Error with high Young's modulus (11.e9 Pa) and small particles (from 100 to 600 microns)

Submitted by limone on Thu, 10/26/2017 - 19:34

Dear All,

I am struggling to make start a simulation with high Young's modulus (11.e9 Pa) and small particles (from 100 to 600 microns).
Here part of my code:


atom_style granular
atom_modify map array
communicate single vel yes
boundary m m m
newton off
units si

# When youngsModulus is > 1.e9 use the command 'hard_particles yes'
hard_particles yes

region reg block -0.120 0.120 -0.120 0.120 -0.001 0.570 units box
create_box 1 reg

neighbor 0.001 bin
neigh_modify delay 0

# Material properties
fix m1 all property/global youngsModulus peratomtype 11.e9
fix m2 all property/global poissonsRatio peratomtype 0.3
fix m3 all property/global coefficientRestitution peratomtypepair 1 0.3
fix m4 all property/global coefficientFriction peratomtypepair 1 0.5

# pair style
pair_style gran model hertz tangential history # Hertzian without cohesion
pair_coeff * *

# timestep, gravity
timestep 0.00001
fix gravi all gravity 9.81 vector 0.0 0.0 -1.0

# particle distributions and insertion
region bc cylinder z 0.0 0.0 0.03 0. 0.065 units box

fix pts1 all particletemplate/sphere 15485863 atom_type 1 density constant 2000 radius constant 0.000600
fix pts2 all particletemplate/sphere 49979687 atom_type 1 density constant 1500 radius constant 0.000100
fix pts3 all particletemplate/sphere 32452867 atom_type 1 density constant 1500 radius constant 0.000250
fix pdd1 all particledistribution/discrete/numberbased 86028157 3 pts1 0.6 pts2 0.3 pts3 0.1

Depending on the radius I set for my particles I get the error "segmentation fault" or this one:


ERROR on proc 2: Too many neighbor bins (../neighbor.cpp:1776)
ERROR on proc 3: Too many neighbor bins (../neighbor.cpp:1776)
ERROR on proc 1: Too many neighbor bins (../neighbor.cpp:1776)
ERROR on proc 0: Too many neighbor bins (../neighbor.cpp:1776)

Any suggestion from anyone ?

Best regards,
Limone

Forums:

paul | Fri, 10/27/2017 - 11:05

You have boundary m m m, which means that the domain expands (= more neighbor bins) as particles attempt to leave it.
And you chose a very unpleasant set of parameters (fine, hard particles) - and a very high timestep.
Thus, everything blows up. Choose a smaller timestep or bigger, softer particles.

Have a look at the work of van Lommen (2014), who investigated the influence of the Youngs Modulus on bulk behavior in DEM simulations,
you can probably get away with softer particles.

Greetings,
Paul

limone | Fri, 11/03/2017 - 13:26

Many thanks Paul for this useful email.

About the boundaries I chose "boundary f f f". I hope it is OK.
About the particle size, Young modulus and timestep, I would need to maintain the particle size. It is really important for my research.

Thanks for the paper you mentioned, It is really interesting:
Lommen, S., Schott, D., & Lodewijks, G. (2014). DEM speedup: Stiffness effects on behavior of bulk material. Particuology, 12(Supplement C), 107-112. doi:https://doi.org/10.1016/j.partic.2013.03.006

In the angle of repose test, they showed that the angle of repose and the bulk density do not change, when the shear modulus G lies in the range from 〖10〗^11 Pa to 〖10〗^7 Pa. I remind that the shear modulus is related to the Young modulus E and Poisson ratio ν, according to the relationship E = 2G (1 + ν). Therefore, by decreasing the shear modulus from 〖10〗^11 Pa to 〖10〗^7 Pa we can get (almost) the same result, in this case the angle of repose, saving computational time, with a time-reduction of about 100 times. Below the value of 〖10〗^7 Pa, the simulations give quite different results.

In this moment I am using the Young modulus and particle dimensions as described in my previous post, but I decreased the DEM timestep (), the CFD timestep as follows:

DEM timesep - timestep = 0.0000001 (1e^-7) [/data/ErgunTestMPI2d_Cluster/DEM/in.liggghts_init]
CFD timestep - deltaT = 0.000001 (1e^-6) [/data/ErgunTestMPI2d_Cluster/CFD/system/controlDict]
couplingInterval = 100 [/data/ErgunTestMPI2d_Cluster/CFD/constant/couplingProperties]

If the simulation is too slow I will decrease a bit the Young modulus, increasing the DEM and CFD timesteps, accordingly.

Just a last question: do you think the boundary f f f can be a good choice now ?

