[CFD] Large Eddy Simulation (LES) 3: Sub-Grid Modelling

[@TimofeyMukha]

Great work, I particularly appreciate the fact that you mention the dependency of the LES equations on the grid when implicit filtering is used. I have a few remarks: - I'm doubtful whether the assumption of isotropic subgrid eddies is in fact being made in eddy viscosity models. The subgrid stresses are certainly not considered isotropic --- they follow the anisotropy of the rate of strain tensor, as per eq. 11. So if your assumption on the eddies is the same as on the stresses, the eddies are in fact allowed to be anisotropic as well. - I understand that you want to show that nu_sgs must depend on the grid size. This is of course correct, but fundamentally there is also a velocity scale involved: to get viscosity you multiply a velocity scale by some length scale. So while nu_sgs accounts for the size and is a function of delta, that is only half the story.

[@fluidmechanics101]

Great clarifications! I have pinned your comment as I think it may be useful for others

[@argayudhistira1113]

I cant believe how happy i am to see you uploaded a new video!

[@argcargv]

One point that should also be noted is that RANS models and LES models are fundamentally different in that RANS uses an ensemble average to arrive at the governing equations whereas LES uses space-time averaging to arrive at the governing equations. This means that the dissipation rate (and other parameters) may have the same units, but they are fundamentally different. The dissipation rate from RANS is defined in the context of ensemble averaged equations and accounts for dissipation that occurs at all scales, whereas the dissipation rate for LES is defined in terms of a space-time average (filter) and only includes unresolved dissipation. They are really apples and oranges and it is easy to get confused if one doesn't recognize that they are fundamentally different.

[@fifinoirdefer9458]

Honestly your videos are so good and so clear, you are single-handedly saving the year of a lost CFD student here, I can't thank you enough. I never gave money to youtubers but I think I'll go on your patreon, you do such a good job I can't express how thankful I am.

[@samoalfredgroup]

Keep up the good work you are doing. I am going to ask all my students to watch all your videos.

[@fluidmechanics101]

Amazing! Thanks 😄

[@Samad-s2x]

Thank you for putting together this series for LES, I am finding it extremely useful! I have one question, you mention that the sub-grid viscosity is a function of mesh cell size at 31:01. So is sub-grid viscosity manually set by the user or does the CFD code compute sub-grid viscosity to ensure that eddies larger than mesh dissipate?

[@fluidmechanics101]

Usually the CFD code uses a default value of the Smagorinsky coefficient which implies that they set the sub grid viscosity. It is left to the user to make sure that their mesh is fine enough for this sub grid viscosity to be appropriate 👍

[@xchen3132]

Finally it comes out, I have been dreaming of this lecture for a long time !!! Thanks!

[@jessbuildstech]

I love this video, thank you so much for explaining it very simply for us - Looking forward to LES 4!

[@sinansavaskan2132]

Perfect tutorial. Just right. Looking forward to next videos on LES

[@ankit_sb]

Thanks for this video. These videos are helping a lot to clarify fundamental things. Looking forward to seeing the next one.

[@johnluffman7954]

This is the best lecture I have ever known.

[@TheManolis1984]

Yes, LES part 3!! Are videos on cell types (tetra, hexa, poly) and k-ω GEKO in your "to do" list?

[@fluidmechanics101]

👍

[@ZiyuZhang-h4s]

A very nice talk about les! Thanks!

[@thunder852za]

Really enjoy these videos.

[@ricksaylor1953]

Excellent! Thanks Aidan.

[@santoshn8450]

Very well explained 👍

[@mohammadjadidi233]

Really enjoy your videos

[@azizouadoud1898]

Excellent thanks for your efforts

[@brandonxu6317]

very useful videos for LES beginners

[@AshokAshok-dt6lu]

Great talk as expected.

[@saikrishna-oy5cl]

As here in LES we are adding sub grid scale shear stress to dissipate the effect small scales ,so to get exact resolved Turbulent Kinetic e energy...Where as in RANS we are having the eddy viscosity(to solve Reynolds stress term) ,Can I think like this ,In RANS eddy viscosity term is there to dissipate effect of fluctuations and we get the average velocity .If I am I wrong can you please clear the misconception....or can I think like this In both the case we are adding the effect of turbulence in the equations .... and I think about 80% we resolve and 20 % we model which is sub grid scale which of the above assumption is correct in the above ? And also...What about the small scale turbulence Kinetic energy here in LES ,How to find it ? And thank you for your great lectures !!!! @Fluid Mechanics 101

[@nilakanthasahoo798]

Excellent...this is no less than a University lecture talk....!!! You planning on becoming a professor?

