This is a very interesting case. I reduced the number of particles from 100K to 10K to speed up the simulation and get an idea what’s going on without waiting all day (always a good idea when troubleshooting, reduce the particle count).
I also changed the VTK output frequency to 0.001 s so we get more detail, and colored the particles by Y-velocity.
In this view, the opacity of the geometry has been reduced from 1.0 to 0.25 and you can see, if you look closely, at around 0.05s, particles are getting inside a region where they should not… but this is a separate issue).
I see that you have disabled the fluid solver, so there’s no drag to slow down particles, but still this looks very unphysical. At first I thought the issue might be that the particles were too closely packed, and are hitting the barrier as a block, sending shock waves through the entire assembly of particles. But opening up the particle spacing (“Show DEM seeding options”) did not really help:
Next I decided to see what 1 particle moving through the system looks like - notice that the system is pretty “bouncy” with the spring coefficients and collision model you have selected - you may want to try a different collision model or change the parameters:
After modeling the single particle I decided to try 2 particles and then I found this:
AHA! - It looks like what is happening here is that the particle is colliding with the very sharp edge of the baffle, and something in the calculation is causing a very high rebound velocity - perhaps due to the very small contact area of the collision. These extremely fast-moving particles are then colliding with the rest of the particles and passing on their kinetic energy, resulting in the swirling mass of particles we are seeing.
So I think we need to either smooth out that rough edge, or experiment with the collision model. Keep trying with a small number of particles until it looks reasonable, then ramp back up.
Here’s one more run, with 20 particles, at an even finer time resolution (vtk_dt=0.001s)
At about 0.055s there is an interesting phenomenon where momentum is transferred from one particle to another along a chain (a la Newton’s Cradle) which causes the rightmost particle to be kicked to the right with a fairly high velocity. But I think most of the high-velocity particles are coming from the collisions with sharp edges, as can be seen at around 0.133 s.
Hi Charles
Thank you for your sincere help in analysing this unphysical solution
and it is really great that you’ve got an interest about particle phenomena of the structure
I thought because of the mesh inside of it, but it is not.
Based on your research, the problem is related with the sharp edge area, which has to be smoothed to ensure the proper collision to particles
for the particle inside the structure, I will change the spring coefficient so that It cannot go through the geometry
Then what I have to do is
shrink the number of particles for smoothed geometry until it gives physical solution
I will keep track of this and if it shows better result then I will reply this again.
I don’t see any files attached to this post. A spring constant of 100 or 500 is very small, please try larger values (100, 1000, 10000) to see if this makes a difference. The issue could also be the des grid setting as it looks like collisions are being missed. Maybe try to increase the neighbor search frequency or change the des grid spacing.
sorry about the missing file, since the file is my research project, I just removed it from the original post
yeah I am tyring to increase the spring constant, and will try your suggestion
I will keep update the result for charles and you later
Sounds good. Please follow Charles’ advice and work with a simplified setup when troubleshooting. You can also attach a different (simpler) setup that represent the issue if you can’t share your exact setup.
Just for comparison, here’s 10K particles with a Hertzian collision model. (I guessed at the Young’s modulus, restitution coeff, etc)
There’s still a lot of bouncing around up top but it’s much better
Thank you for providing me the comparison data, I will use this model later I when I finish your first advice.
Now I am running the case with geometry including the fillet at the tip so that it shows the normal vector on that surface
This is before using fillet
and this is after I used a fillet
once I get result from this, I will upload in this post
