From the main thread:
This was in response to:
"Jerome added a note, with great emphasis, that the Tesla Grohmann equipment has created much better machines, dramatically improving the overall manufacturing efficiency of the Model 3. He also said they are building a giant machine using Tesla’s Grohmann sub-division, something he was clearly eager to share but couldn’t yet talk much about. Hmm. He basically just emphasized that it was a “giant, giant, giant machine” that duplicates everything, is modular, is simple on the modular level, and … is gigantic. We’ll all have to wait a bit longer for more information on that."
From here
I'm not sure we have quite visualised the gigantic machine yet but the important thing here is alien dreadnought is still very much a thing.
I have no idea what Tesla's plans are, but this is my first principles impression about the topic. First, the status quo:
Based on this it's IMO pretty obvious what the "Dreadnought" design should be, roughly:
- The fastest automotive production lines on the planet are over a kilometer long and consist of hundreds of stations, with most of the stations specialized to a single, very small task. Each station has a carefully calibrated time window to perform the step they handle, without slowing down the rest of the production line. They create a new car every 60 seconds or so.
- This design can perform very well when designed well and executed conservatively, but has a number of disadvantages:
- The maximum production capacity of a factory is pretty much set on the design table, years in advance. Each station has a capacity and to improve the throughput of hundreds of stations all stations have to be made faster or more substations have to be added, on a usually cramped production floor.
- As a result the lead time of capacity expansions is 3-5 years, the time to build a new car factory from scratch. This slows down the design cycle to snail's pace: the previous models are to a large extent cast into stone, and the next model has 3-5 years of lead time, sometimes more. This is why advances in the automotive industry are happening so slowly, and this is why there's so much resistance against change.
- But there's production disadvantages as well: the whole production line is exposed to failure of a single critical station. If a station goes down unexpectedly then it immediately stops production in the pipeline up to the next and previous 'overflow line' that can buffer half assembled cars. If the outage is so long that later buffer lines drained or previous ones overflow then the production stoppage avalanches forward and backward and stops the entire production, idling the workforce and stopping a billion dollar capital from earning its investment cost. I.e. losses of millions of dollars per hour are possible.
- In software terms the automotive industry's production method is very monolithic. This design choice originates from the original Ford integrated assembly lines 100 years ago, from around 1913:
- Technological innovations like industrial robots were basically shoehorned into a 1913 factory design. This is similar to the mistake that early steam engine driven factories committed when they converted to electricity: they reused the cramped, high density steam machine factory design which had to be dense due to the physics of the mechanical belt power distribution method, instead of using electricity properly and spreading out their factories. It took about 100 years after the electrification of industry for electricity to be used properly in factory design: big, sprawling buildings with enough space to expand physically, and electricity transported via power lines.
- Anyway, Ford's 1913 design was IMO never seriously changed as new technologies become available - I believe in part due to the long lead times, slow iteration frequency and general technological risk avoidance in the automotive industry. If we take the average iteration duration as 5 years, then the automotive industry is only on it's 20th version of its factory design, with evolutionary and not revolution steps along the way. While the comparison is not apples to apples and thus not fair, there's certainly agile software projects that do 20 revolutionary iterations in a single year.
Pretty much the only drawback that I can see is that this cannot be installed at Fremont where space is at a premium, but requires a large greenfield factory from scratch. The 'carts', the 'conveyor system' and the preemptively less dense installation of modules needs more factory space than is probably available at Fremont. They might still prototype it at Fremont though, as the design will work on smaller and larger scale just as well.
- Multi-purpose modular work stations: the fundamental unit is a "generic industrial station", made by Tesla Grohmann. This is basically a generalized industrial robot that is designed to change its role flexibly AFTER it has been installed on the factory floor. I'll talk more about this later.
- Utilizing the third dimension: beyond stamping machines that require strong, deep reinforced concrete foundations weighing thousands of tons, most of the stations in a car factory don't have to be on ground level. Why not have 3-4 stories instead of a single story? This not only allows better real estate utilization, but also allows flexible expansion of production via the use of elevators and multiple floors.
