Dear Experts,

Is it possible to show in a single simulaiton to show all three types of flow (laminar, transition and turbulent) as can be seen in a Reynolds apparatus. How can I show this using CFD simulation?

Thanks,

Harendra

Hi,

I am not familiar with a Reynolds apparatus, but I assume you are trying to model laminar, transitional and turbulent flows. This can be done in CFX using DNS or LES techniques (need a very fine and carefully constructed mesh, expect long run times) or using the transitional turbulence model. I have no direct experience with the transitional turbulence model but I believe it is designed to model what you are looking at.

Glenn Horrocks

Hi Glenn Horrocks,

Would you please tell me how to activate DNS using CFX-5.7? i did not find it.

Thanks.

-Harendra

Hi,

How do you activate the DNS model????? By definition DNS requires no model!

DNS simply means you resolve every fluid motion right down to the smallest turbulent structures. Therefore you don't need a model at all, the laminar Navier Stokes equations are all you need. The trick is to have a fine enough mesh and accurate enough discretisation to be able to resolve these microsopic structures, yet still have a model which fits on a computer and can run.

DNS is not for the feint-hearted. A lot of issues need consideration over a normal RANS simulation. If you are thinking of doing DNS simulation, have a look at "Turbulence Modelling for CFD" by Wilcox, and chase up some of the references from his discussion on DNS.

Regards, Glenn Horrocks

Actually, I haven't seen anybody using DNS in CFX. Anybody in this forum with experience on that subject (DNS with CFX)? I honestly do not see the point. DNS requires explicit timestepping to my knowledge with small timesteps. CFX is an implicit solver so you are just wasting time with the overhead. But anybody can correct me on this.

Hi Bart,

As you say, CFX is not optimised to do DNS. To have a decent go at DNS you need a solver optimised for the purpose. However, there is no reason why you could not use CFX to do DNS, you will just have to be very patient when waiting for the results!

The CFX 5.8 beta blurb says there has been some changes to the transient timestepping such that it is now possible to get second order accuracy with only one iteration per timestep, that is pseudo-explicit timestepping. It should help the masses of people out there doing DNS with CFX!

Maybe someday when the boss isn't looking I might try to reproduce the DNS results of Kim, Moin and Moser from 1987. They used meshes of up to 4 million points and that is quite doable on a modest cluster these days.

Regards, Glenn Horrocks

Hi,

if you are serious about DNS/LES/DES I would suggest not doing it in CFX. I have seen quite a number of people who have tried but found something missing plus with the coupled solver and iterative time stepping it is unnecissarily slow....great for big time steps...wrong deal for tiny time steps!

There are lots of little details that have to be right in the solver and the post-processing. For example the outlet and opening boundary conditions in an enclosed domain must be non-reflective for ALL wavelengths, some appropriate initialization to stir-up the fluid must be made, all the appropriate averages of variables must be working. I know of people who got started and then a show-stopper was missing.

Don't be convinced that you can sort it out with CCL or User Fortran. There are other codes that have done a lot more DNS/LES work...try Fluent, CFDRC, etc first!

Regards,

Bak_Flow

I don't see how Fluent has any advantage over CFX for doing DNS/LES simulations, and there is certainly no advantage to running Fluent for DES.

To best of my knowledge, the only advantage may be if you were running the Fluent density based explicit solver. If that's the case, however, you will be limited to high-speed flows and using a different solver means you give up the majority of the models.

Not to mention the fact that DNS completely impractical from an engineering point of view (something you seem to care about in other posts) and LES isn't much better.

DES, on the other hand, is very practical and CFX has done a lot of work in this area, including a DES-SST combination.

-Guy

Hi,

allow me to explain my views here...you will get your turn!

1. DNS is (currently) completly impractical for High Reybnolds Number Engineering flows...there are many applications for which it is practical....micro-fluidics, highly viscous polymer and oil flows and mixing, some biological applications, some environmental applications. I merely included it in the set of DNS/LES/DES as flows for which atleast some part of the details turbulent structures are solved...to be complete!

2. The implicit pressure based coupled solver is all that is offered now in CFX-5. This technology is advantageous for large time steps..usually to evolove quickly and physically to the steady state. There is a cost for this...right? There is a cut-off point below which, running smaller time steps require less time per iteration with a segregated-implicit or as you point out an explicit solver as compared to the time required for the coupled set assembly and solution...right??? Most DES/LES work I have seen reuires time scale resolution smaller than this limit and this is faster done with Fluent. Fluent also has non-iterative time stepping and CFX is still working on that.

3. As you point out I am very practical..thank you;-)..and the code that has done the most work and been fully exercised in this area is Fluent, CFDRC, and maybe some others. I don't want to hit bugs or limitations when I could be doing real work...have you run both codes?? The only work I have seen done with CFX has been done by CFX themselves.....why is that?

So I just wanted to point these things out in response to the original question.

XXX

Bak_Flow

I'll concede that there are advantages to running a DNS simulation in an explicit code, but I don't think you can say for sure how the assembly process compares between Fluent's segregated solver and CFX's coupled solver.

As for DES, the whole advantage to DES is that you can take larger timesteps by reducing the LES region to zones with larger turbulent lenght and timescales and avoiding using LES in the near-wall region, so clearly CFX is better. I think you also have to consider how much development effort has been made in this area by CFX's Florian Menter.

As for people using it, I don't have a good grasp of how many CFX or Fluent users are running LES simulations. However, Fluent is more likely to have difficulties getting a solution for steady state problems and encourage users to use LES instead.

Frankly, the proof is in the pudding. Rather than blankly stating that one should use another code, I would encourage prospective users to investigate all of these codes and concentrate on getting meaningful engineering solutions, rather than focusing on a particular modelling method (i.e. only run DES/LES/DNS if you have to, in that order).

-Guy