Answer:
Actually the plate modelling would create many local modes and thats causing this issue. There are 2 ways around this.
1) Use the plate stiffness scale factor to make the plate very rigid for eigenvalue analysis. Boundary change assignment feature would have to be used so that this is done only for eigenvalue and no other analysis.
2) The easier way, use the Ritz vector method.
With that, 90% modal mass participation is obtained. The modified model is attached.
The first method is causing some change, but not enough to lead to 90% participation.
So, either springs would have to be provided at bottom slab, which can be activated only for eigenvalue analysis or Ritz vector method could be used. P
oint spring approach is not recommended since it would also cause many reactions for RS analysis.