[Soc-2017-dev] Weekly Report #11 - Mantaflow Particles and Fluid Guiding

[Nils Thuerey]

Hi Sebastian,

regarding the "velocity bug": without setting the velocities, the updates from gravity & pressure solve accumulate over time, and esp. gravity can lead to huge (and unphysical) increases in velocity. 

For thin velocities I'd recommend to re-extrapolate the SDF after offsetting the initial distance field. We could discuss during one of our meetings...

Otherwise focusing on the guiding sounds like a good plan. What are your next steps for the implementation there?

Cheers,
-> Nils 


=== Technical University of Munich , http://ge.in.tum.de/ ===





> On 12. Aug 2017, at 01:55, Sebastián Barschkis <sebbas at sebbas.org> wrote:
> 
> Hello everyone,
> 
> my progress report for week 11 is now online:
> 
> https://wiki.blender.org/index.php/User:Sebbas/GSoC_2017/Reports/Week_11 <https://wiki.blender.org/index.php/User:Sebbas/GSoC_2017/Reports/Week_11>
> 
> = Week 11 =
> 
> This week I worked on some of the bugs and on fluid guiding. 
> 
> === What's been done ===
> 
> * I looked a bit further into the (still unresolved) velocity bug. As Nils pointed out last week, setting velocities in inflow regions explicitly seems to matter. I found this made sense and so I made a test to prove this point: First I added an inflow  velocity setup to the Mantaflow script that reproduces the bug. Then I compared the outcome with the old setup that did not use inflow velocities.
> 
> * My first impression from the previews is that manually setting inflow velocities definitely improves stability. Setups with open domains now look much alike to setups with closed domains (as it should be). I tested a range of inflow velocities and at no time did I encounter scenes where velocities exploded / "behaved unphysically" (as it happens when there are no inflow velocities set explicitly).
> 
> * So there is a chance that this resolves the issue. For the final verdict though, I'd like to know why setting velocities matters. Right now, I think velocity extrapolation plays a role here too: If no velocities get set in inflow regions, the extrapolation builds up velocities in that region at every step, up to a point where the simulation becomes unstable. 
> 
> * There was a problem with thin obstacles. In the levelset generating function thin obstacle meshes and planes were not considered. You'd always have to make obstacles sufficiently thick with respect to domain size to get an obstacle effect. I created a new setup which fixes this by searching for the nearest points on the mesh surface in addition to finding distances to the mesh surface via raycasts.
> 
> * For fluid guiding I have nothing new or tangible to show yet. I think I have landed a good setup though. More on that next week.
> 
> === Next up ===
> 
> * Continue with last weeks goals (especially particle refactoring and cleanup).
> 
> * No matter what the cost, fluid guiding needs to be finished by next week. (Not trying to sound overly dramatic here. It's just that the week thereafter it's already time to write the documentation).
> 
> === Questions ===
> 
> * No
> 
> Best wishes,
> Sebastián
> <ATT00001.c>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/soc-2017-dev/attachments/20170815/157dffb8/attachment-0001.htm