CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Solving natural convection (

Shljuki July 18, 2012 22:08

Solving natural convection
i have a box model (room) with one cooled surface (ceiling) and one heated surface (window) and i am trying to solve the temperature field inside the room. There is no forced air flow and generated air movement is only due to natural convection, buoyancy effetc. I have a fine mesh, starting with 3mm cell height along the walls and inflation factor of 1.2. I am using k-w with automatic wall treatment. It is a stady state run.

The solution reaches convergancy of about 5E-04. At this stage i stop solving fluids and turbulance and continue to solve only energy and radiation till the temperature field stabilise.
Can you please advise if this methodology seems reasonable? Alse please advise on any other solving approach.

ghorrocks July 19, 2012 06:06


I have a fine mesh
Do you know how many times I have heard that on the forum :)

Unless you have done a mesh sensitivity study and proved you have a fine mesh then you have an unknown mesh.


It is a stady state run.
Buoyant flows at Rayleigh numbers high enough to generate turbulence are almost always transient.


The solution reaches convergancy of about 5E-04. At this stage i stop solving fluids and turbulance and continue to solve only energy and radiation till the temperature field stabilise.
Can you please advise if this methodology seems reasonable? Alse please advise on any other solving approach.
Have a look at this FAQ:

Reasonable - answer = no. You will have a big error.
Other approach - answer = transient solutions are almost always required for these sort of flows.

Shljuki July 19, 2012 08:54

Thanks for your reply.
The meshing approach, first cell size and inflation has been documented as a reasonable good approach for natural convection studies. I have done a sensitivity check on a 2D simplified model against a mesh with Y+ of 0.9. Mesh seems fine.

I have started with transient runs as transient behaviour of flow was expected. The transient runs take too long and they are dependant on initial guess. Since I had 20+ runs and not much time I looked for another solving approach and thought that running only energy and radiation at the end might provide a reasonable solution for comparative studies between the cases.
The most relevant parameter for my study is the heat flux (cooling power or energy) supplied from ceiling surface into the domain and controlled by sensed temperature at the point. I have run one case in transient mode using converged steady state as initial condition and got difference of less than 3% for heat flux, which is acceptable for my analysis.
Is this a typical error or I was just lucky with this one. We have one company licence and unfortunately I won't be able to test each case so I need a help with this issue.

Is there any document or paper that addresses or quantifies this type of error to the solving approach I used in my analysis?

Ayk July 19, 2012 16:43

Is not the SST Model recommended . k-w is well for the near wall regions but the more you get away from the wall it will lose accuracy as far as i know.

There is also a max timestep you can set on buoyancy driven flows in the user manuals. That could help too for better convergence. And are your global imbalances ok ?

ghorrocks July 19, 2012 21:59

The FAQ I linked to describes the process to go through. But if at the end of the day it means a transient run is required then anything else will cause significant error.

Shljuki July 20, 2012 01:01

Thanks Ayk,
global imbalances are OK. The highest imbalance of for energy, still less than 1%.
What turbulence model is recommended for natural convection analysis?

Shljuki July 20, 2012 01:06

Thanks Glenn,
Rayleigh number in the worst case scenario is 6.88*10^8. I assume this indicates transition from laminar to turbulence flow. I will run the case with highest Rayliegh number in transient mode.
Is there a way to plot Rayliegh number in CFX post, or it needs to be calculated?

Felggv July 20, 2012 14:27


I hope you don't bother if I come up with a discussion about mesh here in your topic since you said you had a "fine mesh".

A friend of mine have generated a mesh on ICEM that had been through the quality test of ICEM and got min average 0.99 and max average 1.0.

Should it be called a fine mesh?

It's a Hexadominant mesh done manually by blocking method with 4.0cm sizing HVAC simulation with people represented by rectangular blocks with heat transfer.


ghorrocks July 20, 2012 18:10

Rayliegh number is a global non-dimensional number, so you will need to calculate it.

Sounds like the simulation is at least partly turbulent based on that Ra number.

Felipe - mesh "fineness" has nothing to do with quality. They are independant parameters.

Felggv July 21, 2012 11:27

You mean fine in the sense of small elements?

I misunderstood the meaning of his phrase: "I have a fine mesh".

I thought fine meant good quality.

By the way, if fine doesn't mean small elements, would you explain me or show an explanation of what does it mean?


flotus1 July 21, 2012 12:57

When talking about meshes, "fine" is usually used as the oppsite of "coarse".

Felggv July 21, 2012 13:52

When I read ghorrock's comment about the author of the topic saying he had a fine mesh and about sensitivity analysis and etc made me think that fine was about the quality since fineness is usualy "easy" to see since most computers here where I work cannot run simulations with meshes that are excessively fine.

ghorrocks July 22, 2012 08:22

Yes, I can see now the word "fine" is ambiguous. I can see how people read it differently. My comment was assuming the small size definition of "fine".

Shljuki July 30, 2012 20:27

1 Attachment(s)
Attachment 14858

Wall heat flux [W] on the surface with the highest Rayliegh number: transient versus steady state results - 460 seconds.

ghorrocks July 30, 2012 23:55

Looks like you are getting transient behaviour, but the difference from the steady state run is small. So you will have to decide whether the extra effort of a transient solution is worth the extra accuracy.

All times are GMT -4. The time now is 17:05.