How to Deal with a Stress Singularity

Hello all,


This may require further input from me, but I wanted to begin the discussion.


I have a model as shown below:




I want to find the maximum stress and deflection of the thin plate. Seems simple enough.

I began with a 3D model with two solid bodies and used mesh mating between contact faces. I received results as one might expect; there was a stress singularity along the nodes of the thin plate around the edge of the top of the center column. I plotted results and attempted a straight-line approximation.



I wasn't satisfied with this result, and making mesh refinements resulted in too long of a simulation run-time. So, I tried using a Shell for the top plate and got the following results:

Note: I used surface-to-surface gluing to connect shell and 3D elements of base plate.




NASTRAN reports a maximum Von-Mises stress of ~ 47 MPa at the appropriate location. Refining the mesh settled the maximum stress value, i.e. no stress singularity where stress explodes with decreasing mesh size.


However, one thing I do not understand is, how does the "Weld-Like Connection" option of the Surface-to-Surface Gluing work? Looking at the deformation results show the plate deflecting in Z more than the top of the center column:3.PNG


Plotting Z-Displacement of nodes for center column and plate shows the same discrepency: 

Note: I took the nominal Z position of the plate node as being at the center of the thin plate. Surface normals point up, so maybe nominal Z position is actually the top surface of thin plate? This helps the plot some, but not entirely. (X of orange dot goes to 2.9)




Is this discrepancy real, or does the "Weld-Like Connection" hold true? Are the stress results valid? My local expert suggests the gluing is allowing for relative displacement between mating surfaces and thus the stress results aren't modeling reality.


So, I chose to attempt another type of simulation. 2D Axisymmetric Structural. Here, a 2D mesh allowed me to refine my mesh at the singularity a great deal, thus hopefully providing a more accurate straight-line approximation. Unfortunately, what I found was that decreasing mesh size around the singularity increased both the peak stress AND the stress of the surrounding nodes, as shown below:




So, how can you make a straight line approximation in this case? 


Or, a better question, given the above simulation objective, now that I have completed my exercise... can somebody point me towards the obvious and best solution method for this situation? That would be nice.


Thanks in advance for your thoughts.