LES - Science topic

Different simulation tasks may require different turbulence models. I suggest, you look for literature on your application (or similar applications) as a reference case. And either rely on previous work or do test runs with different models to see which one gets you closest to the reference case.

You need not only to compare your result with the experiment, but also to have a comparison with DNS. If you are applying a wall modeled LES, I would strongly suggest to also try or compare the results with a wall-resolved LES.

Here are ERCOFTAC classic database which might be suitable for your validation:

Thank you very much Mr. Boubaker Dhehibi

First of all, what about the textbooks you are already considering to understand LES?

In general you have to consider first the idea of the evolution of LES into DES and then into DDES:

P.R. Spalart, S. Deck, M.L. Shur, K.D. Squires, M.K. Strelets, and A. Travin. A New Version of Detached-eddy Simulation, Resistant to Ambiguous Grid Densities. Theoretical and Computational Fluid Dynamics, 20(3):181–195, 2006.

As stated by Filippo, in my experience, comparing the velocity profile is never enough in LES, especially for implicitly filtered LES in the channel flow. For example, with certain codes and discretizations, a mass error due to the pressure discretization can directly affect the mean velocity profile and make it look good while all the other quantities would signal a completely wrong solution.

For what concerns the wall and SGS model, in my opinion, a single combination doesn't allow to really take effects apart. Also, there isn't much around on the Werner & Wengle model, which makes it more difficult to understand.

Are you using a specific code or is this an in-house solver?

More generally, you want to build confidence on your tool for this case, but you are exploring a very little portion of the parameter space that is known to strongly affect results, especially at high Re.

The main rule of thumb that I can suggest is that, because the main dynamics at the wall is due to the streamwise streaks, you don't want to represent it uncorrectly. Now, dx+ = 40 is kind of good also for a wall resolved LES so, one effect you might be observing is that your choice of completely equal grid sizes in the 3 coordinate directions is kind of altering (trough the numerics etc.) a local dynamics that is obviously anisotropic and demands a different aspect ratio for the cells. But, again, this is just a guess. There are no two LES codes that perform the same given identical conditions.

I do suppose that your DES model is looking for unsteady (time dependent) solutions?

regards

mico

@Kazi Md. Jahid Hasan, I agree that we can do transient vortex-resolving simulation of flows with OpenFoam. I have used it for such simulations several times. But since it is 2nd order, truncation error can be larger or comparable to modelled properties. See for example Smagorinsky model, where nut is proportional to dx^2. Therefore, I don't think OpenFoam implements LES technology strictly, but most likely it works as something similar.

As i know, there is one third order implementation for OpenFoam by Wildkatze and Arjun Yadav, but I'm not sure whether it is open to community.

The second question is diffusion introduced in implementation of Rhie-Chow interpolation. This artificial diffusion term presents in all transient OpenFoam viscous solvers, regardless of version. Value of this term was unsatisfactory large for some of my problems, that's why I've changed original algorithm. If you want to do LES-like simulations with OpenFoam, then probably you should tune original solver.

Finally, you can take any latest version from openfoam.org (v.9) or openfoam.com (v. 2106) and it should give you similar results if you use incompressible viscous flow model.

Giuseppe Altieri thanks very much for your answer

Leo Cotteleer use DEFINE_TURBULENT_VISCOSITY macro and after writing the UDF, you may hook it in define/model/viscous panel.To be honest, DEFINE_TURBULENT_VISCOSITY macro replaces the new definition of viscosity in turbulent model. If you would like use the default one and add some extra expression, you can do it by this macro, too. You can combine the default turbulent viscosity and others. Please find the attached manual herewith.

best regards

Try the NITA based on fractional time step and compare to see if you have better solutions.

Cells clustering is entirely decided by the mesh generator based on the definitions you made before. If the high building you defined through the boundary condition is placed where the mesh is finer, you have the answer.

Difficult to address any textbook that explain the details about how LES is practically performed. You have to think that there are different codes based on different formulations and different SGS models. You can have big differences from a FD or FV or FEM or SM based LES codes. Often, a LES code is just a derivation of a NS code where an SGS model is added.

