Rando, et al,
I wonder about the failure mechanism of the panels. Perhaps the adhesive used to attach some of the failed panels directly to the roof is not thermally conductive. Or maybe the panel material had poor thermal conductivity. I wonder if that material is the same on the new units now. I wonder what they changed?
perhaps a chemical interaction of panel material and adhesive? It seems to me that there are there are a number of possible failure paths.
It seems to me that bonding the panels to the roof would stiffen the roof up a bit and that makes me question a vibration related failure related to bending. I doubt the deflections are very large. That written if there is data on work hardening of the panel materials that would be interesting to see.
Without data all we can do is speculate, I would very much like to see the panel failure analysis data.