# A new wall function based on the velocity field----strange result

 Register Blogs Members List Search Today's Posts Mark Forums Read

 September 3, 2015, 22:37 A new wall function based on the velocity field----strange result #1 Senior Member   Huang Xianbei Join Date: Sep 2013 Location: CAU,China Posts: 277 Rep Power: 6 Hi,all: Recently, I was implementing a new type of wall function into OF based velocity. As done in OF, the wall function is implemented based on nut or nuSgs for eddy viscosity. While for non-linear model, it's not applicable. Instead, it is needed to implement a wall function based on velocity so that it is more general. As in wall function theory, the tangential velocity is modified, the idea is to devide the velocity into the wall-normal velocity and tangential velocity: U=Uver+Ut the wall-normal velocity can be obtained by: Code: const vectorField wallnormal=U.snGrad()/max(mag(U.snGrad()),VSMALL); const vectorField Uver = (U&wallnormal)*wallnormal; or you can use directly: Code: const vectorField wallnormal=patch.nf(); Then the tangential direction is: Code: const vectorField Ut=(U-Uver)/max(mag(U-Uver),VSMALL); Now, we should calculate yPlusrefer to the wallShearStress utility) Code: const vectorField patchnormal=-patch().Sf()/patch().magSf(); const vectorField wallshearstress=patchnormal&devBeff; scalar walls=mag(wallshearstress[facei]);scalar utau=sqrt(walls); scalar yplus=utau/(ry[facei]*nuw[facei]);// At last, apply the judge condition based on yplus: y>11, u+=ln(Ey+)/K y<=11,u+=y+ Code: if (yplus>11) { Uw[facei]=(log(E_*yplus)/kappa_)*utau*Ut[facei]+Uver[facei]; } else { Uw[facei]=yplus*utau*Ut[facei]+Uver[facei]; } The code is successfully compiled and run with channel flow case in turtorial(OF211), however, the result is incorrect, the pressure gradient is twice the DNS data Does anyone have any idea? here is the code: USpaldingWallFunctionFvPatchVectorField.H USpaldingWallFunctionFvPatchVectorField.C Xianbei

 September 4, 2015, 22:35 #2 Senior Member   Huang Xianbei Join Date: Sep 2013 Location: CAU,China Posts: 277 Rep Power: 6 Another problem I'm facing is the way I calculate the wallshearstress used for yplus calculation. In my code, the magnitude of wallshearstress vector is used, while if I change it to: Ut&wallshearstress to get the tangential component of wallshearstress, then the calculation break down I don't know how do this occur!! edit1: this is due to that the wallshearstress may be 0 and sqrt(0) is not allowed.However, the result is not correct. Does anyone have similar experience? Last edited by huangxianbei; September 5, 2015 at 22:49.

 September 6, 2015, 07:48 #3 Senior Member   Huang Xianbei Join Date: Sep 2013 Location: CAU,China Posts: 277 Rep Power: 6 Well, I found that it may be not possible to implement such a wall function in OF because that the boundary field of U would not correct during the calculation. Therefore, no matter how I change the formulation of Uw, the result is not affected. So it would only be possible to modify the SGS stress just like the process in nuSgsUSpaldingWallFunction. Another problem is that the gradient operation can be used as following: Code: const fvPatchVectorField& U = lesModel.U().boundaryField()[patchi]; const symmTensorField S = symm(fvc::grad(U)); Is there anyone who can tell me how to use the velocity gradient in a wall function as nuSgsUSpaldingWallFunction? Edit 1:solved, use "fvc:;grad(lesModel.U())" Last edited by huangxianbei; September 6, 2015 at 09:03.

