Slashdot Mirror
Elon Musk's Alleged Email To Employees on Tesla's Big Picture (jalopnik.com)
An email allegedly sent by Elon Musk to Tesla staff has announced that the Model 3, which has faced a number of production issues, will go into "24/7" production by June, resulting in 6,000 Model 3 units made per week. But apart from this update, in the email, Elon Musk sheds light on how much he values precision in his cars. An excerpt: Most of the design tolerances of the Model 3 are already better than any other car in the world. Soon, they will all be better. This is not enough. We will keep going until the Model 3 build precision is a factor of ten better than any other car in the world. I am not kidding.
Our car needs to be designed and built with such accuracy and precision that, if an owner measures dimensions, panel gaps and flushness, and their measurements don't match the Model 3 specs, it just means that their measuring tape is wrong.
Some parts suppliers will be unwilling or unable to achieve this level of precision. I understand that this will be considered an unreasonable request by some. That's ok, there are lots of other car companies with much lower standards. They just can't work with Tesla.
Our car needs to be designed and built with such accuracy and precision that, if an owner measures dimensions, panel gaps and flushness, and their measurements don't match the Model 3 specs, it just means that their measuring tape is wrong.
Some parts suppliers will be unwilling or unable to achieve this level of precision. I understand that this will be considered an unreasonable request by some. That's ok, there are lots of other car companies with much lower standards. They just can't work with Tesla.
I wonder if Tesla has failed to use variation simulation tools?
There is no need for precision 10 times greater than other car companies. That is just wasteful! They need to find out WHERE the precision is needed, and HOW MUCH precision is needed. Blindly improving precision "10 times" is ridiculous.
I worked on variation simulation technology in the 1980s. This is the current version of the product I worked on:
https://www.plm.automation.sie...
Hopefully, Tesla is using this or something similar.
I originally ported this code from code written by a university professor at Wayne Statue University in Detroit, and then designed a domain-specific language and implemented a compiler for it, to make models easier to write. (Probably the most important thing I did, though, was to strong-arm my boss into hiring a mathematician to help clean up what was some pretty awful and buggy statistical and geometric-transform code...) The product has changed hands a couple of times since then, before landing at it's current home at Siemens.
The original company that developed this (where I worked) both created the product, and worked with the Detroit automakers on several breakthrough projects that address just where Tesla should be applying this.
For example, the 1984 Corvette C4 was the first car out of Detroit to use BOLT HOLES instead of slots in hood hinges. This was made practical with VSA analysis.
There was a big push for lowered emissions at the time - VSA allowed auto companies to model variability between engines, and predict what percentage would be rejected with a given design.
An important re-design of the FA-18 used VSA modeling extensively, and solved many manufacturing problems with the airframe.
I recall MANY door clearance and other similar fit-and-finish projects.
You could not build today's disk drives at a practical cost without VSA. Every drive manufacturer uses it.
Before VSA, it was largely guesswork. Once you get past a liner stack, it is nearly impossible to work-out by hand. There was some prior use, during WWII. One of the first - if not the first - uses of VSA was in WWII when the technique was developed at Willow Run Labs to solve manufacturability problems with planes being built for WWII. It was done crudely, with a room full of workers on manual calculators...
Professor Greg Gruska at Wayne State dusted off the mothballs in the early 1980s, and wrote some Fortran code to implement it on their mainframe (the code I had to port to IBM PC...) and taught a class in variation simulation analysis. I was the first technical employee at the company that commercialized it.
I believe there was some parallel work in Japan at the time, and there are a couple of competing products.
Did Tesla somehow miss this important analysis technique?
The reason to require 3 nines when only 2 nines are needed is so that when someone misses the spec the vehicle still works. Consider two scenarios:
1) A car with 10,000 parts is assembled. The tolerances were exactly specified, so any tolerance miss creates a non-working car. All vendors meet tolerance 99.999% of the time. 10% of the cars coming off the line won't work. (So you will have to spend money ripping them back apart, more testing of the parts to find the 0.001% of the parts that are bad, etc. Tests with false positive rates lower than 0.001% are hard)
2) A car with 10,000 parts is assembled. The tolerances were over specified, so only 10% of tolerance misses create a non-working car. All vendors meet tolerance 99.999% of the time. 1% of the cars coming off the line won't work.