[@sauravpanda1380]

yes i agree....

[@shobob29]

Thank you for great lectures!!!!!!

[@lenacaban4178]

Great job! I am waiting for more :)

[@emreaygoren1701]

very useful content

[@manojsundararajan6166]

would love for a video from you on DES and its applications.

[@nabekary]

Great vide0. thanks.

[@TariqRidwan]

Superb! 🤓

[@christopherburkhardt4405]

I really like this series on LES! It provides a lot of insight into a topic that is so fascinating. One minor point to watch out for is that some authors like Grinstein, Moin, and Sagaut seem to distinguish between “Implicit LES” (also abbreviated as “ILES” or “MILES”) and “LES with implicit filtering”. The former (ILES) employs no explicit SGS model, but certain features of the numerical scheme induce the desired effects. The latter refers to filtering by volume averaging in finite volume codes instead of explicitly introducing a convolution into the governing equations. Do you think it would be worth clearing that one up in the next few talks?

[@philosopher2425]

Thank you very much for these interesting talks and comprehensive and clear explanations! You've mentioned the term "implicit LES" and refered to it as LES without explicit filtering, i.e. grid filtered, commonly known as Schumanns approach. In the literature "implicit LES" is sometimes used to designate LES without explicit SGS modeling, where one relies on numerical dissipation associated to low-order discretization schemes in order to account for missing eddy breakdown and SGS-level dissipation. Have you heard about this kind of implicit modeling? I would be interested in your thoughts on this topic.

[@fluidmechanics101]

Yes, I have definitely heard of this approach but haven't looked at it in too much detail. I tend to try and focus on the methods used by the mainstream CFD codes, as these tend to be more useful for the majority of the CFD community. Whereas approaches like "implicit LES" tend to be more niche, so I don't cover them in much detail. They are definitely interesting and I would like to look into them some more 🙂

[@philosopher2425]

I recently stumbled across a formula for viscous dissipation epsilon (equation (2) in your video) in the literature, where the strain-rate tensor was used, instead of the velocity gradient tensor: epsilon = 2 nu S_ij S_ij and I was a little confused, because others rely on epsilon = nu du_i/dx_j du_i/dx_j as in this video. Do you maybe know why these different definitions exist?

[@chamocudno]

Hell yeah! Part 3 let's go

[@fluidmechanics101]

😄

[@markoj3512]

I should correct you because my lecturer corrected me when I had my exam. Energy dissipating into internal energy of the fluid not heat. By the way your you have done a great job. 👍

[@밍맹구리]

Thank you for high quality video of LES :) Do you have any plan to make video which is about difference between implicit-LES (i.e. ILES, MILES) and conventional LES model? I am very looking forward to watching your next LES video!!!!!

[@fluidmechanics101]

I will try and cover this in the next few videos. There is a lot to cover!

[@LinkinPark24786359]

Great video!

[@etiennemuller792]

Really great work! I discovered your youtube channel not so long ago and I'm always looking forward to seeing a new video uploaded. Until now I had only a CFD background with the finite element method so you can easily imagine how your videos helped me to get more familiar with the finite volume method. This being said, I partly disagree with the fact that you cannot do a mesh independence study while using an LES approach. I believe it's only true if you are doing implicit LES or LES with implicit filtering. I may be wrong. I really can't wait to see your next videos on sub-grid scales models.

[@fluidmechanics101]

Yep, agreed. The statement about mesh independence studies is only true for implicit LES (like those used in commercial CFD codes like ANSYS Fluent, OpenFOAM and CFX)

[@saikrishna-oy5cl]

@@fluidmechanics101 As here in LES we are adding sub grid scale shear stress to dissipate the effect small scales ,so to get exact resolved Turbulent Kinetic e energy...Where as in RANS we are having the eddy viscosity(to solve Reynolds stress term) ,Can I think like this ,In RANS eddy viscosity term is there to dissipate effect of fluctuations and we get the average velocity .If I am I wrong can you please clear the misconception....or can I think like this In both the case we are adding the effect of turbulence in the equations .... and I think about 80% we resolve and 20 % we model which is sub grid scale which of the above assumption is correct in the above ? And also...What about the small scale turbulence Kinetic energy here in LES ,How to find it ? And thank you for your great lectures !!!!

