"Unboxed" Gen 3 manufacturing Process

[UkNorthampton]

I call BS. It doesn't matter if you do task X and I do task Y on the same car at the same time or we do it on two different cars at the same time. It's still the same amount of time per task. Efficiency is the same. If unboxing allows individual tasks to be faster thanks to better access or something that helps. But parallel by itself doesn't buy you anything.

On a production/conveyor line, you have the same time per station. If you do a few tasks on an unboxed setup, less dwell/walking back time per task.

Plus better access and ergonomics as you mentioned.

 

[aronth5]

I call BS. It doesn't matter if you do task X and I do task Y on the same car at the same time or we do it on two different cars at the same time. It's still the same amount of time per task. Efficiency is the same. If unboxing allows individual tasks to be faster thanks to better access or something that helps. But parallel by itself doesn't buy you anything.

I wouldn't call BS until we have a better understanding of how the improved access will leverage what Optimus can be used for. I suspect in time that will be significant since the unboxed approach appears to reduce the amount of time a worker has to get inside the vehicle. If that means Optimus can perform more tasks than all the better.

 

[Buckminster]

Not sure if its been mentioned on thread or overlooked - Elon said the CT now has full ethernet connectivity between control modules.. This is one of the most important parts of achieving the 'unboxed' assembly.

Along with steer by wire - the constraints of traditional wiring loom solved and in a production vehicle bodes well for Next Gen..

 

[Buckminster]

Ethernet between modules and 48V are just changes in wired architecture. It doesn't eliminate the need for harnesses in the separate body section. Main impact is the size of the harness and the connectors.
Tesla has filed for patents on wiring configurations that shift the connections around, but I don't understand how this would be so critical to the unboxed process.

 

[Buckminster]

We drive a 6 year old BWM i3. It has thermo-plastic body panels which clean up as new still.

Surely Tesla must have looked at this? I believe the primary problem with the i3 was the time and complication of the carbon-fibre and body panels

Any thoughts on this?

Also, in the Munro interview, Elon didn’t agree or disagree to the unboxed method which Munro mentioned.

 

[Buckminster]

https://twitter.com/x/status/1734768107502334326

 

[Gigapress]

I call BS. It doesn't matter if you do task X and I do task Y on the same car at the same time or we do it on two different cars at the same time. It's still the same amount of time per task. Efficiency is the same. If unboxing allows individual tasks to be faster thanks to better access or something that helps. But parallel by itself doesn't buy you anything.

Unboxing does improve efficiency, because it's not just "parallel by itself".

Unboxing gives Tesla:

• More value-added work per unit of floor space.
• Less wasted flow time in which parts are moving from one station to the next.
• Better worker ergonomics due to not needing to lean in and out of the car frame
  • Less wasted time on movement.
  • Less disruption from overuse and head-strike injuries.
  • Probably better quality
  • Better visibility of the work being performed

7 wastes:

1. Overproduction: Producing ahead of what’s actually needed by the next process or customer. The worst form of waste because it contributes to the other six.
2. Waiting: Operators standing idle as machines cycle, equipment fails, needed parts fail to arrive, etc.
3. Conveyance: Moving parts and products unnecessarily, such as from a processing step to a warehouse to a subsequent processing step when the second step instead could be located immediately adjacent to the first step.
4. Processing: Performing unnecessary or incorrect processing, typically from poor tool or product design.
5. Inventory: Having more than the minimum stocks necessary for a precisely controlled pull system.
6. Motion: Operators making movements that are straining or unnecessary, such as looking for parts, tools, documents, etc.
7. Correction: Inspection, rework, and scrap.

What are the 7 Wastes in Lean? | Lean Enterprise Institute

The 7 wastes are common behaviors in business that do not add value for the customer, and therefore, must be eliminated through kaizen.

www.lean.org

 

[Buckminster]

https://twitter.com/x/status/1743330355476578551

Takt time of unboxed could be 10mins or more with ~100 identical stations in parallel. Not sure how many folk think that Tesla will actually use unboxed. ~50% think M2 will be a single gigacasting which doesn't sound unboxed compatible:
Poll: let's build the Robotaxi

 

[Buckminster]

https://twitter.com/x/status/1748161130105577791

 

[UkNorthampton]

https://twitter.com/x/status/1748161130105577791

Teardown person wasn't that great at presentation, wouldn't speculate on OEM chances/changes/retrofit & wanted to sell his reports / services (fair enough).