- Utilizing two dimensions (horizontal space) more effectively: instead of creating a row of machines that have a set ordering at the design table, I'd create a less dense two dimensional layout initially as well, allowing for future growth.
- Utilizing FSD "EV carts" to move units between flexible work stations: I'd use self-driving EV "carts" that carry the partially assembled cars from station to station autonomously. They'd recharge themselves autonomously as well, whenever their battery levels get too low. Note that self-driving carts could also replace elevator systems (which are single points of failure), instead ramps could be used where the EVs carrying the partially assembled cars could move between floors - like in a multi-story parking garage.
- Automated conveyor system that feeds parts to sub-assemblies and the main assembly line.
- Much more parallelism: instead of having a single fundamental major assembly line flow that sub-assemblies feed into, I'd define multi-station 'clusters' of production that receive units in parallel. Any already installed generalized "Grohmann industrial station" could be retooled within a few hours to serve a completely different role: instead of installing a wire harness they could be applying screws or could be fitting windows. This means that if there's an imbalance in the capacity of factory during ramp-up, it doesn't have to be physically modified (or only to a minimal fashion), production capacity could be set aside to improve another workflow. No rails have to be modified and no serial production line gets upset: self-driving EV carts could just re-route their flow utilize the slightly different factory layout.
- Much less manual labor utilized: the Grohmann industrial station is 100% robotic. The main advantage of manual assembly labor is its flexibility - but if this flexibility can be offered at the industrial robot station level, it's a game changer.
- Flexible, gradual ramp-up that has good capital utilization during the ramp-up phase already: because each modular work station is flexible, a new factory with 10,000 units/week final capacity can be ramped up and can already make 1,000 units/week with just around 10% of the capital spent for the full production. The only 'extra overhead' is the large building built - but as we've seen it in Shanghai, large buildings can be built very, very quickly. Also note that even at the 100 units/week stage the factory can already be capital efficient and can expand "itself" without interruption of production equipment: because each station can be programmed to assemble 10% of a car, or just 1% of it, the role of each station can become smaller and more specialized as time goes on and capacity goes up. I.e. you don't have to spend capital, specify, order and wait for hundreds of stations to be built by first parties first, you can literally bootstrap a new factory with a large building and a dozen of work stations to produce ten cars per day or so. The factory will probably be earning money at that early ramp-up stage already! New stations can be installed without disrupting existing production flows - this is inherent in the less dense and autonomously routed and flexibly reconfigurable workflow.
- New stations can also be 'tested' without risking the existing production flow: a couple of units are handled at half speed or a quarter of speed, and the result in QA'd carefully. Note how this spreads out R&D as well: new optimizations can be applied anytime on existing stations as well, there's no pressure to "finalize" the design and put "pens down" at an arbitrary date when hundreds of millions of dollars worth of tooling is ordered. Instead it can all be done gradually and organically, optimizing it as they go.
Gigafactory 1 and starting the production of the Tesla Semi on this platform, with low initial capital investment, would be an excellent approach IMHO.
Note that the tight labor market in Reno is not a problem if indeed their new factory design is almost entirely automated: Gigafactory 1 will be the dream job for industrial robot and production engineers, and I expect Tesla to do a super job attracting talent via luxury amenities. It won't really show up in CoGs as the factory scales up it won't have to hire all that much bigger of a workforce.
Basically I expect Tesla's Dreadnought to eliminate "production hell" entirely, by turning the ramp-up of a factory into a largely software space problem. (They might even be able to simulate a full factory, based on the known capabilities of the generalized work stations.)
There's much more to this topic, but this comment is way too long already ...
This was in response to:
"Jerome added a note, with great emphasis, that the Tesla Grohmann equipment has created much better machines, dramatically improving the overall manufacturing efficiency of the Model 3. He also said they are building a giant machine using Tesla’s Grohmann sub-division, something he was clearly eager to share but couldn’t yet talk much about. Hmm. He basically just emphasized that it was a “giant, giant, giant machine” that duplicates everything, is modular, is simple on the modular level, and … is gigantic. We’ll all have to wait a bit longer for more information on that."
From here
I'm not sure we have quite visualised the gigantic machine yet but the important thing here is alien dreadnought is still very much a thing.
