Waiting4M3
Active Member
It seems to me the blogs invented these headlines, maybe they got too carried away, maybe they wanted to attract some more traffic.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Zhelko Dimic
Careful bull
You're new to Tesla, right?
Big-T
Member
Don't worry Hank, any time a less than positive comment is posted it gets hit hard. The more time I spend here the more I appreciate seeing an objective and/or critical opinion.
I'm well aware of their track record for hitting timelines. If his argument was based on their history, then I would be fine with it (and somewhat agree with it). My point was that it's impossible for any layman to look at a partially finished complex system and figure out how long it will be until it is completed. Tesla is claiming that they've learned from their past mistakes and everything is on track, so that's what I will choose to believe until credible information is presented to the contrary.You're new to Tesla, right?
When you see body-in-white samples doing repetitive cycles for an R&R study of the robot programming, you are starting to get close.
Then pathfinder units need to be evaluated and tested before making a thousand mistakes before you catch it.
Here's what the most modern mfr'g does. Using statistical control analysis, they run enough operations to determine CpK values. This is Process Capability. How often does it screw up? All processes have a failure rate, so you aim for a target, ideally 6 sigma.
We can look at 120 components and tell how many faulty ones will exist in a lot of 1,000,000. But you also need to keep testing sample batches during production to determine the decay or anomalies of the CpK levels.
I was glad to see the silver alpha Model 3 up close but wish we could've walked around it as we could with the model that they had on display at the Gigafactory last July. Rides in the alpha model would've been icing on the cake.
We did some driving between Palo Alto and Fremont but failed to spot any RCs. Of course once we got to the airport, we heard about all of the other sightings that we had missed out on. Doh!
As others have mentioned, we did see some interesting Model 3 related "stuff" in the factory. It was almost like the "cue the deer" moment in Funny Farm and I wondered if it was intentional or pure luck.
Overall we enjoyed the VIP tour and Q&A lunch. Thanks, Tesla!
We did some driving between Palo Alto and Fremont but failed to spot any RCs.
Of course once we got to the airport, we heard about all of the other sightings that we had missed out on. Doh! As others have mentioned, we did see some interesting Model 3 related "stuff" in the factory. It was almost like the "cue the deer" moment in Funny Farm and I wondered if it was intentional or pure luck.
Overall we enjoyed the VIP tour and Q&A lunch. Thanks, Tesla!
Waiting4M3
Active Member
Production can start before you have 6-sigma confidence, you still do the same acceptance tests, and the line not up to 6-sigma just means your rework rate is higher than you want.
Just to get a feel, at 6-sigmal, each process has a fail rate of 2 parts in a billion, if you have 100 steps in the assembly, then your chance of getting a good end-product without rework is 0.999999998^100, or fail rate (bad car) of 0.2 in 1 million. If you have 1,000 steps, then fail rate goes up to 2 bad cars in 1 million, and 10,000 steps gives you chance of 20 bad cars in 1 million. So 6-sigma is very very strict, failure rate is very close to 0.
For early production for sure you won't hit this rate. For example out of 10,000 steps, 1,000 of them are not yet well controlled, and your variability on them is 2x as large as you wanted, so you get 3 sigma, which gives a fail rate of 0.27% per step. Aggregating these 1,000 steps, your yield of getting a good car is 0.9973^1000=6.7%, so almost certainly you will need to rework. If you can fix most of them, and only have 100 problematic steps left, then your yield is 0.9973^100=76.3%, now you're doing a lot better. If you plot the yield vs # of problematic steps using the above example, you get this:
However the last few problems are usually the hardest to fix so they would take longer in terms of time, so if you translate this trend to have time on x-axis, the last few problems (to the right side of the plot) will have their x-axis stretched out, so you end with the "S" curve
Attachments
stopcrazypp
Well-Known Member
For the record, Tesla says the target is 500k cars total in 2018 (previous was for 2020), not 500k Model 3s.
Tesla puts pedal to the metal, 500,000 cars planned in 2018
Waiting4M3
Active Member
As Chris Rock once said, there is no sex in the champagne room, no matter what they tell you.
Yes, there is a bit of hype in that headline. There is still nothing there that justifies the expectations that some people seem to have had regarding the event. It is people that let their expectations get the better of them that caused all the ruckus with the HUD and, to a degree, all of the anti-selling going on now.
If true, I guess I can kiss my estimate (using the posted algorithm) of 10/18 delivery, ba bye.
Yes. It's more of a theoretical starting goal, not an on-going process target. If you can control the process to 6 sigma before production, you are going have far fewer QA issues in the long run. Basically you KNOW what is required to make uniform parts/processes, so you isolate variables as the CpK falls over time. And it will. A more common goal however is 1.67-1.33.
But more importantly a CpK curve of 2.0 (6 sigma) is not making 'perfect parts'. It's making parts that are virtually always within tolerance, but can have a significant range of results if the curve is bell shaped. When manufacturing, having all the parts/processes uniform, allows more tooling wear, more user variation, more material variation, etc. So even a 6 sigma value on a critical hole dia could be insufficient to allow mass production for extended periods due to decay of the tooling, training, or material.
BTW - We see CpK values over 2.0 constantly on individual characteristics. If the tolerance is wide or process cannot vary by design, the numbers get really high.
Yes, Tesla's announcement of the event was worded correctly. The Electrek article headline was slanted towards emphasizing the Model 3 aspect of the event, obviously to entice people to click and read it. Clickbait.
From that article, and others like it, some people got the idea that it was a "Model 3 event". It was not.
I agree. I think first deliveries in July are mostly handmade, and going to board members/investors. Production likely to start in August sometime, but will be very slow initially as they work out kinks.
Yikes.. That room was under video surveillance so I would think Tesla can easily figure out who took those photos (I saw several employees go in/out of a room that had a screen with several cameras from the room). I did see several guys told they can't take photos (after they already did) and then they went through their phone with Tesla employees to delete any photos. I'm actually surprised someone risked it.
Thanks Erik
To share another aspect of modern manufacturing... Tesla has likely created not only a digital twin of the car (all its parts, matierials, etc.) but also a digital twin of the production, i.e., they've modeled the end-to-end production line(s) and run simulations to optimize and validate.
When you're creating a from scratch line, it's what you'd do now. Some hand assembly (shifting to as automated as you can afford) may be done while the line is brought up and debugged -- can be faster than it used to be as it was all run virtually to develop it. This can also be seen in less prototype tooling being created -- don't always need all of it to get to good production tooling nowadays -- instead it's a lot is done virtually. And of course validated physically.
All that said, once you get the line going, you can really ramp up fast. This was alluded to in articles about the gigafactory plans to ramp up M3 battery production -- some manual assembly shifting to a massive ramp up of automated production.
Big-T
Member
I've seen some articles about how Tesla is skipping the "soft tooling" phase of the production line. Hasn't seem to be commented on much here, I suppose it's easy to brush it off....as long as it works. Anyone have any thoughts on whether this could cause some headaches?
If you have faith in your models and simulations, you go directly to production tooling. And remember, this is not Tesla's first goat rodeo. That said, they seem to be pushing hard, so glitches (including supply chain and quality issues) can cause some fits and starts on the path to the targeted production ramp.
