In theory a good escalation system. In practice there must be strong guarantees and trust that there are no repercussions for triggering. Otherwise management will tell you over and over to "pull the andon cord / escalate earle & often" but really no one does.
Trader Joe's cashier bells have entered the chat.
Ironically, the "request assistance" button and accompanying blinking light on top of your stand at self-checkout are the "self-managing" version of that where you as the customer are partly an employee. oh well
One interesting thing you learn over the years is there's a million ways to improve process, and a million ways to avoid it. Virtually every improvement is unlikely; either it's difficult to understand, or difficult to get approval/consensus for, or difficult to implement. If it was easy you'd have done it already. Process won't improve until a motivation arrives that pushes past the difficulties.
Semiconductor manufacturing takes this to the next level with techniques like statistical process control. Units of capacity or work can take themselves offline automatically if something falls too far out of range. Waiting for a human operator is not really feasible when the facility is mostly automated and the feature sizes are atomic scale. There are also concerns with wafers that must be processed at their next step promptly or they will become guaranteed scrap. Human judgement is too slow to sort all of this out on the inner loop.
The ability for one person to stop the entire factory on purpose is not desirable in these facilities. At least not without a quick meeting first. There isn't an actual "stop line" button anywhere. Investors and customers like Apple and Nvidia would never permit this. The best someone could do (without using a fire axe in the datacenter) is to manually whack the EMO button on all the tools in a certain area (e.g. photo) which would effectively stop the line. By the time the 2nd photo tool goes offline, half of manufacturing leads would have their pagers beeping and security would be well on their way. You could try doing this through the information systems, but they likely didn't grant you permissions to flip status flags on 400 million dollar EUV tools at your leisure.
As a kid I used to substitute in an injection molding plant. We worked for four hours, pause, four hours. The plant ran three shifts, 7 days a week.
I worked in a big hall with two rows of molding machines. Each one had a pole with a yellow beacon light at the top.
The machines had fixed cycle times and the operator had to repeat the same steps every cycle. For example on one machine I had to put four small metal cylinders into the mould, close the door, wait, open the door and remove the finished parts. Cycle time was only a couple of seconds usually, with the shortest one I remember being 8 seconds. If you were too slow the machine stopped and the yellow light went on.
You also could stop the machine on purpose. There was a field with 3 by 6 buttons or something with different stopping reasons, toilet break being one of them.
So far, this probably doesn't sound too bad, but to complete picture you have to know two things:
1. Every restart meant throwing away the first one or two minutes of production
2. Foreman had to keep a quota.
That meant, yellow light, foreman came and shouted at you. The buttons were never used. You spent four hours straight doing the same routine every couple of seconds without skipping a beat.
Whenver I think my job is bad I remember that time and I'm glad for what I have.
I have worked on production lines and I’ve worked in the manufacturing and quality engineering side of things.
I know a lot of companies do have that old school production at all cost approach, but the better places I’ve worked at have put quality metrics ahead of productivity metrics.
Usually the metrics are safety, quality, delivery, and cost in that priority.
I know Goodhart’s law, “when a measure becomes a target, it ceases to become a good measure,” gets quoted a bunch on HN, but then nothing else is offered up. In my opinion, having these additional metrics with their own targets serves to balance out and constrain a single metric screwing things up. For example, if productivity is going up by producing lower quality parts, this will be shown in either the quality metrics (e.g., scrap rate, yield) or the cost metrics (e.g., material costs have increased, rework costs have increased).
The other component of it all is that the operators can only produce to the level of the systems they have. They can make some small, point improvements to help, but generally there need to be larger process or system level changes supported to make significant improvements.
I first heard about the Andon cord in this excellent This American Life episode about a GM factory that tried to learn and copy the Toyota method of production. It's still worth a listen and explains the concept well.
13 comments
[ 4.8 ms ] story [ 33.6 ms ] threadThe ability for one person to stop the entire factory on purpose is not desirable in these facilities. At least not without a quick meeting first. There isn't an actual "stop line" button anywhere. Investors and customers like Apple and Nvidia would never permit this. The best someone could do (without using a fire axe in the datacenter) is to manually whack the EMO button on all the tools in a certain area (e.g. photo) which would effectively stop the line. By the time the 2nd photo tool goes offline, half of manufacturing leads would have their pagers beeping and security would be well on their way. You could try doing this through the information systems, but they likely didn't grant you permissions to flip status flags on 400 million dollar EUV tools at your leisure.
I worked in a big hall with two rows of molding machines. Each one had a pole with a yellow beacon light at the top.
The machines had fixed cycle times and the operator had to repeat the same steps every cycle. For example on one machine I had to put four small metal cylinders into the mould, close the door, wait, open the door and remove the finished parts. Cycle time was only a couple of seconds usually, with the shortest one I remember being 8 seconds. If you were too slow the machine stopped and the yellow light went on.
You also could stop the machine on purpose. There was a field with 3 by 6 buttons or something with different stopping reasons, toilet break being one of them.
So far, this probably doesn't sound too bad, but to complete picture you have to know two things:
1. Every restart meant throwing away the first one or two minutes of production
2. Foreman had to keep a quota.
That meant, yellow light, foreman came and shouted at you. The buttons were never used. You spent four hours straight doing the same routine every couple of seconds without skipping a beat.
Whenver I think my job is bad I remember that time and I'm glad for what I have.
I have worked on production lines and I’ve worked in the manufacturing and quality engineering side of things.
I know a lot of companies do have that old school production at all cost approach, but the better places I’ve worked at have put quality metrics ahead of productivity metrics.
Usually the metrics are safety, quality, delivery, and cost in that priority.
I know Goodhart’s law, “when a measure becomes a target, it ceases to become a good measure,” gets quoted a bunch on HN, but then nothing else is offered up. In my opinion, having these additional metrics with their own targets serves to balance out and constrain a single metric screwing things up. For example, if productivity is going up by producing lower quality parts, this will be shown in either the quality metrics (e.g., scrap rate, yield) or the cost metrics (e.g., material costs have increased, rework costs have increased).
The other component of it all is that the operators can only produce to the level of the systems they have. They can make some small, point improvements to help, but generally there need to be larger process or system level changes supported to make significant improvements.
To bring it across to Andon and Lean, Deming literally addressed this as one of his 14 points of quality management that Lean manufacturing follows:
Point 11. Eliminate numerical quotas for the workforce and numerical goals for management.
What’s important here is not that there are buttons in a factory but rather the culture that goes along with their usage.
Point 8. Drive out fear, seems relevant here also.
Sorry you experienced this.
https://www.thisamericanlife.org/561/nummi-2015