Machine tool accuracy checks

Seven ways of detecting CNC machine tools accuracy

A shop-floor guide for engineers and maintenance planners. We cover what each test actually measures, the order to run them in, and how to judge whether a machine is still fit for the tolerances on your drawing.

Get Instant Quote Talk to an Engineer

Laser interferometryBallbar circular testISO 230 referenceThermal drift checks

Detecting CNC machine tools accuracy: seven ways to check positioning accuracy

On this page

1
What matters before you touch a gauge
2
Why detecting CNC machine tools accuracy needs a defined method
3
Prepare the machine and the environment first
4
Reading the numbers without overreacting
5
How often should you run these checks?
6
The seven detection methods, in the order to run them
7
Which method finds which error

Key takeaways

What matters before you touch a gauge

One test never tells the whole storyGeometry, positioning, and dynamic error fail for different reasons and need different instruments.

Run tests cold, then warmTake a baseline at 20 °C, run the spindle for 30–60 minutes, then repeat the same test.

Squareness and straightness come firstPositioning numbers on a skewed machine are meaningless until geometry is corrected.

Log every result with a dateTrend lines over 6–12 months reveal wear long before a part fails inspection.

Section 1

Why detecting CNC machine tools accuracy needs a defined method

A machine that cuts a good first article is not necessarily accurate. It may simply be running a toolpath that hides its errors. Detecting CNC machine tools accuracy means measuring the machine itself, not the part, against a known reference under controlled conditions.

Three error families matter on a vertical machining center. Geometric error covers squareness, straightness, and parallelism of the axes. Positioning error covers the difference between the commanded coordinate and the reached coordinate. Dynamic error covers what happens while the machine is moving under load, at feed, with the spindle running.

A dial indicator and a granite square can find geometric faults. They cannot resolve 2 µm of backlash or a servo mismatch that only appears at 3,000 mm/min. Pick the instrument to match the error family you are chasing.

The order matters. Fix geometry, then position, then dynamics. If you chase servo tuning on a machine with 15 µm of Y-to-Z squareness error, you will tune around a mechanical problem and the result will not hold.

Section 2

Prepare the machine and the environment first

Accuracy tests are sensitive to temperature. Cast iron and steel grow about 11 µm per meter per degree Celsius. A 5 °C swing across a 1,000 mm axis moves the ball screw and the scale by roughly 55 µm. That is larger than most of the errors you are trying to find.

Run the machine at idle for 30 to 60 minutes so the spindle, ballscrews, and castings reach thermal steady state. Record ambient temperature at the start and end of every test. If the two numbers differ by more than 2 °C, the data is suspect.

Clean the table, taper, and any mounting pads. A single chip under a magnetic base or a laser mount shifts the reading by more than the test resolution. Wipe guideways and remove the previous job's fixtures.

Warm-up programs should exercise the full travel of each axis, not just the work envelope you normally use. An axis that only sees 200 mm of travel will show different errors when you command the full 750 mm.

1
Record ambient temperatureLog it at test start and end; note any drift above 2 °C.
2
Check level and foundationRe-level before geometry tests; a settled foundation shows up as twist.
3
Confirm the control is in the right modeDisable cutter compensation and any thermal compensation offsets before measuring.

Section 3

Reading the numbers without overreacting

Compare every result against two references: the machine builder's specification and the tolerance on your actual part. A machine that is 8 µm out of spec but only needs to hold ±25 µm is still productive. Chasing the spec when the part does not require it costs spindle time for nothing.

Look at trends, not single readings. If X-axis backlash was 4 µm last quarter and is 9 µm now, the ball screw nut or thrust bearing is moving. A single 9 µm reading with no history might just be a cold machine.

Separate accuracy from repeatability. Accuracy is how close you land to the target. Repeatability is how close successive attempts land to each other. You can compensate for poor accuracy with the control's pitch error table. You cannot compensate for poor repeatability, because the error changes each cycle.

When a machine fails a test, do not immediately call for a rebuild. Re-level the machine, re-run the test warm, and check the mounts. A surprising share of reported geometry errors come from a foundation bolt that has lost preload.

Section 4

How often should you run these checks?

A production VMC running two shifts should get a spindle runout check monthly, a ballbar test every six months, and a full laser and geometry survey annually. Machines cutting hard materials or running unattended need shorter intervals.

Any of these events should trigger an immediate re-test regardless of schedule: a crash, a spindle replacement, a ball screw change, a move to a new foundation, or a sudden shift in surface finish.

Keep a simple log per machine. Date, ambient temperature, instrument used, and the key numbers. Fifteen minutes of recording saves a full day of arguing about whether a machine was ever in spec.

If you outsource parts, ask your supplier for the last calibration date on the machine that will run your job. A current survey report on a machine with 16 simultaneous 5-axis centers means more than a general quality certificate.

