CNC machining verification: how accuracy is proven before parts ship

• 1
  What CNC machining verification actually measures
• 2
  How the measurement method sets the accuracy ceiling
• 3
  Why setup and thermal drift decide the result first
• 4
  What a real first article inspection report contains
• 5
  When full inspection is worth it and when sampling is enough
• 6
  Where verification stops being useful
• 7
  Matching the measurement method to the feature

Verification is the step where a finished or partly finished part is measured against the drawing, not against the machine's own readout. The machine says where the tool went. The inspection says what the part became. Those are two different numbers, and the gap between them is where scrap comes from.

A drawing calls out a bore at Ø25.000 +0.010 / −0.000 mm. The machinist sets the tool offset, cuts the bore, and the control displays a value. That displayed value is a position, not a diameter. Only a bore gauge or a CMM tells you whether the hole is inside the tolerance band. Verification closes that loop.

The scope is broader than size. Verification also covers geometric tolerance: flatness, perpendicularity, concentricity, position. A shaft can hit every diameter callout and still fail because it is bent 0.03 mm over 200 mm. Surface finish belongs here too. A sealing face at Ra 3.2 μm will leak where the drawing asked for Ra 0.8 μm, even though the caliper reads perfect.

At GreatLight we treat verification as a process step with its own time slot, not as a final favor to the customer. Raw material is checked on arrival, features are monitored during the run, and every part gets a final inspection before it goes in the box.

• 1
  DimensionalDiameters, lengths, depths, hole positions against the tolerance band.
• 2
  GeometricFlatness, perpendicularity, concentricity, true position.
• 3
  SurfaceRa values on sealing, sliding and optical faces.

No inspection is better than its instrument. A rule of thumb in metrology is the 10:1 rule: the measuring device should resolve about ten times finer than the tolerance it is checking. For a ±0.005 mm tolerance, you want a device good to roughly ±0.0005 mm.

Calipers and micrometers resolve to 0.01 mm and 0.001 mm. They are fast and fine for general dimensions, but they measure a point, not a form. A micrometer reads the highest points of a lobed bore and reports it as round. That is why a three-point bore gauge or an air gauge gets used when a bore must be truly circular.

A CMM samples discrete points and fits geometry through them. Sampling density is the whole game. Touch a 300 mm diameter flange at eight points and you will miss a local dip between them. Touch it at forty points and the form error shows up. CMM accuracy also drifts with temperature. A 20 °C controlled room is not a luxury on tight work.

Optical and laser systems fill gaps where contact probing is awkward: thin walls, soft materials, small features, or surfaces you cannot touch without marking. They trade some accuracy for speed and coverage. On a first article, we often cross-check one feature with two different methods before we trust the number.

• 1
  Calipers±0.02 mm practical. Good for stock checks and rough dimensions.
• 2
  Micrometers±0.001 mm resolution. Diameters only, one axis at a time.
• 3
  CMMBest for position and form when point count is high enough.
• 4
  Optical / laserNon-contact, fast scanning, softer accuracy on steep surfaces.

Most out-of-tolerance parts are not cut wrong. They are held wrong or measured wrong. A vise clamped at 40 N·m will deform a thin-walled aluminum housing. The part springs back after unclamping, and every dimension you took while it was clamped is now fiction. Soft jaws, fixturing wax and minimal clamping pressure solve more tolerance problems than a new machine does.

Thermal drift is the quiet one. Aluminum expands about 23 μm per meter per °C. A 500 mm aluminum part sitting 8 °C warmer than the inspection room grows roughly 0.09 mm. That is eighteen times a ±0.005 mm tolerance. Machining generates heat, so a part measured straight off the machine is not the part you ship.

The practical answer is soak time. Let the part reach room temperature before final inspection. On tight work that means 30 to 60 minutes on the bench, longer for thick sections. The same logic applies to the machine itself. A spindle that has run for ten minutes is not at the same thermal state as one that has run for four hours, and the offsets move with it.