What I found fascinating

1) Spent an hour in some client meetings (OEM/Tier 1), realised he wasn't getting through and walked out (key point of what I remembered). Seemed massively fed up with dealing with Legacy management, especially below exec level eg manufacturing heads.
2) He doesn't seem to have ideas on how Brownfield Legacy sites or culture can be reformed. Best start new as Greenfield, ideally with new culture
3) Usual anti-Tesla pundits stunned, didn't seem to have any answers or rebuttals.
4) Customers of Teardown company just want to benchmark Tesla or perhaps (other) Chinese companies. No benchmarks of Euro or USA cars otherwise, irrelevant & non-competitive.
5) Legacy production kit is "Sunk Cost". Fallacy that it may be worth much, although Tesla produce different Y over time & location.
6) Encouraged OEMs/Tier 1 to experiment
7) EV efficiency is king - platforms dead, optimise for each model. Huge implications for product ranges, marketing niches etc. Again tesla/Elon identified this and just went for a few sector-dominating models.
8) Encouraged Legacy to experiment to get to a design that makes sense financially.
9) Tesla/EVs for the win!

 

[kpanda17]

Thing of beauty

having grown up with match box cars, the analogy is amazing
interesting, when one of our match box cars wore too much, the clam shell usually broken open
hope the gen 3 cannot break open

 

[Buckminster]

https://twitter.com/x/status/1750961586699927964

Unboxed confirmed

 

[Buckminster]

Unboxed unlocks efficiency. The ability to assemble subcomponents independently is the next step in optimization.

Picture the current general assembly line: a full painted chassis rolls down the line in its full volume taking up a car sized chunk of space at each station. The conveyor only moves as fast as the slowest process. Each process has a defined section of line and reducing process time means increasing the number of that station (or people doing that task simultaneously).
Mismatch between the time tasks take is inefficiency. Excess capacity at a station is inefficiency. A new less efficient worker can slow the entire line.

Now, concider a station on the line. Worker takes a part from a box, fasteners from another, and attaches the part. How long did it take to make the part? How long did it take to make the fasteners? We don't know, and in terms of final assembly, we don't care. As long as there are parts ready to go, the line can progress at the speed it takes to assemble.

Doors are sort of like this. After painting and removal, they are assembled on separate lines. However, in typical manufacturing, the chassis needs to get the same doors for paint matching, so the GA line is bound to the door line.

What if any door can go with any car? Well then, you can prebuild as many doors as you want off line and buffer them for assembly. Instead of only going at the rate they exit the paint shop, you can assemble as quickly as possible. With asynchonous specialized lines for each subcomponent, you don't need the constant material movement and can group multiple assemblies together for multiple workers to build.

Rather than one person on each of the four doors, you can have four people building the same type of door; buffer, then build the next type. That shifts production speed from what the slowest worker can do to the combined output of all.

Example times to complete a task for four workers: 35, 40, 45, 50 . Separately, a car can only be built every 50 seconds. Combined, a car can be built every 41.8 seconds (a door every 10.4 seconds times 4 doors). Plus, any delays or problems don't impact the main general assembly line thanks to the buffer. Need to go faster? If you add an even slower worker, you speed things up! A 60 seconds per door fifth worker reduces the time to 35.6 seconds per set of 4 doors. That's a 15% improvement!

Tesla is doing this already with interior on pack construction. Unboxed carries it further by making each chassis quadrant a subcomponent that can be parallelized in minimum floor space.

Then, the only limit on production rate is the speed of the final assembly steps. Just eleven major body sections and the glass. This process reduction allows the line the room to duplicate steps as needed to achieve the desired rate. (Yes, paint shop is a restriction, but that can be scaled).

Subcomponent assembly also lends itself to multiple process/ efficiency optimizations, but that's a ramble for a different time.

