The Machine That Wouldn't Behave

For years, one machine dominated my downtime report. Not because it was old, or exotic, or particularly complicated... it dispensed a single component into the product.  The machine's job was to cut this component from a web and drop it into the product. That's all it did.

But it produced two failure modes which appeared on the daily report almost every day: it would cut the component in the wrong spot, or the line would have been down and this machine's name would be listed as the reason.

It was worse on nights. It was close to constant. And it was, for a long time, a particularly confusing malady... not because the failure was complex, but because of when I found out about it.

I never caught it live. I'd hear about it in the daily meeting, hours after the fact, then walk out to the machine to see it running perfectly. No fault code, no obvious wear, no smoking gun... just a machine behaving itself the moment anyone came looking. 

I'd ask the day-shift operators what happened and they'd shrug.  Maybe they would say something mean about night-shift.  

Whatever was wrong with it was intermittent by nature; tied to something upstream of the machine itself, not the machine.

Eventually I faced it enough to understand how it was happening and why. 

The machine had a set of guides that had to be swapped whenever the product running through it changed. The guides held the product tight enough that it couldn't shift as it passed through... because if it shifted, the cut landed in the wrong spot which would cause quality issues... and then the operators would either stop the line to fight with it or make bad product and create a big rework situation. 

It was a simple mechanism once you understood it.

Here's the catch: the operators didn't have to change the guides on every changeover, only sometimes. 

And "only sometimes" is exactly the condition that guarantees a step gets forgotten. 

Those steps in any process are perishable and will fail if not reinforced.  This step wasn't a required, every-cycle action, so it never became automatic. It sat in that gap where the step is real, but not frequent enough to become habit and not consequential enough, in the moment, to get double-checked. So it got skipped. Constantly.

Once the correct guides were in, the machine ran well.  They were reliable enough mechanically and, in our efforts to get them to stop dominating the downtime report, we had restored them to like-new condition.

So I did what most of us do at that point: I went and trained everyone. Every operator, every supervisor, every shift, personally. I explained the mechanism, showed them the failure, made sure they understood exactly why this mattered. 

And it worked. For a while.

Then it drifted back. Not all at once... it was a slow creep back to the old baseline, little by little, until we were right back where we started. 

So I went back and retrained. Same result: improvement, then decay. I could refresh the information, but I couldn't make it stick.

That was the real lesson, and it took me longer to learn than the guides did: information was never the constraint. 

Everyone I trained understood the mechanism. Nobody was confused about what to do. But they couldn't get from being handed the information to the change of habit necessary to avoid the issue.

My observation was that a single training session, however clear, isn't strong enough to displace a longstanding habit. 

Knowledge fades on its own schedule.  And operators don't always pay attention to what they're told.  But I think the real failure was that my words weren't strong enough to change their habits. 

Training treats the problem as an information problem. It isn't. It's a systems problem and systems problems don't get solved by saying the right thing more loudly or more often. 

They require a system built to help people change habits.

That's the piece that eventually became the system behind 180 OPEX. Not another training module, not a poster on the wall: a structure that changes habits. 

I built it, ran it on this exact problem, and the failure mode that had haunted that line for years simply stopped. Not "improved." Stopped. 

And, for the first time, it did not come back.

The machine didn't change. The guides didn't change. What changed was the system supporting our operators.  It finally gave them what they needed.