Large Eddy Simulation - comparing Simulation Methods in OpenFoam or Ansys - why one should use LES
Вставка
- Опубліковано 14 жов 2024
- www.engineerdo.com
This video explains briefly which simulation method is used for what kind of problem. What are the benifits of LES or Large Eddy Simulation compared to RANS.
![[CFD] Large Eddy Simulation (LES): An Introduction](http://i.ytimg.com/vi/r5vP45_6fB4/mqdefault.jpg)
![[CFD] Large Eddy Simulation (LES): An Introduction](/img/tr.png)
As someone who uses RANS models all the time, LES looks like magic to me
Hey Manolis, glad you enjoyed the magic :)
The transition to fully turbulence during external flows is generally characterized at 500,000 and above for external flows. Also, DNS is a method to resolve all scales of Eddie’s including the kolmogrov scales. The matrix inversion can be applied to any solver formulated with an explicit diagonal matrix. However your simulation visualization is very good
Please correct me if I am wrong
Whenever problem is solved using implicit approach we always land in AX = B type linear equarion where A is coefficient matrix which is sparse and diagonal
X can be obtained simply by obtaining A‐¹B but inverting a matrix is computational extensive work when mesh is large and in millions so the best way is to solve via iterative approach such as Conjugate Gradient Algorithm
On the other hand when explicit solver is used there is no need to build coefficient matrix because now the Governing equations will be solved at each grid point explicitly using neighbouring grid points data depending upon the type and order of discretization scheme such as Center Differencing O(3) scheme
Hi, thanks for a nice and concise video. I have just few remarks/questions that I would like to be cleared up to avoid potential confusion (at least for me):
1. Using Re=2300 and 2900 as a threshold for laminar/turbulent flow, isn't this value an estimate for a flow in a smooth straight pipe? And even there, it is just an indicator, not a fixed value. I would be much more carefull stating that anything bellow 2300 is laminar and everything above 2900 is turbulent.
2. Listing 'laminar' as a RANS model, I would like to know how got there. AFAIK (and of course, I might be wrong), turbulence is inherently present in the NS equations and manifests itself more and more as the convective term in the momentum balance gains importance. Consequently, solving the flow using 'laminar' turbulence model with high enough mesh resolution to capture the whole energy cascade is DNS. Thus, I would mention the 'laminar' turbulence model either in a separate category or in the DNS box. To list it in RANS, there is just no averaging in 'laminar'.
3. I thought about having a few more rants on the topic of RANS vs URANS and and the necessity of specifying resolved vs modeled energy ratio for any LES simulation but oh boy, for such a nice video, this is a way to long of a comment.
Thanks for making the effort and sharing your work!
Hey Martin, thanks for your comment. Of course you are right, the Reynolds number just gives you an idea where you are with your flow behaviour. Actually it's rather academic, because in my 10+ years in heavy industrial projects, I've never seen a laminar flow. The choice of the turbulence model can also be academic because it's usually depends also on reliable measurements of the flow which normally are not available. At least in OF laminar means, that no turbulence model is used. So you are write, there will be no averaging in laminar.
It would be very nice to see some more information in the description such as a few more details for the LES. Which filter were used, and numerical schemes for example.
Thanks, Tobi
The laminar to turbulent transition at ~2300 is for flow in pipes, on a flat plate it is ~10^5.
Thanks for pointing that out
Thank you Henrique for such a concise explanation. You did it efficiently just like RANS. Great job. More than happy to subscribe to your channel.
Thanks Mate! Really appreciate the feedback.
In terms of accuracy Turbulence Models are as under
Sparlts Almarts 1 equation --> kw, SST, ke, kw-SST 2 equation Model --> Reynolds Stress Transport 3 equation Model to find u', v' and w' which will be used to find 9.0 Reynolds Stress Tensor components u'u', u'v',u'w',v'u',v'v',v'w',w'u',w'v',w'w' --> Large Eddy Simulation where Macroscale eddies are resolved and microscale and Kormogolov length scale eddies are modelled based on universal pattern of life cycle --> Direct Numerical Simulation where mesh is refined upto Kormogolov length and Simulation time is lower then Kormogolov time scale in order to resolve entire Eddy life cycle from birth to death by viscosity
Really impressed with your videos so far definitely an appreciated source of information for someone who is getting into opensource cfd
Hey Christian, thanks for your comment! We appreciate that!
