What Is Repeatability in CNC Machines?

• 1
  Repeatability vs accuracy, in plain numbers
• 2
  Where the number comes from inside the machine
• 3
  How repeatability is measured and what the spec means
• 4
  When repeatability is not the limiting factor
• 5
  What repeatability means for your drawing
• 6
  How to check a supplier's repeatability claim
• 7
  What the symptom tells you

Put a machine at the same commanded position 30 times and log where the tool actually went. The spread of those points is repeatability. The distance from the average point to the position you asked for is accuracy. They are separate numbers, and a machine can be good at one and poor at the other.

A simple case: the control calls for X100.000 mm. The tool lands at 100.030, 100.031, 100.029, 100.030 mm over four cycles. Repeatability is about 0.002 mm. Accuracy is off by 0.030 mm, because the whole cluster sits in the wrong place. The machine is repeatable but not accurate.

Reverse it and you get the more dangerous case. The tool lands at 100.000, 100.018, 99.982, 100.020 mm. The average is right on target, so the machine looks accurate on a single check, but the spread is 0.038 mm. Parts made on it will not interchange.

For a production run, repeatability decides whether part 1 and part 500 are the same. Accuracy decides whether both of them meet the drawing. You need both, and they are checked with different procedures.

• 1
  RepeatabilitySpread of results when the same command is repeated.
• 2
  AccuracyOffset between the average result and the nominal value.
• 3
  Why it mattersInterchangeability across a run is a repeatability problem.

A CNC machine is a chain of position loops. The control sends a command, the servo drive turns the motor, the ball screw converts rotation into linear travel, and the linear scale or encoder reports back. Every link in that chain has its own variation, and they add up.

Backlash is the classic one. When the axis reverses direction, the screw and nut have to take up a small gap before motion starts again. On an older or worn machine this can reach 0.02–0.05 mm, and it shows up as a step at every direction change.

Thermal growth is the slow one. A spindle running at 12,000 rpm for two hours pushes heat into the housing and the ballscrew. Steel grows about 11 μm per meter per °C. On a 1,000 mm axis, a 5 °C rise moves the zero point roughly 0.055 mm. That is why warm-up matters.

Then come the small ones: servo loop tuning, encoder resolution, guideway preload, and chip build-up on a locating face. None of them is large alone. Together they set the floor for what the machine can hold.

• 1
  BacklashLost motion at direction reversal, 0.02–0.05 mm when worn.
• 2
  Thermal driftAbout 11 μm per meter per °C on steel.
• 3
  Servo tuningPoor gain causes overshoot and hunting at the stop point.

ISO 230-2 is the standard most builders quote against. The machine is commanded to a series of target positions, approached from both directions, usually 5 to 7 targets and 5 to 10 repeats each, at a controlled temperature. The result is reported as a bidirectional positioning deviation and a unidirectional repeatability range.

Read the spec carefully. A catalog number such as ±0.002 mm is often a unidirectional repeatability figure at 20 °C, on a new machine, with the spindle cold. Real numbers on the shop floor are larger. That is not dishonesty, it is a different measurement.

If you want to compare machines, compare like with like. Ask for the standard used, the number of targets, the approach direction, and the ambient temperature. A number without those four items cannot be compared to anything.

In our own shop, we run a ballbar check and a laser interferometer check on the five-axis centers on a schedule, and we re-check after any crash or spindle swap. The tolerance we hold on production parts is ±0.005 mm, which sits well above the machine's own repeatability floor.

• 1
  ISO 230-2Standard test for positioning accuracy and repeatability.
• 2
  Ask for conditionsStandard, target count, direction, temperature.
• 3
  Production tolerance±0.005 mm on parts, tighter than the machine floor.

Repeatability gets blamed for problems it did not cause. If every part is offset in the same direction by the same amount, the machine is repeatable and something else is wrong: a fixture stop packed with chips, a tool that grew after a coating change, or a program zero set on the wrong face.

Cutting force is another limit. A thin-wall aluminum housing at 3 mm wall thickness will deflect under a 12 mm end mill no matter how good the position loop is. The machine returns to the same point; the workpiece moves away from it. That is a fixturing and toolpath problem.

Tool wear is the third. A carbide insert loses 0.02–0.05 mm of size over a few hundred parts in steel. The machine holds position perfectly. The edge does not. Offset the tool on a count, not on a hunch.

So the honest answer is: repeatability sets the floor, and process control sets the ceiling. Improving one does nothing if the other is the binding constraint.

• 1
  Constant offsetPoints to setup, not to the machine.
• 2
  DeflectionLight walls and long tools move under load.
• 3
  Tool wear0.02–0.05 mm over a run in steel; offset on count.

A tight tolerance on a single feature is a different request from a tight tolerance across a batch. If you call out ±0.010 mm on one bore and order 500 pieces, you are asking for a repeatable process, not just a capable machine. The shop has to hold the same result 500 times.

Position tolerance across multiple features is harder still. Two holes 300 mm apart, both within ±0.010 mm of nominal, need the axis to be consistent over that travel. Thermal drift over a two-hour run works directly against this.

Datum choice matters. If the drawing datum is a rough cast surface, the machine can be perfect and the part will still be out, because the datum itself moves from part to part. Pick a machined datum when the tolerance is tight.

For a first article, ask for the inspection report and check the spread, not just the pass or fail. Three parts in tolerance at the same offset tell you the process is stable. Three parts in tolerance at opposite ends of the band tell you it is not.

• 1
  Batch toleranceA repeatability requirement, not a one-off check.
• 2
  Long spansThermal drift over 300 mm of travel shows up.
• 3
  DatumMachined datums make tight tolerances achievable.

Ask what machine does the job, not what the shop owns. A shop with 127 machines may still run your part on the oldest one. Get the model and the axis count, then look up the spec sheet.

Ask how the first article is verified. A CMM report with the measured values, not just the nominal and tolerance, tells you the spread. Reports on request is a normal answer. A refusal is not.

Ask what happens after a crash or a spindle change. Any shop that re-checks geometry after those events is managing repeatability. Any shop that does not is hoping.

And ask about warm-up. A five-axis center that starts cutting cold and holds ±0.005 mm at hour three is either warmed up on purpose or lucky. We run spindles through a warm-up cycle before the first tight-tolerance cut.

• 1
  Machine modelThe spec sheet is public; ask which machine runs the part.
• 2
  Inspection dataMeasured values, not pass/fail only.
• 3
  Warm-upA stated warm-up routine is a good sign.