Hello,

I am busy looking for optimal inlet and outlet boundary conditions for a fully 3D compressible LES-code. However there is not that much literature available on the subject. For the moment I am testing on a Backward Facing Step. All tips and hints are welcome.

Thanks,

Jan

(1). What is the optimal Inlet and Outlet conditions? (2). Try to keep it simple at this stage, move the inlet away from the step first. And make sure that the outlet location is very far downstream from the step. (3). In the laminar flow, the separation bubble length is a function of Reynolds number ( can become very long). In the turbulent flow, it is about six step heights. That is of order 10, so, try a number like 20 step heights. And also make the outer boundary at least 10 step heights. (4). The actual boundary conditions should be consistent with the transient method used.

The optimal outlet boundary condition would be the one with not too much influence on the upstream flow. What do you mean with 'transient method'? And how do I know whether the B.C.'s are consistent with it?

(1). Transient method means the un-steady state equations. (2). Boundary condition consistency means the variables used with the boundary condition must be consistent with the variable used in the equations and the equation itself. You don't want to apply some arbitrary artificial conditions at the boundary.

Any tips where I can read for good introduction to boundary conditions for navier-stokes solvers?

thanks for the responses,

Jan

That's a bit difficult. The problem is that the specification of the inlet turbulence intensity and length-scale (if you've got it) is not really enough. What people usually do is to add a certain amount of white noise with the right turbulence intensity to the given inlet profile and pretend that this is "turbulence". The problem here is that the effective length-scale is actually the (inlet) mesh size and the stuff you put at the inlet as "turbulence" dissipates extremely quickly (low l = high epsilon). I heard some people (Stanford) trying to introduce the correct correlation scale through wavelet transforms at the inlet plane (expensive and a nightmare to put into the code), but they encountered the problem that, although the "stuff" now look like turbulence, it is not correlated (no energy cascade), and again causes trouble.

You've got two choices: 1) it turns out that for backward-facing steps it doesn't really matter what you put into the inlet (provided you put something with the right u'), so as long as you're interested in BFS, you're alright. 2) For more complex geometries, people take a solution of a fully developed channel flow and then feed it into the domain plane by plane, which allows the introduction of the right turbulence into the inlet. Expensive, impractical, but that is the right boundary condition.

I'd be careful with the idea of moving the inlet way upstream and hoping that turbulence would develop before it comes into the region of interest. This may take quite a while (like ~20 channel heights); LES is expensive business as it is and you can't afford to be stingy with mesh resolution far upstream. The cost of your computation for a case like that may easily go up by a factor of 10 or so.

(1). Solving flow over a backward facing step is always much cheaper than shooting at the moon looking for water. (2). I think LES is a fashion like a day-trader hoping to make big money with very little investment. It should keep the super-computer very busy for a long time. And as always, they will discover something useful "by accident". (3). I would say LES is very good for young people also, it should keep them away from shooting at each other. Even when they are sleeping, they are still dreaming about whether the flow is just random or really random.

Hi, 2) if you think LES is a fashion why is so many people using it ?? Because DNS is better, no not always. So time two-equations models does not performe very well for some flow. Like dispersion of gas around a building, mixing process and dispersion of gas inside building with non-isothermal flow. Fire modelling is another case. 3) You do not need a supersomputer just because you are using LES. A high-end PC or workstation will do ! It depends on numeric and implementation. You probebly know why the peak performance of PC and Workstation are very hard to obtain ? (memory vs processor speed) (Like supercomputers - it you don't treat them well they will not behave) New implementation will let you do 200 Mflops on a workstation that you do not have to pay 20M$ or more for. When it comes to LES, one should remember the physic. Of couse it you will always have 2. order statistic for your flow, Parallel computers will help.

regards

Jens

I think you guys started chewing on something interesting here. LES is a nice method in principle, telling nice stories about energy cascade, resolution and subgrid-scale models, but my personal problem here is the boundary. From previous messages, it is obvious that even something as simple as inlet boundary condition is a major pain. And we didn't even start thinking about wall yet (energy cascade going the opposite way, thus breaking the very principle of LES). We simply don't have the money to resolve the boundary layers to the appropriate level for anything more complex than a box in a box. I would personally stick to time-resolved RANS for a while, at least until Speziale and Co. sort out the mathematical arguments behind "uniting" LES and time-resolved RANS (see eg. Speziale, AIAA Journal 36:173-184, 1998). It seems to me that at the moment that we simply don't have the maths for it.

