By Timothy Malone, Axiospec
Most calibration programs pick an interval once and never revisit it. An instrument arrives with a twelve-month recommendation, or the lab puts everything on an annual cycle. Either way, that number outlasts the reason anyone chose it. Both directions cost you. Set the interval too long, and instruments drift out of tolerance between calibrations. Every measurement taken in that window is then in doubt. Set it too short, and you pay to calibrate instruments that were never going to fail. The data to get this right is already in your records. It is the as-found condition you capture at every calibration.
What as-found is really telling you
As-found is the instrument’s condition when it arrives, before any adjustment. Read the right way, it is a pass or fail on the interval that just ended. In tolerance as-found means the instrument was still good at the end of the period. Unless the trend says otherwise, you can treat it as having held up the whole time. Out of tolerance as-found means it did not. Now you have two jobs. Shorten the interval, and work out which measurements taken since the last good calibration are in question. That second job is reverse traceability. It is the one most programs forget.
As-left is the condition when the instrument returns to service. If nothing needed adjusting, as-left matches as-found. Either way, it confirms the instrument left in tolerance and restarts the clock.
Record as-found across enough calibrations, and those pass-or-fail results add up to a number worth acting on. That number is measurement reliability.
The reliability-target method
Reliability is the probability that an instrument is still in tolerance when its interval ends. You get it straight from history. Take the in-tolerance as-found results and divide by the total number of calibrations. Do that over a set period at a set interval, and you have your observed reliability.
Two caveats come with that number. First, it is an estimate. A small group carries a wide margin of error, so a handful of results will not tell you much. Second, a pass is only as good as the calibration behind it. A result that sits right at the tolerance limit deserves a second look, especially on a thin test uncertainty ratio with no guard banding.
Several published methods exist for reviewing intervals. The two main references are NCSLI RP-1 and the international guide ILAC-G24, which is the same document as OIML D 10. The reliability-target method is one of these. It is the one your as-found history unlocks, and it works best on groups of like instruments.
Start by setting a reliability target. Base it on what a wrong measurement would cost you, usually somewhere between 85 and 95 percent. Then compare your observed reliability against it. Below target is the case that matters most. It means instruments are failing more often than you accepted, so you shorten the interval and catch the problem sooner. Above target, the instruments are holding up well, and you can consider a longer interval, though only with care. How far you move in either direction depends on the size of the gap.
A worked example
Suppose you run twenty pressure gauges on a twelve-month interval, which gives you forty calibrations to look at over two years. Start with the case that matters most for quality. If only thirty of the forty came back in tolerance as-found, your reliability is 75 percent, well below an 85 percent target. This is the method earning its keep. The gauges are failing more often than you accepted, so you shorten the interval and catch the drift before it puts more measurements in question. Figure 2 shows this case in the calculator, with a shorter interval near seven months.
Now the other direction. If thirty-six of forty came back in tolerance, reliability is 90 percent, above target, and the gauges are holding up better than a twelve-month cycle assumes. Here you can consider a longer interval, but only with care. Forty results is a small sample, so you step out modestly if at all, stay inside your maximum interval and any manufacturer or regulatory limit, and pull back the moment the next cycle disagrees. Figure 3 shows this case, with the longer interval the tool points toward without telling you to jump to it.
The goal is never to calibrate less. It is to put each instrument on the interval its own evidence supports.
Group your instruments
A single instrument rarely gives you enough calibrations to trust the math. RP-1 handles this by grouping instruments into families that behave alike, meaning the same make and model, similar use, and similar environment. You set and adjust the interval for the whole family. Any instrument that misbehaves breaks out on its own. Two or three data points prove nothing. A family with a couple dozen calibrations between them tells you something real.
Watch for drift, not just failures
The reliability method scores every calibration as a pass or a fail. That is powerful, but it says nothing about how an instrument fails. A gauge that drifts steadily in one direction is a different problem from one that fails at random. It may need adjustment or replacement, not just a tighter interval. So keep the actual as-found deviations, not only the verdict, and watch the trend. A gauge creeping toward its limit will pass as-found right up until the cycle it does not.
What an assessor wants to see
ISO/IEC 17025 expects you to calibrate on a defined schedule, to have a basis for the intervals you use, and to review them. It does not name a method. That puts the burden on you to justify your call. Reliability numbers drawn from your own as-found history are about as strong an answer as you can give. The interval is not a guess or a hand-me-down default. It is set to hold measurement reliability at a target you can point to, backed by data. No software or method makes you compliant on its own. Your procedures and your assessor decide that. But walking into an interval review with reliability figures for each instrument family is a good place to stand.
Why most programs still do not do this
None of this is new. RP-1 first appeared in 1979, more than four decades ago. Programs still run flat, never-reviewed intervals because the data is trapped. As-found and as-left conditions sit on scanned certificates and in spreadsheet cells, cut off from the instrument record. Pulling reliability across a family becomes an archaeology project nobody has time for.
The fix is upstream. Capture each calibration’s as-found and as-left status as a structured record, tied to the instrument and its interval. Then reliability stops being research and becomes a report you run. Adjusting an interval becomes a routine review instead of a project.
You can run this reliability math against your own numbers in the calibration interval calculator at axiospec.com/tools/calibration-interval-calculator.
Timothy Malone is the founder of Axiospec (axiospec.com), cloud calibration management software for manufacturers and calibration labs.
This article is for general information and education. It is not engineering, legal, or professional advice, and it does not replace your own quality procedures, the applicable standards, your equipment manufacturers’ specifications, or the requirements of your accreditation body or assessor. Calibration interval decisions are yours to make and to validate for your own equipment and risk.
Be the first to comment