IMDHMIControls LibraryShop main store

Troubleshooting

FX5U thermocouple readings need a dedicated TC input

Controllers and I/O Published 9 min read

Short answer

A Type K thermocouple on the FX5U built-in analog input cannot yield usable temperature: the voltage channel's count step is worth about 61 °C of Type K signal, and a reading near 200,000 means the program is reading the wrong operand or width, not temperature. Verify the raw 0–4,000 count path, then move the thermocouple to a dedicated input adapter with cold-junction compensation and read degrees directly.

On this page
  1. Why does my FX5U thermocouple reading show almost 200,000?
  2. How fine is the FX5U built-in analog input in millivolts?
  3. What does one count mean in Type K degrees?
  4. How do I verify the raw analog value before scaling?
  5. Why does a thermocouple input need cold-junction compensation?
  6. Which hardware reads a Type K thermocouple on an FX5U properly?
  7. How do I set up the thermocouple adapter in GX Works 3?
  8. When do I keep the native input and when do I fit the adapter?
  9. Questions asked

The number near 200,000 is not a temperature, and no label type or scaling constant will turn it into one on this input. The FX5U's built-in analog channels convert full-scale voltage into a 0–4,000 count integer, and a Type K thermocouple moves only about 41 μV per °C — so the best this channel can ever resolve is roughly 60 °C of thermocouple signal per count. The oversized value itself comes from the data path: a wrong operand, a mis-sized read, or bits merged in from the neighboring register. Prove the raw path with a forced voltage, then fix the measurement where it is actually broken: a dedicated thermocouple input adapter that applies cold-junction compensation and hands the program degrees.

Why does my FX5U thermocouple reading show almost 200,000? #

A value near 200,000 is not a thermocouple signal: the FX5U built-in input produces 0 to 4,000 counts across its documented full-scale voltage, so 200,000 is about fifty times full scale and points to a data-path error rather than a temperature conversion.

The documented conversion span ends at 4,000, which fits in twelve bits of data; 200,000 needs eighteen. The extra digits enter from somewhere other than the measurement, and the candidates in checking order are:

  • A read wider than the documented result: when an instruction grabs more data than the channel's conversion word holds, the neighboring register merges into the value and the magnitude changes entirely.
  • The wrong operand behind the float label: a status word, a setting register, or an unrelated device instead of the channel's conversion result.
  • Scaling applied too early: a multiply or engineering-unit constant evaluated before anyone looked at the raw count hides the fault behind a plausible magnitude.
  • Bit reinterpretation instead of numeric conversion: treating the integer's bit pattern as floating-point changes what the bits mean; it cannot repair a wrong source.

The Float label itself is a red herring. Assigning an integer to a floating-point label converts the numeric value; neither operation repairs a wrong source or a merged word. Watch the operand the program actually reads, as a plain integer, and compare it against the documented span — that single observation separates a data-path fault from a measurement problem. Reading the raw value before trusting any conversion is the same discipline behind troubleshooting non-linear 4-20 mA readings on an S7-1200 analog input.

How fine is the FX5U built-in analog input in millivolts? #

The native channel is documented for a full-scale voltage with a 0–4,000 count conversion result and 115.7 kΩ input resistance, so a thermocouple is barely loaded at all — the problem is never signal loss, only resolution and missing cold-junction information.

Loading arithmetic backs that up: a thermocouple loop presents a few ohms, and a few ohms against 115.7 kΩ shifts a 10 mV reading by under a microvolt, three orders of magnitude below the gap between one count and the next. Signal strength is not the issue either; a healthy Type K delivers roughly 1 mV to 50 mV across normal process temperatures, comfortably inside the input's range. Nothing in this input chain was designed for a thermocouple, which is exactly what the next numbers show.

These figures are the FX5U built-in inputs at the full-scale voltage setting; other input modes and other FX5 analog hardware document their own spans and steps.

What does one count mean in Type K degrees? #

Three lines of arithmetic settle it: the full-scale 10 V span spread over 4,000 counts is 2.5 mV per count, one count at 41 μV per °C is about 61 °C of thermocouple signal, and a 250 °C process point is only a 4-count signal.

Worked example, on the documented voltage span with standard Type K reference data:

  • Counts to millivolts: 10 V is 10,000 mV across 4,000 counts, and 10,000 ÷ 4,000 = 2.5 mV per count.
  • Millivolts to degrees: Type K sensitivity is about 41 μV/°C in the working range (roughly 38 to 43 across it), and one 2.5 mV step ÷ 0.041 mV/°C ≈ 61 °C per count, call it about 60 °C.
  • Degrees to counts: a Type K at 250 °C generates about 10.2 mV against a 0 °C reference, and 10.2 ÷ 2.5 ≈ 4 counts, so the whole operating point sits four steps from zero.
  • The symptom re-checked: 200,000 ÷ 4,000 = 50 full-scale spans, so no valid conversion result on this channel can produce the number you are seeing.

Two conclusions follow. Even with a perfect data path, a converter that steps in 61 °C increments cannot indicate a process measured in tens of degrees; the channel is finished as a thermocouple input. And decimals added after conversion change the display, not the converter — the resolution was lost in the analog-to-digital step, and no instruction gets it back.

How do I verify the raw analog value before scaling? #

Read the channel's conversion result as the documented integer and expect 0 to 4,000 before any scaling: a forced 5.00 V test that returns about 2,000 counts proves the path is sound and puts the fault downstream; anything else convicts the operand.

