CNC CMM Technology: How Coordinate Measurement Confirms Machined Parts

• 1
  What a CMM actually measures
• 2
  Bridge, gantry and arm: which CMM suits which part
• 3
  When to touch the part and when to scan it
• 4
  Alignment and datums decide the result
• 5
  Temperature, fixturing and the error budget
• 6
  Reading a CMM report like an engineer
• 7
  Gage R&R: proving the measurement system itself
• 8
  A first-article CMM run, step by step
• 9
  Which measurement method fits the feature

A coordinate measuring machine is a rigid bridge or arm carrying a probe. The machine records the probe position on three linear axes, and software fits geometry to the collected points. A bore becomes a cylinder. A face becomes a plane. A hole pattern becomes a set of axes relative to a datum. The output is not a diameter reading; it is a reconstruction of the part in space.

That distinction matters. A caliper gives you one number at one place. A CMM gives you position, orientation and form, and it can compare all of them to the drawing at the same time. Position tolerance, coaxiality and profile are natural outputs of the same measurement run.

Typical shop CMMs resolve to 1 μm or better on a temperature-controlled floor. That is finer than the ±0.005 mm machining tolerance we hold on 5-axis work, which is why the CMM is used to verify the process, not just the part.

The measurement is only as good as the probe tip. A worn ruby stylus adds error that no software can remove. Tip qualification against a reference sphere is the first check of every shift.

Bridge CMMs are the default for machined parts up to roughly 1,000 mm. The bridge moves in Y, the ram in Z, and the part sits on a granite table. Granite gives low thermal expansion and high stiffness, so the geometry stays stable through a long measurement run.

Gantry CMMs scale up. The bridge spans a fixed rail and the part stays still, which suits large weldments, frames and housings. If your part is 1,500 mm or longer, a gantry is usually the only practical option because moving a heavy part between setups destroys the alignment you are trying to measure.

Horizontal-arm CMMs reach into deep bores and gearbox cavities from the side. They are slower to program and less stiff than a bridge, but for a transmission housing with internal bearing seats there is no substitute. Portable arms trade accuracy for reach and can be taken to the machine, which is useful for in-process checks on parts that cannot be moved.

None of these is better in the abstract. The choice follows part size, feature accessibility and how much of the tolerance budget you can spend on measurement uncertainty.

Contact probing with a ruby or silicon nitride tip is still the reference method. It is insensitive to surface color, gloss and finish, and it handles deep holes, undercuts and threads. The trade-off is speed: a few hundred points per feature, one at a time.

Optical and laser scanning capture thousands of points per second on free-form surfaces. They are the right tool for turbine blades, impellers, sheet-metal profiles and any part defined by a CAD surface rather than by dimensions. The limit is line of sight. A scanner cannot see the bottom of a Ø6 mm hole 40 mm deep, and it struggles on shiny aluminium without a matte coating.

A practical split: use scanning for form and profile of external surfaces, then confirm critical bores and datum features with contact points. Mixing the two on one report is normal in aerospace and medical work.

Touch probes also reach where a scanner cannot, such as the far side of a rib or an internal O-ring groove. If a feature decides whether the assembly seals, measure it with contact.

A CMM does not know where the part is. It only knows where the probe is. Alignment is the step that tells the software which physical surfaces correspond to the A, B and C datums on the drawing. Get it wrong and every position number on the report is wrong, even though each individual point is accurate.

The common error is aligning to a convenient machined face instead of the datum the designer specified. On a housing that is fixtured three times during machining, the drawing datum may be a face that no longer exists on the finished part, so it has to be established from a functional feature. That decision should be made with the designer, not by the metrology technician alone.

Best-fit alignment is useful for profile work and for checking a part that has no clean datum. It is not acceptable for tolerance verification, because it hides the very deviation you are trying to quantify. Use datum alignment for acceptance, best-fit for process study.

For parts with a Ø400 mm rotary table setup, we align to the same datum scheme used in the CAM program. That keeps the CMM result and the machine coordinate system talking about the same geometry.

Steel grows about 11 μm per meter per °C. Aluminium grows about twice that. A 500 mm aluminium part measured at 26 °C against a 20 °C drawing carries roughly 60 μm of thermal error before the CMM contributes anything. On a ±0.005 mm tolerance that is the whole budget.

Metrology rooms are held at 20 °C ± 1 °C for this reason, and parts are allowed to soak before measurement. Soak time depends on mass: a thin bracket stabilizes in under an hour, a 20 kg casting may need several hours. Measuring a part straight off the machine is a common source of false rejections.

Fixturing matters too. Clamping a thin wall distorts it, and the CMM then measures the clamp, not the part. Light clamping, or support at the same points used in machining, gives a truer picture.

The error budget is the honest frame for all of this: machine error, thermal drift, probe tip wear, fixturing distortion and software fit. If the sum approaches the tolerance, the measurement cannot decide accept or reject with confidence.

A CMM report is a set of fitted features, each with a deviation from nominal. Look first at the datum callouts, then at position and profile, then at size. Size is usually the easiest to hold and the least interesting. Position tells you whether the part will assemble.

Form tolerances are separate from size. A bore can be perfectly on size in average diameter and still be 20 μm out of round. If the report only lists diameter, ask for roundness and cylindricity on that feature.

Watch the number of points per feature. A circle fitted from 4 points is close to meaningless for form; 8 to 12 points spread around the circumference give a usable roundness figure. Software will happily fit a circle to bad data.

Finally, check the reported uncertainty and the stated temperature. A deviation of 8 μm with 5 μm uncertainty is not a clear pass or fail. It is a conversation about the process, and it usually ends with a repeat measurement or a capability study.

Before a CMM result can gate production, the measurement system has to be shown capable. Gage R&R compares variation between operators and between repeated measurements of the same parts. For a tolerance of ±0.005 mm, a total Gage R&R under about 10% of tolerance is a working target; 10–30% is marginal and needs justification.

Most of the variation usually comes from setup, not from the machine. Two technicians aligning the same part to the same datums can differ more than the CMM's own repeatability, especially on parts with short datum surfaces or interrupted faces.

A simple check we run: measure one master part ten times over a shift without re-fixturing. The spread is the machine and software contribution. Then have two operators re-fixture and re-align the same part five times each. The extra spread is the human contribution, and it is almost always the larger number.

For IATF 16949 programs, this data feeds the PPAP package. For one-off prototypes it is overkill, and the money is better spent on a second machining setup.