Hi,

I have worked with CFX (5.5) for some time now on internal flows, like ventilation of buildings.

Often I have to use small physical time steps (0.01-0.001)in order to achieve convergence ?

My problems are well define, mostly retangular geometri and I uses inflated mesh on the walls. The flow are in general non-isothermal. Mesh size are close to 300.000 elements.

Any advise is welcome !

Kind Regards

Roued

Hi

Adding to my first postings.

Look at the velocity vectors. They are all pointing in negativ Z-direction (vertical), except at the inlets and outlets.

Regards

Roued

Hi Roued,

A small physical timestep will not always help. Tipically, you should set your timestep to be equal to the advection timescale for your problem. If you do not know what the advection time is, just look at the timescale output by the solver in your .out file.

Regards, Robin

Hi

Yes I know all about the advection time scale. But it has been impossible to get convergence with this size of time scale.

Most of the time I have to use small time scale, not only at the beginning of the solution process. But thoughout the iterations.

One reason is the difference between the size of mesh elements.

Close to the inlet section the smallest dimension of the mesh element are about 0.005 m and the largest is about 1 - 2 m.

If I use a large number of element, thus making the overall mesh more uniform in size, the convergence is better. But the difference between a coarse and fine mesh is between 5 - 6 times. (Coarse mesh: 300.000 elements - very difficult convergence => small time step, Fine mesh: 2.000.000 elements - better convergence)

But it is not always reasonble to use 2.000.000 for internal flow problems !

Please comment !

Regards

Roued

Hi Roued,

How quickly does your mesh expand from the small scale to the large scale? If the expansion ratio exceeds 2x, it may be causing your grief.

Due to the coupled multigrid solver, grid size does not greatly effect convergence. A fine grid should show the same convergence as a coarse grid, if the mesh quality is the same. If you introduce large expansion ratio's or other poor mesh characteristics, the numerics will suffer.

Robin