Possibly we should first make clear whether we look at this topic (i) as an engineer, who mostly deals with extremely complex geometries and many side-effects in addition to turbulence and is typically payed very well but has hot applications in his neck which leave no time, or (ii) as a physicist who concentrates on the most simple and principal questions of turbulence, e.g. the free decay of turbulent kinetic energy and r.m.s. vorticity in ideal conditions, at frequency and/or wavenumber spectra in ideal conditions, at effective eddy diffusivities in extreme simple cases, or at the behaviour along most simple walls: planar walls. For asymptotic Reynolds numbers, (ii) is solved since 2005 using no single measurement (it took a century), but giving hints how to construct solutions for the much more complex world of the engineer.

Therefore, the question raised by Ahmed Abdelhameed, from hopefully sunny Port Said, is clearly NO - similarly as the two different pictures of quantum mechnics by Schrödinger and Heisenberg do not really help engineer a smart phone; but these pictures/eqns help to principally UNDERSTAND quantum mechnics and the rules how conventional electrics needs to be modified at the very small scales of atomic/moleculear grids to "predict" measurements in these damn semi-conductors correctly.

The turbulence theory of 2005 mentioned above is good for teaching students of engineering or geosciences etc. to get them UNDERSTAND turbulence as a phenomenon; but elaborating specific solutions/modifications is their job : -). So any version of K-omega is principally worth detailed comparisons with dedicated experiments. Real damn expensive work ...

To answer correctly to your question, let me change the "turbulence model" to a "proper formulation" statement.

RANS formulation can never (by definition) produce a unique turbulence model working for all flow problems.

Conversely, LES formulation could theoretically do, providing that a correct grid and a suitable model (such as the mixed dynamic SGS) is used. However, the general BCs. can be a problem also in LES.

Hi,

Some turbulence models are more suitable for strong turbulent flows whereas others are more suitable for transition flows. For example, k-epsilon and k-omega models. This is clear in the governing equations for both models. In some particular cases of laminar flow and you tried to try to use any of turbulence models, k-omega model may give a close solution to the laminar one while k-epsilon model may be away.

This could be a part of the answer. Generally, the governing equation of each of the turbulence models are different.

Regards

Wael

Atleast for low cost (computational) turbulence models like RANS, it's not possible. Even for a single turbulence model like k epsilon, there are many sub classes, each containing empirical constants and assumptions fine tuned for specific types of flows. It's one of the main reasons why you need to validate your RANS results with experiments before you try to use it for any further study. However, for models like LES and DNS, it is a possibility.

No, I am very sure it is practically not feasible due to the different boundary conditions as well as the specific assumptions upon which each model is found valid.For example estimating turbulent heat fluxes using different models, I discovered huge discrepancies among their end results. This of course is expected of models due to the different conditions each model is expected to yield useful result(s).

Reynolds number and boundary condition are two main factors that determines the turbulence model in any flow condition. So better to examine appropriate model for different fluid flow.

I think the clue behind using more than one turbulence model (proper formulation) statement is the physics of viscosity, i mean that for increasing Reynolds number the characterization of the flow changes (the difference between DNS and RANS). More work is needed to investigate the dark phenomena of viscosity.

Molecular viscosity is assumed to be not flow dependent, we could debate if the Newonian model is generally valid. Turbulent viscosity is just an old idea to stabilize a numerical simulation, we should definetley debate about that. Just think about the assumption that a non-linear unresolved convective flux (coming from hyperbolic part of the equation) is transformed in a non linear diffusive term (parabolic part).

But turbulent viscosity means nothing if one does associate the formulation. RANS and LES both introduce a similar viscosity but the physical modelling is totally different.

To just add and give a different perspective,

Note RANS and LES and even Detached eddy model (DES) are solving a modified NS equation than the original NS equation as in DNS.

RANS > DES> LES in descending order of how different are they from the original Navier Stokes.

