CFD discrete adjoint benchmarks
Description
This data set contains meshes for three discrete adjoint benchmarks along with ADflow configuration scripts to reproduce the results in the following paper: Gaetan K.W. Kenway, Charles A. Mader, Ping He, and Joaquim R. R. A. Martins. Effective adjoint approaches in Computational Fluid Dynamics, Progress in Aerospace Sciences, 2019. In this paper we used two CFD solvers that have discrete adjoints: ADflow (see link below to access code) and OpenFOAM (with an internal discrete adjoint). If you are interested in running your adjoint solver for comparison, we recommend to use the provided meshes (either in structured or unstructured format) and the same flow conditions and CPU cores as described in the paper. If you would like to generate your own meshes, we suggest using a similar cell number for each benchmark. When comparing the performance, please report the runtime (in Taubench work unit), peak memory usage (resident set size in GB), number of CPU cores and mesh cells, as well as a description of your adjoint solver and the computing platform (including the Tau reference runtime). The cases are based on the following benchmarks developed by the AIAA Aerodynamic Design Optimization Discussion Group (ADODG): • ADODG3: A low-speed rectangular wing. • ADODG4: The Common Research Model (CRM) wing. • ADODG5: The CRM wing-body-tail configuration. The filename suffixes are: *_Multiblock_Structured_Mesh.cgns: Multiblock structured mesh files for ADflow. *_Overset_Structured_Mesh.cgns: Multiblock overset mesh files for ADflow. *_Singleblock_Unstructured_Mesh.cgns: These meshes are identical to *_Multiblock_Structured_Mesh.cgns except that they are in single block unstructured mesh format. The far field, symmetry, and wall boundary names are FARFIELD, SYMMETRY, and WING, respectively. *_FFD.xyz: Free-form deformation (FFD) files for defining the shape design variables. *_ADflow_Run_Script.py: Python scripts for reproducing the ADflow adjoint results in the paper. See detailed instructions in the scripts. Code_Version.txt: The required codes to produce the results, including links and versions.
Files
Steps to reproduce
1. Download and install the codes as listed in Code_Version.txt. Make sure you use the correct versions. 2. For each benchmark, download the corresponding mesh, FFD, and ADflow_Run_Script.py files. 3. Run the ADflow_Run_Script.py script using the command included in the header.