views:

137

answers:

2

I have a program in SimMechanics that uses 6 derivative blocks (du/dt). It takes about 24 hours to do 10 secs of simulation. Is there any way to reduce the calculation time of the Simulink derivative blocks?

A: 

You don't say what your integration time step is. If it's on the order of milliseconds, and you're simulating a 10 sec total transient time, that means 10,000 time steps.

The stability limit of the time step is determined by the characteristics of the dynamic system you're simulating.

It's also affected by the integration scheme you're using. Explicit integration is well-known to have stability problems for larger time steps, so if you're using an Euler method of integration you'll be forced to use a small time step.

Maybe you can switch your integration scheme to an implicit method, 5th order Runge Kutta with error correction, or Burlich-Storer. See your documentation for details.

You've given no useful information about the physics of the system of interest, the size of the model, or your simulation choices, so all this is an educated guess on my part.

duffymo
A: 

Runge-Kutta methods (called ODE45 or ODE23 in Matlab dialect) are not always useful with mechanical problems, due to best performance with variable time slice setup. Move to fixed time setup and select the solver by evaluating the error order you can admit. Refer to both Matlab documentation (and some Numerical Analysis texts too, :-) ) for deeper detail.

Consider also if your problem needs some "stiff-enabled" technique of resolution. Huge constant terms could drive to instability your solver if not properly handled.

ZZambia
Stiff equations are exactly why I recommended staying away from explicit integration. That's where the stability problems come from.
duffymo