Small features are often ignored in finite element analysis, either by defeaturing them or not meshing finely in their vicinity. This can be fine for overall results like displacement or vibration where they often have little effect. But this may not the case for stress analysis. Consider this model with a slotted hole. The curves at the end of the slot do not become less important for stress as the slot becomes thinner. The stress trends upward with smaller radius, and in fact eventually becomes equivalent to a crack, with infinite stress, as the radius goes towards zero.
A famous real-world example of this it that some early versions of the DeHavilland Comet, the first commercial jetliner and a beautiful aircraft, crashed due to fatigue cracks emanating from the corners of their windows. The cornenrs had fillets that were too small, and the stress riser from that “small feature” caused the fatigue crack.
I was reminded of this recently while trying to modify automeshes in Femap. I came across this setting on their mesh sizing dialog:
If you miss this setting you can play around all you want with “Curvature-based mesh refinement”, “Surface Interior Mesh Growth”, etc and wonder why it is having no effect on the mesh near an important small feature like a fillet. You have to measure the size of the feature. If it is less than 0.0106 in the example above, it will be ignored, so you have to override this setting with a number smaller than your feature. I’m not trying to pick on Femap with this example, I’m sure there is a similar setting in other pre-processors. As mentioned above, the code doesn’t know your are interested in accurate stresses, and the feature may not have a big effect on other results.
What I would like to see in finite element structural analysis interfaces is a master setting like “stress results are important from this analysis”. Various default could key off this, such as turning “midside nodes on surfaces” on and “Curvature-based mesh refinement” . And the default setting of “max size of small feature” could be made smaller.
And I think this is an issue that users who need good stress results should be aware of. I’ve meshed large models and the mesh appears to be of good quality until you zoom in near important features. It may not be following the geometry well there, and adjacent elements may be large compared to the size of the feature.
In the example below, the overall mesh looks good until we zoom in near the fillet around the boss, and the mesh is way too coarse there. In the lower right zoomed in image, telling the mesher that “max size of small feature” should be less than 1 mm (the radius of the fillet) and turning on Curvature-based mesh refinement were needed to get this better mesh.