How to calibrate a pressure transmitter: as-found and as-left

The short answer

Apply five pressures, 0, 25, 50, 75 and 100 per cent of span, first rising and then falling, and record the milliamp output at every point before you adjust anything. That is the as-found data. Express each error as a percentage of span, compare it with the tolerance in your procedure, and adjust only if a point fails. Then run the five points again and record that as the as-left data. Both records together are the calibration; either one on its own is just a note.

What you need, and the 4:1 rule

A calibration needs a pressure source, a reference standard and a way to read the output. In practice that is a hand pump with a reference gauge or, more usually now, a documenting process calibrator of the Fluke 754 or Beamex MC6 class, which generates or measures pressure, supplies 24 V loop power and reads the current in the same box. For the most accurate work a deadweight tester is still the reference, typically quoted around 0.015 to 0.025 per cent of reading.

The reference has to be better than the instrument under test. The common rule is a 4:1 accuracy ratio: the standard should be at least four times better than the tolerance you are proving. Proving a transmitter to plus or minus 0.5 per cent of span therefore wants a standard good to about 0.125 per cent or better. Some quality systems have replaced the flat 4:1 rule with a documented uncertainty budget; if yours has, use that. Either way the standard needs a current traceable certificate, or the exercise proves nothing at all.

The 4-20 mA relationship

For a linear transmitter:

mA = 4 + 16 x (PV - LRV) / (URV - LRV)

and in reverse:

PV = LRV + (mA - 4) / 16 x (URV - LRV)

Span is URV - LRV. Because the span is carried by 16 mA and not by 20, one milliamp is 6.25 per cent of span and 0.16 mA is one per cent. Those two numbers turn a meter reading into a judgement without a calculator.

What the PLC sees

The analogue input module converts the current to an integer. On Siemens S7 hardware a channel configured for 4 to 20 mA delivers 0 at 4 mA and 27648 at 20 mA, with defined over-range and under-range bands either side and a distinct value for an open circuit, so:

counts = 27648 x (mA - 4) / 16

Rockwell and other vendors differ. Some modules hand you a float already in milliamps, others a raw count over a configurable range, so read the module manual rather than assuming 27648. The scaling block in the program then maps counts to engineering units, and it will happily turn a wrong measurement into a confident-looking number on the HMI.

NAMUR NE 43 and why the zero is live

NE 43 defines what a transmitter does when it knows it is broken. The measuring range is 3.8 to 20.5 mA. Below 3.6 mA or above 21.0 mA is failure information rather than a measurement, and the receiving system should raise a fault instead of clamping the value to 0 or 100 per cent.

This only works because the zero is live. If 0 per cent were 0 mA, a cut wire, a dead transmitter and a genuine zero reading would be indistinguishable. With 4 mA at the bottom of the range, a broken loop reads 0 mA, lands well below 3.6, and is unambiguous.

Whether a particular transmitter drives downscale or upscale on failure is a configuration choice, often a jumper or a HART parameter. It has to match the safe direction of the loop and it has to match what the trip logic expects.

The five-point run

Vent to atmosphere, connect, apply full-scale pressure once and release it to exercise the sensor, then start. Apply 0, 25, 50, 75 and 100 per cent going up, then 75, 50, 25 and 0 coming down, letting each point settle. Never overshoot and come back to a point on the way up: that mixes hysteresis into the rising data and hides it. Record the pressure the standard reports, not the value you were aiming for.

As-found and as-left

As-found is the state you found the instrument in, before any adjustment. It answers a question the as-left data cannot: was the plant measuring correctly for the last twelve months? A transmitter found 2 per cent high puts everything made since its last calibration in question, and ISO 9001 clause 7.1.5.2 expects that assessment to be made and recorded. In a GxP plant it opens a deviation.

As-left is the proof the instrument is fit to go back into service. If nothing was adjusted, say so: as-left equal to as-found is a valid and common result.

The as-found series is also the only evidence of drift, and drift is what sets the calibration interval. A team that adjusts first and writes afterwards has no drift history, and therefore no basis for lengthening or shortening the interval.

Worked example: a 0 to 10 bar transmitter

Tolerance is plus or minus 0.5 per cent of span, which is 0.08 mA.

Error as a percentage of span is the milliamp difference divided by 16, times 100. The worst rising point is 50 per cent: 0.130 / 16 x 100 = 0.81 per cent. In process terms that is 0.081 bar, and on a Siemens analogue input about 225 counts.

Look at the shape before reaching for a screwdriver. A pure zero shift moves all five points by the same amount. A pure span error grows linearly from nothing at 0 per cent to a maximum at 100. This one is small at both ends and largest in the middle, which is non-linearity, and no amount of zero and span adjustment will remove it.

Hysteresis is the difference between the rising and falling reading at the same point: 0.025 mA at 25 and 50 per cent and 0.020 mA at 75, so 0.16 per cent of span at worst. Small, and not the problem here.

The decision. Three of five points are outside plus or minus 0.5 per cent, so the instrument fails and must be adjusted. Because the error is a bow and not an offset, the correct action is a sensor trim against the standard, not a re-range. After trimming, the as-left run read 4.002, 8.012, 12.018, 16.014 and 20.006 mA, a worst error of 0.11 per cent of span. Record both tables, raise the as-found result as an out-of-tolerance event, and review what the loop has been used for since the last calibration.

The mistakes people make

Adjusting before recording. The commonest and the worst. The evidence is gone and cannot be recreated.

Three points, rising only. Three points cannot show a bow and a rising-only run cannot show hysteresis.

Calling a re-range a calibration. Changing LRV and URV over HART changes what the milliamps mean. It does not touch the accuracy of the sensor, so a transmitter 0.8 per cent out before is still 0.8 per cent out afterwards.

Reading the HART digital value instead of the current. The digital PV is taken before the digital-to-analogue converter. The PLC reads the current. Measure the current, with a meter in series at the marshalling or PLC terminals where you can.

Calibrating out of the installed position. Mounting attitude puts a zero offset on a differential-pressure cell in particular. Zero it where it will live, at line static pressure, with the equalising valve operated in the right order.

Forgetting the bypass. Put the loop on maintenance override before you start and take it off afterwards. Both halves of that sentence have caused unplanned trips.

What to learn next

The wiring half is a separate skill: see 4-20 mA current loop troubleshooting for when the number is wrong for electrical reasons, and analogue scaling with NORM_X and SCALE_X for the PLC side of the same signal. For the instruments themselves, RTDs, thermocouples and transmitters, and for the commissioning context, PLC I/O loop checking.

Frequently asked questions

How often should a pressure transmitter be calibrated? The interval follows from measured drift and the consequence of being wrong. Annual is a common default for general service, six-monthly or shorter for custody transfer and safety-related loops, and longer intervals are defensible only with as-found history that supports them.

Is five points required by a standard? No standard forces five. Five up and down is the usual compromise; some procedures use three, and safety-related loops often use more. What matters is that the procedure is written down and followed identically every time.

Can I calibrate a transmitter in service? Only with the loop bypassed, the process correctly isolated and the control room informed. Isolating a live differential-pressure transmitter in the wrong order can damage the cell and leave the level reading wrong for the rest of the shift.