CNC Machine Tuning Guide: How Geometry Drift Affects Your Tolerances
A machine can run without alarms and still cut out of tolerance. This guide explains where the error comes from, how much of it you can absorb, and when a retune is the only fix. It is written for engineers and buyers who specify parts, not for machine rebuilder technicians.
What tuning actually corrects
Tuning is not maintenance. Maintenance replaces worn parts; tuning measures how the axes sit relative to each other and brings them back to a known relationship. A VMC with 12,000 hours on it can have a healthy spindle and still place a hole 0.03 mm off nominal, because the table no longer travels square to the column.
The error budget has four parts. Squareness between X and Y, straightness along each axis, spindle axis alignment to the Z travel, and backlash in the ballscrew or rotary table. Each one is small on its own. Stacked together over a 300 mm part, they decide whether you hold ±0.005 mm or fight every setup.
Five-axis geometry adds two more items: the pivot distance between the rotary table center and the spindle, and the trunnion axis alignment. Get these wrong and a part that measures fine at A0 tilts out of tolerance at A90. That is why a five-axis machine needs tuning more often than a three-axis mill, and why the check takes longer.
Practical meaning: a machine does not fail loudly. It drifts. The drift is measurable, repeatable, and correctable, but only if you measure before you cut a production lot.
- 1SquarenessX to Y, Y to Z, Z to X. A granite square and a dial indicator read this in minutes.
- 2StraightnessDeviation of each axis from a true line over its full travel.
- 3Spindle alignmentSpindle axis parallel to Z travel in both planes.
- 4Backlash and pivotLost motion in screws and rotary tables, plus five-axis pivot distance.
How much drift your part can absorb
A tight tolerance does not automatically demand a tight machine. It depends on where the feature sits and how the error enters the cut. A 0.02 mm squareness error over a 100 mm part shifts a hole by 0.02 mm. The same error over a 600 mm part shifts it by 0.12 mm. Size matters more than the spec sheet suggests.
Position tolerance and size tolerance react differently. If a hole is drilled and reamed on the same machine in one setup, machine error mostly cancels between the two operations. If it is milled on one machine and bored on another, the two error stacks add. Good process planning keeps tight features inside a single setup.
Surface finish is a separate axis of error. A spindle with 0.005 mm runout will cut a face with visible chatter at Ra 0.8 μm, no matter how square the machine is. Runout, toolholder condition, and thermal growth are checked alongside geometry because they show up in the same inspection report.
Thermal growth is often blamed for what is really a geometry problem. Aluminum grows roughly 23 μm per meter per °C. A spindle that heats 10 °C over a shift moves its tool tip by a measurable amount. Machines in a temperature-controlled room drift less, but they still drift.
- 1Hold ±0.005 mmSingle setup, short features, warm machine, checked geometry.
- 2Hold ±0.02 mmMost well-kept three-axis work, features under 300 mm.
- 3Fight every partLong parts, multi-setup, cold machine, unmeasured geometry.
Red flags that call for a retune
The clearest signal is a shift in the CMM report that follows the machine, not the part. If two different part numbers, cut weeks apart, both show the same directional bias, the machine moved. If the bias follows one part number, the problem is in the process or the fixture.
A second signal is a change in the sound and finish of a cut. Chatter that appears at a speed that used to run clean usually means the spindle or toolholder has shifted, not that the tool is dull. Check runout before you change feeds.
A third signal is a new operator or a new material. Operators learn to compensate for a drifting machine without telling anyone. When the operator changes, the compensation stops and the parts go out of tolerance. That is a geometry problem being reported as a training problem.
Time-based triggers are blunt but useful. A machine running two shifts in a temperature-varying shop should be checked quarterly. A single-shift machine in a controlled room can run a year between checks. Neither schedule replaces the CMM data, which is the real trigger.
- 1Directional CMM biasSame direction across different part numbers points to the machine.
- 2New chatterAt a proven speed, check runout and geometry before feeds.
- 3Operator changeHidden compensation stops. Recheck geometry.
- 4Calendar checkQuarterly in fluctuating shops, yearly in controlled rooms.
How to verify a machine after tuning
Verification is a test cut, not a printout. A machine can pass a ballbar test and still cut a bad part if the fixture or the thermal state is wrong. The strongest evidence is a test part that exercises the same axes and the same tolerance as your production part.
For a five-axis machine, cut a part that requires all axes to move together. A cone or a hemisphere on a trunnion table is a good choice because it exposes pivot error and axis alignment at the same time. Measure it on a CMM, not with hand tools, and record the numbers against the machine's own history.
Run the test at production temperature. A cold machine reads differently from a warm one. Warm up the spindle for 30 to 60 minutes at the speeds you actually use, then cut. This single step removes more false alarms than any other change in the procedure.
