Branch rans-dev PR, incomplete
#45
Conversation
…on confirmed not to break with the extra variable
… problem doesn't converge even with just the EulerSolver
… terms. wall test problems show mesh dependence, and lack of convergence with small wall elements as a result
Conflicts: test/unit/euler_test_data.cpp test/unit/euler_test_data.hpp
Conflicts: src/physics/fluidflow/euler.cpp
|
@garobed1 and @tuckerbabcock Can you take a look at the details of the build failure? It looks to be related to Adept configuration. Thanks. |
| {"mu", -1.0}, // nondimensional viscosity (if negative, use Sutherland's) | ||
| {"sa-consts", {0.1355, 0.622, 0.666666666666667, 0.41, 0.3, 2, 7.1, 1.2, 0.5, 10, 16, 0.7, 0.9}}, | ||
| //Spalart-Allmaras turbulence model constants, defined in rans_fluxes.hpp | ||
| {"sa-srcs", {1.0, 1.0}} //for debug purposes, disable SA source terms |
There was a problem hiding this comment.
I don't personally like this a default option, I think it can just be an undocumented option that the SA solver can use
There was a problem hiding this comment.
Which one are you referring to, sa-consts, sa-srcs, or both?
There was a problem hiding this comment.
I mean the sa-srcs. If it's just something for debugging then I would rather it not be part of the "public" interface of default options.
There was a problem hiding this comment.
That's fair. As of now it is 'required' to be set, but I think I'll just remove the option altogether since I know the solver is correct now.
|
It looks like the previous issue with the CI on this PR (related to Adept configure) has resolved itself, and the new issue is the issue with Hypre not being compiled with |
|
I merged |
This is a pull request to merge
rans-devintodevas it is. I merged the latestdevbranch intorans-devprior to this, and all tests pass.I leave the branch in an essentially incomplete state since I am going to work with another code for the sake of my upcoming MDO conference paper. While the Spalart-Allmaras turbulence model as it is implemented here produces correct results, the solver strategy is not robust and very often fails when trying to solve problems on RANS-type meshes. I believe the best way to improve robustness is to implement a feature that decouples the Navier-Stokes equations and the SA equation for the start of the solution and solves two separate systems, then recouples the equations after a sufficient residual norm decrease and finishes solving. I may revisit this myself after the paper.
All of the RANS tests that are active pass, although I disabled a few of the tests in
test_rans_integ.cpp:SAViscousSlipWallBC JacobianandSASourceIntegrator::AssembleElementGrad. I am uncertain why the slip wall integrator test fails, but I am reasonably certain that the source term integrator fails due to a finite-differencing error, but I haven't been able to directly prove this.I haven't added a regression test for the SA model due to the volatility of the solver. I can try to make one if it would be helpful, along with making other requested changes before merging.