FLIP leaks through collision geo container - Gatorade Shower Test

[venator11387]

I am relatively new to Houdini and I am starting to use FLIP. I started doing a test where a character gets splashed with a Gatorade Shower/ Bath:

As seen in the video attachment below, the FLIP particles leak through the cooler container where I put them (the collision guide geo is shown in the video). I have tried everything from upping the division size, thickness, substep, density, etc. but it still leaks through especially when it moves. I'm pretty much at a loss on how to fix this. I've attached the scene file below with the cooler geo.

FlipLeakVid.zip

FLIPgatoradeFiles.zip

[Nerox]

Hi Vincent,

I can't solve it but I do notice that the orientation of the primitive (polygons/faces) normals are not perpendicular to the primitive. I can't be sure but this might cause the trouble.

The trouble is that I can't figure out how to reset or reorient the primitive normal. Some one else might know the trick to this one.

Cheers,

[mangi]

Just at first glance

your collision geometry is closed

mangi

[venator11387]

Just at first glance

your collision geometry is closed

That's weird, how did you get that?! The primitives look fine on my end as well. Is it because I am using Houdini 12.0.683?

Edit: So I checked my scene using H12.1.185 and parts of the collision are not only closed, but have no normals on some parts. Unless someone knows, I think I may just start over to see what's up.

Edited February 9, 2013 by venator11387

[venator11387]

So after a couple of days of trying to work this out I am pretty much at the same point. I used the method of a separate thick collision geo around the container from Peter Quint's Fluid Winespill Tutorial. However, for some reason instead of going through the geometry (though some still leak through) the particles disappear in the areas they used to spill through. In addition, it seems like the FLIP particles disappear even though I filled up the container.

I included two vids that have the geo wireframed to show what's going on.

Has anyone ever had this kind of trouble with FLIP? :/

Leak_regularCollisionGeo.mov

Leak2_SeperateCollisionGeo.mov

Scene.zip

Edited February 11, 2013 by venator11387

[Nerox]

Hi Vincent,

Don't know if you've noticed it but its not just the leaking, in your example the water loses 80% of its volume as well. Have you posted this on sidefx.com? I would like to advice you to do so. You might have a chance one of the developers will have a look.

[rayman]

try this

• 
  If you are animating source/collision geometry at the SOP level, you need to tell the FLIP Solver to look for point-based collision velocities. On the FLIP Solver’s Collisions tab, set Velocity Type to “Point”. It’s usually a good idea to add a Trail SOP to your collision object set to “Compute Point Velocities”. Once you do that you should see good collision velocities in the visualization.
  (When animating geometry for use with dynamics, remember to use $FF (fractional frame) or $T (time) in animation expressions instead of $F (frame number) for proper sub-frame interpolation.)
  

[venator11387]

Hey Pavel,

I was actually animating the cooler on the object level with the FLIP object parented to it. I tried out your method, but a problem I came across is that even though the collision geo and the FLIP object are being animated (using transform sops), the FLIP particles will not move along with them (as seen in the attachment below). Unless that can be dealt with I won't know for sure if your method will work. As you can see though, there are still some leaks and popping occuring.

Nick, I have not tried SideFX yet, but thanks for the suggestion I will definitely do that asap.

Leak_NEW1.mov

[venator11387]

So after taking a better look (sorry for the amateur mistakes and posts) I noticed that though the container geometry moves, after turning on Show Collision Guide Geometry, the collision geo does NOT move with it.

Is there a way in the dop network to have the static object collision geo match the movement of the container geometry, which is animated by a transform sop?

[Nerox]

Toggle 'Deforming Geometry' on the collision object in DOP's.

[venator11387]

Thanks Nick! Sorry for the late reply. That actually did produce better results, but the amount of issues were still so great that I decided to switch to another method. I am now using an emitter for the process.

One thing I am trying to figure out is, how do I control the FLIP particle birth rate? I tried using a Pop Solver dop connected to the FLIP solver. Then within the pop I tried to use a Location and then a Source node, but with no success.

[johner]

Just got a chance to look at this...

You've got a thin collision object with very fast movement, which is very challenging for a fluid (liquid) solver. The main thing to think here is substepping. Basically you need to give the solver a chance to see as much of the collision motion as possible, which means it needs to solve multiple times per frame, i.e. more than the default of 2 substeps.

The short answer here is: increase the substeps on the DOP Network to 8, while setting the max substeps on the FLIP Solver to 1. I've attached the result, which doesn't require any separately-modeled collision geometry (although that's never a bad idea with thin collision geometry).

https://s3.amazonaws...torade_pour.mp4

Long answer: this is more complicated than just increasing the FLIP substeps, unfortunately.

At first you'd think, OK I need more substeps, just increase the Max Substeps parameter on the FLIP Solver. This doesn't really work because the solver decides to substep based on the velocity of the fluid, which only speeds up after the collision object has started moving, which means that crucial first movement isn't substepped enough and you get leakage right when the object starts moving.

OK, I'll keyframe the Min Substeps to 8 right before the collision object starts moving. Nope, in this version of Houdini those parameters are Set Initial on FLIP's Gas Substep node and can't be animated, although that has been changed recently.

Fine, just set Min Substeps to 8 on the FLIP Solver. Normally this would work, and in fact if you slow down the animation by a bit, say slide the final two keyframes from frame 24 to 36, this does work (frankly I'd argue the animation is a bit too fast here - those coolers are heavy!). The problem here is that while the FLIP Object is being solved 8 times per frame, the cooler Static Object is still only being evaluated once per frame, since it's only the FLIP Solver that is substepping. That means the position, rotation, velocity, and angular velocity of the collision object are only being calculated at frame intervals, with any substep values being linearly interpolated from one frame to the next. So your nice smooth Bezier animation curve is discarded in favor of linear interpolation during the frame, and you get instantaneous accelerations and such with which the fluid can't keep up.

However, if you substep the entire DOP Network, then both the collision object and the FLIP Object are being evaluated / solved 8 times per frame, so the collision object gets nice smooth values for position, velocity, etc. and the FLIP Object can keep up well. In this case that's your best bet.

(You also had some particles in your initial fluid state that were inside the collision object; they just got pushed out the bottom on the first frames, not really a leak.)

Cooler_substeps.hipnc

Edited February 19, 2013 by johner

[venator11387]

Wow, Thanks John! Turning the up the sub steps definitely has made things better!

[Alexpdennis]

this is a cool thread.. thanks for all the leads.. but i have done what you say and i still cant get the particles to stop shedding volume.. i also dont even have any animation. if someone could take a look at this or point me in a direction, that'd be awesome. thanks so much

martiniGlass.hipnc

Edited March 1, 2013 by Alexpdennis