I am trying to model unsteady confined vortex flow using Fluent for my postgrad [i.e. structured grid].But I am having problems implementing the deforming grid model for 3D rotationally cyclic conditions. So if anyone has any information on 3D rotationally cyclic deforming grids, turbulence models and numerical schemes for both low and high speed swirling flows it would be very much appreciated since this area of CFD is relatively new to me.

Thanks

I did my PhD on the simulation of trailing vortices and the development of turbulence in a vortex. I can tell you straight away that usual (statistical) turbulence models such as k-epsilon are definitely not adapted to rotating flows as they greatly overpredict the levels of turbulence. What I would suggest is for you to perform a DNS or LES of your phenomena, depending on your Reynolds number. If you have access to a code in cylindrical coordinates it would be even better.

Fabien

First of all I must thank you for your time and advice.

However I had a fair idea that the isotropic k-e was fairly useless in such a flow. Also fir it's inability to model coriolos forces associated with swirling flow. However as far as I am aware there are 3 turbulence models available

1. The K-E turbulence model {can be used a initial answer

to speed convergence)-(patched) 2. The RNG K-E turbulence model 3. The R.S.M

However I an not two sure what DNS and LES are and why there are used. However if any info., can be ascertained in terms of good book's, please do not hesitate, to send me the reference.

Thanks oliver

Oliver:

Well, DNS=direct numerical simulation LES=large-eddy simulation. DNS is just a simulation of directly discretized Navier-Stokes equations without any turbulence model. Of course it is limited to low Reynolds numbers, otherwise there can be accumulation of turbulent energy at the cutoff scale which is basically the scale of the local grid size. This accumulation only happens if there are actually scales of turbulence which are smaller than the cutoff scale. In that case the grid acts as some sort of low-pass filter on turbulence. Then you have to use LES which includes what is known as a subgrid-scale model. These models are usually just refined versions of mixing-length models (Smagorisnky) but there are also "dynamic" versions (such as Germano) which adapt to local levels of turbulence. At Onera we used another dynamic approach which was as effective and much less expensive as far as CPU than the Germano model. I do not recall what type of commercial software you are using, by I believe that codes such as StarCD and Fluent now include the Smagorinsky model. However, I must warn you: if you really want to get good results in your simulation, I would suggest that you use a dedicated finite-difference based DNS/LES code instead. There now are compact hermitian schemes (Lele) which are very precise (6th order in space) with spectral-like resolution. Swirling flows are very complex flows and require very very tight grids in order to capture all the phenomena, especially in situations of transition. If you really must use a statistical NS code, make sure you are using a RSM model WITH corrective terms to account for the effects of CURVATURE. Otherwise you will get an over-prediction of turbulence levels just like a k-epsilon. If you want more information, feel free to e-mail me.

Oliver,

What you are trying to do should be possible with Fluent. If you give me some more details I can probably help you out.

Regards,

Jonathan

Are you saying that Fluent has a turbulence model which works well in swirling flows? The RNG model is slightly better than the standard k-epsilon, but you can't seriosly say that it in general works very well in swirling flows? Or are you refering to a different model?

Do you have the option to include a Richardsson number corrections in your turbulence models? Do you offer any non-linear models? Do you have any RSM based models (EARSM or differential RSM)?

I agree! A Richardson correction (CURVATURE!!!) is absolutely necessary if you are using any type of statistical turbulence model. I still think that only DNS/LES can treat unsteady and developing phenomena in swirling flows. With usual NS codes (including commercial software!) you won't be able to get the same results.

Yes, I have to second that opinion, swirling flows is a typical example of flows where most turbulence models fail completely and very few, if any, models perform satisfactory. The Richardson number correction, or something similiar, is an absolute necessity if you are using a standard k-epsilon model. RSM models have the potential to model the curvature effects better, but they are often too difficult to use numerically. I agree that LES is an interesting and promising approach, but second moment closures (RSM) is an alternative, or don't you agree?

