NC Prototype Crushed It: What ±0.001 mm Really Takes

• 1
  Key takeaways
• 2
  Why an NC prototype crushed it to ±0.001 mm
• 3
  Thermal drift, tool wear and fixturing: the real error budget
• 4
  Which features can hold ±0.001 mm and which cannot
• 5
  How the result is proven, not just claimed
• 6
  When a micron-level callout is worth the cost
• 7
  Tolerance feasibility by feature type

The claim that an NC prototype crushed it to ±0.001 mm sounds like a machine specification. It is not. A typical 3-axis or 5-axis machining center positions its axes far finer than that, but positioning accuracy and finished part accuracy are different quantities. The cutter, the workpiece, the fixture and the room all move during the cut, and those movements are often an order of magnitude larger than the servo resolution.

On a real prototype run, the achievable tolerance is set by the tightest link in a chain: spindle thermal growth, tool deflection, chip load variation, clamping stress, and the temperature of the part when it is measured. A part machined at 24 °C and measured at 20 °C changes size before the probe touches it. For aluminium 6061, the coefficient of thermal expansion is about 23 × 10⁻⁶ per °C. A 100 mm feature shifts roughly 0.009 mm over a 4 °C difference. That single effect is already several times larger than ±0.001 mm.

So when a shop reports a micron-level result, the number usually comes from a narrow set of features, measured under controlled conditions, after a finishing operation. It is a process achievement, not a catalogue capability. Understanding which features were measured, and how, tells you whether the same result is repeatable on your part.

Every cut carries an error budget. Thermal drift is usually the largest term on a long cycle. A spindle running at 12,000 rpm warms the headstock and the tool holder, and the Z axis grows as a result. Over a two-hour cycle, that growth can reach 0.01–0.02 mm on an uncontrolled machine. A temperature-controlled room at 20 ± 1 °C reduces the swing, and a warm-up cycle before the first cut removes the rest.

Tool wear is the second term. A carbide end mill cutting 7075 aluminium wears slowly, but the same tool in 17-4PH stainless or Inconel wears fast enough to move the finished size within a single part. For tight work, we change tools on a fixed cycle count rather than waiting for a surface finish problem to appear. Cutter compensation then keeps the nominal size stable across the batch.

Fixturing is the quiet one. Clamping a thin wall distorts it while it is held, and the part springs back when the clamps release. A bore that measured 0.003 mm oversize on the machine can measure on size after unclamping, or the reverse. For features below 0.005 mm, the fixture design matters as much as the cutting strategy.

• 1
  Warm-up firstRun the spindle 20–30 minutes before the first tight cut.
• 2
  Light finishing passes0.05–0.1 mm radial engagement keeps deflection predictable.
• 3
  Probe on the machineIn-process probing catches drift before the part leaves the fixture.

Feature size drives feasibility more than material does. A short, stiff feature with a rigid support behind it is the easy case. A Ø10 mm pin, 15 mm long, turned in one setup on a mill-turn center, can be held to a micron with a ground finish and a controlled room. The cutting forces are small, the tool is short, and the part does not move.

The hard case is a long, slender feature or a thin wall. A 300 mm aluminium plate 2 mm thick will deflect under its own clamping and under the cutter. No amount of machine accuracy fixes that. The tolerance on such a feature should be opened to ±0.02 mm or the design changed to add ribs, a thicker section or a stress-relief step.

There is also the question of datum. A ±0.001 mm callout on a hole that references a face machined in a different setup inherits the setup error. For micron work, the datum and the feature should be cut in the same setup, on the same machine, without re-clamping. This is one reason 5-axis machining helps: fewer setups means fewer datum transfers.

A tolerance claim is only as good as the inspection behind it. A caliper or a micrometer read by hand has a resolution around 0.01 mm and an operator-dependent repeatability that is worse. To validate a micron-level result, the part has to be measured on a CMM in a temperature-controlled room, with a probe calibration traceable to a known standard.

Measurement uncertainty matters too. If the CMM itself carries ±0.002 mm of uncertainty, a reading of 10.001 mm does not prove a ±0.001 mm part. Good practice is to keep the measurement uncertainty at roughly one quarter of the tolerance being checked. That usually means a CMM, not a height gauge, and it means letting the part soak at 20 °C before the first touch.

We inspect 100% of parts before shipment, and reports are available on request. Raw material certificates, in-process checks and final inspection notes can all be attached to the shipment. If a drawing calls for a micron-level feature, we will say which features we can prove and which we cannot, before the job starts.

A micron-level tolerance on a prototype is worth paying for when the feature controls function. A bearing seat, a spigot that sets alignment, a valve bore, a lens mount: these are the features where a few microns change how the assembly behaves. In those cases, the extra cost of a controlled room, a grinding step and a CMM report is small compared with the cost of a failed fit test.

It is usually not worth paying for on non-functional geometry. Cosmetic edges, clearance holes, cable routing slots and mounting tabs rarely need better than ±0.05 mm. Putting a tight tolerance on those features does not improve the part. It adds inspection time, slows the cycle, and can push the quote up without changing the result.

A practical approach is selective tolerance. Mark the two or three features that matter, keep them tight, and leave the rest at the general tolerance block. This is the pattern we see on well-prepared drawings, and it is the pattern that keeps a prototype affordable while still proving the critical dimensions.