Finite element solver

[borisb2]

Testing out fracturing with finite element solver in H15. I find it hard to produce stable and good looking results. Simple tests may work fine but when it comes to simulating 400.000 tets the story is a little different. So far I tested various combinations with the following:

- concrete building made of several solid objects. Colliding rock transfers pintoanimation-attribute in sop-solver to dop-objects to release parts.

- dopnet substepping / FEM-solver substepping (fixed 1, 2 and 3, adaptive 1-2, 2-4, 2-6, 3-10)

- max linear solve 1024 / 2048 / 4000 / 10000

- collision passes 5 / 15 / 30 / 40

 

So far it NEVER looks nowhere near a stable simulation. Pieces that are colliding with the ground snap sometimes back or rotate weirdly, I also cant get rid of stretching tets here and there. If I would use more stubsteps, the sim-times would sky-rocket. The last setup (about 350.000 tets, embedded geo, 7 solid objects) simulates about 20-25 frames, then suddenly freezes in the sim, meaning no error, I could still escape, it just doesnt continue. (wtf?)

 

At MPC I'm used to work with kali (DMM-solver), so the idea would be to test out how good houdini can compete. But at this point I better not start comparing both. Houdini FEM just doesnt seem to be ready for hires-fracturing. And I do remember when testing the free DMM version in maya (despite all issues) it never had these solving issues.

Any more attributes/setups I could test out to get it more stable? Could sidefx maybe switch to implement the DMM-solver instead using the current ?

 

Edited May 30, 2016 by borisb2

[symek]

Interesting insight (minus redundant sarcasm), thank you! Isn't it a very specific scenario of simulating concrete (high stiffness generally gives solvers a headache) in a physical scale of buildings (lots' of mass). I wouldn't be surprised if SESI have't tested its solver for this narrow case and it can't handle t well. Did you try to scale your simulation down? You definitely should send the scene to SESI and ask for a solution at the support. 

IFAIK FEM solver was design with soft bodies in mind (like most people seem to use it), but it should handle any case of course. 

You can always consider of making sims in Maya (sarcasm back)

[Guest tar]

17 hours ago, borisb2 said:

Any more attributes/setups I could test out to get it more stable?

 

Email the file into support so they can test it. The more real-world examples the better they can build their tech with.