[pdp.aero]

Hi folks.


I just wanna confirm that in the current version 7.1.1 the periodic boundaries are frozen and mesh on those cannot be deformed? right?
If it is frozen, I wanna open an issue.


Have seen an issue but why they cannot be deformed?


I have a test case with transitional periodic boundaries that needs mesh deformation on those faces because ffd box passes through them. By having them frozen, I get negative cell volume on adjacent cells.


Best,
Pay

[pcg]

In general the FFD deformation would not respect the periodicity, hence the freezing of such boundaries, can you make a smaller FFD?

[pdp.aero]

Originally Posted by pcg

In general the FFD deformation would not respect the periodicity, hence the freezing of such boundaries, can you make a smaller FFD?

Hi Pedro,


Gotcha, but if a pair of periodic boundary deform exactly the same, periodicity maintains.


I assume symmetry boundaries are also have frozen? right?


I can reduce length of the FFB box in lateral direction but it is not a right thing to do because the it will have intersection with the wing and makes those intersected edges to be frozen. Consider a wing section between to parallel domains. The flow could be transnational periodic in span direction and those two parallel domains cutting the wing have two edges that needs to be deformed with the rest of the wing but both domains also need to be kept periodic.


Is there somewhere I can read more about config options in details?


I see this in the config file:


% MESH DEFORMATION PARAMETERS (NEW SOLVER) %
%
% Use the reformatted pseudo-elastic solver for grid deformation
DEFORM_MESH= YES
%
% Moving markers which deform the mesh
MARKER_DEFORM_MESH = ( airfoil )
MARKER_DEFORM_MESH_SYM_PLANE = ( wall )


does this "MARKER_DEFORM_MESH_SYM_PLANE" deforms the symmetric planes? or does this deform sym planes?


Then there is this:


% Definition of the FFD planes to be frozen in the FFD (x,y,z). ---> coordinates?
% Value from 0 FFD degree in that direction. Pick a value larger than degree if you don't want to fix any plane.
FFD_FIX_I= (0,2,3) ---> i, j, k indices?
FFD_FIX_J= (0,2,3)
FFD_FIX_K= (0,2,3)
%


% There is a symmetry plane (j=0) for all the FFD boxes (YES, NO)
FFD_SYMMETRY_PLANE= NO ---> how does this affect deformation?




Thanks a lot!



Best,
Pay

[pcg]

Hi Pay,

Symmetry boundaries are allowed to "slide".

The *DEFORM_MESH* options do not affect the mesh deformation that SU2_DEF uses (yet, folks are working on that).

The FFD_FIX_I/J/K options are active when FFD_CONTINUITY is set to USER_DEFINED (which is what I would recommend).
They are lists of I or J or K planes that you want to fix (not triplets of coordinates as the example could suggest), I usually leave them empty.

FFD_SYMMETRY_PLANE overrides the internal logic to detect intersection, this only matters if FFD_CONTINUITY is set to 1st or 2nd derivative (I never used it though).

There is no good source of documentation for those options I'm afraid... (I had to open the code and go look for what that last one does).

I understand that with careful setup of the FFD and optimization you can make it preserve the periodicity. However, this does not guarantee that "free" boundaries will remain periodic after deformation, to do so we would need a proper symmetry BC for the elasticity problem.
The shape_optimization script cannot do any of that careful manipulation anyway and so we opted for freezing those boundaries.

If you want to free those boundaries you can modify the file CVolumetricMovement.cpp, in function CVolumetricMovement::SetBoundaryDisplacements and in the first "for" loop in that function add an exclusion for PERIODIC_BOUNDARY (similarly to what it is done in that loop for symmetry and internal boundaries).

Regards,
Pedro

[pcg]

You may also get way with defining those surfaces as symmetry planes when you call SU2_DEF but not when you use the code.
Or you can try an FFD box that includes those surfaces and setting them also as design markers. If you guarantee the periodicity of the FFD this might be ok.
This is all untested of course.

[pdp.aero]

Originally Posted by pcg

You may also get way with defining those surfaces as symmetry planes when you call SU2_DEF but not when you use the code.
Or you can try an FFD box that includes those surfaces and setting them also as design markers. If you guarantee the periodicity of the FFD this might be ok.
This is all untested of course.

Thanks a lot Pedro for helping me out.


I understand your point on free form deformation of periodic boundaries. Well, they shouldn't be free free but together free.


I am not using the python shape optimization script. I use FADO and I have already written a glyph (i.e. Pointwise's script) that reads SU2's mesh and remesh the deformed surface through FADO during the optimization. The mesh sure is gonna be periodic. The deformation is also will remain periodic because flow is periodic for this case and points deform in only z direction.


I have read the issue on github on this deformation of periodic boundaries. I dont't wanna change something in SU2 for this particular case unless I have to cuz results must be reproduced later by some others based on the official SU2 release. Looking for a work around.


I do appreciate your comments. I will give them a try and probably will write back here how they turn out.


Have already set it up with FFD_SYMMETRY_PLANE and 1st DERIVATIVES CONTINUITY; though results weren't smooth at j=0 symmetry plane.



Best regards,
Pay