[JanH]

Hi all,

Within my thesis for the TU-Delft I use SU2 for gaining sensitivities of lift and drag of a wing w.r.t. the FFD control points. When I compare the (converged) sensitivities with Finite Difference its a match everywhere, expect for the last three rows of FFD control points near the trailing edge. See the attachment, where control point 0 is the upper control point at mid-span at the leading edge, control point 11 is the upper control point at mid-span at the trailing edge. The mesh and ffd-box definition I used is also attached.

When I look at the senstivity comparison done in the release paper of SU2 (paper AIAA 2013-0287, page 51) the control points near the trailing edge are not included in the comparison.

Is this a known problem? Is there a fix?
I am using v3.2.8 with Euler and have not tried v4.0 nor RANS yet, but are the sensitivities near the trailing edge improved for that version or governing equations?

I already tried refining the mesh, a mesh with a blunt trailing edge and changing the numerical method and order for the adjoint solution, but without success.

For my thesiswork I am deforming whole airfoil sections at once, therefore I cannot simply exclude the TE from my design variables neither can i make the surface sensitivities 0 at the trailing edge (with the sharp edges option in the config file).

I hope someone can help me with this problem,

Thank you in advance,

JanH

[hlk]

Originally Posted by JanH

Hi all,

Within my thesis for the TU-Delft I use SU2 for gaining sensitivities of lift and drag of a wing w.r.t. the FFD control points. When I compare the (converged) sensitivities with Finite Difference its a match everywhere, expect for the last three rows of FFD control points near the trailing edge. See the attachment, where control point 0 is the upper control point at mid-span at the leading edge, control point 11 is the upper control point at mid-span at the trailing edge. The mesh and ffd-box definition I used is also attached.

When I look at the senstivity comparison done in the release paper of SU2 (paper AIAA 2013-0287, page 51) the control points near the trailing edge are not included in the comparison.

Is this a known problem? Is there a fix?
I am using v3.2.8 with Euler and have not tried v4.0 nor RANS yet, but are the sensitivities near the trailing edge improved for that version or governing equations?

I already tried refining the mesh, a mesh with a blunt trailing edge and changing the numerical method and order for the adjoint solution, but without success.

For my thesiswork I am deforming whole airfoil sections at once, therefore I cannot simply exclude the TE from my design variables neither can i make the surface sensitivities 0 at the trailing edge (with the sharp edges option in the config file).

I hope someone can help me with this problem,

Thank you in advance,

JanH

Continuous adjoint sensitivities near sharp edges are known to be inaccurate due to the assumption of a smooth surface during the derivation of the equations.
The current method of addressing this is to identify sharp edges and filter out the sensitivities near them. This reduces the effect that sharp edges have on the sensitivities near sharp edges.

The settings to control this filter are:

SENS_REMOVE_SHARP (set to true to use the sharp edge filter)
REF_SHARP_EDGES (controls coefficient used to determine the location of sharp edges)
SHARP_EDGES_COEFF (controls how far away from a sharp edge the sensitivities are removed (this setting is also used in a limiter, but the two methods generally aren't used at the same time))

[CarlosLozano]

You may want to take a look at this paper:

https://www.sciencedirect.com/scienc...45793011003641

where that precise problem is examined.