Tool wear adds a third drift. A carbide end mill wearing 0.02 mm over a 500-part run will walk the last parts out of tolerance unless the operator re-measures and re-offsets. In-process monitoring catches this. Checking the first part and the last part catches nothing in between.

• 1
  ClampingThin walls deform. Soft jaws and light pressure keep the geometry honest.
• 2
  Soak time30–60 minutes on the bench before final measurement.
• 3
  Tool wearRe-measure mid-run; a worn tool drifts the whole batch.

A first article inspection report is not a certificate. It is a document that maps every drawing callout to a measured value, an instrument, and a pass or fail. If a report does not name the instrument, you cannot judge the number. A flatness of 0.01 mm means one thing from a surface plate and dial indicator, and something else from a CMM with sparse points.

The ballooned drawing is the backbone. Each dimension gets a number, and the report lists that number, the nominal, the tolerance band, the actual reading, and the deviation. That layout lets a design engineer scan for the features that are close to the edge, which is usually more useful than the pass column.

Material traceability belongs in the same packet. A 17-4PH stainless part needs a heat lot number that ties back to the mill certificate. Without it, the mechanical properties are an assumption. For medical work under ISO 13485, the report also carries process records: who ran it, on which machine, with which program revision.

Reports are available on request for our production runs. We do not ship a report nobody asked for, because the paperwork should match the risk of the part. A bracket and an implant do not need the same packet.

• 1
  Ballooned drawingEvery callout numbered and traceable to a measured value.
• 2
  Instrument namedThe report states the device and its resolution.
• 3
  Material lotHeat number tying the part to a mill certificate.

Full inspection of every feature on every part is slow and expensive. Sampling is defensible when the process is stable and the feature is not safety-critical. The logic is statistical: if a CMM run of the first ten parts shows a tight spread well inside the band, the process is in control, and checking one part per hour is reasonable.

The exception is any feature where a single failure causes a functional loss or a safety event. A combustion seal face, a medical implant thread, a hydraulic port on a flight component. Those get 100% inspection regardless of process stability, because the consequence of a missed defect is not proportional to its probability.

Feature accessibility also drives the decision. A deep cross-hole at 8× diameter is hard to measure repeatably. If you cannot measure it cleanly, you cannot sample it meaningfully, and the better answer is often to change the design or the process so the feature becomes measurable. Good metrology starts at the drawing, not at the inspection bench.

We run 100% inspection before shipment on production orders, with in-process monitoring during the run. That covers raw material check, dimensional monitoring, and final inspection. The reports come out of the same records, so the numbers in the packet are the numbers we acted on.

• 1
  SampleStable process, non-critical feature, low consequence of failure.
• 2
  100% inspectSealing faces, safety features, hard-to-measure critical geometry.
• 3
  RedesignIf a feature cannot be measured, its tolerance cannot be defended.

Verification confirms that a part matches a drawing. It cannot confirm that the drawing is right. If a clearance is modeled at 0.05 mm and the real assembly needs 0.15 mm, every inspection passes and the assembly still binds. That is a design problem wearing a quality costume, and no CMM report will surface it.

There is also a floor on what machining can hold. At ±0.005 mm, a 100 mm aluminum part sits near the practical edge of three-axis milling when you account for thermal drift and tool wear. Push below that and you are grinding, lapping, or rethinking the tolerance. A tolerance that no process can hold is not a quality target, it is a cost multiplier.

Surface finish behaves the same way. Ra 0.2–0.8 μm is achievable on our machines with the right tool path, but it demands specific feeds, speeds and often a finishing pass. Asking for it on a non-functional face adds cycle time for no benefit. Put the tight finish where it does work: sealing, sliding and optical surfaces.

The useful question is not how tight we can measure. It is which features actually need the tight band, and whether the drawing says so. Engineers who mark critical dimensions and leave cosmetic faces loose get faster parts and cleaner reports.

• 1
  Drawing errorsInspection cannot catch a design that is dimensionally correct and functionally wrong.
• 2
  Process floorBelow ±0.005 mm on aluminum, expect grinding or lapping, not milling.
• 3
  Finish placementRa 0.2–0.8 μm only where the function demands it.