[@fluidmechanics101]

I think you are almost there. You can think of RANS and LES together in the following way: the turbulence model models the effect of the turbulence that is not resolved by increasing the dissipation of gradients in the velocity field.

[@fluidmechanics101]

Also, you can calculate the small scale TKE (as you call it) from the local sub grid viscosity and the local length scale. Have a look in the CFD manuals to find the formulae 👍

[@saikrishna-oy5cl]

@@fluidmechanics101 Ya I got it...waiting for more lectures on LES(sub grid scale models)...and thank you for your reply.... You are the best...I have ever seen !!!

[@felizbanking2951]

Hello can you discuss DELAYED DETACHED LARGE EDDY SIMULATIONS. can you make videos on how to understand DDES

[@jerrinjob6210]

Can you please do a video on TVD and flux limiters ? Thanks in advance.

[@sachinhegde2472]

Can you make a video, explaining the process of CFD simulation? (From geometry of the object to be simulated to the solvers to solve the algebraic equations). Maybe not too detailed, but considering the major aspects and pointing out to other videos of yours wherever necessary for more details. That would help me a lot. Thank you in advance

[@fluidmechanics101]

Yes, I am planning to do this eventually. I think it would be very helpful to many people 👍

[@s.mohammadhosseini6450]

You are amazing!!

[@rayansoppi2380]

Merci beaucoup !

[@JaatLostInFrance]

Amazing videos, really learning a lot from your videos. They are treasures, to be honest in the LES field. Following your series I have got one doubt when you increase the overall viscosity, aren't you increasing effects on those really big sizes eddies too which are successfully resolved by the mesh? Any help on this will be really helpful.

[@fluidmechanics101]

Great question. Honestly I am not sure on this one. LES is a very difficult field and I am still learning myself

[@anurajmaurya7256]

Please make video on variational multiscale turbulence model

[@pv597]

I have a question, how does one "marry" LES with, for example, PISO or VOF? I don't see where LES fits in the over-all solving process. Is it just giving a viscosity value to each cell? Is it changing the velocity field or pressure value? I'm new to CFD, and I have a general understanding of how each algorithm/method works, but I still don't see how does all fit into the Finite Volume Method (the general matrix, so to speak). Best regards and thank you for such incredible videos!

[@fluidmechanics101]

The principle is exactly the same. Both RANS and LES work with the finite volume method. The only difference is that you have a much finer mesh and smaller time step for LES so that you can resolve the fluctuations on the mesh. The Reynolds Stress or Sub Grid stress are just additional forces (remember the divergence of a stress is a force) that you apply to the control volumes. The sub grid stress is much smaller than the Reynolds stress. But the principle is the same 👍

[@pollovrhd]

@@fluidmechanics101, so when you set a simulation, for example, in Fluent, you set it to use PISO as the Pressure-Velocity coupling and LES for the turbulence model, which one is solving the Navier-Stokes Equations? Which one is building the matrix or calculating the pressures and velocities? I assume that VOF only gives the values of density and viscosity at each cell, proportional to each phase.

[@pv597]

@@pollovrhd From my understanding (correct me if I'm wrong) is that PISO is the algorithm (sequence of instructions, such as solving the momentum equations or the pressure value), LES is the one that actually solves these values (the computer does the calculation, but LES tells it how) and VOF tracks the free surface cell by cell.

[@fluidmechanics101]

Almost. PISO gives the order in which the equations are solved. The finite volume method is actually the method which assembles the equations and solves them. LES and RANS are just additional modifications to diffusion (in various ways) to account for the effect of missing (unresolved) turbulence. VOF is the method used to specify extra sets of equations for different phases

[@fluidmechanics101]

Sorry, didn't read your whole post! Yes VOF gives you velocity and density in each cell

[@kennettallgren640]

Nice talk as always! Do anyone know if some attempt to model turbulence as an ensemble of eddies superimposed on each other as a computational model? I mean could an individual eddy be represented by a center point, rotation direction and strength. Idk fi this makes any sense :). Could you do any calculations from this?

