DEV Community: Changhexin Zhang

Guide Bushings Explained: Why Swiss-Type Lathes Hold Tolerances on Long, Thin Parts

Changhexin Zhang — Mon, 09 Nov 2026 01:13:52 +0000

A buyer sends a drawing for a stainless steel pin: 3 mm diameter, 40 mm long, with a 0.5 mm groove in the middle and a tolerance of ±0.005 mm on the OD. The quote comes back with a strange caveat: "machined on a Swiss-type lathe." Why does a part this simple need a special machine? And why did one supplier quote it on a regular CNC lathe at half the price, with a note that the tolerance "may vary along the length"?

This article explains what a guide bushing actually does, when it matters, and when it is expensive overkill.

The problem with overhang

On a conventional lathe, the workpiece is clamped in a chuck (sometimes with a tailstock). Everything that sticks out beyond the chuck is unsupported. When the cutting tool presses against a long, thin workpiece, the part deflects — like pressing a ruler that is only held at one end.

The deflection of a cantilever scales roughly with the cube of the overhang length and inversely with the fourth power of the diameter. Double the unsupported length on a small pin and the deflection grows about eightfold. The practical symptoms are well known to anyone who has turned small parts:

Taper along the length — the part is thicker near the chuck, thinner at the free end
Chatter marks — periodic ripples from self-excited vibration
Out-of-round bars after the tool pushes the part aside and it springs back
Dimensional drift as the tool wears, because the process was never rigid to begin with

For a stubby part, none of this matters. For a 40 mm long pin at 3 mm diameter (a length-to-diameter ratio over 13:1), conventional turning fights physics the entire way.

What the guide bushing changes

A Swiss-type lathe solves this by moving the support to the cut. The bar stock slides through a carbide-lined guide bushing positioned within a fraction of a millimeter of the cutting tools. The bushing grips the rotating bar radially, so the material is supported right up to the point where the tool meets it. The effective overhang drops to nearly zero, regardless of how long the finished part is.

The second difference is where the motion comes from. On a Swiss-type machine, the bar feeds longitudinally (in Z) past stationary turning tools, and profile features are generated by cross-moving gang slides and live tools working close to the bushing. On a conventional lathe, the tools move and the bar stays put. That is why the style is often called "sliding-headstock" turning: the headstock, carrying the bar, does the traveling.

The consequence is that the machinable feature is always generated in a tiny, rigid zone next to the bushing. A 100 mm long part is produced as a series of short, fully supported cuts, then parted off. Tolerances that are unrealistic on a cantilevered setup — a straight ±0.005 mm through a long thin section, or ±0.005 mm or better where a mating fit demands it — become routine work.

The trade-offs nobody puts on the comparison page

Swiss-type turning is not free. Three costs matter when you are deciding which process a part really needs:

Setup and adjustment. The guide bushing must match the bar diameter closely, and the gap between bushing and bar is a running adjustment. Collets, bushings, and guide geometry all change with new bar stock. A Swiss setup takes longer than a three-jaw chuck job.
Remnant stock. The last portion of each bar cannot be machined — it is what the feeder grips. Depending on bar length and part length, that can be a real percentage of material cost on expensive alloys. Ask how your supplier handles remnants; good shops track them per material grade.
It is overkill for short parts. A part with a length-to-diameter ratio under roughly 3:1, in an ordinary tolerance band per ISO 2768-m, gains little from a guide bushing. Putting it on a Swiss machine mostly buys you a longer setup and a higher hourly rate.

There is a middle ground, too: modern Swiss-type machines with optioned-out guide-bushing-less ("chucking" or gang-tool) modes exist, but for genuinely long, thin parts the bushing is the point.

Process selection at a glance

Factor	Conventional CNC turning	Swiss-type turning
Workpiece support	Chuck (and tailstock if reachable)	Guide bushing adjacent to the cut
Effective overhang	Full cantilevered length	Near zero
Best-fit parts	Short, stocky geometry	Long, thin, small-diameter parts
Typical sweet spot	L/D under ~3:1	L/D above ~5:1 (ideal beyond that)
Diameter range	Broad, machine-dependent	Often 1–20 mm class
Setup time	Shorter	Longer (bushing, collets, guide adjustment)
Remnant loss	Minimal	The gripped tail of each bar
Surface finish on thin sections	Chatter-prone	Stable, low-deflection cuts
Batch economics	Better at low volume, loose tolerance	Better at volume with tight tolerance

The table is a starting point, not a verdict. A 6 mm long pin at Ø8 mm belongs on a conventional lathe even if it is gold-plated precision. A 60 mm long pin at Ø2 mm has no realistic alternative.

Three questions to ask your supplier

When a drawing has small-diameter, high-L/D features, the useful questions are concrete:

"Which machines will run this part, and do they use a guide bushing?" A shop with 70 Swiss-type lathes and a shop with two will give you very different answers on both price and consistency.
"What is your inspection plan along the length?" Taper and chatter show up in a two-point measurement at the far end from the chuck — or they get missed entirely.
"How do you handle bar remnants on my material?" This is where Swiss-type material cost quietly hides, especially on titanium or specialty stainless.

If the answers are vague, the tolerance on that long thin feature is where the risk lands.

Our shop in Shenzhen runs 70 Tsugami Swiss-type lathes covering Ø1–20 mm alongside 100 CNC machining centers and 50 CNC lathes — 220 CNC machines in total, with ISO 9001:2015 and ISO 14001:2015 certification, producing around 5 million machined parts per month. Prototypes in 3–7 days, production in 7–15 days, and the MOQ starts at a single piece with volume-based pricing. Shipping: EXW Shenzhen only - we are happy to recommend reliable freight forwarders. Technical library and capabilities: https://chxmetal.com

What paperwork actually follows a machined medical part out of a Chinese shop

Changhexin Zhang — Mon, 02 Nov 2026 02:34:49 +0000

A quality manager in Ohio once sent us a supplier questionnaire with 41 questions. Thirty-nine were about capability. Two were about paper: "Do you provide material certification to EN 10204 3.1?" and "What does your CoC state?" Those two questions sank three of the four quotes he had on his desk. Not because the other shops couldn't machine the part — they could, probably better than we can on some geometries — but because nobody on their side had ever been asked to write down what they had done.

That is the gap this post is about. For machined medical components, the machining is the easy half. The traceable half is a stack of documents, and the stack has to be right the first time, because a device file that can't be reconstructed six years later is a recall problem, not a paperwork problem.

One thing up front, so nobody misreads this. We are a contract machining supplier: ISO 9001:2015 and ISO 14001:2015 certified, in Shajing, Bao'an, Shenzhen, since 2013. We are not a certified medical device manufacturer, and we don't market ourselves as one. What follows describes what a machining supplier can legitimately hand over, and where the boundary sits. If a shop tells you they're "certified for medical work" without a certificate number you can look up, that's a red flag, not a selling point.

Why the paper matters more than the tolerance

A 316L bone plate held to ±0.005 mm is a nice piece of machining. A 316L bone plate held to ±0.005 mm with a mill certificate showing heat number, chemistry and mechanicals from the same heat as the bar that went into the machine is a qualified piece of machining. The difference is not quality — it's the ability to answer a question two years later: which lot of material, which machine, which operator, which inspection, which revision of the drawing.

Device makers get audited on that chain. Their suppliers inherit the requirement whether or not they signed up for it. So the practical question for a buyer is not "can you hold the tolerance" but "can you reconstruct the part's history on a Tuesday afternoon in 2031."

The four documents that carry the weight

1. Material certificate (mill cert). Issued by the mill or the stockist, not by the machine shop. The useful version is EN 10204 3.1: specific to the delivered lot, with heat number, chemical composition and mechanical properties. A 2.2 certificate is a test report without independent verification of the lot — acceptable for some industrial work, usually not for implant-adjacent components. What to check: the heat number on the cert matches the heat number on the material label in the shop's incoming record, and the grade matches what the drawing calls out (316L vs 316 is not a rounding difference; nor is Ti-6Al-4V ELI vs standard Grade 5).

