1.7 Calibration and Field Checks

After this section you can: read a certificate, say what it does not cover, run a hand-held calibrator check, and decide what to do with a dropped sensor.

A sensor's sensitivity is a measurement, and like any measurement it was made somewhere, by someone, against something, with an uncertainty. The calibration certificate is the record of that measurement. Most sensors on a shaker are calibrated by comparison, back to back: the sensor under test is mounted on a reference sensor on a calibration exciter, so both see the same motion, and the ratio of their outputs times the reference's known sensitivity is the test sensor's sensitivity, at each frequency on the list. ISO 16063-21 is the standard for calibration by comparison. The reference sensor has its own calibration, made against a higher standard, and that chain of calibrations back to a national standard is traceability. MIL-STD-810H Method 514.8, paragraph 4.2.3, asks for calibration traceable to national standards, and NASA's General Environmental Verification Standard (GEVS, GSFC-STD-7000B), in paragraph 1.12, asks for calibration under a recognized program and names ANSI/NCSL Z540.1, Z540.3 or ISO/IEC 17025. A certificate from a laboratory accredited under one of them states its uncertainty and its traceability, and a certificate that states neither is a number with nothing behind it.

The uncertainty is worth reading, because it is spent out of the test's tolerance. Commercial comparison services quote expanded uncertainties of roughly one to two percent across the band most shaker work uses, and larger toward the ends of their range. GEVS says in paragraph 1.13 that its test tolerances include measurement uncertainty, and Method 514.8, in paragraph 4.2.2.2, asks that the complete test control system produce no more uncertainty than one-third of the tolerance the method sets for sine. The sensor's calibration is one part of that budget. The mount, the cable, the conditioner, the converter and the sensitivity entry are the rest, and a chain that has not been proven end to end has an uncertainty nobody knows.

A certificate says what it covers, and what it covers is narrow: this serial number; its sensitivity at a reference frequency, and its response at the frequencies listed, at the amplitude used; on the mount used, usually a stud at a stated torque with grease, on a stiff fixture; at the temperature of the calibration laboratory; on the date it was done. The mount used for calibration is the reference for everything that follows, which is why 1.4 measures every other mount against the stud. What the certificate does not cover is everything about the test: this mount on this fixture, this cable, this conditioner, the sensitivity typed into this channel, the chamber's temperature, a mounting block it was not calibrated with, frequencies outside its list, and anything that has happened to the sensor since the date at the bottom. The certificate says the sensor was right. It cannot say the channel is right today.

How often a sensor goes back is the lab's calibration program's decision, and a year is the common interval. Some events send it back early: a shock beyond its rating, a repair, a reading in a field check that disagrees with its certificate. A sensor past its due date does not go on an article, whatever it read last time.

The field check closes part of the gap between the certificate and the channel. A hand-held calibrator is a small battery-powered shaker with its own reference sensor and control loop, built to put one known motion into whatever is mounted on it. The common class runs at 1 g rms at 159.2 Hz. The frequency is chosen so that $2\pi f$ is 1,000 radians per second, which makes velocity and displacement fall out in round numbers (the arithmetic is the author's): 1 g rms is 386.1 in/s², so the velocity is 0.386 in/s rms and the displacement 0.000386 in rms, or 9.81 mm/s and 9.81 µm in metric units. One setting checks an acceleration, a velocity or a displacement channel. The calibrator has its own certificate and tolerance, a few percent, and its sheet gives the heaviest sensor it can drive; a sensor heavier than that is checked some other way.

The check is run through the whole chain, which is what makes it worth running. The sensor goes on the calibrator by its stud at its torque, on the cable it will use, through the conditioner it will use, into the channel it will use, with the sensitivity, units and detector the test will run. The screen should read 1 g rms, within the calibrator's tolerance and the chain's. When it does not, the size of the miss usually names the fault:

• 1.41 g, high by a factor of 1.41: the detector is set to peak and the calibrator is giving rms (the reverse, a reading of 0.707, needs a calibrator whose level is set in peak, which the common 1 g rms class is not);
• 2.83 g: a peak-to-peak detector against the rms calibrator;
• off by 9.81: meters per second squared read as g, or the reverse;
• off by the ratio of two sensitivities: the channel carries another sensor's number, as in "The sensor that was not the one in the table" in 1.6;
• off by the conditioner's gain: the sensitivity has been entered in two places (1.5);
• off by a few percent more than the tolerances allow: the sensor itself, which goes back for calibration.

The reading is logged against the channel, with the date and the calibrator's serial number. It is the first line of the channel's record for the test.

On a setup of sixty or a hundred channels the check through every chain before every test is real time, and the floor will ask what may be relaxed. The control, limit and abort channels and the independent monitor are never relaxed: an error on any of them is an error in the test or in the evidence. For response channels a lab may write a lighter rule into its procedure, for example the full calibrator check when a sensor is mounted for a test series and a TEDS read with the bias and the tap of 2.5 at each test after that, provided the rule names which channels it covers and the test's owner, the customer where the plan names one, has agreed to it in the plan. A check skipped at the console without that agreement is a deviation, and is logged as one.

The check has limits. It is one frequency at one level. It does not prove the response near the top of the band, the linearity at 50 g, the mounted resonance on the fixture, or the axis the sensor will be on, because during the check the sensor is on the calibrator and not on the fixture. Those are proven in place, by the polarity check, the tap, the noise-floor run and the comparison with the last low-level run, and by the rigid-body check at the first low level (4.4). This section introduces the end-to-end check; 2.5 gives it as a procedure, channel by channel, before the article is at risk.

A dropped sensor is judged, not condemned. A general-purpose accelerometer carries a shock limit on its datasheet of several thousand g, and one manufacturer's calibration guidance puts a hand-height drop onto concrete at typically 2,000 g, with the caution that ordinary handling can change a sensor's characteristics and sometimes damage it. So a drop onto the bench is usually inside the rating and a drop onto the floor is near it, and nothing on the outside shows which. The response is a check, not a tag: read the bias, run the calibrator check against the certificate, and look at the time history on the calibrator for ringing or a shifted baseline. A sensor that reads as its certificate says goes back to work. A sensor whose reading has moved, or that fell from height onto a hard floor, or that is a miniature or shock type with a lower limit, or whose lab's program requires it, goes back for calibration. The author's practice is the check first and the judgment from what it shows, because a rule that retires every dropped sensor retires the drawer.

Many IEPE sensors now carry a transducer electronic data sheet, TEDS, under IEEE 1451.4: a small memory chip in the sensor that the conditioner reads over the same cable. For an accelerometer it holds the model, the serial number, the sensitivity, the reference frequency and the calibration date. It removes the typing, and with it most of the wrong-serial errors. It does not prove that the calibration is current, because the date in the chip is only compared with today if the system is set to do so. It does not prove that the sensitivity in the chip was rewritten at the last calibration. And it does not prove that this sensor is at this location on this axis: the chip travels with the sensor, not with the point on the fixture. When a sensor comes back from calibration, its TEDS is read against its new certificate before it goes back in the drawer.