Room heated by a radiator not convering
 

Hello,

I am trying to simulate a room heated by a radiator. There is no mass exchange and the boundary conditions are shown in the image below.

I have built a structured mesh that is refined close to the boundaries and in my opinion is reasonably good.

However, when I simulate it, there is no way that it is converging. I have tried using all pressure-velocity couplings and different turbulence models.

Does anyone have any idea on where can the problem lie? Thanks

Natural convection problems are notorious for not converging to low residuals values. I would imagine the turbulence models would have basically no effect on convergence. For example, IcePak and the old AirPack use a zero-equation turbulence model when solving similar problems. I am not sure what the second image shows, as it is mostly black.


First step would be to look at the computed flow field. Does it make sense? If not, this can potentially help identify problems. Also look at contours of the residuals. These will help to ID places where things are not adequately resolved or where recirculation areas are not going to allow the solution to converge. Sometimes it makes sense to compare a flow field and a flow field at several hundred iterations later. If they are basically the same, then you are achieving convergence.



A better approach to natural convection problems is to set monitor points on important areas or points. For example, in an electronic assembly, we would create a temperature monitor point on a chip to see that this value converges. For your problem, may there are several surfaces where you want the solution. Create monitors for temperature on these. As long as the residuals are not increasing, and they are converged as well as we can, we can often trust the solution. Note, sometimes there will be some oscillation in the residuals due to recirculation zones in flow field.

1. How are you defining convergence? You should not simply expect residuals of 1e-6 for convergence.
2. What parameters are you monitoring to say that your simulation converged.
3. As previously mentioned natural convection problems would not converge properly in steady state. If you are using ideal gas equation to solve your problem, try switching to bousinessq approximation. It might help. Your temperature difference is not that high. If Beta(del T) << 1, then using Bousinessq is valid. It will help with convergence.