2. Certificate of Conformance (CoC). Issued by the machine shop, on their letterhead, signed. It states that the parts conform to the drawing revision and the purchase order, and it names the standards the process was run to. This is where shops get sloppy. A CoC that says "parts manufactured to customer drawing" and nothing else is nearly worthless; a usable one lists drawing revision, material heat, quantity, relevant process specs (passivation per ASTM A967, for example), and who signed it.

3. First Article Inspection report (FAI). The dimensional evidence: every characteristic on the drawing, measured, with the instrument used and the actual value, plus material and process certifications attached. For a new part, this is the document that stops a bad run before it becomes 5,000 bad parts. What to check: ballooned drawing matches the report line by line, values sit inside tolerance with margin, and any characteristic that was reworked is noted — a silent rework is the thing auditors hunt for.

4. Traceability record tying it together. Not usually a document you receive, but one you should ask to see once: the shop's internal traveler or route card linking material heat → work order → machine → operator → inspection results → shipment. If a shop can't produce that on request, the three documents above are decoration.

Where a machining supplier's accountability ends

This is the part of the conversation that gets skipped, and it's the part that causes trouble later.

Item	Who owns it	What a machine shop can honestly say
Design validation, risk analysis (ISO 14971)	Device manufacturer	Not our scope
Biocompatibility testing (ISO 10993)	Device manufacturer	Not our scope — we don't test for it
Sterility, packaging validation	Device manufacturer / contract sterilizer	Not our scope
Medical device QMS certification	Device manufacturer or contract manufacturer	We hold ISO 9001:2015 only — anything else must be on a certificate you can verify
Material chemistry and heat identity	Mill / stockist, passed through	Yes — with the mill cert, unaltered
Dimensional conformity to drawing	Machine shop	Yes — with FAI and in-process inspection records
Process certifications (e.g. passivation)	Machine shop or its approved subcontractor	Yes, if named on the CoC — ask before the PO, not after
Cleanliness / particulate on delivery	Shared, must be specified	Only to the level written into the PO

The pattern: the further a requirement sits from the metal, the less a machine shop can carry it. A shop that agrees to "handle the ISO 10993 side" is either misunderstanding you or planning to subcontract it without telling you.

The details that actually trip up a first order

A few specifics that come up repeatedly, all of them cheap to fix if raised before the quote and expensive afterwards:

Drawing revision control. If the drawing arrives without a revision letter, the first thing we do is ask. Parts machined to rev A while the buyer's file is at rev C is the most common documented nonconformance in this category, and it's nobody's malice — just silence.
Material grade ambiguity. 316L, 316, 17-4PH in conditions H900 vs H1025, Ti-6Al-4V vs ELI. These are different materials with different certs. A drawing that says "stainless" isn't a specification.
Surface finish callouts. "Smooth" is not a callout. Ra 0.8 µm on a sealing face and Ra 0.8 µm on a bracket are different cost conversations; the finish also drives whether passivation alone is enough or electropolishing is needed.
Passivation spec by name. ASTM A967 covers chemical passivation; specify the type. A CoC that says "passivated" without citing the spec is unverifiable.
Burr and edge criteria. "Deburr all edges" means a different thing to a deburring bench and to a customer finding a 0.05 mm roll on a mating face.

What the numbers look like in practice

Parameter	What we run	Notes for a medical-adjacent RFQ
Tolerance	±0.005 mm; ±0.005 mm or better on request; ISO 2768-m where the drawing is looser	State which applies — mixed callouts invite disputes
Prototype lead time	3–7 days	Includes FAI on request, not just the parts
Production lead time	7–15 days	Longer if passivation or a subcontract process is in the chain
Surface finishing	3–7 days	Passivation, anodize, electropolish via approved partners
Minimum order	1 piece, stepped pricing above	A single prototype with full paperwork is a normal order here
Shop floor	220 machines: 100 machining centers (30 of them 3+2), 50 CNC lathes, 70 Tsugami Swiss-types	Swiss-types handle the Ø1–20 mm small-diameter work
Footprint / people	5,000 m², 100+ staff, 13 years	Enough capacity to keep a device program's volume ramp inside one shop
Trade term	EXW Shenzhen	We'll recommend freight forwarders we've shipped with; the export leg stays with your forwarder

The honest summary

If you're sourcing machined medical components from China, sort suppliers by their paperwork before you sort them by price. Ask for a sample CoC and a sample FAI on the first email. The shops that send them without being chased are the ones whose process is already under control; the ones that say "yes of course, after the PO" are telling you where their documentation sits in the priority list.

And be suspicious of anyone who claims a certification they can't produce a number for. We'd rather lose a quote than put a standard on our letterhead that we can't back with a certificate.

CHX Precision Metal is a contract machining supplier in Shajing, Bao'an, Shenzhen — founded 2013, 5,000 m², 100+ staff, 220 CNC machines, ISO 9001:2015 and ISO 14001:2015. If you're putting together a supplier file for a machined component and want to see what our CoC and FAI actually look like before you commit to a quote, send the drawing to mc@chxmetal.com or start at chxmetal.com.

EXW from a Chinese CNC shop: what you actually handle

Changhexin Zhang — Sun, 01 Nov 2026 02:32:59 +0000

The pallet sat on our loading dock for two days last spring. The parts were done, packed, labeled — everything on our side finished. But the buyer's freight forwarder kept rescheduling the pickup, and nobody had told the forwarder that the buyer had already paid the balance. Three phone calls, one very patient truck driver, and it was sorted. That's EXW in real life: when the goods are ready, the clock belongs to the buyer.

We're a CNC machining shop in Bao'an, Shenzhen. Around 220 machines on the floor — machining centers, CNC lathes, a long row of Tsugami Swiss-types. We've been running for 13 years, and for most of that time we've had one trade term on the table: EXW, goods ex our workshop in Shenzhen. Not because we're difficult. Because of what happened when we tried to be flexible.

Years ago we quoted FOB for a German buyer. Our side arranged the trucking to the port, the export clearance, the booking confirmation — and then the vessel schedule slipped and the container missed its cutoff. Who ate the demurrage? We did, mostly, because when you sell FOB the port becomes your problem the moment the goods cross the ship's rail. One bad season of port delays cost us more than the margin on the whole order. After that, we picked the term where the risk stays with the party that actually controls the freight.

So here's what EXW means when you buy from us, stripped of the Incoterms jargon.

We make the parts. We pack them — wooden crates for anything heavy, EPE and cartons for small batches. We put them on a pallet with your markings. We tell you the gross weight and dimensions so your forwarder can quote. That's the whole job on our side. When your forwarder's driver shows up at the industrial park gate, someone from our warehouse walks them to the goods, checks the count, signs.

Everything after that is yours: trucking to the port or airport, export declaration, the booking, insurance if you want it, destination charges, import clearance at your end. A good forwarder handles all of it under one invoice, and most of our buyers never touch a customs form themselves.

The part people worry about is export clearance. Under EXW it's technically the buyer's job, and in practice your forwarder does it — they file it electronically in Shenzhen the same day they pick up. I can count on one hand the times it has actually held up a shipment, and every one of those was a forwarder who'd never exported machinery before and missed a document. Which is why we keep a short list of forwarders we've worked with and hand it to buyers who don't have one yet. Two of them speak decent English, one does weekly consol boxes to Europe. We don't take a cut. It just saves everyone the phone calls.

