CNC Machining Quality Inspection Guide: FAI, CMM Reports and Sampling
A buyer-focused guide to CNC inspection plans, FAI, CMM reports, measurement uncertainty, Cp/Cpk, AQL sampling, gauge R&R and supplier quality evidence.
For a buyer, a quality report is valuable only when it proves that the delivered parts meet the drawing under an agreed measurement method. A long report can still miss the features that control assembly, leakage, motion, or safety. This guide shows engineering and procurement teams how to build a practical inspection plan for CNC machined parts—before the purchase order is released.
1. Start with function, not a percentage of dimensions
Do not ask a supplier to inspect “10% of dimensions” without defining which dimensions matter. Classify drawing requirements by consequence:
- Critical: failure may affect safety, regulatory compliance, sealing, interchangeability, or primary assembly.
- Major: failure can reduce performance, life, fit, or downstream process capability.
- Minor: failure is unlikely to affect use but may affect appearance or workmanship.
Critical characteristics normally need an explicit method, frequency, record, and reaction plan. A cosmetic edge break and a bearing-seat diameter should not receive the same control.
2. Match the measuring tool to the characteristic
| Characteristic | Typical method | Buyer question | Common risk |
|---|---|---|---|
| Outside diameter | Micrometer or air gauge | Read at several axial and angular positions? | One reading misses taper or lobing |
| Bore diameter | Bore gauge, air gauge, or CMM | Is size required, or also cylindricity and position? | Two-point size does not prove form |
| Hole position | CMM or functional gauge | Which datum frame and material condition apply? | Wrong alignment makes the report meaningless |
| Flatness | CMM, surface plate, or scan | How many points and what filtering strategy? | Too few points miss local peaks |
| Runout | Indicator with defined datum setup | How is the part restrained and rotated? | Fixture error becomes part error |
| Surface roughness | Profilometer | Cutoff, evaluation length, direction, and parameter? | “Ra” alone does not define the test |
| Thread | GO/NO-GO gauge plus feature checks | Are pitch diameter, depth, and post-finish condition controlled? | A GO gauge alone does not verify every requirement |
3. Use a sensible measurement-resolution rule
A practical starting point is to use equipment whose resolution is substantially finer than the tolerance being verified. A 0.001 mm display does not automatically mean 0.001 mm accuracy. Ask for instrument accuracy, calibration status, environmental conditions, fixturing method, operator procedure, and total measurement uncertainty.
Example: a bore is specified as 20.000 ±0.010 mm, giving a total tolerance of 0.020 mm. An instrument resolution of 0.01 mm is clearly inadequate. Even 0.001 mm resolution must be supported by suitable accuracy, repeatability, temperature control, and a method capable of detecting form error.
4. Drawing interpretation must be agreed first
ISO 1101 defines the symbol language used for geometrical tolerancing. The inspection plan must use the same datum system, modifiers, and requirement interpretation as the drawing. Before production, resolve ambiguous datum targets, simultaneous requirements, profile boundaries, free-state conditions, and whether dimensions apply before or after coating.
A CMM can produce precise numbers from the wrong alignment. The first article review should therefore include the datum setup, probing strategy, point distribution, scanning or discrete-point method, and any software filtering.
5. First Article Inspection is not production control
FAI verifies that the selected manufacturing route can produce a conforming part. It should connect every drawing balloon to a measured result, method, instrument, and acceptance limit. Use it after a new design, tooling change, material change, program revision, supplier transfer, or long production interruption.
FAI does not prove that every future part will conform. Production control also needs in-process checks, final inspection, sampling or 100% verification where justified, and a documented reaction when a result approaches or exceeds a limit.
6. Capability calculations: useful only with stable data
Cp = (USL − LSL) / 6σ
Cpk = min[(USL − mean) / 3σ, (mean − LSL) / 3σ]
Example: a shaft has limits of 9.980 to 10.020 mm. A stable process has a mean of 10.006 mm and estimated σ of 0.003 mm.
