Dear all,

Currently I'm working on goal-oriented error estimation by using the adjoint solution as weighting function. At this moment I am looking for a steady compressible Navier-Stokes and Euler code that incorporates the adjoint as well. Since such codes are not available yet (or institutes don't want to share...), I have to write it myself for which I intend to use OpenFOAM because of it's clear structure and being open source (I'm a fan of GPL anyway!!).

Now my question is: are there OpenFOAM users already working on an adjoint solver for flow problems (maybe for the purpose of geometry optimisation)? Or maybe you have useful tips about how to implement it the best way?
I am thinking about solving the analytic adjoint equations (instead of the more common 'transpose the system' way) which requires a new adjoint boundary condition implementation as well. In that case I have to add new source code to OpenFOAM and rewrite the main program for the particular adjoint equations, so it probably will be a hard job....
Hope to hear from you!

Kind regards, Jelmer Cnossen

[hjasak]

Yes, there is adjoint work going on. However, I am not prepared to share any data or code at this time.

Hrv

[braennstroem]

Hi Jasak,

I wonder, how the work on adjoint methods is going on. Would be nice, if there is something to share!?

Regards!
Fabian

[braennstroem]

Hi,

if there is nothing to share :-( does anyone have some good hints/links/papers about that topic and how complicated that would be to implement this into openfoam!?

Regards!
Fabian

[eugene]

AIAA-2007-3947

Very basic and has some fairly glaring inaccuracies, but its a start.

[braennstroem]

Hi Eugene,
thanks for the 'paper'! Is there a chance that the code will be implemented in the public openfoam release any day!?

Regards!
Fabian

[peita]

Hello,

I am student of mathematics in Prague. I am trying to use adjoint solver on external dynamics. I use the same solver as the one in AIAA-2007-3947 on flow over car (something that looks like beetle, more insect than car) with drag coefficient as objective function. But it has has really strange behaviour, after few hundreds iterations adjoint pressure, adjoint velocity components and phia have values over 10^60 and are still increasing. Don't anyone know what is wrong?



I have following boundary conditions:



40 m/s primal velocity on input,zero gradient on output, fixedValue 0 on car and ground, symetryPlane on the top

For primal pressure: zeroGradient on ground, inlet and car, fixedValue 0 on outlet and symmetryPlane on top.

For adjoint velocity: fixedValue 0 on inlet and outlet, zeroGradient on ground and symmetryPlane on top, on car I have unit vector with direction oposite to primal flow.

For adjoint pressure: zerogradient everywhere except inlet, here fixedValue 0.

Thank you very much for any hint!

Petr Simanek

[Martin80]

Hello,
is there an easy way to get AIAA 2007-3947 ?
b.r., martin.

[Anne Lincke]

Hey Peita,

are you still facing those problems?
Maybe your adjoint velocity outlet condition is not the right one.
Which cost function are you evaluating and where?

Anne

[Anne Lincke]

Hey Martin,
you can find the paper here

http://www.aiaa.org/agenda.cfm?lumeetingid=1379

Regards
Anne

[geeknick]

I have looked around assuming that it would be relatively easy these days to find a nonogram online solver, can someone suggest a site please. Thanks in anticipation.

[smrutimaya]

I found a nice description for modifying AdjointShapeOptimizationFoam. Its incorporating a cost function related to aero-acoustic noise. It may be helpful to some.

http://www.diva-portal.org/smash/get...FULLTEXT02.pdf

The title is : Numerical Method for Shape Optimization from Wind Noise Perspective