Calculations with some "noise" [SOLVED]

[Auradrummer]

Hello guys.

I'm implementing a model where a board is supported by four coils, one in each corner. The result is very good, and the board behavior is rellistic. But some coils are flickering. I'm sure that this isn't limitation of the engine. There is some instruction that I have to set to make the calculations smoother?

[JernejL]

Coils? you mean springs?

[Auradrummer]

Yes, yes, springs. What is the difference between 'coil' and 'spring'? Here in Brazil we understanding them the same way.

[Dave Gravel]

Here in Canada a coil looking similar to a spring but it's really not a spring.
Here the coil is use in electronic stuff and it is static.
Only the way how it is make looking similar to a spring, but it can't move.
http://www.germes-online.com/direct/dbi ... d_Coil.jpg
http://media.digikey.com/photos/Epcos%2 ... 182A30.jpg
http://www.rocketroberts.com/techart/im ... e_coil.jpg
http://www.isplc2006.org/b2b/pics/Loadi ... eries_.jpg

[JernejL]

Yes, yes, springs. What is the difference between 'coil' and 'spring'? Here in Brazil we understanding them the same way.

Coil is like a lightbuilb coil, heating coil or an induction electronic element used in radios.
Spring is the ellastic mechanical steel part which is used in car wheels, etc..

Anyway what you need is a better spring functions, and there's quite a few, but i never made springs myself, i can lookup for some useful works when i come home.

[Auradrummer]

Ah, OK. Here those coils are called "spiral". I learnt one more!

[Auradrummer]

I discovered one thing:

My system, as I already said, is made of four springs and one table. The processing is made in the following sequence:
spring3; spring2; spring1; spring0; table ...

As this is an recursive operation (the result of one depends directly of the result of the other), I'm accumulating error till get to the table again. I think the right way to balance the system is calculate them intercalated, like:

spring3; table; spring2; table;spring1; table;spring0; table;

There is some natural way to make Newton do this? Like some instruction, timeStep (I don't know how to use it), thread (not know too)...

[Auradrummer]

I've noticed something else:

The spring that is more unstable is that have the more unstable timeStep. I think the problem could be here.

[Dave Gravel]

Is this only for a simulator with just some few objects ? or it's for a game with many objects ?
If it is only simulator with low object count you can try to use solver 0 and not solver 1.
Because solver 0 have more precision and it is surely more stable with this type of simulation.
Maybe a fixed timestep can help too,
About the thread count I don't think that have any special effects on the simulation in self maybe on the speed.

[Auradrummer]

It is for a game, with too much objects. In the fact, I'm trying to create a car, but studying the physics since the beginning. I'm convinced that I'll never be successful using already made functions if I don't understand the basics.

I didn't set any solver. How can I do this?

[Dave Gravel]

I talk about NewtonSetSolverModel.
0 egal more precision and 1 egal less but faster... you can use higher value too.

I don't remember by default if newton beta 2 use 0 or 1 I think it using 0.
The problem with a game the solver 0 can come hard to deal if you have a big object count.
The other problem if you use solver 1 it is faster but all joints lose more precision.

You need to find the best for your use.

[Auradrummer]

Nope, solver made no difference (fortunately I don't need to use higher solver, the problem is in another area)

I'm plotting graphics of the flickering value and analyzing it. The amplitude and frequence of the flicker do not changes.

I'm studying the codes of Newton Demos. Your car works fine in the demos but fail if I try to implement. So, the theory that this is some setting I'm missing is getting stronger.

all the samples, as I saw, have timeStep as follows:

Code: Select all

CHiResTimer time;
timeStep = time.GetElapsedSeconds();
NewtonUpdate (nWorld, 1/60)

I printed the 'timeStep' variable, and it is 0,17 or (1/60).

I can use NewtonUpdate (nWorld, timeStep) ?

I hope I was clear (I'm very tired right now)

[Auradrummer]

The mistake here is that I'm calculating the spring forces and then applying them to the table in different functions. I concentrate everything in only one function and I got a very better result.