Hi, again It depends what you need! If you are interessted in overall flow features, make the inlet boundary condition simple, -> use a velocity profile superimposed with white noise(random) corresponding to the right turbulence level. If you need better agreement for 2.order statistics, make the inlet channel longer. If you need good 2. order statistics, (the is big!)do a LES of the inlet flow conditions and supply the instantanious velocity profile to the second LES where you do the simulation of backward facing step.

Boundary at the walls: go for the same structure. 1.Think about how many grid point you would use. Put most of them where the flow is complex. (at the wall or internal obstracles) 2.use simplified boundary condition for rest of the boundaries. van Drist wall damping, Log law/Power law profile or perhaps One of the better wall function by Piomelli or those developed at CTR. 3. Best boundary conditions: Resolved all wall boundary (== many grid point) But unless that you are interested in some special feature close to the wall, you would not see majure difference in the overall flow field. Since most flow is not complex in the hole flow domain.

regards

jens

(1).First of all, fashion means popular and change-in-time. It is somewhat consistent with your observation. (2). When something is not a fashion, it remains unchanged and becomes classical. (3). If one look at LES as a natural extension to the transient flow calculations, which has occupied many hours of CFD researchers time already, it shouldn't be surprised that LES is a fashion. It can "easily" put his transient flow calculations into the real world, which is a very important step. That's why I said big return with very little investment. (4). In order to capitalize this opportunity, all one has to do is to find an ideal application problem to show the advantage of using the LES approach. Such opportunity definitely has a much higher probability of success than shooting at the moon looking for water. As long as the water is essential, no one will worry about a slight inpurity of man-made turbulence in it. (5). So, for such problems, the use of the LES definitely will continue. (6). On the other hand, I don't think it is a simple task to simulate the initial conditions of a turbulent flow, including the inlet conditions. The situation is identical to the problem of simulating the stock market. It is nearly impossible, because of the large number of people involved in the market place. (7). It looks like that the Reynolds-averaged-equation approach is similar to the study of the stock market index trend, while the LES approach is something like insider-information scheme which can be quite profitable sometimes. (8). Perhaps, before one can specify the initial turbulent flow conditions, it is necessary to study the parametric behavior of the turbulent flow first, that is a systematic representation of the random flow field (inverse of the Reynolds-averaged solutions). (9). For those who are already in the stock market, the insider's information is always very attractive to them. So, if someone has already shot at the moon for water, relatively speaking, the LES approach is nothing but the transient flow calculations with a flavor of day-dream (or nightmare depends on the day of the month).

(1). About the inlet boundary conditions, there are two possible configurations. (2). In one configuration, one can place a flat plate with a sharp leading edge in a free stream at some turbulence level. In this case, the leading edge should take care of itself. The free stream condition should include the turbulence in the transient simulation. In other words, the time-averaged TKE and the velocity profiles should be uniform. (the transient inlet flow is not uniform) (3). In the other configuration, the inlet station is placed somewhere on the flat plate where there is already a boundary layer. In this case, it is not known how to simulate the transient velocity profile such that the time-averaged flow parameters will be the steady-state profiles. For this reason, it is easier to take the first approach. (4). So, for the problem of flow over a backward facing step, it is easier to place the problem(finite length flat plate with sharp leading edge and a finite length downstream flat plate after the step) inside a large uniform flow environment. In this way, it is easier to specify the inlet free stream, far-field, and downstream boundary conditions. The downstream will return to the free-stream at a large distance. In this case, a single far-field condition should be enough. Otherwise, the specification of boundary conditions would be rather complicated.

Hi Dr. Jasak,

Just an information: Prof. Speziale passed away in April 1999. (Please look at www.icase.edu in the reasearch quarterly article) I wish we all could have seen that answer from him. Indeed a great loss to our research community

Mayank