A stable offset that grows from input 1 to input 8 and ignores probe swaps is a terminal temperature gradient at the card, not sensor noise. Cold junction compensation adds each terminal's difference from the card's compensation sensor straight into the reading. Four 130 CFM fans 2 inches below the base of the panel behind these numbers brought an 8 °F spread down to 2 °F or less per card.
When every channel on a thermocouple input card sits at a steady offset that grows from input 1 to input 8, and swapping probes moves nothing, the probes are fine. The card's cold junction compensation is reading a temperature the terminals do not share. Heat from the card stack and the panel leaves one end of the terminal row warmer, and the difference lands in the readings one for one: a transition junction 5 °F away from the compensation sensor shifts its channel about 5 °F. In the panel behind these figures, an 8 °F climb across every card disappeared once air moved through the base. This page shows the two tests that confirm it and the fix that held.
Symptoms and what each points to
| What you see | What it points to | First move |
|---|---|---|
| Offset steady within ±0.3 °F and growing from input 1 to input 8 | Terminal temperature gradient at the card feeding cold junction compensation | Swap test, then an IR check of the terminal row |
| Probes swapped between inputs 1 and 8 leave the pattern in place | The error follows channel position, so the probes are not the fault | Jumper test at the card |
| One probe walked across all 8 inputs reads higher at every step | Same confirmation: the error belongs to the input position | Jumper test at the card |
| About 4 inches of copper or thermocouple wire jumped at the card reads consistent | The spread lives in the junction temperatures of the field wiring or the terminal row | Even out the terminal temperatures, then retest |
| Readings wander or jump with load | A different fault: noise, a loose or mixed-metal termination | Start with erratic analog input readings: check the base first |
| Card display, PLC value and HMI screen all agree | The chain is consistent, so the error sits upstream, in the measurement circuit | Test at the card terminals |
In the case behind this page, the first four rows were true at once: Type K probes with ungrounded tips and 6 to 10 ft of lead, all 8 inputs tied to one process point, and the climb showed up in the shop and again at a desk during initial programming. Shielded cable and ferrite cores had changed nothing, which fits. Interference usually shows up as wandering values, and these readings held a steady, position-locked offset.
How the terminal temperature gets into the reading
A thermocouple loop produces microvolts in proportion to the temperature difference between its measuring junction, out at the process, and its transition junction, the point where the thermocouple alloy meets copper at the card or its terminal block. Cold junction compensation measures the temperature near those transitions and adds it after the conversion, which is what turns a microvolt reading into degrees. Two facts follow. First, a transition junction 5 °F from the compensation sensor shifts that channel 5 °F, and the proportion holds at any size, because the card's own conversion cancels the microvolts. Second, the compensation term is added at the card, so no correction downstream, in the PLC scaling or on the screen, can remove it.
Every alloy change in the path adds another junction with its own temperature. Thermocouple wire landing on a copper terminal block creates one; if that block sits at a different temperature from the sensor, the same error appears. Card designers hold the transitions together on an isothermal block, a mass of metal at one temperature, and a row heated unevenly defeats that intent.
Direction tells you where the sensor sits. In this case the input 1 terminal area ran 5 to 6 °F hotter than input 8 while channel 1 read lower, so the compensation sensor reads a temperature close to the hot end's. The exact position lives in the card's documentation; the fix is the same wherever it sits: even out the row.
Checks: the swap test and the jumper test
- Record all channels at once with the process steady. Swap the input 1 and input 8 probes and record again. If the low-to-high climb stays with the channel numbers, the card side owns the fault. If it moves with a probe, test that probe on another input or substitute a known one.
- Take the field wiring off and jump each input at the card with about 4 inches of one material, copper wire or thermocouple wire. Leave extension wire and intermediate terminal blocks out of the test. Consistent readings across the channels put the spread in the junction temperatures of the field wiring or the terminal row itself.
- Reinstall the field wiring and confirm the original readings return, so the test itself changed nothing.
- Shoot the terminal row with an IR gun while the climb is present and compare the input 1 area with the input 8 area. Treat the number as a screening measurement of the surface. In this case the input 1 area read 5 to 6 °F hotter, matching the climb's direction and size.
Skip the grounding and shielding experiments until these two tests say something. In this case, lifting the DC common from earth ground changed nothing, and shields and ferrites changed nothing, exactly what a steady position-locked offset predicts. Interference has its own signature, and the checks for wandering analog values start at the base and its wiring.
