FLIP Randomly shooting out particles

[7ofDiamonds]

Hey all!

I'm having an issue with a FLIP simulation I'm trying to achieve. The particles are randomly shooting out (after about frame 40).

I cannot pinpoint if it's a problem with the collision or something else.

 

I've attached a HIP file (H14). I would appreciate any help. Thank you!

 

 

FLIP_test.hipnc

[Atom]

I think that is just normal operation. The simulation looks ok to me on H15 after I reduced your Uniform Divisions down from 300 to the default of 30. The shoot offs are just the wave recoiling against the curved surface of the broken globe.

Edited October 27, 2015 by Atom

[TheDude]

for the collisions, I'd use proxy volume, and plug by size to be at 1. in sops do a vdb sdf of the collision geometry, which youll point to on the static object. Also instead of using dop level substeps, on the flip solver slap max substeps up to like 16 and put the cfl condition at about 5. Unfortunately due to the friction, bounce, and seperation all working together, it'll need to resolve a lot. the cfl condition will do more steps faster and more efficiently than the dop level substeps.

[7ofDiamonds]

I think that is just normal operation. The simulation looks ok to me on H15 after I reduced your Uniform Divisions down from 300 to the default of 30. The shoot offs are just the wave recoiling against the curved surface of the broken globe.

Hmm, if I turn down the uniform divisions lower than 300, I get particles leaking in H14.. However I was able to solve the shooting issue by turning velocity smoothing.

 

for the collisions, I'd use proxy volume, and plug by size to be at 1. in sops do a vdb sdf of the collision geometry, which youll point to on the static object. Also instead of using dop level substeps, on the flip solver slap max substeps up to like 16 and put the cfl condition at about 5. Unfortunately due to the friction, bounce, and seperation all working together, it'll need to resolve a lot. the cfl condition will do more steps faster and more efficiently than the dop level substeps.

Thank you so much for the tip on the substeps in FLIP instead of the DOP level. Things seems to be much faster! 

I'm not sure what you meant by "do a vdb sdf". Should I convert the geo to SDF? There are a couple of VDF SDF nodes and I'm not particularly sure which one I should be using. Would appreciate if you could explain this step a little bit further. 

 

In addition to that, how would I make sure my collision geometry is accurate since the surface is really thin? I'm not sure if making the geometry thicker would be ideal since the water needs too looks as if it flows down the sides.

Edited October 28, 2015 by 7ofDiamonds

[TheDude]

you can put down a 'vdb from polygons' node, and this defaults to create a sparse sdf volume. usually for collision geometry 1 million voxels is more than adequate. now in dops on the static object, the way you're sampling can be made much faster. Change the mode to volume sample, and division method by size. then in the proxy volume, just point to the vdb you created. if you slap the division size to 1 it defaults to the same resolution as the source. Now with your geometry you can choose how accurate you want the collision to be based on the voxel size on the vdbfrompolygons, so dont worry about altering any geo