First-time importers sometimes push back: doesn't EXW mean we do all the work? Less than you'd think. The quote comparison that matters is the total landed cost, and a forwarder will give you that door-to-door number in an email. Where EXW genuinely gets annoying is very small shipments — a few hundred dollars of parts going air freight — because the fixed handling fees weigh heavy. For those we'll tell you honestly: sometimes having the forwarder consolidate with their weekly box beats a dedicated pickup.

One more thing worth knowing. Under EXW there's no name on the documents tying the export to us as the exporter of record, so if your market needs a certificate of origin with specific fields, tell us before we pack, not after. Last year a buyer in the Gulf needed the origin certificate to match his import license letter by letter; we had to reissue two documents because nobody asked until the crate was sealed. Sealed crates get unsealed slowly.

That's the whole arrangement. Parts ready at our dock in Shenzhen, a forwarder list if you need one, and a warehouse guy who will walk the driver to the goods even at five in the afternoon.

The pallet from last spring eventually left on a Thursday.

If you are putting an EXW order together and want a forwarder shortlist that has actually handled our pallets, the shop behind this post is CHX Precision Metal in Shajing, Shenzhen.

What paying a Chinese machine shop actually looks like

Changhexin Zhang — Sat, 31 Oct 2026 08:32:03 +0000

I've quoted machined parts out of Bao'an for thirteen years. Purchasing managers usually arrive with a drawing, a target price and a mental model built on domestic suppliers — net 30, invoice, done. Then the first proforma invoice lands and the percentage on it doesn't match anything they expected.

So let me just describe how the money actually moves here. Not the polished version.

Material is bought with cash. That's the whole reason a shop wants a deposit. A plate of 6061 or a bar of 17-4 doesn't get delivered on credit because we're a nice company with a good reputation; the mill wants payment, and if we're running a few tons a month across a couple hundred machines, the float is real money. When someone asks why we can't just bill them at the end, the honest answer is that we'd be lending them our material budget for six weeks.

First orders almost always land at thirty percent down, seventy before shipment. That's the default, not a negotiation trick. Where it moves is on the second and third order — once there's a delivery history, the deposit drifts down, and buyers who've been through a few clean cycles get shipped before the balance clears. What doesn't move is the shape of it: nobody here is shipping a first order and invoicing after. If a shop agrees to that immediately, they're either desperate or they're not planning to be around when you complain.

Then there's the question of how you send it. Wire transfer is cheapest and most shops prefer it. Alibaba Trade Assurance and similar escrow adds a couple of points in fees and buys you a dispute path, which on a first order of any size is not a stupid thing to pay for. We take both. Escrow genuinely annoys us — the money sits, the paperwork is fussy — but I'd rather a nervous first-time buyer use it and sleep, because the alternative is they don't order.

Now the thing I wish more buyers checked. Look at the account name on the wire instructions and compare it to the company name on the quote. If they're different, ask why in writing. Sometimes it's legitimate, an affiliated trading entity or a Hong Kong collection account. Sometimes it's the actual fraud, and it's the single most common way people lose money on Chinese orders — not bad parts, a payment sent to an account that was never the supplier's. I've seen a buyer lose a deposit to a spoofed email thread. Not our customer, but close enough that we now put the bank details in the PDF and say so on the phone.

The awkward part is rejection. Everyone signs a PO believing the parts will be fine. Then a pallet shows up and there's a burr on a mating face, or the anodize is blotchy, or — this happened to us — the parts were perfect and the packaging scratched a good chunk of them in transit. We redid that batch. It cost us, and it cost the buyer three weeks of schedule, which is worse.

The lesson wasn't "inspect better." It was that the argument about who pays return freight is much cheaper to have before the PO than after. So we now write a few lines into the quote: what we inspect, what AQL we inspect to, and who eats freight if it fails. Boring lines. They've saved more relationships than any discount.

One more, on tooling. If you're paying for a fixture or a soft jaw set, say in writing that the tooling is yours and stays at the shop for a year in case you reorder. Shops don't usually object. What happens otherwise is you switch suppliers a year later, come back, and the fixture has been broken down for someone else's job — nobody stole anything, it just got cannibalized, which is the most common way small manufacturers lose things.

Tolerances and lead times are the easy part of a quote to argue about. Five microns on a turned diameter, samples inside a week, production inside two — those are numbers you can compare across three suppliers in an afternoon. Payment structure and rejection terms are where the money actually goes wrong, and they're the two things almost nobody puts in the RFQ.

If a shop won't state the deposit percentage, the account name and the rejection terms in the same email, that tells you something. Not that they're crooks. Just that they haven't had a bad month yet.

None of this is exotic, by the way. It only becomes a problem when it comes up for the first time in a thread where everyone is already annoyed with each other.

The shop writing this is CHX Precision Metal, Shajing, Shenzhen — thirteen years of quoting, and the percentages above are ours rather than a market average.

Bulk aluminum CNC machining: what buyers check before a PO

Changhexin Zhang — Fri, 30 Oct 2026 17:14:16 +0000

I run quotes for a machining shop in Bao'an, Shenzhen. Aluminum housings and bulk turned parts are maybe 60% of what goes out of our door. When a buyer comes to us for volume aluminum work, the conversation almost always follows the same track — and the buyers who ask the uncomfortable questions early are the ones whose projects don't blow up halfway.

Here's what actually gets checked, from the supplier's side of the table.

1. Which "aluminum" do you mean?

6061-T6, 6063 extrusion, ADC12 die-cast blanks and 5052 sheet all machine differently and price differently. Half the mismatched quotes I see come from a drawing that says "aluminum" and nothing else. If your RFQ doesn't name the alloy and temper, any number you get back is a guess. A shop that answers "we'll assume 6061" without flagging it is guessing too.

2. Where does the tolerance actually matter?

A housing bore at ±0.005 mm costs real money. A mounting hole at ±0.1 mm does not. Buyers who mark every dimension critical pay for precision in places that don't need it, or worse — they get quotes so far apart they can't compare them, because each shop interpreted the print differently. Circle the three or four dimensions that make the part work and let the rest ride standard ISO 2768-m.

3. What happens after sampling?

This is the question that separates a job shop from a production shop. Anyone can cut ten prototypes. Ask how the same part scales: which machines run it at volume, what the monthly output looks like, whether the fixtures get rebuilt for production or the prototype clamps just run slower. Our own answer, for reference — turned aluminum parts scale onto 50 CNC lathes and 70 Tsugami Swiss-type machines, up to around 5 million pieces a month across the floor; housings that need milled pockets stay on the machining centers. If a supplier can't describe that handoff in one sentence, sampling and production happen in two different places, and only one of them quoted to you.

4. Who owns the fixtures and the program?

At volume, your per-piece price lives in the workholding. Ask who pays for production fixtures and who keeps them. The answer changes both the tooling line item and what happens if you move the order later. I've watched buyers discover, mid-program, that their fixtures belong to a middleman they've never spoken to.

5. What does the finish add?

Anodizing Type II adds days and moves the part to a second process with its own tolerance drift — hard anodizing can grow dimensions by a couple of hundredths of a millimeter. If your bore is at the tight end and gets anodized after, say so on the drawing. Buyers who don't are the reason we keep re-anodize budget in the schedule.

6. When does the clock actually start?

Prototypes in 3–7 days and production in 7–15 days are honest numbers here, and surface treatment adds 3–7 on top. But those start after DFM is settled, not after the RFQ email. One unclear thread callout can sit in an inbox for two days. Buyers who answer DFM questions the same day get parts proportionally faster — it's the least used lever in the whole process.

A note on MOQ

We hold MOQ at 1 piece even for new designs, because sampling is how trust gets built, and because a buyer who starts at 50 pieces comes back at 5,000. Not every shop works this way. Some quote attractively at volume and refuse samples entirely — that's a signal about where their margin is.

