CNC machining in CO: how accuracy is actually produced

• 1
  What accurate CNC machining in CO really depends on
• 2
  Machine geometry: where the last 0.01 mm comes from
• 3
  Tool deflection and the limits of a light pass
• 4
  Heat: the error that grows during the shift
• 5
  Material behavior changes the accuracy plan
• 6
  How accuracy is verified, not claimed
• 7
  Which process fits which accuracy problem

Accuracy is not a property of a machine alone. It is the sum of positioning error, tool deflection, workholding stiffness, thermal growth and measurement uncertainty. A 5-axis center rated at ±0.005 mm can still produce a 0.05 mm drift if the fixture moves 0.01 mm under a 2,000 N cut and the spindle warms up 4 °C over an eight-hour run.

The practical question is not which machine is best. It is which error source dominates your part. A thin-wall aluminum housing at 1.5 mm wall thickness is deflection-limited, so cutter geometry and stepover matter more than machine class. A 300 kg steel manifold is mass-limited, so fixture rigidity and thermal soak matter more than spindle speed.

When engineers evaluate CNC machining in CO, we usually ask for three things before quoting: the tolerance band, the datum scheme on the drawing, and the material condition. Those three answers decide the setup, the number of operations and the inspection plan. Everything else follows.

So accuracy is designed, not selected. The rest of this page walks through each error source, the parameters that control it, and the cases where a different process is the better call.

A 3-axis machine moves the tool in X, Y and Z. Every surface that is not parallel to one of those axes needs either a second setup or a re-fixture. Each re-fixture adds its own positional error, typically 0.01–0.03 mm when a vise is reset by hand. That is why a part with four angled faces can hold ±0.005 mm in theory and lose it in the second operation.

A 4-axis mill adds a rotary table, usually Ø400 mm or smaller, so the part can index around one axis. The error budget now includes the table's runout and the rotary encoder resolution. This handles parts with faces around a single axis: shafts, gear housings, valve bodies.

Simultaneous 5-axis machining keeps the part in one fixture and tilts both the tool and the table. The error source shifts from re-fixturing to the machine's rotary pivot accuracy and the post-processor's correctness. When the post is wrong, the error is systematic and repeats on every part, which is easy to detect but expensive to fix.

For parts under 200 mm with three or more non-orthogonal features, 5-axis usually wins on total error even though the single machine is less rigid. Fewer setups beat a stiffer spindle.

• 1
  One setup, one datumEvery re-fixture re-introduces the tolerance stack.
• 2
  Rotary table runout adds directlyCheck it before blaming the cutter.
• 3
  Post-processor errors repeatThey show up as a consistent offset, not random scatter.

A 6 mm carbide end mill with 40 mm of gauge length deflects roughly 0.02 mm under a 300 N radial load. That single number explains most out-of-tolerance walls on aluminum parts. The fix is mechanical, not software: shorten the gauge length, increase the diameter, or reduce the radial depth of cut.

Radial engagement (ae) drives the force far more than axial depth (ap). Dropping ae from 50% to 10% of cutter diameter can cut deflection by more than half while keeping the same material removal rate, because the higher feed per tooth compensates. This is the basis of high-efficiency milling and it is the first thing we change on a drifting wall.

Feed per tooth matters too. Too low and the tool rubs, work-hardens stainless and burns the edge. For 304 stainless we typically stay above 0.05 mm per tooth on a 6 mm cutter. Below that, tool life drops and surface finish gets worse, not better.

Thin-wall parts need support, not more passes. Adding a sacrificial rib or filling the pocket with a low-melt compound changes the stiffness of the part, which is usually the weaker link.

A spindle that runs at 12,000 rpm for two hours grows 20–40 mm in Z relative to a cold start. On a 300 mm tall part, that is 0.02–0.05 mm of taper from the first cut to the last. No controller compensates for it unless the machine has thermal compensation enabled and calibrated.

The countermeasure is boring: warm up the spindle, keep the coolant at a stable temperature, and avoid measuring a cold part. A part at 35 °C measures about 0.004 mm longer per 100 mm than the same part at 20 °C in aluminum. If the drawing calls ±0.005 mm, that alone can fail the part.

This is why we inspect at 20 °C ±2 °C and why we let parts stabilize before the final check. It is also why a supplier that ships immediately after the last cut may report a dimension that changes in transit.

For long runs, the first-off part should be measured after thermal equilibrium, not immediately. Otherwise the process window is set against a moving reference.

Aluminum 6061 and 7075 cut cleanly and hold tolerance well. They also move after machining. A 7075 plate released from a rolled condition can warp 0.1 mm over 200 mm within a day as internal stress redistributes. Rough, stress-relieve, finish is the standard sequence for tight plates.

Stainless 304 and 316L work-harden. A dull cutter or a dwell in the cut raises local hardness and pushes the next pass off. Rigid setups and constant feed matter more than on aluminum. 17-4PH adds a heat-treat step, so the tolerance plan has to account for growth during aging.

Titanium Ti-6Al-4V has low thermal conductivity, so heat goes into the tool. Cutting temperatures above 600 °C at the edge cause rapid wear and dimensional drift. High-pressure coolant and conservative surface speed, often 40–60 m/min, keep the process stable.

Plastics are the opposite problem. POM and PEEK move with temperature and clamp pressure. Light clamping, sharp cutters and air blast instead of flood coolant usually hold dimensions better than heavy fixturing.

A CMM report is only meaningful if it states the temperature, the datum alignment method and the number of points per feature. A report that lists one measured value per dimension tells you almost nothing about a bore's roundness or a face's flatness.

For production parts we check raw material certificates, monitor in-process dimensions on the machine, and run a final inspection before shipment. Reports are available on request. The qualification rate across these checks is 99.99%, which is a measure of the process, not a marketing number.

Gauge repeatability and reproducibility matters. If the same part measured twice by two operators differs by 0.004 mm, the process cannot be controlled to ±0.005 mm. That is a measurement problem, and it has to be solved before the machining problem.

For medical and automotive work, the inspection plan is tied to the certification the part falls under: ISO 13485 for devices, IATF 16949 for automotive. The paperwork is part of the deliverable, not an afterthought.