Dear all, I am working on CFX to model a laminar flow through a micro channel with a diameter of 1 mm. Because of the narrow channel and a high viscosity the velocity is very low (<0.02m/s). I am working with Reynolds numbers about one. For Reynolds numbers higher then 3, I had no problems achieving convergence. By decreasing the velocity, the RMS values aren't running below a value of 1e^-4. This problem seems to be grid independent. I increased the number of elements with factor 80 and still had the same problems. Because of the solution of those not converged runs showed arbitrary behaviour of the flow field I think altering the convergence criteria is not a good idea. Maybe someone has an idea how I can handle this problem with CFX. Or is there any other possibility to solve it with a different method (e.g. DNS)? Thanks

Matthias

Hi Matthias,

Assuming your grid is fine enough, you are already doing DNS. Running the solver in double precision may help.

Regards, Robin

Hi,

Or to put it another way, your Reynolds number is so low that the flow is laminar with no turbulence. Therefore there are no turbulent structures to resolve so DNS is not applicable. If your simulation is mesh independant and fully converged then a laminar simulation should be very accurate.

As Robin says, try double precision numbers. That is certainly the first thing to try.

Glenn Horrocks

Hi Robin and Glenn, In connection with the same problem i would like to ask what should be done if double precision is also not working.Let me make my problem more clear: I am simulating a backward facing step with full geometry(NO SYMMETRY) with inlet ,outlet and wall.The level of grid size is less than what people have used for LES. Now my problem is also after switching on the double precision , the solution is not converging and oscilatting between 10^-3 and 10^-4. Why it is so?Half of the geometry putting symeetry converged well for the same reynolds number. Please suggest.

Regards

Manu

Ok, with double precision it works perfekct. Thanks PS.: Manu, I would suggest your grid is not accurate enough and symmetry is easier to solve, because the boundary values are constant (in angle direction) for every iteration step. Without the symmetry they might change from cell to cell for each iteration step to the next step which comes from the less accurate grid.