Check also if the regime is not unsteady (chaotic, turbulent, periodic) due to high flow characteristic number (Reynolds, grashoff, ...): this if the problem is correctly set up (modelling, boundary conditions,...).

thank you Dear Aziz

thank you Dear Lucky

thank you Dear Ridha

How do you fix pressure? Is that the best way?

Did you try a coupled solver? (or, if you tried coupled, did you try segregated?)

Can you come up with a better initial condition? Or a different way to generate a 'close' solution?

Would you call your convective flow turbulent or laminar? Please specify.

In the former case it might be a problem of the closure because most engineering closures where buoyancy is included are often highly speculative. An expensive recommendation is DNS. A cheaper approach is to choose the right algebraic boundary-layer description and outside the attached turbulence theory. Good luck!

If you use segregate algorithm you may expect extremely slow convergence, especially if you deal with natural convection at high Rayleigh number. That is why in SIMPLE-like algorithms the pressure correction is derived based on assumption that pressure force and inertial force are in balance, and this is not the case in your problem. At the beginning of you should start with small pressure and velocity under-relaxation factors, let say 0.05 (maybe even lower) and afterwords you can increase them (but not too much).

What does it mean that convergence get stuck? Maybe your convergence criterion is to high.

Dear Helmut 

it is laminar flow

Dear Behnam, sorry, then FLUENT should actually do it because it should resolve all the near-wall (fine-scale) flow structures. Twisted flow neats etc . A heavy job, good luck!

Hi

may by i can help but i must see the model.

I modeled this:

watter tank with heater - https://youtu.be/Hf0KTi_9HhE

chilled beam - https://youtu.be/2A4xEE90cHU

My best guess is that the flow is turbulent. For bottom heated case, the flow become unsteady around Ra=1800, which means that the temperature difference is order of a few degree, then forget about steady state solution. To resolve such issue you need to use turbulent model. Of course almost all turbulent models are not accurate. Howeverfor industrial problems, the results are acceptable.

Thank you Deat Abdulmajeed 

Thank you Dear Richard

i'll check these .

Behnam,

pl read my paper  to get an idea for deep tank simulations: "3D Time-Dependent Simulations of Onset of Transient Convection in Evaporative Cooling of Liquids."

You have to check your grid size vis a vis the flow velocity for steady-state simulations.

kk

You need to check Ra number, if it is >108, then the flow could be turbulent

The response of S. Etaig needs more explications. As indicated by Dr. Abdulmajeed and my response, you should define the configuration to speak on critical number of thresholds of transitions from a regime to other. The cases of differentially heated cavity and Rayleigh Benard one differ totally (means the side of heating). Also, the Prandtl number and the configuration (square, rectangle,...) are also of more importance in defining the critical point of transition. I mean:

         _________Adiabatic_______                    _________T_cold    _______

T_hot |                      T_cold         |                    |                                        |

         |                                        |   differs from  |                 | 

         |                                        |                     |                                        |

        |_________Adiabatic_______|                     |_________T_hot     ______| 

For the first, square and Prandtl=0.71:

untill Ra~1.8*10^8 (Laminar), ---------chaotic----->Ra~5*10^9-----------------developped turbulence       ------>

Well, you should start somewhat experimenting with the value of the kinematic viscosity and document the various behaviours of your virtual fluid. They you should find out whether it is a pure Stokes flow or at the threshold of convectiv turbulence. But as I am not fluent in FLUENT, it also may be a wrong option etc. Do you try non-stationary embedding or a stationary direct solver with iteration? Cheers, HZB