Just an addendum to Gerald's warning. Some programs put ficticious (Zienkiewicz) torsional stiffness (normal to the plate) to avoid singularities, so the problem might not fail with a beam inducing a torsional load into the plate - but the results would be meaningless.
If memory serves me right, the primary reason the large mass method was used was that Nastran and some other older FEA codes did not originally have base excitation capability and thus it was approximated with the structure attached to a large mass with force excitation (to the large mass).
The rule of thumb aspect ratio for thin shell theory has to do with the actual part being modeled not the elements. You can decrease the size of the elements for refinement and, if they are thin shell formulated elements, the analyis wiill converge to thin shell theory.
FeaGuru is right. If you have an inside corner, you're going to have infinite stress at that point - and the more you refine the grid the more you will approach infinity. That's where you have to make engineering judgements as to whether the stress raiser is a problem for brittle parts...
Minoand is right about coparing DOFs. If the element math is right then both the constant strain tet and the higher order elements will converge to the "theoretically" correct answer - but it will take a lot more constant strain tets to get there.
The easiest way to do it is to overlap at least one row of plate elements with the solid elements so that the plate moments are reacted by a plane of nodes (rather than a line of nodes) on the solids. The issue you will have to decide for yourself is whether modelling that way really reflects...
1. Take some courses that cover FEA theory, not just how to use canned programs.
2. Read all the good books on the subject (Bathe and Gallagher are some easy to read authors).
3. Start with simple problems that you can also solve by hand and compare results using different element types and...
Visual C++/Windows character questsion: Is there any easy way to imbed some Greek characters (particularly, theta and phi) into a C++ character string for use in the SetDlgItemText command? I use to be able to print them using ASCII decimal codes 233 and 237 on older DOS C compilers (Borland)...