5 Methods to Measure the Dimensional Accuracy of Machining Parts

A digital caliper reads outside length, step, and depth in one pass. Resolution is typically 0.01 mm, but real accuracy is closer to ±0.02–0.03 mm because of jaw flex and operator feel. Use it for stock checks, deburring verification, and features with a callout of ±0.05 mm or looser.

An outside micrometer does better on a single dimension. A 0–25 mm micrometer with a friction thimble holds ±0.002 mm repeatability when the part and the tool sit at the same temperature. That makes it the right pick for shaft diameters, plate thickness, and ground surfaces.

Pin gauges answer a different question: does this hole fall inside its limits? Class ZZ pins in 0.01 mm steps tell you go or no-go without reading a number. For a Ø6 H7 bore, a 6.000 mm go pin and a 6.012 mm no-go pin settle the call in seconds. They do not tell you the actual size, only the fit.

These methods share one weakness. They measure one point at a time, so roundness, taper, and position error stay hidden. A caliper cannot find a bore that is 0.03 mm oval.

• 1
  Best forSimple outside dimensions, quick first-article checks, incoming stock
• 2
  Typical uncertainty±0.02–0.03 mm caliper; ±0.002 mm micrometer
• 3
  Skip whenThe drawing controls form, position, or a bore tighter than IT7

A granite surface plate plus a height gauge builds a reference plane the part can trust. The plate is flat to a few micrometres across its whole face. A carbide-tipped height gauge with a dial or digital head then reads step heights, shoulder locations, and hole centers from that plane.

Gauge blocks set the master. A stack wrung to 50.000 mm lets you zero the gauge against a known length instead of trusting the factory scale. This matters on parts where a 0.01 mm error in setup becomes a 0.01 mm error in every reading.

The method handles medium-complexity work well: a milled pocket depth, a boss height, the distance from a datum face to a dowel hole. It struggles with angled features and free-form surfaces, where the contact point moves as the gauge slides.

Temperature control is the real limit. A 100 mm steel part grows about 0.0012 mm per °C. Over an 8 °C swing between a cold truck and a warm bench, that is roughly 0.01 mm of drift before anyone touches a gauge.

• 1
  Best forStep heights, datum distances, hole centers from a flat face
• 2
  Typical uncertainty±0.005–0.01 mm with a good plate and block stack
• 3
  Skip whenFeatures are angled, curved, or buried inside the part

A bridge or gantry CMM touches the part with a ruby stylus and records X, Y, Z for every point. Software fits the points to a nominal model, then reports size, position, form, and orientation against the drawing. This is the method most engineers mean when they ask for a dimensional report.

A typical shop-grade CMM holds volumetric accuracy around 2 + L/300 μm. A 200 mm feature therefore carries roughly ±0.0027 mm of machine error, well inside the ±0.005 mm callouts we hold on bearing bores and mating faces. Probe and stylus add their own share, so a clean, rigid setup still matters.

The CMM earns its cost on position and form. True position of a bolt circle, perpendicularity of a shaft to a face, flatness of a sealing surface: these are hard to capture with hand tools and simple on a CMM. Scanning heads go further, collecting thousands of points to map a whole contour.

The trade-offs are time and access. A full first-article report can take an hour per part. Deep bores and undercuts may need a star stylus or a rotary table. Very soft materials, such as un-anodized aluminium, can be marked by the stylus if contact pressure is not reduced.

• 1
  Best forGD&T, position, form, first-article inspection reports
• 2
  Typical uncertainty≈2 + L/300 μm at 20 °C
• 3
  Skip whenThe part is a loose-tolerance bracket or a one-off visual check

Structured-light and laser scanners capture a dense point cloud without touching the part. They suit thin walls, delicate features, and free-form surfaces a stylus cannot follow. A full part can be digitized in minutes, which makes them useful for reverse engineering and for comparing a machined part to its CAD model.

Accuracy depends on the field of view. A handheld scanner may hold 0.02–0.05 mm over a large part. A fixed system with a small field can reach 0.005 mm on a short feature. The rule is simple: smaller view, better number.

Shiny and transparent surfaces fight the sensor. Anodized aluminium, polished stainless, and clear plastics need a thin matte coating or a spray before scanning. That coating is removable, but it adds a step and a small thickness error.

Scanners shine at full-surface comparison. They are weaker at a single tight bore, where the laser cannot see past the edge. Pair a scan for form with a CMM or pin gauge for critical fits.

• 1
  Best forFree-form surfaces, thin walls, CAD comparison, reverse engineering
• 2
  Typical uncertainty0.02–0.05 mm handheld; down to 0.005 mm on small fields
• 3
  Skip whenThe part is mirror-polished or the feature is a deep blind bore

Trial cutting measures the process, not just the part. The operator cuts a small portion of the surface, measures what came off, and adjusts the tool offset before finishing the rest. Two or three passes bring the dimension into range without scrapping the workpiece.

This method earns its place on large or expensive parts, where a full re-cut is not an option. A 4,000 mm frame, a one-off casting, or a part with a long lead time all justify a slow start. It also catches setup errors: a wrong work offset shows up on the first trial cut, not the last.

The limit is that trial cutting only covers the feature being cut. A part can pass its trial cut and still fail on a feature machined later. Combine it with a final inspection step so nothing ships unverified.

At GreatLight, trial cutting runs alongside in-process monitoring and a 100% inspection before shipment. Raw material check, in-process checks, and final inspection reports are available on request.

• 1
  Best forLarge, costly, or low-volume parts where scrap is unacceptable
• 2
  Typical uncertaintyDepends on the on-machine probe; often ±0.01 mm
• 3
  Skip whenThe part is small and cheap enough to scrap without a second thought