If you want to focus mainly on FV-based LES, you can read the historical paper of Schumann. The CTR has a lot of material but you should read also the fundamenta papers published on journals. I suggest also to see the general forum of cfd-online where many details are discussed.

You can find also some files in my RG home page with some lectures about LES

But, I strongly suggest to ask for specific issues about LES to get useful answers.

If you have the resources (time and equipment ) go for LES.

Ayaboe Edoh

in RANS-like formulation the numerical discretization has much less implication as the turbulence model acts on all the scales.

I don’t know the specific issues in PANS but only for the case of a scale separation the numerical discretization assumes relevance

Documents that analyze LES grids like this: "Grid-point requirements for large eddy simulation: Chapman’s estimates revisited" or "Grid Construction Strategies for Wall-Resolving Large Eddy Simulation and Estimates of the Resulting Number of Grid Points".

I think Muhammad provide you a good references

Dear N.C. Markatos

Completely agree with you in the computational context, not for the real-world scenarios. What I mean is somehow deal with the butterfly effect which implies even the farthest smallest decimals may have a role to play in the holistic simulation of turbulence like in nature.

Regards,

Hamed

Several researchers in this field: Zheltovodov, Bushnell, Borovoy.

The real difference is that LES is NOT a turbulence model but a different formulation of the governing equations based on local filtering of the variable.

That strongly differs from the formulation based in the statistical averaging (RANS/URANS).

The variables that are computed are totally different.

As I mentioned above, in principle such a LES simulation with CHT is possible, but only on the expense of massive Linux Cluster computation time. If I say "massive ressources" then I mean something like several hundreds of CPU cores.

What are the reasons:

1) LES by itself is already a rather expensive simulation method. Usually a good quality LES takes a couple of days on - let's say - 100-200 CPU cores. Mesh resolution depends on the simulated Re number and resolved flow/geometry features, but it is absolutely clear, that you cannot do a good LES on 1 million mesh elements. LES is in any case full 3d and even for a backward facing step you need a certain minimum resolution in the 3rd coordinate direction. And it is forbidden (physically invalid !!) to apply symmetry boundary conditions in a scale-resolving simulation like LES. Even periodicity is a wrong assumption, sine real turbulence is chaotic but not periodic or symmetric.

2) Second, combining LES with conjugate heat transfer is extremely expensive. The reason is, that the time-scales of the LES and the thermal conduction processes in solids are orders of magnitude different in terms of applicable and required time-scales. In the LES you need very small time-scales in order to resolve the motion of large and medium size turbulent eddies and their turbulent fluctuations for pressure, velocity, temperature. In the solid you need to run from your thermal initialization (which is usually not a thermal equilibrium state) to the fully developed temperature distribution in both the solid and fluid domain. Since it is a transient simulation, you cannot apply different timescales in the different domains, like this is usually done in steady-state simulations in order to speed-up this process of finding the new thermal equilibrium, after temperature distributions have fully developed. So you need to run the LES for a substantially linger time for combined LES/CHT in comparison with doing "only" a full LES simulation without CHT. This usually makes LES/CHT so expensive, that with "normal" computational resources it is not possible to do it. But at the peak of science with supercomputers it can be done.

3) In order to quicker arrive at the seeked thermal equilibrium state in fully developed LES conditions one can apply a "trick". In order to let the temperature distribution develop itself more quickly - in particular in the solid domain - one can artificially increase the thermal conductivity value of the solid material by 1-2 orders of magnitude over a certain simulation time period. By that the solid material can quicker heat up to the desired thermal state from the applied boundary conditions and one can save simulation time. But afterwards, if the thermal conductivity value has been set back to normal, one need to run still a larger number of LES through-flow times for the geometry in order to get rid of the influence of this artificially increased physical parameter in the obtained solution.

Finally - I have tried with colleages in an industrial project with VOITH Hydro a similar CFD analyses. And we finished this project with a bit mixed success. The computational resources (~350 CPU cores at that time several years ago in 2014) were not yet really sufficient to obtai a 100% good result of SAS_SST/CHT simulation. But maybe you like to study our publication:

http://www.drthfrank.de/publications/2014/140515_ACUM_2014_CKasprzyk_et_al.pdf

Best regards,

Dr. Th. Frank.

