Two inspection reports sat on one table, both describing a batch of 500 aluminum housings. The supplier's, checked with an air gauge, cleared every bore; the buyer's CMM flagged twelve parts, the hole pattern 0.06 mm outside its 0.05 mm true position callout. Nobody had lied. The teams had never agreed which CNC inspection methods would judge each feature, so they were not measuring the same thing, and that is the origin of most cnc part inspection disputes.
This page answers three questions: which method resolves the tolerance, how readings are judged at the boundary, and what belongs in the purchase order.
Why the inspection method matters more than the inspection equipment
A bore at Ø12 +0.02/0 mm leaves a band 0.02 mm wide. A vernier caliper resolving 0.02 mm cannot even print 12.014. It shows 12.00 or 12.02, and its resolution equals the whole band. A bore micrometer resolves about 0.001 mm, or 5% of the band. An air gauge repeats to roughly 0.1 µm and finds a 4 µm taper neither could see.
Each carries a different measurement uncertainty, the interval around a measured value within which the true value lies. Choosing among CNC inspection methods is choosing how wide that interval is, so match the method to the tolerance band, starting with CNC machining tolerances.
The cost of getting it wrong is never the inspection fee. It is good parts scrapped and a line stopped.
The six families of CNC inspection methods
Six families of CNC inspection methods cover everything on a shop floor, and a seventh, 3D scanning and laser probes, sits in the comparison table below as a specialist branch of coordinate metrology. What matters is not the count but which question each one answers. For the broader process context, start from our overview of CNC machining.
Hand-held and bench gauging
Calipers, micrometers, height gauges and dial indicators — cheap, immediate, and dependent on the hand holding them: inspectors can differ by several microns. Use them on noncritical dimensions and patrol checks, never as evidence in a cnc part inspection report.
Dedicated gauges and go/no-go
Pin gauges, thread plug and ring gauges and custom fixtures return a verdict, not a value, clearing a part faster than anything here. They yield no trend data, and a burr can jam a plug gauge and invent a failure. See burrs and edge breaks.
Air gauging (pneumatic gauging)
Air gauging meters back pressure behind a nozzle as its gap to the workpiece changes; non-contact and self-cleaning, so it belongs beside the machine. Repeatability runs about 0.1 µm at high magnification and 0.5 to 1 µm in everyday shop setups, while the usable range is typically 20 to 200 µm, so each plug covers one diameter and zeroes on two masters. One reading gives diameter, roundness, taper and straightness. Never position.
Optical and vision measurement
An optical comparator projects a magnified silhouette against a chart overlay, around 12.5 µm (0.0005 in) in two dimensions. A vision measuring system uses telecentric optics and sub-pixel edge extraction to reach roughly 5 µm (0.0002 in), with CAD overlay. Both excel where probe force would deform a feature; both fail on reflective surfaces and burrs.
Coordinate measuring machines
A coordinate measuring machine (CMM) moves a probe along three axes and records points. Bridges suit prismatic parts, gantries handle panels, and arms travel to parts that cannot move. Touch-trigger probes take discrete points; scanning probes sweep thousands and serve form and profile better. Only a CMM returns size, form, orientation and location in one coordinate system, so it carries nearly all GD&T judgment.
When the doubt is about location rather than size, our true position tolerance guide sets out bonus tolerance at MMC, virtual condition and how a round zone differs from a square one.
Surface roughness measurement
Stylus instruments report Ra, Rz and their relatives, and Ra is not a property of the surface alone, as the next section explains.
Inspection equipment compared: accuracy, speed, cost and best-for
Figures below are references only. Accuracy claims are the easiest thing to misread across all CNC inspection methods.
