Learning Forum

[dodonn4]

Hi everyone,

I am working on a fluent simulation to model natural convection heating inside a bag of liquid. I was able to successfully simulate it and get accurate results, then I changed the geometry slightly. The changes to the geometry were not significant at all, the general size and shape are still the same. It is relatively simple geometry too; just an elliptical cylinder that tapers inwards towards the top.

However, now I noticed that my results significantly vary with different mesh sizes. For example, if I have 150k+ elements in my mesh my results are around 20-30% different than when I have 100-130k elements. And if I have over 250k elements in my mesh the simulation fails.

Why does the mesh have such a significant impact on the results?

[Dio]

Most of studies need this kind of different tries. It is called mesh independence check. You start your simulation with less elements with appropriate quality than increase the number of elements to see difference. At some level the difference will be less, it means you dont need to increase your number of mesh. So this is normal. Just keep increase your number of mesh with desirable mesh quality, then make a excel file for results or you can parameterized it to see the results.
Huge differences can be caused from the low quality, or not enough convergence. As I know some heat transfer studies need 10e-6 at the residuals parts.
For geometry, if location from the geometry you changed is important for the results, than very small changes can make differences on the results, I guess.

[dodonn4]

Thanks for the reply.
Performing a mesh independence check makes sense, but what I don't quite understand is how my simulation failed when I increased my mesh to a certain number of elements. I expected to get results similar to what I saw with fewer mesh elements, but the results were all over the place during the simulation until it eventually failed. Could this indicate an issue in my geometry or setup in fluent?

[Dio]

You can check by parameterizing your inputs and outputs.
For example, in mesh window, you can paremeterize number of element, nodes, orthogonal quality (or which one do you want for mesh quality)
If your results are dependent on specific location on geometry, paremeterize that location mesh settings, too. ( like bias factor, number of elements or something else)
for setup, you can do same thing for your inputs and outputs like inlet velocity, output temperature, pressure, Gr number etc.
With this, you can change your mesh and see the both mesh quality and result change using parameter set window. This can be a way to compare mesh quality with results.
Maybe your simulation needs low aspect ratio, reason for fail may comes from this.
Maybe your formulations for results are not proper, I am not expert on this formulation but I have been working on to understand method in ansys, too
As far as I know huge differences or sudden changes in solution part is about mesh quality. Mesh transition should be smooth, especially for sensitive calculations like heat transfer applications.

[Rob]

If the solution is changing as you refine the mesh, what is happening? As says it could be aspect ratio, cell quality or that you're now picking up another flow feature that's moving/changing quickly and therefore making the solver less stable.