'A Lot of Hoped-for Automation Was Counterproductive', Remembers Elon Musk (bloomberg.com)

Thursday Elon Musk gave a surprisingly candid interview about Tesla's massive push to increase production of Model 3 sedans to 5,000 a week. An anonymous reader quotes Musk's remarks to Bloomberg: I spent almost the entire time in the factory the final week, and yeah, it was essentially three months with a tiny break of like one day that I wasn't there. I was wearing the same clothes for five days. Yeah, it was really intense. And everybody else was really intense, too... I think we had to prove that we could make 5,000 cars in a week -- 5,000 Model 3s and at the same time make 2,000 S and X's, so essentially show that we could make 7,000 cars. We had to prove ourselves. The number of people who thought we would actually make it is very tiny, like vanishingly small. There was suddenly the credibility of the company, my credibility, you know, the credibility of the whole team. It was like, "Can you actually do this or not?"

There were a lot of issues that we had to address in order to do it. You know, we had to create the new general assembly line in basically less than a month -- to create it and get to an excess of a 1,000-cars-a-week rate in like four weeks... A lot of the hoped-for automation was counterproductive. It's not like we knew it would be bad, because why would we buy a ticket to hell...? A whole bunch of the robots are turned off, and it was reverted to a manual station because the robots kept faulting out. When the robot faults out -- like the vision system can't figure out how to put the object in -- then you've got to reset the system. You've got to manually seat the components. It stops the whole production line while you sort out why the robot faults out.
When the interviewer asks why that happens, Musk replies, "Because we were huge idiots and didn't know what we were doing. That's why."

Comments

Re:So Musk Admits...

[bgarcia]

Would that be the same 120 year old trillion dollar auto industry which said it was impossible to design and develop an electric car that people would actually buy?

Re:best way to do it

[SWPadnos]

From TFS;

That seems to me a lack of fault tolerance in the overall design approach. A major oversight if so. An entire line should not shut down due to one fault.

Well, yes and no. Assembly lines are tuned so that you produce the correct quantity of each item in the same amount of time it takes to complete the "full product". It's called Takt time. There are just over 10,000 minutes in a week, so the production goal of 5,000 cars per week requires a new Model 3 to roll off the line every 2 minutes.

Let's say they're building the battery packs as part of the assembly line, and it takes 60 minutes to build each pack. Just for grins, let's assume that it takes 10 minutes to install that battery pack. This implies that you need 6x as many battery assembly stations (or sub-assembly lines) to make batteries as you need battery installation stations.

If you have a robot installing batteries, and that robot fails, then you want to stop making batteries before you fill up all the available space with battery packs you can't use (because you can't install them). Similarly, you no longer have chassis with batteries to move further down the line, so anything after the battery installation has to stop, at least for that line.

There is slack in the process - it might be 59:30 to build the packs or 9:18 to install them, and there would likely be some parts storage as well, so the line could whether an employee going to the bathroom or something. It couldn't whether a continuous failure or a permanent change in process time (ie, it may take a human 13 minutes to install the pack and inspect their work).

Most assembly lines work this way. The more slack you have, the less efficient your line is (because you have all this extra time ...). The less slack you have, the less you can tolerate a fault.

I wonder how much $$$ was lost on that automation equipment now collecting dust.

I'd bet most of the lost money is whatever was spent on programming the units, and the lost opportunity cost for both the hardware and its configuration. The robots themselves are likely still usable, but they'll need to refine how they're used in order to recoup that investment.