Coding Error Forces 2019 Subaru Ascents to be Scrapped

Conceivably, they may have to dismantle the entire vehicle to get to a hidden weld, which could require compromising the structural integrity of the vehicle if dismantling is equivalent to cutting at any step in the dismantling process.

My initial thoughts were that the omission may have affected the integrity of the bodyshell or it requires a factory jig to ensure correct alignment for the weld.

Perhaps scrapping the 293 cars is cheaper and simpler than one potential lawsuit though.

Getting to bare metal would require so much of the vehicles interior to be removed, the chances of doing it without damaging said interior would be next to none, then doing the spot welds depending how close they are to the exterior could affect the paint, which would in turn require repainting, which would in turn require stripping the whole vehicle back to get the whole paint job even [edit: to get a perfect finish on a portion of a repair indistinguishable from the rest of a car is pretty hard]

To to go from "as new" to "Bare metal" to back again would be a logistical nightmare. There'd be so many little things "oops, damaged that plastic piece with a screw driver...binned !". And even if you did manage to get the welds in, you'd probably have to re-schedule a whole bunch of new parts for production, just for that batch.

The cost to pull that off would outweigh a new car - as it would all be skilled automotive labour (remembering you couldn't reverse the robotics that assembled part of the car in the first place).

People would be expecting to buy new cars that would have to be perfect. Even if by some miracle you did pull off getting the cars back to perfect condition, imagine if (when) the new owners found out?

291 would represent x%age of cars for that model year. Just bin them.

Disclaimer: I work for a Subaru competitor, any opinions are solely my own.

1. There are conditions where you cannot repair a vehicle and then sell it as a new vehicle (eg when you have to cut and weld). In these cases it's actually more cost effective to scrap it and rebuild it. Quite likely, most of the components of these vehicles will be used again on a vehicle with the same VIN. Scrap doesn't necessarily mean crush and recycle.

2. The weld process may only be certified on bare, unpainted sheet metal. After the paint process, basically none of the sheet metal is bare. Even the unpainted parts are coated with a pre-primer. Fun fact: sheet metal must be welded in the plant within a month of being stamped. (Aftersales consumer repairs are all manual work, so the same restriction doesn't apply)

3. As another commentor stated, even if Subaru had to eat the entire cost of production for these ~300, that is much better than a lawsuit. 300*50,000 = 15 million (probably more like 7-10 million). A single lawsuit could blow that away.

4. Cars are not really like airplanes. Airplanes have replaceable body components. Cars don't really have replaceable body components. A whole airframe for a large aircraft is probably worth millions. A welded auto body costs less than 5k in the manufacturing plant. A plane may take between a few days to a month to manufacture. A car plant makes hundreds to a thousand cars in a day.

> If missing welds were discovered on an airliner

Airliners are riveted together, as rivets are more inspectable and reliable.

Too much risk for the manufacturer. These welds are part of the unibody - what protects you in a crash.

A fun anecdote to explain the absurdity that is Auto manufacture.

When Ford made it's own vehicles in Australia, the engines were made in Melbourne, while the body and final assembly of the two was in Geelong. That's an hour away on the busiest freeway in the country

A friend is a process Engineer and did a big investigation into what they could do when the engines were late arriving due to traffic jams. Of course the whole thing was Just In Time because reasons, so they needed the engines to arrive just at the right time.

At the end of a multi-year study it was concluded the most cost-effective thing to do when the engines were not ready to put into cars that were ready for them, was..... let the cars fall off the end of the line, crush them, and eventually crush the engines when they showed up.

Any attempt to re-integrate them, or salvage stuff, or push them back on the start of the line, or anything else you want to think of costs more money than the above solution. They ran the numbers every which way, and as shocked as they were, it's the truth.

It was a rumor for a long time that every single time the production line was stopped, even for five seconds, it cost $1Mil. It was investigated, and found to be higher than that.

Here’s an interesting series of articles on Risks to the Public by the ACM Special Interest Group for Software Engineering: https://dl.acm.org/results.cfm?query=Risks+to+the+public&Go....

From TFA: “[Subaru] found an Ascent with the missing welds during a routine sampling inspection on 21 July”

The affected vehicles were built between 13th and 21st.

Not sure how long QA takes for a vehicle, but they found it within a week.

Yep, someone forgot to put those welding points and others didn't observe them missing.

> I'm interested why over-automation caused so many problems in manufacturing plants.

A robot will uncritically do what you tell it.

An experienced welder will tell you you're an idiot when you're an idiot.

They may be referring to the actual process of perfecting the automation. Bespoke CI/CD isn’t an instantaneous thing of beauty and efficiency gains.

I think for many operations automation can be perfect, but other operations are really challenging to automate well today.

However, I do believe if you were to design the vehicle for optimal automated assembly throughout, it could significantly change the situation.

But a design optimized for 100% automated assembly is probably not going to be as servicable. Imagine for example if wiring harnesses were built as rigid modules which were captive within the unibody when it was welded together. This could probably simplify the automated assembly, eliminating a step of snaking a flexible bundle of wires through passages, but then nobody is ever going to be replacing the wiring harness module.

I do think things can be done in the vehicle to improve the automation without compromising servicability, but it might not be economical, or maybe it just requires more control over the entire supply chain to tailor everything to the vehicle's assembly process.

It's basically because these production lines are incredibly complex and getting tiny details right is important.

When you have humans running a line, they can notice and adapt to the countless possible issues that occur (like "the alignment here is often slightly out" or "tightening X here causes problems for the people in the next station"). Their top speed is theoretically less than a automated line, but they are quite adaptable and robust.

Since automated lines are less adaptable and robust, so you have to get them just right. Any tiny little issue can cause surprising problems down the line and implementing fixes is quite slow because you can't just describe the problem and required fix to a human and leave them to work it out.

I think the best way to set up a production line is to exclusively use humans at the beginning, and carefully introduce automation once the process is understood and stable

One of the limit is reliability. Robot don’t self maintain... The more elaborate the task, the more complicated the maintenance to ensure acceptable failure rate. At some point you spend as much time fixing as doing the job. Especially in cases like here where the throughput target is craaaaazy.

My guess is that the complexity of integrating sensor information and maintaining calibration isn't refined to the tolerances needed. In theory, you only need to detect physical resistance and visual variations (not like the assembly line workers are licking the bolts). In practice, you can have a series of bolts within tolerance limits, but causing a warp an attached piece that lies outside the preceding sensors' detection ranges. A human might step back and see that the door is bending weirdly, the door bend might not be noticed for several stations, causing more disruption.

That being said, car parts in general seems so unnecessarily complex. Too many curves, too little compartmentalization and modularity, etc. At this point in history, I would expect the average person to be capable of practically disassembling most cosmetic pieces in under an hour and with little research, but it's an obnoxious process of hunting down hidden screws and plastic clips in precise sequences. If cars were built in a reasonable manner, it should have been trivial to repurpose most of the parts in these vehicles.

I was picturing some oversized Subaru-branded Rube Goldberg machine, with the final step being a shiny car being tossed into a crusher.

Ok, we've revised the title to disambiguate it.