Leading the Intelligent Factory: Chapter 4
See trouble before it costs you
Slow changes do not trip alarms. Read every tag every night.
Spread 1
Slow trouble does not trip an alarm
The decision. Every shift, someone looks at a trend and decides whether a change is worth acting on. Act too early and you chase noise. Act too late and a drift becomes a break, a trip or a reject.
What goes wrong today. Your alarms are built for fast, unsafe events. Slow changes do not trip them: a flow that creeps up over three weeks, a valve output that froze, a cycle too slow to see on a trend screen. Each shift reads the same trend a little differently. When something does swing, the fix often goes to the symptom. Moisture cycles, so the crew trims steam.
What AI changes. A machine can read every tag every night against its own history. It sorts what it finds into plain types: drifted, stepped to a new level, flat-lined, swinging more than usual, stopped reporting. A short list reaches the team by morning. When something cycles, frequency analysis tests it against every tag on the machine and points to the source.
What still needs people. Deciding what to do, going to look at the valve, and confirming the fix held.
Questions for Monday morning
- Which changes on our process would we not see until they cost us a break or a trip?
- Does every shift answer the same drift the same way?
- How many of our control valves have not moved in a week?
- Who reads the overnight exceptions, and what is done by 9 a.m.?
In practice. Envoy's daily process reporting scans every tag overnight and puts a short list in front of your engineers each morning.