Fluid River

[Korhon]

Hi all!

Heres my atempt to create some kind of flooding effect. I`ve been trying to create largescale water sims latley and having some problems.

This is the same kind of setup that Miguel Perez did a while ago(in this post: http://forums.odforce.net/index.php?showtopic=6311&st=12 ). Levelset fluids with the cusp_values olt.

My biggest problem is to get enough details in the simulation. In this test i used 500 resolution, but still its not enough to get fine details in collisions.

I can use the cusp values to emitt particles, but i think i need more details in the mesh to get a realistic look.

http://physbam.stanford.edu/~fedkiw/papers...ford2001-04.pdf In this article they speak about some way of combining levelsetfluids and particle fluids. Is this something that could be done in dops without tooo much work?

Heres the test:

http://www.gimpville.no/files/gimpville/Wa...velSet_test.mov

simulationtime 16hours on a dual quadcore 2.33, used 3.8gb ram

[vmuriel]

Hehe, 500 resolutions unit ... great !

Edited May 29, 2008 by vmuriel

[cellchuk]

hey it look pretty nice m8.

[pclaes]

very interesting test! looks pretty cool too. Although I understand where you want to go. The container based fluids tend to give inaccuracies in the representation of the volume. (some of the fine detail is lost). Have you tried using particle fluids at all? I believe fedkiw uses a technique which uses a high number of particles driven by a container based fluid. Because he uses these particles, no volume is actually lost. (whereas in container based fluids you do lose detail and volume around the splashes).

In houdini you can use the velocity field of one fluid to affect the particles. I can't remember it of my heart, but there are some files here on oddforce dealing with having a fluid drive particles.

I posted one of my files on 3dbuzz: http://www.3dbuzz.com/vbforum/showthread.php?t=162944 based upon various shaders and implementations from other users here and the help files. Even though that one is to simulate something resembling fire, you might be able to use some of it for driving water.

Something I personally want to look into is harvesting some of the gpu power to help calculate particle based fluids. (check google for some results). Probably using python for the gpu - but I will find out more about this as I go.

Good luck with it, I'm curious to see your progress!

[Korhon]

Thanks for the good feedback!

Pclaes: I havent tried particlefluids much in houdini yet. Will try to simulate the same scene with particlefluids to see if its possible to get finer details.

But from my experience with realflow i know that to get large scale watersims you need insane amounts of particles. So i was kinda hoping i could use levelset fluids for my overall watershape, and then use particle fluids to "pimp my waves" where i need to

But I will have a look at your files! Looks very interesting to drive particles with a fluidscontainer.

And simulating with the gup looks awsome! http://www.youtube.com/watch?v=xMjgssWkRE0

Thanks!

[pclaes]

It does require a lot of particles, so I think your approach of the general volume with levelsets is a good one.

Another trick typically used would be to use your high level velocity field to drive several lower resolution particle fluid sims, then merge the different caches together. This generally works well and gives faster results than doing one huge sim trying to solve everything in one go. I did this for the simulation of a jam material for the fine details around the hands of my character ( http://forums.odforce.net/index.php?showtopic=7151 ).

In Houdini it can be tricky to get stable results from emitting new particles (particle fluids) from an object -normal particles should be fine though- (say from the high speed velocity cusping parts), because sudden velocity increases might push your system to emit a big number of particles at once, causing a big local increase in concentration of particles. This can cause the particles to explode as their inner forces become too high too quickly. On top of that you will have to deal with interpolation issues. (you might have seen those banding effects in particle emissions). I am sure you do not wish to go to the subframe level to solve the interpolation of your particles.

Subframe interpolation of the final mesh will most likely be impossible because you can't blend between meshes (different topologies). But you can probably find a way to blend between the velocity vector fields of two frames and drive those lower resolution particles fluids with it.

This guy has done some amazing work on the gpu type of fluids. It is contained, but the speed gains are just insane.

http://www.cs.caltech.edu/~keenan/project_fluid.html