CNC Machining of High-Precision Parts

• 1
  What actually limits accuracy in CNC machining of high-precision parts
• 2
  How 3-axis, 4-axis, and 5-axis setups change achievable accuracy
• 3
  Material behavior and its effect on high-precision parts
• 4
  Tolerancing, datums, and inspection for high-precision parts
• 5
  Surface finish, edge quality, and secondary operations
• 6
  Matching process and tolerance to the part

A CNC machine does not cut to a number. It cuts to a position, and that position carries error. The error comes from four places: machine geometry, thermal growth, cutting force, and the tool itself. Each behaves differently, and each scales with part size.

Machine geometry is the first limit. A three-axis mill with 0.010 mm positioning error cannot hold ±0.005 mm no matter how slow the feed. The ball screw, the linear guide, and the spindle runout all add up. On our five-axis centers, spindle runout is held under 0.002 mm, which leaves room for the other errors.

Thermal growth is the one most shops ignore. A spindle running at 12,000 rpm warms 3–5 °C in the first hour. On a 300 mm aluminum part, that is roughly 0.007 mm of expansion. You cannot inspect that away after the cut. You have to let the machine warm up or compensate in the program.

Cutting force bends the tool and the part. A 12 mm end mill at 2 mm radial depth in 4140 steel deflects 0.01–0.03 mm. The part also springs back after the cut. This is why finishing passes use light radial depth, high spindle speed, and sharp tools. Roughing cuts move metal; finishing cuts set the tolerance.

• 1
  GeometryBall screw, guide, and spindle error stack up before the tool touches metal.
• 2
  ThermalSpindle and ambient heat shift dimensions during long cycles.
• 3
  ForceTool and part deflection is biggest in deep cuts and hard materials.
• 4
  MetrologyA tolerance you cannot measure is a tolerance you cannot hold.

Every additional axis removes a setup. Each setup is a chance to lose 0.005–0.015 mm in re-clamping. That is the real reason five-axis matters for high-precision parts, not the simultaneous motion itself.

A part with features on four sides needs four setups on a 3-axis mill. Each setup re-datums the part. Even with a good vise and a probe, you lose repeatability. A five-axis center machines those features in one setup, so the datum stays fixed and the relationship between features stays tight.

Four-axis mills sit in the middle. They handle cylindrical parts, slots around a shaft, and hole patterns on a rotary table. If your part rotates around one axis, four-axis is often enough and cheaper to program than full five-axis.

The trade-off is access. Five-axis machines reach undercuts and steep walls, but the rotary axes add their own error. A tilted rotary table at 30° position error of 5 arc-seconds moves a point 100 mm from center by about 0.0024 mm. It is small, but it is not zero. For tight flatness on a single face, a rigid three-axis machine can still win.

• 1
  One setup winsFewer re-clamps means less datum shift.
• 2
  Four-axis for round partsRotary table handles slots and hole patterns around one axis.
• 3
  Five-axis for complex geometryReaches undercuts but adds rotary error to the stack.

Aluminum 6061 and 7075 cut clean and hold tolerance well. They also move after machining. A thin 7075 wall can spring 0.02 mm within hours as internal stress releases. We rough, stress-relieve when needed, then finish. That sequence costs a day and saves a scrapped batch.

Stainless 304 and 316 work-harden. A dull tool rubs instead of cutting, and the surface hardens under the cut. That pushes the tool off line and ruins the finish. Sharp carbide, correct feed per tooth, and no dwell in the cut keep 304 predictable. For 17-4PH, heat treatment condition matters more than the alloy name.

Titanium Ti-6Al-4V and Inconel are a different problem. They conduct heat poorly, so the cutting edge runs hot and wears fast. Tool deflection is high because the material resists the cut. Tolerances tighter than ±0.010 mm on thin Inconel walls are a struggle even on good machines.

Plastics like POM, PEEK, and PC expand with heat and clamp pressure. A PEEK part measured at the machine at 20 °C may read differently in a 23 °C inspection room. For high-precision plastic parts, agree on the measurement temperature and the clamping method before the first cut.

• 1
  Stress reliefRough, relieve, finish for thin aluminum walls.
• 2
  Work hardeningSharp tools and steady feed prevent 304 rub.
• 3
  Heat and clampPlastics need agreed measurement temperature and light fixturing.

A drawing with ±0.005 mm on every dimension is usually a drawing that has not been thought through. Tolerance stack-up matters. If five dimensions in a chain each carry ±0.005 mm, the assembly can drift ±0.025 mm. Assign tight tolerance only where the function needs it.

GD&T helps when it is used honestly. A position tolerance of Ø0.02 mm at MMC is measurable and meaningful. A flatness callout of 0.005 mm on a 400 mm surface is hard to verify and expensive to hold. Ask what the feature does before you tighten it.

Inspection has to match the tolerance. Calipers read to 0.02 mm at best. For ±0.005 mm you need a CMM or a micrometer with a known temperature. We inspect 100% before shipment, with raw material checks, in-process monitoring, and final reports on request.

The measurement environment counts too. A part measured on a warm shop floor and re-measured in a metrology lab can differ. Steel grows about 11 µm per meter per °C. On a 500 mm steel shaft, a 3 °C difference is 0.016 mm. That is larger than the tolerance itself.

• 1
  Tighten selectivelyOnly the functional dimensions need ±0.005 mm.
• 2
  Match the tool to the calloutCalipers cannot verify a 0.005 mm tolerance.
• 3
  Control temperatureA 3 °C shift can exceed the tolerance on long steel parts.

Tolerance and finish are linked. A Ra 0.2–0.8 μm finish needs a light finishing pass, a sharp tool, and a rigid setup. The same cut that holds ±0.005 mm often produces Ra 0.8–1.6 μm, which is fine for most mating surfaces. As-machined at Ra 1.6–3.2 μm suits non-critical faces.

Burns and edge break matter on high-precision parts. A 0.05 mm burr on a sealing face causes a leak. We control this with tool path strategy, not just deburring by hand afterward. A chamfer or radius written into the model is cheaper than a manual fix.

Finishes change dimensions. Anodizing adds 5–25 μm per surface depending on type. Hardcoat adds more. If a hole must stay within ±0.005 mm after anodizing, the machinist has to cut undersize on purpose. Tell your shop the final finish before they program the part.

Plating and coating also affect fit. Electroless nickel adds an even layer, but threads and press fits still shift. For parts that assemble after coating, agree on the pre-plate dimension and who owns the final measurement.

• 1
  Finish drives the cutRa 0.2–0.8 μm needs a dedicated light pass.
• 2
  Burns are dimensionalA 0.05 mm burr can fail a seal.
• 3
  Coating adds materialAnodize and plating shift dimensions by 5–25 μm per surface.