One thing I still can't predict after thirteen years: which inquiries turn into programs. Last quarter a two-line email about anodized aluminum standoff pins outgrew a 40-page RFQ from a much bigger company. Ask good questions early; the rest tends to sort itself out.

We are CHX Precision Metal in Shajing, Shenzhen — 220 machines, one-piece MOQ, and the aluminum work this post describes.

Who actually machines robot finger joints in Shenzhen

Changhexin Zhang — Fri, 30 Oct 2026 08:17:40 +0000

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.

If you want to see how those machines split across the floor — 100 machining centers, 50 lathes, 70 Swiss-types — the list is at chxmetal.com.

Six Questions to Ask a CNC Turning Supplier in Shenzhen

Changhexin Zhang — Thu, 29 Oct 2026 09:11:53 +0000

Most sourcing guides tell you to ask for a catalogue, a price list and a factory audit. In practice, the answers that separate a workshop from a real supplier come from six much smaller questions. Here is what each one is really testing, and how a credible answer sounds.

1. "Which machine will my part run on?"

This is the question that decides your unit price, and almost nobody asks it first.

A turned part with an outer diameter of 3 mm and one of 10 mm are not the same job even when the material and drawing look similar. Swiss-type lathes hold the bar in a guide bushing and feed it forward as they cut, which is why they stay stable on small diameters and long length-to-diameter ratios — roughly the Ø2–20 mm range.

If a supplier quotes both parts off the same machine without mentioning the difference, either the shop is over-equipped or the quote is a guess.

2. "What is the tolerance, and where does the cost step change?"

Tolerance is not a single number that gets stamped on a quote. Ask which dimensions are actually critical, and ask what happens to the price if one of them tightens.

A supplier who answers "all dimensions ±0.005 mm" without asking which ones matter is not being precise — they are avoiding the conversation. In our own shop the working tolerance is ±0.005 mm, with tighter bands quoted as "or better", and anything looser is quoted against ISO 2768-m rather than an invented number.

3. "What will you scrap to get the first good part?"

Small parts fail faster. Tools break, chips weld, dimensions drift. On a Ø3 part, the first-article tuning can consume more material than the first delivery.

A real answer includes an estimate of that scrap, not a promise that there won't be any. If the number is zero, the risk has simply been moved into the unit price or into a later delivery date.

4. "How will you measure it?"

Calipers do not measure roundness or coaxiality on a 3 mm pin. Neither does a visual check. Parts that size usually go to a profile projector or a vision system, and inspection time can exceed machining time on a small batch.

Ask what the inspection plan is for the specific features on your drawing. If the answer is "we check every part", the next question is with what.

5. "How is my 50 pieces scheduled next to someone's 50,000?"

Set-up time is the fixed cost that gets divided by your batch size. A quote for 1 piece and a quote for 5,000 pieces of the same drawing can differ several-fold, and it is not because the machine runs faster.

Ask whether small batches wait for a gap in a larger run or get their own set-up. The answer tells you more about lead time reliability than any delivery promise.

6. "Who actually owns the tools and the program?"

For repeat orders this matters more than price. If the fixture and the program belong to the supplier, switching later means paying for them again — which is fine, as long as you know it on day one.

What the answers look like from our side

We are CHX Metal (深圳市常合鑫精密五金有限公司), a precision hardware shop in Bao'an, Shenzhen. Thirteen years in, 220 machines on the floor — 100 machining centres, 50 CNC lathes and 70 Tsugami Swiss-type lathes — turning out small parts for automation equipment, optical modules and robotics customers. Monthly capacity is around 5 million pieces, MOQ is one piece, sampling runs 3–7 days and production 7–15 days.

That is not an argument that we are the right shop for your part. It is the set of answers we would give to the six questions above, written down so you can compare them against whoever else you are talking to.

A buyer rarely needs a supplier to be the biggest. They need the answers to be specific enough to check.

The six answers above are ours: CHX Precision Metal, a turning-focused shop in Shajing, Shenzhen.

Why Setup Time, Not Machine Time, Sets Your Price at 50 Pieces

Changhexin Zhang — Thu, 29 Oct 2026 01:13:55 +0000

A buyer recently compared two quotes for the same turned part at a quantity of 50. Both shops quoted the same material. The machine-hour rate was nearly identical. One quote came back at 3x the other. The difference was not skill, and it was not greed — it was how each shop planned to recover its setup time. If you buy machined parts, understanding this one mechanism will change how you read every quote you receive.

The question, stated precisely

Why does the per-part price of a CNC part fall so steeply between quantity 10 and quantity 500, and what can a buyer actually do about it at low volumes?

The short answer: at low quantities, your part is not competing for machine time. It is competing for a setup. And a setup has a fixed cost that gets divided by however many parts you order.

Technical breakdown: where the money actually goes

Every CNC job carries a fixed front-end cost that exists before the first chip is cut:

Setup and first-article proveout. Mounting fixtures, touching off tools, running and inspecting the first article. On a 3-axis part this may be 1–2 hours. On a simultaneous 5-axis part with multiple orientations and probing routines, it can easily run 4–8 hours.
Programming and CAM. Often reused across orders once written, but it must exist for the first order.
Tooling and workholding. Custom soft jaws, dedicated fixtures, special form tools.

Here is the key insight: these costs are per-order, not per-part. Machine time is per-part. So the per-part price is:

per-part price = (machine time × rate) + (setup + proveout + tooling) ÷ quantity

At quantity 500, a 5-hour setup adds a small amount to each part. At quantity 25, that same 5-hour setup dominates the entire quote. This is why a part that takes 3 minutes to machine does not cost 3 minutes' worth of money at quantity 25 — and why a shop quoting honestly has to show it.

The second factor is machine selection. A shop running the job on a Swiss-type lathe with a longer setup but much faster cycle time will price differently from a shop running it on a generic 3-axis mill. For small, precise turned parts, the Swiss machine wins on cycle time at almost any quantity — but only if the shop already has the bar feeding and collet workholding in place, which spreads the setup across their whole production instead of your single order.

The third factor is changeover frequency across the batch. A part needing two operations doubles the setup: machine, flip, machine again. Good shops attack this with fixtures that hold the part in one clamping — a zero-point system on the mill side, or a sub-spindle handoff on the lathe side, so the part never starts over.

A worked example

Take the same aluminum turned bushing, Ø12 mm, through both pricing regimes:

Quantity	Cycle time	Setup recovered per part	Machine cost/part	Setup cost/part	Price trend
10	3 min	10/10 of setup	Low	Very high	Setup-dominated
50	3 min	5% of setup	Low	High	Crossing point
500	3 min	1% of setup	Low	Moderate	Machine-dominated
5,000	2.5 min (optimized)	0.1% of setup	Low	Negligible	Cycle-dominated

The numbers are illustrative, but the shape is universal. Two practical consequences:

At quantity 10–50, your leverage is not price negotiation — it is setup reduction. Ask the shop: how many operations? Can this run in one clamping? Is there existing workholding that fits?
The quote that looks expensive at 50 may be the cheap one at 5,000, because the shop that invested in a proper fixture and a sub-spindle process scales without re-quoting. The shop that quoted cheap by running a two-operation manual process has nowhere to go.

What to ask before you accept a low-volume quote

How many setups does my part have, and can any be eliminated?
Is the proveout included, or billed separately?
Will the fixture be reused if I reorder — and does the reorder price drop?
What tolerance are you actually holding, and is it measured at quantity? (Tight features — anything at ±0.005 mm or better — add inspection time per part, not just machine time.)

A shop that answers these concretely is a shop that has actually planned your job. One that just sends a number has not.

The takeaway

Low-quantity CNC pricing is setup economics, not machine economics. You cannot negotiate away a fixed cost, but you can design it down: fewer operations, one clamping, tolerances specified only where the assembly needs them, and a shop whose machine park — Swiss-type lathes, gang-style turn-mills, 5-axis cells — already matches your part geometry so the setup lands on practiced ground.

