Precision CNC parts explained

• 1
  What makes a machined part a precision CNC part
• 2
  Tolerance stacks decide where the real limit sits
• 3
  3, 4 and 5 axis: what each axis count actually buys
• 4
  Why setup count drives both accuracy and cost
• 5
  How precision is verified before parts ship
• 6
  When precision machining is the wrong answer
• 7
  Material choice changes the achievable tolerance
• 8
  Which machine type fits which part

A precision CNC part is a component cut to a tolerance band so narrow that the remaining error budget is close to the measurement uncertainty. In our shop that usually means ±0.005 mm (±0.0002 in) or tighter on the controlling dimensions. The feature itself is rarely exotic. A bore, a face, a slot. What changes is how much deviation the assembly will tolerate before it stops working.

Three things separate precision work from general machining. First, the tolerance band. Second, the geometric relationship between features, not just the size of each one. Third, repeatability across the run, so part 1 and part 500 mate the same way. A part can hit ±0.005 mm on a single feature and still fail if the bore-to-face perpendicularity drifts.

The reason engineers care is functional, not cosmetic. A bearing seat that is 0.01 mm out of round will transfer vibration. A valve body with a mismatched port angle will leak at pressure. Precision is the cost of keeping a system predictable.

Tolerance on a drawing is a per-feature number. Tolerance in an assembly is a stack. If a shaft sits in two bores with a combined positional allowance of 0.02 mm, and the mating bracket adds another 0.03 mm, the assembly can bind even though every individual part passed inspection. This is the most common reason a first article looks fine and the pilot build does not.

The practical fix is to identify the one or two dimensions that control function and hold those tight, while loosening everything that only controls fit-up or clearance. Tightening every callout on a print multiplies cost without improving the assembly. Loose callouts on non-critical features let us use faster passes and fewer setups.

For reference, our standard achievable band is ±0.005 mm, and surface finishes run from Ra 0.2–0.8 μm on fine work up to Ra 1.6–3.2 μm as-machined. A fine finish is sometimes a bearing requirement and sometimes only a sealing requirement. Those two cases justify very different process choices.

A three-axis mill moves the tool in X, Y and Z. The part stays in one orientation. That is enough for plates, brackets, pockets and most prismatic work, and it is the cheapest way to make them. The limit appears when a feature faces a direction the spindle cannot reach without re-fixturing the part.

A four-axis machine adds rotation around one axis, usually A. The part can be indexed or turned while the tool cuts, which suits cylindrical bodies, cam profiles, and parts where holes must be drilled at several angles around a centerline. A Ø400 mm rotary table covers most of this work. It is a strong middle option when the geometry is rotational but not free-form.

A five-axis machine adds a second rotary axis, so the tool can approach the part from nearly any angle in one setup. The gain is not speed for its own sake. It is fewer setups, fewer fixtures, and no accumulated re-datum error. On a part with five angled faces, three-axis work may need four setups; five-axis does it in one. Each removed setup removes a chance to lose position.

Every time a part comes off a fixture and goes back on, you re-establish a datum. That re-datum introduces error, and the error is not random: it stacks. A part made in four setups carries four chances for a 0.01 mm shift. The same part in one setup carries one. This is the single biggest reason complex geometry migrates to five-axis.

Setup count also sets the labor content. Fixture build, first-article check, and in-process verification repeat with each op. For low-volume runs, the setup time can exceed the cutting time. For high-volume runs it amortizes, but the accuracy penalty does not disappear.

The practical question during review is simple: how many distinct tool approach directions does this part have? Count them. If the number is one or two, three-axis tooling is usually the right call. If it is four or more, or if the faces are compound angles, five-axis almost always wins on total cost.

A tolerance is only real if it is measured. Our flow is raw material check, in-process monitoring, then final inspection on 100% of parts before shipment, with reports available on request. For tight features, that means CMM verification of the controlling dimensions and the geometric callouts around them.

Temperature matters more than most people expect. Aluminum expands roughly 23 μm per meter per °C. A 200 mm aluminum part measured 5 °C above the calibration temperature can read about 0.023 mm long. That is four times a ±0.005 mm band. Good shops measure at controlled temperature, or correct for it, and say which one they did.

Surface finish is verified separately. A stylus or optical check on a Ra 0.8–1.6 μm requirement is routine. Pushing to Ra 0.2–0.8 μm usually means a finishing pass with a smaller stepover, a sharper tool, or a different operation entirely. It is a process decision, not a polishing afterthought.

Not every tight-tolerance part should be milled. If the geometry is a thin shell under 1 mm wall with large unsupported spans, chatter and distortion will fight you on every pass. Sometimes a casting plus finish machining holds the shape better, and sometimes sheet metal fabrication is the honest answer. We quote all three, so the comparison is real.

If the annual volume is high and the material is aluminum or zinc, die casting followed by a light finish cut often beats cutting the whole part from billet. The machining budget moves to the two or three faces that actually need tolerance. Everything else comes out of the mold at near-net shape.

Deep internal channels with no straight-line access are a different problem. Additive processes can build geometry that no end mill can reach. The trade is surface finish and material properties. A machined sealing face on a printed body is a common hybrid, and it works well when the print is treated as near-net stock.

Aluminum 6061-T6 is the default for precision work because it cuts clean, holds dimension, and is stable after stress relief. 7075 gives higher strength but is more prone to movement after heavy stock removal, so rough and finish passes may need to be separated. 2024 behaves similarly and is common in aerospace brackets.

Stainless 303 and 304 machine well and are widely used for shafts and housings. 17-4PH (SUS630) holds tolerance after heat treatment, which makes it a frequent choice for parts that need both corrosion resistance and strength. 316L is common in medical and food-contact work.

Titanium Ti-6Al-4V and Inconel are cut at much lower surface speeds. Tool wear is fast, and the heat goes into the cutter rather than the chip. Tolerances are still achievable, but cycle times rise and the process window narrows. PEEK and POM move with temperature and moisture, so they need different inspection timing than metals.