Yes we can do swirling flows in Fluent. We have several examples of swirling flows including highly swirling flows such as vortex seperators and several of our customers model these flows often. We usualy use a RSM turbulence model for this type of flow. RNG won't handle very highly swirling flows. We also have LES available.

There has been a very interesting study published by O. Zeman in 1995. I can't remember if it was in JFM or Phys. Fluids, but he studied turbulence in a 2D vortex (Lamb-Oseen) with both a k-e with Richardson correction and a RSM model with curvature terms and 3-rd order closures which also included Richardson correction. The conclusion was that the Richardson-corrected k-e model predictions were way too high.

I'd like to say that most phenomena in swirling flows and particularly in vortices (aircraft trailing vortices, flap vortices, submarines and ships) are unsteady by nature. I showed in my PhD thesis that turbulence cannot be sustained in a trailing vortex, be it 3D or 2D. If I hadn't had the kind of tools I used, i.e. a 3D finite difference LES code with a Cray C-98 for computing power (I had a 200x200x51 grid and used hundreds of hours of CPU), I would never have been able to obtain those results.

A RANS model could work even if the main flow is unsteady, provided that the time-scales (turbulent and main-flow) are well separated. Do you think that this is not the case? i.e. do you think that the unsteady effects in swirling flows are important for the turbulence modeling? Have you got any physical explanation as to how this happens? This is very interesting ... especially for all researchers that are working hard on adapting RANS models to swirling flows ;-)

Ah, I didn't know that you have a working RSM model in Fluent. Which model is this? Does it work well numerically? How much more expensive is it than a k-epsilon model?

I would guess that LES is a bit out of reach for most industrial applications today, considering the typical Re-numbers and that you have to do a full transient 3D simulation, don't you agree?

One last observation: Going to RSM or LES introduces a new problem in swirling flow predictions - how do you accurately set the inlet conditions? I've seen several papers where they have examplified this problem. Often the predictions become very sensitve to how you prescribe your inlet profiles when using RSM or LES.

I'm glad to hear that you say that the RNG model does not work for higly swirling flows. The RNG theory is basically a theory for the intertial subrange and it can not be expected to improve the predictions for swirling flows when the reynolds stress tensor becomes anisotropic, although the RNG approach does give rise to terms in the epsilon equation which are similar to the Richardson number curvature correction.

We have had the standard Gibson/Launder model for some time. We now also have the quadratic pressure-strain model of Speziale, Sarkar and Gatzki. I haven't used the latter much yet so can't say much about it, but certainly Gibson/Launder is very stable. The extra expense seems to be almost entirely from the greater number of equations rather than the stiffness of the equations set, at least in most cases.

I agree LES is very expensive for industrial type problems because of the fine mesh and time step requirements for such problems.

We can set boundary conditions of all variables can be made to vary in space and time as required either using profile files or user defined subroutines. It is knowing what these boundary conditions really are that is often the tricky bit.

The RNG modification of the epsilon equation does help in swirling flows up to a point but is not enough for highly swirling flows. As you say anisotropy of the stresses then becomes to great for RNG to cope.

It's always fun to fool around with a commerical code. My suggestion is : take the steady-state case first to work out the geometry, boundary condition set-up . Then you can easily study the effect due to the numerical schemes, turbulence models systematically. All you need to do is a "click" to select the option and set the number of iterations. It's so simple that almost anyone can do that. Someone has a name for it , it's called "CFD Mechanics". You simply follows the instructions and get the work done right. Based on my experience of using some else's codes, the hard part is to determine which result should be used in the presentation. If you present a paper in a meeting and someone ask you a question about the power-law scheme used in the paper, do you think you can answer the question consistently with the code listing? You need to do some home work exercises with commercial codes first before you actually try to solve your problem. If you don't know exactly how the methods ,the models, and the schemes are implemented in the code, then you will have a hard time to explain your results to the audience. You can't just say: I used code-A , selected options B and C, now I have results D and E. The user's guide says B is power-law and C is second-order, so I don't know which one you prefer. I think it's o.k. to use any code to solve a problem, but you need to have the ability to verify that the second-order option is actually second-order accurate. This is a problem with most commerical codes. You have to take their words for it.

