The SU2 team and Pointwise recently teamed up and ran one of the meshes generated for the AIAA High Lift Prediction Workshop using the latest release of SU2. Information regarding the workshop, the meshing process, solver setup, and the SU2 configuration file are all here.
In an effort to assess the state-of-the-art with regards to computational fluid dynamics (CFD) and its predictive capabilities, AIAA has organized a number of workshops, such as the High Lift Prediction Workshop (HiLiftPW). The HiLiftPW is one where an aircraft in landing/take-off configuration is used as a benchmark case, exercising current CFD solver technologies. Pointwise was an active participant in HiLiftPW-3 and generated a family of unstructured, hybrid, hybrid overset, and high order meshes for the NASA High Lift Common Research Model (CRM) geometry. These meshes represented the committee supplied meshes for participants to use when preparing the required test cases.
Following the workshop, it was found the hybrid hex-prism-tet mesh was used more frequently by participants than the other meshes provided. With that in mind, we decided to take a brief look at how the mesh was constructed and run a simulation using the latest release of SU2, making the configuration file available for those who would like to reproduce our results.
Hybrid Mesh Generation
The strategies used to prepare the hybrid hex-prism-tet mesh are unique to Pointwise and represent the latest in hybrid viscous meshing technology. For instance, the geometry itself was prepared by assembling surfaces into unique meshing regions. Pointwise’s bottom-up meshing process relies on these regions to define the surface mesh topology. Quad-dominant surface meshes were then generated with anisotropic stretching applied in areas to resolve curvature and reduce surface cell count, and ultimately, volumetric cell count. In practice, anisotropic stretching on the surface can reduce volumetric cell counts by 2-3x over pure isotropic surface level refinement approaches. The viscous volume mesh was generated using the anisotropic tetrahedral extrusion technique (T-Rex) where unstructured hexes and prisms are constructed in an advancing normal fashion near the boundary and transition to isotropic tetrahedra in the off-body region. To read more about the specifics of the geometry, the workshop meshing guidelines, and details regarding the meshing process, click the link below.
SU2 v6 Falcon
SU2 v6 Falcon, the latest version of the open-source, multiphysics simulation and design code, was released on Valentine’s Day of 2018. SU2 is a collection of software tools written in C++ and Python for the analysis of partial differential equations (PDEs) and PDE-constrained optimization problems on unstructured meshes with state-of-the-art numerical methods. SU2 is a leading technology for adjoint-based optimization. The code is open source and freely available on GitHub. Everyone is encouraged to get involved in the project, and recently the SU2 International Developers Society was unveiled to help organize, inspire, and connect contributors from all around the globe.
SU2 v6 is packed with new features, representing roughly a year of work by developers all around the world. At the 2nd Annual SU2 Developers Meeting in late December 2017, it was reported that SU2 v6 contains over 74 successful GitHub Pull Requests worth of contributions into the master branch, featuring over 1,385 commits containing 18,193 new lines of code.
A standout new addition is the Delayed Detached Eddy Simulation (DDES) technique for computing scale-resolving flows, developed in collaboration with the department of Aeronautics and Mechanics of the Aeronautical Institute of Technology (ITA-Brazil). The implementation includes state-of-the art unstructured non-zonal Grey Area Mitigation (GAM) methods, specifically designed to accelerate the RANS to LES transition. This DDES extension will both provide a much needed community framework for further improvement of hybrid RANS/LES methods and position SU2 to meet the demands of new industrial applications.
Alongside the new software version, the SU2 project website received a major upgrade. The new site is based on GitHub pages and Jekyll, which made for a nice development experience. However, the main motivation for the change was in keeping with the spirit of open-source: the new website is entirely open and modifiable through GitHub, just like the main source code repository. Since the documentation for SU2 and tutorials were moved over into this same structure, everyone can now easily write docs and examples to share with the rest of the world through Pull Requests to the project website repository.
A standard set of SU2 configuration parameters have been chosen for the high-lift CRM case that are also a good starting point for any compressible RANS case in external aerodynamics.
Thanks to feedback from the community, the way in which convective schemes are specified in SU2 has been improved. In this case, the Roe upwind scheme with 2nd-order reconstruction (MUSCL_FLOW = YES) and a slope limiter was chosen. In order to activate the slope limiter for the upwind methods, SLOPE_LIMITER_FLOW must be set to something other than NONE. Otherwise, no limiting will be applied to the flow variable gradients during the 2nd-order reconstruction. Limiting is not applicable if MUSCL_FLOW = NO, as there is no higher-order reconstruction, and thus, no need to limit the gradients. Setting MUSCL_FLOW= NO with an upwind method results in a 1st-order accurate scheme in space. Several limiters are available in SU2, including the well-known Venkatakrishnan limiter for unstructured grids. It is recommended that users experiment with the VENKAT_LIMITER_COEFF value for their own applications. It should be mentioned that the MUSCL reconstruction and slope limiting apply only to the upwind schemes. If you choose one of the centered convective schemes, e.g., JST or Lax-Friedrichs, there is no reconstruction process. JST and Lax-Friedrichs are 2nd-order and 1st-order by construction, respectively, and the scalar dissipation for these schemes can be tuned with the JST_SENSOR_COEFF and LAX_SENSOR_COEFF options, respectively.
Lastly, several improvements to the ILU preconditioner for solving linear systems have been made within SU2 v6, including the ability to specify a higher level of fill-in if desired (zero fill-in remains the default). The ILU preconditioner can offer better convergence that the other available preconditioners and may allow for higher CFL numbers when solving the systems to tighter tolerances. It is recommend that folks experiment with the ILU preconditioner for their implicit time integration problems, but remember that there is always a trade-off between convergence of the linear system with each nonlinear iteration and the computational cost, especially with higher levels of fill-in. For the high-lift CRM case, the GMRES linear solver with ILU(0) was selected. The linear solver convergence tolerance is set to 1e-10 (relative), which is sufficiently low such that the maximum number of linear solver iterations (5) is executed for each nonlinear iteration of the flow solver.
The preliminary results for the lift and drag coefficients are in line with the other participants in the HiLiftPW at 1.72 and 0.169, respectively, for the 8 degree angle-of-attack case. However, the reader is encouraged to run the case and perform more detailed analysis of the results, including studying pressure coefficient distributions and surface streamlines. The effect of including a 2nd-order turbulence model could be investigated. Readers can also try the other available grids provided by Pointwise or some of the other flow conditions provided by the high-lift prediction workshop.
We invite you to download a copy of SU2 v6 and the the configuration file for the high-lift CRM case and try it for yourself. It has been posted to GitHub within the new SU2 project website repository, and you can access the file at the link below.