[alpharays]

Hi all,

Anyone have experience using adjoint solver?
I'm having problem to understand how to generate new design point positions after "Compute Mesh Sensivity".

Mentioned in tutorial "Alternatively, you can export the new Sensitivity w.r.t. designPointPositions table to a file and then import it, along with the designPointPositions_1.csv file (original design points) into the external optimization tool of your choice." Then using an optimization tool of our choice to compute new design point positions.

Anyone, please give me advice.

Thank you in advance
Regards

[meb]

If you can export sensitivity data at surface, you can sculpt the surface using a morpher. You have to take care because usually adjoint data are noisy and need to be filtered.
This approach is very well established in the platform of RBF4AERO (www.rbf4aero.eu). The Morpher Tool is the Stand Alone version of RBF Morph (www.rbf-morph.com). At the moment we haven't examples about STAR-CCM+ adjoint but we have all the tools to implement this.

[me3840]

I'm not sure if there's a reason to use an external morphing utility, STAR-CCM+ has a morpher built in. But the question is once you have the adjoint sensitives, how do you choose a displacement for each point? That's the role of the optimization code. You can make a very simple one which is algebraic just using field functions if you wish, but anything more complex would require another code to figure out how to modify different positions.

[meb]

We have a very well established implementation within ANSYS Fluent, that now has been extended to OF as well.
Radial Basis Functions mesh morphing allows to benefit of adjoint information in complex and challenging conditions. For instance you can reshape a wall of pipe whilst preserving a detail that is immersed in the flow (say a valve).
I'm not an user of Star and I'm not aware of the flexibility offered by the in house morpher.
To fully answer to the technical detail "How To" I can details the OF approach that could be seamlessly extended to OF.

You have to options:
(a) adjoint sculpting
(b) shape parameters steepest descent gradient method

(a)

(1) After a full CFD run and a full adjoint convergence the user exports on a file the sensitivity maps on surfaces to be reshaped.
(2) Sensitivity information can be given as a point and a vector (i.e. a displacement) or as a face and a value (the normal has to be retrieved in this case).
(3) Aforementioned field is then filtered and normalised (usually raw data are noisy). We use a technique based on a decimation of the cloud containing sensitivity and then a regression to match the whole field.
(4) An RBF problem is defined using the filtered data as centres. Fixed points are added on surfaces to be preserved (usually a deformable buffer of uncontrolled nodes is left to accommodate transition between adjoint controlled faces and fixed ones). Encaps can be added to limit the morphing actions.
(5) A batch command allows to update the full CFD volume mesh according to the RBF (the intensity can be adjusted).
(6) The workflow can be automated using a new acceptable shape as the starting point of a new step. Usually the step is decided by defining the maximum deformation at each step.

(b)
The adjoint data exported are projected onto a basis of predefined shape parameters (defined in advance using the morpher). In this case the workflow is.
(1) Compute the derivatives vs each shape parameters
(2) Build the gradient (or feed a gradient based advanced method using derivatives)
(3) Set the amount of shape variation to be allowed
(4) Morph the mesh, recompute CFD and adjoint
(5) Iterate

I can add references to relevant paper.

[alpharays]

Originally Posted by meb

We have a very well established implementation within ANSYS Fluent, that now has been extended to OF as well.
Radial Basis Functions mesh morphing allows to benefit of adjoint information in complex and challenging conditions. For instance you can reshape a wall of pipe whilst preserving a detail that is immersed in the flow (say a valve).
I'm not an user of Star and I'm not aware of the flexibility offered by the in house morpher.
To fully answer to the technical detail "How To" I can details the OF approach that could be seamlessly extended to OF.

You have to options:
(a) adjoint sculpting
(b) shape parameters steepest descent gradient method

(a)

(1) After a full CFD run and a full adjoint convergence the user exports on a file the sensitivity maps on surfaces to be reshaped.
(2) Sensitivity information can be given as a point and a vector (i.e. a displacement) or as a face and a value (the normal has to be retrieved in this case).
(3) Aforementioned field is then filtered and normalised (usually raw data are noisy). We use a technique based on a decimation of the cloud containing sensitivity and then a regression to match the whole field.
(4) An RBF problem is defined using the filtered data as centres. Fixed points are added on surfaces to be preserved (usually a deformable buffer of uncontrolled nodes is left to accommodate transition between adjoint controlled faces and fixed ones). Encaps can be added to limit the morphing actions.
(5) A batch command allows to update the full CFD volume mesh according to the RBF (the intensity can be adjusted).
(6) The workflow can be automated using a new acceptable shape as the starting point of a new step. Usually the step is decided by defining the maximum deformation at each step.

(b)
The adjoint data exported are projected onto a basis of predefined shape parameters (defined in advance using the morpher). In this case the workflow is.
(1) Compute the derivatives vs each shape parameters
(2) Build the gradient (or feed a gradient based advanced method using derivatives)
(3) Set the amount of shape variation to be allowed
(4) Morph the mesh, recompute CFD and adjoint
(5) Iterate

I can add references to relevant paper.

Yes, can you give relevant paper for my reference?

[meb]

Originally Posted by alpharays

Yes, can you give relevant paper for my reference?

Sure.
You can have a look to our RBF4AERO project web page where papers are available for download.
http://www.rbf4aero.eu/

On RG you can have a look to:
https://www.researchgate.net/publica...N_APPLICATIONS
https://www.researchgate.net/publica...an_RBF_Morpher

[alpharays]

Originally Posted by meb

Sure.
You can have a look to our RBF4AERO project web page where papers are available for download.
http://www.rbf4aero.eu/

On RG you can have a look to:
https://www.researchgate.net/publica...N_APPLICATIONS
https://www.researchgate.net/publica...an_RBF_Morpher

Thank you for your prompt reply. Will have take a look at papers first.