 

[MC3OZ]

They need to "make the machines, which make the machines which makes the machine" - hard to copy - bullish.

Hard to determine the right thread to speculate on this topic, but I chose this one...

We can describe "unboxed Gen3 as follows....

The car is built in separate sections in particular the "skateboard" rear casting, structural battery pack and front casting.. parts are added to each of these sections with good operator density as shown on investor day.

For this (section building) phase Operator could be:-

• A Human.
• An Optimus Robot.
• A generic Tesla machine, perhaps using arms similar to Optimus arms.

Less likely options are:-

• A conventional (KUKA) Robot
• Station specific manufacturing equipment, designed and built by Tesla.

This (section building) consists mainly of task humans previously did in GA, materials many be fragile and there are lots of steps.

The additional sections are the sides and possibly the roof, it seems likely that these can be built via fairly traditional "body shop robots". The roof could be a casting , or stamped panels, if may be joined to the sides before being joined to "skateboard" in the "assembly" process.

Assembly - joining the skateboard and sides,,, this could be a highly specific machine built by Tesla perhaps incorporating some conventional robots and some equipment using Optimus robots arms.

Cladding - adding painted panels to the vehicle body formed in assembly - if conventional robots are used we need to ensure they can not dent or scratch the panels. This is likely to be a highly specific machine built by Tesla.

For the Gen3 process, Assembly and Cladding could be the steps that determine the factory run rate. Good solutions are required and this equipment will be needed in al Gen3 factories.... it meets the definition of "hard to copy"...

For the skateboard sections these need to arrive at assembly process fast enough to keep up with the required run rate, at the assembly station, components for at least 4 separate lines need to merge... The roof and sides may come in via traditional line, but IMO skateboard sections could travel on carts, perhaps the sides will also travel on carts... So perhaps these cars "dock' as part of the assembly process.

The fact that skateboard sections need to arrive a final assembly on carts, then raises the question of how long they spend on the carts. I can see a model where carts visit various line stations in turn collecting parts,. This may be "grab and go" with assembly being done elsewhere.... So the line is line of carts and in the event of failure/rework, a cart simply drops out of the normal line.

For Optimus itself, I don't consider the hardware "hard to copy" but the software and the rate of learning may be hard to replicate. Any robot based on Optimus like arms and vision should inherit these attributes.

I don't know what Tesla has in mind, but I would consider attaching detachable Optimus arms to the carts, instead of using Optimus robots.

The advantages compared to Optimus are:-

• Firm heavy base, movement or arm should not induce motion or vibrations.
• No problem with balance, falls or movement in unexpected directions..
• Work while in motion should be possible,
• Relative coordinates of the skateboard section should be fairly fixed, with vision and touch adding granularity.

The advantages compared to heavy industrial robots are mainly:-

• Cost,
• Vision and touch.
• Speed/ease of training
• Gradual phase in...
• Adaptable to flexible "cart line" process.
• Can be made safe for humans working in close proximity. (padding, dock-button, etc)

I am sure there are trade-offs and what I am imagining might not be what Tesla intends to do.

 

[Artful Dodger]

I don't know what Tesla has in mind, but I would consider attaching detachable Optimus arms to the carts,

Lol, I don't think it'll be that complicated (or exotic, or weird). They'll start with human operators on a GA line which has been simplified as much as possible. They'll use motion capture and video to train bots to replace those humans, one-for-one, as soon as practicable and cost effective. Eventually, most of the operators will be bots.

 

[mongo]

Hard to determine the right thread to speculate on this topic, but I chose this one...

We can describe "unboxed Gen3 as follows....

The car is built in separate sections in particular the "skateboard" rear casting, structural battery pack and front casting.. parts are added to each of these sections with good operator density as shown on investor day.

I don't think the front and rear will attach to the pack before they attach to the sides. Physical crossection would require a lot of fixturing and the dimensional tolerance of the front/rear to sides is tighter than the pack to frame.

