CNC calibration: how we keep peak accuracy on 127 machines
Calibration maps what a machine actually does back to what the controller thinks it is doing. This page explains the seven checks that hold a CNC machine inside ±0.005 mm, how often each one has to repeat, and the signs that no software offset will save you.
What actually changes between one CNC calibration and the next
A CNC machine is a stack of steel, cast iron and ballscrews sitting in a room that heats up and cools down every day. When a servo drives the X axis 100 mm, the controller assumes the screw turned by exactly the right amount. It did not. Thermal growth, wear on the ballscrew, and small geometric errors all pull the real position away from the commanded one.
Calibration measures that gap and writes it back into the control as compensation. A laser interferometer reads true axis position along the full travel. A ballbar traces a circle and reveals squareness, backlash and servo mismatch in one test. Each measurement ends with numbers you can act on, not an opinion.
Drift is rarely dramatic. A machine that cut a 50 mm bore to 49.998 mm in January may hold 50.006 mm in July and pass everything in between. Nothing failed. The room got warmer, the screw grew, and the error walked out of tolerance quietly. That is why CNC calibration is a schedule, not a repair.
Geometry matters as much as linear position. If Y is not square to X, every 45° wall is wrong even when each single axis reads perfect. Squareness errors scale with part size, so a small error that is harmless on a 100 mm bracket becomes visible on a 1,200 mm frame.
How often CNC calibration needs to repeat
Monthly is not a rule, it is a starting point. Frequency follows how hard the machine works and how tight the tolerance is. A 3-axis mill roughing aluminium to a ±0.05 mm print can go six months and never show a problem. A 5-axis center holding ±0.005 mm on titanium runs a short check every week.
Thermal drift pushes the schedule more than wear does. A shop that runs one shift in a temperature-controlled room sees far less movement than a shop running 24 hours with doors opening to a warm yard. We log ambient temperature next to every calibration record so we can correlate drift with conditions instead of guessing.
After any event, recalibrate regardless of schedule. That list is short: a crash, a spindle replacement, a ballscrew or bearing change, moving the machine, or a floor repair under the castings. One shifted anchor pad is enough to pull squareness out of spec.
Records matter for the customer, not just the shop. An aerospace or medical buyer wants to see that the machine condition was measured before the run started. Our reports go out on request with the last calibration date, the measured error, and the compensation applied.
When CNC calibration cannot fix the part
Calibration corrects the machine. It cannot correct the process. If a wall is tapered because the tool is deflecting, the machine is doing exactly what it was told. Compensation values will not remove the taper.
Temperature at the part is the other hard limit. A 100 mm aluminium part grows about 0.0023 mm per °C. In a shop swinging 5 °C, that is over 0.01 mm of size movement before the tool touches metal. No axis compensation covers a part that is being measured at a different temperature than it was cut at.
Fixturing and workholding sit in the same category. A vise that lifts the part 0.01 mm on clamping, or a thin wall that springs back after the cut, will produce out-of-tolerance results on a machine that just passed every check.
Fixture, tool, and thermal errors are separate budgets. When a part fails, measure each one before assuming the machine drifted. That order of thinking saves a lot of unnecessary recalibration work.
Seven checks that define a full CNC calibration
Run in this order; each one assumes the previous passed.
- 1Level and foundationRecheck level with a precision spirit level or electronic level to 0.02 mm/m. Clean pads, retorque anchor bolts, and confirm no floor cracking under the castings.
- 2Squareness between axesSweep a granite square or use a ballbar circle test. Squareness should hold within 0.010 mm over 300 mm. Correct mechanical alignment before touching software.
- 3Linear positioningLaser interferometer along full stroke, 20–30 points per axis. Record lead error, repeatability and backlash. Backlash above 0.005 mm on a finishing axis needs screw or thrust bearing work.
- 4Straightness and parallelismIndicator or laser straightness on X, Y and Z. Parallelism between spindle axis and Z travel should stay inside 0.010 mm over 300 mm.
- 5Spindle and tool taperCheck spindle runout (under 0.003 mm at the taper), taper contact with blueing (80%+ contact), drawbar force with a gauge, and thermal growth after 30 minutes at working RPM.
- 6Rotary and 5-axis geometryOn 4-axis and 5-axis machines, verify rotary centerline position and tilt axis offset. On a Ø400 mm rotary table, centerline runout should stay under 0.005 mm.
- 7Volumetric and verificationConfirm with a ballbar circle-diamond test and a known artifact or test cut. Compare against the last record. Sign off only when the machine cuts the artifact inside ±0.005 mm.
Check interval by machine type, tolerance band and duty cycle
Use this to set your own schedule; adjust for shifts per day and room temperature control.
| Machine / duty | Tolerance held | Suggested interval | Trigger for early recalibration |
|---|---|---|---|
| 3-axis mill, 1 shift | ±0.05 mm | 6 months | After any crash or spindle change |
| 3-axis mill, 3 shifts | ±0.025 mm | 3 months | Part size drifting on one axis |
| 4-axis mill-turn, 2 shifts | ±0.010 mm | 6–8 weeks | Backlash rise on the rotary axis |
| 5-axis center, light duty | ±0.005 mm | 4 weeks | Ballbar roundness worsens |
| 5-axis center, 24 h titanium | ±0.005 mm | Weekly short check | Spindle growth over 0.010 mm |
| Any machine, after a move | Any | Before first cut | Floor or anchor disturbance |
Machine error or process error?
If the error follows the axis and repeats on a test artifact, recalibrate the machine. If the error follows the part shape, the wall thickness or the batch, fix the tool, the fixture or the temperature first.
Common questions about CNC calibration
Does a new machine need calibration before its first job?
Yes. Installation alignment and a laser positioning check should happen before production, not after the first rejected batch. Transport and rigging move castings.
We treat the first calibration as the baseline record. Every later measurement is compared against it, so a missing baseline turns every future drift into a mystery.
Can software compensation replace mechanical repair?
Up to a point. Control compensation handles lead error, pitch error and small squareness offsets well.
It cannot recover lost repeatability. If backlash or thrust bearing wear makes the axis land in a different place every time, compensation just follows the noise instead of removing it.
How do we know if the error is thermal or mechanical?
Run the same test cut cold and again after 30 minutes at working spindle speed. A mechanical error stays roughly the same. A thermal error grows with run time and shrinks after the machine cools.
Log ambient temperature with each measurement. Two or three records usually make the pattern obvious.
What does a calibration report contain?
Measured error per axis, backlash, squareness, spindle runout, and the compensation values written into the control. We add the date, ambient temperature and the technician's name.
Reports go out on request with the shipment, which helps buyers who audit incoming parts.
Does calibration affect lead time?
Scheduled checks fit inside normal planning. A short weekly check on a 5-axis machine takes minutes and does not stop a job.
The exception is a crash or spindle change. That machine comes out of the schedule until it passes again, and we quote the new date rather than ship parts from a machine we have not verified.
Which certifications cover the calibration process?
Our quality system is certified to ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022.
Calibration records sit inside that system, alongside in-process monitoring and the final inspection that every part goes through before shipment.
Send us the drawing and the tolerance
Tell us the material, the tolerance band and the quantity. You get a quotation and a free DFM analysis within 12 hours, and parts ship in 3–5 days once production starts.
12-hour quote100% inspection±0.005 mmNo minimum order quantity
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