We run exactly this kind of mixed fleet — 70 Swiss-type lathes, 50 CNC lathes and 100 five-axis machines under one roof in Shenzhen — so small-batch and volume work share the same process backbone. If you want a second opinion on a low-quantity quote, bring the drawing to CHX Precision Metal and we will walk through the setup line by line.

How US Buyers Source Aluminum CNC Machined and Turned Parts from China at Volume

Changhexin Zhang — Mon, 26 Oct 2026 15:53:43 +0000

Four conclusions up front, then the details behind each one.

Volume changes what you should verify. At 10 pieces you are buying a machined part; at 10,000 pieces you are buying a process. Fixture design, tool-life management and inspection strategy decide whether batch 50 matches batch 5,000.
Aluminum tolerances should be layered, not uniform. Put ±0.005 mm (±0.0002 in) on the features that mate or seal, and let general dimensions run to ISO 2768-m. A drawing where every dimension carries the tightest band costs more to inspect than to machine.
Turning at volume is two different machines. Chuck lathes and Swiss-type sliding-headstock lathes solve different problems; a supplier who runs both can quote your whole part family from one drawing package.
Landed cost to the USA is knowable before you order. FOB, CIF or DDP, HS classification and freight mode can all be stated on the quote, so the comparison between suppliers happens on real numbers.

What "volume" actually means for aluminum CNC work

Buyers use "mass production" to mean anything from 500 pieces to 5 million. The process differences matter more than the word:

Volume	What the shop must show you
1–50 (prototype)	First-article report with measured values, DFM notes on the drawing
50–1,000 (pilot)	Fixture concept, cycle time estimate, process sheet
1,000–50,000 (production)	Tool-life plan, SPC or trend data on key features, capacity per week
50,000+ (continuous)	Dedicated fixtures, multiple-machine balancing, packaging and freight plan

A supplier who quotes 100,000 pieces without mentioning fixtures, tool life or inspection ratio is quoting a wish, not a process.

Aluminum grades: what changes between them

Grade	Typical use	Machining	Anodizing consistency	Relative price
6061-T6	Housings, brackets, heatsinks, structural parts	Good, chips break cleanly	Uniform, takes color well	Baseline
6063	Extrusion-adjacent profiles, enclosures	Good	Uniform	Slightly below 6061
6082	European-spec structural parts	Good	Uniform	Close to 6061
5052-H32	Sheet-metal parts, covers	Poor for machining (gummy)	Very uniform	Low
7075-T6	High-stress, thin-wall structural parts	Moderate	Patchy risk on colored finishes	1.5–2×

Two practical notes for US prints. First, if the drawing is European (EN or DIN standard parts), 6082 may be the specified grade; substituting 6061 is a design change, not a shop decision. Second, for cosmetic anodized housings, 6061-T6 runs with fewer color-lot problems than 7075, which carries zinc and tends toward patchiness in dyed finishes.

Tolerance layering with ISO 2768-m

Nominal length	ISO 2768-m band	Critical features
0.5–3 mm	±0.1 mm	±0.005 mm
3–6 mm	±0.1 mm	±0.005 mm
6–30 mm	±0.2 mm	±0.005 mm
30–120 mm	±0.3 mm	±0.005 mm
120–400 mm	±0.5 mm	±0.005 mm or better
400–1000 mm	±0.8 mm	quoted per feature

The gap between a general band and a critical band is 20 to 40×. A part with 8 mating dimensions and 40 general dimensions is clean to build and inspect when the drawing says so explicitly. When all 48 dimensions carry ±0.005 mm, inspection time can exceed machining time, and the quote reflects that whether or not the function requires it.

Thin aluminum walls add a second constraint: clamping force. Walls under 1.5 mm distort under a vise and spring back on release. Volume-capable shops design the fixture before the program, leave stock for a stress-relief step between roughing and finishing, and measure after the part leaves the fixture — ask about all three.

Turning at volume: chuck lathes versus Swiss-type

Attribute	CNC chuck lathe	Swiss-type sliding headstock
Work envelope	Larger diameters, chuck-held	Ø2–20 mm bar-fed, long parts
Deflection control	Depends on setup	Guide bushing supports the stock at the cut
Back-side operations	Often a second setup	Sub-spindle completes in one cycle
Typical parts	Flanges, hubs, larger fittings	Pins, connectors, medical components, threaded fittings

The distinction matters when your part family includes both. A shop running 50 chuck lathes and 70 Swiss-type machines can keep the whole family on one purchase order, with consistent inspection standards across the two lines.

Volume consistency in turning comes from boring basics done relentlessly: tool-life counters with forced changes, trend charts on the two or three features that matter, and first-article reports carrying measured values rather than an approval stamp. Ask to see a redacted report before the first PO.

Anodizing and size: plan the film

Anodic film grows about half outward and half into the base metal. A Type II sulfuric finish at 10–15 µm adds roughly 5–8 µm per surface; a Type III hard coat at 50 µm adds about 0.025 mm per surface, and holes shrink by a similar amount. Three planning rules:

State on the drawing whether dimensions apply before or after finishing.
Mask or pre-compensate mating dimensions.
Expect dark colors to run thicker films than clear, with a correspondingly larger size shift.

Landed cost to the USA

The quote should separate: part price, freight mode, Incoterms (FOB Shenzhen, CIF or DDP to your door), HS classification, and any tariff line that applies to your HTS code. Air express reaches most US cities in 5–8 days; sea freight runs 25–35 days for larger batches. For repeat programs, sea freight plus a safety-stock agreement usually beats air on total cost, but that is a procurement decision, not a hidden one — the numbers should be on the table.

Quality documents to expect

FAI (first-article inspection) report with measured values
CMM dimensional report per batch or per sampling plan
EN 10204 3.1 material certificates traceable to heat lot
Batch traceability and, on request, PPAP-style packages for automotive programs

Certification-wise, ISO 9001:2015 and ISO 14001:2015 are the baseline documents to verify (and verify means checking the certificate number and validity, not the logo on the website). IATF 16949 and ISO 13485 are sector-specific systems; a supplier who is implementing them will say so plainly, and one who claims everything should be asked for certificate numbers.

Frequently asked questions

Q: What is a realistic MOQ for aluminum CNC parts from China?
One piece is workable when the shop runs single-piece scheduling on the same line as production. Tiered pricing then reflects setup amortization as volume grows. If a supplier refuses singles entirely, that tells you their line is organized for continuous flow — fine for mature programs, restrictive for development.

Q: How fast can prototypes and production ship?
Prototypes 3–7 days, production batches 7–15 days, surface treatment adding 3–7 days when required. Air express 5–8 days door to door in the USA; sea freight 25–35 days. Quotes within 12 working hours when the drawing package is complete.

Q: Can you hold ±0.005 mm across a whole aluminum part?
On selected critical features, yes, with CMM verification. Across every dimension of a typical part, no shop does this economically — and if one claims to, ask how inspection is priced. Layered tolerancing is the standard engineering answer.

Q: What do you need from us to quote?
A 2D PDF or 3D model, material grade, annual and per-order quantities, finish requirements, and target delivery. DFM feedback comes back with the quote — if a tolerance is looser than the function requires, the quote should say so.

The parameters above reflect the operating standards of a 220-machine aluminum and turning production line (100 machining centres including 30 positioned five-axis, 50 CNC lathes, 70 Tsugami Swiss-type machines) and are given for engineering comparison. Fit and process feasibility for a specific part should be confirmed on a first article.

