CNC Precision Production: How Accuracy Holds at Volume

• 1
  What CNC precision production really controls
• 2
  Where 5-axis kinematics help and where they do not
• 3
  Thermal growth is the largest hidden error in CNC precision production
• 4
  Measurement closes the loop or the loop drifts
• 5
  Material behavior and post-processing shift the achievable band
• 6
  From one prototype to a 10,000+ part run
• 7
  Machine configuration by part geometry

A tolerance callout is a promise about a distribution, not about one part. Cutting one pocket to ±0.005 mm is a setup problem. Holding that band on part 4,000 is a process problem: the machine geometry, the thermal state, the fixture, the tool, and the measurement loop all have to stay inside the same window for hours.

CNC precision production is the discipline of keeping those five variables bounded. When a shop quotes ±0.005 mm (±0.0002 in), it is claiming that the sum of machine positioning error, thermal growth, fixture deflection, tool wear, and gauge uncertainty stays under that band. Each source is small on its own. They add up.

The useful question for an engineer is not which machine a shop owns. It is which error sources the shop measures and compensates, and how often. A 5-axis center with no thermal logging can be less repeatable than a 3-axis mill that runs a warm-up cycle and a probe check every few hours.

• 1
  Positioning errorBall screw pitch error and encoder resolution set a floor the machine cannot beat.
• 2
  Thermal driftSpindle and ball screw growth can move the tool 10–30 μm over a shift.
• 3
  Fixture stiffnessA flexible setup moves under cutting force and shows up as taper or chatter.
• 4
  Tool wearFlank wear changes both size and surface finish as the edge dulls.

Simultaneous 5-axis motion moves the tool along three linear axes (X, Y, Z) and two rotary axes at once. The gain is not speed. The gain is that you can reach a face, a fillet, and an undercut in one setup, with a short rigid tool, at an angle that keeps the cutting edge loaded correctly.

That matters for impellers, turbine blades, complex castings needing finish passes on sculpted surfaces, and any part with features on five sides. In those cases the alternative is three or four setups, and every setup adds a datum shift. Each re-clamp contributes its own error, often 10–20 μm if the fixture is not designed for the part.

For prismatic parts with features on two or three faces, 5-axis adds little. A 3-axis mill with a good vise and a probe cycle can be more predictable, because there are fewer moving elements to characterize. The 27 three-axis machines in a production cell exist for exactly this reason.

The boundary is geometric. If a part has no undercut, no compound angle, and no sculpted surface, the extra rotary axes buy setup reduction only. Pay for 5-axis when a setup reduction removes a real error source or a real handling risk, not as a default.

A spindle running at 12,000 rpm warms up over the first 40–90 minutes. The housing grows, the tool holder grows, and the tool tip position moves. On a linear axis, ball screw heating adds more. Total drift on an unmanaged machine can reach 20–30 μm, which is four to six times the tolerance band on a tight part.

The fix is not a colder room. It is a known state. Shops that hold ±0.005 mm run a warm-up cycle before the first cut, log spindle and bed temperature, and check a master artifact at intervals through the shift. If the artifact moves, the offset is corrected before the next batch, not after the parts are scrapped.

Material matters here too. Aluminium 6061 and 7075 conduct heat away quickly and cut at high speed, so the part stays cooler but the tool wears faster. Stainless 316L and 17-4PH work-harden and run hotter at the edge. Titanium Ti-6Al-4V concentrates heat in the tool because it conducts poorly. Each material changes the drift rate.

For long parts, a 4,000 mm bed grows more than a compact one. A 1 °C rise on a 4,000 mm steel bed adds roughly 48 μm of length. That is why large parts are often roughed, allowed to cool, then finished in a second pass.

• 1
  Warm-up cycleRun the spindle 40–90 minutes before the first precision cut.
• 2
  Master artifact checkProbe a known part at fixed intervals to catch drift.
• 3
  Rough then finishLet large parts stabilize before the finishing pass.

A tolerance you cannot measure is a tolerance you cannot hold. For ±0.005 mm, a coordinate measuring machine (CMM) with a stated uncertainty well below the band is the baseline. Optical comparators and height gauges cover features a probe cannot reach, such as narrow slots and thin walls.

The measurement plan has to match the feature. A CMM touch probe gives good size and position data but can deflect thin walls. Surface finish needs a separate stylus or optical gauge; Ra 0.2–0.8 μm is a fine finish that a visual check cannot confirm. Thread and bore gauges handle features probes cannot enter.

In-process probing changes the economics. Probing a datum on the machine after each clamp lets the control shift the work offset to the actual stock position. That removes stack-up from rough castings and forgings, where the as-cast surface can vary by 1–2 mm. Without probing, that variation lands directly in the tolerance.

The reporting side matters for regulated work. Aerospace, medical, and automotive programs usually require first article inspection reports, material certs, and in-process records tied to a lot number. If the shop cannot produce a traceable report, the tolerance claim is not auditable.

100% inspection before shipment is the baseline at GreatLight: raw material check, in-process monitoring, and final inspection, with reports on request. The point is not the paperwork. The point is that a drift caught at hour three saves the parts made at hour four.

Machinability sets how much of the tolerance budget goes to tool wear. Aluminium 6061-T6 and 6082 cut cleanly and hold size well. Aluminium 7075 is stronger but more notch-sensitive, so sharp edges and light finishing passes matter. Copper alloys such as C36000 machine freely; beryllium copper and pure copper C110 gum up and need different geometry and coolant strategy.

Stainless behaves differently across grades. 303 is free-machining, 304 and 316L work-harden and need constant feed to stay under the hardened layer, and 17-4PH (SUS630) is often machined in the solution-treated state then aged. Titanium TA2 and TC4 (Ti-6Al-4V) need low surface speed and high coolant pressure because the chip carries little heat away. Inconel and Hastelloy push tool life down further and usually add a finishing pass to recover the band.

Post-processing can move a part outside its tolerance. Anodizing builds an oxide layer that adds roughly half the coating thickness per surface, so a hardcoat at 25 μm grows a dimension by about 25 μm across a diameter. Electroless nickel and plating do the same. Passivation and bead blasting remove little material but can round a sharp edge.

The practical rule is to machine to a pre-plate dimension when a coating is specified, and to say so on the drawing. If a shop finishes to nominal and then plates, the part comes back oversized and the tolerance was never really held.

A prototype and a production run stress different things. Prototypes stress reach and setup flexibility: can the shop make the feature at all, in one or two setups, and turn it around fast. A 5-axis cell with no minimum order quantity handles this well because the fixture cost is not amortized over volume.

Production runs stress repeatability and cycle time. The fixture becomes dedicated, the tool path is optimized, and the control offsets are locked. The risk shifts from reach to drift: tool wear across thousands of parts, chip evacuation, and fixture wear at the clamp points. A shop that ran the prototype is not automatically the shop that should run the volume.

The transition point is usually where the fixture changes. If the production fixture is a different design from the prototype fixture, the datum strategy changes and the first article must be re-qualified. Skipping that step is how a part that passed prototype inspection fails at part 500.

Volume also changes the inspection strategy. One-piece inspection on a prototype is affordable. On a 10,000-part run, sampling plus in-process probing plus a final audit is the workable model, with full inspection reserved for critical features. The drawing should mark which dimensions are critical, so the inspection plan can match.

• 1
  Qualification rate99.99% across inspected shipments.
• 2
  Lead timeQuotation and DFM analysis within 12 hours; parts ship in 3–5 days.
• 3
  Order sizeNo minimum order quantity, from one prototype to 10,000+ parts.