How to Demonstrate the Accuracy of a CNC Machine
A claim of ±0.005 mm means nothing until it is measured. This guide shows the test sequence our engineers run to demonstrate the accuracy of a cnc machine, what each instrument proves, and when a test stops being worth the setup time.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
- 7
What actually proves machine accuracy
What accuracy of a cnc machine actually measures
Accuracy of a cnc machine is the difference between the position the control commanded and the position the tool actually reached. That is a single number per axis over a defined travel, and it is not the same as resolution or repeatability. Resolution is the smallest increment the control can command, often 0.001 mm or finer. A machine can resolve 0.001 mm and still land 0.03 mm away from target.
Repeatability is the spread of results when the machine returns to the same point many times. It is usually tighter than accuracy, often by a factor of three to five. A machine with ±0.005 mm accuracy may repeat to ±0.0015 mm. When a drawing calls for a ±0.005 mm tolerance, both numbers matter, but repeatability governs whether a production run holds the band.
There is a third number engineers forget: geometric accuracy. Squareness between X and Y, parallelism of the spindle axis to Z, and straightness along a 4,000 mm travel all add to the final position error. A laser measures them, but only if the setup is aligned to the machine axes, not to the floor.
Finally, thermal state defines the baseline. Cast iron and steel grow roughly 11–12 μm per meter per °C. A spindle that warms 8 °C during the first hour shifts the tool point more than 0.02 mm on a large machine. Test cold, and you prove nothing about the afternoon.
- 1Positional accuracyCommanded point vs. actual point, per axis, over full travel.
- 2RepeatabilitySpread of returns to one point; typically 3–5× tighter than accuracy.
- 3Geometric accuracySquareness, straightness, parallelism; adds directly to position error.
- 4Thermal driftRoughly 11–12 μm/m per °C on steel and cast iron structures.
Which instruments demonstrate the accuracy of a cnc machine
A laser interferometer is the reference method for linear positioning. A stabilized beam fires from the head to a reflector mounted on the table or spindle, and the system reads displacement to sub-micron resolution. A full test on a 1,000 mm axis takes 1–2 hours and produces a plot of error versus position, plus backlash and reversal values. This is the data a customer should see when a supplier claims tight positioning.
A ballbar test runs a 100–300 mm circular path while a telescoping bar with a precision encoder measures radial deviation. It takes 10–15 minutes per circle and exposes servo mismatch, scale errors, and squareness problems in one trace. Roundness of the polar plot is the quick read; the numeric diagnosis separates backlash from following error.
Spindle error motion is measured with a master ball and a capacitance or inductive probe, usually per ISO 230-7. Taper runout alone does not cover it, because thermal growth and bearing preload change the error vector at speed. Run the spindle at production rpm for 30 minutes, then measure.
For the whole chain, cut a test artifact. A stepped pyramid, a bored circle, or a NAS 979-style circle-diamond-square part exercises interpolation, tool change, and thermal behavior together. Measure it on a CMM and compare the results with the laser data. If the two disagree, the machine setup is wrong, not the instrument.
- 1Laser interferometerLinear positioning, backlash, straightness. 1–2 hours per axis.
- 2BallbarCircularity, servo mismatch, squareness. 10–15 minutes per circle.
- 3Spindle analyzerError motion at speed after a 30-minute warm-up.
- 4Cut artifact + CMMEnd-to-end proof of the machine, tool, and program together.
Conditions that decide whether the numbers are real
Temperature is the first thing to control. ISO 230-3 recommends a stabilized environment, and in practice that means 20 ± 1 °C with the machine soaked for at least 12 hours. A machine moved into the room yesterday will still be settling. Measuring before the soak produces data that looks great and fails on the first production day.
Preload and clamping come next. Test with the fixtures and vise you will use in production, or state clearly that the test was run bare. Clamping force moves a thin plate; a 0.5 mm deflection at the vise jaws shows up as a position error far from the contact point.
Compensation settings must be frozen and recorded. Pitch error compensation, backlash compensation, and thermal compensation all change the raw numbers. Note the parameter values in the report, and test with them active, because that is how the machine runs. Turning compensation off to get a cleaner plot defeats the purpose.
Finally, document the probe and the calibration chain. A laser system traceable to a national standard, with a calibration certificate within the last 12 months, is what makes the result defensible. An uncalibrated dial indicator on a magnetic base is not evidence.
- 1Soak 12 hoursMachine and workpiece at 20 ± 1 °C before any measurement.
- 2Production fixturingTest with the vise or fixture, or declare the test as free-standing.
- 3Freeze compensationRecord pitch, backlash, and thermal parameters in the report.
- 4Traceable calibrationLaser certificate within 12 months; note the reference standard.
Where accuracy demonstrations go wrong
The most common failure is testing a cold machine and publishing the number. A cold spindle sits near ambient, and everything looks tight. Two hours into a production run the ballscrew and spindle grow, and the same program drifts out of tolerance. Always report the thermal state alongside the measurement.
The second mistake is measuring a small envelope. A laser test over 200 mm of a 1,000 mm axis says nothing about the far end, where ballscrew sag and pitch error accumulate. Test over the travel you actually use, and state the tested length in the report. A ±0.005 mm result over 200 mm is not a ±0.005 mm machine over 1,000 mm.