Step by step

The seven detection methods, in the order to run them

1
1. Straightness and squareness with a granite square and indicatorMount the square on the table, sweep the long face with a 0.002 mm indicator along the full X travel. Then sweep the short face for Y. Acceptable squareness on a general-purpose VMC is typically 0.010–0.020 mm over 300 mm. If it is worse, adjust the column or check for a twisted bed before doing anything else.
2
2. Spindle taper runout and axis-of-rotation checkInsert a certified test bar in the taper. Measure runout at 25 mm and 300 mm from the gauge line with a 0.001 mm indicator. Taper runout should stay under 0.005 mm; the far-end reading tells you about spindle tilt, not just bearing wear. Rotate the spindle by hand, not under power.
3
3. Linear positioning with a laser interferometerSet up the laser head and reflector along one axis. Command moves in 25 mm or 50 mm increments across the full travel and record the error at each point. Run each direction twice. The spread between forward and reverse readings at the same point is backlash. Compare results against the machine's specification.
4
4. Repeat positioning at a single pointCommand the axis to one target 20 to 30 times, approaching from the same direction. The peak-to-peak spread of readings is the repeatability. Most VMCs hold 0.002–0.005 mm. If repeatability is good but accuracy is poor, the fix is in the compensation table, not the mechanics.
5
5. Circular interpolation with a ballbarClamp the ballbar between the table and the spindle with a 100–300 mm radius. Run a full 360° circle at 500 mm/min and then at 3,000 mm/min. Read the roundness value, then the individual signatures: squareness shows as a tilted oval, backlash as a step at quadrant change, servo mismatch as a stretched oval. Record both feed rates.
6
6. Cutting a test artifact and measuring it on a CMMMachine a standard test piece with pockets, bores, steps, and a circular boss. Measure it on a CMM at 20 °C. This is the only test that captures tool deflection, spindle growth under cutting load, and control lookahead together. Compare roundness of the boss against the ballbar prediction.
7
7. Thermal drift monitoring over a production shiftTouch off on a fixed reference feature every 30 minutes for 4 to 8 hours. Plot the offset against time. A machine that drifts more than 0.020 mm over a shift will not hold ±0.010 mm on long parts. Use the curve to schedule warm-up time before critical cuts.

Method selection

Which method finds which error

Match the instrument to the failure mode you are chasing.

Method	Error found	Typical resolution	When to use
Granite square + indicator	Squareness, straightness	0.002 mm	After a move, crash, or foundation work
Test bar runout	Spindle tilt and bearing wear	0.001 mm	Spindle noise, taper chatter, poor finish
Laser interferometer	Positioning and backlash	0.001 mm over 4,000 mm	Annual calibration or after ball screw work
Repeat positioning test	Repeatability at one point	0.001 mm	Before rewriting the compensation table
Ballbar circular test	Servo mismatch, backlash, squareness	1–2 µm roundness	After tuning drives or changing parameters
Cut artifact on CMM	Combined static and dynamic error	Depends on CMM	Before releasing a tight-tolerance job
Thermal drift log	Growth over a shift	0.001 mm	Long parts, lights-out runs, summer months

Test the machine, not the part

If your drawing calls for ±0.005 mm, you need a machine survey that proves the machine can hold it. We run laser, ballbar, and artifact checks on our own 127 CNC machines and can share the relevant survey with your quote.

Get Instant Quote Talk to an Engineer

FAQs

Questions engineers ask about accuracy testing

Can I detect CNC machine tools accuracy without a laser interferometer?

Yes, for some error families. A granite square and a 0.002 mm indicator will find squareness and straightness faults. A ballbar will find servo mismatch and backlash.

What you lose is absolute linear positioning data over long travel. If your parts are longer than 500 mm and call for ±0.010 mm, a laser survey is the only reliable way to see the error curve.

What does a ballbar roundness value of 15 µm mean?

On a 100 mm radius test, 15 µm roundness is roughly 0.015 mm of radial error. That is workable for general machining but marginal for tight-tolerance work.

Look at the shape, not just the number. A tilted oval points to squareness, a flat step at each quadrant points to backlash, and a stretched oval points to servo mismatch between axes.

How much thermal drift is acceptable?

For a machine expected to hold ±0.010 mm, keep drift under 0.010 mm over the cutting period. If the drift curve is still climbing after four hours, the machine has not reached steady state.

Add warm-up time before the first critical cut, or move the critical features to later in the shift when the curve has flattened.

Do I need to re-test after a control parameter change?

Yes. Servo gain, feed-forward, and backlash compensation values all change the dynamic behavior of the axes. A ballbar test before and after the change takes 20 minutes and tells you whether the tuning helped or made things worse.

What is the difference between positioning accuracy and repeatability?

Accuracy is the gap between the commanded position and the actual position. Repeatability is the spread of actual positions when you command the same point many times.

A machine can be repeatable but inaccurate, and that case is fixable with compensation. A machine that is inaccurate and also not repeatable needs mechanical repair.

How long does a full accuracy survey take?

On a standard vertical machining center, a geometry check and laser survey take four to six hours including warm-up and setup. A ballbar test adds one hour.

Plan for a full day if you want the results written into a report with the artifact cut and CMM measurement included.