[@fluidmechanics101]

You could look into the 'synthetic eddy method'. This is essentially what you are describing and is used to specify I let conditions for LES

[@kennettallgren640]

@@fluidmechanics101 Thanks! I will look into it.

[@Mohamed-oz8bj]

Hi Adian, Excellent work. Just wanted to ask about the total time recommended to run LES Model. Several authors have used roughly 10^-5 or 10^-6 as the time step size but i did not get anyone mentioning the total time. Someone said about 5 sec, another person argued that 5 sec is too much maybe 0.5 sec. What is your suggestion. Thanks in advance.

[@fluidmechanics101]

Hi Mohammed, the total time will depend on your specific problem. You need to run enough time for the flow to 'wash through' the domain twice. You can calculate this roughly as 'volume of your domain / volume flow rate through your domain'. People often refer to this as the 'hydraulic retention time'

[@Mohamed-oz8bj]

@@fluidmechanics101 Thanks a lot. i think the formula you have provided 'hydraulic retention time' will be very helpful. i will start with that and lets see where it ends. You are amazing Aideal. God bless you.

[@khaledebraheem]

Hi. I am deeply grateful to you. Thank you. In understanding RANS vs LES, I am a little bit confused here. In RANS: 1- Navier-Stokes Equation 2- Decomposition of instantaneous u to (U Mean) + (u' fluctuation) 3- Time-averaging /Output: an equation f(U Mean, normal u'u' and shear u'v' ... )/ 4- We need to replace u'v' with U Mean also /While doing that new artificial term Mu(t) is up / 5- Modeling (e.g. K-e) to get Mu(t) /So in brief, we replaced the (instantaneous u) with (U Mean) and Mu(t) which itself is replaced by k and e for example/ On the other hand, In LES: 1- Solve for instantaneous u directly. NO decomposition here. 2- Statistically-averaging to get U Mean. This is not the same average as above. Right?? 3- u' fluctuation = instantaneous u - U Mean 4- Calculate resolved turbulent kinetic energy k(res) 5- Calculate k in SGS by adding an extra term. Thank you.

[@fluidmechanics101]

For RANS you are 100% correct. For LES you have all the key stages. I think the part that you are missing is that the instantaneous u that you are solving for in LES contains = U mean + u' resolved. There is another fluctuating component that is below the grid size. Ureal = U mean + u'resolved + u'subgrid So when we do the statistical averaging in LES (to calculate U mean) we aren't including u'subgrid. This is why Umean in LES is slightly different to Umean in RANS (assuming the RANS model were perfectly correct). I know this is very confusing but I think you have all the stages that you need for your understanding and you are correct 👍 also, great job on your explanation

[@lizarettflavour]

LEEEES =) Thank you, Aiden =)

[@Shri_RRaamm]

Hi Adian Sir, What wall function should i use for omega? can you please give us a idea? Thanks

[@fluidmechanics101]

If you are using LES then you aren't solving an equation for omega, so you don't need a wall function. If you are using RANS (I suspect) then check out my video for omegaWallFunctions. Broadly speaking what you do is apply a fixed value to the face of the cell in contact with the wall. The value that you apply depends on the height of the cell centroid from the wall (y+). In OpenFOAM there is only one boundary condition to choose from for omega, so this shouldn't be a problem!

[@Shri_RRaamm]

@@fluidmechanics101 I think there is no specific videos for omega wall function????

[@fifinoirdefer9458]

@@Shri_RRaamm You haven't look really far kzbin.info/www/bejne/aGe0knaBa9x3ppo

[@Shri_RRaamm]

@@fifinoirdefer9458 that is the turbulence model but i was talking about omega wall function. is it not different?

[@fluidmechanics101]

Ah sorry guys I think I forgot to do a video on omegaWallFunctions 😄well spotted!

[@ShamJam85]

The filtering operation has not been touched anywhere perhaps

[@joaquinaranciaga7479]

Not only you cover the contents clearly and without unnecesary redundance, but also the rythm of your speech would be very hard to improve. Not to mention that you do not make a single 'hesitation' pause. Congrats!

[@fluidmechanics101]

Very kind! Thank you 🙂

[@ghebat4716]

the goat

[@Bryanbar93]

Wow