loganoz / horses3d

HORSES3D: A high-order discontinuous Galerkin solver for flow simulations and multi-physics applications
https://loganoz.github.io/horses3d/
MIT License
118 stars 24 forks source link

Gerasimos dev2 #23

Closed Jerryntk closed 2 years ago

Jerryntk commented 2 years ago

This is a Spalart-Almaras turbulence model implementation for RANS simulations. The model has been taken by the work of Joshi et al. [1,2]. The model parameters and implementation can be found in src/libs/physics/spallartalmaras. Any turbulence model can be added to the Turbulence.f90 file and the rest is taken care of. To run a RANS case you need to use the horses3d.nssa solver. The method has been tested and works with the time integration methods implemented.

I have included two test cases, a Manufactured solution and the Flat plate (Re=6,000,000) from NASA turbulence research. They can be used as a guide for setting up other simulations.

The solver .nssa contains the -DNAVIERSTOKES -DSPALARTALMRAS flags so it uses all the segments of the code enclosed by those flags

Limitations/Considerations for using the solver:

The wall distance calculator is based on the calculation of the distance to the closest degree of freedom and can be found in HexMesh.f90. That means that for stretched meshes the approximation can incur error. This error might be prohibitive for obtaining very accurate results (Error in Cd > 1e-4). If the analytical geometry is known_, there is an implementation of an optimization algorithm (COBYLA) which can be used to find the minimum distance accurately. In general for hex meshes which are not skew the implemented method should work adequately.

The approximation of the geometry. A current restriction which extends to other solvers in HORSES3D. If the geometry is of higher order than the solution we downgrade the mesh polynomial order. That means that a Q3 mesh will be transformed to Q2 before calculating the metric terms for a P2 solution. The bad approximation of the geometry can have a significant impact on the results. Mesh refinement near the geometry might be necessary to capture the geometry with lower polynomial order.

Coupling has only been performed for the compressible Navier-Stokes solver. Extension to the other solver should be fairly straightforward following the steps in this solver.

References: [1] S Joshi, A Hurtado-de-Mendoza, J Kou, K Puri, C Hirsch, E Ferrer, "Design of a non-linear hp–multigrid scheme with static near–wall p–adaptation for RANS in a high–order flux reconstruction framework" [2] Oliver, T.A.: A high-order, adaptive, discontinuous Galerkin nite element method for the Reynolds-Averaged Navier-Stokes equations. MIT, Cambridge, Dept. of Aeronautics and Astronautics (2008)