In all turbulence models closure is extremely crucial to its performance and need not be eddy viscosity based approach for LES.

In RANS, mostly eddy viscosity based approach is used.

In LES, only the small scales having less information from the boundary are used for closure. So, one LES model can still be used for a wide range of flows. But still cannot be universal

RANS, on the other hand models even the larger scales influenced by geometry and boundary, so have become more problem

dependent.

Some advanced methods like Reynolds stress models which calculate the spatial temporal evolution of the turbulent stresses with certain source and sinks as closure are aimed to perform better. They capture some physics better from the closure of Reynolds stress models and hence address wider range of flows. But, " one-model" for all flows is quite ambitious since we cannot address the dynamics of the flows at unresolved scales quantitatively at a single go.

Ahmed Abdelhameed

You raise an acute problem. Methodically we are in a similar situation like Newton "some days ago": To test his THEORY of gravitation in practice, he took our beautiful planet Earth as a point mass. This computation could still be made manually, with success. Today such computations are made with much more detail, but the THEORY is the same: Newton's law!

So in turbulence we are in a similar situation: we have a THEORY (resting on Kolmogorov 1942, Baumert 2005, 2013, see attachment) which functions for idealized practical problems like e.g. isothermal flow along a plane surface, and it PREDICTS the Karman constant as 1/SQART(2*Pi) =0.399 without using a single measurement.

This theory may now be advanced for the case of non-isothermal problems or cooking (convective flows), but nobody has an interest. Meanwhile I learned why: the reason was explained by Sreenivan on the BACKPAGE of APS News of June 2018. And I repeat here what Falkovich and Sreeni already stated in 2006: "The diversity of problems in turbulence should not obscur the fact that the heart of the subject belongs to physics." I thus encouraged a colleague to think about turbulence production by surface waves in oceans and lakes, and I encouraged people to think about a finite Reynolds number version of my theory. Both problems are actually not too complex but one needs to have the right inspiration. If members of this forum are interested, I could help more directly via baumert (at) iamaris.org

Strictly speaking, it is not suitable. The turbulence model has its scope of application. Some are only suitable for laminar flow, and some are only suitable for weak or moderate inverse pressure gradient separation. Different turbulence models demand different y plus. Actual engineering problems often have various complex shapes, often with laminar flow and weak separation and large separation areas. And grid generation will cause a large difference in y plus near the wall surface. In some places, it obviously does not meet the requirements of the turbulence model for y+ on the grid near the wall. Although it seems most reasonable to mix various turbulence models to adapt to different states, it need too much effort and is difficult. Maybe you can use the turbulent stress model adaptively. But How to use it is more troublesome.

Probably not. Since turbulence has a

lot of vortexes with different scale. For large scale vortexes，they have strong dependence on boundaries and therefore，it is hard to build a general model for RANS to ac curately depict them without considering the specific flow boindaries.

Dear @Jianming Liu, you added this answer: " ... Some are only suitable for laminar flow, ...". A TURBULENCE model for LAMINAR flow? Did I correctly understood???

Dear Jianming Liu

Laminar flow applications does not need turbulence model , DNS is enough .

what we hope to find is just ONE turbulence model for turbulent flow cases.

Regards

The physics of turbulence is still unclear. There are hypotheses. And a lot of engineering models.

there is a lot of confusion I frequently read about the use of "turbulent model" decoupled from what acts in the "formulation" by means of the turbulence model....

Laminar flow does not need a model but we must be aware of the fact that often laminar, transitional and turbulent regime coexist in the same flow problem. The correct question should be about a proper formulation that can be used indipendently from the assumption on the regime. While for the RANS formulation the answer is negative, that is the turbulence model affects a laminar regime, in LES formulation the answer is different. Proper SGS model can be used such that a laminar regime is preserved automatically simply because the effect of the model vanishes. That is possibile with the dynamic formulation as well as with the dynamic mixed models. However, that does not mean that a universal model exists.

