5 replies to this topic

[Netvudu]

Hi there.
Trying to solve a problem with a scene I tried going back to a simple setup of a vase full of water and I´m finding it much more difficult than anticipated.

Check the attached scene, please. As it is, with 120 divs for the collision volume, the particles still disappear. I have to activate Laser Scan to get something sort of useable...and still it´s tough to fill a glass. Anything I´m missing here?

I´m trying this on the latest build. Previous builds show the same behaviour.


 TestFLIPVase.hipnc   1021.19K   44 downloads

[Erik_JE]

More substeps and thicker collision geo is your best bets.

[johner]

Yes, a filled glass / vase setup is harder than it should be in FLIP, unfortunately.   FLIP is really good at large, dynamic scenes, but less good at more static ones.  That said, there are things you can do to fix this setup, mostly having to do with resolution.  A thicker collision object would help, as Erik notes, but the main problem is that the collision field used by FLIP is too low-res in your test simulation res.  

If you enable the Collision guide on the FLIP object in your file, you can view the fluid's collision field; the object's collision SDF is sampled into this field when calculating fluid pressure. In this case you'll see lots of holes in the field, as it's too low resolution to represent the SDF accurately at your test sim resolution.  

However, the Particle collision detection method uses the object's hi-res SDF directly.  Usually these two methods represent about the same thing, but in a case like this you've got accurate particle collisions, but the fluid solver isn't really seeing the collision properly, so there's not enough fluid pressure keeping the particles in hydrostatic balance.

A thicker collision object helps this, but probably the best answer is to increase your collision and simulation resolution.  You can enable Collision Separation on the FLIP Object and specify a higher res collision field than the other FLIP fields for exactly this reason.  Even then you'll probably need to use an overall higher resolution simulation to fully resolve the collisions.

In the attached I've made those changes and a few more:
- changed the initial boolean ops to create fluid particles with less of a gap to the vase wall
- re-enabled reseeding which also helps fill in gaps between the fluid and the vase wall
- increased the SurfaceExtrapolation parameter slightly which (you guessed it) helps with the same thing
- increased minsubsteps a bit so you get more substeps even when the fluid is barely moving.

With these changes you get a fairly static sim, although you still start to get spurious internal currents after a few seconds.  But obviously is the  fluid is really static there are better ways to model it .

Attached Files

•  TestFLIPVase_hires.hipnc   1018.63K   64 downloads

[Netvudu]

Thanks for the detailed explanation johner. It makes sense.

So, is this the proposed method by SESI for such situations, or are we supposed to be using (what I though it was legacy) the Particle Fluid solver?

[johner]

Netvudu, on 03 May 2012 - 02:05 PM, said:

So, is this the proposed method by SESI for such situations, or are we supposed to be using (what I though it was legacy) the Particle Fluid solver?

FLIP is the preferred method for liquid simulation as of H12.  You can usually get acceptable results with "filled container" type FLIP sims, although at the moment it's definitely harder than it should be to prototype those at low resolutions.

Just as an example, attached is an flipbook of adding an emitter to the above setup to fill the vase (this is really more like filling a barrel - the vase is 2 meters tall).

Attached Files

•  vase_fill.mp4.gz   5.9MB   184 downloads

[JuriBryan]

hey try to keep everything at the settings that you have right now and just change the "sign sweep Threshold" to something like 1 or below it.
then change the "max sign sweep count" to something like 9.
It takes a bit longer to calculate the volume, but it gives you a rock solid collision mesh even with lower division settings.
And its only calculated ones at the first frame when its needed, so the time is alright.