Physics Based -- What the Hell Does that Mean?

I know about a little physics -- enough to be dangerous. Enough to wonder what people mean when they say their sofware is "physics based."

I hear "physics-based" a lot at the GPU Technology Conference. For example, virtual dummies, roughly the shape of humans, are being flung through virtual wall, much to enjoyment of the assembled. The "simulation" is termed physics based. It is said in a way that insists that, yes, that is exactly how a body would crash, how fast it has to be flung before the wall starts to break up, how the pieces of the wall fly through the air, etc.

Oh, yeah? I have not studied realism in crash test dummies personally. I have witnessed the effect of a large calibre bullet bust up a thick steel plate. The shock wave from the impact of the bullet made a chunk of the plate break off on the opposite side and fly around the room -- an effect known as spalling. I pondered the equations that would model the shockwave. They were intense. The modeling of a dummy through a wall -- hmm... let me see. You could assume the dummy is an articulated collection of rigid body masses but what about the point of contact? The local deformations would be key in determining local failure of the wall at point of impact so rigid bodies need not apply. The wall itself was not simple, composed of vertical structural members and several layers. This was a far more complex problem than my bullet/plate problem. 

Now do you think NVIDIA was solving all those equations. I think not.  It's for fun and games. The results will not protect crews in armored vehicles, or protect office occupants from catapulted bodies. (Didn't they use to dead plague victims into besieged cities?). We see more "physics" applied to splashing water. I think modeling fire and explosions is absorbing our brightest minds -- for the movie industry. 

It would be nice if the industry adopted a level of physics based. At the lowest level, say level 1, would be if F=ma could be found any where in the code, even if it used for only one object. At the highest -- and possibly unattainable theoretical limit -- the magical physics based level 10, where every molecule in the picture behaves according to known laws, with each contributing to the motion and interaction of the bodies of which they are part. 

Comments

[Joe Aggie]

Physics-Based simulation is a term used by Computer Graphics folks and IEEE junta. It has nothing to do with FEA ! These computer geeks in Stanford coined this term to attract more money from venture capitalists (just like they cooked up most other phrases in computer/software/internet technolofy today).
Ok, what is Physics-based simulation -- hmmm, don't know. But I know it is totally different from FEA :)

[Deelip Menezes]

Mark: "All FEA simulation results are wrong. Its just a question of how wrong."

Suddenly I don't feel like flying in a plane again. Ah! But then thats the reason why we have this wonderful thing called 'Factor Of Safefy'. Love it.

[twitter.com/burhop]

Greg makes a good point.

All FEA simulation results are wrong. Its just a question of how wrong.

[Greg]

*All* simulations involve many assumptions, abstractions, simplifications - whether it be in geometry, material properties/ behavior, interactions (contact etc), or simply just the inputs/boundary conditions. So whether you solve F=ma with rigid bodies or something a little bit more sophisticated, the appropriateness of the simulation is always to do with the context in which it belongs.

This is something that FEA codes today do not really try to address - there remain lots of opportunities (and such a long way to go) to deliver on the 20+ year promise of "Designer FEA".

I've seen some nifty youtube videos lately showing simplification of geometry for use in FEA, and somewhat time-compressed solutions to these problems. The fact remains that the user *still* needs to know:
- why the model needs simplification
- what features are causing the "problems"
- why the new/simplified model is still an appropriate representation of "reality"
- why the use of 1 or two linear tets through the thickness of a rib in bending is BAD (saw that faux pas in a recent youtube video)
- once the FEA tool has given me output data (because that's all it does provide), what analysis must I do to be able to make decisions on it...

So I guess my conclusion is: even after you're done wrestling with your explicit dynamic code in order to see the effects of a dummy thrown against a wall, how much better is the result than the one computed in your playstation? Can you prove it? Will it change the decisions you make on your design of the wall?

[twitter.com/burhop]

I do think its interesting to watch the advancement here.

Virtual World simulation - I hate to say "game" since the scope has gone beyond that - favor speed over accuracy so there is a difference like you point out. Dumbed down geometry is often used. Still, math is math. We just need faster CPU's.

A good URL for more info is the Second Life Wiki which talks about their physics engine http://wiki.secondlife.com/wiki/Physics_engine

Heres a good Youtube video too:

http://www.youtube.com/watch?v=mXmgJvbvtbM

I'm looking forward to someday going beyond a "drop test" to things like "thrown across the room" test, and "drive it like you stole it" test.