(The figures above come from CHX Precision Metal in Shajing, Shenzhen — *chxmetal.com** has the machine list behind them.)*

How to Vet a Liquid Cooling Cold Plate Supplier: Skived Fin, Friction Stir Welding or Vacuum Brazing

Changhexin Zhang — Mon, 26 Oct 2026 15:11:09 +0000

Most cold plate projects fail for reasons that have nothing to do with the machine tool. They fail because the drawing specified a flatness callout that nobody measured, or because the leak test was a five-second bubble check, or because the supplier quoted a brazed assembly and delivered a gasketed one. The machining itself is usually the easy part.

This guide is written for engineers and sourcing managers who need to separate a supplier that builds custom cold plates from a catalogue vendor that resells standard heat sinks. It covers the four part types, the three manufacturing routes, the four specs that decide whether the part works, the questions to ask, and the five parameters you need before anyone can quote honestly.

1. First, separate two kinds of supplier

This is the step that gets skipped, and it costs the most time.

A catalogue vendor sells heat sinks and cold plates by part number, from stock, in fixed footprints with published thermal resistance curves. That is a legitimate business, and if your design happens to match an existing part number, buying it is faster and cheaper than machining one.

A build-to-print shop starts from your 2D drawing or 3D model, quotes per feature, and holds no finished inventory. It exists to turn a drawing into a part.

Dimension	Catalogue vendor	Build-to-print shop
Basis of quote	Part number, size code, quantity break	Your drawing, feature by feature
Stock	Ships from inventory	No finished stock, schedules to order
Custom footprint	No, or requires a new tooling charge	Yes, that is the normal case
Minimum order	Packaging or carton quantity	1 piece is common
Lead time meaning	Ships now	Setup + machining + inspection + finish
FAI report	Usually not provided	Expected, with measured values
Can you change the drawing	No	Yes, and DFM feedback is part of the value

Three questions settle it: "Will you make this to my drawing, or select a part number?", "Will you run one piece?", and "Will you send DFM comments before quoting?" A shop that answers all three toward build-to-print is the right kind of supplier. A shop that returns only a number, with no comment on the drawing, is usually filling in a template.

If what you actually need is a standard extruded heat sink, buy the standard heat sink. Custom machining is the wrong tool for that job, and pretending otherwise wastes both sides' time.

2. Four part types, four different problems

"Cold plate" is used loosely. The four families below have different critical features and different failure modes.

Cold plate (flat plate with internal channel). The critical features are the sealing face flatness, the channel cross-section, and the bond quality between plate and cover. Failure mode is leakage at the seal or a blocked channel from burrs and braze flash.

Cold rail (long, narrow, often extruded or machined bar). The critical feature is straightness over length and the flatness of the mounting face that contacts the die or the module. Failure mode is poor contact pressure at the ends, which shows up as a hot spot rather than a leak.

Manifold (distribution block with multiple ports). The critical features are port position, internal cross-drill alignment, and thread depth. Failure mode is flow imbalance between branches, or debris from cross-hole burrs migrating into a pump.

Quick-disconnect housing / connector body. The critical features are the O-ring groove dimensions, concentricity between the bore and the groove, and surface finish in the sealing zone. Failure mode is a slow weep that only appears after thermal cycling.

Ask a supplier which of these four they have actually shipped, and ask for the inspection records from one of them. The answer tells you more than a capability list.

3. Three routes to a sealed channel, and how to choose

The channel can be created by skiving, by machining plus friction stir welding, or by machining plus vacuum brazing. The choice drives cost, lead time, thermal performance and risk.

Route A: skived fin. A skiving tool peels a continuous fin from a solid block of aluminium or copper. Fin thickness around 0.3 mm and fin density up to roughly 1.5 mm pitch are achievable, with no bond line between fin and base — the fin and base are one piece of metal. Shortest thermal path per unit cost, and no braze or weld joint to inspect. Limitation: the channel geometry is essentially straight fins, so it does not suit complex flow paths, and thin copper fins are harder to skive than aluminium.

Route B: machined channel plus friction stir welding (FSW). The channel is milled into the base, a cover plate is friction stir welded on, and the joint is a solid-state weld that stays below the melting point. Suits long straight seams, gives a fine grain structure in the weld zone, and avoids the filler metal and flux of brazing. Limitation: the process needs a rigid fixture and a machine with enough downforce, and it leaves a weld bead that must be machined flush if the face is a sealing surface.

Route C: machined channel plus vacuum brazing. The channel is milled, filler metal is placed, and the assembly is brazed in a vacuum furnace. Suits complex internal geometries, multi-layer stacks, and copper assemblies. Limitation: the braze joint is the thermal and mechanical weak link, flux residue and braze flash can obstruct channels, and furnace load size caps part dimensions.

Criterion	Skived fin	Machined + FSW	Machined + vacuum braze
Thermal path	Shortest (single piece)	Good (solid-state joint)	Depends on braze coverage
Geometry freedom	Straight fins only	Straight seams, simple cavity	Complex cavities, multi-layer
Tooling cost	Skiving blades, per fin pitch	Fixture + FSW machine	Fixture + furnace load
Typical prototype lead time	5–10 days	7–15 days	10–20 days
Joint inspection	Not applicable	Weld bead, visual + section	X-ray or section, coverage %
Main risk	Fin damage in handling	Fixture rigidity, weld flash	Voiding, flux residue, distortion

If your channel is a set of straight fins under a flat lid, skiving is usually the lowest-risk route. If you need a serpentine path with a flat sealing face, machined plus FSW is the cleaner joint. If you need a stacked or multi-cavity internal geometry, brazing is often the only way, and then the inspection plan matters more than the process choice.

4. The four specs that decide whether it works

1. Flatness of the contact face. This is the number that governs thermal contact resistance, and it is the number most often quoted as a drawing note and never measured. A practical target for a cold plate in the 100 mm class is 0.02 mm over 100 mm, measured on every part rather than sampled. A plate that is flat on the bench and 0.05 mm out after the fittings are torqued in has not been controlled, it has been measured once.

2. Surface roughness in the sealing zone. Ra 1.6 µm is a workable starting point for an O-ring seat; a lapped finish is needed where a metal-to-metal seal or a thin gasket is used. Roughness outside the seal band matters much less than people assume — the fin field and the outer surfaces can be left as-machined.

3. O-ring groove geometry. Groove width, depth and corner radius set the squeeze percentage, and squeeze sets whether the seal holds over thermal cycling. Groove depth tolerance tighter than ±0.05 mm is normal for this feature, and the groove should be concentric to the bore within the drawing callout, not "as machined".

4. Leak test method. Ask three things: pressure, medium, and hold time. Pressure decay at 1.5 times the working pressure is a common acceptance criterion; a bubble test in a water tank is cheaper and less sensitive. Hermetic parts often need helium mass spectrometry with a stated leak rate. A supplier who cannot state all three numbers in the quote has not thought about the failure mode.

5. Six questions worth asking before you send a drawing

Which route do you propose for this geometry, and why? A supplier who answers "brazing" for every part is selling one process, not solving your problem.
How is flatness measured, on what fixture, before or after fittings are installed? The answer should name the instrument and the clamping state.
What is your leak test specification — pressure, medium, hold time, acceptance criterion?
Is the sealing face machined after joining, or before? After joining is the only answer that controls distortion.
Who does the surface finish — anodising, chromate conversion, nickel plating — and how does the finish thickness affect the groove? Hard anodising grows into the groove and changes squeeze.
Can you show a section or X-ray of a brazed joint from a previous job? Coverage percentage, not a verbal assurance.

6. Five parameters you need before anyone can quote

Quotes for the same drawing differ by 2× for explainable reasons. Supply these five and the spread narrows:

Heat load and flow rate (W and L/min) — determines whether fin density matters at all.
Working pressure and test pressure — sets wall thickness, joint type and test cost.
Fluid (water, water-glycol, dielectric, refrigerant) — rules out certain alloys and platings.
Interface: die size, mounting hole pattern, torque spec, and whether a TIM is applied by you or the supplier.
Quantity and life: prototype only, pilot, or production volume, plus expected annual usage.

