There are many metrics you could use to examine the quality.
Measure the RMSD to a known homolog (ie. what the Phyre2 server used as a threading template. If there are any regions of high RMSD that would suggest they aren't super likely.
Measure dihedral angles for the protein chain for a Ramachandran plot, if you have any outliers, it suggests your simulation has an implausible bond angle for the backbone.
There are loads of webservers for this kind of thing, so I'd start by just googling a few to see what you can find.