Question about particles settling under gravity in DEME

[Prabhat Paudyal]

Hello everyone,

I have been trying to replicate the 2D Goldenberg et al experiment in DEME. It was replicated as a validation study for Chrono::GPU (as described in the 2021 paper), so I figured it could be easily done in DEME too. However, I am having some issues with generating particles and having them settle under gravity. I have been using a nested for loop to generate the initial positions of particles and while I don't have any issues with the generation of positions (see attached paraview screenshot), I encounter the excessive velocity run-time error while settling the particles under gravity. I have tried to decrease the step size, increase the box domain dimensions to eliminate boundary effects, relax the physics (decrease the Young's modulus of the terrain), and increased the initial safety distance of the particles to no avail. 

I have attached my code to this message and I have been modifying the 2D ball drop demo. 

Thanks and regards,

Prabhat

[Ruochun Zhang]

Hi Prabhat,

To answer your question: You probably see a warning in your output file,

""""

WARNING! At least one clump is initialized with a position out of the box domain you specified.
It is found at 0.378, 0, 0.076383 (this message only shows one such example).
This simulation is unlikely to go as planned.
"""

This is the problem. It appears that the points you sampled live in x∈[0, 0.4], but the x range of your simulation world is defined as something like [-0.25, 0.25]. This will make the particles to be initialized in unexpected locations, cause large initial penetrations, and then destabilize the simulation. I fixed the world size definition, and it seems to fix the problem. The extremely fine step size you are using is not necessary: The physics in this simulation is simple.

By the way, in the upcoming paper about DEME, we will include a Goldenberg test example, and provide the associated demo script. If you want, we can let you know when it is officially published.

Thank you,

Ruochun

[Ruochun Zhang]

BTW you also need to give x and z a bit of room on their left, meaning not starting from 0, but starting from something like -terrain_rad, so that you don't have initial penetration there, either.

Ruochun

[Prabhat Paudyal]

Hi Ruochun,

Thank you so much for your response. I would very much appreciate it if you notify me once the paper gets published.

Thanks,
Prabhat

[Ruochun Zhang]

Hi Prabhat,

It's published and can be found here https://www.sciencedirect.com/science/article/pii/S001046552400119X?via%3Dihub

Thank you,

Ruochun

[Prabhat Paudyal]

Thank you Ruochun for letting me know. One more thing: when can we expect to see the demo script of the Goldenberg test example in the repo? I don't think it is there. 

Thanks,

Prabhat

[Prabhat Paudyal]

Hi, I'm just following up on my message here. 

[Ruochun Zhang]

Hi Prabhat,

Bona will be following up on this question on this thread soon.

Thank you,

Ruochun

[Bonaventura TAGLIAFIERRO]

Hi Prabhat,

you will find the new case at https://github.com/projectchrono/DEM-Engine/blob/main/src/demo/DEMdemo_ContactChain.cpp.

Note that he initial particle positioning was created as a list (outside of DEME). This allows keeping the template concise while only leveraging DEME's features (such as initiating simulations from files, changing properties on the fly, single instance control).

Thank you,

Bonaventura

[Prabhat Paudyal]

Hi Bona,

Thanks for sharing the source code. I will look into it and will reach out if I have any questions/confusion.

Thanks,

Prabhat