Problem in results when using QCR2000 option fo SA turbulence model.

[fomistul-cfd]

Hello everyone,

Describe the bug
I encounter an odd behaviour for SA turbulence model where when using the QCR2000 option for SA, results for viscous drag component and pressure drag component are shifted, this does not occur with other options.
The case is from this discussion(#2376) , here the results for Sa-neg-QCR2000-R when conducting an grid convergence study converge to wrong values for CDv and CDp, so I looked into the issue, when using SST and SA model without QCR2000 option the results converge to desired value and are in line with results from this article .

Below are the results for grid convergence of 3 cases where the outlier is the case with QCR2000 option.

JST	SA		cdp	cdv
		r0	0.043435	0.000968
		r1	0.011374	0.002303
		r2	0.00318	0.004735
		r3	0.001569	0.006745
		r4	0.00129	0.007067
		r5	0.001253	0.006835
JST	SST		cdp	cdv
		r0	0.043475	0.000958
		r1	0.011196	0.002317
		r2	0.002961	0.004501
		r3	0.001496	0.006218
		r4	0.001301	0.006779
		r5	0.001257	0.006725
JST	SA-neg_QCR2000_R		cdp	cdv
		r0	0.043495	0.000888
		r1	0.011623	0.002106
		r2	0.003469	0.004408
		r3	0.001807	0.006457
		r4	0.001546	0.006842
		r5	0.00144	0.00659

Here are the results for R4 grid where when using QCR2000 option create a jump in pressure component drag and a lowering in viscous drag , the overall drag seems to be similar but the components are different. A problem with the QCR2000 could be the explanation why the results for grid convergence from this discussion(https://github.com//discussions/2376) are not in line with expectation.

	cdp	cdv
SA	0.00129	0.007067
SA-neg	0.001285	0.007066
SA_neg_ft2	0.001285	0.007066
SA_neg_ft2_Rotation	0.001285	0.007032
Sa_neg_ft2_QCR2000	0.001482	0.006853
SA_QCR2000	0.001482	0.006853

Below is the archive with grids and config file used.
JST.tar.gz

• OS: Ubuntu v24
• SU2 Version: [e.g., v8.4.0]

[pcarruscag]

And why would you expect the results with QCR to be the same as without?
JST (at least as implemented in SU2) is not a very good scheme for detailed comparison of viscous drag, the numerical dissipation on walls can be significant.

I suggest using the Roe scheme, and case with known results from other codes, i.e. something used in the drag or lift prediction workshops, maybe there are QCR-2000 results for the DSMA661 https://su2code.github.io/vandv/dsma661/

[fomistul-cfd]

Hi @pcarruscag , I tried running with ROE, SLAU2, AUSMPLUSUP2 schemes, the results are virtually the same, I used JST result as an example because it doens't require a limiter ( so i doesn't require a limit_coeff that needs to be changed when using different refinement grids).

I expected the values ( especially the pressure drag, to be close to those expected from here https://highfidelitycfdverificationworkshop.github.io/papers/rans.pdf (page 8) ( Cdp ≈0.00125 and Cdv≈0.006630), the SU2 cases without QCR2000 option are close to those values, and the outliner is the one with this option included . I also run the test case in ANSYS Fluent with SA model and SA-QCR2000, the resulted values for Cdv and Cdp were basically identical and in line with those expected.

The case is from https://highfidelitycfdverificationworkshop.github.io/ , a workshop similar to HLPW and DPW series workshops.
https://github.com/HighFidelityCFDVerificationWorkshop/2024RANS/tree/main/Joukowski_Results here are some results from different codes that give similar values using SA-neg-QCR2000-R turbulence model.

The case that you sugested have results only for SA and SST in NASA TMR database. I will try to run to run the flat-plate and 2D bump channel and compare the results tomorow to see if it apear also in those cases.
https://turbmodels.larc.nasa.gov/flatplate_saqcr.html
https://turbmodels.larc.nasa.gov/bump_saqcr.html

[fomistul-cfd]

Hi @pcarruscag , I ran the case of bump chanel https://tmbwg.github.io/turbmodels/bump.html , that is part of VandV suite for SU2 https://su2code.github.io/vandv/Bump_Channel/ , when viewing the results for total forces the results are in line expected value, however, in components a trend similar to Jukowsky case is visible where pressure component of drag converge to a different value that expected, pressure drag is overestimated, and viscous one is underestimated when using SA-QCR2000 turbulence model.

Maybe there is some problem in how SU2 calculate the components when QCR2000 option is active, when inspecting pressure and surface friction contour , the SA and SA-QCR are basically identical.

I will try to integrate the values by hand to see if my guess is right.