Filippo Maria Denaro Thank you for your suggestion. I have enough computational power and i have checked papers too. some people used RANS, some LES and DES. I was thinking that which technique will give me the good results

Agree with all above. 10 - 20 local dn+ ( local wall unit). 50 if you have a good wall modeling. Use 1-2 dn+ if you want DNS-like solution, say 5 -10% error relative to DNS.

The practical question is how to find dn+.

In practice, there is no way to know that for arbitrary shapes body. Just do a simulation on reasonable grid. Compute overall total shear stress. This value should be in the same order of magnitude of the actual value. Use that to determine your next grid.

You can use some smart means/analytical cases to guess your first grid.

Well, you could perform several analyses. For example you can compute the TKE over several planes normal to the streamwise direction and associate the evaluation of the energy spectra.

A qualitative approach could be the identification of the vortical structures.

I don't think it exists a unique approach.

The concept of LES, while an attractive alternative to the prohibitively expensive direct numerical simulation, poses some significant problems on its own. The main difficulty is that there is no common agreement on the form of a subgrid scale model. In the best known case of the Smagorinsky-Lilly formula, we have a simple algebraic relationship involving the magnitude of a strain rate tensor and the length scale determined by the grid size.

In the more general case, we can propose various expressions involving the characteristic velocity and length scales. Given the different physical mechanisms and geometric constraints affecting vertical and horizontal turbulence, it is advisable to have different values for mixing lengths mentioned in the question.

We should remember, however, that evaluation of diffusivities based on the concept of a mixing length is only an approximation. The precise formula for the mixing coefficients in the LES model can be obtained from the realistic sub-grid-scale model.

Please CITE :

Zeineldin Elserfy

as you see from the literature, there are many proposals for generating inflow conditions for LES (and DNS). I can not say that one specific formulation is the best.

It is known that Cf=wall shear stress (tau wall)/0.5*rho*v^2.....from openFoam you can use the postProcess command to calculate wallShearStress.

So, it is either you can calculate Cf from wallShearStress and then divide it manually by 0.5*rho*v^2 (using an excel sheet for example) .......or you can create your own post processing function in openFoam by modifying the wallShearStress function.

Are you looking at density variation (compressible) and phase change at super critical temperature ?

Gravitational effects in horizontal ducts are observed at very low velocities such as free-forced convection situations .

i would advise dense phase model in cfd to assess such situations.

please elaborate the problem so that we can discuss in detail.

i would suggest to look into physics so that we can choose appropriate model for analysis

If you wanted to model it with perfect accuracy, that would require a DNS model with a high order of accuracy in terms of your numerical grid resolution, a lot of computational power and time at your disposal.

To get an some understanding at least (not perfect accuracy) of the eddies and drag associated with something like a car wheel LES would describe the effect of the larger eddies, as the name suggests, whereas RANS is averaged, again as the name suggests, I recommend reading this thread on stack exchange for some more understanding of the differences "What are the differences between an LES-SGS model and a RANS based turbulence model?"

good luck

Olexiy Shynkarenko Thank you for the information Olexiy. Would it be possible to share some weblink or author information as i tried to google but couldn't get any relevant hit.

Thank you,

Sid

Hi Joemon

As initial preperation step for Ansys meshing, I use spaceclaim to prepare the model. In spaceclaim, even if you have two volumes (LES inside RANS volume) you can define the topology option as 'merge'; then conformal mesh will be created later in Ansys meshing. However, ELES allows non-conformal mesh so it is not a big problem if the two volumes are not conformal, as long as the grid size difference is not beyond the allowing threshold. For the overlapping surfaces of the two volumes, you need to define them later as 'interior' in Fluent. Once they are defined as 'interior' you can define them as 'Les rans interface'. For non-conformal mesh, you need to use 'create/edit mesh interface dialogue box' to define the overlapping surfaces of LES and RANS volumes as interfaces.

