
[Sponsors] 
May 18, 2005, 06:34 
Yes currently the sgs viscosit

#21 
Senior Member
Join Date: Mar 2009
Posts: 854
Rep Power: 15 
Yes currently the sgs viscosity is isotropic. Most anisotropic models generate stresses directly rather than an anisotropic viscosity and this would typically be implemented as a correction on a simpler model which can be implemented implicitly in the momentum equation like the one currently implemented. Take a look at the implementation of the LRR or LaunderGibson RANS models and you will see what I mean.
> Maybe by deriving the wall function of > "fixedValueFvPatchVectorField"? No, it's not that simple. If you want to use an anisotropic sgs viscosity model you will have to use an tensorialviscosity form of the laplacian in the momentum equation or use the correction approach I suggested above. 

May 18, 2005, 07:33 
Thanks Henry,
my actual probl

#22 
Member
Rolando Maier
Join Date: Mar 2009
Posts: 89
Rep Power: 10 
Thanks Henry,
my actual problem is to implement a LES wall function which should calculate the wall stresses not only depending on the gradient but also on other information (which is equivalent to an anisotropic wall viscosity). Following your suggestion I´ll implement it in the LESmodel itself. Rolando 

June 22, 2005, 06:06 
Hello,
I´ve been using OpenFO

#23 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Hello,
I´ve been using OpenFOAM for a short time. I wanted to do some primitive LES calculations of a repeated channel geometry. I used wall functions at the walls, to save mesh resolution. As SGSmodel I used the Smagorinsky model. As starting solution I used a mean velocity field, which I randomly disturbed. I solved the problem by using channelOodles. The solution procedure seems to work well, but the results are not as I expected them. The velocity of the wall nearest cells seem to fit the wall function. But the velocity profile in the inner field seems to be "too laminar". May it be the cause, that my disturbation routine doesn´t create divergence free fields? (I hoped that the solution procedure smoothes this deficiency). Ralph 

June 22, 2005, 06:13 
Your initialisation is fine, t

#24 
Senior Member
Join Date: Mar 2009
Posts: 854
Rep Power: 15 
Your initialisation is fine, the solution procedure will fix any errors in the initial fields.
What Re are you running? What Courantnumber? What discretisations schemes? How fine is your mesh? Is it graded appropriately? What coefficient are you using for the Smagorinsky model and does it suit the Re you are running with? 

June 22, 2005, 06:21 
In my experience, random pertu

#25 
Senior Member
Eugene de Villiers
Join Date: Mar 2009
Posts: 725
Rep Power: 14 
In my experience, random perturbations will seldom cause transition to turbulence of an LES flow. As you mentioned, the perturbations aren't divergence free and will be swiftly damped out by continuity enforcement.
Aside from this though, you shouldn't really be using wall function to simulate a channel flow. Your nearwall cell's y+ should be below ~20 otherwise your results might be wildly inaccurate and you risk desturbing the nearwall turbulence generation cycle (that accounts for nearly all the turbulence production in a periodic channel flow). LES wall functions in Foam are in fact intended for use solely with the DES approach, which is formulated for seperated flows. So structured perturbations + smaller y+. 

June 22, 2005, 07:26 
Thanks Henry, thanks Eugene,

#26 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Thanks Henry, thanks Eugene,
I just looked at my settings and the results of the calculation. Henry: Re = 40000 Co = about 0.8 I use the same discretisation schemes as in the channelOodles tutorial, except the ddtScheme. I use the CrankNicholson ddtScheme. I use a uniform mesh. As I use a wall function, I didn´t grade it. The cell size is chosen in a way to provide a (time averaged) y+ of 19.5 The Smagorinsky coefficients are chosen as ce=1.65 and ck=0.07 (which should result in a Smagorinsky constant of Cs=0.12). I´m not quite sure if the coefficients fit the Renumber. How do they depend on Re? (I chose Cs=0.12, because I´ve seen it in a publication). Did I choose wrong values? Eugene: I chose a wall function, as I wanted to reproduce a result I´ve found in a publication. My y+ are 19.5 (time averaged). Is there a way in OpenFOAM to create structured, divergence free perturbations of an initial field? Ralph 

June 23, 2005, 04:16 
Hello,
could anybody give me

#27 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Hello,
could anybody give me a hint, if there is a lack in the parameter configuration I´ve chosen. See above (Wednesday, June 22, 2005  05:26 am). As I use a wall function, I don´t use "vanDriest Delta" but "cubeRootVol Delta" instead. Is that the right way? Ralph 

June 23, 2005, 04:30 
Your operating parameters are

#28 
Senior Member
Join Date: Mar 2009
Posts: 854
Rep Power: 15 
Your operating parameters are fine but Eugene may be right that with this nearwall resolution you will not be able to simulate any sensible turbulence genration process there and it dominates the rest of the flow. When you say you want to reproduce published results for this case are you using the same setup? Same mesh? Most importantly same type of LES wallfunction?
The way we produce good initial turbulent flowfields is by mapping from previous solutions. 

