Temperature measurement in the field

Why is my thermocouple reading maxed out or jumping all over the place?

Saturated high is almost always an open circuit. A jumping reading is almost always a loose terminal, water in a junction box, or a grounding problem.

Most transmitters and PLC thermocouple cards default to upscale burnout. An open element drives the output into the high fault band and the indicator goes to full scale or a sensor-fault code. NAMUR NE 43 puts the valid measurement band at 3.8 to 20.5 mA and reserves ≤3.6 mA and ≥21.0 mA for faults. What comes out of the device depends on how it was configured. A Rosemount transmitter in its standard, non-NAMUR configuration saturates high at 20.8 mA and alarms high at 21.75 mA; configured to NAMUR, the same device saturates high at 20.5 mA and alarms high at 22.5 mA. Set to downscale burnout it saturates low instead — 3.75 mA standard, 3.6 mA NAMUR. Same open circuit, opposite symptom. Check the burnout setting before you read anything into the direction. Then work it in this order:

• Disconnect at the head and ohm the element. A Type K loop reads continuity, typically single digits to a few tens of ohms depending on length and gauge. An open element reads OL. What you want is open or not open. The exact number does not matter.
• Check polarity at every terminal block in the run. One reversed pair makes the indication fall as the process heats. That is a different signature from a saturated reading.
• Open the junction box. Water, green corrosion, a backed-out screw — on a wandering reading that is the first place to go.
• Confirm the shield is grounded at one end only, and find out whether the sheath is grounded to the well. A grounded-junction thermocouple on a heated or electrically noisy vessel can put hundreds of millivolts — sometimes volts — of common-mode voltage on a non-isolated input.

An intermittent that only shows up hot is usually thermal expansion at a crimp, or a cracked junction. Not the transmitter.

Which wire is positive on a Type K thermocouple — what do the yellow and red colors mean?

On a Type K, yellow is positive and red is negative, under the US code in ASTM E230/E230M and the older ANSI MC96.1. In that code red is the negative leg on every type.

The extension-grade jacket is yellow for Type K. Inside, yellow is chromel (Ni-Cr) and positive; red is alumel (Ni-Al-Mn-Si) and negative. IEC 60584-3 agrees with none of that. Type K there is a green jacket, green positive, white negative, and under IEC white is always the negative leg. So identify the standard before you land the wire. A yellow-jacketed cable from one continent and a green-jacketed cable from another are the same thermocouple type.

If the print has worn off, a magnet settles it. Alumel, the negative leg, is ferromagnetic below its Curie point of about 152 °C and gets pulled hard. Chromel does not. One exception: the compensating extension alloys used in place of true KX cable are not necessarily magnetic. The magnet is reliable on the thermocouple itself and on KX extension, less so on a compensating cable.

Can I use regular copper wire to extend a thermocouple instead of thermocouple extension wire?

No. Copper moves the reference junction to wherever the splice is. On a Type J or Type K the error is roughly 1 °C for every 1 °C of difference between that splice and the instrument's cold-junction sensor.

A thermocouple instrument measures its own terminal temperature and adds back the millivolts that correspond to it. The correction only works if the thermocouple alloys actually reach those terminals. Splice to copper in a field junction box that sits 15 °C hotter than the control panel and the reading lands about 15 °C low. It also wanders all day as the sun moves across the box. A splice colder than the panel reads high.

The 1:1 rule holds for Types J and K because their output per degree at ambient is close to their output per degree at process temperature. The noble-metal types are different. Type S puts out about 6 µV/°C near ambient — the ITS-90 table gives 0.113 mV at 20 °C and 0.173 mV at 30 °C — against about 11.7 µV/°C at 1000 °C (9.587 mV at 1000 °C, 10.757 mV at 1100 °C). Type R has close to the same shape. So a 15 °C splice error injects roughly 90 µV, and at 1000 °C that lands as about 8 °C at the indication rather than 15.

Use matching extension wire the whole way — KX for Type K, JX for Type J — including the terminal blocks. On a long run, or hostile routing, the better answer is not more thermocouple wire. Put a head-mounted transmitter at the sensor and run 4–20 mA on ordinary shielded copper pair. That takes out the reference-junction problem, the noise pickup, and the cost of several hundred feet of alloy cable in one move.

Type J vs Type K thermocouple — what's the difference and can I swap one for the other?

They are not interchangeable. A Type J element read by an instrument configured for Type K shows about 398 °C when the process is at 300 °C — an error near 98 °C.

