Gradients are Always 0 for Unsteady Shape Optimization

[ErcanUmut60]

Hello everbody,

I am trying to use Discrete Adjoint method for a launch pad optimization (45deg wall) for an unsteady flow. I applied an FFD box over the wall and tried to set up a configuration for adjoint method. I am new to the Adjoint Methodology all-together. I have a couple of problems and questions for anybody who might be able to help.

1- I am trying to obtain a deformation over a partial area of the wall. I tried different FFD boxes as you can see from the pictures. I went over some Test Cases. There was a supersonic wedge case which applied STATIC_TOTAL_PRESSURE as objective function. In my case its static pressure but the idea is the same. I wanted to make some similar ffd boxes but I am not sure if they are adequate or not. I would be glad if I can get any feedback about them.

2- No matter of the FFD box placement, I get gradients of zero as shown below. I examined the Tutorial, Unsteady Adjoint Airfoil, and approached my problem in a similar fashion. What could be the reason for this?

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

Design variable (FFD_CONTROL_POINT_2D) number 0.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 1.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 2.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 3.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 4.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 5.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 6.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 7.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 8.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 9.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 10.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 11.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 12.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------

Design variable (FFD_CONTROL_POINT_2D) number 13.
SURFACE_STATIC_PRESSURE gradient : 0
-------------------------------------------------------------------------`

3- When I use a constraint, say lift as in the tutorial, the simulation continues with other designs. But when I set OPT_CONSTRAINT=NONE simulation stops as shown below. Why does this occur?

Found: mesh_out.su2
New Project: ./
Removing old designs in 10s. Done!

Sequential Least SQuares Programming (SLSQP) parameters:
Number of design variables: 14 ( 14 ) 
Objective function scaling factor: [0.0001]
Maximum number of iterations: 100
Requested accuracy: 1.0000000000000003e-18
Initial guess for the independent variable(s): [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, 0.0]
Lower and upper bound for each independent variable: [(-0.001, 0.001), (-0.001, 0.001), (-0.00
1, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001)
, (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001, 0.001), (-0.001
, 0.001)]

  NIT    FC           OBJFUN            GNORM
    1     1     0.000000E+00     0.000000E+00
Optimization terminated successfully    (Exit mode 0)
            Current function value: 0.0
            Iterations: 1
            Function evaluations: 1
            Gradient evaluations: 1

4- Is there any way to select a pressure point (probe) somewhere near the nozzle as an objective and make the shape deformation over the wall according to that point? Or is it possible to reading outlet values and making deformation according that?

[bigfooted]

I do not see what is going on here, how the ffd box is positioned and what the design really looks like, can you post some more pictures, or a link to the testcase? First thing to check is if the adjoint has converged nonzero solution

[ErcanUmut60]

Yes of course.

https://su2code.github.io/tutorials/Unsteady_Shape_Opt_NACA0012/
https://github.com/su2code/TestCases/tree/57532f042f8b88882aa59019654078ab2f39aef7/cont_adj_euler/wedge

These are my references mainly. I am trying to use adjoint methodology to reduce the high pressure over the wall during the ignition of the rocket. My ffd box settings are

% -------------------- FREE-FORM DEFORMATION PARAMETERS -----------------------%

% Tolerance of the Free-Form Deformation point inversion
FFD_TOLERANCE= 1E-10
%
% Maximum number of iterations in the Free-Form Deformation point inversion
FFD_ITERATIONS= 500

FFD_DEFINITION= (BOX, \
0.25, 0.1, 0.0, \
0.1, 0.25, 0.0, \
-0.30, -0.15, 0.0, \
-0.15, -0.3, 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= (1, 6, 0)

% ----------------------- DESIGN VARIABLE PARAMETERS --------------------------%
%
DV_KIND= FFD_SETTING
% Marker of the surface in which we are going apply the shape deformation
% NOTE: for deformation the outlet should be a MARKER_SYM to hinder the mesh being ripped apart.
DV_MARKER= ( wall )
%
% Parameters of the shape deformation

DV_PARAM= ( 1.0 )

and you can see the geometry below. I am trying to conduct an Unsteady Shape Optimization since pressure is varying with time in my case. Other inputs,

% ----------------------- DESIGN VARIABLE PARAMETERS --------------------------%

DV_KIND= FFD_CONTROL_POINT_2D
DV_MARKER= ( wall )
DV_PARAM= ( BOX, 0.0, 0.0, 1.0, 1.0)
DV_VALUE= 0.2
DV_FILENAME= mesh_motion.dat

% ------------------------ GRID DEFORMATION PARAMETERS ------------------------%

DEFORM_LINEAR_SOLVER_ITER= 1000

DEFORM_NONLINEAR_ITER= 1

DEFORM_CONSOLE_OUTPUT= YES

DEFORM_LINEAR_SOLVER_ERROR =  1E-6

DEFORM_STIFFNESS_TYPE= WALL_DISTANCE

% --------------------- Optimization Parameters ----------------------------%

OPT_OBJECTIVE= SURFACE_STATIC_PRESSURE * 10.0
OPT_CONSTRAINT= NONE
OPT_ITERATIONS= 10
OPT_GRADIENT_FACTOR= 0.1
OPT_RELAX_FACTOR= 1

OPT_BOUND_UPPER= 0.1
OPT_BOUND_LOWER= -0.1

DEFINITION_DV=(19, 1.0 | wall | BOX, 0, 2, 0.5, 1.0); (19, 1.0 | wall | BOX, 0, 3, 0.5, 1.0); (19, 1.0 | wall | BOX, 0, 4, 0.5, 1.0); (19, 1.0 | wall | BOX, 0, 5, 0.5, 1.0); (19, 1.0 | wall | BOX, 0, 6, 0.5, 1.0);  (19, 1.0 | wall | BOX, 1, 2, 0.5, 1.0); (19, 1.0 | wall | BOX, 1, 3, 0.5, 1.0); (19, 1.0 | wall | BOX, 1, 4, 0.5, 1.0); (19, 1.0 | wall | BOX, 1, 5, 0.5, 1.0); (19, 1.0 | wall | BOX, 1, 6, 0.5, 1.0) 

How can I check if the adjoint has converged nonzero solution?