Mesh Refinement Study Problem
 

Hi,

I am performing a mesh refinement study on the volute of a pump.

While increasing elements my measure begins to converge, then jumps at around 9 million elements. All solutions have converged to 1E-06 RMS residuals.

I have attached a plot of the mesh refinement for reference.

Has anyone ever seen anything like this before? and if so what are some possible solutions or insight to what is going on here?

Here is some other info
ANSYS CFX
k-epsilon turbulence model
default tetrahedral mesh
10 inflation layers at walls .003 [in] smallest layer, w/ growth rate of 1.2

I have ran similar geometries with these settings and not had this problem in the past.

Thanks in advance.

For each mesh refinement level, you may also want to monitor convergence of your metric (forces, integrate pressure over a particular surface, etc.) in addition to monitoring the residuals of the conservation equations. Also, are there any issues with cell quality? You mention use of tet cells and as well as a turbulence model, is there a proper boundary layer mesh near the solid walls?

You could be experiencing hysteresis. If you are using local time stepping, the trajectory the solution takes will have yet another dependency on the grid and how it is laid out. So when the grid is refined it must be refined everywhere in a similar fashion. If the refinement occurs some here and some there then that my trigger the hysteresis. Or, global time stepping should be used. But, that can be slow.

Can you share your measure with us? is it a force or a moment on the pump.

It's not uncommon for changes in flow features as you change the mesh quality, as you start to pick up extra flow features that you couldnt with a smaller mesh size.

I would think consider different mesh types and different mesh types also, as these may have a significant effect. If the pump blades are at all separated then this may explain how things are shifting around so much.