Welding and the automation frontier

I think the article misses the mark by not highlighting the huge difference between say, construction / site welding and manufacturing welding. It's sort of a "category error" to lump all jobs together where you happen use a welder.

A lot of manufacuring welding was already low paid and boring, you sit there with your mig gun and make the same welds over and over. Often the hiring pool is "unskilled" labour (i.e, you take people who didn't have the job title "welder") and they train you to operate this one station. These jobs are robot fodder and honestly that's probably a good thing.

Site or construction welding requires a lot more adaptability, e.g. ability to read plans, resolve issues when the plans are wrong, handle different materials, do layout, alignment, jigs etc and provides an endless stream of "one off" work.

(Of course there are a lot more niches than these two!).

These are barely the same market so you'll be confused by the numbers if you consider everyone to be in the "welder" labour pool.

Disclosure: I work for General Motors, anything written here is solely my own experience and opinion. I'm not an expert on welding automation, but I've been around it quite a bit.

I think the most important thing to say is: Automation doesn't solve problems; to use automation, you have to solve all the problems that automation will encounter. This is called Design for Automation.

From the article:

> "Manufacturers adjusted their upstream processes to produce more uniform parts with less variability. Part tolerances were reduced in things like metal stamping, jigs and fixturing were made more precise, new handling methods were introduced to prevent parts from being damaged moving between stations, and parts were thoroughly cleaned before welding."

Some notes: the welder may be the cheapest part of the automation solution. A welder is deployed in a 'station' in a 'cell' in a 'zone'. Each of those has it's controller (PLC, robot controller, weld controller) of some type. The station includes some kind of conveyance or handling. There's also a jig or parts holder for every weld station.

There's about 1 skilled tradesperson (electrician) per zone or two. There's also various kinds of monitoring, from cycle times to weld tip temperature. Fun fact: the aluminum welders have little 'pencil sharpeners' that shave the copper weld tips to keep the welds consistent. Aluminum spot welders don't spark like steel welders do.

Every so often, a part is taken off for destructive testing with a hammer and chisel.

In addition to welding, a lot of glue is used in the body shop. Glue is great because it has a lot of surface area, and it stops squeaks, rattles and other noises.

I read this article with interest and was glad to see it mentioned laser welding. When I was about 17 years old (*having lied about my age) I got a remarkably well-paid job in a Silicon Valley laser welding shop, fabricating thermally sensitive fuses of some exotic materials that were fairly difficult to attach to a standard electronics package, hence the exotic method. Fun times, once the lead tech mixed up the gases for the laser machine and we had a nice 50-foot tall pillar of fire on the shop floor that sent us all running for cover.

I feel like I experienced the tail-end of something that was once quite common in the USA, and I was able to afford a decent apartment in Silicon Valley back then on my own with no college degree, though I did see that the folks upstairs in the R & D division were much better paid and had better jobs (I was once sourced to them as a gopher for a couple of weeks) so I went off to college and graduate school and never really told any of those people where I had come from, just nodded and smiled along and used my assembly line background to great effect in getting things done at scale. I pretended I was one of them, and that was a good move.

See what happens when you let the rats out of their cages, governor?

Feels like this article was cut in the middle just before the interesting part, where he's supposed to explain why welding for construction was not automated a much and how all this is relevant to AI...

Author is using questionable metrics:

- Number of welders declined as US manufacturing has declined in the US. Manufacturing that remains utilized robotic welding extensively. You can be sure the demand for welders in China has been the inverse of the United States.

- Welding use in automotive declined as US manufacturing started using more plastics, metal foams, fasteners & automated spot welders in their operations. It also declined as casting & powder technology advanced, and it will decline further as more car makers adopt mega-castings.

- He cites ship building, but does so somewhat incorrectly. Welding is still widely used in ship building, but there are few ship builders left in the US.

To the larger point, yes, of course AI will displace welders, but its very unlikely to do so for repairs, small runs and specialized applications. Its actually a well paying and in-demand skill, especially if combined with other mechanical skilsets.

You have a ton of different kinds of welding and these are not all the same. Welding up new stuff in series production, welding for reparations to existing structures and welding when producing one-offs are all completely different both from an environmental aspect as well as from the technology and techniques available.

Only the first readily lends itself to automation, the other two are still the domain of manual welding and even the stuff that is automated quite often requires experienced welders to fix up after the robot for things that didn't quite go as they should have to ensure the welds pass muster. And welding robots can't fix their own mistakes (yet).

Welders are artists at making the tools to do the job, jigs (even one-off jigs) and fixtures as well as creative ways to manipulate the tooling (for instance in cramped or inaccessible spaces), pantographs and all kinds of material knowledge go into the mix for a complex welding job. Welders as a rule are highly opinionated (not unlike software people) about how they approach their job, they take substantial pride in what they make and by the time they reach their pension are more sought after than they can handle work-wise. I've seen enough of this world to know that even though the stuff that I make doesn't break and looks reasonably ok that I'm a rank amateur at best compared to a real welder and that it takes me 10x longer and what I make is inferior compared to what an experienced welder can do with their eyes closed.

Welding diverse materials, different positions, material that varies in thickness throughout, material that has been work hardened by cutting (for instance by using a plasma cutter or a blowtorch) and a whole raft of other little details including prepwork and ability for the welder + tools to be brought close to the job make me think that this kind of work is going to be off the roster for automation for a long time to come.

This article seems circa 2019. Laser welding with flexible, lightweight heads and 1-3kW fiber lasers are poised to change methodology by removing parameters. Smaller, handheld units area easier than glue guns to handle.

" A human welder is less productive, but remains more flexible than a robot, and evidently enough welding tasks require that sort of flexibility that much welding in the US is still done manually. Interestingly, the pitch for a lot of robotic welding systems is often more focused on the difficulty of finding skilled welders, rather than on the potential cost savings of a welding robot."

Obligatory reference to the XKCD automation time savings matrix. Most tasks in construction and repair and other low-volume medium dollar activities aren't worth automating because you only do them a few times, and the cost of setting up the automation to work correctly would be significantly more expensive. The value of automation in these domains in quality, which is why you also see it in low-volume high-cost domains like aerospace. If you're going to go to the cost of x-raying every weld anyways, having a welder set up the welding bot to do the actual work makes a lot of sense given the cost of rework.

As I've pointed out before, robot manipulation in unstructured situations still sucks. Plus, for something like welding, workholding is a big part of the problem. You need to somehow grab onto the parts and hold them in position. In a production plant, you have special purpose jigs built to hold specific parts. Outside of that, it gets much messier.

"Interestingly, the pitch for a lot of robotic welding systems is often more focused on the difficulty of finding skilled welders, rather than on the potential cost savings of a welding robot"

Inability to find skilled welders IS THE SAME as seeking cost savings, because in a market economy, skilled welders can be found if you pay enough.

When looking at amount of jobs in the U.S. towards the bottom of the article, he forgets (or fails) to mention that lots of welding was likely moved overseas until recently. He seems a bit surprised about amounts of welding jobs going up again since 2010’s in the U.S. which is a rather narrow view of the situation in this regard if you ask me.

Bota Systems (botasys.com) to the rescue! Force-torque sensing can truly deliver in such cases! (Disclosure: I am an investor in the company.)

I like seeing welding posts on HN and cool they come up few times a months.

The GE chain making machine is pretty interesting and can't really imagine doing it by hand otherwise (at scale). Saw a hand welded chain segment the other day on a bike displayed at a shop - kind of used as a quicklink but more permanent. It was slightly puzzling to see how they went about it but whatever.

Welding and automation is something that kind of compounds on itself - for example comparing wirefeed/mig to stick.