- Cp = 0.040 / 0.018 = 2.22
- Upper Cpk = 0.014 / 0.009 = 1.56
- Lower Cpk = 0.026 / 0.009 = 2.89
- Cpk = 1.56
The spread is small but the mean is shifted upward. Do not accept a capability number without the dataset, subgroup method, sample size, stability evidence, gauge suitability, and treatment of outliers.
7. Sampling does not guarantee zero defects
ISO 2859-1:2026 provides attribute-sampling schemes indexed by acceptance quality limit for lot-by-lot inspection. AQL is a decision tool for sampling; it is not permission to ship a known defect rate and it is not a zero-defect guarantee.
The purchase order should identify the governing standard and edition, lot definition, inspection level, AQL by defect class, normal/tightened/reduced rules, and what happens after rejection. Critical characteristics may require 100% inspection, automated error-proofing, or a capability-based control plan rather than ordinary attribute sampling.
8. What a useful dimensional report should contain
- Part number, revision, purchase order, lot, quantity, and serial or cavity identification
- Drawing balloon number and exact characteristic description
- Nominal value, upper limit, lower limit, actual result, units, and pass/fail
- Datum reference frame and material-condition modifier where applicable
- Instrument ID, calibration status, method, and inspection date
- Part temperature or controlled-room condition for temperature-sensitive work
- Inspector or approver, report revision, and traceability to material and finish certificates
- Clear notation for subcontracted processes and post-finish measurements
9. Buyer review checklist for a CMM report
- Confirm the report matches the correct part and drawing revision.
- Check that reported datums match the feature-control frame.
- Review how datum features were simulated and constrained.
- Check the number and distribution of points or scan paths.
- Confirm whether size, form, orientation, and location are reported separately.
- Look for unit, sign, rounding, and coordinate-system errors.
- Review results close to a limit, not only red failures.
- Confirm coated or heat-treated features were measured at the required stage.
10. Measurement-system analysis
If a tolerance is tight or a characteristic is used for process capability, request evidence that the measurement system is suitable. A gauge R&R study can separate repeatability from reproducibility, but the study design must reflect the real range of parts, operators, fixtures, and method.
For destructive tests, very small lots, or unique parts, conventional gauge R&R may be impractical. Agree on an alternative such as reference standards, repeated checks, correlation studies, uncertainty budgets, or independent verification.
11. Reaction plan: define it before a failure
A complete control plan answers four questions: who stops production, how suspect material is contained, how the root cause is verified, and what evidence is required before shipment resumes. Cover segregation, traceability, reinspection, concession requests, corrective action, and notification timing.
12. RFQ inspection requirements buyers can copy
- Provide a ballooned FAI report for all drawing characteristics before production approval.
- Record 100% of critical characteristics; use the agreed sampling plan for major and minor characteristics.
- Measure GD&T requirements from the drawing datum reference frame and state the CMM strategy.
- Measure finish-sensitive dimensions after coating unless the drawing explicitly states otherwise.
- Supply material and finish certificates linked to the manufacturing lot.
- Retain inspection records for the agreed period and obtain written approval before deviation.
Frequently asked questions
Does a CMM report prove a part is good?
Only for the characteristics, setup, strategy, uncertainty, and part condition represented by the report. It does not automatically prove surface integrity, material condition, internal defects, or every unmeasured part.
Should every dimension be inspected on every part?
Not necessarily. Inspection effort should follow risk, process capability, lot size, customer requirements, and the cost of failure. Critical characteristics may justify 100% control; stable low-risk features may use rational sampling.
What should I do when a value is close to the limit?
Review measurement uncertainty, process trend, gauge suitability, and the likelihood of drift. A technically passing result can still signal that the process needs centering or additional monitoring.
Request an inspection-plan review
Send Zync Precision your CAD model, drawing, revision, quantity, critical features, required certificates, and end-use risks. We can review measurement access, datum strategy, post-finish dimensions, report format, and inspection scope before quotation.