 September 7, 2015, 05:42 #4 Senior Member   Huang Xianbei Join Date: Sep 2013 Location: CAU,China Posts: 277 Rep Power: 6 Hi, all: After further investigation, I found that the reason why wall function on U got incorrect result was that the field never changed! Here is two steps of the calculations: case1: without wall function: Code: Time = 0.2 Courant Number mean: 0.101138 max: 0.121578 DILUPBiCG: Solving for Ux, Initial residual = 0.0152247, Final residual = 8.25801e-06, No Iterations 2 DILUPBiCG: Solving for Uy, Initial residual = 1, Final residual = 3.17434e-06, No Iterations 3 DILUPBiCG: Solving for Uz, Initial residual = 0.0458469, Final residual = 1.27448e-07, No Iterations 3 DICPCG: Solving for p, Initial residual = 1, Final residual = 0.0439203, No Iterations 4 time step continuity errors : sum local = 6.97808e-06, global = -4.07795e-19, cumulative = -4.07795e-19 DICPCG: Solving for p, Initial residual = 0.0352028, Final residual = 5.3519e-07, No Iterations 37 time step continuity errors : sum local = 1.72154e-10, global = -1.35659e-19, cumulative = -5.43454e-19 Uncorrected Ubar = 0.133433 pressure gradient = 0.000336634 ExecutionTime = 0.04 s ClockTime = 0 s Time = 0.4 Courant Number mean: 0.101256 max: 0.121545 DILUPBiCG: Solving for Ux, Initial residual = 0.0386084, Final residual = 1.19873e-08, No Iterations 3 DILUPBiCG: Solving for Uy, Initial residual = 0.333311, Final residual = 1.1725e-07, No Iterations 3 DILUPBiCG: Solving for Uz, Initial residual = 0.055974, Final residual = 4.27262e-08, No Iterations 3 DICPCG: Solving for p, Initial residual = 0.883414, Final residual = 0.0426073, No Iterations 4 time step continuity errors : sum local = 6.88827e-06, global = -2.91983e-19, cumulative = -8.35436e-19 DICPCG: Solving for p, Initial residual = 0.0802868, Final residual = 5.44705e-07, No Iterations 38 time step continuity errors : sum local = 6.15091e-11, global = -1.27589e-18, cumulative = -2.11133e-18 Uncorrected Ubar = 0.133557 pressure gradient = -8.86117e-05 ExecutionTime = 0.06 s ClockTime = 0 s case 2: with wall function on U boundary condition: Code: Time = 0.2 Courant Number mean: 0.101138 max: 0.121578 DILUPBiCG: Solving for Ux, Initial residual = 0.0152247, Final residual = 8.25801e-06, No Iterations 2 DILUPBiCG: Solving for Uy, Initial residual = 1, Final residual = 3.17434e-06, No Iterations 3 DILUPBiCG: Solving for Uz, Initial residual = 0.0458469, Final residual = 1.27448e-07, No Iterations 3 DICPCG: Solving for p, Initial residual = 1, Final residual = 0.0439203, No Iterations 4 time step continuity errors : sum local = 6.97808e-06, global = -4.07795e-19, cumulative = -4.07795e-19 DICPCG: Solving for p, Initial residual = 0.0352028, Final residual = 5.3519e-07, No Iterations 37 time step continuity errors : sum local = 1.72154e-10, global = -1.35659e-19, cumulative = -5.43454e-19 Uncorrected Ubar = 0.133433 pressure gradient = 0.000336634 ExecutionTime = 0.11 s ClockTime = 0 s Time = 0.4 Courant Number mean: 0.101256 max: 0.121545 DILUPBiCG: Solving for Ux, Initial residual = 0.0386084, Final residual = 1.19873e-08, No Iterations 3 DILUPBiCG: Solving for Uy, Initial residual = 0.333311, Final residual = 1.1725e-07, No Iterations 3 DILUPBiCG: Solving for Uz, Initial residual = 0.055974, Final residual = 4.27262e-08, No Iterations 3 DICPCG: Solving for p, Initial residual = 0.883414, Final residual = 0.0426073, No Iterations 4 time step continuity errors : sum local = 6.88827e-06, global = -2.91983e-19, cumulative = -8.35436e-19 DICPCG: Solving for p, Initial residual = 0.0802868, Final residual = 5.44705e-07, No Iterations 38 time step continuity errors : sum local = 6.15091e-11, global = -1.27589e-18, cumulative = -2.11133e-18 Uncorrected Ubar = 0.133557 pressure gradient = -8.86117e-05 ExecutionTime = 0.13 s ClockTime = 0 s As can be seen ,they are exactly the same! That means the boundary field is not modified during the calculation when wall function is applied! How can I make it take effect during the calculation? In the code, fixedValueFvPatchVectorField::evaluate() is called. In the description, this function is to evaluate the patch field. How can it be that the patch field is not changed??? Any idea is welcome! Xianbei

 September 9, 2015, 22:25 #5 Senior Member   Huang Xianbei Join Date: Sep 2013 Location: CAU,China Posts: 277 Rep Power: 6 As U.correctBoundaryConditions() is called in the solver, it's strange that U is not corrected. The correctBoundaryConditions is explained here, it will call the evaluate() function in the boundary condition applied. So in my code, U.correctBoundaryConditions() should call evaluate() and correct the velocity field, however, no change is seen.

September 30, 2015, 03:54
#6
Member

Jason Tan
Join Date: Sep 2014
Posts: 47
Rep Power: 4
Quote:
 Originally Posted by huangxianbei As U.correctBoundaryConditions() is called in the solver, it's strange that U is not corrected. The correctBoundaryConditions is explained here, it will call the evaluate() function in the boundary condition applied. So in my code, U.correctBoundaryConditions() should call evaluate() and correct the velocity field, however, no change is seen.
hello,
I read your thread and I know what you research is about DNS, and so do I, Now I face a hard question is that how do calculate y+ in DNS,you know there is no any codes about calculating y+ in DNS. I saw your answer in other thread about y+ about DNS,but you didn't show how,Can you show me? my email is : tzqfly2009@163.com

September 30, 2015, 07:52
#7
Senior Member

Huang Xianbei
Join Date: Sep 2013
Location: CAU,China
Posts: 277
Rep Power: 6
Quote:
 Originally Posted by tzqfly hello, I read your thread and I know what you research is about DNS, and so do I, Now I face a hard question is that how do calculate y+ in DNS,you know there is no any codes about calculating y+ in DNS. I saw your answer in other thread about y+ about DNS,but you didn't show how,Can you show me? my email is : tzqfly2009@163.com
Hi:
For a DNS calculation , a simple way to achieve this is using the velocity gradient. If the mesh is fine, you can use gradU=u1/y1, here, 1 represents the value at the first node near-wall. tau_wall=nu*gradU, u_tau=sqrt(tau_wall), so you can calculate the y+

Xianbei

 Thread Tools Display Modes Linear Mode

 Posting Rules You may not post new threads You may not post replies You may not post attachments You may not edit your posts BB code is On Smilies are On [IMG] code is On HTML code is OffTrackbacks are On Pingbacks are On Refbacks are On Forum Rules

 Similar Threads Thread Thread Starter Forum Replies Last Post uhlondon STAR-CCM+ 3 June 25, 2015 08:44 alfaruk CFX 8 December 3, 2013 17:51 nennbs OpenFOAM Native Meshers: blockMesh 7 April 17, 2013 05:42 ivanyao OpenFOAM Running, Solving & CFD 1 October 12, 2012 09:31 stephane baralon Main CFD Forum 11 September 2, 1999 04:05

All times are GMT -4. The time now is 14:23.