You haven't attached the simulation. However, I am going to guess that you put a "fixed" constraint around your outside edges. As your plate "balloons", it will pull the edges inward toward the center. The capscrews do not prevent this.
Thanks. I have attached the simulation with the assembly file itself. You can see it by clicking the study 1 tab at the bottom left corner.
I opened the assembly. Usually (but not always) the primary issue is the boundary conditions. Where it is "fixed", the Simulation assumes absolute rigidity, which is not as in real life. Hence, the plate assembly will be much more stiff than actual. It will deflect much less than shown by lab test data. Also, I noticed the "Global contact" is set to "bonded", which glues the plates together where they contact. This 'bonding' tends to stiffen-up the assembly considerably.
I noticed there are at least two planes of symmetry as well, so you could use a 1/4 symmetric plate assembly. This will reduce meshing and computation time substantially.
I would approach this by:
1. Create a new Configuration and section the model (cut off three quarters and leave one quarter for analysis).
2. Install symmetry boundary conditions on the cut (exposed) faces.
3. Install "no penetration" contacts between surfaces of the two plates that come into contact.
4. Install "fixed" fixtures on only the inside surfaces of the bolt holes (not quite realistic, but this may be close enough).
Try a sample run on this. It will take a while to run due to the "no penetration" contacts (compared to without them). The assembly should be much softer. If still too stiff, it might be because it does not acount for the elasticity of the screws. They can actually stretch.
To add another step, I would install "Bolt connectors" between the plates to account for the elasticity of the bolts (but remove the fixed restraints on the inside bolt hole surfaces of the 2mm plate). I am assuming here that the 5mm plate has screw threads, but the 2mm plate does not.
I hope this helps!
I just want to add, if not already done, obtain a hand-calculation estimate for a plate deflection from Roarke's or another source (for a rigid perimeter). I typically see lab test data has an error spread of about 10%, so the Roarke's formula and FEA should agree at least within about that range.