How to prevent FLIP fluid from leaking?

[magneto]

Hi,

I made a simple scene with a bowl but the FLIP fluid is leaking out of the bowl right off the bat. Is there a way to prevent this? I played with the collision settings for bullet, but it seems it's limited. I turned off "Polygons as Convex Hull", but then I don't see anything for the collision. Is this normal? Should it at least not show me the same geometry when I am only showing the collision geometry?

I increased the substeps in the FLIP solver to 20 and it reduced the number of particles leaked but still it didn't completely prevent it. Is this the right workflow?

Here is how it looks without increasing the substeps:

Thanks

[JonathanGranskg]

Looks like a collision volume issue. Do you have enough divisions on the collision volume? And check the collision field on the flipobject, its resolution affects what's finally used for collisions... Not sure what else it could be.

[magneto]

Thanks, that solved it but which collision is being used for the simulation since this is a static object? Is it the RBD Solver tab or the Bullet Data tab? I know for RBD objects, you can choose the type using the Rigid Body Solver DOP, but I don't have it in this simulation.

Or should I always assume the RBD Solver tab, if there isn't anything that uses the Bullet Solver? Just seems like there is no indication of what's being used.

[JonathanGranskg]

If someone else is able to answer this that would be great, I'm not sure if my answer is correct.

I think the volume based collisions will always be used for interaction with fluids, at least that's the result I got when playing around with it just now...

[yongbin]

you can use thicker object only for collision. It's cheating, but works well.

[jkunz07]

Your particle separation looks to be quite low in the attached image, I would start with lowering the particle separation before increasing the substeps.

[Annon]

Where are you sticking your collision in the flip solver?

[magneto]

Thanks guys.

@YongBin: You mean model a separate thicker object?

@Christian: It looks like this:

@JonathanGranskg: So if the geometry is used in both RBD and FLIP sim, then effectively you will have to define 2 different collisions, and make sure they look identical even though they have different tools (RBD vs Bullet)? Anytime the mesh is changed, you will have to modify 2 collisions?

@jkunz07: This was the default value. I will try lowering it along with the substep.

[Annon]

Oh no, do it like this:

magneto_flipCollision.hip

[yongbin]

Yep, that's what I mean.

[magneto]

Thanks Christian, yours is much better. Mine ended up like that because I used the shelf tools, so not as optimal I guess.

Btw when I play the animation in your scene, I see some particles penetrating the bowl object deeply. Is this normal? Here is the pic:

[Annon]

You have to think of the centroid of the sphere, that's the point that's colliding, not the sphere, that's just a visualisation....

Higher resolution sim... Mesh.

Also a higher res collision as well I bet.

Edited August 19, 2013 by ChristianW

[magneto]

Thanks Christian, I thought the same for the centroid, but thought the ones that are almost fully enclosed inside the bowl geometry might be problematic. I will try higher res collision

[johner]

Note that Particle collision detection doesn't work with Source Volume-style collisions, as it relies on there being a collision DOP Object to collide with. In that case you'll need to use Move Outside Collision.

Also enabling Collision Separation on the FLIP Object and setting it to 1x or even 0.5x of your Particle Separation will give you a higher-res collision field even when just testing at low particle resolutions. Remember there are two collision resolutions at play here: the resolution of the collision SDF that you set on either the Static Object or the FluidSource SOP; and the resolution of the FLIP solver's collision field, into which it samples the SDF's of the various collision objects. If the collision field is too low-res, the solver won't really "see" the collisions, no matter how high resolution the original collision SDF.

Edited August 19, 2013 by johner

[magneto]

Thanks johner, that really clears up alot of details. What setting should I increase in the flip solver's collision tab though? There is velocity and surface extrapolation, but not sure if they are it. Or should I play with the Volume Fraction Method?

Also how do you guys visualize the volume based collisions accurately? RBD object's collisions are clear because you can see the geometry. But with the volumetric collisions, it seems it's harder to judge, especially when I am inside the DOPnet where both the geometry and the collision object disappears.

[Annon]

Just visualise collisions on the flip collision...

Play with it Ryan, you have enough experience with this stuff to work it out.

[Sifis]

Ryan check this out:

http://www.sidefx.co...journal=default

"Fixed a long standing bug in the FLIP Solver that caused phantom velocities along the bottom of curved collision containers such as bowls or cups."

Major changes in the collision extrapolation method. Maybe this is what you're looking for.

Edited August 21, 2013 by Sifis

[magneto]

Thanks alot guys, I didn't realize there was a bug on it, but glad that it's fixed

[johner]

Just so you're aware, this bug had to do with liquid standing or settling into a curved container (well, really any container, curved or otherwise). Prior to this fix, it was difficult to make a standing glass or bowl of water without phantom velocities around the bottom edges.

But everything else in this thread about making sure the solver can properly "see" you collision in the first place still holds.

[magneto]

Thanks johner, that's good to know. I didn't know there was an issue like this since I never tried it. So I guess the FLIP solver is always improving? Do you guys know of any other issues or things that it's lacking currently?