High volumefraction particles + periodic geometry

Submitted by tangl on Tue, 05/06/2014 - 11:46

Hi,

I am a new user of LIGGGHTS and I need it only for generating a 3D packing geometry. The 3D packing consist of a cube with five different particle sizes. I am dealing with two issues.

1) The volume percentage of the particles I need is 60%-70% and this high volume percentage is difficult to achieve. Therefore I start with a larger simulation box to place the particles and use fix deform to "shrink" the box to the correct dimensions to achieve the required volume percentage of the particles. This works but the only thing is that the particles are "squeezed" into each other (see attachment). How can I prevent this? I have read about rigid particles but do I have to assume each particles as rigid? Or other option I can think of is to adjust the particle-particle interaction but I'm not sure how to adjust this correctly to prevent penetration of particles.

2) The second issue is that I need to generate a periodic geometry. So for example, if a particle is partly cut by the top plane of the box, the cut part should be inserted at the bottom plane of the box at the same location. The same should hold for left-right plane and front-behind plane. See attachment to make this more clear. I know periodic boundary conditions are possible but I need the periodic geometry so somehow this should already be done with the placing of the particles. Does anyone knows how to solve this?

In the attachment also the input file I'm using can be found. Thanks a lot!

Regards,
Lisa

AttachmentSize
Image icon Deformed simulation box443.22 KB
Image icon Initial simulation box409.75 KB
Image icon Periodic packing geometry166.67 KB
Plain text icon Input file3.05 KB

Forums:

jtvanlew | Wed, 05/07/2014 - 23:37

are you absolutely certain that the particles are _actually_ overlapping each other? To me that looks like paraview point sprites representing their radius as larger than the physical radius. If you have assigned a young's modulus to each particle type, then there's not a strong chance that they are overlapping each other to the degree seen in your photo. How are you graphically drawing those particles in your system?

From your input file it looks like you're using periodic conditions already. I believe the way LIGGGHTS handles the periodic conditions is not to 'cut' the particle but to create a kind of ghost particle at the opposite boundary that behaves like its originator. then when you visualize it in paraview or something, you'll see particles extending beyond the boundary box and on the opposite side you'll see indentations where that other 'half' of the particle is influencing its neighbors. Are you not getting expected periodic behavior from some other measurable way?

jon

tangl | Mon, 05/12/2014 - 14:29

Hi Jon,

thanks for your reply. First I also thought it was just a visualization issue in Paraview and that I just accidentally used wrong setting to graphically draw the particles. Therefore, as a check I used Matlab to read the centre-points and (phydical) radius from the output file and plotted the particles in Matlab. Then I see the same overlap of particles as in Paraview so it seems that the particles are really overlapping each other. Any other suggestions to prevent this overlap?

Concerning the periodic boundary conditions, I know I'm using them already. Assume that LIGGGHTS handles the periodic conditions as you explained with a ghost particle, is it then possible to write the centre point and radius of the ghost particles to the output file as well? Then I can plot the ghost particles and see if the geometry is actually periodic.

Regards,
Lisa

jtvanlew | Tue, 05/13/2014 - 02:09

see sbateman's response below about the overlap. that might solve the first problem.

there might be a simple solution if you only want a qualitative confirmation that your boundaries are periodic. You could use an opensource software called "ovito" for visualizing the liggghts dump files (without even needing to lpp conver them first). With Ovito, when you open the dump file, there is an option (in the top right) to 'add modification'. if you look down, you can 'show periodic images'. a little context window will open below it and you can specify which directions are periodic and how many "periods" to visually repeat. With that shown, you can zoom in and around and see how the boundaries are interlocking each other.

if you want a quantitative confirmation, you should look through the forum for the contact force dump / visualizer (originally made by mark bentley, if i remember correctly) there are flags in that contact force dump that indicate if a pebble is periodic as well as which particle is in contact with which. that can confirm to you that particles are in contact across periodic boundaries.

at least that is how i'd handle it if i were you.

i don't know of a built-in way to do what you said (re-drawing ghost particles).

jon

tangl | Thu, 05/22/2014 - 17:10

Hi Jon,

thanks for mentioning Ovito. It is really a handy tool. I used a geometry with only 5 particles to check if I get the desired periodic geometry. With "show periodic images" I saw that it is correct and furthermore, you can export the periodic images and then the output file will contain the center point and radius of the particles of the periodic images as well. So I don't need to redraw the ghost particles within LIGGGHTS.