Type J is iron/constantan, about 54 µV/°C near 100 °C (the ITS-90 table gives 4.726 mV at 90 °C, 5.269 at 100, 5.814 at 110). Averaged from 0 to 100 °C it is 52.7 µV/°C; at 0 °C, about 50.4 µV/°C. The iron leg rusts, which makes J a poor choice in wet, humid, or oxidizing service. Type K is chromel/alumel, about 41 µV/°C near 100 °C, and it holds up far better in an oxidizing atmosphere.

Upper limits are a function of wire gauge, not of the type alone. The recommended upper temperature limits for protected thermocouples in conventional closed-end protection tubes are, for Type K, 1260 °C (2300 °F) at 8 AWG, 1090 °C (2000 °F) at 14 AWG, 980 °C (1800 °F) at 20 AWG, and 870 °C (1600 °F) at 24 and 28 AWG. For Type J: 760 °C (1400 °F) at 8 AWG, 590 °C (1100 °F) at 14 AWG, 480 °C (900 °F) at 20 AWG, and 370 °C (700 °F) at 24 and 28 AWG. Those are guides, not absolutes. The wire does get used above them, at the cost of life and stability. And a 1/4 in Type K mineral-insulated assembly is nowhere near 8 AWG, so do not read 1260 °C off a catalog and assume it applies to the probe in your hand.

The millivolt tables are what make them non-swappable. At 300 °C a Type J produces 16.327 mV, a Type K 12.209 mV. Feed 16.327 mV to a Type K input and it reads about 398 °C (the K table gives 16.313 mV at 398 °C and 16.355 at 399). Feed 12.209 mV to a Type J input and it reads about 226 °C. Accuracy is not the reason to prefer one: standard limits of error above 0 °C are identical for both under ASTM E230, ±2.2 °C or ±0.75% of reading, whichever is greater, tightening to ±1.1 °C or ±0.4% for special limits. Below 0 °C they part company. The standard gives no initial tolerance values for Type J below 0 °C at all, because of the characteristics of the materials. If the service runs below freezing, that alone is reason to look at Type T or an RTD.

Why does my infrared thermometer read low on shiny stainless steel, and what emissivity should I set it to?

Bright stainless sits near 0.15 emissivity. It emits very little infrared and reflects nearly everything around it, so a gun left at the default 0.95 reads far below the true surface temperature.

Working values: mirror-polished stainless 0.10–0.20 (some published metal tables run lower still on the brightest finishes), mill-finish or dull stainless 0.20–0.35, heavily oxidized or scaled steel 0.80–0.90, most paints, rubbers, plastics, and insulation jacketing 0.90–0.95. Treat all of those as indicative ranges, not specifications. Do not guess on bare metal. The number swings with finish, oxide, and even the direction of the grain.

Do this instead. Stick a piece of matte black electrical tape (ε ≈ 0.95) on the surface, give it a minute to reach the metal's temperature, set the gun to 0.95, and shoot the tape. If you then need the bare-metal setting, adjust emissivity on the bare surface until it matches the tape reading. Keep the whole spot on the target — a 12:1 instrument at 6 ft sees a 6 in circle. Shoot near perpendicular; off-angle shots collect more reflection and less emission.

Know what you are getting, though. An IR gun reads the outside of the pipe, not the fluid. On an insulated line, a steam-traced line, or anything with a low film coefficient, the wall can be tens of degrees off the process. If the number has to be right, use a contact probe or the installed thermowell.

What's the difference between an RTD and a thermocouple, and which one should I use?

PT100 RTD below about 400 °C, thermocouple above it. The RTD is roughly six times more accurate and far more stable. The thermocouple goes hotter, responds faster, and survives more abuse.

An RTD is a platinum resistor. 100.00 Ω at 0 °C, 138.51 Ω at 100 °C, about 0.385 Ω/°C averaged over that span. IEC 60751 Class A tolerance is ±(0.15 + 0.002|t|) °C, which is ±0.35 °C at 100 °C. But every class is only defined over a stated range, and the range depends on how the element is built. Class A covers −100 to +450 °C wire-wound and −30 to +300 °C thin film. Outside its window the device is a Class B part, whatever the datasheet headline says. The tighter Class AA, ±(0.1 + 0.0017|t|) °C, is narrower still: −50 to +250 °C wire-wound, 0 to +150 °C thin film. The 2022 edition of IEC 60751 added W and F designations to make the wire-wound versus film distinction explicit. A thermocouple is two dissimilar wires making a small voltage — about 41 µV/°C on Type K, standard tolerance ±2.2 °C or ±0.75% of reading.

