Documentation / Calibration / Intervals

I — What an interval is

A calibration interval is a bet

The bet is that the instrument will stay inside its acceptance limits for the period, and that if it does not, the consequences are tolerable. Shorten it increases calibration cost, downtime and spare inventory. Lengthening it increases the quantity of product measured between a drift and its discovery.

Reference · No product claim on this page

Almost every interval on a gage register is inherited. It came from a manufacturer’s datasheet, from the interval the previous quality manager used, or from the round number someone typed when the register was first built. Each is a legitimate starting point. The difficulty is that a starting value tends to become permanent, because nothing in the ordinary running of a calibration program ever asks the question again.

II — The quantity being controlled

The controlling quantity is end-of-period reliability: the probability that an instrument is still in tolerance when it arrives for its next calibration. Choose a target, model reliability as a function of elapsed time from historical as-found data, and solve for the time at which the curve crosses the target.

choose R_target        the reliability the program requires    
model R(t)             from historical as-found results
solve for t            the interval that holds R(t) ≥ R_target

The whole method rests on one input. R(t) is estimated from as-found data: what the instrument read before anyone adjusted it. An instrument returned with a certificate that says only PASS has reported nothing about where it was inside the limits, and a run of those carries no information about drift at all.

III — Why most programs cannot do this

What the method needs	What the program usually has
As-found values, numerically	A pass/fail conformity statement
Per check point across the range	A single overall verdict
With sign	An unsigned magnitude, or nothing
The acceptance limits that applied at the time	The limits in force today
A history long enough to fit a curve	Certificates in a filing cabinet, as PDFs
Grouped by instruments that actually behave alike	Grouped by whatever the register calls them

Every row on the right is a records decision made years earlier, usually without knowing it foreclosed this analysis. That is the recurring shape of this whole subject: the interval is not hard to compute, it is hard to have the data for, and the data is only ever collected by accident unless someone decided to collect it.

IV — The interval is not the only lever

Intermediate checks between calibrations change the arithmetic more than the interval does. They do not establish that the instrument is accurate; they report it has not moved since the last check. That is enough to shorten an exposure window from a quarter to a week, which is the difference between quarantining a quarter’s production and quarantining a week’s.

An interval decision made without that in view optimizes the wrong variable. The cost of a longer interval is the product measured between a drift and its discovery rather than the calibration itself, and a check standard on a shelf is usually the cheaper way to recover it.

V — What to be careful about

Method names travel further than the documents that define them. Interval-setting methods are commonly cited by short designations that appear in secondary literature and in some software interfaces. The recommended practice most often credited for them is paywalled and not publicly reproduced, so those designations are not repeated here. If a tool or a consultant names one, ask which document defines it and check the wording against that document before it goes into a procedure.

A certificate that recommends an interval is making an assumption it cannot make. The manufacturer has no knowledge of the conditions of use, duty cycle or environment. A datasheet recommendation is a reasonable starting point; an interval set on a calibration certificate requires the owner's agreement before it governs.

VI — Sources and boundaries

ILAC-G24, Guidelines for the determination of recalibration intervals of measuring equipment, is the freely available international reference, related to OIML D 10. The North American recommended practice on the same subject is paywalled.
Commonly quoted end-of-period reliability targets sit in the 85–95% range. That range is widely repeated in the interval-analysis literature and is not reproduced here as a specific recommendation, because the right target is a function of what the instrument is used on.
No clause numbers from IATF 16949 or AS9100 appear on this page, and no interval-method designations are cited, for the reason given in section V.

When a gage fails calibration → What a Digimatic frame carries → All documentation →