Time step too small with maxCo=1
 

Dear all,
I'm running unsteady simulation of vertical axis wind turbine with sliding meshes and using the pimpleFoam solver with OF 2.1.

I'm trying to reproduce a simulation from a paper, using the same conditions (mesh size, inlet velocity, ...).

My problem is that when I try to refine the mesh in the boundary layer, the time step decreases like hell (down to 1e-9 when I'm solving the boundary layer to y+=1), and the simulation time prediction is many centuries... This seems normal because I'm fixing the Courant number to one. However, from what I've seen in the literature, such simulations can be achieved much faster with the same boundary layer resolution with other solvers such as fluent for instance.

So the question is: is this small time step OpenFOAM specific ? Does anyone know why such simulations can be achieved with fluent for example ? I'm stuck because of this, and forced not to resolve the boundary layer and to use a wall function which impacts the results...

Thanks in advance for your answers.

Regards,
Daniel

Hi Daniel,

Have a look at the links below and start playing around with maxCo with different settings for nOuterCorrectors and tolerance levels and different values for the relaxation factors. Just make sure you do get the "Pimple loop converged in XX iterations" comment.

http://www.cfd-online.com/Forums/blo...hm-part-i.html
http://www.cfd-online.com/Forums/blo...m-part-ii.html

Regards,
Tom

Hi Tom,
thanks for the answer. I've already looked at this awesome tutorial, but it didn't solve my problem.

• If I use PISO, I keep maxCo to 1 which takes too long because of the small time steps.
• If I use SIMPLE, I keep maxCo to 5, what increases the time step. This is good, but the inner iterations take more time until the 'pimple loop converges'. And if I wand the convergence to be reached faster I need to increase the residual tolerance and therefore decrease the accuracy. So in the end, it doesn't change a lot. And I fear that if I increase the Courant number too much, it will not converge. Do you know what's the limit ? From what I've read, it can be dangerous to set Co>5. Are there any other tricks to get a reasonable simulation time ?

Thanks !


Regards,

Daniel

Hi Daniel,

It is pretty much case-specific on how high you can go. I have ran simulations with maxCo at 250 and still got accurate results, but on other projects I had to go back to somewhere around 10. The best way is running some tests to find the cut-off. You have some parameters to play around with (relaxation factors, tolerance levels, number of correctors/nonOrthogonalcorrerctors, etc.). My advice would be to just take a simplified version of your problem and do a sensitivity study, I expect that you will learn quite a lot.

Regards,
Tom

Hi Tom,

that sounds like a good idea. I didn't know so high courant number could be used, depending on the case. I'll try to play around with the different parameters and see what's the maximum Co.

Thanks a lot,
Daniel

Hi,

I've juste found out that using maxCo=300 still gives me accurate results ! I've reduced my simulation time from 2 weeks to 8 hours !

I thought the long simulation times were OpenFOAM specific, but the solution is just to find the right parameters. That's great news, because I can stick with OpenFOAM for the future !

Regards,
Daniel