How does the above sound to you?

Daniel

It is difficult to use UDF to define a modified SGS model in Fluent. So we choose OpenFoam to do the job.

You can also try using modal techniques such as spectral POD, or Empirical Modal decomposition to extract coherent features from the flow.

Why shouldn't any species transport model (like Eddy Dissipation) not require temperature patching? Table-based models like equilibrium or flamelets do not, but if you use some form of species transport, you will have to provide an ignition source at some point to initiate combustion.... otherwise, you'd only simulate mixing of fuel and oxidizer. What can happen is that after temperature patching, the flame will not stabilize and the hot exhaust products are transported out of the computational domain.This would indicate that your flow field is not as stable as you'd want. Sometimes it also helps to not only patch temperature, but also some intermediate species like CO or H2 (I assume you talk about the combustion of natural gas).

Be careful with nomenclature here, by the way: in FLUENT, non-premixed combustion models are mixture-fraction-based models. This is not to be confused with non-premixed combustion in general which just means that fuel and oxidizer are separately injected into the combustion chamber. Thus, you can simulate non-premixed combustion with some kind of species transport model, which is in fact done quite often.

The filter width depends on the formulation: implicit or explicit filtering. In a software like Fluent the approach to the filtering is implicit and the filter width tends to zero according to the grid size tending to zero. This way the LES solution tends to be a solution of the unfiltered NS equations.

The only congruent way to work is performing the 3D LES and then averaging in time the results to get a statistically steady solution according to the 2D RANS.

K-epsilon model

That's probably the case, but you can't compute the mass flow rate in that way for a turbulent flow. That formula is valid for laminar cases.

Try using a pressure gradient in x direction only

The stress tensor in the Newtonian model writes as

T = [-p+(lambda+2/3mu)div v]I+2mu D₀

therefore if lambda+2/3mu=0 is assumed, also if the flow is not divergence-free this term has no relevance.

Taylor microscale can be estimated by means of the autocorrelation.

While there is no universal answer to your question about hybrid computation, more and more people consider that the directive approach (OpenACC / OpenMP) to be the nominal way to go for porting large legacy codes, mostly because OpenAcc / OpenMP directives can be added incrementally in the code.

An interesting emerging alternative, which is probably more adapted to new codes aiming both at performance and portability across architectures (x86_64 multicore, nvidia GPU, AMD GPU, ARM, ...) is to use a "smart" library like Kokkos (developped at Sandia NL) or RAJA (devel at Livermore NL) or others to implement shared memory parallelism. These libraries provides "generic" concepts to express parallel loops (for, reduce, and so on). These libraries focus on node-level parallelism, and need to be used in combination with MPI for large scale distributed computation.

Another major advantage of these libraries, when compared to the directive approach, is that, and this is especially true for Kokkos, is that they also provide "architecture-aware" data containers. E.g. Kokkos provides arrays (called Kokkos::View) for CPU/GPU/... which allows you to access data like this data(i,j,k), very intuitively, and simplify the memory management when dealing with GPUs. Note that Kokkos library is a core building piece of the large parallel linear algebra package Trilinos.

I think that portability and more precisely "performance portability" is a keep concept today, that is being able to develop codes that can run efficiently on multiple platforms, not just one.

Rajasekarababu,

In most of one phase industrial applications, RANS is a pragmatic choice. This is explained by the robustness, ease of use and efficiency of the RANS.

However most of multiphase flows are highly transient and steady state approach does not work.

Some industrial applications expose limitations to the RANS modelling approach. For these applications, LES and DES provide a viable alternative at the expense of increased complexity and simulation time.

Regards

Thank you all above for this very interesting discussion. I hope you allow me to join. I have been giving this topic some thought as well over the past years, in the framework of the shallow water equations and the formation of dynamic eddies in URANS computations (or to be precise U-SWE computations) with steady boundary conditions. In other words, any dynamics that emerged in these computations came purely from the nonlinear excitation of discretization errors (which is I guess the topic of this discussion).

