CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Software User Forums > SU2 > SU2 Shape Design

Drag Coefficient Gradient: 0 in 2d adjoint optimization

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   March 24, 2017, 13:11
Default Drag Coefficient Gradient: 0 in 2d adjoint optimization
  #1
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
hi,
I'm running a shape optimization for a 2D drag minimization problem on a simple supersonic aerofoil. The flow and adjoint solutions converge reasonably, but for some reason the gradient at each design variable is zero.
I have the complete FFD definition in both the cfg and SU2 grid files (attached).
anyone have any ideas as to where i might be going wrong? i've tried editing the FFD and DV definitions but nothing seems to help. i'm using V5.0.0 but as far as i can tell the FFD definition is correct for that version.
any ideas really appreciated.
Tom.
CFG file (FFD AND DV DEFINITION STUFF ONLY):
Code:
DV_KIND= FFD_CONTROL_POINT_2D 
%
DV_MARKER= ( UPPER, LOWER )
%
DV_PARAM= ( FFD_FOIL, 4, 2, 0.01, 0.01)
%
DV_VALUE= 0.0
%
FFD_TOLERANCE= 1E-6
%
FFD_ITERATIONS= 100
%
FFD_DEFINITION= (FOIL, 0.0, 0.1, 0.0, 1, 0.1, 0.0, 1, -0.1, 0.0, 0, -0.1, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0)
%
FFD_DEGREE= (4, 2, 0)
%
FFD_CONTINUITY= 1ST_DERIVATIVE
%
OPT_OBJECTIVE= DRAG * 0.001
%
OPT_CONSTRAINT= ( MAX_THICKNESS > 0.06 ) * 0.001; ( LIFT = 0.0772 ) * 0.001
%
OPT_ITERATIONS= 100
%
OPT_ACCURACY= 1E-4
%
OPT_BOUND_UPPER= 0.1
%
OPT_BOUND_LOWER= -0.1
%
DEFINITION_DV= ( 15, 1 | foil | FFD_FOIL, 0, 0, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 1, 0, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 2, 0, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 3, 0, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 4, 0, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 0, 1, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 1, 1, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 2, 1, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 3, 1, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 4, 1, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 0, 2, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 1, 2, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 2, 2, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 3, 2, 0.1, 0.1); ( 15, 1 | foil | FFD_FOIL, 4, 2, 0.1, 0.1)
su2 mesh (definition parts only - whole mesh too big to upload!)
Code:
NDIME= 2
AOA_OFFSET= 0
AOS_OFFSET= 0
NELEM= 17464

# 2D MESH GOES HERE BUT NOT INCLUDED #

NPERIODIC= 1
PERIODIC_INDEX= 0
0.000000000000000e+00    0.000000000000000e+00    0.000000000000000e+00
0.000000000000000e+00    0.000000000000000e+00    0.000000000000000e+00
0.000000000000000e+00    0.000000000000000e+00    0.000000000000000e+00
FFD_NBOX= 1
FFD_NLEVEL= 1
FFD_TAG= FFD_FOIL
FFD_LEVEL= 0
FFD_DEGREE_I= 4
FFD_DEGREE_J= 2
FFD_BLENDING= BEZIER
FFD_PARENTS= 0
FFD_CHILDREN= 0
FFD_CORNER_POINTS= 4
0    0.1
1    0.1
1    -0.1
0    -0.1
FFD_CONTROL_POINTS= 15
0    0    0    0    0.1    0
0    1    0    0    0    0
0    2    0    0    -0.1    0
1    0    0    0.25    0.1    0
1    1    0    0.25    0    0
1    2    0    0.25    -0.1    0
2    0    0    0.5    0.1    0
2    1    0    0.5    0    0
2    2    0    0.5    -0.1    0
3    0    0    0.75    0.1    0
3    1    0    0.75    0    0
3    2    0    0.75    -0.1    0
4    0    0    1    0.1    0
4    1    0    1    0    0
4    2    0    1    -0.1    0
FFD_SURFACE_POINTS= 199 
#ONLY TWO POINTS INCLUDED IN THIS POST FOR CONCISENESS#
LOWER    104    9.928418222292663e-05    5.000597183305870e-01    5.000000000000000e-01
UPPER    1    1.000000000000000e+00    5.000000000000000e-01    5.000000000000000e-01
tomp1993 is offline   Reply With Quote

Old   March 27, 2017, 06:47
Default
  #2
New Member
 
WANG,Hao
Join Date: Sep 2016
Location: Xi'an China
Posts: 19
Rep Power: 6
Haowang is on a distinguished road
Hi tom,

I would recommand you to check whether the FFD box intersects with the airfoil surface. Or you could also check the screen output of SU2_DEF and SU2_DOT modules. In some case, it will tell you that some planes has been fixed to ensure the continuity of the surface. If so, the FFD control points won't move and the corresponding gradients will be zero as well.

