How to handle large chunks of missing data for a subset of variables?

Follow the full discussion on Reddit.
I have a scenario where I will be modeling many different parts on many different machines using IoT sensor data. That alone is going to be a large amount of models. However, an added feature (that I was not told about) will be a list of questions that will be asked of the users on top of the fully automated sensor data. The idea is to raise an alert when a part might be malfunctioning, then ask the user a list of questions that cannot be found via existing sensors and give them an updated score.