In these investigations, I experimented with the effect of the nonlinear advection scheme and the grid resolution on the emergence of eddies. Results indicate that most computations are performed with a 'numerical' Reynolds number which is not at all linked to the fluid viscosity (at least for practical grid resolutions), but to the artificial viscosity, effectively emerging from truncation errors from the discretization (and mostly from the nonlinear advection discretization). For numerical Reynolds numbers that are low enough (coarse grid, 'poor-quality' advection scheme), the results were steady and only resembed the time-averaged solution. For higher grid resolution and/or better advection schemes (not necessarily higher order, but with better conservation properties), the solution became time -dependent with vortex shedding (flow around a cylinder, flow around groynes in rivers).

Due to its link to the numerical viscosity, the aforementioned numerical Reynolds number is automatically space-, direction- and (in case of flow dynamics) time-dependent, resulting in an artificial/numerical solution.

For commonly-applied grid resolutions in practical environmental flow applications, the numerical Reynolds number is orders of magnitude lower than the theoretical one, based on the fluid viscosity. However, - and this is the main thing and also the reason why ILES methods were developed - this artificial solution resembles in some/many cases what happens in reality. In other words, despite the numerical/artificial nature of the (numerical) eddy viscosity, the flow dynamics and eddy formation, shedding frequency, etc. shows good agreement with reality, at least for certain advection schemes that exhibit certain conservation properties (momentum or energy).

To come back to the topic question:

In my opinion, URANS with/without a turbulence model and LES without explicit filtering differ really only in the turbulence model. The dynamic effects seen in URANS are artificial and in both models greatly depend on the grid resolution and the discretization of the nonlinear terms, whose discretization errors cause dispersion and spurious modes, which might look 'physically-correct'. The question for me then remains: do we except this and say that the results are valid despite being obtained for the wrong reasons in URANS (discretization errors), or do we strive for a correct solution that was obtained for the right reason (like in LES)? And perhaps linked to this: how do we know for sure whether a certain method always provides adequate results, for the wrong reasons?

My apologies for this long response.

Best wishes,

Frank Platzek

LES is so wide but also so niche that, in my opinion, a course is only useful if targeting basic stuff only (but with useful insights). Once you build that background correctly, the student can go deep in the preferred direction without the risk of being messed up by the so common copy&paste sources in the field.

Also, base material is nowadays largely available on most topics. To be really useful and new on a niche topic, that material should instead include also the minimal details, including implementations. Basically, the one who prepared it should have worked on that beforehand.

You can think of a two-part course. A first, basic one, but wide enough to cover all the LES arguments, giving the correct view. And then a second one, with each lecture dedicated to a single topic.

But you see how this can easily require a huge effort, not proportionate with respect to the scope (i.e., how many students are going to be interested in, say, implementation details of inlet specification methods, or commutation error treatment, or whatever?).

Thanks Kazi. However, it seems I will opt for SUREG procedure anyway. However, this leaves me to figure out n parameters of n equations, which I can solve in GAMS. I was hoping for to find a more elegant solution, but I cannot figure it out.

see this similar post https://www.researchgate.net/post/LBM_code_OpenLB_or_Palabos

I saw it has been done. Thank you very much.

LES is not a model but a totally different formulation...

You can try to set the code using a 3D model and a high order FV-based upwind reconstruction for the convective terms, if available. This is the ILES previously recommended. No turbulence model must be supplied, you should work as it were laminar conditions.

However, be careful that you need to solve the BL over the wall, that requires a weel refined grid close to all the walls.

You will have to wait that the solution develop a statistically energy equilibrium, then you have to continue the simulation for several time-unit and save the evolution of the fields to do a post-processing of the statistics. Depending on the Reynolds number and computational power, that could be quite long, also some weeks.

Good luck

LES and RANS are two different worlds! The latter is based on statistical averaging, leading to steady equations (eventually also in 2d). A sort of time contribution is in the URANS brut it can be debated the le of the temporal mean.

LES is based in a local filtering, the equations are unsteady and 3d. As a consequence, the meaning of the turbulence model as well as of the final solution is totallui different.

