CNC Parts Inspection Standard: What Gets Measured and Why

• 1
  What a CNC Parts Inspection Standard Actually Defines
• 2
  Tolerance Stack-Up and Process Capability
• 3
  Matching the Instrument to the Feature
• 4
  Sampling Plans and What Full Inspection Really Means
• 5
  Where Measurement Error Comes From
• 6
  Reading an Inspection Report Without Guessing
• 7
  Which Inspection Method Fits Which Feature

A CNC parts inspection standard is not a single document. It is the chain of decisions that turns a drawing into a number you can trust. The drawing states nominal size, tolerance and datum references. The standard tells you how to hold the part, which instrument to use, how many pieces to check, and how to record the result. Break any link and the number loses meaning.

The chain starts with the datum. A bore measured from three different faces gives three different readings, even on a perfect part. So the first question on any inspection plan is not which gauge to use but which surface locates the part. That is why ASME Y14.5 and ISO 1101 put datum feature symbols on the drawing instead of leaving the choice to the shop floor.

A second decision is unit and rounding. A drawing tolerance of ±0.05 mm and a gauge that reads 0.01 mm are a reasonable pair. A gauge that reads 0.001 mm on the same feature adds digits, not certainty, unless the machine and the thermal environment can support them.

Finally, the standard fixes the acceptance rule. A feature at the tolerance limit is acceptable or not depending on whether the limit is inclusive, and on whether the measurement uncertainty has been subtracted. Without that rule, two inspectors can disagree on the same part and both be correct.

• 1
  DrawingNominal size, tolerance zone, datum references, GD&T controls.
• 2
  MethodFixture, instrument, probing strategy, environment.
• 3
  SamplingHow many pieces, at what intervals, full or partial.
• 4
  RecordReport format, traceability, pass/fail rule.

A tolerance on a print is a requirement, not a prediction. Whether a process can hold it depends on the machine, the fixture, the material and the thermal state of the part. We normally work to ±0.005 mm where the geometry allows it, but that figure only holds when the setup is rigid and the part is not thin-walled.

Process capability is the practical number. If a feature runs at Cp 1.0, roughly three parts per thousand fall outside a ±3σ band. Push the same feature to ±0.01 mm and the process becomes comfortable. This is why an engineer should ask what the feature does before tightening a tolerance. A locating bore that sets gear mesh needs the tight number. A clearance hole for an M6 screw does not.

Surface finish behaves the same way. Ra 0.8–1.6 μm is a normal machined finish on aluminum and steel. Ra 0.2–0.8 μm needs a finer step-over, a sharper tool and often a second operation. Each step down in roughness adds cost and adds measurement time, because a stylus or optical profilometer has to be brought in.

Geometric tolerances interact with size tolerances. A shaft can be within its diameter limits and still not fit, because straightness or cylindricity consumed the clearance. Inspectors check the size first because it is fast, then the form and orientation because those are what actually cause assembly failures.

• 1
  Size firstFast, cheap, catches gross errors.
• 2
  Form nextStraightness, flatness, cylindricity; needs a CMM or roundness tester.
• 3
  Orientation lastPerpendicularity, parallelism, angularity; datum-dependent.

Hand tools answer most questions on a production floor. Calipers and micrometers cover outside diameters, lengths and thicknesses at roughly 0.01 mm resolution. Bore gauges and pin gauges cover holes. Height gauges and dial indicators cover step heights and runout. These are fast and cheap, and on a ±0.05 mm feature they are the right choice.

A CMM takes over when the feature is geometric. Position, profile, concentricity and true position all need a probe to touch many points and fit a mathematical feature. A bridge CMM in a temperature-controlled room is the reference method, and it is also where the datum question gets settled, because the software aligns to the datum features named on the drawing.

Optical methods cover what a probe cannot reach. Vision systems measure small holes, slots and edge profiles without contact force. Laser scanners capture freeform surfaces and compare them to the CAD model as a color map. For a thin blade or a soft plastic part, contact probing can deflect the feature more than the tolerance allows, so optical is not a luxury there.

Surface finish uses its own instruments. A contact profilometer drags a diamond stylus across the surface and reports Ra, Rz and Rmax. Portable units with a skid are common on the floor; skidless units are used for curved or short surfaces. Measure the same direction as the function, because turning and milling leave different roughness along and across the feed direction.