Hao Wang
Haowang is offline   Reply With Quote

Old   March 27, 2017, 11:00
Thumbs up
  #3
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
Thanks for your help!
so, you are right and it is fixing planes which are intersecting.
is there a way to switch this off without changing the geometry or size of the FFD box? or should i simply re-make the FFD box so that its control points lie entirely outside of the geometry? should it matter if an FFD connector (edge) is penetrating the surface or is it simply a case of relocating the control points?
tomp1993 is offline   Reply With Quote

Old   March 27, 2017, 11:57
Thumbs down
  #4
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
i have made a new FFD definition with no points within the aerofoil and no connectors (edges) penetrating the surface, but for some reason am still having the same response to SU2_DEF:
Code:
------------------------- Surface grid deformation ----------------------
Performing the deformation of the surface grid.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1. 
FFD Blending using Bezier Curves.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 4. Control points: 10. Surface points: 199.
Writing a Paraview file of the FFD boxes.

----------------- FFD technique (parametric -> cartesian) ---------------
Checking FFD box dimension.
Checking FFD box intersections with the solid surfaces.
SU2 is fixing the planes to maintain a continuous 2nd order derivative.
Update cartesian coord        | FFD box: FFD_FOIL. Max Diff: 0.52505.
Writing a Paraview file of the FFD boxes.
not really sure where to go from here?
also, when i run shape_optimization.py, the log_adjoint file says the FFD box is different (see corner points, surf points...), which it isn't...
Code:
---------- Start gradient evaluation using surface sensitivity ----------
Read the FFD information from mesh file.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 0. Control points: 4. Surface points: 0.
Check the FFD box intersections with the solid surfaces.
-------------------------------------------------------------------------
tomp1993 is offline   Reply With Quote

Old   March 28, 2017, 00:00
Default
  #5
New Member
 
WANG,Hao
Join Date: Sep 2016
Location: Xi'an China
Posts: 19
Rep Power: 6
Haowang is on a distinguished road
Quote:
Originally Posted by tomp1993 View Post
i have made a new FFD definition with no points within the aerofoil and no connectors (edges) penetrating the surface, but for some reason am still having the same response to SU2_DEF:
Code:
------------------------- Surface grid deformation ----------------------
Performing the deformation of the surface grid.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1. 
FFD Blending using Bezier Curves.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 4. Control points: 10. Surface points: 199.
Writing a Paraview file of the FFD boxes.

----------------- FFD technique (parametric -> cartesian) ---------------
Checking FFD box dimension.
Checking FFD box intersections with the solid surfaces.
SU2 is fixing the planes to maintain a continuous 2nd order derivative.
Update cartesian coord        | FFD box: FFD_FOIL. Max Diff: 0.52505.
Writing a Paraview file of the FFD boxes.
not really sure where to go from here?
also, when i run shape_optimization.py, the log_adjoint file says the FFD box is different (see corner points, surf points...), which it isn't...
Code:
---------- Start gradient evaluation using surface sensitivity ----------
Read the FFD information from mesh file.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 0. Control points: 4. Surface points: 0.
Check the FFD box intersections with the solid surfaces.
-------------------------------------------------------------------------
As far as i know, only when your FFD box doesn't intersect with the airfoil surface, theoretically all the control points set as design variables can move.
I think you need to check the ffd_box.dat file to make sure whether the ffd box has been changed or not, because from your screen output, it doesn't say that any plane has been fixed.
As for your log_adjoint file, I didn't encounter this problem before. I guess you didn't use SU2_DEF to set your FFD box.
Haowang is offline   Reply With Quote

Old   March 30, 2017, 07:31
Question
  #6
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
Hi again,
so i have tried using SU2_DEF to set the FFD box instead of the Pointwise script, and have exactly the same issue. SU2_DEF executed without any problems and i can't see anything obviously wrong with the .su2 file which it outputted.
the adjoint solution is reasonably converged and shows the surface sensitivities as expected.
once again, log_adjoint outputs the following:
Code:
---------- Start gradient evaluation using surface sensitivity ----------
Read the FFD information from mesh file.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 0. Control points: 4. Surface points: 0.
Check the FFD box intersections with the solid surfaces.
-------------------------------------------------------------------------

Design variable number 0.
Perform 3D deformation of the surface.
Update cartesian coord        | FFD box: FFD_FOIL. Max Diff: 0.
Evaluate functional gradient using the continuous adjoint strategy.
Drag coefficient gradient: 0.

