|
[Sponsors] |
[LES]how to learn large eddy simulation from scratch alone? |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
October 29, 2013, 08:41 |
[LES]how to learn large eddy simulation from scratch alone?
|
#1 |
Member
garry
Join Date: Oct 2013
Posts: 41
Rep Power: 12 |
dear all:
i am a first year graduate. my teacher let me learn les now, but there is nobody had learned about it in our research group. i have learned cfd and les about two months, but the progress is little. now i am reading the book "Large Eddy Simulation for Incompressible Flow", and i realize that from scratch to les code is a long way. i want to know how can i learn les fast and what i should do now??? |
|
October 29, 2013, 09:11 |
|
#2 | |
Senior Member
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,839
Rep Power: 73 |
Quote:
The learning processs of LES is not simple, in no way can be fast, it requires time and effort. You must start having good basis of CFD, mathematics, fluid dynamics. Attending some courses is mandatory, also because you can have contacts with people and compare yourself. Generally, my opinion is that preparing a person in the LES field is a PhD contest. |
||
October 29, 2013, 09:11 |
|
#3 |
Senior Member
|
My opinion is that there is no fast in learning LES. It is a field where numerics meets physical modeling leading to infinity^2 possibilities. Also, there is a lot of personal flavour and much less detailed explanations. The basis, of course, is very simple: you have to resolve any dynamically significant scale.
You can start by reading Sagaut, Lesieur, Garnier, Pope... but then you have to go back and start again, and again while getting practical experience. Still, the basis is as simple as adding a grid based mixing length model to a non diffusive 2nd order finite difference code for 3D unsteady Navier-stokes equations. |
|
October 29, 2013, 21:26 |
|
#4 |
Member
garry
Join Date: Oct 2013
Posts: 41
Rep Power: 12 |
FMDenaro and sbaffini
thank you. in my opinion, after a general understanding, learn les from a simple code may be better for the understanding of the process. is it right? |
|
October 30, 2013, 03:29 |
|
#5 | |
Senior Member
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,839
Rep Power: 73 |
Quote:
No ... studyng an LES code you learn only how one formulution is implemented, which is, actually, only a step in a learning process of LES. As Paolo said, when you try to implement a simple LES code, it is common you need to come back to some theoretical questions and so on ... To tell the true, if you want to become just a code user, then just learn how using Fluent by the handbook. Learning LES is a totally different issue. Good luck |
||
October 30, 2013, 04:49 |
|
#6 |
Senior Member
|
I'll make an example to elucidate on the matter exposed by Filippo. One of the most used benchmark for LES is the turbulent flow in a plane channel at some friction Reynolds number Re_tau.
Now, you would be tempted to do LES of this case with some SGS model in some code. However, what usually happens is, more or less, this: 1) LES of the same case without a SGS model and without explicit filtering is strongly dependent on the specific code adopted. i.e., some codes give you underprediction of some quantities, some other will overpredict, some will be just as good as a DNS. Adding a SGS model of a certain type, of course, will help some codes but not some others as their original behaviour is quite different. 2) At this point, some theory guy will tell you "Hey, this not LES, this is just crap... you have to filter your equations explicitly". Then you answer "Ok, what term should i filter?" and he will tell you: "Uhmm, it depends, you can filter your convective term in the momentum equation, or all the non-linear terms (what about continuity equation for density based solvers? Is it linear or not?), or maybe everything (ADM)" 3) At this point, if you are a novice, you should already be pretty confused, but you decide to go along and apply some filter to some terms. For a fixed code, the results will still change according to your filter. You then go back to the theory guy with something like: "Hey dude, what the f...?" and he replies: "First of all, is your filter symmetric as your convective term discretization? Is your filter conservative? Does it commute with your numerical derivative operators?" These are, mostly, conflicting requirements but, as a confused novice, you don't even understand all of them. 4) As you are gonna do something, you decide ""F... symmetry, i'll go commutation error free as in Stanford (how could they be wrong? they invented LES!!!)". So you pick up a filter and a SGS model (let us assume that your filtered results without SGS model somehow convince you enough to not change the filter width). Which SGS model? "Everybody is talking about this Dynamic Procedure, how good it is for the near wall behavior, let's go Dynamic Smagorinsky!!!" 5) After your first attempts, the code keeps crashing with NaNs. Then you go back to the theory guy, who says:"Well, you should be aware that you need some clipping, averaging of the dynamic constant to do the job". "Which one?" you reply... "Well, i use this" he says, adding "...and remember to compute the filter widths correctly". At this time you usually start panicking but, at this point, it's too late to go back RANS and you keep going. 6) After a while you keep working on that, you might be able to produce some results but then you freeze in realizing that, especially for the channel case, they might be completely out. If you still have any reason to go over, you call the theory guy at the phone, who will finally tell you: "You should be aware that there are good and bad SGS models just like in RANS. Also, this case might require some pretty fine grid to get good results. Finally, keep in mind that filtered Navier-Stokes equations are different from the Navier-Stokes equation and depend on your filter, there is no reason for your results to agree with DNS ones." While a huge "REALLY???" is coming out from your mouth, you hear some voices/noises from the phone and ask: "Dude, where are you, i can't hear you very well". And he says: "I'm with some chicks on the beach. Good luck with your LES code. Bye". Of course, with this story i was kidding, the theory guy never ends up on the beach with some chicks. |
|
October 30, 2013, 05:07 |
|
#7 |
Senior Member
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,839
Rep Power: 73 |
... Great! Today wheater is good to go on the beach |
|
October 30, 2013, 05:49 |
|
#8 |
Senior Member
|
You can think of LES as assembling IKEA stuff. On the paper, it is very clear and you don't see any problem coming. Then you end up putting much more effort than planned, wasting most of your time, your hands broken. Every box is full of warnings, but you think they are for childs and you don't care. No single piece fits the others as expected. Even in the end, when the thing is built and it actually works, you have probably done something wrong, but it works, so who cares?
|
|
October 30, 2013, 08:18 |
|
#9 |
Member
garry
Join Date: Oct 2013
Posts: 41
Rep Power: 12 |
haha, funny man. But, it sounds really helpful. Thanks. There is an old saying in China--haste makes waste.
|
|
Tags |
beginner, cfd, learner, les |
Thread Tools | Search this Thread |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Y+ value for Large Eddy Simulation | yannossss | Main CFD Forum | 15 | March 21, 2017 23:41 |
Introductory books on Large Eddy Simulation | q.z. | Main CFD Forum | 5 | September 20, 2016 05:32 |
Large Eddy Simulation Question | Jing_min | FLOW-3D | 6 | May 23, 2013 02:36 |
Large Eddy Simulation with too high Mach number | Roland R | CFX | 4 | November 25, 2010 16:43 |
Large Eddy simulation | Andreas Hauser | Main CFD Forum | 1 | May 20, 2000 20:33 |