Coordinated Measurement Machine: How It Works and When to Use It

• 1
  What the coordinated measurement machine actually measures
• 2
  Probe types and what each one is good for
• 3
  Accuracy, repeatability and uncertainty are three different numbers
• 4
  Datums decide whether the numbers mean anything
• 5
  How CMM data feeds back into CNC machining
• 6
  When a coordinated measurement machine is the wrong choice
• 7
  Matching the measurement method to the feature

A coordinated measurement machine records the position of a probe tip in three axes. The bridge or gantry moves the probe, a scale on each axis reports its position, and the controller turns three numbers into one point in space.

One point tells you almost nothing. The value comes from many points. Touch 12 points on a bore and the software fits a circle. Touch 25 points on a plane and it fits a plane. From those fitted features it derives diameter, flatness, position and angle.

The machine is not a camera and it is not a scanner. It is a contact instrument with a known probe tip radius. Every point is corrected for that radius before fitting, which is why tip calibration is the first thing to check when readings drift.

Typical bridge machines hold ±0.005 mm in a temperature-controlled room. That number is not the accuracy of your part. It is the accuracy of the machine under ideal conditions, and it sets the floor for everything measured downstream.

Touch-trigger probes fire a signal the moment the stylus deflects. They are fast, cheap to replace, and fine for prismatic features at ±0.005 mm to ±0.01 mm. They are the default for checking hole position and hole diameter on a machined housing.

Scanning probes stay in contact and stream thousands of points per second. They cost more and need more setup, but they show form error that a touch probe hides. If a bore is round within 3 μm but lobed in three places, only a scan will show it clearly.

Optical and laser probes measure without contact. Use them on soft plastics, thin sheet, or parts you cannot touch without leaving a mark. The trade-off is surface finish sensitivity: a polished face reads differently from a bead-blasted one, so laser data needs its own calibration.

Stylus choice matters as much as the probe head. A long stylus flexes more and amplifies error. For deep bores we switch to a shorter stylus or a star configuration, and we re-qualify the tip after every change.

Accuracy is how close a reading sits to the true value. Repeatability is how close repeated readings sit to each other. A machine can repeat to 1 μm and still be 8 μm off the true size. Buyers often quote one number and mean the other.

Uncertainty combines both, plus temperature, fixturing and the part itself. ISO 10360-2 defines how a CMM is tested with gauge blocks and length bars. Ask for that test result rather than a brochure figure.

Thermal drift is the quiet error. Aluminum expands about 23 μm per meter per °C. A 300 mm aluminum part moving 3 °C shifts roughly 21 μm, which is larger than the tolerance band on many jobs. We let parts soak at 20 °C ±1 °C before final inspection.

Fixturing adds its own error. A part clamped on three points and squeezed in a vise can spring back after release. Measure in the free state whenever the drawing calls for it, or note the clamping condition on the report.

A CMM does not know where the part is. You tell it by building a datum reference frame from features on the part. Three points on a primary plane, two on a secondary, one on a tertiary. Get that order wrong and every position reading shifts.

The drawing datum and the inspection datum must match. If the drawing calls A as the large face and you align to three pads on the back, position results will disagree with the customer's incoming inspection by tens of microns.

For cylindrical parts we often build an axis datum from a long bore or a spigot. That is more stable than a short face, and it matches how the part sits in its assembly. Position then reads as runout around the true axis.

Keep a documented alignment routine per part number. Repeatability across shifts comes from the routine, not from the operator's memory.

Inspection is only useful if it closes the loop. When a batch of 5-axis machined brackets shows a consistent +0.012 mm shift on one bore, that is not random. It points to thermal growth on the spindle or a worn tool, and the offset can be corrected in the next setup.

We use CMM results to confirm first-article parts before a run starts. A report on the first part at 20 °C tells the machinist whether the program, the fixture and the tool offsets are all in agreement. Fixing that before 500 parts is cheaper than sorting after.

For die-cast and vacuum-cast parts, CMM data separates machining error from casting error. If the raw casting is already 0.15 mm out of position on a boss, no amount of tool offset will fix it. The tool path needs to shift instead.

We machine to ±0.005 mm and hold Ra 0.8–1.6 μm on functional faces, then inspect 100% before shipment. Reports are available on request. Raw material checks and in-process monitoring sit between those two points.

If you only need to know whether a shaft is 20.00 mm or 20.05 mm, a micrometer answers in five seconds. A CMM adds setup time and costs more per part. Use the simple gauge.

If the part is 4,000 mm long, few CMMs can take it. Our largest machining envelope is 4,000 × 400 × 150 mm, and a part that size is usually checked with a laser tracker or a portable arm instead.

If the surface is soft or the part is flexible, contact probing can deflect the feature you are trying to measure. Optical methods or a low-force scanning head are the better route.

And if the geometry is a simple 2D profile with a tight tolerance, an optical comparator or a vision system often gives a faster, clearer answer than a full 3D alignment.