Open brendankeith opened 4 years ago
Hi!
I will just add the images and the table in the post so that they are easier to see.
This is the problem @brendankeith is trying to solve, one snapshot of the sensitivities and the final shape one gets.
If we take a look at the objective function, it does not go down:
itr, f, df_abs[%], df_rel[%], norm_df, step_size, time_stamp
1, 1.27204E+09, 0.00000E+00, 0.00000E+00, 8.65154E+07, 1.00000E+00, Fri Mar 20 12:02:05 2020
2, 7.21772E+10, 5.57411E+03, 5.57411E+03, 6.73333E+09, 1.00000E+00, Fri Mar 20 12:02:05 2020
3, 3.65729E+12, 2.87413E+05, 2.81839E+05, 2.90892E+11, 1.00000E+00, Fri Mar 20 12:02:06 2020
4, 1.16725E+14, 9.17605E+06, 8.88864E+06, 4.74951E+11, 1.00000E+00, Fri Mar 20 12:02:06 2020
5, 1.59801E+18, 1.25625E+11, 1.25616E+11, 1.42290E+15, 1.00000E+00, Fri Mar 20 12:02:06 2020
6, 6.42661E+15, 5.05219E+08, -1.25120E+11, 8.34923E+13, 1.00000E+00, Fri Mar 20 12:02:07 2020
7, 2.84295E+16, 2.23494E+09, 1.72972E+09, 3.42314E+13, 1.00000E+00, Fri Mar 20 12:02:07 2020
8, 2.53899E+16, 1.99599E+09, -2.38948E+08, 6.87294E+13, 1.00000E+00, Fri Mar 20 12:02:07 2020
9, 6.17331E+17, 4.85306E+10, 4.65346E+10, 5.62109E+14, 1.00000E+00, Fri Mar 20 12:02:08 2020
10, 2.86949E+15, 2.25581E+08, -4.83050E+10, 2.91385E+12, 1.00000E+00, Fri Mar 20 12:02:08 2020
@brendankeith What element types are you using? Can you compare the mdpa files and try with the same as in the test case?
Hi @armingeiser I was using ShellThinElementCorotational3D3N, however, in the example you are using ShellThinElement3D3N. I made the change and it seems to have fixed everything. THANKS!
@brendankeith There is an error missing in the code, as Corotational elements do not work for sensitivity analysis...
I have set up a shape optimization problem based on the shape optimization example https://github.com/KratosMultiphysics/Examples/tree/master/shape_optimization/02_Strain_Energy_Minimization_3D_Shell. I used a longer shell and, instead of applying a point load, I am looking at a distributed load which grows from one end of the shell to the other.
The sensitivities I am getting back are very suspicious and lead to a somwhat crinkled non-symmetric shape (see attached figures). I am wondering what I am doing wrong.
Note that included is a picture of the initial sensitivities coming from the shape optimization example. Also, the objective function doesn't decrease throughout the optimization process.
@marcnunezc @armingeiser @MFusseder
data.zip