SimScale CAE Forum

CFL Number for AMI simulation


#1

Hello All,

I have been trying to simulate an oscillating airfoil using AMI. In the transient case, the CFL number exceeds the limits. I tried using different sized grids and tried slower velocities and smaller time steps. And the only time it actually ran to completion was for very small time steps, not suitable for my intended use.

I was wondering if there was any other way to control the CFL number and also does the angular velocity of the AMI region affect the CFL number?

Thanks
Vishal


#2

Project Link https://www.simscale.com/projects/vt_vijay/amitest_1/


#3

Hi Vishal,

let me tag my colleagues @Anware, @Get_Barried as well as @vgon_alves here who might help you out with this issue.

Cheers!

Jousef


#4

Hi @vt_vijay

How long do you intend to run this transient simulation for?

Just a side note and if I remember correctly, the CFL number is like an information factor in a sense. Your results cannot transgress the “speed of information” if not you will run into possible continuity problems. If your cell size is very small and your inlet speed very slow, then in order to keep the CFL to less than 1 you will need very small time steps to allow the information source (the fluid that is flowing) to progress through the entire computational domain.

I believe I saw some simulations allowing a CFL of more than 1 in order to use bigger timesteps. You can take a look at it here. Other than that and to my knowledge, there isn’t any other way to really control the CFL number and it always has to be a fine balance between computational cost in terms of the timestep and the mesh fineness that will influence accuracy.

Yes I believe so as it directly affects how the mesh is “computed” between the rotating and stationary interface. You can find documentation about it here. I would think that with an increase of angular velocity, while the computational domain of the cell remains the same, the velocity between the mesh increases. Hence, the larger the velocity the faster information “moves” from one cell to the next and vice versa. From this I would expect the timestep required to keep CFL less than 1 to be smaller the slower the angular velocity and also vice versa.

Again, my lack of experience with transient simulations and even more rare rotating transient simulations may result in my above statements being inaccurate, so do read through the documentation and double check on the possible conjectures that I have arrived at.

Hope this helps!

Cheers.

Regards,
Barry


#5

Hi @vt_vijay, The above points by @Get_Barried are very important. That said, I believe the issue here is divergence. A diverged simulation will cause the error because it might have a really high predicted velocity and therefore throw the CFL warning which can be misleading. When running AMI simulations the interface between the region and the AMI region is from interpolation, and if the quality of the interface is poor then the simulation diverges.

I would try increasing the mesh refinement of the interface (looking at your project I would say by at least two levels) and see if this issue is mitigated.

Please keep us posted and update if this does or doesn’t solve your problem.
Best,
Darren