Vision measurement, CMM and gauges: choose by the characteristic

What question does the drawing ask?

Begin with the characteristic, datum and tolerance. A visible diameter in silhouette, a hole-to-hole distance and a three-dimensional positional tolerance are different tasks. Specify the drawing revision and whether inspection is intended for process feedback, product acceptance or both. A quick trend check can be valuable without being sufficient for final release. The method must answer the actual acceptance question, including features that are not visible from the proposed camera position.

Why can a good camera still give a poor measurement?

If the part sits differently, the measured edge can move even when its size does not. Lens perspective, focus, burrs, oil and the lighting direction all influence the image. Telecentric optics may reduce sensitivity to depth changes within their designed range, but they do not remove fixture error, thermal changes or uncertainty in where a rounded edge is defined. Validate the whole measurement chain, including loading and software, rather than quoting pixel size as measurement accuracy.

How should the methods be compared?

Choose representative parts spanning the working range, including near-limit parts. Establish suitable reference values and record how they were obtained. Run the proposed production method with repeated unloading and loading, appropriate operators or fixture configurations and realistic environmental variation. Compare systematic differences from the reference as well as variation between repeated readings. NIST separates repeatability, reproducibility, stability and bias; averaging repeated measurements does not automatically remove a systematic error.

Use the existing Gauge R&R guide for the study design. Where methods disagree, inspect the definition of the feature: a CMM probe, an optical edge and a gauge may interact with a burr or radius differently. Do not force numerical agreement by changing offsets until the reason for the difference is understood.

What happens near a tolerance boundary?

Agree the acceptance decision rule before commissioning. A result close to a tolerance limit may need a defined guard band or reference remeasurement based on the measurement uncertainty and customer requirements. Avoid repeated testing until a favourable reading appears. Record the initial result, remeasurement reason and final disposition so the process does not silently convert measurement variation into accepted product.

What should a feasibility request include?

Send the drawing revision, characteristic list, datums, required cycle, sample range and available reference measurements. Describe the loading method and production surface condition. Ask which characteristics the proposed station can cover, how the remaining features will be checked and how uncertainty will be established. For one-face fixtured components use the machined-part measurement scope; for single-file bolts, pins and turned parts use the fastener scope.

Fixtured machined-part measurement · Fasteners and turned parts · Gauge R&R study

Confirm that both methods measure the same feature

Before comparing results, define the measurand: the quantity being measured, including its datum and relevant conditions. A vision system may detect a projected silhouette while a contact instrument samples a physical surface at defined points. A hole affected by a chamfer can therefore produce different values depending on which edge or depth is used. Neither result becomes correct merely because its instrument has a higher resolution.

Write the feature definition in the acceptance plan and use identified parts for comparison. Record cleaning condition, orientation, temperature conditions where relevant, fixture and reference method. NIST’s bias guidance addresses systematic difference from a reference; it does not remove the need to choose an appropriate reference in the first place. (NIST e-Handbook, 2.4.5 Analysis of bias) A reference instrument with an unsuitable probe strategy or unsupported datum can also produce misleading comparisons.

South Africa’s Measurement Units and Measurement Standards Act provides the national measurement context, and the accreditation legislation establishes the framework for calibration and conformity-assessment accreditation. (South Africa: Measurement Units and Measurement Standards Act 18 of 2006) (South Africa: Accreditation for Conformity Assessment, Calibration and Good Laboratory Practice Act 19 of 2006) Check the actual calibration scope and uncertainty for the required feature. A general accreditation logo or certificate for another measurement is not sufficient evidence for this comparison.

A hypothetical optical sampling calculation

Hypothetical example: a camera covers an 80 mm field across 2,432 pixels. Object-plane sampling is 80 / 2,432 = 0.0329 mm per pixel. A 0.20 mm feature spans about 6.08 pixels. These calculations describe image sampling only. They do not establish that the station measures to 0.0329 mm, nor do they show that it can accept a particular drawing tolerance. Edge interpolation, optics, contrast, calibration and presentation must be evaluated separately.

Subpixel algorithms can estimate an edge between pixel centres, but that estimate still depends on image formation and calibration. Do not convert a software decimal display into a metrological capability claim. Similarly, a gauge with fine display increments can have uncertainty larger than its least significant digit. JCGM 100 sections 4–6 distinguish evaluation and combination of uncertainty contributions from display resolution. (JCGM 100:2008, Guide to the expression of uncertainty in measurement)

A useful experiment compares the same parts across the production range and examines the difference against size and position. A constant offset suggests a different problem from an error that grows across the field of view. Repeat at relevant orientations and heights to expose optical distortion, perspective or datum effects. Keep the original data rather than reporting only an average difference that can conceal opposing errors.

Use the methods as complementary controls

Vision is useful when the required external feature is visible, product presentation is controlled and the complete measurement method meets the agreed decision requirement. A CMM can support more complex spatial definitions and reference investigation, subject to its own probing strategy and uncertainty. A dedicated gauge can provide a direct functional or go/no-go check for a particular feature. The best production system may use more than one of these methods at different stages.

For an inline camera, test repeated loading, fixture wear and the permitted contamination condition. For the reference method, preserve the calibration and measurement procedure. NIST’s gauge R&R and repeatability guidance explains why short-term repetition is only part of characterising a measurement process. (NIST e-Handbook, 2.4 Gauge R&R studies) (NIST e-Handbook, 2.4.4.1 Analysis of repeatability) A stable laboratory result does not automatically describe variation on a vibrating production fixture.

Agree the handling of results near tolerance limits before production. If the uncertainty or method disagreement prevents a defensible release decision, use the specified reference or review route. Do not silently widen drawing limits to reduce optical rejects. The product owner should approve any decision rule and understand its implications for false acceptance and unnecessary rejection.

Keep a reference record that survives maintenance

For each critical characteristic, retain the drawing revision, measured feature definition, calibration record, reference part identities, raw paired measurements and analysis of differences. Link the camera recipe and fixture revision to that record. Replacing a lens, moving the camera or changing a locating feature can invalidate the relationship even if the software continues to display plausible dimensions.

Use stable check artefacts or reference parts appropriate to the method for routine monitoring, and define how damage or drift is detected. A check part is useful only while its reference condition remains known. Record a failed check and its affected production interval before restoring the system; deleting an outlier from a trend is not a corrective action.

The dimensional-measurement application page describes station design, and the gauge R&R article explains a variability study. This comparison guide helps select and correlate the measurement methods. It does not claim that camera resolution alone replaces a CMM, that contact measurement is automatically correct or that any instrument is suitable for every tolerance.