I'm developing one equipment. I also do dimensional analysis to determine the governing dimensionless groups of the flow in that equipment. Then I simulate the flow in various sizes of equipments to validate the dimensional analysis. In the models, there is the volumetric heat source, and I add bouyant effect to them. I use 'total energy' , second order upwind, and solve by using double precision. All test cases were computed untill residuals had reached the respective minimum. For the high magnitude of heat source (means high flow velocity), the difference of dimensionless groups (I derived) between models (with dynamic similarity) are about 4% (very good!). However for the small magnitude of heat (low speed), the difference between the largest (model height=100 m) to the smallest model(model height=5 m) is about 30%.

I think that there might be some things in CFX that when I turn it on, the results will be better. Do anyone have suggestion?

Thank yoiu very much for your kind consideration

Are you modelling the density using the Buesineseq approximation or using a true compressible fluid?

a true compressible fluid. Any idea?

Whats the reynolds # at the lower heat source? Could be the turbulence model you are using is inappropriate. e.g. if the flow is in a transitional regime the k-epsilon will be bad.

Actually the flow is turbulent, but I found that laminar model provides the reasonable results for the first approximation. Then I run them all with laminar model.

Why do you say the flow is turbulent?

Which dimensionless parameter do you look at to figure this out?

You may need to run laminar over a range of heat source values until you find the heat source value where the flow does not converge to a steady result anymore at a given timestep. Then you sort of know a bound where you need to start using a turbulence model.

Next thing to do is to activate the turbulence model and decrease the heat source until you start noticing problems/departures from a physically reasonable result.

The flow may be transitional, in which case LES/DES/SAS may need to be used.

Note in all this though that mesh independent answers should technically be checked at each heat source value... more work!