Bullet and scale

[haggi]

Hi,

I'm experiementing a bit with bullet and fracturing and did some simple tests. If I know it correctly then the simulation base unit in Houdini is meter, so 1 unit (e.g. a basic polybox) is 1 meter large. If I now try to simulate real geometry like a bottle or a cup of tea, then my geometry is maybe only 0.3 units large for a bottle. These are my results:

First, the real size, 0.3 units large, no substeps, convex collision geo:

bullet_01_convex.mov

As we can see, the collisions are quite inaccurate and the simulation is jittering a lot. That's simply because the collsion geo is inaccurate. So I switched to concave and tried again.

bullet_01.mov

That's more accurate but still jittering a lot. So I switched back to convex and modified the collision padding to get a better representation.

bullet_01_convex_collpad.mov

Collision is better, but we still have a lot of movement in the geometry. So maybe it helps to increase substeps? I increased them to 10 on the dop node.

bullet_01_convex_collpad_sub.mov

As you can see that was not such a good idea, the parts are jumping around like crazy. So Now I started with an increased scaling of the geometry and increasing the gravity at the same time twice, once with convex once with concave.

bullet_02_concave.mov

bullet_02_convex.mov

Now the convex sim starts to look better. Just for testing, I increased the scaling up to 3 and did it again:

bullet_03_concave.mov

bullet_03_convex.mov

Looks very much better. Now I tried with my real geometry:

bullet_04_concave.mov

still some jittering, and not enough bounce, so I increased the bounce.

bullet_04_convex_bounce.mov

And a bit more bouncing.

bullet_04_convex_bounceHi.mov

Now it looks more the way I want it. So my final conclusion is: Even if it may be physically correct to simulate a 30 cm bottle in real size, the only way to make it work as expected is to scale it up several times, simulate and scale it back down. Is this correct and can anyone confirm this? Maybe I'm completly wrong and simply missed some of the many settings in Houdini so I attached my scene. Here the uniform scaling of "transform6" node in the "alembic" network is connected to gravity to make it automatically higher if the geometry grows and bounce of groudplane and geo is higher than default. 

Scene file:

bottle.zip

bullet_01_convex_hull.mov

[Farmfield]

Yes, in the Cg Society destruction course, I think Pawel Grochola says 2x the scale for most sims, even doing something like a smaller building, if you have a lot of detail, at least.

And it seems to work similarly in the other direction, you get to a point where too large simulations might start to behave weirdly and you need to scale them down. In both cases, you might have to compensate, tweaking the timescale, to get it looking correct for it's intended scale.

That being said, I might be completely wrong and/or someone of the really smart people here will come in and explain it better.

[haggi]

You mean you manipulate the time instead of gravity? That's interesting, never tried this.

[carlo_c]

I've scaled down large scale sims to get better results and vice versa as said previously. Probably just comes down to what you think looks best and matches the reference you have! 

Edited August 22, 2016 by carlo_c

[Farmfield]

17 minutes ago, haggi said:

You mean you manipulate the time instead of gravity? That's interesting, never tried this.

Gravity only works in one direction and though you might want to play with that too, increasing it with lower time scales to compensate for things falling at max velocity falling "too slow" - but primarily I would primarily tweak the timescale - bigger things move "slower", smaller moves "faster", so to speak.

[haggi]

Ah, great. Thanks for the input.