Pre and post processing for DEM-Engine

[Weigang Shen]

Hi Everyone,

I would like to generate a block which consists of a assembly spheres. Is it possible to generate the block by other code, and then load it into DEM-Engine via a file containg the informations (such as position and radius) of the spheres?

Thank you,

Weigang

[Ruochun Zhang]

Hi Weigang,

Yes. Again,  DEME only cares about the position and radius of the initial spheres, so you can supply it however you want. If you have a file that records this information using your own format, then the easiest thing is to have your own C++ or Python script that reads it into arrays and then feeds them to the solver. If your file shares the format with DEME's standard clump output file, or your file is simply the output of WriteClumpFile, then you can conveniently load clump positions and orientations by ReadClumpXyzFromCsv and ReadClumpQuatFromCsv (examples are in GRCPrep_Part1 and 2). Note that currently, sphere radius cannot be read from standard clump output files since it's part of the clump template information thus not in the clump output.

Thank you,
Ruochun

[Weigang Shen]

Hi Ruochun,

Thank you for your reply! I have a question about the clump. Do the particles in a clump interact with each other via any force model?

Thank you,

Weigang

[Ruochun Zhang]

No. That's because the sphere components in a clump are just shape representations, and they have no physics, only material properties. If they did have interaction with each other then every clump would've immediately exploded upon simulation starting.

Thank you,

Ruochun

[Weigang Shen]

Thank you for your reply! I have two questions about the demo "DEMdemo-Fracture-Box".

(1) After i have run this demo for one day, there is still nothing outputed. Is it fact that the DEME will run a long time once a fracture model is used?

(2) How to tell DEME the contact force models between different familys? I did not see any snippet which undertakes this work, although the demo has at least two kinds of contact force model. The question i really want to ask is that, if we have more than three familys in our model and there are more than two familys consist of spherical particles, how to tell DEME the  contact force model between different familys, and tell DEME the contact force model between the particles of a family in the same time.

Thank you,

Weigang

[Ruochun Zhang]

Hi Weigang,

I would like to answer Question 2 first. You can use variables AOwnerFamily and BOwnerFamily to refer to clump (or owner) A's and clump B's family numbers directly in the force model. So you can write if statements that treat each case accordingly. This is mentioned in the DEME paper: You can check out Table 2.

For Question 1, the situation is more complicated. We are actually able to reproduce the problem you encountered (simulation freezing), but on consumer/gaming cards only. It runs perfectly on data center cards and that's what we used for generating this experiment, therefore it did not appear a problem for us. It's likely due to different GPU architectures running the same code differently, we'll try to find a solution and let you know in the following days, so stay tuned. By the way, you are using a gaming card to run the simulations, correct?

Thank you,

Ruochun

[Weigang Shen]

Hi Ruochun,

My card is indeed a gaming card NVIDIA GeForce RTX 2070 Super. Could you recommend a card?

Thanks,

Weigang

[Ruochun Zhang]

Hi Weigang,

It's not like recommending a card; rather, if you happen to have a data center card available (A100, A5000 etc.), you can circumvent this problem for now by using them. Otherwise, let's hope we find a solution very soon, or at least find a workaround. We'll keep you posted.

Thank you,

Ruochun

[Ruochun Zhang]

Hi Weigang, 

The problem appears to be fixed now. Please pull the newest main branch and try the fracture demo again. I suggest you do a clean rebuild from an empty directory, because only a force model file is changed and if you just "make", cmake probably thinks nothing needs to be done. Or you can simply manually copy the modified cu file over to the appropriate location in the build folder.

The problem was that we left an uninitialized variable in the force model file. On data center cards, the compiler zeros it out automatically, but on consumer cards it does not. We therefore missed it initially. Should be all good now.

Thank you, 

Ruochun

--
You received this message because you are subscribed to the Google Groups "ProjectChrono" group.
To unsubscribe from this group and stop receiving emails from it, send an email to projectchron...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/projectchrono/1910af21-d1e3-43dd-a779-0c229e6150a6n%40googlegroups.com.

[Weigang Shen]

Hi Ruochun,

Thank you for your great work. The demo now is runs well. I have a question about the Fracture-Box demo. After the bond between two particles breaks, is it right that the contact model between them change defaultly from ForceModelWithFracture to Hertz–Mindlin model?

