FLIP Viscosity - Collision object grid stepping artifacts

[Dave Stewart]

Hi All! I'm wondering if anyone has run into this and found a good solution, it's been popping up in my low-to-mid viscosity sims for a while. Perhaps I'm missing something stupidly obvious? Collision geo/field are fine and the interface between fluid and collision object looks great before turning on viscosity, then I get these grid artifacts that are tough to get rid of. Reducing Surface Extrapolation way down in Volume Collisions helps but doesn't completely remove the artifacts. Also, using a high slip-on-collision value helps, but I usually don't want the sliding behavior with sticky viscous fluids. Reducing Grid Scale reduces the relative grid size and helps too, but the artifacts are still there, just smaller.

Attached file is somewhat of a worst-cast example, just curious if other folks have dealt with this before. I know I can run some post-sim tweaks to clean up the collision surface intersection, but I've also noticed that the artifacts reveal themselves on the outer surface of the fluid if it gets thin enough, which is not so easy to fix...

01 - viscosity disabled

02 - viscosity enabled

03 - viscosity enabled, surface extrapolation 0.01

Thanks in advance for any help/ideas!!

--Dave

DStewart_FLIP_ViscousCollisionStepping.hip

[nigelgardiner]

have you tried increasing the sub steps on the solver when you turn viscosity on?

Machine tied up but I'll have a look at your hip when i can - that's an interesting artifact.

cheers

Nigel.

[Atom]

I can confirm your original findings. It does look like viscosity creates a stepping pattern. My guess is that behind the scenes the viscosity is fetching a lower resolution of the collision object.

 

This would be a nice example file to submit to SESI as a bug.

[anthonymcgrath]

shrink the collision shape a small amount perhaps to help it sit on the surface a little more?

[Dave Stewart]

Thanks for taking a look at this guys! It seems that it may be a limitation of the viscosity solver, as it's decoupled from the pressure solve. Sounds like this may get addressed in a future version though!

https://cs.uwaterloo.ca/~elariono/stokes/index.html

I'll go ahead and submit the file to SFX to see what they say about it.

Nigel, increasing substeps doesn't seem to have any affect on the artifacts.

Atom, it could be that or just a velocity matching issue relative to the grid scale. Got some interesting thoughts on this in the Houdini Artists FB group.

Anthony, I did try that and it helps with the fluid surface/collision interface, but not when the artifacts appear on the outside surface of the fluid as it gets very thin.

Cheers,

--Dave

 

[nigelgardiner]

Hi Dave

Thanks for that link. There's a link to a github for download of a new stokes microsolver which sounds like it could address your issue.

cheers

Nigel.

[michaelb-01]

Hey Dave,

 

I know this is quite old now but I've had problems with collisions when using viscosity in the past too. I've found I need to increase the exterior band voxels on my collision vdb to greater or equal than the surface extrapolation on the flip solver (default of 0.5). I wonder if this could fix your problem too? 

[Rgoldade]

The stair stepping is most likely caused by the solid-object boundary conditions in the viscosity solver. The stokes solver is mostly to couple pressure and viscosity at the free surface (i.e., liquid-air boundary) so I suspect you would see the same artifacts. In 16.0 we improved the solid-object boundary conditions but it's not perfect. Fortunately I do have a potential solution, I still have to work through the math and test it.

[Dave Stewart]

On 1/24/2019 at 8:29 AM, Rgoldade said:

The stair stepping is most likely caused by the solid-object boundary conditions in the viscosity solver. The stokes solver is mostly to couple pressure and viscosity at the free surface (i.e., liquid-air boundary) so I suspect you would see the same artifacts. In 16.0 we improved the solid-object boundary conditions but it's not perfect. Fortunately I do have a potential solution, I still have to work through the math and test it.

Thanks for the info Ryan, makes sense. SideFX confirmed the issue a while back, and my current solution is still the same though I haven't opened this test file in H17. Reducing surface extrapolation, lowering grid size and adding a bit of viscous slip helps, but doesn't eliminate the problem. 

Curious to hear more about your possible solution!

--Dave