It should be very useful an anomaly detection working by template.
This should make the training and creation of the model monitoring properties of all the things of such template, creating so a model for all the "fleet".
When it is in monitoring phase, it should tell you how is the status of this property respect the "fleet model", something like:
"the temperature is higher than the 80% of other things" (and possibly with a time-filter, like "since 1 hour" for example).
With this information I probably will to a check to see if temperature is going wrong or not ....
An even better implementation would be if it was possible to get the FLEET distribution trend (at template level), and compare it with the thing distribution trend, and possibly make a distribution chart.
If difference between the two charts is too big it should give an alert.