Dear Pals,

I am simulating a duct in CFX. it is a inverted L shape duct. the inlet is located in the horizontal duct and the outlet is in the vertical duct. the vertical duct is bigger in size when compared to the horizontal. my problem is when the flow is coming out of the horizontal duct, circulations are formed at the two sides of the vertical duct as the velocity of my flow is bit higher (30m/s). so the solution is not converging it is keep on oscillating in 10e-3. i am using k-epsilon turbulence model and even i tried SST k-omega. can anyone help me to resolve this.

hi abdul! Do u have a mesh adapted to your turbulence model? if i were you i will try to use an adaptated time step at first. Ansys published a very interesting doc concerning fluid flow separation. try to find it. by

A few more details about how good your mesh is and about your boundary conditions would give more options.Try refining your mesh a little at the corner where the two ducts meet!

Thank you Omer for your reply. my mesh quality is around 0.35 and the min angle is about 41. i am using pressure inlet and massflow outlet boundary conditions.

Try using a very low physical timescale. If ur still having problems with convergence try a local timescale factor. But to make sure you have achieved convergence, swithch back to an auto timescale and if there is no change in your imbalances and mass flow then your case is converged. Your problem could be due to large difference in velocity timescales.

Pressure inlets are generally not a good idea. Is there a way you can have massflow or velocity inlet, and pressure outlet? It would really help you with the convergence.

hai omer i tried that conditions too but the result is same. it is oscillating in 10-3.

Trying changing your autotimestep to a large physical timestep.

It might work...

Hi Omer, Why do you think pressure inlet is not a good option compared to mass flow and velocity b.c.s.

[Daniel Frederik]

Sometimes convergence is not achieved even thou the mesh quality is ok. That can be due to transient effects in the flow. Coarsening the mesh can help to make it converge because the sransient effects will not be detected by the mesh. The resukts obtained this way have to be treated with caution. In my opinion the best way would be to do a transient analysis in this case.

[Meister]

I would try different timesteps in a user-function and run every step for lets say 100it. Starting with the largest step (e.g. 0.1) to the smallest (e.g. 1e-6). You can easily see if there is a significant change in the convergency. There is also some features to force the simulation in a steady-state solution but probably it is best to do a transient simulation and do a average over time after.

[JuPa]

The original post was made 7 years ago. I'm sure he's achieved a converged solution by now assuming he's not using a timestep of a picosecond.

[Meister]

uhhh...so he should better change to the latest Version in CFX as well - they have better convergency