ENH: yPlus: add option to disable wall function contributions (#1773)

Summary

See the self-explanatory commit messages.

Resolved bugs (If applicable)

#1773

Details of new models (If applicable)

Small test cases: GL1773-pre-fix.zip vs GL1773-post-fix

Test-case characteristics:

• One-dimensional smooth-wall plane channel flow, ReTau=5200
• Number of cells = 20
• nu = 0.000192827 [m2/s]
• The case was designed to produce => -1*sqrt(mag(wallShearStress)) = 1
• The y1+ set (expected) = {0.05, 0.5, 1, 5, 10, 20, 30, 50, 100, 1000}
• kOmegaSST/SpalartAllmaras, simpleFoam

Main field of interest:

• mag(wallShearStress) function object returns ~ 1 for all test cases, therefore the expected y1+
• yPlus function object (min) results for lowerWall rounded up to 2 decimals:

y1+ = 0.05:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	0.12	0.05
nutkWallFunction	0.00025	0.05
nutUSpaldingWallFunction	0.05	0.05
nutUBlendedWallFunction	0.05	0.05
nutLowReWallFunction	0.05	0.05

y1+ = 0.5:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	0.18	0.50
nutkWallFunction	0.025	0.50
nutUSpaldingWallFunction	0.50	0.50
nutUBlendedWallFunction	0.50	0.50

y1+ = 1:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	0.34	1.00
nutkWallFunction	0.1	1.00
nutUSpaldingWallFunction	1.00	1.00
nutUBlendedWallFunction	1.00	1.00

y1+ = 5:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	3.02	5.00
nutkWallFunction	2.75	5.00
nutUSpaldingWallFunction	5.00	5.00
nutUBlendedWallFunction	5.00	5.00

y1+ = 10:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	9.12	10.00
nutkWallFunction	8.31	10.00
nutUSpaldingWallFunction	10.00	10.00
nutUBlendedWallFunction	10.00	10.00

y1+ = 20:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	20.00	20.02
nutkWallFunction	19.30	20.01
nutUSpaldingWallFunction	20.00	20.00
nutUBlendedWallFunction	20.00	20.01

y1+ = 30:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	30.09	30.09
nutkWallFunction	29.91	30.12
nutUSpaldingWallFunction	30.08	30.08
nutUBlendedWallFunction	30.09	30.09

y1+ = 50:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	50.16	50.16
nutkWallFunction	50.51	50.19
nutUSpaldingWallFunction	50.17	50.17
nutUBlendedWallFunction	50.16	50.16

y1+ = 100:

Wall function	Pre-fix yPlus	Post-fix yPlus (when useWallFunction=false)
nutUWallFunction	100.01	100.01
nutkWallFunction	100.96	100.01
nutUSpaldingWallFunction	100.01	100.01
nutUBlendedWallFunction	100.01	100.01

Risks

• Compiled with Gcc-7.4.1/Clang-9.0, DPInt32Opt
• No regression issue: 1-pre-fix-yPlus.zip vs 2-post-fix-yPlus.zip (Test: incompressible/pisoFoam/RAS/cavity)
• No changes in the default behaviour.
  • User can now select the new behaviour by useWallFunction=false which is by default true.

changed milestone to %v2012

added bug enhancement integration: openfoam.org labels

changed the description

changed the description

changed the description

All looks good - would also be good to see what we can do to improve the performance of the nutU and nutk wall functions in the y+ < 10 range

I think there is no issue in the prediction performance of nutU and nutk wall functions in the y+ < 10 range. The predictions made by the modified yPlus FO for the same flow field prove this.

The problem is the yPlus prediction functions inside these wall functions:

For example, say nutkWallFunction:

In calcNut():

const scalar yPlus = Cmu25*y[facei]*sqrt(k[celli])/nuw[facei]; is computed and used only for the log region with respect to yPlusLam_.

But, when Foam::tmp<Foam::scalarField> Foam::nutkWallFunctionFvPatchScalarField::yPlus() is invoked, yPlus() returns pow025(Cmu_)*y*sqrt(kwc)/nuw; irrespective of yPlusLam_, hence irrational yPlus predictions for the viscous sublayer.

Potential solution-1: Introduce yPlusLam_ branching/blending into yPlus(), which already exists in calcNut() funcs via blend() func.

Potential solution-2: Deprecate wall functions' yPlus functions in favour of the official y+ definition (y+=y*utau/nu) since i) the definitions do/will not match with the correct definition of y+ ii) blending approaches will make yPlusLam_ branching complex.

resolved all threads

mentioned in issue #1838 (closed)

The issue has been revealed more severe and involved than expected at the submission of this merge request.

To devise and provide a more generic solution, this minimal-extent merge request is now closed, so that all necessary changes from all relevant parties can be reviewed under a single merge request.

Please keep an eye on (or if already submitted at the time of your reading, navigate to) the oncoming merge request.

closed