[stuntmanmike]

Hi all,

I am currently simulating turbulent diffusion flames in CFX for my bachelor thesis and I am encountering some convergence problems.
I have a cylindrical, structured mesh (see pics attached) with an inlet (2cm diameter) at the bottom. The Mesh has a diameter of 2m and a height of 10m.
I am injecting pure propane with an inlet velocity of 250m/s. I am using the standard k-eps model, the eddy dissipation model, the discrete transfer model and the magnussen soot model.
For transient simulations with a time step of 1e-4s my momentum and mass residuals go below 10^-4. The species residuals, however, remain above.
The flow looks almost steady in CFX-post.

When I am doing steady simulations, however, I get quite bad convergence (all residuals are above 10e-4 and oscillating ). I am using a physical time step of 0.25.
Although, the result looks reasonable to me. From my understanding I should also have good convergence for steady simulations, as the transient flow reaches quasi steady state.
Am I right? Maybe my mesh is bad (although it seems ok, from my understanding)? Perhaps, someone could enlighten my about that? I would be very grateful.

[ghorrocks]

This FAQ discusses this issue is more detail:

http://www.cfd-online.com/Wiki/Ansys...gence_criteria

Yes, there are cases where a transient run converges to a steady solution but a steady state simulation fails to converge. This is the sign of a very numerically unstable simulation. Combustion models are well known to be numerically unstable so this is not a surprise in your case.

[Opaque]

For the steady simulation, I would reduce the timestep a bit more, say 5 to 10, and see at what value the residuals settle now. I would restart from the previous run you did using 0.25 [s], not from a cold start.

[stuntmanmike]

Thanks for the replies!
I reduced the physical time step to 0.05s (was that what you meant, Opaque?) and continued the steady simulation for another 150 time steps. Unfortunately, it doesn't help much. The residuals are still not dropping. Just the oscillations become a bit weaker.
Maybe it's just not possible to get a steady solution here? One thing, that I thought of was, that maybe the large difference of cell sizes in my mesh leads to difficulties. As the inlet is very small, compared to the total mesh, the O-grid produces very small cells at the inlet, while the cells at the boundary of the cylinder are much bigger. I tried to get a smooth transition but maybe that's still a problem? That is basically the only thing I can think of because I doubt that there is anything wrong with my physics.

[ghorrocks]

If you read the FAQ I linked to it suggested increasing the time step size, not decreasing it.

My first post said:

Yes, there are cases where a transient run converges to a steady solution but a steady state simulation fails to converge. This is the sign of a very numerically unstable simulation. Combustion models are well known to be numerically unstable so this is not a surprise in your case.

And that remains the case. The numerical instability is caused by the rapid gradients caused by the combustion (as combustion usually has a very sharp combustion front). Anything with rapid gradients is hard to model. So this model is always going to be difficult to converge as it is a combustion model.

[stuntmanmike]

I also tried increasing the physical time step to 1s. It doesn't help, though. I also tried the local time scale factor, as suggested in the FAQ and the residuals went down. But when I return to a physical time scale for the final steps it goes up again. So this also doesnt work.

So, as ghorrocks said, it might be not possible to have the steady solution converging. I guess I will leave it at that then and stick to the transient simulations. Thanks for your help!!