All turbulence models are calibrated against only few cases of experimental or numerical data. A model performing well for a particular case of problem may or may not perform well for other problem of practical application.

So turbulence models can not be universal.

It is not possible to use one turbulence model for all practical fluid dynamics applications. This is because each model got developed pertaining to a particular application and has been seen that it has failed severely while it was tried to replicate another physics. For example, while analyzing down-force for the front wing of motor race cars it was found that the SST k-omega model could replicate the physical observations whereas the RNG k-epsilon model could not. Similar results (differences with respect to available models) are well available in literature and the choice of having a unanimous model for all type of practical applications is yet to be arrived at.

In theory, a physical model of turbulent flows can be constructed on the basis of moment methods for solving the Boltzmann equation. Look my article "The internal structure of turbulence".

DNS, it’s the model of Rozhdestvensky B.L., may be completely physically inadequate.

With regard to other models, we can say that they resemble the theory of caloric.

Possibly we should first make clear whether we look at this topic (i) as an engineer, who mostly deals with extremely complex geometries and many side-effects in addition to turbulence and is typically payed very well but has hot applications in his neck which leave no time, or (ii) as a physicist who concentrates on the most simple and principal questions of turbulence, e.g. the free decay of turbulent kinetic energy and r.m.s. vorticity in ideal conditions, at frequency and/or wavenumber spectra in ideal conditions, at effective eddy diffusivities in extreme simple cases, or at the behaviour along most simple walls: planar walls. For asymptotic Reynolds numbers, (ii) is solved since 2005 using no single measurement (it took a century), but giving hints how to construct solutions for the much more complex world of the engineer.

Therefore, the question raised by Ahmed Abdelhameed, from hopefully sunny Port Said, is clearly NO - similarly as the two different pictures of quantum mechnics by Schrödinger and Heisenberg do not really help engineer a smart phone; but these pictures/eqns help to principally UNDERSTAND quantum mechnics and the rules how conventional electrics needs to be modified at the very small scales of atomic/moleculear grids to "predict" measurements in these damn semi-conductors correctly.

The turbulence theory of 2005 mentioned above is good for teaching students of engineering or geosciences etc. to get them UNDERSTAND turbulence as a phenomenon; but elaborating specific solutions/modifications is their job : -). So any version of K-omega is principally worth detailed comparisons with dedicated experiments. Real damn expensive work ...

Dear... Professor Helmut Ziegfeld

Thank you so much for conclusive and constructive declaration about turbulence.However, may be some one like Printdle(Boundary Layer Theory) has the ability to figure out the clue of turbulence leading to ONE turbulence model.

Best wishes and Regard

Dear Ahmed,

the point is boundary conds. With a one-eqn model like Prandtl's, it's impossible to treat boundaries correctly. I just explained this issue to a Chinese colleague: the eddy viscosity/diffusivity is governed by TWO turbulent scales: length and time. And you need therefore separate eqns for the two because their boundary conds follow different rules/philosophies. That's the reason why Prandtl's scholars introduced two-eqn models, although incorrect choices like K-eps etc., but at least two state variables allowing for separate boundary conds.

I suppose, exactly like the question was stated, you indeed CAN USE just ONE turbulence model for all applications. That is entirely fine. But you will see errors of different quantitative scale for the different applications with that selected just one turbulence model and the real question is, whether you can live with these error ranges for the quantities of your interest? Sometimes remaining model errors will be small, for other applications they might become larger. Are they then still acceptable for your underlying purpose of doing CFD?

Good 2-eq. turbulence models to use in this sense of the question (possibly as a not too bad starting point for all following investigations) are the SST or the GEKO turbulence models.

Best regards,

Dr. Th. Frank.

Dear Ahmed & Sergey, I believe that the MAJOR CRITERION to discriminate between all of the many different turbulence models is: whether the model under consideration is able to PREDICT universal constants describing the statistical behaviour of turbulence in space and time.