Hi Every body, I am working on DNS of Diffusion flames, in this i am getting problem at Boundary conditions, let me explain about my problem: I am taking the LES data as Boundary conditions to my DNS, but this LES ,velocity components are becoming nagative, so what can do to over come this problem,

Thankyou in advance

hi, do you use periodic boundary condition or something else? If you use periodic BC, then in fact no BC is needed, if not, I wonder what kind of method you employ. Thanks.

Hi George, Thankyou for ur advise, I think I didnt explain my problem very clear. now i will explain with diagram:

*********************** * * *********************** above diagram is my domain for LES of diffusion flame, its length is 20D where D is the nozzle diameter of the jet. now i want to do DNS of the flame portion wich is in between 3D to 17D, for this i am taking the data of LES at 3D and 17D,which will be used as BC for DNS. while taking this LES data , i am getting some Velocity components nagative at 3D and 17D, this is my problem.. offcourse in y direction(along the width of the flame) i am taking periodic B.C

I will be happy if u give ur suggestions and ur help in solving this problem. Thankyou

hi, mahiboobswami, glad to see your description. Though I'm still not very clear with your project, I try to give my suggestions:

1, Since you do DNS in the middle part of your domain, the mesh is more fine than the other two parts, is it right? maybe this jump from corse to fine mesh cause some problems in finite difference;

2, I guess you are using time marching to obtain stable solution, is it right? If so, how do you provide the BC every time step? Maybe you just stored the LES results at two boundaries beforehand, and I think it is maybe good for the inlet(3D), but may be not ok for outlet(17D) since the outlet turbulence structure can not be preset but computated.

Maybe I'm wrong, sure we can discuss more about it since I'm also doing something about LES now.

Hi George,

u understood my problem, what ever u r telling is right, i mean I am taking the more fine grid for DNS, i am doing interpolation of LES coarse grid data to DNS fine grid . second one is, it is time marching, if i run my LES for 20,000 time steps , i will run my DNS also for the same number of time steps , for each time step of my DNS, it will read the corresponding time step of LES data. thankyou for ur suggestion, i think now my problem is clear to u

hi, mahiboobswami, now I seem to understand your problem.

Thus I'm almost (90%) sure that you cannot use those LES data at outlet (17D). The reasons are:

1. LES, as its name implies, only large scale eddy are simulated, those more delicate structures are modeled by SGS stuff. However, DNS sure give out some more delicate structures and this incompatiblity may cause a lot of trouble.

2. I do not know if you just directly use those LES data at outlet or not. Because according to theoretical analysis, in subsonic system, at the outlet, you suppose to impose one variable.

My suggestion is maybe you can try other kind of BC at outlet. According to my experience, NSCBC method is a hopeful choice. But before you do such a big change, maybe you can check your code again. I guess that you use ghost cell. So just trace those ghost cell to find if they give unreasonable values since I had made such kind of mistakes.

If you decide to try NSCBC, one thing may need to notice: you should incorporate this method into your code rather than calculating boundary values seperately. I do not know why, but my experience is like this.

Good luck!

Hi George, Thankyou for ur suggestion, i am also thinking of NSCBC, one thing i want to know is, presently i have code which is running for LES of Combustion(diffussion flames), BC used for this is Dirichlet and Neumman BC. i want to know, if i remove SGS model from this LES code and increasing the grid ressolution very fine(approximately nearly equal to Kolmogorov scale), will it work or not,,

my main goal is i have to do 3D , DNS of Diffussion flame, Please give ur valuable suggestion on this,,if it works I am the lucky guy, I can save much of my time..

hi, mahiboobswami, my opinion is that your code sure can used for DNS, however, I'm not sure if those B.C.s can be applied or not. Because though my case is a rather "pure" one without chemical reaction, the Dirichlet and Neumman BC still cannot be applied (at outlet). The reason is that, the Dirichlet BC will either eliminate the fluctuation at boundary which should be there if a fix value is imposed or give incompatible field with that of interior domain if stored data are inputted (at outlet). As for Neumman BC, I only tried extrapolation (1st order and 2nd order) and found that the code converges really slowly and may blow up in more time steps.

After all, you can try. Esp. Neumman BC. I'm sure your code can be used for DNS and I'm curious if those BCs can work or not. Would you pls tell me your trial results? I mean if they work or not? Thanks ahead!

Hi,

George, I have allready started my trial by removing SGS model with Neumman BC. the estmated simulation time is 35 days, still now it is twenty days is over,daily i am observing instanteneous results, they are fine, but i am also having doubt as the time step increases , it may diverge,, but i have positive belief on this and i am also eagerly waiting for this results, i will inform u soon after getting my results. are u working for ur PhD or doing job, my self is doing my PhD at Germany, basicall i am from India. thankyou for ur valuable suggestions.

hi, Mahiboobswami, glad to know your code works. I'm a Chinese and is working for my Ph.D. too. So good luck and I'm longing for your results.

George