| Method | Typical accuracy or repeatability | What it measures | Speed per part | Relative cost (capital) | Best for | Cannot do |
|---|---|---|---|---|---|---|
| Caliper | About 0.02 mm resolution | Length, OD, ID, depth | Seconds | $ | Noncritical dimensions | GD&T, true position |
| Micrometer and bore gauge | About 0.001 mm resolution | OD, ID, thickness | Seconds | $ | Diameters to ±0.01 mm | Position, relationships |
| Go/no-go and dedicated gauges | Pass or fail, no value | Threads, hole size, fit | Seconds | $ to $$ | High-volume screening | Numeric results, trend data |
| Air gauging | Repeatability to about 0.1 µm; range 20 to 200 µm | Diameter, roundness, taper, straightness | Seconds | $$ | Precision bores in oily shop conditions | True position, 3D geometry |
| Optical comparator | Approximately 12.5 µm (0.0005 in) | 2D profile, thread pitch | Fast | $$ | Turned shafts, flat profiles | 3D contours, internal features |
| Vision measuring system | Approximately 5 µm (0.0002 in) | 2D and 2.5D geometry, CAD overlay | Moderate | $$$ | Thin, soft or micro features | Reflective, transparent or burred edges |
| Bridge CMM | Per ISO 10360-2, E0,MPE = A + L/K µm with L in mm, for example 2.2 + L/300 or 3.0 + L/333 | Size, form, orientation, location, run-out | Moderate, programmed | $$$$ | GD&T, datums, first article | Very soft parts |
| 3D scanning and laser probes | Typically ±0.01 to 0.05 mm, depending on system and part | Point clouds, deviation maps | Fast | $$$ | Sculptured surfaces, reverse engineering | Tight GD&T at ±0.01 mm |
| Surface roughness tester | Depending on parameter and cut-off | Ra, Rz, Rq | Under 1 min | $$ | Finish verification | Any dimensional result |
Two things follow from the table above. Maximum permissible error grows with measured length, so claims such as "CMM accurate to ±0.002 mm" mean nothing without the ISO 10360-2 formula. Cost is per piece, not per machine: an air plug clears a bore in seconds, costing a fraction of a CMM cycle over a run. Pair every result with traceable measurement records.
Which gauge belongs to which tolerance band
This table turns the six CNC inspection methods into a purchasing decision, using the ISO 2768 general tolerance classes for linear and angular dimensions, and ISO 22081 for general geometry.
Rule of thumb: the gauge should resolve the tolerance by about 10:1, with 4:1 as the floor.
| Tolerance band | General tolerance class | Recommended primary method | Acceptable alternative | Why not step down |
|---|---|---|---|---|
| ±0.5 mm and looser | Very coarse to coarse (v, c) | Caliper, visual | None | Anything tighter is wasted cost |
| ±0.2 mm | Coarse (c) | Caliper | Height gauge | No extra decision value |
| ±0.1 mm | Medium (m), mid ranges | Caliper or height gauge | Vision system | Certainty you cannot act on |
| ±0.05 mm | Fine (f), nominal sizes to 6 mm; individual callout above | Micrometer or bore gauge | Vision system, CMM | A caliper's 0.02 mm resolution takes a fifth of the 0.10 mm band, which is 5:1 |
| ±0.02 mm | Outside the general classes | Micrometer or bore gauge, air gauge for bores | CMM | 0.02 mm against a 0.04 mm band is 2:1, so a caliper spends half the band |
| ±0.01 mm | Individual callout | CMM, air gauge for bore diameter | Vision system, small parts | Hand tools consume the band in uncertainty |
| ±0.005 mm and tighter | Individual callout | CMM in a temperature-controlled room, with stated uncertainty | Air gauge, diameter only | Needs an uncertainty budget, not a better machine |
| Position, profile, run-out | Not covered by general tolerances | CMM, or fixture with indicator | None | Hand tools cannot build a datum frame |
| Bottom line | Match the method to the band, not to the habit | Below ±0.01 mm, use a CMM or air gauge plus a written uncertainty budget | Anything looser spends money on nothing | Stepping down leaves no band left to prove |
Behind it sits a working convention, not a standard requirement, and the arithmetic is less dramatic than the slogan. A ±0.05 mm bore has a 0.10 mm band, so a 0.02 mm caliper resolves a fifth of it, which is 5:1. That clears the 4:1 floor and misses 10:1. Call it what it is: a fifth of your tolerance spent on the gauge, and the argument stops being about the caliper. At ±0.01 mm the same caliper is at 1:1 and has nothing left to decide with. A design for manufacturability review at quote stage is the cheapest place to strike it.
If you want a second opinion on which method fits which feature, send us the drawing with the CTQ list.