I wrote some notes about https://www.researchgate.net/publication/313280320_Lecture_on_LES_-_Part_I_and_II

1) The distinction appears from your choice of grid refinement close to the wall. Of course, then a proper SGS model have to be used ( Enhanced Wall Functions in https://www.sharcnet.ca/Software/Fluent6/html/ug/node514.htm).

2) see 1)

3) yes, any CFD code that solves the unsteady 3D form of the NS equations can perform a DNS.

Generally one sees at the spatial correlations, if they reaches zero, the lenght is sufficient.

Torque is an important factor in much of the equipment on a factory floor. Measuring torque is often something that's misunderstood, which can lead to over- under-designing of measurement systems. This article addresses the many techniques and tradeoff of torque measurement techniques.

Torque can be divided into two major categories, either static or dynamic. The methods used to measure torque can be further divided into two more categories, either reaction or in-line. Understanding the type of torque to be measured, as well as the different types of torque sensors that are available, will have a profound impact on the accuracy of the resulting data, as well as the cost of the measurement.

In a discussion of static vs. dynamic torque, it is often easiest start with an understanding of the difference between a static and dynamic force. To put it simply, a dynamic force involves acceleration, were a static force does not.

The relationship between dynamic force and acceleration is described by Newton’s second law; F=ma (force equals mass times acceleration). The force required to stop your car with its substantial mass would a dynamic force, as the car must be decelerated. The force exerted by the brake caliper in order to stop that car would be a static force because there is no acceleration of the brake pads involved.

https://www.eetimes.com/document.asp?doc_id=1273988

Have a look here

This estimation is generally used for channel flows where you have that x and z are directions of flow periodicity (homogeneous flow). These spatial directions are discretized by uniform grid steps in the whole domain. Conversely, along the wall normal direction (y) the resolution of the BL requires at least 3-4 nodes within y+=1, then the grid is via via enlarged approaching the center of the channel.

I suggest to read this paper

Yes, I am aware ;-) I also worked and proposed some approach for LES using upwind.

But I think that in the LES community there are different schools that approach the LES equations in two ways, as either real filtered PDEs or numerical unresolved (by the discretization of equations and domain) equations. In my paper I tried to express my idea about those differences. I think that expecially the FV approach for LES was quite misunderstood.

I never worked using hybrid formulation. I suspect that there are few strong theories on what solution one really get in the blended regions...

Could you expand your question with the details of your problem? Why are you working with LES in such a problem?

Dear Professor Ivan Kovalets,

Thanks for the answer. Data assimilation for modification of boundary conditions at the measured surface would be a great idea. I will try in this way.

This way is quite simple but you need a very long inlet to let the flow correlate physically...but in some case where the separation is driven by the geometry (e.g. the backward facing step) that can work quite well

no there isn't a follow up evaluation yet as far as I'm aware although we recommended one be conducted and the department people certainly hoped there would be. it might be worth contacting the department. regards

Eileen

Hi,

it is common to assume a K₀ stress state, i.e. the lateral stress is the vertical stress multiplied by K₀, where K₀ is the earth pressure coefficient at rest. Jaky's formula can be used to estimate K₀ for normally consolidated soils:

K₀ = 1 - sin(phi), where phi is the friction angle.

See reference for other estimates.

Regards

Reference: R. F. Craig, Craig’s Soil Mechanics, 7th Edition, E & FN Spon, London, New York, 2007.

We can say the system is hyper-chaotic system because the LEs are +, +, 0, -.

There is no requirement related to the large Lyapunov exponent.

System is meant to be much more sensitive to initial conditions for higher Lyapunov exponent.

