3D design optimization in a rotating frame?
 

Hi there,

I'm trying to perform an optimization case in SU2 using a relatively simple 3D mesh in a rotating frame. I'm using a Caradonna & Tung rotor and I've based my cfg file on the cfg file provided in the tutorials for the optimization case of the NACA0012 (Euler). First I'd like to get it working for the Euler equations before moving on, in order to save time. When running shape_optimization.py the flow solution converges relatively quickly (as it should), but afterwards the adjoint solution is giving me trouble. I've tried several things to fix this, like using a coarser/finer mesh, adjusting the reduction factor of the CFL number of the adjoint problem (CFL_REDUCTION_ADJFLOW), using or not using multigrid, lowering the rotational speed (default is 2500 rpm, meaning that the used MACH_MOTION=0.877) but nothing seems to work. Every time the residuals of the adjoint problem increase (sometimes slowly, sometimes quickly, depending on the configuration) and after a while an increasing number of non-physical points appear. My main question is: before I waste too much time trying to fix my case, has anyone yet succeeded in running an optimization case for a rotating frame in 3D? For 2D it works, a normal flow analysis with a rotating frame in 3D works and a normal optimization case in 3D with a rigid frame works, but I haven't found a successful combination of them yet.

Thanks in advance for your help, I'm an inexperienced CFD user, so any help is welcome!

Hi,

Yes, we have had some success running design in these scenarios, and you might be interested in the following references:

http://adl.stanford.edu/papers/AIAA-2012-3018.pdf
http://adl.stanford.edu/papers/AIAA-2013-2580.pdf

However, the adjoint can be notoriously difficult to converge, and in our experience, these problems often had issues near the leading edges of the blades (when solved in the rotating frame). Many of the ideas that you tried above are good, and we regularly also try limiting the absolute value of the adjoint density variable to avoid divergence. You can do this with the following option:

%
% Limit value for the adjoint variable
LIMIT_ADJFLOW= 1E6

You might print out the adjoint solution and identify where in the mesh you are having issues, and then set the maximum value to something reasonable to keep this region from diverging.

Hope this helps,
Tom

Thanks a lot for your quick and helpful reply, Tom! I'm definitely interested in the papers you referred to. I hope that I can find a solution soon, the largest residuals indeed occur at the leading edge (at the tip of the blade, to be exact).

Just out of curiosity: in the first paper you refer to you also use the Caradonna & Tung rotor and you've used a periodic BC there. Have you also used multigrid? So far I haven't been able to use multigrid in combination with a periodic BC, I always get an error when it tries to set the multigrid structure in that case.