As you state, avoiding extremely small features that aren't likely to
affect the flow helps a lot in meshing the geometry in the same way as
meshing an FEA model. Tiny surfaces, gaps caused by poor modeling
(geom checks, tolerance settings) will also cause problems.
On the other hand if they are large relative to the size of the flow
field being modeled, rounds, chamfers, leak gaps, holes, bolt heads,
etc can have a big effect on the flow results so they may be necessary
to get accurate CFD results.
Real assemblies may have small clearances or gaps that in reality leak
flow. If these leak gaps are small relative to the flow field of
interest it may be desirable to eliminate them from the model. This
requires some judgment or simple 1D calculation to determine whether
the leakage is important to the flow results of interest.
It depends on the meshing tool used, but often a solid model of the
fluid must be constructed. This can be done by using the solid model
of the part to cut a fluid volume.
Finally, it is usually desirable to extend the fluid zone inlet and
exit length so that the flow boundary conditions at the inlet and exit
are separated from the region of interest. This requires either adding
actual upstream and downstream geometry to the model, or creating some
simplified geometry that mimics the effective area and perhaps the
flowpath curvature of the actual geometry.
Dave
My perennial comment about not using relative accuracy applies here.
If your models are full of holes, the mesh monkey will have to fill
them in, and may hideously misinterpret your intent.
Make sure you sit the mesher down in front of your screen & explain
what's going on, as they are social retards almost without exeption,
and won't necessarily tell you if they get stuck. They'll just make
some crap up & blow your budget on something you can't use, then blame
you. Killing them is frowned upon, even though by any rational
measure, they aren't properly human, as they aren't designers.
The bitter voice of experience.