In the book "Turbulent Flows" by S.B. Pope (2001), the author explains non-dimensional wall units in physical terms instead of computational. The Reynolds stress at the wall must be zero, so the stress at the wall must be purely viscous (i.e. frictional). One can define a length scale for the purely viscous stress at the wall called the viscous length scale, defined as

d_v = nu*(rho/tau_w)^(1/2) = nu/u_tau, viscous length scale [m],

nu = kinematic viscosity (molecular), [m^2/s],

rho = fluid density, [kg/m^3],

tau_w = wall stress, [N/m^2],

u_tau = (tau_w/rho)^(1/2) friction velocity, [m/s],

A friction Reynolds number can be defined as

Re_tau = u_tau*d/nu, dimensionless,

d = a distance, [m],

such that if the distance is the viscous length scale d=d_v, Re_tau = 1. Non-dimensional "wall units" are x+ = x/d_v, y+=y/d_v, z+=z/d_v, so if you estimate the viscous lengthscale d_v then you can estimate normalized grid dimensions for your simulation.

The spanwise and streamwise resolution needed for turbulent boundary layer LES depends on your choice of wall model / your choice of SGS model. For incompressible LES, I recommend taking a look at the chapter on LES wall models in the book "LES for incompressible flows" by P. Saguat (2006). The author does a thorough review of the contemporary state of the art for LES wall models, and the author notes that part of the problem for LES of turbulent boundary layers is that turbulent kinetic energy is produced in the "buffer layer" just outside of the viscous sub layer in a turbulent boundary layer (Re_tau = order(1-10)). Therefore you either have to have a very fine resolution LES, like the y+=1 requirement that you stated, and/or use a wall model.

You already have some tracks to follow with the first two answer, but I would like to present a different track for explaining this.

Fully resolving the NS equations (DNS) involves that you are able to resolve, with your mesh, all turbulent scales down to the smaller one: the Kolmogorov length scale. However, the latter greatly depends on the Reynolds number of your flow which explains why it is not practical (or even feasable) to solve high Re number flows using DNS.

To tackle this problem, the idea is to seperate the scales (large/small) by spacially/temporally filtering your field. That way you actually only need to have a mesh fine enough to solve the large scales, and then, model the small scales. Thus, the cost of the simulation can be greatly reduced. 

Also, I would like to support the reference suggestion from Leonardo Araujo as it is a great reference for LES. There is also a version covering compressible flows if you need.

"Large Eddy Simulation for Compressible Flows", Garnier, Eric, Adams, Nikolaus, Sagaut, P.

 I think pethyagoda sir is doing some modifications in barbus group classification s recently in Indian sp

first, you did not provide details about your problem so that discussing BC.s in your setup is not possible...

In general:

1) Use dynamic SGS model, the turbulent viscosity will be determined by the flow conditions

2) Walls:Use natual BC.s for velocity if you have 3-4 grid nodes within y+<1

3) Inflow: many proposals exist, depending on your computational resource you can choose the best for you. Have a look at the Sagaut's book

Just out of curiosity, you say your time step ~10^-06 s? How many cells does your mesh possess?

You definitely must have other informations regarding the time of your problem. For instance, you may want to simulate a period/cycle of breathing, or perhaps two. If they take, say, two seconds each you would need at least 2 million time steps. Perhaps you may also consider using fractional time step to assure numerical stability in such a complicated CFD problem.

a no-model LES is the only way to check for the effect of  the SGS model and isolate from the numerical errors..

Usually, I set the fixed pressure gradient, this way the non-dimensional equations are expressed directly in terms of the v+ velocity.

The effect of overestimation in the average velocity profile is quite common in some LES codes. But what about your energy spectra profiles?

I attach a paper of a project we developed few years ago.

You should take a look at this:

In so far as explicit LES is concerned, there are no practical obstacles to combine it with AMR. Say, set the filter length to the local size of the grid and the model will run stably. There are, however, theoretical (and thus accuracy) issues related to commutativity of filtering and differencing operations. For discussion see Marsden,  Vasilyev, and Moin, Construction of commutative filters for LES on unstructured meshes, J. Comput. Phys. 175 (2002) 584–603, and references therein.

Piotr

I think you will have to write UDF's for this kind of post processing. Just like a turbulent flow experimental researcher processing an instantaneous hot wire output.

Dynamic Samgorinsky model will be appropriate as sgs model for LES.

Dear Dr. Thornber and Dr. Bensow,

Thank very much for your answers.

Many thanks for your time and reply.

I want to know that which procedure has lower simulation time with same accuracy?