Third, mixing accuracy and repeatability in one figure. Buyers see a single number and assume it covers both. Report positional error, backlash, and repeatability as three separate values, each with its test length and feed rate.
Last, ignoring the workpiece. Machine accuracy sets the floor, but tool deflection, fixturing, and material springback add on top. A 4 mm end mill in aluminium with a 20 mm overhang deflects measurably under a 0.1 mm depth of cut. Prove the machine, then prove the process.
- 1Cold test, hot claimReport ambient and spindle temperature with every plot.
- 2Short travel testState the tested length; pitch error grows with distance.
- 3One number for two metricsSeparate positioning, backlash, and repeatability.
- 4Machine onlyTool and fixture errors sit on top of machine error.
How to demonstrate the accuracy of a cnc machine, step by step
- 11. Clean and level the machineWipe ways, remove chips, check level to 0.02 mm/m on both axes. A machine out of level twists the bed and skews every later number.
- 22. Run a 30–60 minute warm-upExercise all axes and the spindle at production rpm. Cold starts understate thermal error by more than half.
- 33. Measure linear positioning with a laserTest each axis over full travel in 20–50 mm increments, three runs forward and back. Record backlash and reversal values separately from positioning error.
- 44. Run a ballbar circle testUse a 150–300 mm radius at 300–1,000 mm/min feed. Three circles at different feeds separate servo tuning from geometry errors.
- 55. Check spindle error motion at speedMount a master ball in the taper, measure with a capacitive probe after warm-up. Report radial and axial values separately.
- 66. Cut and measure a test artifactMachine a stepped pyramid or circle-diamond-square part, then measure on a CMM. Compare against the laser data for agreement within 0.005 mm.
- 77. Repeat the laser test coldShut down, let the machine cool 4 hours, and re-run one axis. The gap between cold and hot readings is your thermal drift figure.
- 88. Issue the report with raw dataInclude plots, compensation parameters, probe calibration, and ambient temperature log. Summary sheets without raw files are not evidence.
Which test to use for which claim
Match the instrument to the question the customer is asking.
| Test method | What it proves | Typical time | Best for |
|---|---|---|---|
| Laser interferometer | Linear positioning, backlash, straightness | 1–2 hours per axis | Large travel, tight ±0.005 mm claims |
| Ballbar circle test | Circularity, servo mismatch, squareness | 10–15 minutes | Fast health check, servo tuning |
| Spindle error motion | Radial and axial error at production rpm | 30–45 minutes | Fine boring, high-speed finishing |
| Cut artifact + CMM | Whole process: machine, tool, program | 2–4 hours | Customer acceptance, first article |
| Dial indicator sweep | Basic runout and squareness only | 5–10 minutes | Pre-shift check, not a claim |
| Temperature log | Thermal drift over a shift | 4–8 hours | Long-cycle parts, large frames |
Prove the machine, then prove the process
A laser plot proves positioning. A cut artifact proves the whole chain. Ask for both, plus the thermal state, before you accept any accuracy claim.
Questions engineers ask about accuracy testing
How often should a CNC machine be accuracy tested?
For production machines holding ±0.005 mm, a laser check every 6–12 months is typical, with a ballbar check quarterly. After any crash, spindle replacement, or move, test before releasing the machine back to production.
A quick ballbar circle once a month catches servo drift early and costs 15 minutes. Full laser reports are usually tied to customer audits or first-article submissions.
Can I demonstrate accuracy without a laser interferometer?
Yes, but the claim is weaker. A ballbar plus a cut artifact measured on a CMM covers geometry and the full process chain. What you lose is the per-axis linear positioning plot and the reversal values.
If the customer only needs process capability, the artifact is often the more convincing evidence, because it includes the tool and program. If they ask for positional accuracy over 1,000 mm, you need the laser.
What ambient temperature should the test room hold?
20 ± 1 °C is the working target for a machine claiming ±0.005 mm. The machine should soak at that temperature for 12 hours minimum before measurement.
Where the shop cannot hold ±1 °C, use 20 ± 2 °C and state it in the report. The number is less important than recording it, because it explains variation between test days.
Does compensation hide a worn machine?
Pitch error compensation corrects a repeatable error, so it genuinely improves positioning on a healthy machine. It cannot fix a ballscrew with random backlash or a spindle with growing runout.
A good report shows the compensated result and the raw error before compensation. If the raw error grows year over year, the screw or the bearings are wearing and the compensation table is just masking it.
How do I compare two machines that were tested differently?
Ask for four things: tested travel length, feed rate, thermal state, and whether compensation was active. Without those, two plots cannot be compared.
Re-run the same test program on both machines under the same conditions. A shared artifact cut on both, measured on the same CMM, settles the argument faster than any datasheet.
What accuracy can GreatLight hold on production parts?
Our shops hold ±0.005 mm on qualified features, with fine finishes from Ra 0.2–0.8 μm when the drawing calls for it. We run 127 high-precision CNC machines, including 16 simultaneous 5-axis centers and a Ø400 mm rotary table, and inspect 100% of parts before shipment.
Inspection reports are available on request. Upload a drawing and we return a quotation with DFM feedback within 12 hours.
Send us a drawing and see the numbers behind the claim
Upload your part and we return a quotation, DFM feedback, and the inspection method we will use, within 12 hours.
12-hour quote100% inspectionNDA on request±0.005 mm capability
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
