and watched EDEM Webinar yesterday where they sayed that it is possible to scale up particles and then with calibrations You can get pretty fine results.

Best Regards,Neven

AnjanaKittu | Wed, 04/11/2018 - 00:50

Hi,
for my simulation, I have to get stress, strain and strain rates in specific regions in an assembly. Could anyone suggest me how to code that for the specific regions? I call this measurement spheres. Any help is appreciated.

Thank you,
Anjana