Keep the records. The value of a tuning log is not the absolute number, it is the trend. A machine that moves 0.005 mm a year is stable. A machine that moves 0.005 mm a month needs a repair, not another tune.
- 1Test part over printoutExercise the same axes and tolerance as production.
- 2Five-axis cone or hemisphereExposes pivot error and axis alignment together.
- 3Warm up first30 to 60 minutes at production speeds.
- 4Track the trendRate of change matters more than the absolute value.
When retuning is not the answer
Retuning cannot fix a worn ballscrew. If backlash exceeds the compensation range in the control, no amount of geometry correction will hold tolerance. The machine needs a screw, a bearing, or a rebuild. Tuning a worn machine produces good numbers on a ballbar and bad parts on the table.
Retuning also cannot fix a process problem. If the fixture is not rigid, or the tool is too long for the cut, or the part is being machined in a state of residual stress, the machine is not the cause. In those cases, a full geometry check confirms the machine is fine and redirects the effort to the right place.
There is an economic boundary too. A 20-year-old three-axis mill with a worn spindle may not be worth a full rebuild if the work it does is only ±0.05 mm. The right decision is to keep it, tune it lightly, and route tight work to a machine built for it. Not every machine needs to be a precision machine.
- 1Worn ballscrewBeyond control compensation. Replace, do not tune.
- 2Rigid process problemFixture, tool length, residual stress. Tuning will not help.
- 3Economic limitMatch the machine to the tolerance the job needs.
Which action fits the symptom
Match the observed behavior to the likely cause and the correct response.
| Symptom | Likely cause | Correct action |
|---|---|---|
| Same directional bias across part numbers | Machine geometry drift | Full geometry retune |
| Bias follows one part number only | Fixture or process | Check fixture, not machine |
| Chatter at a proven cutting speed | Spindle or toolholder runout | Measure runout first |
| Error grows across a shift | Thermal growth | Warm up and check compensation |
| Backlash beyond comp range | Worn ballscrew or bearing | Repair, not tune |
| Part fine at A0, off at A90 | Five-axis pivot error | Five-axis geometry retune |
| Tolerance looser than ±0.05 mm | No precision need | Light check, no rebuild |
The verdict
If the CMM bias follows the machine, retune it. If it follows the part, fix the process. If the screw is worn, repair it. Tuning is the right answer only when geometry is the cause.
Common questions
How often should a CNC machine be tuned?
It depends on duty cycle and environment more than on the calendar. A machine running two shifts in a shop where the temperature swings should be checked quarterly. A single-shift machine in a controlled room can run a year between checks.
The real trigger is data. When the CMM report shows a directional bias that repeats across part numbers, the machine has moved and needs a check regardless of when it was last tuned.
Can I tune a machine myself?
Basic squareness and level checks are within reach of a shop with a granite square, a dial indicator, and a test bar. That covers most three-axis geometry work.
Five-axis pivot and trunnion alignment is a different job. It needs a ballbar or a calibrated test part and the control's kinematic parameters. A wrong entry there can make the machine worse than it was.
Does a retune change the machine's accuracy spec?
No. Tuning restores the machine to the accuracy it was built to. It does not raise the spec. A machine rated at ±0.005 mm is tuned back to that number, not beyond it.
If a machine cannot be brought back to its original spec, the cause is wear, and the fix is a repair or a rebuild, not another tuning pass.
Why do parts drift over a long shift?
Thermal growth is the usual cause. Aluminum expands about 23 μm per meter per °C, and a spindle warming 10 °C moves the tool tip by a measurable amount. The part grows too, so the effect is not always obvious.
Warm up the machine for 30 to 60 minutes at production speeds, and check whether the control has thermal compensation enabled before assuming the geometry is wrong.
What does a geometry check cover?
Squareness between axes, straightness over full travel, spindle alignment to the Z travel, and backlash in screws and rotary tables. On a five-axis machine it also covers pivot distance and trunnion alignment.
A full check ends with a test cut at production temperature, measured on a CMM and recorded against the machine's history so the trend is visible.
When is a light check enough?
When the work the machine does is looser than the drift. A machine cutting at ±0.05 mm does not need a full five-axis geometry pass if the CMM shows it is holding that band.
Match the check to the tolerance of the job. Over-maintaining a roughing machine costs money that is better spent on the machines that cut tight work.
Send us the drawing, get a machining plan
We run 127 high-precision CNC machines across three plants, with 16 simultaneous five-axis centers and 100% inspection before shipment. Send your part and we will tell you which machine holds it and how we verify it.
12-hour quote and DFM±0.005 mm capabilityISO 9001 / IATF 16949 / ISO 13485
More from GreatLight
We publish setup notes, tooling trials and inspection data from the factory floor.