The rest of the differences are practical. An RTD needs no cold-junction compensation and runs on ordinary copper lead wire; a thermocouple needs matching extension wire all the way to the terminals. A thermocouple can be made from very fine wire with an exposed junction and respond in under a second, where an RTD in a well takes tens of seconds. On range, IEC 60751 tolerance classes run to 600 °C for wire-wound Class B and 500 °C for thin film, and the standard's resistance table extends to 850 °C. In practice the sheath, the cement, and the seals limit an industrial assembly well before the platinum does. Type K runs past all of it.

Vibration is the tiebreaker people forget. RTD elements are wound or thin-film deposited, and they crack under sustained shaking. On a reciprocating compressor discharge, or a hard-shaking line, a thermocouple usually outlives the more accurate device.

What is the difference between 2-wire, 3-wire, and 4-wire RTD, and does the extra wire really matter?

The extra wires cancel lead resistance. It matters: on a PT100, every 1 Ω left uncancelled is about 2.6 °C of error.

2-wire puts the full round-trip lead resistance in series with the element. A 100 ft run of 20 AWG copper is 1.015 Ω per conductor, 2.03 Ω for the pair, and reads about 5.3 °C high. Nor is it a fixed offset. Copper drifts 0.393%/°C, so that pair moves another 0.24 Ω — roughly 0.6 °C — across a 30 °C ambient swing.

3-wire cancels the leads by assuming all three conductors are identical. The instrument measures the element plus one lead, then subtracts a second measurement that captures another lead. Only the difference between those two conductors survives, so 0.1 Ω of mismatch is 0.26 °C. Good enough for essentially all process control — as long as all three conductors are the same gauge, the same length, and in the same cable. 4-wire forces current through one pair and senses voltage on the other, which carries no meaningful current. Lead resistance and lead mismatch both drop out entirely. That is what calibration labs use.

Two other things worth knowing. Excitation current self-heats the element: 1 mA through 100 Ω is 0.1 mW, negligible in flowing liquid, not negligible in still air or in a dry element sitting loose in a well. And if a long run is unavoidable, a PT1000 cuts lead error by 10× (3.85 Ω/°C instead of 0.385), while a head-mounted transmitter that converts to 4–20 mA right at the sensor removes the problem completely.

How do I wire a 3-wire PT100 to a transmitter or PLC analog input — which wire goes where?

Find the two wires that read near 0 Ω to each other. That is the pair joined at one end of the element, and it goes to the two terminals the input treats as common. The odd wire goes to the remaining terminal.

The IEC 60751 color set for a 3-wire PT100 is two red conductors and one white. Do not wire on color alone. JIS practice and individual manufacturers differ, and field cable gets replaced. Meter it first. The two conductors bonded to the same end of the element read under about 1 Ω to each other, which is just lead resistance. Either of them to the odd conductor reads the element plus one lead: about 109.73 Ω at 25 °C, 100.00 Ω at 0 °C, plus a fraction of an ohm of wire.

On most transmitters and analog cards the terminals are marked 1-2-3, or RTD+, RTD−, COMP, with two of them internally tied together (often 2 and 3). The bonded pair goes to those two. The odd conductor goes to the third. Swapping the odd wire with one of the pair rarely damages anything, but it defeats compensation entirely, and the reading then drifts with the lead-wire temperature.

Two installation rules decide whether the compensation works at all. Run all three conductors in the same cable, same gauge, same length. And if the input only accepts two wires, put the jumper at the sensor head, not at the card. A jumper landed at the card compensates for nothing.

What temperature is 110 ohms on a PT100, and how do I convert RTD resistance to degrees?

110.00 Ω on a PT100 is 25.7 °C (78.2 °F). Above 0 °C you convert with the Callendar–Van Dusen equation from IEC 60751.

Forward: R(t) = 100 × (1 + At + Bt²), with A = 3.9083×10⁻³ and B = −5.775×10⁻⁷. Solved for temperature: t = (−A + √(A² − 4B(1 − R/100))) / 2B. Below 0 °C a third term is added, R(t) = 100 × (1 + At + Bt² + C(t − 100)t³), with C = −4.183×10⁻¹². That branch has to be solved iteratively or read off a table. For a PT1000, divide the measured resistance by 10 and use the same equations.

