RANS optimization of Onera M6 wing
 

Hi

I have been trying to do RANS optimization on Onera M6 wing
using SU2 v 2.0.4. I am using the same Mesh-FFD box definition as
used in the Euler optimization testcase of Onera wing.

But the SU2 flow solver is blowing up during second design cycle (first mesh deformation).

I am attaching the end of run.log file and the .cfg file with the post.

Please suggest how to run this case.

Thanks
Anant

Hi Anant,

My guess is that the spring analogy is not able to correctly deform the viscous mesh for this case, and thus, the solution immediately diverges upon the second design cycle (there may be very low quality cells left over from the deformation). Can you please visualize the deformed mesh that is output from SU2_MDC?

Fortunately, we are getting close to releasing a developer version of SU2 that has a new mesh deformation algorithm based on the linear elasticity equations which is much more robust, especially for viscous meshes with high aspect ratio cells.

All the best,
Tom

Hello

I have been trying to do RANS optimization on Onera M6 wing,
for quite sometime, but haven't been able to do it.

I am using the hybrid coarse mesh provided with the onera testcase.
The adjoint solver doesn't seem to converge.

The .cfg file is attached with the post.

Has anyone tried this case ?

Thanks
Anant

Hi Thomas

I tried RANS optimization for Onera wing case with SU2 2.0.5 version.
But at the start of simulation, it is showing some warnings:

WARNING: unrecognized option in the config. file: GRID_DEFORM_SOLVER.
WARNING: unrecognized option in the config. file: GRADIENT_METHOD.

Why is this coming ?

I have attached the .cfg file with the post.

Thanks
Anant

Hi

In continuation of my previous post regarding problems being faced while doing RANS optimization for Onera M6 wing,
I am attaching the FFD box definition (initial and after 1st deformation).

It can be observed that there is some issue while deforming the mesh. Is this due to incorrect gradients been calculated ?

Also, I want to know that if I want to reduce the step size of deformation, how can I do it ?
Should it be done to rectify the deform error ?

While start of the simulation, these warnings are coming:

WARNING: unrecognized option in the config. file: ONESHOT_CAUCHY_EPS.
WARNING: unrecognized option in the config. file: GRID_DEFORM_SOLVER.
WARNING: unrecognized option in the config. file: GRADIENT_METHOD.

I am also attaching the configuration file.

Thanks
Anant

Hi

I have been trying to do RANS shape optimization on Onera M6 wing
(hybrid coarse mesh) for quite some time, but have been unable to do so.

When I see the CD gradient of the RANS case after first design cycle,
it is observed that the gradients are very high ( almost 10 times that
obtained for euler optimization case).

This results in large mesh deformation and the flow solver blows up
during second design run.

I am attaching the .cfg file and CD gradient plot.

Can anyone tell how to address this issue ?
Can someone share the input files if he has tried RANS optimization
case on Onera M6 wing ?

Thanks
Anant

Hi Anant,

A couple questions...

1. Are you able to successfully perform a deformation of the mesh using SU2_MDC and the FEA method for an FFD control point? These types of viscous meshes with tight spacing and high aspect ratio cells can be difficult to deform.

2. Have you played with the scale factor for the objective function (and constraints) in the config file? This factor multiplies the cost function and gradient before giving it to the optimizer, and it is particularly important for the first step in the optimization process. If the gradient is not on the order of a meaningful dimensional perturbation of an FFD control point, the first step of the optimizer will cause the mesh deformation to fail due to too large of a step being taken.

Hope this helps!
Tom

Hi Thomas

I was able to run RANS optimization case on Onera M6 wing using
your suggestion (playing with the scale factor).

The simulation ran upto 45 iterations after which I stopped it.
However, the adjoint calculation folders, ADJOINT_DRAG and ADJOINT_LIFT
are not present in every design folder.
Some are having it and some do not.

Why is it so ?

Thanks
Anant

Hi Anant,

The 'DESIGNS' folder carries the information for every functional evaluation, but note that, depending on the optimization algorithm, the number of functional and gradient evaluations are not always equal. The optimizer will often compute multiple functional evaluations per gradient evaluation in order to find an appropriate step size. Therefore, some of the candidate designs did not also require a gradient evaluation (and therefore it's not found in the folder). At the end, the optimizer should print out the total number of functional and gradient evaluations that were required.

Cheers,
T