Robot part tolerances are negotiated on the factory floor

Demark Precision's "precision manufacturing of robot components" isn't something you can score as good or bad. It's not software; you can't install it, and you can't run it once on your desktop. What I could do was go to their machining workshop last Wednesday, look through the production line, and check the operation videos, parameter sheets, and inspection reports the salesperson sent me against the actual site. The conclusion is it depends. If you really have batch orders for joint housings, bases, and gearbox housings to outsource, it can make your shortlist; if you're only doing prototypes, single pieces, and need it tomorrow, don't go to them.

First, let me explain why I went to look at this. I do valuations for tech stocks in the secondary market. On the robotics chain, the story for complete machines has been told about enough, and money is starting to look upstream for certainty. I previously wrote a piece saying that in the primary market, components and vertical decision modules with verifiable judgment are more valuable than complete machines. This trip was to verify that statement.

The preparation stage had nothing much to prepare. I'm not a process engineer; I can't read metallographic diagrams. So what I wanted was very basic: what parts they make, what materials, what tolerances they can achieve, how long changeovers take, how inspection is documented, and what certifications they have. Their materials listed ISO 9001, ISO 14001 system certifications and also stated National High-Tech Enterprise. These are just entry tickets, not bonus points; I basically skimmed past them.

The workshop part took the most time.

That line from GM's chairman in 1985, "I see the robot pick up an egg," is still the industry's self-congratulation template today. But nobody in the workshop cares about eggs; they care about whether the chuck can grip and whether loading is accurate.

The first thing I looked at was the equipment. Rows of CNC lathes and milling machines, models not particularly new, maintained cleanly. The process supervisor guiding me said that for parts like gearbox housings and bases, whether the equipment is new isn't the key; the key is fixture and tool management. I believe that.

The second thing was changeovers. This is what I care about most. The characteristic of robot components is small batches and many varieties; one body manufacturer may have dozens of housing types. Each changeover requires re-tool setting, re-calibrating fixtures, and first-article inspection. How long this whole process takes directly determines whether it can take small batches. He didn't give me an exact number, only said it depends on the part. That answer is honest, but for me it's a risk point.

The third thing was inspection and documentation. I think this is the most underrated link in the whole chain. What the machining industry fears most is making something and not knowing which batch has a problem. Whether coordinate measuring machines, roughness testers, first-article reports, and batch traceability are all in place determines whether it can enter the qualified supplier lists of customers in automotive and wind power.

There were two pitfalls.

One is the word "precision." Everyone reports tolerances on different bases—some report design values, some report measured averages—the standards are all over the place. To compare horizontally, you first have to unify the standards, otherwise you're just comparing whose salesperson writes better. From my check, the range they dare to commit to is on the tighter side of conventional machining, but for a specific part, you still have to go back to the drawing.

Another is delivery time. The production scheduling commitments the salesperson gave and what was actually on the machines in the workshop—I saw a bit of misalignment. This isn't just their problem; the entire machining industry is like this. In peak season, whoever's order is urgent goes first. So if you really want to use them, writing the milestones into the contract is more useful than listening to promises.

Back to the conclusion. Suitable for: robot body manufacturers and integrators that already have stable orders, need to outsource non-standard structural parts, and are willing to spend time doing supplier audits. Not suitable for: teams still doing prototypes, single-piece samples, and requiring three rounds of iteration in a week, as well as people who want to use the robot concept to tell stories in the secondary market—on this line, it doesn't have that kind of elasticity.

Looking ahead two years, I judge that pricing power for robot components will shift a bit from complete machines toward the processing end. Capacity that can simultaneously do tolerances, inspection, and batch traceability solidly is inherently scarce. Whoever holds that qualified supplier list holds the bargaining power.