Define machine/line states
Agree on a small state model such as running, planned stop, unplanned stop, blocked, starved and changeover where appropriate. The exact model should fit the process.
Create events from transitions
A downtime event needs start, end, duration, resource and production context. Automated detection can identify time; human input may still be needed for cause.
Reason-code design
Reason hierarchies should be specific enough to support action but small enough that users can classify events consistently.
Prioritize recurring loss
Pareto analysis by total duration, frequency or impact can reveal different problems. A frequent one-minute stop and an occasional four-hour failure deserve different responses.