Dear Sirs,
there seems to be a bug which is causing an error in the interaction between mesh surfaces and granular particles depending on the processors specification.
I try to run the attached case with LIGGGHTS_2.0.4 using "nice -19 mpirun -np < in.device"
For some processors choice, like "processors 4 1 1" everything works fine. The particle touching the edge of the hole in the center of the box rebounds with finite velocity. However, for many other processors specification (for example, "processors 3 1 1") the surface of the particle does not see the wall. Instead, the particle "passes through" the wall, and when its center of mass reaches the wall height at the edge, an "explosion" occurs, i.e. the particle rebounds with a huge velocity which is totally unrealistic for the problem.
Maybe I am missing something like how to map the particles or something like that? Or is it really a problem of the mesh-granular interaction implementation of LIGGGHTS?
Thanks so much in advance for the help!
Eric
| Attachment | Size |
|---|---|
 Box_with_hole.tar_.gz | 2.79 MB |
ckloss | Mon, 07/30/2012 - 16:52
I'll have a
I'll have a look!
Thanks,
Christoph
ericparteli | Wed, 08/01/2012 - 15:02
Thank you, Christoph, just
Thank you, Christoph,
just run a simulation with more particles and a different initial condition and the simulation crashed at some point *also* with that "good" processor choice (4 1 1).
So the existence of that problem, namely that a particle may pass through the mesh geometry, is clearly not particular to specific processor choices.
So far the problem always occurred at the edge where the hole in the box begins - maybe there is a problem if the surface is composed of two planes that make an angle? Although I did not try other configurations with sharp edges.
Or maybe it is a problem of interpretation of the .stl file by LIGGGHTS (edges not well interpreted)?
Please tell me if you think I could further help to find the source of this problem.
Best
Eric
ckloss | Sun, 08/05/2012 - 14:21
Hi Eric, I assume it has to
Hi Eric,
I assume it has to do with fix rigid. Edge contact is implemented and tested quite thoroughly, so I don't think this is the issue. I will check and come back to you
Christoph
ericparteli | Sun, 08/05/2012 - 19:42
spherical particles
Hi, Christoph,
thanks for the hint and thanks for helping me with this.
I just run a simulation with spheres instead of using hybrid molecular granular.
The problem persists, please see the case in the attachment, which includes a movie and the script files.
While this simulation does not exclude that a problem my exist in the fix rigid as you suggested, it does show that either something else in LIGGGHTS is causing this error or there is something wrong in my script files.
Maybe my .stl file is such that LIGGGHTS is not interpreting correctly the edges (an error in the .stl file)? It is difficult for me to figure out whether this might be the case. I just produced the hole in the box using the "Bohrung" option in Inventor. Indeed, all particles that fall within the hole without interacting with the edges behave correctly. Only those for which the center of mass falls at the edge (or sometimes with a small offset) show this problem.
I also produced a box with a circular hole using the same procedure in Inventor, and similar problem occurs.
Thanks again for all your help and time.
Best
Eric
ckloss | Tue, 08/07/2012 - 23:41
Ok, got it... it was a
Ok, got it... it was a parallelization problem after all.
Will be fixed in 2.0.6
Best regards,
Christoph
ericparteli | Mon, 08/13/2012 - 11:54
simulation with 2.0.6 working correctly
Dear Christoph,
indeed the simulation with LIGGGHTS 2.0.6 is now working correctly regardless of the processor choice, both for spherical particles and for the "rigid molecule" style!
Thanks so much for your excellent work and for your help on this!
Best regards from Erlangen
Eric
ckloss | Mon, 08/20/2012 - 21:37
You're welcome! Cheers,
You're welcome!
Cheers, Christoph
sgeerster | Tue, 03/29/2016 - 21:27
Seeing explosions/popcorn behavior
Hello All,
I seem to be having an issue similar to what Eric was experiencing, although I'm using a version of LIGGGHTS-PUBLIC downloaded recently, i.e. sometime since 2015.
I have particles falling into a funnel which then discharges in to a dish below the funnel nozzle. The funnel and dish are generated and imported as *.stl files from a CAD program. As particles pass the funnel nozzle, I get jumps/explosions or behavior that resembles the popping of corn.
I believe it's a parallelization issue. I've run two identical funnel/dish problems on two different parallel computing systems. One system exhibits the popping behavior while the other does not.
Any suggestions?
Thanks,
Steve
ckloss | Mon, 04/25/2016 - 21:18
Hi Steve,
Hi Steve,
the issue mentioned above was very specific to fix rigid (which is deprecated in the meantime) and has long been fixed. If you can post a minimal set-up which shows the possible issue in < 1 min on <=4 cores, I'll be happy to take a look!
Best wishes
Christoph