[Tommy Chen]

Hi Stanford guys

I have run a continuous adjoint computation of the RAE2822 airfoil in the test cases.
In the surface_adjoint.dat file and the surface_adjoint.csv file, I found that the surface_sensitivity variable is a one dimensional scalar instead of a 2D vector?

Why is the surface sensitivity of 2D airfoil scalar instead of vector ?

Many thanks

[Tommy Chen]

So is this 1D surface_sensitivity variable the absolute value of the 2D gradient vector?

If then , what is the variable that difines the direction of the gradient?

[fpalacios]

Originally Posted by Tommy Chen

So is this 1D surface_sensitivity variable the absolute value of the 2D gradient vector?

If then , what is the variable that difines the direction of the gradient?

Hi,
In this link your will find a good reference of the implementation https://adl.stanford.edu/papers/AIAA..._March2012.pdf

Anyway, the computed sensitivity is with respect to displacements in the normal direction. So it is always an scalar... e.g. how change the drag of the aircraft when you move a point on the surface in the normal direction to the surface.

Best,
Francisco

[Tommy Chen]

Originally Posted by fpalacios

Hi,
In this link your will find a good reference of the implementation https://adl.stanford.edu/papers/AIAA..._March2012.pdf

Anyway, the computed sensitivity is with respect to displacements in the normal direction. So it is always an scalar... e.g. how change the drag of the aircraft when you move a point on the surface in the normal direction to the surface.

Best,
Francisco

Thanks Francisco,

So if the swept angle of the wing or the span and chord length of the wing are also design variables,which leads to the change in the plan form, then the normal-to-surface sensitivity is still all that we need to calculate the gradient of the objective function ?

[fpalacios]

Originally Posted by Tommy Chen

Thanks Francisco,

So if the swept angle of the wing or the span and chord length of the wing are also design variables,which leads to the change in the plan form, then the normal-to-surface sensitivity is still all that we need to calculate the gradient of the objective function ?

Yes, it is going to work, and the gradient in the normal direction is everything that you need. The only thing that you should keep in mind is that the sensitivity on geometrical singularities is not right so... some small inaccuracies are expected if you move singular points (fix or round leading edges is a good idea to avoid that situation).

Best,
Francisco

[Tommy Chen]

Originally Posted by fpalacios

Yes, it is going to work, and the gradient in the normal direction is everything that you need. The only thing that you should keep in mind is that the sensitivity on geometrical singularities is not right so... some small inaccuracies are expected if you move singular points (fix or round leading edges is a good idea to avoid that situation).

Best,
Francisco

I get your point, many thanks.

So actually I can use all the normal-to-surface sensitivity and the normal vector (the absolute value of which is the surface element area) to project the gradient to any other design variables developed myself.