Missing any of these turns a quote into a range, and a range is what causes the "why is this one three times the other" conversation two weeks later.

7. Our own numbers, as a reference point

Not a recommendation to choose us — a concrete coordinate to compare other quotes against.

Location and scale: Shajing, Bao'an, Shenzhen; 5,000 m² plant; exports to 50+ countries.
Equipment: 220 CNC machines — 100 machining centres (30 of them 3+2 axis, mainly aluminium), 50 CNC lathes, 70 Tsugami Swiss-type lathes.
Tolerance: ±0.005 mm on critical features, ISO 2768-m for general dimensions, verified on a CMM at first article, run start and lot release.
Cooling work: skived fin, machined channel with FSW cover, machined channel with vacuum-brazed cover, plus cold rails and manifold blocks; copper and aluminium, with nickel plating on copper where the fluid requires it.
Inspection: flatness checked per part with a record, Ra measured in the seal band, leak test at 1.5× working pressure by pressure decay, helium test on request.
Lead time and MOQ: prototypes 3–7 days, batch production 7–15 days, surface finishing adds 3–7 days, MOQ 1 piece with quantity price breaks.
Certification: ISO 9001:2015 and ISO 14001:2015. IATF 16949 and ISO 13485 are in progress and not yet certified — we do not claim them.

Our business is build-to-print. You send a drawing or a 3D model, we come back with a process plan and a quote, usually within 24 hours, with DFM comments naming the tolerances that cost money and buy nothing. We do not sell standard parts and we do not hold finished inventory.

FAQ

What is a realistic flatness callout for a cold plate?

0.02 mm over 100 mm is achievable and measurable per part. Tighter than that on a large plate pushes you into lapping and changes the cost structure; the question to ask is whether the thermal model actually needs it.

Skived fin or machined fin — which performs better?

Skived fin has the shorter thermal path because fin and base are one piece of metal with no bond line. Machined fin wins when the channel geometry is not a straight fin field. For straight fins under a flat lid, skiving is generally the better cost-performance choice.

How do I check a brazed joint I cannot see?

Ask for coverage percentage from an X-ray or a destructive section of a sacrificial part from the same furnace load. "We braze it properly" is not an inspection result.

Does hard anodising affect the O-ring groove?

Yes. The coating grows into the surface and reduces groove depth, which changes squeeze. Either mask the groove, machine it oversize before coating, or specify the finish on the mating surfaces only.

What leak test should I specify?

Pressure decay at 1.5× working pressure with a stated hold time and a stated drop threshold is the usual industrial baseline. For dielectric or refrigerant loops, add a helium test with a leak rate specification.

Can I get one piece for a prototype?

At a build-to-print shop, yes — that is the normal starting point. At a catalogue vendor, usually not, because there is no part number for your geometry.

Tolerance, lead time and MOQ figures in this article reflect our own production standards and are given for comparison. No assessment is made of any specific company; verify any supplier on site before placing production orders.

(CHX Precision Metal, Shajing, Shenzhen — *chxmetal.com*.)

How to Vet a Titanium (TC4) and PEEK CNC Machining Supplier in China

Changhexin Zhang — Mon, 26 Oct 2026 15:05:08 +0000

A titanium bracket and a PEEK end effector rarely come from the same shop, even though both are "CNC machining". The material decides the tooling, the coolant, the fixture and the inspection — and a quote that ignores those differences is a quote you will renegotiate later.

This is a vetting checklist for buyers sourcing titanium (Ti-6Al-4V / TC4) and PEEK parts from China: what to ask, what the answers should sound like, and which numbers actually matter.

Two kinds of supplier, and only one of them is relevant here

Before comparing prices, establish which business you are talking to.

	Catalog parts supplier	Build-to-print machine shop
Prices from	Part number, spec table, quantity break	Your 2D drawing / 3D model, feature by feature
Stock	Ships from inventory	No stock; scheduled per order
Custom geometry	Declines, or asks you to pick a standard size	This is the entire business
MOQ	Carton or pallet	1 piece is a normal answer
First article report	Usually not offered	Expected, with measured values
Will they challenge your drawing	No	Yes — that is the DFM feedback you are paying for

Three questions separate them:

"This custom geometry — do you make it to my drawing, or do I pick from your catalog?"
"Will you run one piece?"
"Before quoting, will you send a DFM note flagging tolerances that are tighter than the function needs?"

A shop that pushes back on your drawing is usually costing the part from the process. A shop that only returns a number is quoting a template.

Titanium: what changes the process

Four grades cover most requests:

Grade	Designation	Tensile (approx.)	Typical use	Machining difficulty
2	Commercially pure Ti	345–450 MPa	Chemical pumps, marine hardware, heat exchanger plates	Moderate
5	Ti-6Al-4V / TC4	900–1000 MPa	Structural brackets, UAV airframes, fasteners, medical instruments	Medium-high
9	Ti-3Al-2.5V	620–720 MPa	Hydraulic lines, formed pressure parts	Medium
23	Ti-6Al-4V ELI	860–965 MPa	Implant-grade, low-temperature toughness	As Grade 5, plus cleanliness

If the drawing says only "titanium", a supplier may quote Grade 2 and deliver a part that is 40 percent weaker than the Grade 5 you designed for. Naming the grade and the material standard (ASTM B348 or GB/T 2965) is the cheapest risk control available.

Why titanium is slow — three physical facts

Thermal conductivity around 7 W/(m·K). 6061 aluminium is roughly 167. Heat from the cut has nowhere to go except into the tool edge, which is why cutting speed runs at about one third of the aluminium value for the same tool.

Work hardening. Titanium hardens in the cut. A light pass over an already hardened surface accelerates wear instead of improving finish, so the process has to keep constant engagement rather than dwell.

Chemical reactivity at temperature. Titanium reacts with tool materials when hot, which is why tooling has to be dedicated. Cutters that have run titanium should not touch aluminium or stainless, and the same applies to fixtures and the cleaning station. Contamination does not show up in dimensional inspection; it shows up later as porosity in a weld or pitting after anodising.

Add low elastic modulus (about 110 GPa, roughly half of steel) and thin walls deflect about twice as much for the same clamping force. Fixture rigidity becomes the limiting factor, not machine accuracy.

Six questions for a titanium supplier

Are titanium tools kept separate from aluminium and steel tools? The answer should be yes, including fixtures and cleaning.
How is coolant delivered? Through-tool high pressure is the strong answer. The target is the cutting edge, not the workpiece.
Have you machined walls under 1 mm? Show the fixture approach. A real answer describes support, material removal sequence and measurement after release.
Can material certificates be traced to the mill heat number? Yes, with EN 10204 3.1 available on request.
Is stress-relief annealing in-house or subcontracted? Most machine shops subcontract vacuum annealing. Ask who does it, how long it takes, and whether finish machining happens after annealing.
What inspection closes the loop? CMM as the reference, measured values in the report rather than a pass stamp, SPC on critical characteristics for volume programmes.

PEEK: what changes the process

PEEK is not one material. Four grades cover most of the demand, and they behave differently enough that a quote saying only "PEEK" is not a quote.

Grade	What it is	Key property	Typical part	Machining note
Unfilled	Neat PEEK	Highest elongation and chemical resistance, lowest stiffness; continuous service around 250 °C, density near 1.30 g/cm³, tensile around 100 MPa	Seals, insulator bodies, fluid components	Cuts easily, moves easily — annealing is mandatory
GF30	30% glass fibre	Stiffer, better creep resistance, abrasive to tooling	Structural brackets, pump housings, valve bodies	Change inserts on a count, not on failure
CF30	30% carbon fibre	Stiffest, lowest thermal expansion, electrically conductive	Wafer handling end effectors, precision fixtures	Abrasive plus conductive — dust and static control
Bearing grade	Graphite / PTFE additions	Low friction, wear resistance	Bushings, thrust washers, sliding pads	Surface smears easily; parameters matter

The annealing sequence is the tolerance

PEEK parts that fail inspection usually failed weeks earlier, in the process plan. The sequence has to be: rough machine → stress-relief anneal → finish machine. Annealing after finish machining only releases part of the stress; the stress from the finishing cut stays in the part.

