Consider the following geometry (0°/90° stacked):
which produces the following reasonable mesh (matrix hidden):
When we add a z-binder yarn (note: the full unit cell is not shown; this is a reduced boundary)...:
(This is a 0.15 power-ellipse cross section, with the major axes touching)
We produce a mesh with what I believe are two separate (related?) problems. The first one is illustrated in this picture:
In the image above, we see an excessive amount of elements being generated along the width-axis of the yarn. This is a big problem for FEA because the tiny elements perpetuate throughout the thickness of the model (and this isn't even the full unit cell). Note that this does not appear to be an artifact related to tolerance. The tolerance in the image above is visibly crude (overlapping yarns fully touch until the last element-length), and refining the tolerance causes even more elements to be generated.
Here is an unrealistically crude tolerance case where most of the stuffer is voided by matrix, yet the tiny elements still exist along the width-axis of the yarn (matrix not shown):
Next, examine the elements along the path of the binder yarn:
Notice that the element lengths along the outside face of the binder yarn are much longer than the typical length. It appears as if the mesher steps along in the global coordinate direction when generating nodes, rather than the local centerline tangent direction. This becomes less pronounced as the angle of the binder yarn drops down.
Here is another geometry that exhibits only the second problem, to a lesser extent (but still undesirable):
and mesh:
Note the very long elements along the face of the binders.
