FLIP mesh problems, particle fluid surface grief

[exel]

Hi all, hoping someone can help with this --

I've got a beach break sim, waves crashing on the rocky shore, etc.  As the wave crests and curls over to crash, there's problems with the interior surfaces.

While trying to eliminate variables and track down the issue, I've been working with a smaller section of the sim and turned off all the filtering on the mesh so it looks a bit crunchy right now, but hopefully this shows things clearly enough:

(front view)

(back view):

--There it is, frames 98-99, a big part of the mesh just disappears, *pop*!  This was done with the particle fluid surface SOP by the way, Houdini 15.5.565 .  I've opened it up and have been digging around to try and diagnose the problem and so far all I can do is cause that shape to disappear on a different frame, sometimes it reappears for a few frames.

   So far I think it has something to do with how the surface layer of particles gets combined with the source @surface field out of DOPs.  I get the same problem with or without using the Fluid Compress node.  I've been all over the place with the erosion scale, droplet size, etc., but nothing is improving things

If I look at the particles coming out of DOPs, I can see that the interior detail is there, nothing's popping from the sim:

, and if I view the surface field coming from DOPs, that looks more like what I would expect:

 

Anyway, I'm running out of ideas on how to fix this.  I'd really like to take advantage of the fluid compression and the features that Particle Fluid Surface offers, but the "unioning" of the VDB from particles and the compressed fluid SDF isn't working so well.  If I give up and just use the surface field straight from DOPS, I lose a lot of control over the detail and breakup of the splash areas.

Any advice or tips are most welcome!

Houdini 5.5.565 / Mac OS X 10.11.1

 

 

 

[exel]

...for what it's worth I managed to improve it a bit after changing a lot of settings:

--cranked up the particle bandwidth on the fluid compress node (caches now take up twice the disk space).  The particles are still being packed but I think it's keeping pretty much all of them now, no culling going any more

--cranked the erosion scale high enough until the particle fluid surface's internal "union" operation stopped wiping out the big shapes

--turned off "rebuild SDF" on the VDB From Particle Fluid node -- for some reason this was key, none of the results were working until I changed this...

Some of the smaller shapes still pop on and off, but I might be able to live with this, I'm hoping to generate enough bubbles and foam in there to hide the remaining crappiness...

 

[michael]

JASON?? is that you?

Jeff said he suspects it's NANs

can you send this test scene in as a bug?

[exel]

4 hours ago, michael said:

JASON?? is that you?

Jeff said he suspects it's NANs

can you send this test scene in as a bug?

It is indeed myself Mr. G, what's up man!  In an attempt to solve one problem, I caused others -- blocky voxel shapes knocked out of the mesh on certain frames (attached).  Something in the way it's resampling the compressed surface to match the VDB from particle fluid is causing Tetris-shaped holes in my mesh.  I think I managed to fix that as well, sadly by removing the compressed fluid surface from the data and making the particle fluid surface node use only the particles.  I think I'm losing most of the optimization and savings from the Fluid Compress node but the mesh is working a bit better.

   The scene I'm working with is kind of a beast, not particularly portable but if I can reproduce this in a more basic setup I'll send it in, thanks!

 

[crozer]

hey @exel, i only want to mention that using a spherical surfacing method in the particle fluid surface SOP can yield better results, especially if the mesh is abruptly going from thick to non-existent (eg. flickering as holes appear). this plus a low voxel scale and to compensate some smoothing.

i'm using houdini 16.0.504.20 by the way.
hope it helps!