Traceability for image-based measurement

What traceability actually means

It is not a certificate on a wall. It is a property of a result: this number can be followed back, step by step, to a realisation of the metre, and at every step somebody has stated how much uncertainty the step contributes.

A micrometer has a short chain: the instrument was compared against gauge blocks, which were compared against a laboratory standard, which was compared against a national one. An image measurement inserts an extra and unusually fragile link: a scale factor that a person derived, on a particular day, for a particular optical configuration. Everything downstream inherits it.

That is not a weakness of image measurement. It is simply the link auditors go to first, because it is the one most often undocumented.

The chain, link by link

1. National standard. The definition of the metre, realised by a national metrology institute.
2. Accredited laboratory. A laboratory operating to ISO/IEC 17025 calibrates artefacts against standards traceable to that realisation, and issues a certificate stating the value and its uncertainty.
3. Your calibration artefact. The stage micrometer, graticule, ring gauge or gauge block that certificate refers to. It has an identity, a certificate number, and an expiry.
4. Your scale factor. Derived by imaging that artefact under a stated optical configuration on a stated date by a stated person.
5. Your measurement. A pixel count multiplied by that factor.
6. Your report. The document that lets somebody else follow links 5 back to 1.

Break any link and the ones above it stop meaning anything. The most common break is between 3 and 4: an artefact with a perfectly good certificate, used to derive a scale factor that nobody recorded the circumstances of.

The questions auditors actually ask

In rough order of how often they come up:

"How do you know one pixel is that size?"

They want the artefact, its certificate, the date, and the configuration. "It came set up like that" is the answer that ends well for nobody.

"When was it last calibrated, and what triggers a recalibration?"

A schedule is good. A schedule plus a list of events that force an unscheduled recalibration — objective changed, camera replaced, system moved, fixture height altered — is much better, because it shows you understand what the calibration depends on.

"Show me the measurement you are reporting."

This is the question image measurement answers better than almost any other technique, and only if the report carries the image. A number in a table says a feature was 12.04 mm. An image with the measurement drawn on it says which feature, measured where, between which two points. A CMM report cannot do that.

"What is your uncertainty?"

The honest components are the artefact's certified uncertainty, the pixel span you calibrated over, edge-location repeatability, lens distortion across the field, and operator variation in placing the endpoints. You are not expected to have solved all of them. You are expected to know they exist.

"Who took this, and can they?"

Operator identity on the report, and training records behind it.

"Could someone else repeat this?"

Which is the real question behind all the others.

What belongs on the report

A defensible image measurement report carries, at minimum:

• Part identity — name, number, drawing number and revision, batch
• The instrument, and the software version that produced the sheet
• The magnification or optical configuration used
• The calibration in force, and ideally the artefact behind it
• The date, and the operator
• The measured image, with the overlays that produced each value
• The values, their units, the limits applied, and the verdict
• A scale bar, so the image stays interpretable after it is resized (why this matters)

The unit column deserves particular emphasis. A results table of bare numbers is a table of ambiguities, and the ambiguity is invariably discovered at the worst possible moment.

What Measuret contributes

Software cannot make a measurement traceable. Only the artefact, the procedure and the record do that. What software can do is stop the record being lost, and make it hard to produce a sheet that hides its own basis.

• The part and test information block is a grid you shape to your own quality system — part, drawing, report number, batch, material, inspection standard, sample size, instrument, magnification, operator, shift — with rows you can add and delete, and a saved set so the next report starts from the same fields rather than from nothing.
• The measured image is on the sheet, at full printable width, with the overlay composed at the source image's own resolution. The evidence and the number travel together.
• Units are a report column, not an assumption.
• Unmeasured rows can be printed, marked Not measured, rather than silently dropped — so a checklist that was not completed looks like a checklist that was not completed.
• Verdicts distinguish Pass, Fail and Indeterminate, so a row that cannot honestly be judged is not quietly counted as a pass (more on this).
• Measurements are held in source-image coordinates and converted through the calibration, so zoom, window size and export cannot alter a recorded value.

The remaining links — a certified artefact, a recalibration trigger list, and the discipline to fill the information block honestly — are yours.