It seems that contacts are a bit on and off, so could be due to many reasons, causing problems with convergence.

Also at one iteration the model moves a lot so to much movement could cause contacts to be on and off and they never catch properly.

So that I think is the main issue here. TO address this:


I would suggest to use sub steps say 20 initial/minimum and maximum 100 say, so to apply the load gradually.

Also one can put an enforced displacement in first step to get the contacts to activate nicely (easier with imposed motion), and then in a second step start applying the load. 


(if this does not help post again with images of your model so we can have a look).