Thank you,

Weigang

[Ruochun Zhang]

Hi Weigang,

Note that in general, if you use a custom force model, then that model alone is used, there is no fallback so I wouldn't call that anything has "defaulted" to Hertz–Mindlin. Specific to the Fracture-Box demo, it's just that the force model (ForceModelWithFracture.cu) is written in a way that when the bond between two particles breaks, the "else" branch of that big if statement is executed, and something similar to Hertz–Mindlin is enforced.

I said it's similar since it's not a pure spring–damper model now, it's a spring–damper model with the resting location (zero-force location) being the penetration depth the moment when the bond broke. This implementation avoids a potential problem with a pure spring–damper model: Without the adjusted resting location, a newly broken-away pair of particles would see the bond suddenly gone but there were still left-over penetration, therefore immediately pushing each other away with high velocity, resulting in non-physical results. Shout-out to Bona for implementing this.

Thank you!

Ruochun

[Weigang Shen]

Hi Ruochun,

Thank you for your reply. In fact, I am puzzled that there is not any sentence in the  Fracture-Box demo to declare the contact model between spheres and the compressing plate. So, how to declare the contact model, especially for the case in which there has been another contact model, such as the  ForceModelWithFracture.cu?

Thank you,

Weigang

[Ruochun Zhang]

Currently, DEM-Engine custom model functionality permits only one script, meaning that in principal, you are writing a unified force model that goes for sphere-sphere, sphere-boundary, and sphere-mesh. Just think of the sphere-compressor contact as a special case, which is the contact between a normal sphere and an infinitely-large sphere.

About enforcing different models for different objects, you can associate different objects with respective family numbers, then use if statements to differentiate types of contacts. I talked about this in one of the earlier replies in this thread, please go back and have a look.

Thank you,

Ruochun

[Weigang Shen]

Hi Ruochun,

Do you mean that we can add a if statements in the ForceModelWithFracture.cu to  differentiate types of contacts?

Thank you,

Weigang

[Ruochun Zhang]

Hi Weigang,

Yes. And that's exactly what we did in the fracture demo to begin with, is it not?

Thank you,

Ruochun

[Weigang Shen]

Hi Ruochun,

I am currently modelling the impact of a rock block against a plate. However, as shown below, the spheres penetrate through the plate without any contact with the plate. Could you give some suggestions?

Thank you,

Weigang

[Ruochun Zhang]

I assume you are doing it via a custom force model. In that case, this problem is very similar to the exact demo we provided (which runs). You have to tell us what you changed in order for us to give suggestions.

Thank you,

Ruochun

[Weigang Shen]

Hi Ruochun:

The script is attached below.

Thank you,

Weigang

[Ruochun Zhang]

Hi Weigang,

This is because you prescribed the linear motion of family 1 (which is for all the particles) to be -1.0. When the velocity is prescribed, the simulation entities won't accept the influence of physical contacts, and will keep the prescribed velocity no matter what. This is exactly what we see in the simulation. Removing that line should fix it. You probably meant to add an initial velocity.

Ruochun

[Weigang Shen]

Hi Ruochun,

After removing the line which prescribe the linear motion, the particles can contact with the plate. However, if we have to prescribe a velocity for the particles, where should we add this line?

Thank you,

Weigang

[Ruochun Zhang]

That depends on whether we are on the same page about the concept of prescribing velocity. You understand that when the particles come in contact with the plate, they cannot keep whatever velocity you assigned them before, if you respect the contact force. So you either respect the contact force or do not respect it. By assigning a prescribed velocity to a family, you choose not to respect the contact force's influence on the entities, and always enforce this velocity. 

What I understand that you wish to achieve is to ensure the particles have a certain velocity at the moment they come in contact with the plate. That is not a velocity prescription. That can instead be done in some other ways. For example, initialize the simulation such that the particles are touching the plate, and give the particles an initial velocity then start the simulation. But I don't know if I understand what you wanted to do correctly,

Thank you,

Ruochun

[Weigang Shen]

You  are right. In fact, the scenario i would like to model is that the impact of a rock block onto a palte after a freefall. I would like to set a initial velocity for the rock block according to the freefall height. So, how to set a initial velocity?

Thank you,

Weigang

[Ruochun Zhang]

You can SetVel to simulation entities, such as particles, before the simulation starts. You can for example see how it is used at line 150 of DEMdemo_GRCPrep_Part1.

Thank you,

Ruochun

[Weigang Shen]

Hi Ruochun,

Thank you for your reply! Could you tell us about the data format of the input xyz generated by the HCPSampler?

Thank you,

Weigang