• 1
  Hand tools±0.01 mm resolution; diameters, lengths, simple steps.
• 2
  CMMGeometric tolerances, true position, first article reports.
• 3
  Optical / visionSmall features, soft parts, freeform surfaces.
• 4
  ProfilometerRa, Rz; check direction against the functional direction.

100% inspection before shipment is our default, and it means every part passes through a dimensional check against the drawing before it is packed. It does not mean every feature on every part is measured on a CMM. That distinction matters when you write a purchase order, because the two arrangements have very different cost and lead time.

Full CMM inspection of every feature is a first article activity. The first part off a new setup gets the full report: all dimensions, all geometric controls, material certificate and finish check. Subsequent parts get a shorter check on the features that the process can actually drift. On a stable setup, that is often the tightest tolerance and the thinnest wall.

Statistical sampling follows ANSI/ASQ Z1.4 or ISO 2859-1 when a lot is large and the process is known. A general inspection level II plan on a 1,250-piece lot draws 80 pieces at AQL 1.0. The plan assumes the process is in control. If a tool change or a material batch change happened mid-lot, sampling has to restart, because the population is no longer one population.

For prototype and low-volume work, sampling is not useful. One part out of three tells you little. Instead, inspect the part completely, then use the report to correct the CAM program before the next run. That is the fastest route from a first article to a stable production process.

• 1
  100% dimensionalEvery part, key features, before packing.
• 2
  First articleFull feature list, CMM report, material and finish records.
• 3
  StatisticalLarge stable lots; AQL plan per ISO 2859-1.
• 4
  PrototypeInspect fully, then feed results back into the program.

Temperature is the largest single source. Aluminum expands about 23 µm per meter per °C. A 500 mm aluminum part measured at 25 °C instead of 20 °C is roughly 0.06 mm longer than its certificate says. On a ±0.05 mm tolerance, that alone can flip the result. Steel is about 11 µm per meter per °C, so the same part in steel moves half as much.

Fixture-induced distortion is the second source. Clamping a thin plate flat on a plate can flatten a bow that the part will recover once released. The measurement is then correct for the clamped state and wrong for the free state. The fix is to measure in the free state, or to specify the clamping condition on the drawing.

Probe and stylus geometry matter more than most people expect. A ruby ball of 2 mm diameter cannot enter a 1 mm slot, and it reads a radius on every sharp edge it touches. Probe deflection adds a small error on every point, which is why CMM software compensates and why calibration spheres are checked at the start of each shift.

Operator technique is the last source and the hardest to control. Consistent probing speed, consistent datum sequence and consistent reporting format remove most of it. Two inspectors following the same written procedure will agree far more often than two inspectors improvising.

• 1
  ThermalSoak the part; record the temperature in the report.
• 2
  ClampingMeasure free state unless the drawing says otherwise.
• 3
  ProbeBall diameter limits feature access; check calibration.
• 4
  TechniqueWritten procedure, fixed datum sequence, fixed report format.

A useful inspection report repeats the drawing callout in the same order, states the nominal and the tolerance, then gives the measured value and the pass/fail result. If the report lists only a final verdict, you cannot tell whether a marginal feature was one step from failing. Ask for the measured values, not just the checkmarks.

The report should name the instrument and its calibration status. A micrometer with an expired calibration certificate is a number without a pedigree. For CMM results, the report should state the alignment, the datum features used and the number of points per feature. Ten points on a Ø200 mm bore is thin coverage; a hundred points is not.

Material and finish records belong in the same package. A dimensional report on a part made from the wrong heat of 17-4PH is worthless. We include material certificates on request, and finish thickness or coating weight is recorded for plated and anodized parts because it changes the final dimension.

Keep one first article report per part number and revision. When the revision changes, the report is stale. That single habit catches most of the assembly problems that surface months later, and it gives a buyer something concrete to compare against a supplier's claim.

• 1
  Values, not checkmarksNominal, tolerance, measured, result.
• 2
  Instrument traceabilityModel, calibration date, resolution.
• 3
  Material and finishCertificates and coating records.
• 4
  Revision controlOne report per part number and revision.