 

[Discoducky]

Happy to write more on this, but here's a few quick thoughts

• Designing out human required labor
  • Which are the hardest things that humans absolutely must do and what could Optimus do and what could automation do
• New gigapresses and not just new dies for existing presses. These will most likely have new dimensionality and will integrate even more parts into the casting, thus reducing parts even more.
  • I'd expect that they'll consider removing the need for "B" surfaces and "inner" panels
    • One piece gigapress bonding to one piece final exterior surface
• New paint shop to ensure that all parts match and 10x run-rate of existing
  • This is an entirely new machine and process to achieve unheard of accuracy across a wide amount of vehicles
  • NOTE: I hope they don't end up doing this and instead go with cold rolled and no paint. Should be overall higher sustainability
• New ECUs as you'll need to test and verify functional features prior to the front and back are connected.
  • I could see workarounds to this as you could have a software shim for the missing parts in the diagnostic code and be hooked up to surrogate ECU
• New structural connections and methods for attaching these at incredibly high success rates with redundancy
  • Front and rear glass get installed "unboxed" so the machines that handle these after glass install cannot flex them and or damage the fit and finish
• New low voltage electrical connections and I could see this working as a one step process where one giant machine brings the sides, front and back together at once and it all "snaps" together electrically
  • Once low voltage completes diagnostics, then the high voltage system is attached and the car is effectively finished electrically

 

[MC3OZ]

Lol, I don't think it'll be that complicated (or exotic, or weird). They'll start with human operators on a GA line which has been simplified as much as possible. They'll use motion capture and video to train bots to replace those humans, one-for-one, as soon as practicable and cost effective. Eventually, most of the operators will be bots.

I agree that is where they will start.

make the machines, which make the machines which makes the machine" - hard to copy

Obviously doesn't apply to humans,,, and IMO it doesn't apply to Optimus hardware, it would not apply to robot-carts. but the question is which robot form factor works best in a factory environment.. Just getting humans to do the GA jobs isn't hard-to-copy, using any kind of bot is hard-to-copy.

Hard-to-copy might simply be limited to assembly and cladding, they are the harder to solve problems and probably critical in terms of the run-rate, so that is where the initial focus will probably be.

For the sequence of assembly, I agree that pack might need to be added after the sides are attached...

IMO Tesla will use a variety of tools and the best tool for the job, even if they need to invent the tool.

Inventing the tools not only gets the job done faster and cheaper, it makes the whole process harder to copy, especially if multiple new tools are invented. The fact that a tool is "weird" and may take time to perfect/integrate doesn't stop it being part of the final solution... The only real question is if that tool is the best possible tool for that job.

I did hear somewhere that Tesla were working on robots on wheels, that seems to me like it could be a step beyond simply autonomous carts. Robot-carts might be able to charge via induction while on the job for true uninterrupted 24/7 operation.

While Optimus can walk, that will not be as good a wheels on a flat level surface.

Why I posted was I noticed speculation on "hard-to-copy" in the main thread while on holidays, including which tools could be used. It was obvious to me that the main criteria was the "the best tool for the job" and to understand that, we need to understand the job and the strengths and weaknesses of various tools in the context of that job..

For something like Gen3 unboxed, R&D on developing the best possible tools pays off.

 

[MC3OZ]

Paint is another area where a new tool will need to be invented, But that is probably just panels riding though a paint shop on carts, perhaps with a visual AI based check on paint quality and some automated correction/improvement process if needed.

The possibility of vertically stacking painting in multiple colors at the same time was something I mentioned previously.

The add building new tanks at the south end of Austin as part of the new extension, these could be for paint.

So why have paint?

Mainly due to consumer preference especially when making perhaps up to 10-15 million Gen3 cars per year.

If the panel material can be used unpainted, that also means that only the customer facing side needs to be painted,

If panel are sometimes painted, Tesla needs to ensure they are not scratched during the installation process.

 

[MC3OZ]

New machines are possibly also needed to make parts including parts like seats that Tesla already make.
if possible. Tesla needs to reduce the cost of seats for Gen3 cars..

In general, lower cost can be:-

• More vertical integration.
• More automation.
• Less/cheaper raw materials.
• Lower capex.
• Lower prices on higher volume parts supplies.
• Smaller factory footprint.
• Faster line speed.
• Fewer parts and processes,

.
Any path to lower cost and faster production ramps is well worth pursuing