PDA

View Full Version : Undesireable Bullet Chains behavior



Chrusion
08-21-2014, 12:19 PM
I have an 8' dia. drum covered with 256 chains (5 links each) arranged in 16 rows of 16 chains. Each link is a round ring about 3" in diameter and 3/4" thick.

Each chain has a half-round link anchor on the drum surface. Bullet is working fine initially in dropping the chains onto the drum surface (they initially all point radially outward). BUT, when I slowly rotate the drum (8 rpm), the chains hanging on the downward moving side start to float and clump up in a loose knot as if gravity is not pulling down on them. The chains on the opposite side are pulled upward and lay on the drum surface without issue. The chains that have fallen into a clump on the top of the drum remain clumped until almost at the bottom... they don't fall and hang from their anchors as they pass the horizontal X plane. After a few hundred frames, many of these clumped up chains randomly fall off the drum for no reason... there is no aggressive motion of the top link within the anchor that is overwhelming the collision detection... they're not jittering, being hit, or bouncing around inside the anchor ring, they just "decide" to fall thru the anchor, even though I've doubled the calc frame rate from 200, to 400, to 800. I've turned all object bounciness to 0.1%, decreased/increased friction (as low as 10% up to 80%), raised/lowered linear/angular dampings (up to 5% and down to 0.1%), etc. Collision margin on the chains is has been from 0.1" to 0.02" and 0.125" for the drum and anchors.

Nothing is "unclumping" the chains during drum rotation. I've increased their mass to astronomical values (880 tons - solid) and down to real-world weights (1,600 kg / 1,280 links = 6 kg each - solid). Same for gravity... -9.8 to -20. When clumped, the rings maintain a 1/4 to 1/2" collision margin as each link looks like it's floating.

I'm using a single layer for all the chains (in Bullet as Parts type and Convex Pieces shape) and a layer for the drum collider, and a layer for the anchors, all parented to a rotator null.

Any pointers?

jwiede
08-21-2014, 01:11 PM
Really difficult to visualize some of what you're describing, screenshots might help. Any chance you can provide a scene package (full, or cut down to just the problem)?

Chrusion
08-21-2014, 01:57 PM
Yeah... sorry about that. I prepared this while you replied.

Scene is from scratch and object is layers pulled from the 50 layer object.

New scene is acting the same (clumping) but with more breakage.

Chrusion
08-22-2014, 06:35 AM
After another set of tweaks, I'm getting a little better result, but still, a majority of the chains are sticking to the drum (90% friction, 0.1% bounce, 5% damping) and individual links are sticking to each other (the clumping effect - 90% friction, 0.1% bounce, 3% damping, 5% glue) as they rotate around the bottom half. The links are not falling and hanging from themselves and their anchors. Some chains remain stuck to the drum even at the very bottom of rotation and going up the other side.

So frustrating.

Chrusion
08-22-2014, 11:22 AM
For giggles I upped the dampings from 3% to 30% on the chains and, giggle, they're behaving a lot better! Only a few are sticking to the drum and remain clumped up, but none are breaking off. The rest are falling/hanging much more realistically now, much less floating. No idea why those two params would do this... seems counter intuitive... more damping = more "gravitational" pull?

The scene above has the drum turning at half the normal rate. I upped it back to 8 rpm and still no break away chains and now getting a bit of centrifugal effect.

I wish Bullet was MULTI-THREADED! This waiting around for 45 mins for 600 frames to sim each time a tweak is made is why it's taken all week to get this thing behaving right. I mean, Bullet is cooking on the sim right now at a whopping 15% CPU load according to task mgr for the last 15 mins. Really? Why?

Chrusion
08-22-2014, 12:36 PM
Hmmm... the last param tweak looks much better. I can better see now that the clumping (very much reduced now) is being caused by what looks like magnetic attraction of the chain links to neighboring links and collision shapes. Floating has been eliminated, but a few chains are visibly being pulled together as they slide on the drum surface. Some links are "magnetically" attached to some of the collision paddles and some to the drum surface. Strange, strange, strange.

Any idea which param could possibly be responsible for this?