|
[Sponsors] |
July 20, 2024, 02:57 |
Usage of SU2_DOT
|
#1 |
New Member
Viv Bone
Join Date: Jun 2024
Posts: 15
Rep Power: 2 |
Hello,
I'm looking to use SU2 to manually compute derivatives of an objective function with respect to flight conditions (Mach, AoA) and some shape design variables. I'm not looking to perform shape optimization here; I only need to evaluate gradients at a range of input points. I've parameterized my geometry with a 3D FFD box, which I am manually adjusting by setting the DV_PARAM and DV_VALUE entries in my config file. The following steps are running correctly: + mesh deformation using SU2_DEF + evaluation of primal solution using SU2_CFD + evaluation of adjoint solution using SU2_CFD. Using the SENS_MACH and SENS_AOA outputs, I've validated my adjoint solution against finite difference gradients. When I run SU2_DOT <config_file>.cfg, I get the following error message: Surface sensitivity not found in file Here, my simulation is 3D, so my surface adjoint results are output as `surface_adjoint.vtu'; could this be the cause of the problem? If I run the same steps on the QuickStart example, everything works and I get a gradient in the `of_grad.dat' file. Thanks for any help. |
|
July 21, 2024, 22:33 |
|
#2 |
New Member
Tongtong
Join Date: Jan 2024
Posts: 24
Rep Power: 2 |
Hi MaxGains,
Could you please describe the problem in more detail or provide a screenshot of the error? From my experience, the SU2_DOT module uses the result file from SU2_CFD (restart_adj.dat, which should be renamed to solution_adj.dat) to obtain the projection results for the design variables. Tongtong |
|
July 22, 2024, 22:50 |
|
#3 |
New Member
Viv Bone
Join Date: Jun 2024
Posts: 15
Rep Power: 2 |
Thanks Tongtong.
As a reference point, the QuickStart examples works fine for me. After running the primal and adjoint simulations and renaming the restart files, my directory contains the following files: Code:
adjoint.vtu flow.vtu history.csv inv_NACA0012.cfg mesh_NACA0012_inv.su2 solution_adj_cd.dat solution_flow.dat surface_adjoint.csv surface_flow.csv Code:
SU2_DOT inv_NACA0012.cfg Code:
of_grad.dat I'm now trying to replicate this workflow for my own example. I first converge my primal and adjoint simulations, then rename restart_adj_cl.dat to solution_adj_cl.dat. I run my job file Code:
SU2_DOT run_adjoint.cfg Code:
----------------- Physical Case Definition ( Zone 0 ) ------------------- Input mesh file name: mesh_deformed.su2 Input sensitivity file name: surface_adjoint. -------------- Surface deformation parameters ( Zone 0 ) ---------------- Design variables definition (markers <-> value <-> param): FFD (control point) <-> WING3D <-> 0 <-> ( WING, 0, 0, 0, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0 <-> ( WING, 1, 0, 0, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0.0406 <-> ( WING, 0, 1, 0, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0.0406 <-> ( WING, 1, 1, 0, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0 <-> ( WING, 0, 0, 1, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0 <-> ( WING, 1, 0, 1, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0.0406 <-> ( WING, 0, 1, 1, 1, 0, 0 ) FFD (control point) <-> WING3D <-> 0.0406 <-> ( WING, 1, 1, 1, 1, 0, 0 ) ---------------- Design problem definition ( Zone 0 ) ------------------ CL objective function. -------------------- Output Information ( Zone 0 ) ---------------------- Output gradient file name: of_grad.dat. ------------- Config File Boundary Information ( Zone 0 ) --------------- +-----------------------------------------------------------------------+ | Marker Type| Marker Name| +-----------------------------------------------------------------------+ | Far-field| FARFIELD| +-----------------------------------------------------------------------+ | Symmetry plane| SYMMETRY| +-----------------------------------------------------------------------+ | Heat flux wall| WING3D| +-----------------------------------------------------------------------+ Three dimensional problem. 2057 grid points. 1920 volume elements. 3 surface markers. 120 boundary elements in index 0 (Marker = WING3D). 192 boundary elements in index 1 (Marker = SYMMETRY). 120 boundary elements in index 2 (Marker = FARFIELD). 1728 hexahedra. 192 prisms. Identify vertices. Storing a mapping from global to local point index. ----------------------- Preprocessing computations ---------------------- Setting local point connectivity. Checking the numerical grid orientation of the elements. All volume elements are correctly orientend. All surface elements are correctly orientend. Identify edges and vertices. Setting the bound control volume structure. Storing a mapping from global to local point index. Reading surface sensitivities at each node from file. Error in "virtual void CPhysicalGeometry::SetBoundSensitivity(CConfig*)": ------------------------------------------------------------------------- Surface sensitivity not found in file. ------------------------------ Error Exit ------------------------------- Here, since my case is 3D, my surface adjoint file is surface_adjoint.vtu (rather than surface_adjoint.dat). Could this be causing the problem? Regards, Viv |
|
July 23, 2024, 15:05 |
|
#4 |
Senior Member
bigfoot
Join Date: Dec 2011
Location: Netherlands
Posts: 676
Rep Power: 21 |
I don't think the reader can read paraview vtu files. You need SURFACE_CSV in OUTPUT_FILES I guess.
If this is indeed the problem, maybe we can change the error message. |
|
July 24, 2024, 23:54 |
|
#5 |
New Member
Viv Bone
Join Date: Jun 2024
Posts: 15
Rep Power: 2 |
Yep -- that was the problem!
Very easy to convert between the two with vtk library though |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
CPUs vs GPUs for CFD? | hami11 | Hardware | 22 | July 23, 2024 16:59 |
Fluent time usage plot and save to file | svantevid | FLUENT | 1 | December 12, 2023 05:01 |
Large memory usage for steady simulation | Wenqiang | SU2 | 2 | May 15, 2020 14:40 |
New workstation for different usage scenarios - CPU and RAM | natem | Hardware | 6 | August 7, 2013 03:47 |
OpenFOAM Solver/BC usage description | murrayjc | OpenFOAM | 3 | August 25, 2009 05:48 |