#and so on...#
whereas SU2_DEF outputs this:
Code:
------------------------- Surface grid deformation ----------------------
Performing the deformation of the surface grid.
1 Free Form Deformation boxes.
1 Free Form Deformation nested levels.
FFD box tag: FFD_FOIL. FFD box level: 0. Degrees: 4, 1. 
FFD Blending using Bezier Curves.
Number of parent boxes: 0. Number of child boxes: 0.
Corner points: 4. Control points: 10. Surface points: 199.
Writing a Paraview file of the FFD boxes.

----------------- FFD technique (parametric -> cartesian) ---------------
Checking FFD box dimension.
Checking FFD box intersections with the solid surfaces.
SU2 is fixing the planes to maintain a continuous 1st order derivative.
Update cartesian coord        | FFD box: FFD_FOIL. Max Diff: 0.525.
Writing a Paraview file of the FFD boxes.
i am quite confused as they seem to be reading the FFD information differently? SU2_DEF is fine but shape_optimization.py just doesnt seem to be reading the FFD information in the same way?
thanks for the help.
tomp1993 is offline   Reply With Quote

Old   March 31, 2017, 05:19
Default
  #7
Super Moderator
 
Tim Albring
Join Date: Sep 2015
Posts: 195
Rep Power: 7
talbring is on a distinguished road
Hi Tom,

it looks like shape_optimization.py uses an older version of SU2_DEF (the line "FFD Blending using Bezier Curves." was added with changes in 5.0. Make sure that the environment variables SU2_RUN and PATH are pointing to the correct directories.

Tim
talbring is offline   Reply With Quote

Old   April 11, 2017, 10:25
Thumbs up
  #8
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
Thanks Tim that was absolutely bang on,
turns out i had SU2_HOME pointed to 5.0.0 but SU2_RUN pointed to the previous version.
Tom
tomp1993 is offline   Reply With Quote

Old   April 12, 2017, 05:07
Default
  #9
Super Moderator
 
Tim Albring
Join Date: Sep 2015
Posts: 195
Rep Power: 7
talbring is on a distinguished road
Glad that it solved the problem. Thanks for letting us know.
talbring is offline   Reply With Quote

Old   August 29, 2017, 13:44
Default
  #10
Member
 
tom pratt
Join Date: Jan 2016
Posts: 55
Rep Power: 6
tomp1993 is on a distinguished road
Hi,
so; i thought this issue had been resolved by following your advice; and whilst you were correct; unfortunately the issue is not solved.
the gradient i am finding is still zero.
i have also included AOA as a design variable, and the gradient for this is non-zero as expected.
log_adjoint.out shows the following:
Code:
---------- Start gradient evaluation using sensitivity information ----------
Evaluate functional gradient using Finite Differences.

Design variable number 1.
Performing 3D deformation of the surface.
...
Design variable (ANGLE_OF_ATTACK) number 0.
DRAG gradient : 0.00274491
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 1.
DRAG gradient : 0
-------------------------------------------------------------------------
...
etc
whilst log_geometry.out shows:
Code:
 -------------------- Objective function evaluation ----------------------

Station 1. Plane (yCoord): 0.
Max. thickness: 0.0599312, 1/3C thickness: 0.0533524, 2/3C thickness: 0.0533529
1/4C thickness: 0.045046, 1/2C thickness: 0.0599898, 3/4C thickness: 0.0450466
Area: 0.0400106, Twist angle: 0, Chord: 0.999996

------------- Gradient evaluation using finite differences --------------
Design Variable not implemented yet

Station 1. Plane (yCoord): 0.
Max. thick. grad.: 0. 1/3c thick. grad.: 0. 2/3c thick. grad.: 0.
1/4c thick. grad.: 0. 1/2c thick. grad.: 0. 3/4c thick. grad.: 0.
Area grad.: 0. Twist angle grad.: 0. Chord grad.: 0.

-------------------------------------------------------------------------

Design variable number 1.
Perform 3D deformation of the surface.

Station 1. Plane (yCoord): 0.
Max. thick. grad.: 0. 1/3c thick. grad.: 0. 2/3c thick. grad.: 0.
1/4c thick. grad.: 0. 1/2c thick. grad.: 0. 3/4c thick. grad.: 0.
Area grad.: 0. Twist angle grad.: 0. Chord grad.: 0.