Example: an 8 °F climb from input 1 to input 8 on a Type K card
Inputs, from the case this page works from:
- Card and sensors: 8-channel thermocouple input card, Type K, ungrounded tips, leads 6 to 10 ft
- Test condition: all 8 inputs tied to one process point
- Readings: an 8 °F climb from input 1 to input 8, steady within ±0.3 °F
- Terminal row, IR screening: input 1 area 5 to 6 °F hotter than input 8
Rule, in words: the card converts the loop's microvolts to degrees on the Type K scale, then adds the temperature its compensation sensor reads. A channel's error equals the sensor's temperature minus that channel's transition junction temperature.
Arithmetic: put the sensor near the hot end, at terminal 1's temperature.
- Channel 1: error = T(sensor) − T(terminal 1) = 0 °F
- Channel 8: terminal 8 sits 5.5 °F cooler, the midpoint of 5 to 6 °F, so error = T(sensor) − (T(sensor) − 5.5 °F) = +5.5 °F, channel 8 reading high
- Gap between channels 1 and 8: 5.5 °F, climbing toward input 8
The observed climb is 8 °F, so the terminal gradient accounts for 5.5 °F of it and the surface reading leaves 2.5 °F unaccounted: an IR gun reads the terminal's surface, the alloy junction can sit a couple of degrees from that surface, and the sensor sits somewhere off the very end of the row. Both numbers agree on the cause and the direction.
Result: a terminal-to-sensor spread of roughly 8 °F produces the observed 8 °F climb. The same steady gradient produces the steady offset, which is why the readings only wander within ±0.3 °F.
Decision: when the reading spread matches the terminal spread within about 2 to 3 °F, the card is healthy and the layout is the fault. Equalize the terminal row and hold the spread at 2 °F or less per card, a figure the card's maker counts as within specification.
The fix: move air through the base and land the wire short
- Mount the base vertically on the panel so convection runs along the card row. Both bases in this case were vertical already, so check yours before adding hardware.
- Move air through the base. The change that held: four 130 CFM, 4-inch-square AC fans mounted 2 inches below an 11-slot base carrying ten thermocouple cards and one RTD card. Every card came down to 2 °F or less, and all 80 thermocouples landed within 3 °F of each other.
- Land thermocouple wire at the card directly where you can. Every alloy change adds a junction whose temperature enters the measurement, so keep unavoidable terminal blocks close to the card and in the same air.
- The same trapped warmth that skews the card also stresses nearby screens; ghost touch on a touchscreen at high temperature has checks of its own before any screen gets replaced.
After the fix
- Trend the channels over a full shift with the enclosure closed and hold the spread within 2 °F per card.
- Re-shoot the terminal row with the IR gun. The 5 to 6 °F end-to-end difference should be gone or down to a degree or two.
- Re-run the swap test with the field wiring back in place, then verify one known process point before returning the loop to service.
- When the value reaches a screen, confirm the screen agrees with the card and the PLC; in this case all three matched, which is the outcome you want once the layout fault is gone. The display side has its own routine: display a thermocouple temperature on an HMI panel.
One safety note: until a corrected loop is checked against a known process temperature, treat its readings as suspect; anything that trips or controls on them should be validated against a known process temperature before you trust it.
Recommendation: when thermocouple readings hold a steady offset that climbs with channel number, run the 4-inch jumper test at the card before touching probes, shields or grounding. If the jumpers read consistent, even out the terminal row by moving air through the base and land the thermocouple wire as short and as direct as you can, holding the result to 2 °F or less per card. If the jumpers still climb, keep the card and its local thermal environment under investigation with the card's maker.
Questions
Why do copper jumpers make all channels read the same?
With the same metal on both legs, the loop carries almost no thermoelectric voltage, so every channel reports a temperature close to the card's compensation sensor. Thermocouple-wire jumpers work nearly as well, because a junction twisted or welded 4 inches from the card sits near the terminal temperature. The jumper test is a discriminator between card and field wiring. Once the field wiring is back, verify one known process point before trusting the loop.
Can a per-channel offset in software correct the gradient?
Only while the gradient holds still. The offset tracks the temperature difference across the terminal row, which moves with card loading, enclosure load and season, so a fixed software correction drifts with it. Keep software offsets for a known sensor trim. For a position-dependent spread, even out the terminal temperature; that removes the error where it starts and keeps the correction from being wrong half the year.
Does the same gradient hit an RTD card on the same base?
Not the same way. An RTD card measures resistance at the probe, so terminal temperature adds none of the thermoelectric error that thermocouple wiring carries; lead resistance and terminal contact still matter on their own. In the panel behind these figures, ten thermocouple cards shared one base with a single RTD card, and the airflow change was made for the thermocouple rows.
Planning information, not legal or engineering advice. Rules differ by place: check codes and permits with your local authority and your licensed trades.