How a CMM inspection actually runs, step by step
Every CMM inspection runs in four stages, and costly mistakes happen in the first.
Drawing review — deciding what to prove, not what is easy to measure
The programmer identifies CTQ characteristics first, then the datum scheme, because datum order describes how the part sits in its assembly and dictates how it is located. A part clamped for convenience rather than to its datum reference frame measures perfectly and still fails in service.
Fixturing, temperature and probe selection
Aluminum expands at roughly 23 µm per meter per °C. A 200 mm aluminum part measured 5 °C above the 20 °C reference of ISO 1 has grown about 23 µm along its longest axis. That is 58% of the 0.04 mm band that ±0.02 mm allows, or 115% of the tolerance itself read one-sided. A shop measuring parts straight off the machine loses every argument with a temperature-controlled customer.
Datum alignment and measurement execution
The operator builds a coordinate system by the 3-2-1 method, measures the datums, then the features. Point count and distribution decide credibility, and six points on a bore support one conclusion.
Before arguing over a number, check the measurement: our gage R and R guide separates equipment variation from appraiser variation, and neither costs much to fix.
Evaluation, uncertainty and the report
Every line should carry the measured value, the tolerance band and the judgment. "TP 0.05: PASS" tells you almost nothing; showing the X and Y deviation and actual hole diameters reveals whether a part passed on merit or borrowed margin from its size tolerance. Equipment ID, calibration dates and operator then connect the numbers to a first article inspection report.
How accurate is a CMM, really?
CMM accuracy is given by ISO 10360-2 as E0,MPE = A + L/K, in µm with L in mm, commonly 2.2 + L/300 or 3.0 + L/333. The constant A dominates on short parts; the L/K term takes over as measured length grows. A single number like "±0.002 mm" means nothing without it.
CMM accuracy is a function of length, not a single figure. ISO 10360-2 specifies the maximum permissible error of length measurement as E0,MPE = A + L/K, in µm with L in mm, and specifications commonly read 2.2 + L/300 or 3.0 + L/333. The constant term dominates on short parts and the length term further out. On a 2.2 + L/300 machine, E0,MPE is 2.37 µm at 50 mm and 4.87 µm at 800 mm. On a 3.0 + L/333 machine it is 3.15 µm and 5.40 µm. The same machine, twice the permissible error, which is why "accurate to ±0.002 mm" is a claim about the constant term only.
Acceptance testing happens once, at installation; what protects you is reverification, repeated on a cycle the user sets, typically annually. Ask for the latest reverification report, along with calibration certificates traceable to ISO/IEC 17025. The chain itself is public: NIST in the United States, NPL in the United Kingdom, and BIPM internationally hold the length standards those certificates point back to.
Then settle how boundaries are judged. ISO 14253-1 sets the default rule: whoever claims, pays. A supplier proving conformance must land inside a zone shrunk by its own uncertainty; a customer proving non-conformance must land outside one expanded by theirs. CNC inspection methods differ mainly in how wide that gray band is.
GD&T: what each method can and cannot verify
Two parallel symbolic languages describe geometry. ISO 1101:2017 defines the form, orientation, location and run-out symbols, with ISO 1660:2017 covering profile; the North American counterpart is ASME Y14.5-2018, with its mathematics in ASME Y14.5.1. ASME Y14.5 and ISO 1101 encode the same ideas with different defaults, and the default is where drawings fail.
The trap is this. ASME Rule #1 applies the envelope principle: size limits also control form unless the drawing says otherwise. ISO applies the independence principle per ISO 8015: size and geometry are met independently unless a relationship is stated, with datums in ISO 5459. Read an ISO drawing with ASME habits and you assume form is controlled when it is not.
Which GD&T callouts need a CMM?
Not every GD&T callout needs a CMM, and not every true position callout does either — a functional position gauge will clear a hole pattern faster than any CMM, as long as the drawing says it governs.
Sorted by what genuinely needs a coordinate measuring machine:
- Flatness, straightness, roundness — a surface plate and indicator often suffice for simple parts.
- Parallelism, perpendicularity, angularity — height gauge with indicator, or CMM when the datum is not a face.