Still didn't solve the overlap of particles but I'm glad that I could obtain the periodic geometry. Thanks!

Regards,
Lisa

sbateman | Mon, 05/12/2014 - 21:05

In your input file, you don't set the timestep, so it defaults to 1e-8 seconds for units si. Then you only run for 1000 steps.

I really don't think you can expect to get a reasonable packing with only 1e-5 seconds of simulation time. You've basically just imploded the box, and the particles haven't had time to explode back outward.

PaulWinkler's picture

PaulWinkler | Fri, 05/16/2014 - 14:50

Hi,

the overlapping you produce result from the short time range which is considered in the simulation. Through the inertia you end your simulation, when the particles even start to accelerate for moving to there new positions. After the rearrangement, when the particles settled and are in rest it's time for an overlap check. Since I deal also with dense sphere packings (not shrinking the box but let the particles grow) overlapping can be a problem for very dense packings.
Is there a way in LIGGGHTS to limit overlapping to a given rate? Just for the skin depth maybe? Or is it possible with on-board commands to determine overlapping for all particles, since it is calculated for inter-particle forces anyway?

Regards,
Paul

tangl | Thu, 05/22/2014 - 10:06

Hi,

do you happen to know what is a reasonable simulation time so that the particles can rearrange while the box is shrinking? And you mentioned that after the rearrangement, it is time for an overlap check. I agree that is need to be done but does LIGGGHTS do this automatically or is this an option I need to define in my input file?

Regards,
Lisa

sbateman | Thu, 05/22/2014 - 18:06

Generating a good dense granular packing is a lot harder problem than you realize. Here's what I would suggest to get started:

Choose an appropriate timestep. See the documentation for fix check/timestep/gran to get some idea. It will depend on your material properties, particle sizes, etc.

If you are going to generate the packing by shrinking with fix deform, do some quick calculations to determine an appropriate rate of shrinkage and the number of timesteps it will take. In other words, how fast should your particles be moving? How much should they be allowed to overlap? How much time do the particles need to relax after each position remapping? Read through the documentation very carefully and run some simple tests to get some idea.

Also, read some papers about granular jamming and see if it applies to your system. As the volume fraction increases to the random-close-pack limit, you might start seeing some jamming effects.

Selim Duneskater | Wed, 05/21/2014 - 21:04

Hi all,

In order to achieve high values of the volume fractions, juste shake your simulation box (eg. vertical vibrations).

Best regards,

tangl | Thu, 05/22/2014 - 09:05

Hi all,

thanks for all your comments. I will try to use "ovito" to check if it is a periodic geometry.

Concerning the overlap, I already had a feeling that it maybe had something to do with the simulation time. I will set the timestep and see if I still get the overlap of particles.

I will let you know if I could solve the issues with your advice. Thanks again!

Regards,
Lisa

tangl | Thu, 05/22/2014 - 09:28

I want to have a volume fraction of 0.6-0.7. I haven't try shaking yet, but it is an option.

Selim Duneskater | Sat, 05/24/2014 - 13:49

Hi,
You may go up to 0.64. Google the work on compaction by Patrick Richard et al. (Université de Rennes, F).

zamir | Fri, 05/23/2014 - 17:50

Shaking a dense packing would indeed loosen it, but, if you continue to shake it, it will redensify. The solid fraction vs time plot will end up looking like a sine wave. There is a lot of literature that shows this in physical agggregate packing. The best amplitude and period of the vibration will need a small study to identify though.

Another approach is to reduce the friction of the particles whicle applying a static force. When you acheive your desired solid fraction, reset the friction back to your desired simulation value. This has also been shown in DEM literature.

Later,
Zamir

PaulWinkler's picture

PaulWinkler | Tue, 05/27/2014 - 16:49

Hi Zamir,

thanks for the hint with the sine wave, since I am civil engineer things in my universe normally don't move. ;-)
We are setting friction to zero and compact then with a static pressure. This works very well, but now I am dealing with determining the value for the pressure to get a realistic volume fraction / porosity...

Regards,
Paul

atul2018 | Fri, 10/22/2021 - 12:42

Hello All

I also want to generate a dense packing of 60% solid fraction which seems more difiicult than i expected. Is someone found a sophesticated way to generate dense packing, any more suggestion?

Best Regards
Atul