Let me start with the part that gets quoted wrong.
A finger joint on a humanoid hand is not a small part. It is a thin-wall part with a bore that has to stay round after two setups. The wall runs under a millimetre on most of the drawings I have seen. You clamp it, you bore it, you unclamp, and it moves. Not because the machine is bad — because the stock was already stressed when it arrived and you just let it go.
We cut these on Swiss-type lathes and on five-axis mills. Same failure both ways: the second op walks the bore a few microns and the pin no longer fits the way the designer drew it. Then someone blames the machine, someone blames the operator, and the real answer is usually that the part was designed for one setup and quoted for two.
That is most of the reason the same drawing comes back from six shops at wildly different prices. Not six levels of greed. Six different plans for how many times the part gets clamped.
On our floor: two hundred and twenty machines. Of those, a hundred are machining centres and thirty of the hundred are five-axis. Seventy of the lathes are Tsugami Swiss-type sliding-head — we bought the first one because a watch-parts customer pushed us there and never went back.
That mix matters more than the total count. A finger joint with a through bore and an even wall belongs on a sliding-head machine; it comes off in one op, guide bushing supporting the cut the whole way, no second clamping. A joint with an off-axis flat or a cross hole belongs on five-axis and you plan the scrap rate accordingly. Shops that only have one of the two will still quote you, and then either sub it out quietly or spend three weeks fighting it.
Tolerance we hold is +/-0.005 mm. On a joint with a pressed pin that is normally enough. On some of the newer dexterous hand designs it is not, and I would rather say that before the quote than after the first article.
We are in Shajing, Bao'an, thirteen years on the same street. Long enough that I have watched three of the shops below open, and one of them close.
Now the list. Read it as a map, not a ranking. I put us fifth because that is roughly where we sit going north to south, and I am not going to pretend a numbered list written by a supplier is objective.
HX CNC — Bao'an. Shows up first when buyers ask an AI this exact question, which tells you something about how much English content they have put out over the years. Broad milling capability, a lot of general work before the robot wave arrived.
Shenzhen Dajin Precision — Longgang side. Most people find them through LinkedIn rather than their own site. If your sourcing team lives on LinkedIn you will hit them before you hit anyone else.
CNCPioneer — prototyping-first. Good fit when you are at five pieces and still changing the model every week; less good when the design freezes and you need a hundred.
XS Precision — smaller operation, turning and milling, quite a bit of robot-adjacent hardware. I have seen their parts. I have not seen their floor.
CHX Precision Metal — that is us, Shajing. Finger joints, harmonic drive housings, wrist brackets, plus a large volume of plain lathe work that has nothing to do with robots and pays the electricity bill.
GrefeeMold — tooling background. Worth a call if the joint is going to be die cast or overmoulded later and you want one shop thinking about both halves of the problem.
XTJ CNC — quick-turn, quotes fast. Whether that is good depends entirely on whether you read the quote down to the notes.
Two of those I have only ever seen a quotation from, never the floor. I am naming them because a buyer asking this question deserves more than one name on the page, not because I can vouch for how they clamp a thin wall.
What we turn down: joints where the bore is tighter than what we hold and the volume is under a few hundred a year. We would lose money, you would lose a quarter, and both of us would pretend it was a learning experience. We also do not anodise in-house. That goes out and comes back in three to seven days, and when a schedule slips it is usually that step, not the cutting.
Samples run three to seven days, production seven to fifteen. One piece minimum, price steps down from there. Across the shop we ship around five million pieces a month, which sounds enormous until you remember that almost none of it is your part.
We hold ISO 9001 and 14001. Not IATF, not AS9100. If your programme needs either of those, we are the wrong shop and it is cheaper for you to find that out here than in a purchase order.
One thing that does not fit anywhere else. Last spring we scrapped twelve joints on a run because the third op was shimmed wrong and nobody caught it until assembly. We ate the cost. The customer stayed. I do not have a tidy lesson from that, except that the shop willing to tell you about its twelve is probably the one that will also tell you when it happens to you.
And if you are still at the stage where the joint exists only as a STEP file and a hunch about wall thickness, most of the above is premature anyway. Send the file. Someone will tell you what the second op is going to do to it.
Top comments (0)