- True position, concentricity, profile — CMM inspection territory: each needs a datum reference frame. Air gauging gives roundness and taper, never position.
- Run-out — a bench center and dial indicator handle it, with no CMM required.
The pattern to take away: callouts that need a datum reference frame need a CMM, and callouts that a fixture can answer do not.
Surface roughness: the parameter, the cut-off and the current standard
Ra is the average of a filtered profile, and the filtering is slippery. Ra is not a property of the surface alone. It is the result of a sampling length, the cut-off λc, and lengthening that cut-off raises the reported Ra. When two labs disagree about surface roughness measurement on identical parts, the cause is a filter, not a metal.
Two more variables decide the number: stylus tip radius, since a larger tip bridges narrow valleys, and traverse direction, since scanning with the lay reads lower. A usable record states cut-off, direction and location, which is what your surface roughness callouts should specify.
The standards have moved too, and much supplier documentation has not. Three of the four standards in that note have moved. ISO 4287 was withdrawn in December 2021 and replaced by ISO 21920-2:2021. ISO 1302 gave way to ISO 21920-1:2021, and ISO 4288 to ISO 21920-3:2021, which now covers operators such as sampling length and filters. ASME B46.1 remains the North American equivalent. A drawing citing ISO 4287 stays valid as frozen under change control, but a newly issued drawing should cite ISO 21920-1, especially for cosmetic surfaces such as brushed aluminum finishes.
What to put on the drawing and in the PO
Get the general-tolerance note right, and stop citing a withdrawn part
The title-block note ISO 2768-mK is half obsolete. The "m" points to the ISO 2768 general tolerance tables for linear and angular dimensions. The "K" points to Part 2, which has been withdrawn: new drawings cite ISO 22081 for general geometry.
Recommended wording: cite those general tolerance tables by the edition quoted on your drawing, then name ISO 22081 for general geometry. Never write a bare "ISO 2768", because an undated reference means "current version", and with a revised edition in publication at ISO/TC 213, such a note can change its own meaning.
Name the inspection method for CTQ features
Write it feature by feature in the technical agreement: bore position by CMM, bore diameter by air gauge, cosmetic face by visual comparison against a limit sample, thread by go/no-go plug gauge to class 6H. "Supplier shall select the appropriate method" hands the argument to whoever holds the part.
Specify sampling, report format and the decision rule
Sampling should be explicit: full layout on first article, then AQL sampling or 100% on CTQ characteristics, with reduced plans justified by capability data. Reports should carry actual values, tolerance limits and a verdict, plus equipment ID, calibration due date, inspector and date. That is the minimum a cnc part inspection record needs to survive a dispute six months later.
Write the decision rule down too. A shaft drawn at 20.000 ±0.01 mm that reads 20.009 mm at the supplier and 20.011 mm at goods-in is not cheating. It is two honest measurements sitting inside each other's uncertainty. With ISO 14253-1 named in the PO, neither party can prove its claim and the part is dispositioned against the agreed rule instead of the loudest voice.
Ask for the environment and the calibration chain
Request three documents at quotation stage, not after a dispute: the ISO 10360-2 acceptance or reverification report for the machine that will measure your part, calibration certificates under ISO/IEC 17025 accreditation, and a measurement uncertainty budget for your characteristic size. Those answers reveal more than any brochure about how to evaluate a CNC supplier. Where parts enter regulated products, IATF 16949 for automotive and ISO 13485 for medical devices also expand what you may demand.
To get those four drawing clauses and those three documents into your next order, ask us for the inspection plan we would run on your part.
What inspection costs you in money and calendar time
Tolerances are priced in provability, not difficulty. Moving from ±0.01 mm to ±0.005 mm pays for extra first-part verification, lighter finishing passes, in-batch wear compensation, fewer re-clamping operations, and higher sampling with uncertainty statements.
Documentation is calendar time, not machine time. A full first article report takes a day or a week, and expedited orders usually stall at the report rather than the spindle, which is worth remembering when negotiating CNC machining lead time.
One lever lowers cost and risk together. Spend inspection intensity on the features that make the assembly work and let the rest fall back to general tolerances. It is the principle behind what drives CNC machining cost.
