First of all, you have to determine the dependent variable like temperature or Nusselt number, then start to increase the number of mesh and make a table or graph to estimate the change in the results.

when you start to have the close results with a little difference in decimal of the results you can depend on the number of the new mesh. This call mesh dependence which means any increasing in the number of mesh will not change the results.

Good Luck

I would try the largest size of your device with sensitivity study and figure out what is the largest size that will work there. Then I would translate this mesh to the smallest element, and figure if refining it further does change the results. If not, I would go with this setup for all cases. If it affects your results, refine the mesh in the smallest case, and translate that to largest - and repeat until you find it.

I also assume that the mesh size is significantly smaller than the element you test, and that the range of sizes is such, that the smallest mesh from the large sample will still be sufficient for the small one. Otherwise, you may have to come up with just some general dimensionless number that describes your mesh, such as "my element always had 20 cells in its length and the first mesh boundary layer element had depth of 1/50 of length in each case". You would have to validate such rule against at least two, extreme variants of the parametric study.

I think you can perform the grid independence test with one of your geometrical models. After you have found the mesh size that gives relatively accurate results with an acceptable computational cost, I believe you can base on that finding and make an assumption that the mesh size should be applicable to other geometrical models. In my opinion, this should be good enough if your geometry does not change a lot. 

If I understand what you are doing, changing the lenght you are working with different Reynolds numbers. That means you have different flow problems depending on the value of Re. Thus, each problem requires different grid resolution and has a different solution you should study by performing a grid convercence analysis.

Hello,

Thanks a lot for your replies. I'm starting to understand what should I do, however, if it wouldn't be so much, can any of you recommend me some bibliography? in order to perform an adecuate "grid convergence analysis" that actually validates -statisticaly speaking- the solution's independance from the grid for this parametric situation.

Regards.

Thank you so much, those were really helpful.

First of all, you have to determine the dependent variable like temperature or Nusselt number, then start to increase the number of mesh and make a table or graph to estimate the change in the results.

when you start to have the close results with a little difference in decimal of the results you can depend on the number of the new mesh. This call mesh dependence which means any increasing in the number of mesh will not change the results.

Good Luck

I think Grid independence means the change of one, two or more parameters in different time and comparing them for example comparison between the change of the temp with the wind change of the fan-coil in winter and summer. So comparison between mesh and their accuracy. All the results must be collected in table or graph.