Hello, I just found out about that site and I hope I will get some help here.

I'm studying the flow in a guide tube of a pressurized water reactor. I did a first mesh sensitivity study with 4 different meshes in steady state (from approximately 200000 to 800000 elements). The results show that the solution converges better and faster with increasing elements and the residuals are of the order of 10^(-6) with the 800000 mesh wich is good. So I thought I was approaching a mesh independent solution.

When I calculate the solution on a mesh with even more elements (approximately 1100000 elements) the results show that the convergence is well slower and I can't get residuals less than 10^(-4). In addition the solution is totally different compared to the results obtained before. I tried with even more elements (1500000) and the convergence is still even slower and the solution still different...

That's why I want to have advices from regular CFD users. Thank you.

I don't understand why the solution changes like that as the previous results seemed good and the solution was well converging with very low residuals... Should I try again with even more elements?

This is the issue with meshes. You need to make sure your mesh doesn't effect your answer as you have just said it does. If it converges thats ok, but it doesn't necessarily converge on the right answer. A mesh study is required and it needs to be comparable to a known (measured) system. Interestingly enough, I'm in the same business and a colleague of mine did something very similar in CFX. If you wish, I could examine his thesis and point out anything that may be of use to you.

-W

Very Sorry, its early in the morning and I'm babbling about mesh studies. You've done the study. My colleague did an 800k node mesh and a 2mil node mesh and found that it was sufficiently close to excepted lab values. Anything more and he was wasting time, anything less and he would get different answers. Again, we was doing a normal path up a channel in a PWR so if I'm guessing your model right, you aren't too far off. -W

One thing to note about mesh refinement is that when strong shear layers are present, the turbulent viscosity may not be sufficient to prevent large scale turbulent structures from developing. If you are running CFX, this problem can sometimes be suppressed by increasing your timestep (a smaller timestep will allow the vortices to propegate). This is often the source of convergence problems when meshes are refined.

You should also try to maintain the same mesh characteristics or topology. Changes in mesh quality, such as increases in volume ratio or aspect ratio, can effect the solution and convergence characteristics.

Also try and follow many of the existing guidelines on getting convergence. You can search the forum for a lot of this as well as review the on-line documentation. The most common parameter to play with is your timestep (try both increasing and decreasing it), and you may also want to try running with double precision turned on, which can reduce the effects of grid quality.

Regards, Robin

Thank you a lot for these advices this site is awesome.