What Is Repeatability in a CNC Machine?

• 1
  Key takeaways
• 2
  What repeatability in a CNC machine actually measures
• 3
  The mechanical chain that decides repeatability
• 4
  How to measure repeatability without a metrology lab
• 5
  When repeatability eats your tolerance
• 6
  Where repeatability decides the outcome
• 7
  How we hold repeatability on the floor
• 8
  What each error source looks like on the machine

Repeatability in a CNC machine is the spread of end positions when the same axis is commanded to the same coordinate again and again. Command X100.000 mm fifty times and you get fifty slightly different arrivals. The width of that cluster, usually expressed as a plus-or-minus band or a total range, is the repeatability figure.

Accuracy is a different number. Accuracy tells you how far the average arrival sits from the nominal coordinate. A machine can be accurate and poorly repeatable, or repeatable and consistently offset. The second case is common and easy to fix: adjust the compensation parameter and the whole cluster slides onto the target.

This is why machine builders quote positioning accuracy and repeatability as two separate specifications. On a well-maintained vertical mill, bidirectional repeatability typically falls between 0.002 mm and 0.006 mm. The accuracy figure might be three to five times larger before compensation is applied.

The distinction matters because the two errors behave differently in production. A constant offset produces every part shifted by the same amount, which a fixture offset or tool offset can absorb. Scatter produces parts that wander inside the tolerance band, and no single offset can pull them back.

• 1
  RepeatabilityThe width of the landing zone after many identical moves.
• 2
  AccuracyHow close the center of that zone sits to the commanded point.
• 3
  ResolutionThe smallest step the encoder or scale can report. It caps both figures.

Motion travels through a chain: servo motor, coupling, ball screw, nut, linear guide, saddle, column, spindle. Every joint in that chain can absorb or release a few microns. The ball nut is usually the first suspect. Preload removes axial play, but preload is a spring force, and springs relax.

Backlash appears when the machine reverses direction. Command a move in plus X, then minus X, and the tool may not start moving until the control has taken up 0.003 mm of lost motion. On a machine with a worn nut or a loose thrust bearing, that lost motion is unpredictable and therefore hurts repeatability.

Thermal growth is the second big driver. A spindle running at 12,000 rpm for two hours can grow 20–40 μm in Z. The ball screw grows too, roughly 11 μm per meter per degree Celsius for steel. On a 1,000 mm travel, a 3 °C rise shifts the far end of the axis by about 0.033 mm.

Servo tuning sits on top of the mechanical chain. A loop that is too soft leaves following error and overshoots on reversal. A loop that is too stiff excites the structure and produces a short ring at every stop. Both show up as scatter when you measure with an indicator.

The shop-floor test is simple. Mount a dial indicator or a lever gauge on the table so the stylus touches the spindle nose or a gauge pin in the holder. Command the same position fifty times and record every reading. The peak-to-peak spread is your practical repeatability.

Do it bidirectionally. Approach the point from plus X and from minus X in alternating cycles. The difference between the two clusters is backlash. Run the test cold, then run it again after a 30-minute spindle warm-up, and compare. If the spread grows after warm-up, you are looking at thermal or preload behavior, not at the control.

For a fuller picture, a ball-bar or a laser interferometer traces the whole travel and reports straightness, squareness and reversal error. Most job shops do not need that daily. A 50-cycle indicator test on each axis, run monthly, catches most degradation early.

One rule keeps the data honest. Measure at the tool, not at the position readout. The control reports where it thinks the axis is. The indicator reports where the axis actually arrived. Those two numbers diverge exactly when repeatability is failing.

• 1
  Cold and warmAlways record both. The delta tells you more than the absolute value.
• 2
  Same direction, then reversedTen cycles each way, minimum, before you trust the numbers.

A tolerance band is a budget shared by several contributors: machine repeatability, thermal drift, tool wear, fixture error and material springback. If the drawing calls for ±0.025 mm and repeatability already consumes ±0.008 mm, you have roughly ±0.017 mm left for everything else. That is workable. If repeatability consumes ±0.020 mm, the job is fragile.

The rule of thumb we use is that machine repeatability should stay under one third of the total tolerance. Below that, the process is stable and inspection is mostly confirming what you already expect. Above one third, every part becomes a decision, and scrap rates climb without an obvious cause.

Tight features feel this first. A bore with a positional tolerance of Ø0.05 mm at MMC, a bearing seat with an interference fit, or a mating face that must sit flat within 0.01 mm. These are the features where scatter shows up as assembly problems rather than as a rejected dimension on a report.

For looser work, repeatability barely enters the conversation. A bracket with ±0.2 mm on hole position will pass on almost any maintained machine. Spending money on a tighter machine for that part buys nothing.

Medical instruments and implants live at the tight end. A surgical guide that must align with a patient-specific model has no room for scatter, and the material is often 316L or titanium, which is unforgiving to cut. Repeatability in the machine keeps the second, tenth and hundredth piece identical to the first.

Automotive and EV parts sit in the middle. Transmission housings, inverter plates and battery tray interfaces carry position tolerances in the 0.02–0.05 mm range across thousands of parts per run. The first part is easy. The last part is the test.

Aerospace brackets and structural fittings combine tight position with thin walls. Thin walls deflect under cutting force, so the machine must hold the same path every time or the deflection changes and the wall thickness varies. Robotic and automation components follow a similar pattern, with bores that must accept bearings without selective fitting.

Across all of them, the engineering meaning is the same. Repeatability is what makes a process predictable. Once a process is predictable, inspection can sample instead of screen, and the quote can reflect the real cost.

We run 127 high-precision CNC machines across three wholly-owned plants, including 16 simultaneous 5-axis machining centers and 16 mill-turn centers. Every machine follows a warm-up routine before the first production cut. Spindle and axis temperatures are logged, and a machine that drifts outside its band is pulled for service before the next run starts.

Inspection is 100 percent before shipment. That covers raw material verification, in-process monitoring and a final check, with reports available on request. For repeat jobs we compare the first article against the last article of the run, which is where repeatability problems become visible.

Tolerances are held to ±0.005 mm where the drawing requires it, with surface finishes from Ra 0.2–0.8 μm on fine work up to Ra 1.6–3.2 μm as-machined. Materials range from 6061 and 7075 aluminium to 17-4PH stainless, Ti-6Al-4V and Inconel, plus engineering plastics such as POM and PEEK.

The plants hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022. For engineering teams, those certificates matter less than the process behind them: documented warm-up, scheduled ball screw checks and a calibration log per machine.