Worked example: an audio front panel and a knurled knob
The front panel carries three demands at once. Cosmetic-face flatness and the six-hole pattern need a true position judgment, so they go to a CMM: a hole pattern can only be judged inside one datum reference frame. Surface flaws and grain direction belong to visual inspection against limit samples. And fit dimensions move in the anodizing bath, so anodizing and final dimensions must be reconciled: inspect after treatment, or agree pre-plate dimensions in writing, on custom amplifier front panels.
The knurled knob needs a different mix. Outer diameter and bore fit are quick on a micrometer at prototype volumes and on an air gauge once quantities rise, since air gauging will not mark a visible surface. Knurl profile and pitch defeat a touch probe, because the tip lands differently on every tooth, so those go to a vision system.
One knob, three CNC inspection methods. Nothing about that part is unusual. A per-feature plan written into the notes for CNC machined audio knobs beats any single "best" machine.
Frequently asked questions
What is CMM inspection?
CMM inspection is dimensional measurement on a coordinate measuring machine: a probe touches points on a finished part, and software compares those coordinates against the drawing inside a reconstructed datum reference frame. It is the only common method that returns size, form, orientation and location in one coordinate system, so it carries most GD&T decisions.
How accurate is a CMM?
CMM accuracy follows ISO 10360-2 as E0,MPE = A + L/K, in µm with L in mm, for example 2.2 + L/300 or 3.0 + L/333, so permissible error grows with the length measured. A single figure such as "±0.002 mm" therefore means nothing without the formula, the length, and whether it describes resolution, repeatability or maximum permissible error. Always ask for the latest reverification report.
What is the difference between a CMM and an optical comparator?
An optical comparator projects a magnified two-dimensional silhouette onto a screen for comparison against a chart overlay, typically around 12.5 µm accuracy, and is fast on profiles, threads and chamfers. A CMM measures discrete points in three dimensions with a tactile probe: slower and costlier, but able to judge position, datum relationships and every geometric tolerance. A comparator cannot see depth.
Which inspection method should I use for ±0.01 mm?
Use a CMM for anything at ±0.01 mm, or an air gauge where the feature is a bore diameter and quantities justify a dedicated plug. Hand tools are wrong here, because a caliper's 0.02 mm resolution consumes the entire tolerance band in uncertainty and leaves nothing to decide with. Below ±0.005 mm, require a written uncertainty budget rather than simply asking for a better machine.
Do I need a CMM inspection for every part?
No. Choosing among CNC inspection methods starts with the tolerance: use a CMM when a geometric tolerance applies, when several features must be judged against one shared datum reference frame, or when the tolerance is ±0.01 mm or tighter. Plain lengths and diameters looser than that are faster and cheaper on micrometers, bore gauges or dedicated fixtures, and high-volume feature checks on repeat dimensions belong on go/no-go gauges instead.
What should a CNC part inspection report include?
Every line should carry the measured value, the tolerance band and the judgment, because "PASS" alone hides how much of the tolerance the part consumed and whether a borderline result came from the process or from bonus tolerance. Alongside those, require part number and revision, ballooned drawing reference, equipment ID with calibration due date, inspector name and date, and measurement temperature where tolerances are tight.
How do I know if a supplier's measurements are trustworthy?
Ask for three documents and judge the response rather than the claims. The first is the ISO 10360-2 acceptance or reverification report for the machine that will measure your part, dated inside its stated cycle. The second is calibration certificates for the masters it relies on, issued under ISO/IEC 17025 accreditation. The third is a measurement uncertainty budget for your characteristic size.
What to do on Monday
Circle the CTQ characteristics on your most troublesome drawing, name one method per characteristic, and request those three documents when you ask for quotations rather than after an argument. That single page of notes is what turns cnc part inspection from an argument into a procedure.
We have machined parts for 23+ years, run 200+ CNC machines, and hold IATF 16949, ISO 9001 and ISO 13485. We routinely hold ±0.01 mm on customer-drawn features, quote within 24 hours and ship samples in 3–7 days. Send a drawing with the critical characteristics marked and we will quote it with the inspection plan attached, including which of these CNC inspection methods we would apply, and why.
Related capabilities: CNC milling and CNC turning.