Cheers, Engineerdo
Nice visualization! Now I know why my hair gets messy in a convertible :D
Thanks Ale X, stay tuned. The simulation with a comparison of a wind deflector will follow soon ;-)
Hi everybody, I am Ian, honestly, I think I am too late to learn, now I am a fourth-year student in mechanical engineering but I am new in Ansys
I really want to learn Fluent, but I face a problem that I don't really know when I have to choose the right turbulent mode
I see on youtube they just click K omega SST, K epsilon(realizable & RNG), LES,Spalart-Allmaras, RANS, and so on, but I don't really know what case I have to choose and do my turbulent mode is right or not
do you have a recommendation for me to learn about it in a simple way?
thank you so much
Hi Ian, in a nutshell all models are built to match a different case. If you want to get a good fit between your simulation and the reality you will need a measurement and then you can try if one of the turbulence models comes close to your measurements.
How many cells did you use in your mesh?
Hey, does ansys work in Windows 10 Home edition ??
And one more thing... Which software should i use.. OpenFoam ir Ansys?
I am very confused.. Please help me out..
Ansys should work with windows, however thats not openFoam ;-)
Great. I want to do this simulation on golf ball , Do you send tutorial of this simulation?
thanks this helps me with my mental experiment thinking about twin jet nozzles emitting converging propulsion force to crash inro each other behind a fighter plane, and wondering if it would be tantamount to sonic the hedgehog
That's definitely a interesting idea, even if I donot understand it :-D
Hey I have a question, I have a small propeller about 18 inches. The rotational speed is about 6710 RPM. And I swear I've searched a lot on the topic but there is not actually much info on how to properly simulate a propeller. So far I have tested Reynolds Stress Eq. K-omega and K-epsilon. However I really don't know which is better for my needs. I have used a lot of K-omega for fixed wings at low levels of aoa and low reynolds (about a million Re). Which method should I use for rotating propellers?
Hey Mich, thats the beauty of a real project. Nothing is for sure which on the other hand give a bit of freedom. I would start with a RANS model using k-omega because it is usually pretty stable. It's always possible to increase the amount of work that goes into a model. In the beginning of a project I usually the try to get a solution with smallest effort in time possible. A LES will take much more time than a RANS simulation.
A validated model will always need measurements, which are in 99% of the cases to expensive to carry out
Great video
Wait until there is an AI model which can increase efficiency by getting the mesh state equations close to reality, then allowing DNS to converge each time step. I feel AI will learn how the NV equations work, and this will make our lives amazing as sim engineers.
Hey Brandon, that's a nice idea. However, I'm not so sure if it will be that huge improvements. In comparison we engineers are already using things like optimisations in different forms and from my point of view the use cases are limited. Of cause you can improve to a level which can be reached by trial and error. 95 % of my work to make good assumptions and to know where to look for problems not so much the numerical or geometric improvements. Often we know very little of the boundary conditions and the operator. The algorithms are always limited to the data you feed them with and I assume that AI will lead probably to an even increasing false assumptions because they will generate very good results to poor boundary conditions and will lead to a false interpretation of the results. I have witnessed this on several conferences where results of simulation reached 0.1 % of error but the error (or uncertainty) of boundary conditions was >20 %.
What if the AI algorithm wrongly predicts an outcome? Who takes liability of a failure? The AI developer or the engineer?
Can you tell me how to visualize flow like your video (at 3.03s ) in Ansys??
Hey Mehedi, I don't know how you do it in Ansys. However, I can tell you how you do it in paraview. You can export your result data as vtk and open it in paraview. Then you clip the data by scalar (magnitude of the velocity) of the velocity.
A other option is to use semi-transparent slices in your simulation results. You can use the higher velocities to be transparent.
nice one, but if you could show how to do it from the beginning that would be appreciated from the engineering community!
Nicely put!
Thanks Nada :)
Hey great vid! Liked and subscribed!
Great content!
Thanks Vivi K, for your feedback.
please give me th link for installing ANSYS
Bu kadar fark ettiğini görünce şoke oldum
Sure mate
It must be URANS and the flow pattern beneath the car is very bizarre.
What do you mean by a bizarre flow pattern?
@@engineerdo6568 Beneath the car, the velocity in the car-adjacent cells should not be zeros?
Good Work👍
Thank you! Cheers!
Porsche 1964, this is a good car.
nice one, but if you could show how to do it from the beginning that would be appreciated from the engineering community!
Will do
@@engineerdo6568 I am subscribing your channel. I hope to see it soon