Leading the Intelligent Factory: Chapter 5
Fix the right thing first
The list of problems is always longer than the people to fix them. Rank it by risk and value, and know what kind of fix each item needs.
Spread 1
The loudest problem is not always the biggest
The pressure. Breaks, trips and repeat failures cost you every week. Your maintenance and engineering time is fixed. Every hour spent on the wrong problem is an hour the right one keeps costing you.
What goes wrong today. The list gets ranked by who complains loudest, or by what changed last. A loop that cycles gets retuned three times, when the valve is sticking. A rise in breaks gets blamed on the newest change, while the real shift happened a week earlier. Loops that sit at a limit, with no control range left, never make the list at all.
What AI changes. This is the decision of what to fix first. A machine can grade every loop on the process and every tag around an event. It separates causes from effects, so the tags that move because of a problem drop off the list. And it says what kind of fix each item most likely needs: tuning, the valve, or more capacity.
What still needs people. Weighing safety, the outage window and the cost of the fix. Walking down to the valve. Deciding which trial to run.
Spread 2
A real example: two loops and a break study
A paper machine loop health report graded 200 loops from two hours of one-second data. Two were red, and they cycled together every 3 minutes: machine chest consistency and headbox level.
The consistency loop’s output showed a triangle-shaped pattern. That is the signature of a sticking valve, and retuning does not fix it. The headbox level showed no such signature. It was most likely riding the consistency cycle. So the order was clear: stroke and repair the consistency valve first, then see if the level still cycles. Retuning the level first would have treated the effect.
The same machine saw sheet breaks double, from 0.95 to 1.94 a day. We tested all 3,299 process tags for a change at the same point. 281 changed. After grouping tags that move together, 35 groups were left: 11 possible causes, 7 effects of the breaks themselves, and 17 building and facility tags.
The strongest cause signal was the furnish. Hardwood share jumped from 1.8% to 10.6% when breaks rose. Refiner pressure also dropped, but six days later, so it was most likely a reaction, not the cause. The team got two questions to answer, not 3,299 tags to chase.
Questions for Monday morning
- How is our fix list ranked today, and who ranks it?
- How many of our loops run in manual, and why?
- Which loops sit at a limit with no room left to control?
- When a loop cycles, do we check the valve before we retune?
- When breaks or trips rise, do we know what changed first?
In practice. Envoy's loop health report grades every loop on the process and states the most likely kind of fix for each one: tuning, the valve, or capacity.