Machining accuracy CNC: what actually controls it

• 1
  Accuracy, repeatability and resolution are three different numbers
• 2
  Where the machine itself sets the floor
• 3
  Heat is the slowest and largest error source
• 4
  Tool wear, deflection and chip load
• 5
  Setup, workholding and measurement error
• 6
  Tolerance ranges and what they cost in process

People use machining accuracy CNC as one phrase, but the shop floor treats it as three separate numbers. Accuracy is how close the tool lands to the commanded position. Repeatability is how close it lands to the last part it made. Resolution is the smallest increment the control can command, often 0.001 mm or finer.

A machine can be repeatable without being accurate. If the ballscrew is mapped wrong, every part comes out 0.03 mm too deep, but part 2 matches part 1 exactly. That is a calibration problem, not a capability problem, and a laser interferometer test fixes it.

The reverse is worse. A machine that averages the right size but scatters ±0.02 mm part to part cannot be corrected by offsets. That scatter is what eats your tolerance band, and it is why we quote capability per feature rather than per machine.

Resolution is mostly marketing. A 0.0001 mm scale on a machine with 0.01 mm thermal drift buys you nothing. Ask for the repeatability figure measured on a test cut, not the encoder spec sheet.

• 1
  AccuracyCommanded position vs. actual position, checked with a laser or ballbar.
• 2
  RepeatabilitySpread across a run of identical parts; the number that sets your real limit.
• 3
  ResolutionSmallest command step. Useful only if the machine is stable enough to use it.

Rigidity comes before control. A 500 × 310 × 200 mm compact travel machine with linear guides and a Ø400 mm rotary table behaves very differently from a 4,000 × 400 × 150 mm gantry. Long travels amplify every small error: a 0.005 mm per 300 mm pitch error compounds across a 2 m part.

Spindle condition matters more than most people expect. Runout of 0.005 mm at the taper becomes 0.005 mm on the wall of every bore, plus the cutting force error it induces. We check spindle runout on a schedule and retire spindles before they drift past the tolerance band they are assigned to.

Axis count changes how many setups a part needs, and each setup is a fresh chance to lose 0.01 mm. A 16-station simultaneous 5-axis setup can cut five faces in one clamping. That is often worth more than a marginally stiffer 3-axis machine.

• 1
  Travel sizeLonger axes mean larger accumulated positioning error.
• 2
  Spindle runoutShows up directly in bore diameter and wall thickness.
• 3
  Axis countFive-axis cuts reduce setups, which protects position tolerance.

A machine warms up over the first two to three hours of a shift. Ballscrews, spindles and the frame all grow at different rates, so the relationship between commanded and actual position drifts. On a 750 × 1,150 × 550 mm machine, that drift can reach 0.02 mm before the castings stabilize.

The fix is boring but effective: warm-up cycles, controlled coolant temperature, and stable shop temperature. We run a warm-up program before first cut and avoid opening loading doors to outside air on cold mornings. Parts measured right off the machine can read 0.01 mm different once they cool.

Material matters here too. Aluminium 6061 and 7075 move roughly twice as much per degree as stainless 304. A thin aluminium rib can distort from cutting heat alone, then relax after clamping is released. Rough, rest, then finish is the usual answer.

If your drawing calls ±0.005 mm on a 300 mm aluminium frame, ask whether it will be measured at 20 °C or on the shop floor. The answer changes the process.

Every cutter deflects under load. A long 6 mm end mill cutting 6061 at full depth may bend 0.02 mm at the tip, and the wall it leaves is that much thin. Short tools, light radial engagement and a finishing pass at 0.2 mm depth keep the deflection predictable.

Wear is gradual, so the last part of a run is usually the worst one. On a 10,000 part run we set tool-change intervals from measured wear rather than a fixed clock, and we keep a first-article and a last-article measurement for the record.

Chip load is the other lever. Too low and the tool rubs, work-hardens stainless and burns the edge. Too high and the finish degrades. For 316L we typically run 0.05–0.08 mm per tooth finishing, with plenty of coolant. For PEEK, air blast beats flood coolant because moisture absorption changes the part after machining.

Tool holding matters as much as the tool. A worn collet with 0.01 mm runout doubles the effective error on a small cutter. We replace holders on a schedule.

• 1
  DeflectionGrows with tool length; keep length-to-diameter under 4:1 for finishing.
• 2
  WearTrack it by part count and measured size, not by hours.
• 3
  Holder runoutAdds directly to the feature tolerance; check with a dial indicator.

Clamping force bends parts. A thin plate held in a vise bows, gets machined flat, then springs back when released. Soft jaws machined to the part profile, or a vacuum fixture for thin walls, removes most of this. For thin-walled aluminium housings we often leave 0.3 mm and finish after stress relief.

Datum strategy drives the final number. If a drawing dimensions hole positions from a face that is machined second, the tolerance chain includes the first setup. Calling out datums that reflect the functional assembly usually lets the shop hold a looser, cheaper tolerance.

Measurement is part of the error budget. A caliper on a curved surface reads differently from a CMM touch point. We run 100% inspection before shipment with raw material checks, in-process monitoring and final inspection, and reports are available on request. If your incoming inspection uses a different method, agree on the method before the first run.

One more thing. A part that measures well on a cold CMM can still fail in assembly if the functional requirement is fit, not size. Tell us the fit.