Precision CNC: how accuracy is actually made

• 1
  Precision CNC setup count, not spindle speed, controls your cost
• 2
  What simultaneous 5-axis motion actually changes
• 3
  Where ±0.005 mm comes from and where it does not
• 4
  Choosing between milling, turning, and mill-turn
• 5
  Inspection is part of the process, not a final gate
• 6
  Matching the process to the part

A machine tool does not care how complicated your drawing looks. It cares how many times the part has to be unclamped, turned over, and re-zeroed. Every additional setup adds a new datum, and every new datum adds stack-up error. On a 3-axis mill, a part with features on four faces may need three or four setups. On a precision CNC 5-axis center with a Ø400 mm rotary table, the same part can often be finished in one or two.

This is why two quotes for the same part can differ by a factor of three. The metal is not the variable. The number of times a human touches the part is the variable. A single setup also means the feature-to-feature relationships stay locked to one coordinate frame, so a bore and the face it sits on do not drift apart.

The practical rule: count the faces that carry toleranced features. One or two faces, and a 3-axis machine with a good vise is usually the cheapest correct answer. Three or more, and you should be pricing simultaneous 5-axis work instead of a chain of fixtures.

Setup reduction is also the main reason to combine turning and milling on one platform. A mill-turn center with 16 stations can finish a shaft and its cross-drilled flange without a second chucking, which removes the runout error that comes from re-gripping a finished diameter.

A 3-axis machine moves the tool in X, Y, and Z while the part stays still. The tool always approaches along one fixed direction, so any undercut, deep pocket wall, or angled port has to be reached by tilting the part instead. A precision CNC 5-axis center adds two rotary axes, so the tool axis can stay normal to a curved surface while it cuts.

The benefit is not just reach. Keeping the tool normal to the surface lets you use the side of the cutter on contoured walls instead of the tip. That spreads the cutting load, holds a better surface finish, and lets a shorter, stiffer tool do the job. Short tools deflect less, which is where the last few microns of tolerance come from.

The trade-off is programming and verification time. A 5-axis toolpath has to be checked for collisions between the holder, the table, and the part. Complex parts may need a stock model and a full simulation before the first cut. That front-loaded effort is why 5-axis work pays off on geometry, not on simple prisms.

There is also a rigidity limit. Rotary axes are stacked, so the further the part sits from the trunnion, the more leverage the cutting force has. Long thin parts on a 5-axis table can chatter even when the same part is stable in a 3-axis vise.

A tolerance on a drawing is a promise about the finished part, not about the machine. A precision CNC shop can hold ±0.005 mm (±0.0002 in) on a dimension that is reachable and measurable. It cannot hold that on a 300 mm thin-wall pocket that moves as the material is removed, or on a dimension measured from a rough cast surface.

Three things set the floor. First, thermal state: a part that warms during roughing and is measured hot will read differently when it cools. Second, workholding: a vise can distort a thin wall by more than the tolerance before the cutter ever touches it. Third, metrology: if the shop cannot measure a feature to one fifth of the tolerance, the number on the drawing is a guess.

This is why a good DFM review asks which dimensions are functional. A bearing bore at ±0.005 mm is worth the cost. A clearance hole at ±0.2 mm is not. Tightening every dimension on a print does not improve the part, it just moves money from the features that matter to the ones that do not.

Surface finish follows the same logic. Ra 0.8–1.6 μm is a normal machined finish. Ra 0.2–0.8 μm usually needs a finer step-over or a secondary operation, and it should be specified only on sealing or sliding surfaces.

The decision is driven by the part's dominant shape, not by the shop's favorite machine. If most of the material is removed from a block and the critical features are prismatic or contoured, precision CNC milling is the base process. If the part is a body of revolution, turning is faster and holds roundness more easily because the part spins on its own axis.

Parts that are round but carry off-axis features, like a shaft with a milled flat or a cross-drilled flange, sit in the middle. They can be done on a lathe with live tooling, on a mill with a rotary table, or on a mill-turn center. Mill-turn is usually the best answer when the off-axis features are toleranced relative to the turned diameter, because everything is cut in one chucking.

Plate and frame parts are a separate case. When stock is already close to final thickness and most features are through-cuts, a 3-axis machine with a large travel envelope handles the job with less setup than a 5-axis center. That keeps the hourly rate down and the schedule short.

One more boundary: very deep cavities and long slender bores push you toward EDM or a specialized boring process. Milling can reach them, but tool deflection and chip evacuation start to fight the tolerance.

A dimension that is checked only after the last operation is a dimension you cannot control. In-process monitoring is what keeps a run stable. Probing on the machine after roughing tells you where the stock actually sits, so the finishing pass can be offset to match. That is how a casting with variable stock still ends up on nominal.

For tight features, the measurement method has to match the tolerance. Calipers are fine for ±0.1 mm. A bore at ±0.005 mm needs a bore gauge or a CMM with a known probe tip and a controlled temperature. If the inspection method and the tolerance do not agree, the report is noise.

Material traceability runs alongside dimensional checks. A raw material certificate lets you tie a finished lot back to a heat number, which matters for aerospace and medical work. Final inspection before shipment, with reports on request, closes the loop.

The honest limit: a first article proves the process was correct at that moment. It does not prove it stays correct for 10,000 parts. That is why in-process checks are spaced across the run rather than clustered at the start.