FLIP in realworld scale - strange behavior

[au@]

Hello guys!



In order to get realistic splash I scaled down my scene to real world size.

FLIP volume is 8x8x8cm now. I didn't change Hip File Options (I work in meters) but I decreased particle separation.

When I begin my simulation fluid starts to "boil" and doesn't react to falling geometry.

Increasing "Minimum substeps" to 2 calms down fluid a bit but doesn't completely solve this problem.

Spatial scale doesn't affect the fluid in this case.



What am I doing wrong? Please, help!

FLIP_realworld_scale.hip

[lukeiamyourfather]

Try many more substeps like 25 or 50 (maybe even more than that). Change it at the DOP network level so it affects all of the solvers, not just the FLIP solver.

[au@]

Thank you very much! Problem solved.

 

High substeps increase simulation time.

What is the best tactics for such scenes?

Work in meters and use low spatial scale values?

[danw]

The only tactics are to try and figure out a reasonable number of substeps for the scale you want to represent.

 

If you're modelling reality, gravity is always ~9.8m/s^2... so the smaller scale you want to represent, the more grid cells fluid is typically going to travel over in a single frame.  For FLIP fluids, you typically don't want it travelling over more than 2 or 3 cells in a single substep, tops.  If you're representing an 8cm splash, and you've got a tiny particle separation to give you enough resolution in a sim that size, then gravity is going to tend to make the fluid travel across 10s of cells per frame, so you need to keep factoring up your substeps until you get to at least no more than 2-3 cells per substep.

 

There's not really any trick.  Small scale fluids travel quickly relative to their absolute size... you need more timesteps to simulate them.  You can only get away with less on large scale sims specifically because you're sacrificing resolution for scale, and assuming a distant enough camera position to get away with it.

[lukeiamyourfather]

There's no meaningful correlation between substeps and scale. It's about how much movement there is between each frame. It just happens to be that the movement between each frame is significant in this small scene, but not because it's small. The same could be said for a huge scene like water tower with an airplane flying into it, there would still be significant movement between each frame and it would require many substeps to solve. Not because it's big or small but because the movement between each frame.

[danw]

Sure, there's no hard-and-fast rule about it... but a smaller scale sim is going to require much smaller voxels to resolve to an acceptable level of detail.  Whatever happens to be going on in your sim in terms of behaviour and forces, if you attempt to resolve that same simulation with larger or smaller voxels, the one with the smaller voxels will require higher substepping in order to behave correctly.  Smaller scale sims tend to have smaller voxels, because you're focusing much closer and so require much higher local detail.

My point was only that for a typical simulation of standing water, acting under gravity alone, the resulting motion from that gravity is going to tend to cover more voxels per frame if you're simulating a small, focused region with smaller particles separation, than if you're focusing on a larger area and correspondingly increasing your particle separation.

 

There's no definitive correlation, but I'd argue there's certainly a meaningful enough one to make a ballpark estimate of the number of substeps you'll require.

[au@]

Thank you for your clear and helpful explanation.