I have been trying to model the flow in a very thin channel using COMSOL Multyphysics. The laminar flow module does not converge no matter how I change the meshes, or use ramping loads and such. The k-epsilon turbulant model converges however, when I plot the CellReynolds the values doesnt exceed 5. Streamlines with turbulant viscosity dont seem fine. The cellReynolds and streamlines are shown in the pictures. What do you think can be the problem?
Dear/Morteza
If the flow is laminar so you do not need turbulence model just use DNS directly. But if the flow is turbulent you can use one equation turbulence model with fine mesh and over relaxation factors for stream wise velocity components
Perhaps in your laminar model, you need to add isotropic diffusion term
Your question is posed in such a way that is difficult to help. Provide details about your flow problem: dimension of the geometry, characteristic flow parameter, fluid.
Furthermore, your figures say nothing of really helpfull. When you wrote "convergence" what do you mean, convergence to a steady state? In a deterministic sense the flow can be time-dependent also if it were laminar. On the other hand, in statistical sense the RANS formulation is steady by definition so that, depending on your problem it can be used or not and you should switch to URANS.
Dear Morteza, few considerations:
1) The behaviour you find is no surprise: RANS model are design to provide easy convergence, whereas Navier-Stokes is intrinsically prone to develop the Richardson cascade, and therefore, a steady solution is not possible. The small structures can have the same wave-numbers of the p and q waves, which can trigger numerical instabilities.
2) Try to repeat your simulation at Re = 1 - 10. It should converge (somewhere) in that regime. If it doesn't there is a serious problem, or your grid elements are highly skewed.
3) run away from COMSOL fast. It works on very simple problems, but it is terrible for fluid dynamics.
4) Use a proper CFD code, I recommend OpenFOAM.
Good luck
First, how are you are defining the Reynolds number, Is the length of the channel part of it ?
Thank you everybody for your help. This was very helpful and gave me lot to think about and study.
Mohamed Jardak are you asking me about the definition of the reynolds number or how I define that in Comsol? Comsol has it intrinsically, and I just chose to plot the reynolds number contours. As far as I know reynolds number is not concerned with the length of the tube directly.
Andrea Boghi Thank you for the notes. Well I chose Comsol for its multiphysics capabilities. The other notes I will consider.
Aren't you solving the incompressible Navier-Stokes equations ?
If so, Comsol must account for a Reynods number or to
other parameters that are related to it. Sorry, I never used Comsol to be able to help further.
Low Reynolds number flow does not necessarily imply steady behavior. There can still be flow instabilities. Like a cigarette plume (Low Re), which will break up at some point anyway.
It seems that there are some free shear layers in your flow, these are inherently unstable and will induce some kind of unsteadiness in your flow. You need to either resolve this unsteadiness or model it with a turbulence model (which are generally not suited for low Reynolds number flows).
So you need to do an unsteady simulation, maybe even refine the mesh to resolve all flow details, use a smaller time step, those kind of things. If you don't, the problem will probably diverge.
Dear Teymoori
I don't know Comsol. But based on my experience on CFD I guess your problem does not deal with flow regime. The geometry is not clear for me, however, I have some suggestion for you:
1) First of all, you should prove (at least for you) that the flow is laminar or turbulent (Based on Reynolds number and so on).
2) Check your boundary conditions. There may be an ill posed boundary condition problem.
3) The problem may be inherently unstable and shall be solved transient. I guess this case is more probable. Since you succeeded to solve the problem by adding artificial viscosity (i.e. turbulent viscosity) which means that the problem is stable now. To check this case, I can suggest to solve the laminar flow in transient form (time dependently)
4) And as the final point, I am not sure about your interpretation of convergence. In some problems, the residuals can not be as small as the targets. However, it doesn't mean that the problem is not converged. Some source terms in equations (like buoyancy force included in momentum equation in natural convection) prevents residuals to get very small. So, judge also the solution to be if it is acceptable or not.
Hossain Nemati Thank you for the valuable input. I will surly consider your points.
- Badges
- Science topic