Anchor points worth memorizing:

• −200 °C = 18.52 Ω  |  −100 °C = 60.26 Ω  |  0 °C = 100.00 Ω
• 25 °C = 109.73 Ω  |  50 °C = 119.40 Ω  |  75 °C = 128.99 Ω  |  100 °C = 138.51 Ω
• 200 °C = 175.86 Ω  |  300 °C = 212.05 Ω  |  400 °C = 247.09 Ω  |  600 °C = 313.71 Ω

The 0.385 Ω/°C shortcut is fine near ambient. Further out the curve bends downward: at 175.86 Ω the linear method gives 197 °C against a true 200 °C, about 3 °C of error. Use the equation or a table for anything going on a calibration record.

What's the difference between a 2-wire, 3-wire, and 4-wire 4-20mA transmitter?

This is about how the transmitter is powered, not how the sensor is wired. 2-wire is loop-powered on the same pair that carries the signal. 3-wire has a separate supply that shares a common with the signal. 4-wire has fully separate power and output pairs.

A 2-wire transmitter draws its own operating current out of the loop. Its entire internal circuit has to run on less than the lowest current it will ever output — under about 3.5 mA, below the downscale alarm point. The terminal voltage window is device-specific; read it off the nameplate rather than carrying it in your head. A Rosemount 644 temperature transmitter, for example, is specified for 12.0 to 42.4 V DC at the terminals. That window sets the load limit: max loop resistance = (V_supply − V_min) / 0.022, using 22 mA so the loop still works at the upscale alarm. With a 24 V supply and a 12 V minimum transmitter, that is (24 − 12) / 0.022 = 545 Ω. HART needs at least 250 Ω of real resistance in the loop to work, and the HART specification accommodates total loop resistance from 230 to 1100 Ω.

A 3-wire transmitter takes +24 V and a common, then sources 4–20 mA on the third conductor referenced back to that same common. Convenient. But the output shares the power return, so voltage drop in the common conductor lands directly on the signal. A 4-wire transmitter has its own power (24 V DC or 120 V AC) and a separate 4–20 mA output, usually galvanically isolated. Use it where you need the isolation, or where the device needs more power than a loop can supply.

One more thing: a 3-wire or 4-wire output may be sourcing or sinking. Match it to the receiving card, or the loop will simply sit dead.

How deep should a thermowell be inserted into a pipe for an accurate temperature reading?

The common minimum is 10 times the tip diameter of wetted immersion in flowing liquid, and more than that in gas or low-velocity service. The tip typically wants to sit at least a third of the way across the pipe ID, and never touching the far wall.

The error you are fighting is stem conduction. A well is a metal rod bolted to a nozzle that sits at ambient, and it drags the tip back toward the pipe-wall and flange temperature the whole time. Film coefficient decides how much immersion you need, not process temperature. Gas gives up heat to the well far more slowly than liquid does, and so does liquid at low velocity. The 10× figure is the widely published minimum for a well or sheath in liquid. Gas, steam, and low-velocity service need more, and vendor guidance commonly runs to 15–20×. Treat that as a starting point rather than a code number, and check it against the sensor manufacturer's own installation data for the service.

So a 3/4 in tip in liquid wants at least 7.5 in of wetted immersion, measured from the tip. That is not the U-length on the drawing. U is measured from under the flange face or thread, so a long nozzle, a standoff, or insulation build-up eats into the wetted length before the process ever sees the well. It is also why 2 in and smaller lines usually need an elbow installation with the well pointed upstream into the flow, or a local pipe expansion. A short well poked into a 2 in line is measuring the pipe wall as much as the process.

Two things worth doing once the well is set. Insulate the nozzle and the exposed head — on outdoor or steam-traced lines that can be worth several degrees by itself. And make sure the sensor actually bottoms out against the tip. A spring-loaded sensor that was never pushed home leaves an air gap at the closed end, and turns a fast measurement into a slow, wrong one.

Why did my thermowell snap off inside the pipe?

Almost always vortex-induced vibration. Flow past the shank sheds alternating vortices. When the shedding frequency approaches the well's natural frequency, it resonates and fatigues at the root.

The break is at the base, where bending stress is highest. It is a fatigue fracture, not an overload. ASME PTC 19.3 TW-2016 is the calculation that prevents it, and it checks four things: frequency, dynamic stress, static stress, and pressure rating.

The frequency check compares the vortex shedding frequency against the well's natural frequency as a ratio, r = f_s / f_n. The limit is not a single number. r < 0.8 applies only to low-density gas that meets the standard's damping conditions; for liquids and other media the limit is r < 0.4. A second resonance hides inside that band. In-line excitation occurs at twice the shedding rate, which puts an in-line resonance at r = 0.5, and the standard treats that case separately. The 2016 edition (para. 6-8.5) allows a design to pass through the in-line resonance only when the medium is gaseous, the in-line resonance range does not overlap continuous plant operation, and the well withstands 10¹¹ cycles without exceeding the allowable flexural fatigue stress. Shedding frequency comes from a Strouhal number that sits near 0.22 through the ordinary turbulent range; the 2016 edition makes it a function of Reynolds number rather than the fixed value the older standard used. Run the actual calculation, at every operating condition and not just the design point. Do not work from this summary.

