[luk1]

Hello,
I am running RANS simulations in a vertical Backward facing step domain and I am using CFX.

It is a counter-gravity flow and behind the step, there is a heated wall in order to get mixed convection.

Of course, I expect a steady-state solution but I either find backflow at the outlet from time to time or the whole time. Only about 0.2% of the area and 3.5% of the faces are affected by backflow. In the picture attached you can see where the velocity in streamwise direction is < 0.

The settings are as follows:
• Boussinesq buoyancy (it is valid here for sure)
• „Thermal energy“ equation for heat transer
• SST model
In the out-file I can see that the momentum equations are not solved sufficiently and not as good as the other equations. --> see picture. Also, it shows only „ok“ for those and not „OK“. The assymetry of the flow and T field is comparatively strong (left vs right side of the channel in streamwise view).
The problem does not occur when the heat flux is reduced by 90% (transient convection) or 99% (forced convection). So I am pretty sure the buoyancy must be responsible for this issue.

What I did so far:
• I have doubled the length of the domain behind the heated wall and I still get the same results. The backflow at the outlet occurs in the same cells.
• I have varied the buoyancy reference temperature: inlet temperature first (300K), then the average (310K) of inlet temperature and highest temperature calculated in the simulation.

Do you have any suggestions what else I could try?

[ghorrocks]

I think you will find your initial assumption of a steady state solution is incorrect. Buoyancy simulations frequently result in large scale transient flow features.

The back flow you are getting is probably correct: The hot, buoyant fluid will rise on the heater side. To conserve mass (and volume as this is a constant density simulation) there needs to be a region of reduced flow, or if the buoyancy is enough (or the through flow rate low enough) a region of reversed flow.

So everything you report appears as expected. My suggestion is you will need to do this as a transient simulation.

[luk1]

Thanks for your reply!

As I wrote, this simulation was done using the SST model. I ran a similar simulation for almost 2000 iterations and I could see an oscillation of all equation residuals and I got unsatisfying results (I checked the U and T profiles, my monitor points and some characteristic key numbers like skin friction coefficient along the wall).

In a simulation where I use Reynolds stress models (BSL RS and Omega RS model) all the equation residuals plunge down below my convergence criteria (RMS 1e-6) and remain roughly constant from this point on. I received proper, reasonable results there.

I also tried it for both models that are combined to SST, say the k-eps and the k-omega model. I found good results and quick convergence for the k-omega model while the k-eps failed to reach my convergence criteria and it was visible that in the flow and temperature field and so on.

So I assume that everything is working fine in the near wall region and that the problems occur in the free stream region. But I don’t know what might be the reason for such a behavior and how to fix that (if possible). Do you have any ideas?

Additional information: The mesh I am using is considered to be independent after I did a mesh independency study with Omega RS. And I assume that there shouldn't be problems for SST due to the mesh quality if it is sufficient for a RSM.

[ghorrocks]

So I assume that everything is working fine in the near wall region and that the problems occur in the free stream region.

Why assume? You should use the post processor to check this directly and then you do not need to assume.

But I don’t know what might be the reason for such a behavior and how to fix that (if possible).

I do not know what behaviour you are talking about, or why you want to fix it. That is why the reason for you doing this simulation is important. Why are you doing these simulations? What are you trying to learn?

But so far all I see is some differences between the different turbulence models. This is normal, each turbulence model has its own strengths and weaknesses - and you would expect differences between them all.