Best regards,
Limone

Rachel | Mon, 11/20/2017 - 19:17

Dear Limone, that's a good point. I will also follow your step in my simulation. Have you found a good way of solving such big Young' module problems? I am using the Young's module to be 5.0e7 Pa, and as you said, I think this module cannot be decreased in the simulation coz the relatively large difference mentioned in Lommon(2004). One more question, do you know whether the pission ratio will significantly affect the results of simulations? Any suggestions are greatly appreciated. Many thanks.

Rachel

limone | Fri, 01/12/2018 - 17:47

Hi Rachel,

This DEM aspect (the "time-step - Youngs modulus - particle size" relationship) has been and still it is a big/difficult issue.... A possible way to workaround it is to use the "coarse graining"

In Daniel Schiochet Nasato, Christoph Goniva, Stefan Pirker, Christoph Kloss, Coarse Graining for Large-scale DEM Simulations of Particle Flow – An Investigation on Contact and Cohesion Models, Procedia Engineering, Volume 102, 2015, Pages 1484-1490, ISSN 1877-7058, https://doi.org/10.1016/j.proeng.2015.01.282.
"Discrete element modeling is a numerical method that simulates particle dynamics based on individual particle
collisions. Therefore computational effort increases with the increase of the number of particles simulated, which
may result in prohibitive computational times when simulations of more than 10 million particles are necessary. "
"Coarse graining allows reducing the computational effort by replacing individual particles by representative
parcels, substantially reducing the required number of particles to represent a process. However analytical
considerations and verifications are necessary to ensure that the physics is captured correctly. "

In case you do not want to use "coarse graining", or you do not have access to it in your DEM, you need to take a decision... often between the Youngs modulus and the particle size.... Several papers still consider particle sizes larger than 1 mm, even though several real particle size distributions are less than 1 mm. A similar comment could be done between the Youngs modulus in numerical tests and the real value of the materials.

My suggestion is to proceed similarly to Lommen et al.(2014)...Which means trying to perform the same numerical test with 3 or 4 values of Young modulus and check how a couple of significant quantities in your system change, by varying the Young modulus... If the variation is really negligible you can go ahead with the smallest Young modulus (which variation is small) and in the paper you are gonna write you can justify this choice with an evidence (even though several reviewers will attack that approximation)....

I hope it is clear :)

Cheers,
Limone

Rachel | Sun, 01/14/2018 - 03:56

Thanks Limone, I will try your suggestion soon to check the effects of Young's modulus. Also, if you are convenient, I have another question. I just restart the results of a pure DEM code in a new CFDEM coupling code. I found the void ratio changes sharply when I start the new CFDEM coupling code. For example, in my pure LIGGGHTS code, I get the void ratio to be 0.5, but when I restart the results in a CFDEM code, I found the void ratio change from 0.5 to 0.4 very soon. I cannot figure out why, do you have any ideas? Thank you very much.

Rachel

limone | Mon, 01/15/2018 - 16:49

Hi Rachel,

About this question I do not how to help you. I am sorry. But I would suggest to put this question in the LIGGGHTS/CFDEM forum, maybe with more details... otherwise a bit difficult to understand what you are looking for exactly...

"Add New Forum Topic" in https://www.cfdem.com/node/add/forum/0

BR,
Limo

AndresMM | Tue, 11/21/2017 - 10:03

Hey Limone,

are you certain that your CFD-Timestep, DEM-Timestep and coupling-interval agree with each other? Your coupling interval is 100, but the CFD-Timestep is 10xDEM-Timestep

Cheers,
-A

limone | Fri, 01/12/2018 - 17:22

Hi -A,

I have just checked the three quantities and they are exactly what I wrote. But you are right, the coupling interval should be 10 with those DEM and CFD timesteps (something strange)...... BTW the simulation was working.... Let me try putting coupling interval as 10....

Riccardo Maione's picture

Riccardo Maione | Fri, 11/03/2017 - 17:05

Hello limone,

the f f f boundary condition stands for fixed, meaning that your box (the one you choose with the create box command) will remain fix and particles, exiting this box will be automatically eliminated.
I deduced that you are making angle of repose test by setting a pile is that right? what I would suggest is to choose an adapted simulation box (so that particles belonging to the pile won't be eliminated erroneously) and to use the f f f boundary condition

best regards,
Riccardo

limone | Fri, 01/12/2018 - 17:25

Many thanks Riccardo!

I am doing fluidized beds (later also a angle of repose test as well)....... I am using the f f f as you suggested and the simulation is working well :)

Thanks a lot (and sorry for the late reply - was busy with other tasks),

BR,
Limone