The phenomenon you've seen is common and correct. If you insist using one model to cover laminar, transition and turbulent, you can choose RNG model.

You are doing fluid flow analysis inside a pipe flow at different speeds, that ranges from laminar to Turbulent flow. Any model depends number of empirical constants. You are doing internal flow therefore depends number of parameters of the wall. Therefore it is better to summarize the constant required in the model based on the surface as well as boundary conditions, pressure gradient etc.  

Thank you for the quick response. I have another question please.

Assuming that I am using the RNG model, but then should I actually keep changing the mesh refinement per each Re in order to get the optimal Y+ value? If so then, again, I do expect that changing the mesh could actually have an influence on the velocity field as well. So I can not be sure that a different behaviour that I observe is only due to the different inlet conditions, but also due to a different mesh?

I thought of running the mesh independence test per each simulation but this is just too impractical, specially that I perform 3D simulation as my geometry is not axis-symmetric.

Thanks

For laminar flow, you probably can just used one set of mesh. Of course, mesh dependency has to be checked for that mesh. In transition and turbulent, your mesh distribution will depend on the Reynolds number. That's one reason that turbulent flow is so difficult to solve.

Hello,

why don't you use the k-kl-omega model instead. This model is made for solving laminar and turbulent flows, as well as transition (to some extend). Otherwise, if you use a turbulence model and you start with having tke in your initial field, you will end up with a turbulent flow, although your Reynolds number considered referrers to a laminar flow. There are some papers online dealing with/describing the problem of using RANS models)

Further information on the k-kl-omega model can be found in cfd-online. As far as I know it is correctly implemented in of 2.3.1.

Theoretically you have to check mesh independence for each Reynolds number. However, resolving the wall with y+

Thank you Tobias. As far as I know that the mesh should be arranged such that the first cell point does not fall into the viscous sublayer (using the wall function). Accordingly, I thought the optimal y+ value should be 20 < y+ < 30. no?  Or do you recommend not using the wall function at all in this case?

Thanks,

Hi Abdel, 

Generally any RANS model can solve the laminar flow almost accurately. The problem gets complicated as it approaches turbulence. Some models such as kOmegaSST or kEpsilon will solve turbulent flows well, while transition flows are solved better by other models. You will need to run some tests yourself using those models and compare which models give you better results overall. 

Regarding the yPlus values, if you are using wall function models then you want to have y+>15. And If you are using low-Re or models with enhanced wall treatment, then you will want to make sure y+ is

Thanks a lot Nikesh, you made things much clearer for me now. I am simulating pipe flow but I do actually have some zones with flow separation. I think that is one reason why I could not get good results using the wall function; pressure drop was very much overestimated.

I have some doubts about the boundary conditions of the turbulence fields in case I am not using a wall function. I assume it should be a fixed value of Zero at the wall, no? Also are the values at the inlet boundary of great importance? I calculate them using some give formulae.

Do you think that LES would be able to solve the problem of simulating laminar, transition and turbulent flow using the same model?

Thanks

You are looking for a RANS model that can do transition to turbulence.  A large number of them can do this (including some low Re k/e models, v2-f model, etc).  

Practically, there is no reason a RANS model cannot be used for laminar flow.   Put in a small k (or large eps) and you will get a small eddy viscosity that is much smaller than the laminar viscosity.  There are RANS models that will automatically turn off and do DNS if a DNS mesh is provided.

Physically, RANS models contain enough physics to make the distinction.  As the ability to predict B.L. transition demonstrates.   Very few models are tuned to predict it well (this is not a typical tuning case) - but most of them do it.

Mesh independence of turbulence models is a different issue all together.   Some models (like K-omega) are renowned for not being mesh independent.  In theory, there is no reason 2D RANS cannot be mesh independent.  Actually, 3D time-resolved RANS should transition to unsteady solutions and  chaos (just like Navier-Stokes does) when the mesh is fine enough to allow those types of solutions (because those capture the most physics for the mesh provided).   So the idea of mesh independence for a turbulent flow is a complex one.  I have two papers showing mesh independence of RANS models as they go from RANS to LES to DNS looking solutions/behavior.  .   

@Perot Thanks a lot for the comprehensive reply, I very much appreciate it.

I am interested to read the papers you mentioned above. Could you please give me their titles? As I found that you have 81 publications.

Thank you

J. B. Perot & J. Gadebusch, A Stress Transport Equation Model for Simulating Turbulence at any Mesh Resolution.Theoretical and Computational Fluid Dynamics. 23 (4), 271-286, 2009.

J. B. Perot & J. Gadebusch, A Self-adapting Turbulence Model for Flow Simulation at any Mesh Resolution, Physics of Fluids, 19 (11), 115105, 2007.

Everything is on:

http://www.ecs.umass.edu/mie/tcfd/Publications.html

if researchGate doesn't have it yet.

Have you considered a LES approach? 

@Andrea, I am considering both to verify my results