-------------------------------------------------------------------------
I have initialised the FFD box using SU2_DEF with V.5.0.0.
the description in the .su2 grid is as follows:
Code:
 NPERIODIC= 1
PERIODIC_INDEX= 0
0.000000000000000e+00   0.000000000000000e+00   0.000000000000000e+00
0.000000000000000e+00   0.000000000000000e+00   0.000000000000000e+00
0.000000000000000e+00   0.000000000000000e+00   0.000000000000000e+00
FFD_NBOX= 1
FFD_NLEVEL= 1
FFD_TAG= FFD_FOIL
FFD_LEVEL= 0
FFD_DEGREE_I= 4
FFD_DEGREE_J= 1
FFD_BLENDING= BSPLINE_UNIFORM
BSPLINE_ORDER_I= 2
BSPLINE_ORDER_J= 2
FFD_PARENTS= 0
FFD_CHILDREN= 0
FFD_CORNER_POINTS= 4
-0.05   0.06
1.05    0.06
1.05    -0.06
-0.05   -0.06
FFD_CONTROL_POINTS= 10
0       0       -0.05   0.06
0       1       -0.05   -0.06
1       0       0.225   0.06
1       1       0.225   -0.06
2       0       0.5     0.06
2       1       0.5     -0.06
3       0       0.775   0.06
3       1       0.775   -0.06
4       0       1.05    0.06
4       1       1.05    -0.06
FFD_SURFACE_POINTS= 398
...
etc
(i have also tried with all xyz,ijk values of the corner and control points as opposed to just xy and ij as shown above; the result is the same!)
Thanks in advance!
Tom.
tomp1993 is offline   Reply With Quote

Old   July 19, 2018, 08:30
Default
  #11
New Member
 
aravindh
Join Date: May 2018
Posts: 1
Rep Power: 0
aravindh_kulandaivel is on a distinguished road
Quote:
Originally Posted by Haowang View Post
As far as i know, only when your FFD box doesn't intersect with the airfoil surface, theoretically all the control points set as design variables can move.
I think you need to check the ffd_box.dat file to make sure whether the ffd box has been changed or not, because from your screen output, it doesn't say that any plane has been fixed.
As for your log_adjoint file, I didn't encounter this problem before. I guess you didn't use SU2_DEF to set your FFD box.
Hello,
I have the same problem. My SU2 is fixing the planes, so no deformation occurs when visualized in PARAVIEW. But ffd boxes are changing as per ffd_box.dat file. Could you please help me to solve this issue?

Thanks in advance
Aravindh
aravindh_kulandaivel is offline   Reply With Quote

Old   August 1, 2019, 16:14
Default FFD not interracting with geometry
  #12
New Member
 
Angu Praveen
Join Date: May 2019
Posts: 13
Rep Power: 3
angupraveen is on a distinguished road
Dear all,

I know this is a long shot, but you are the one who can help me here.

I am trying to optimize the wing section of a BWB aircraft. I am currently facing a problem where my FFD is not interacting with the geometry.

I am not aware of how to get over this. So basically I ran the SU2_DEF and generated the FFD and then set up a few control points(198 in my case) as design variables and then run SU2_CFD and SU2_CFD_AD after that.

Once all the calculations are done, I wanted to do the sensitivity analysis before going forward with the optimization. Unfortunately, my FFD intersects the geometry and all the gradients are "Zero". I tried changing the FFD_CONTINUITY to 'NO_DERIVATIVE', '1ST_DERIVATIVE' and also 2ND_DERIVATIVE. But the results are the same.

It would be really helpful if you can help me bypass this issue,

I've attached a few pictures and the outputs for your reference.

Looking forward to hearing from you.

Kind regards

Angu


----------------- FFD technique (parametric -> cartesian) ---------------
Checking FFD box dimension.
Checking FFD box intersections with the solid surfaces.
The FFD planes i=0, i=10, intersect solid surfaces.
SU2 is fixing the planes to maintain a continuous 2nd order derivative.
Update cartesian coord | FFD box: WING. Max Diff: 0.

Design variable (FFD_CONTROL_POINT) number 0.
DRAG gradient : 0
-------------------------------------------------------------------------


angupraveen is offline   Reply With Quote

Reply

Tags
ffd, gradient, su2 adjoint

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
question regarding LES of pipe flow - pimpleFoam Dan1788 OpenFOAM Running, Solving & CFD 37 December 26, 2017 15:42
Periodic flow using Cyclic - comparison with Fluent nusivares OpenFOAM Running, Solving & CFD 30 December 12, 2017 06:35
Gradient Computation: finite differences and adjoint method doan.nak SU2 3 November 22, 2017 06:37
Question about heat transfer coefficient setting for CFX Anna Tian CFX 1 June 16, 2013 07:28
Solidification drag coefficient atulverma FLOW-3D 1 May 14, 2009 10:24


All times are GMT -4. The time now is 04:24.