Measurement and gauging

A part is not good because it was made well. It is good because somebody measured it and the measurement was trustworthy. Those are different claims, and the second one is where most quality arguments actually live.

A bridge-type coordinate measuring machine in a clean, temperature-controlled inspection room, its touch probe lowered onto a machined cast component fixtured on the granite table, with a rack of ring and plug gauges on the wall behind and an even, cool light throughout. — The room is held at 20 °C for the same reason the machine sits on granite. A measurement is worth exactly what the conditions around it are worth, and no more.

Choosing an instrument — the 10 to 1 rule

A chart of measuring instruments placed against the tolerance range each is suited to. A steel rule, vernier caliper, micrometer, bore gauge, height gauge with a granite plate, coordinate measuring machine and a form tester are shown on a logarithmic scale of resolution from one millimetre down to one tenth of a micrometre. For each the typical resolution and the smallest tolerance it can honestly police under the ten to one rule are stated. — Pick the instrument from the tolerance, not from what is on the bench. Anything that cannot resolve about a tenth of the tolerance is reporting mostly its own variation.

Instrument	Resolution	Honest for tolerances down to
Steel rule	0.5 mm	± 2.5 mm
Vernier caliper	0.02 mm	± 0.1 mm
Micrometer	0.001 mm	± 0.01 mm
Bore gauge (comparative)	0.001 mm	± 0.01 mm
Height gauge on granite	0.001 mm	± 0.01 mm
CMM	0.0005 mm	± 0.005 mm and geometric tolerances

The CMM, and what it is actually for

Hand instruments measure sizes and distances. A coordinate measuring machine does something categorically different: it establishes the datum reference frame in software, then evaluates position, profile and runout against that frame.

That is the only practical way to check a full geometric scheme on a complicated part — and it is also why the datum scheme from datums and datum reference frames has to be right before anyone writes a CMM program.

Functional gauges are the opposite approach and equally valid. A plate with pins at true position, sized at virtual condition, either goes on or it does not. No numbers, no trends — but it answers the only question the assembly line cares about in two seconds, and it accounts for bonus tolerance automatically because the geometry does the arithmetic.

Temperature, which is not a detail

Every dimension on every drawing is defined at 20 °C unless it says otherwise.

steel, 1 metre long, 1 °C rise  ->  about 12 µm longer
aluminium, same                 ->  about 23 µm longer

Measurement system analysis — what it is really asking

A diagram explaining gauge repeatability and reproducibility. The upper panel shows the tolerance band with the true part distribution inside it and the measurement error distribution added, illustrating how a wide measurement variation pushes conforming parts outside the limits and pulls nonconforming parts inside. The lower panel is a bar chart of gauge repeatability and reproducibility as a percentage of tolerance with three zones marked, under ten per cent acceptable, ten to thirty per cent conditional, and over thirty per cent unacceptable, with the consequences of each labelled. — What you record is the true value plus the measurement error. A noisy gauge scraps good parts and ships bad ones at the same time, and tightening the process does not help.

Two separate questions:

Repeatability — one operator, one gauge, the same part, several times. This is the equipment.
Reproducibility — different operators, same parts. This is the method and training.

Together they give Gauge R&R as a percentage of the tolerance:

Gauge R&R	Verdict
Under 10 %	Acceptable
10 – 30 %	Conditional — depends on how critical the characteristic is
Over 30 %	Unacceptable. The measurement is consuming the tolerance

Agree the method before the parts exist

Probing strategy, filtering, datum simulation and clamping force all change the reported value on a real part. Two competent laboratories following different but reasonable methods can each be confident and reach opposite conclusions about the same component.

Settling that up front turns a future disagreement into an arithmetic check. Leaving it open turns it into a commercial dispute with no technical resolution available.

What to take away

Choose the instrument from the tolerance, at roughly 10 to 1 resolution.
A caliper reading two decimals is not a 0.02 mm instrument.
A CMM builds the datum frame in software and evaluates geometric tolerances directly.
A functional gauge gives no numbers and answers the assembly question instantly.
Everything is defined at 20 °C. Steel moves about 12 µm per metre per degree.
Let parts soak before inspection. A hot part is a wrong part.
Repeatability is the equipment; reproducibility is the method.
Under 10 % acceptable, over 30 % not. A bad gauge scraps good parts and ships bad ones.
Agree the inspection method before production, or the dispute has no technical answer.