As Hamtramck’s assembly line tried to gain speed, the computer-guided dolly wandered off course. The spray-painting robots began spraying each other instead of the cars
When a massive computer-controlled ‘robogate’ welding machine smashed a car body, or a welding machine stopped dead
automation works best when it’s added incrementally to a production process that’s already working smoothly
Industrial robots don’t use machine learning or neural networks.
They only do one thing. Why is it so difficult to make them work ?
They have to do one thing very precisely and without interfering with each other.
Example:
Concerning the article:
That is bullshit. We are in the 2020s, not the 1980s. Which new large scale automation is still based on relay logic? Non! Because that stuff is technically irrelevant by now!
I don´t get the point. Do people want cars from 1985 back because more of them can be made?
Volkswagen has manufacturing issues all the time and they have been in the business for a long time. You hear about them investing a billion Euros here or there, but the electric cars Volkswagen makes are pathetic compared to Tesla.
I know that, but I just assumed, that if the task doesn’t require the robot to use any kind of AI, then it’s trivial.
Elon Musk believed, that he could make 6000 cars per week, and allowed as many customers to pre-order, as he thought he could manage.
The problem, is that after the pre-orders were made, he discovered, that he could only make 2000 per week, which means that the customers have to wait longer to get their Tesla, and are starting to loose their patience.
One machine does one thing but then add hundreds of machines doing hundreds of things all having to be in time with each other. It’s not plug and play.
This is typical Elon jumping in with both feet and realizing that doesn’t work, which the article explains.
Besides unions and not every plant has them, why else do you think established auto makers still use humans?. People are easier to configure.
Closest thing to this topic I know is just creating macros
Basic mistake is to start playing with actual time, shit just goes wrong like that, and ideally you want to create loops without time so shit works exactly as wanted, but if you simply cannot figure out how to do that then you just have to do it like brute
Former industrial automation tech chiming in here. Your code can only be as fast as the machines it controls. The slowest machine is what everything has to be set against.
Much harder than it sounds. You can spend months tuning the code to reflect the reality in the field.
A lot of automation programers don’t have mechanical experience so it is a bit of trial and error.
For example I was working on a natural gas plant. Sometimes throughout the day we would have low flow alarms that would stop the system. After breaking into a pipe we learned that when a certain pump kicked on the check valve required a minimum pressure above what we thought to fully open. And the variable frequency drive didn’t ramp up fast enough to overcome it.
Also with robots/automation if you replace a person with it you need a plan as to what to do if it fails. Do you have redundancy or a way to take over manually until the problem is fixed.?
Hell even networking can get tricky since latency can effect equipment that runs in real time.
Normally you do this sort of swap over in phases so if an issue is found it can be corrected before moving on. It looks like they tried to do it all at once so it becomes troubleshooting a Rube Goldberg machine.
If it was a brand new facility tons of testing occurs before the final commissioning.
Edit: also your sensors are not the same as how a human sees the world so a thing that looks basic to you may be misinterpreted by the computer.
The complex movements needed for certain task are easy for people since out brain is more of an approximation engine than computer. Tuning “pick up part” is a lot more logic than you would think because each motor and servo need their I/O.
That was a good perspective to read. I sometimes think with engineering machines that engineer things, people do forget how fraking hard it is.
This is vital, and one of the things SCM sets out to optimize. Finding constraints and making things more efficient only works if you’re realistic about what your constraint is.
I feel like this guy is trying to reinvent all these wheels, when there are time tried practices that could solve all his company’s problems. Shit has been saving companies money since the 1940s.