HI! There,

Is there any one who had experience in simulating stirred vessels?

I encountered the following problems:

(1) how to create periodic boundary conditions?

(2) In 2d simulation, the residual can be reduced to 10-3 or even 10-4. But in 3d simulation, the fluent can only reduce the residual from 1.0 to 1.0E-1, and can not reduce it any more, though the rotating speed is still slow (say 5 -- 10 rpm).

I don't know what's wrong.

Could any one help me?

Ray

(1). First of all, it is almost useless to watch the residual values. I have never use that in my CFD codes. I don't know why it is still being used. So, you will have to tell me the meaning of the residuals reaching 10-3, 10-4 or 10-1. If CFD is about watching the residual values, then you will be out of job before you even enter the job market! (2). So, don't talk about those meaningless numbers, instead, talk about your mesh, your boundary conditions, numerical schemes, turbulence models, etc.... (3). For example, mixing means flow separations. In this case, which turbulence model did you use? (4). Apparently, you have to introduce something like a flat plate into the vessel in order to stir up the flow. And I am assuming that you are doing it by rotation. So, the simplest model you can build is to put just one blade into the vessel ( hopefully, it is cylindrical ). A flate plate rotating in a vessel is just like flow over a finite flat plate in Cartesian world. And everybody knows that flow over a flat plate will create separation bubble behind and also unsteady wake. (5). Perhape, it is a good idea to go back to the flow over a flat plate(normal to) problem first. The problem is almost identical to the stirred vessel problem. (6). After you have conquored the flow over (normal to) a finite flat plate problem, you can put it into a cylindrical vessel. And solve this simple stirred vessel problem. The periodic boundary conditions are reserved for the advanced users. So, there is no sense of using it right now. (or wait until you have solved the above mentioned two simple problems.) (7). By the way, don't think that flow over (normal to )a finite flat plate is easy to handle. When you have transient flows with separation, it must be very difficult.

Dear John, I think that the residuals do mean a lot in CFD computations ie: if you are interested in not only the gross flow patterns but also numbers such as Nusselt number say for problems involving fluid flow with heat transfer.

It is also a very good idea to mointor the resdiuals to make sure that your results make physical sense (some people might call this a purely academic exercise but let us not forget that CFD was born in universities).

Regards Mahesh

Beside the residuals one has to monitor several important flow variables during the calculation. Then one can see if there are still changes in the flow field. Residuals are just showing you if your going into the right direction. But they don't tell you when your computation has finished without any other information available.

The stirred vessel problem is 3d and transient. If there are also brakers at the wall one has to run it using sliding mesh or a "multiple rotating frames of reference" approach.

It is also very important to modell the real geometry of the paddel !!

(1). How do you express the physical sense or the Nusselt number in terms of the residuals, mathematically?

Your convergence problems with Fluent do not surprise me. I believe that Fluent uses a sequential solver which is not very robust. I am a Ph.D. student at Louisiana State University and I am working to develop a turbulence model to accurately model stirred tank flows. Your tank seems laminar so turbulence modeling instabilities are not affecting your convergence. Your convergence problem is either caused by poor grid refinement or the solver may be slow to converge on such a complex flow. If your tank is baffled, remove the baffles and try that geometry. You can also try different initial conditions. Try using an initial condition where the velocity is initially rotating at the impeller speed. Next try an initial condition where the initial tank fluid is quiescent. You can also try solving the problem steady state instead of transient to see if that will converge. You may email me if you would like to discuss this further.