Geometry and velocity drive the failure. A long, straight, thin well has a low natural frequency. Shedding force rises with the square of velocity while shedding frequency rises in direct proportion to it, so a well that was fine at 3 m/s can both quadruple its loading and double its excitation frequency at 6 m/s. The fixes are a shorter U-length, a tapered or stepped shank, a heavier root, or a support collar. And rerun the calculation whenever velocity, density, or U-length changes, not only at original design.

Be clear on one thing. A thermowell is part of the pressure boundary. A broken well in a live line is a containment and mechanical-integrity problem, and it belongs with a piping or mechanical engineer before it belongs to the instrument shop.

How do I calibrate an RTD or thermocouple transmitter using a decade resistance box or thermocouple simulator?

Replace the sensor with a known resistance or millivolt source at five points — 0, 25, 50, 75, and 100% of range — and compare the transmitter's output against 4.000, 8.000, 12.000, 16.000, and 20.000 mA.

Know what this proves. The element is out of the circuit, so you are checking the transmitter's input, linearization, and output. It is a transmitter calibration, not a temperature calibration. It says nothing about whether the sensor in the well is telling the truth.

• PT100 ranged 0–100 °C: source 100.00, 109.73, 119.40, 128.99, and 138.51 Ω.
• Type K ranged 0–400 °C: source 0.000, 4.096, 8.138, 12.209, and 16.397 mV, all referenced to 0 °C.
• Wire the decade box in the same 3- or 4-wire configuration the real sensor uses, with short leads of matched length. On a 3-wire input, extra clip-lead resistance is indistinguishable from process temperature — 1 Ω of it is 2.6 °C.

On thermocouple work the reference junction decides everything. Either let the calibrator apply its own internal cold-junction compensation for the selected type, or set the transmitter to a fixed 0 °C reference and source raw table millivolts. One or the other. Never both. Use actual thermocouple extension wire from the calibrator to the transmitter: copper test leads move the reference junction to the calibrator and add the temperature difference between the two ends as straight error.

Run every point upscale and then back down to expose hysteresis, and record As-Found values before you touch a trim. A transmitter that is out at exactly one point of a five-point run is telling you something other than "needs adjustment."

How do I calibrate an RTD using a dry block calibrator?

Put a traceable reference probe in the block alongside the RTD under test, at the same depth, and read both at every setpoint. The block's own display is a controller readout, not a standard.

Immersion is the dominant error in dry-block work, usually larger than anything wrong with the sensor. Published guidance puts minimum immersion at about 15× the probe diameter, or — where the sensitive length is known — the sensitive length plus 10× the diameter. Insert both probes to the full depth of the insert well and match the two depths exactly. If you know where the sensing elements sit, line their centers up. Use inserts drilled to a close fit for each probe. A sloppy hole leaves an air gap between the probe and the block, and the reference and the test unit stop seeing the same temperature. The axial gradient in a dry block — the difference from the bottom of the well to the top — is typically the single largest uncertainty component. A probe short of full depth is not merely less accurate. It is reading a different place.

Allow real stabilization time. 10 to 20 minutes per setpoint is common practice, not a value fixed by any standard, and nothing gets recorded until the reference has settled to within the block's stated stability. Work the setpoints in one direction, then repeat in the other. Write down As-Found readings at every point before you trim anything.

Two limits to be honest about. Calibrating a loose element does not prove the installed measurement. Well fit, insertion depth, and stem conduction in the actual pipe are usually larger than the sensor's own error. And a dry block with a shop reference probe will not produce a calibration certificate with a stated uncertainty traceable to a national standard. That takes an accredited calibration laboratory. Better to say so than to issue a sticker that implies more than it can support.

When this page runs out

When this page runs out of usefulness, here is where to go. If you need an accredited calibration certificate with a stated uncertainty, an accredited calibration laboratory is the right call. If a thermowell has failed in a live line, that is a pressure-boundary question for a piping or mechanical-integrity engineer before it is an instrument question. If a device is in warranty and genuinely defective, send it to the manufacturer. What is left over — a loop that reads wrong, a well that reads slow, a panel of wire nobody documented, an application and an instrument that were never matched to each other — is field work. That is what we do: support@bridgesindust.com.