June 23, 2005, 04:43 
Thanks Henry,
I´m glad to hea

#29 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Thanks Henry,
I´m glad to hear it´s not the simulation parameters that are the problem. I use the same mesh and the same setup as published. The wall function I use, is not the same, but I think it´s quite similar. What I´m actually doing is to variate the mesh resolution and check the influence on the solution. As I do not have a good previous solution, would it be possible to calculate the solution in a NONperiodic channel with a turbulent inlet boundary condition and map that solution on the periodic channel? Is it correct, that I do NOT use a vanDriest damping for the Smagorinsky model in conjunction with a wall function? Ralph 

June 23, 2005, 04:57 
You could use a NONperiodic c

#30 
Senior Member
Join Date: Mar 2009
Posts: 854
Rep Power: 15 
You could use a NONperiodic channel with a turbulent inlet boundary condition to obtain a starting solution.
Wallfunctions are already a nearwall model and it would be inconsistent to use them with vanDriest damping unless they were designed to be used in this way. If you want to reproduce published results it is very important you use the same wallfunction because you will find your results will be totally dictated by it as the effect of the nearwall region creates the turbulence in flow. 

June 23, 2005, 05:02 
Thanks a lot Henry

#31 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Thanks a lot Henry


June 23, 2005, 05:13 
P.S. Channelflow takes a ver

#32 
Senior Member
Join Date: Mar 2009
Posts: 854
Rep Power: 15 
P.S. Channelflow takes a very long time to evolve, many thousands of timesteps, how many have you done so far?


June 23, 2005, 05:33 
I´ve done 10000  20000 timest

#33 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
I´ve done 10000  20000 timesteps in the different attempts.
But there seemed to be reached a steady state much earlier. 

June 24, 2005, 06:14 
Here is a channelflow perturba

#34 
Senior Member
Eugene de Villiers
Join Date: Mar 2009
Posts: 725
Rep Power: 14 
Here is a channelflow perturbation code. It was used to study the instability of sinusoidal low speed streaks to shear and the subsequent transition to turbulence. As such, it is a "natural" turbulence initiator in that it mimics some of the mechanics of the nearwall turbulence cycle.
perturbU.tgz 

June 24, 2005, 07:32 
I think that is what I was loo

#35 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
I think that is what I was looking for all the time.
Thanks a lot Eugene for publishing it. Ralph 

July 8, 2005, 07:36 
Does anyone know, if there is

#36 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Does anyone know, if there is a Re number dependency of the VanDriest coeffiecients Aplus and Cdelta (used in turbulenceProperties/vanDriestCoeffs)?
Ralph 

July 8, 2005, 08:54 
Probably. Both are known to be

#37 
Senior Member
Eugene de Villiers
Join Date: Mar 2009
Posts: 725
Rep Power: 14 
Probably. Both are known to be somewhat variable and Aplus at is calibrated for a fully developed boundary layer. Sorry I don't have more information to hand.


July 8, 2005, 09:06 
Thanks Eugene,
I´ll take the

#38 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Thanks Eugene,
I´ll take the channelOodles parameters and leave them unchanged. By the way, what does "maxDeltaRatio" stand for? Ralph 

July 8, 2005, 09:33 
maxDeltaRatio is the maximum c

#39 
Senior Member
Eugene de Villiers
Join Date: Mar 2009
Posts: 725
Rep Power: 14 
maxDeltaRatio is the maximum cell to cellneighbour ratio of delta, the SGS length scale. It is used when you choose "smooth" for your method of determining delta size.
Typically you would use a combination of vanDriest, smooth and cubeRootVol. cubeRootVol is the base length scale based on the cube root of the cell volume. "smooth" will smear this distribution so all neighbouring values are within maxDeltaRatio of each other and vanDriest will override "smooth" near the walls with its damped viscous length scales. 

July 8, 2005, 09:36 
Thanks for the explanation Eug

#40 
Member
Ralph
Join Date: Mar 2009
Posts: 40
Rep Power: 10 
Thanks for the explanation Eugene.
I got it :) 

Thread Tools  
Display Modes  