Tornado definately is a swirling flow. It's also fairly stable, very slow moving. In laboratory simulation, the tornado has shown to be fairly stable. Has any one computed this type of flow using two-equation models? The other type of swirling flow is the dump combustor flow with swirl. Many people have studied this type of flows using two-equation model. I don't know whether there is a big white elephant in a swirling flow or not. DNS and LES are transient in nature. But after several hundred hours of calculation, it's going to be very difficult to keep the numerical solution in phase with the real flow. Actually, the two-equation model is not hopeless at all. At least, the turbulence kinetic energy equation is o.k.. The one which requires special attention is the dissipation equation. The equation was overly simplified right from the begining, that is there are many terms missing. Under simplified conditions, these terms don't have any effect. That's why the equation was simplified. There are two approaches to include the missing terms: one is to model the effect separately in a new term ( added to the equation), the other is to look for the source of missing term ( the form ) and then try to model the identified missing term or terms. ( I have looked into both approaches ten years and had done some calculations. )

Dear Prof. Larsson,

How are you, Sir ?

Perhaps it may not be of decency for people like me from commercial CFD software vendors to take every chance here in this forum to make a sales pitch about their products. But I couldn't resist coming on here for this one. Please forgive me.

Fluent Inc. has offered second-moment closure models since late 80's (ASM and DRSM later). As you can imagine, we had to offer them to our users because our major client base was initially in power generation/environment industry where they often ought to deal with cyclones, swirl combustors, etc. the kind of applications warranting use of turbulence models other than k-epsilon model. So I can say we, as a company, have hands-on experience in the second-moment closure models.

Presently, we have a DRSM (Differential Reynolds-Stress Model)in both FLUENT 4.4 (4.5)and FLUENT/UNS4.2 (FLUENT 5). Our DRSM implementation seems to be pretty robust. And it's nice to see the DRSM implemented on our unstructured mesh based solver. I know some people out there are always suspicious whenever we commercial software companies say anything about our products. Good reason to be suspicious. But we do sometimes cool stuff !

By the way, you've sounded to advocate much nonlinear EVM. Do you really think nonlinear EVMs are the way to go ? I've seen a number of publications where they employed NLEVMs. However, I find most of the validations done for backward-facing step, airfoils, etc. for which some two-equation models do bequally good jobs. The stress-driven secondary flows in ducts with sharp corners predicted by NLEVM doesn't impress me, because no one cares when it comes to industrial flows. Do you know of any computations using NLEVM for highly swirling flows in cyclones ?

Best regards, Sung-Eun Kim

You're very welcome to respond to queries here, although the discussion tends to get a bit boring if every call for help from a potential customer generates a bunch of "we have the best product on the market, lets continue the discussion in private emails..." from every CFD code vendor.

About NLEVM and EARSM models. I do think that they offer a physically sound approach to handling several major problem with classical Boussinesq based eddy viscosity models. I'm thinking about stagnation flows, immediate response to curvature etc. Surely they aren't a general solution, and predicting secondary effects still remains a big challenge. My experience from using different variants of non-linear models is that most are not much more numerically difficult to use than linear models. The same is definitely not true for differential RSM models, and I do believe that most DRSM models still have a big weakness in their handling of walls and low-Re effects. Seems like the turbulence modeling community is heading towards LES now though...

About swirling flows - not really my area so I'll have to pass on that question I'm afraid. Sorry.

What do you think will be (or is) the most promising development in turbulence modeling for industrial applications within the next few year?

[pramod]

I am doing analysis of turbulent swirling flow. Can anybody tell me which model is used in this analysis? Which is the best software package to solve this type of models? Please give me idea of doing this..

Regards,
pramod