Practical consequences:

Measure after the part has cooled and stabilised, not at the machine while it is warm.
Light clamping with soft jaws and support under thin sections — PEEK deflects far more than aluminium under the same vise pressure.
Account for thermal expansion: PEEK expands roughly four times as much as steel over the same temperature rise. A part that measures correctly at 20 °C can be out of tolerance at 80 °C service.

Why PEEK edges turn white

Whitening is stress whitening, and it comes from rubbing rather than cutting: a dull tool, a feed that is too low, or too much heat. It is not cosmetic. The white layer is a stress concentration and a crack initiation site.

The correct response is to change the tool and the parameters. Polishing the surface removes the evidence and leaves the stress. When you ask a supplier how they handle whitening, "we polish it" and "we change the cutter and the feed" identify two very different shops.

The tolerance conversation

±0.005 mm is a two-sided band with 0.010 mm total width. On a 20 mm bore that is 0.05 percent of nominal, which is why it cannot be a blanket tolerance on a whole drawing.

Nominal length	ISO 2768-m (medium)	What is held on critical features
0.5–3 mm	±0.1 mm	±0.005 mm
3–6 mm	±0.1 mm	±0.005 mm
6–30 mm	±0.2 mm	±0.005 mm
30–120 mm	±0.3 mm	±0.005 mm
120–400 mm	±0.5 mm	±0.005 mm or better
400–1000 mm	±0.8 mm	quoted per feature

The gap is 20 to 40 times. A part with 8 critical dimensions and 40 general ones machines and inspects cleanly when the tolerance is applied where it matters. The same part fully toleranced at ±0.005 mm can double in price with no functional gain, because inspection time can exceed machining time.

Five things actually limit the achievable band, and machine accuracy is only the first: fixture rigidity, tool wear and thermal growth, material behaviour, the datum scheme, and feature geometry (a deep narrow slot reaches the tool's limit long before the machine's positioning accuracy matters).

Three questions worth asking:

"On which features do you hold ±0.005 mm?" A supplier answering "all dimensions" has not thought about it.
"Does the first article report give measured values or a pass stamp?" Measured values, with distribution, are the only useful version.
"Can you supply SPC data on critical characteristics?" For volume programmes, receiving inspection needs it to avoid 100 percent checking.

Lead time and minimum order

"Lead time" hides three different clocks. Split them:

Item	What to pin down	Typical
Quote turnaround	How fast, and does it include a DFM note	Within 24 hours is achievable
Prototype	From PO to parts shipped, material included	3–7 days
Production	From PO to batch complete	7–15 days
Surface finishing	Separate from the numbers above	Add 3–7 days
Minimum order	Will they run one piece; are there quantity breaks	1 piece, with price breaks by quantity
Documentation	First article, material cert, inspection records included	Should ship with the parts, not as an extra

A shop quoting "7 days" that means 7 days of machining, plus 5 days of material and 4 days of finishing, is describing a 16-day delivery. Writing the six lines into the PO is how that gets avoided.

What to send for a quote

2D drawing with tolerances, datum scheme, surface roughness; or a 3D model plus a tolerance note
Material grade and standard
Quantity for prototype and for production
Surface finish requirement, including any cosmetic zone
Target lead time and any hard date

With those five, a shop can quote. Without the datum and the material grade, it can only quote a range.

If you want those five inputs turned into a quote, the shop is CHX Precision Metal, Shajing, Shenzhen.

Why a Machined Part Moves After It Leaves the Fixture: Stress, Warping and What to Do About It

Changhexin Zhang — Mon, 26 Oct 2026 02:49:35 +0000

A buyer once sent us the same aluminum housing twice in one month. Both parts measured flat within 0.01 mm on the vendor's inspection report. Both arrived visibly bowed. This is not a rare story — it is one of the most common precision-machining failures, and it almost never shows up on the machine. Here is what actually causes it, and what to specify so it does not happen to you.

The question, stated precisely

Why does a part that measures perfect on the machining center end up out of tolerance after it is removed from the fixture, packed, or a week later?

The technical breakdown: it is almost always stress, not accuracy

When a CNC machine removes material, three things release energy at the same time:

Residual stress inside the raw stock. Extruded, rolled or forged material carries locked-in stresses from its own manufacturing. Milling away one side of a stressed plate is like cutting one string on a pulled bow — the part moves to rebalance. Aluminum plates from the same batch can differ noticeably in this behavior.
Cutting-induced stress. Heat and force at the cutting edge compress and stretch the surface layer. Deep, aggressive roughing in thin walls builds up more of it.
Clamping force. A vise or fixture that pulls a flexible part flat will "help" it pass inspection while clamped. Release it, and the true shape reappears.

The machine itself is rarely the problem. A machine that holds 0.005 mm will happily produce a part that moves 0.05 mm the day after — because the geometry changed inside the material after cutting stopped.

The numbers: what to expect before you get burned

Factor	Typical effect on flatness	Mitigation
Stress-relieved (T651/T7351) plate vs standard extrusion	Difference of 0.02–0.10 mm on a 200 mm span after heavy material removal	Specify stress-relieved temper up front
Roughing → release → finishing vs single-pass to final size	0.03–0.15 mm improvement on thin-walled parts	Write rough-then-finish into the process
Roughing → release → finishing with an intermediate rest	Further 0.005–0.02 mm gain on long parts	Accept extra lead time
Deep roughing on thin walls (<1.5 mm wall)	Largest single contributor to post-machining warp	Control depth of cut, leave finishing stock
Final skim pass with light clamping	Recovers geometry the fixture was hiding	Insist inspection happens unclamped

Two takeaway numbers worth remembering:

A part is not "flat" until it is measured free of the fixture. Anything else measures the clamp, not the part.
Symmetry matters. Removing equal amounts of material from both sides keeps the stress balance; machining everything from one side guarantees bending.

What to actually specify (the DFM answer)

Ask for stress-relieved stock on parts with high material-removal ratio or walls under 2 mm. It costs more per kilogram and saves the whole batch.
For critical flatness, require a rough → release → finish sequence, and say so on the drawing notes. Suppliers who already machine this way will not blink.
Put the inspection condition on the drawing: "flatness measured in free state, unclamped." Without it, every party can be technically right and the part can still be wrong.
For aerospace-grade alloys, an intermediate thermal treatment between roughing and finishing is the standard answer; for most industrial parts, a rest period plus symmetric machining is enough.
Quote tolerances that match function. A spacer that rides inside a machined pocket does not need 0.01 mm flatness; an optical or sealing face does.

Conclusion

Post-machining movement is a materials-and-process question, not a spindle-accuracy question. The fix is specified by the buyer, executed by the supplier, and verified by unclamped inspection. Parts that arrive flat are the product of a sequence, not a lucky machine.

About the author

CHX Precision Metal is a CNC machining manufacturer in Shenzhen, China, operating since December 2013 across a 5,000 m² facility with 100+ staff and 220 CNC machines — including 100 five-axis machining centers, 50 CNC lathes and 70 Tsugami Swiss-type lathes. We hold ±0.01 mm standard tolerances (±0.005 mm with grinding) under ISO 9001:2015 and ISO 14001:2015, with 3–7 day prototyping and 7–15 day production, from a MOQ of 1 piece. If your last batch taught you what stress relief means, tell us about the part: https://chxmetal.com — or write to mc@chxmetal.com.