To complete the remark of @Icome please specify whether your simulation is done in static or dynamic.
In statics, the authorized translations are only in one direction and always with respect to a frame of reference, whether for example a face in relation to another face or a "coordinate system".
You will notice to support my point that for the constraints, for example X if you give a millimeter, you are not offered X and -X, but only one of the two.
(it's pear or cheese but not both: even for the €120 menu)
If I understand, you have four frames under your feet and you want to know the force on one or two of the four bolts (one bolt is better in my opinion).
So on the two plates that are the least stressed to the tear (downwind plate or opposite the forces) you put connections on between the ground or the virtual wall. On the other two plates you can then measure the reactive force on your beveling bolts.
Obviously, if you put anchor bolts on all your turntables, you will have results, but they will not give quite the maximum effort on the turntables to the wind because part of the forces is absorbed by the structure which must also be deformed somewhere.
Another simpler way to do it: you put remote loads on each of your plates, which are fixed by anchor bolts on a virtual wall.
It's normal that the calculation is longer because with a virtual wall you manage the contact: so the calculation is done by iteration (if your foot/leg deforms, then your plate will not press on its entire surface but probably only on the inside or outside).
It is also necessary to think about putting lateral constraints to avoid a sliding of the model laterally (this can hinder the convergence of the calculation or give results of aberrant displacements). Attaching a straight line (edge of a sole) to limit these movements can be a good solution
I would like to hear from @froussel (it's also one of the advantages of blocking on an edge ;-) ) and I add that if the mesh is too fine it has considerable consequences on the delay because there are then millions of elements to process.
It is better to start with a slightly coarser mesh that refines the very constrained areas. On this kind of exercise it does not normally exceed 5 minutes especially for a tank.