This appears to be an extension of the "Kanzen" technique originally used by the Japanese car manufacturers. It took them from essentially not competitive in the US to a dominant position in very few years.
while (sig==sig) sig=!sig;
Does he think other manufacturers make sloppy parts on purpose?
The process defines the margin. Tesla injection molds/presses/machines tools will _not_ be 10x more precise then others. I don't care how often Musk has them change the tooling.
I worked for a manager that didn't understand dimensioning. She added two 0s to a dimension on a drawing before sending it to a contract manufacturer. Those would have been some insanely expensive 0s, if they hadn't been just undoable. Specing an O.D. to 0.25000 inches. How do you even do that?
Tesla has _much_ bigger problems. If the documented test cars are a guide, Teslas will _all_ be junked within a few years of going out of warranty. Motor sets (22k$US) are being replaced, on average, every 2 years. Cars are being totalled by insurance companies after being caught in heavy rains with the windows open (flooded battery pack, about 50% probability that seal will hold). 'Fender benders' cost 30k$US to fix.
He is distracting. The fact is that Tesla has made some incredibly bad decisions beyond electric power...The door handles jump to mind (talk about a _dumb_ unnecessary complication). 100% aluminum body. Auto Driving vapor. An electric drivetrain was a reach, adding all sorts of 'seemed like a good idea' bells and whistles doom the company.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Maybe the domestic US manufacture is not that good, but any of the premium German or Japanese manufacturers will be that precise.
I worked in a Detroit factory. We made parts for Ford, GM, Toyota, BMW, and Lotus. Same machines. Same processes. A dirty secret of Japanese and German precision that Musk needs to learn is they both outsourced the finer details to suppliers with a century of experience in Michigan.
Precision manufacturing only coming from Germany or Japan is marketing copy. Today, parts are like programmers. They can come from anywhere.
Klipstein's Law: Tolerances will accumulate unidirectionally toward maximum difficulty of assembly.
THE SOFTWARE, IT NO WORKY!!!
Yes. And that Japanese cars were built to last. Buy an American car in the 70/80s and they broke. Soon. Any many Japanese cars were still running 10-20 years later.
This is probably before your time - a common issue now.
Every single American car I owned up to 2000 broke. Or wore out extremely fast. I then tried a Mazda. The only part that failed in that was the transmission, guess what, made in the USA.
This is why 20 years later, I will still never purchase a car that is American or built in America. I need a reliable vehicle.
This is my only exception to buying non American goods. But I need to get to work and home from work without a tow truck.
Recall the bailout of Detroit. Recall the warranties were 5/5,000 years instead of 10/100,000
No, the parachute way is needlessly complicated. Parachutes would add more weight than just putting some extra rocket fuel on would, you've added hardware that is ONLY used for recovery and has additional failure modes, and on top of that it has been established that the rocket wouldn't survive without an entry burn, because the engine exhaust pushes the entry shockwave away from the rocket and keeps it from melting. With propulsive the whole way, the only needed complications are the software needed for landing and the ability to relight engines - the latter of which they would need anyway for the final landing.
I'll add to this that SpaceX *tried* the parachute way first, and the stage broke up in the upper atmosphere every time before it could even deploy the parachute.
Inferiour to a Windows PC, most definitely not.
In the Windows 3.1 era, that may have been the case. However, once Win95 became available, it completely spanked the Mac in every possible way, including reliability and window management. Apple made no headway on Copland and aggressively defied adapting a taskbar or similar mechanism for window management to avoid the stigma of copying Microsoft. By 1995, productivity was way, way higher on a PC as Apple refused to adapt to new ways of working, and continued pushing their views of apps using fixed amounts of memory and having multiple windows open per app, a la Desk Accessories. That just covered the usability, mind you -- I don't even talk about performance per dollar. It was a mess, and every student at my university saw Macs as a joke. Only the teachers were die-hard Apple fans.
It was actually school policy that our departments were only allowed to buy Macs, no matter how much we begged the school to let us buy PCs. It was much easier and faster for me to leave class, go to my dorm, do my assignment on my $800 PC, print everything out, and walk back to class than to do the work in the lab on a brand new $5,000 Mac.
Having recently seen Amiga die, I was certain Apple would also be out of business by the end of the century. That very nearly happened, had Apple not reached an epiphany: they finally accepted they couldn't design or maintain their own OS, and they should give up and buy someone else's. Classic Macs were a trainwreck.