  1. Locate the documented storage for the built-in channel's conversion result in the FX5 analog-control manual; it lists where the value lands and in which format for the configured mode.
  2. Force a known input. Disconnect the thermocouple, apply 5.00 V DC from a loop calibrator to the channel terminals, and expect about 2,000 counts — half of full scale, by the per-count arithmetic in the section above. Repeat at 1.00 V and expect 400; two points tell an offset error from a scale error.
  3. Watch that operand live in GX Works 3 as a plain integer while the voltage is applied. If it tracks the forced voltage inside the documented span, the path is sound; if it does not, correct the operand and the read width before touching anything else.
  4. Re-scale only from documented endpoints, in floating-point arithmetic so integer division cannot truncate the fraction; a hand check such as 1,240 counts × 2.5 mV = 3,100 mV confirms the multiply before the program runs it.
  5. Build forced-point checks like this into commissioning for every analog channel; the remote I/O commissioning checklist for a modular station lays out the same channel-by-channel verification.

Why does a thermocouple input need cold-junction compensation? #

A thermocouple measures the difference between its measuring junction and the terminals it lands on, so the terminal temperature must be measured and subtracted; a plain voltage input has no sensor at the terminals, so the result drifts with panel temperature.

The millivolts correspond to a temperature difference, not a temperature. On standard Type K data a 250 °C junction produces about 10.2 mV, but only against 0 °C terminals; with the panel at a realistic 30 °C the same junction delivers about 8.9 mV, and looking 8.9 mV up in the table lands near 220 °C — 30 °C low, which is exactly the terminal temperature. Compensation closes the gap: measure the terminal temperature with a second sensor, compute the emf that junction would contribute, add it to the measured millivolts, and convert the corrected total to temperature.

A general-purpose voltage input has no terminal sensor and no linearization for the K curve, so both jobs fall to the program, and a finer-resolution voltage module would only shrink the count step while leaving the terminal temperature unmeasured. Better resolution on the wrong quantity is still the wrong quantity. The mechanism deserves its own depth: why thermocouple cold-junction compensation matters explains the reference junction and the correction loop.

Which hardware reads a Type K thermocouple on an FX5U properly? #

Move the Type K to a dedicated thermocouple input adapter — the 4AD-TC-ADP identified for this CPU family — which applies cold-junction compensation internally, linearizes the curve, and reports temperature in degrees instead of raw counts.

That adapter does internally what the voltage path cannot: it measures the terminal temperature per channel, applies the correction, linearizes the thermocouple curve, and hands the program engineering units. Before ordering, confirm in the FX5 analog documentation the complete catalog designation, support for your exact CPU unit, the supported thermocouple types and measuring ranges, and the mounting rules; those details decide fit, not the short designation alone.

Wire it like the millivolt circuit it is: polarity consistent end to end, Type K extension or compensation cable from probe to terminals instead of copper, shields landed per the manual, and the low-level leads routed away from power and drive wiring — de-energize the cabinet section before opening the terminals, because induced noise on a thermocouple circuit reads directly as temperature.

An op-amp stage, the other fix people reach for, amplifies the millivolts but adds its own offset and drift, and it still leaves the terminal temperature unmeasured — the 30 °C error from the section above would ride straight through it, only larger on screen. Amplification is not the missing function; compensation is.

How do I set up the thermocouple adapter in GX Works 3? #

Add the adapter to the FX5U project's module configuration in GX Works 3, set the channel type to K, and read degrees directly from the adapter's documented output range — no millivolt stage, no lookup table, no scaling constants.

  1. Add the adapter in the project's module configuration so the CPU detects it at power-up and allocates its data.
  2. Set the channel to thermocouple input, select Type K, and set the engineering unit the manual specifies; configuring Type K thermocouple cold-junction compensation walks through the compensation side of that setup.
  3. Read temperature in the program from the adapter's documented output range, using the data type the manual states for the channel; the values arrive compensated and linearized, so the mV math and lookup tables come out of the logic rather than getting patched.
  4. Prove the channel before trusting it: source 250 °C from a thermocouple calibrator and expect a reading near 250 °C within the documented accuracy, or sink the junction in an ice bath and expect near 0 °C. Checking a thermocouple with cold-junction compensation covers the verification procedure.
  5. Trend the value through a warm-up and compare it against a second reference; a correctly compensated channel holds steady while the cabinet temperature climbs.

When do I keep the native input and when do I fit the adapter? #

If the reading feeds a temperature indication, alarm, or interlock, fit the thermocouple adapter and retire the float experiment: even a flawless program leaves the built-in voltage input resolving Type K in about 60 °C steps, which no temperature loop can tolerate.

  • Thermocouple temperature: dedicated adapter, Type K configured, degrees out; the built-in input tops out near 60 °C per count and no software raises that ceiling.
  • Bench confirmation that a thermocouple is alive: the native channel is fine — its coarse count steps will show millivolts appearing as the junction heats, with the terminal temperature factored in by hand.
  • Standard process signals: keep the built-in input, which is documented for 4–20 mA into 400–2,000 counts at 10 μA resolution and for voltage input into 0–4,000 counts.
  • Any analog channel reporting an impossible value: repair operand and width first, then choose hardware; a corrected data path sometimes dissolves the fault that started the hardware hunt.

For the FX5U-32MR/DS in this fault: order the thermocouple adapter, wire the Type K with extension cable, select type K in GX Works 3, verify against a calibrator, and delete the float-label experiment. From the next power-up, the program reads a temperature the hardware actually measured.

Questions asked

Can I get raw millivolts out of the thermocouple adapter?

The adapter applies cold-junction compensation as part of the measurement and reports the compensated temperature, so there is no clean way to read the uncompensated thermocouple voltage from it. If a diagnostic genuinely needs millivolts, source a known temperature from a calibrator and compare against the Type K reference table, or measure the loop with an isolated millivoltmeter. Design the control around degrees, not volts.

How accurate will a Type K reading be once the adapter is fitted?

Total error is the sum of the adapter's documented accuracy, the cold-junction sensor's tolerance, and the thermocouple wire's own tolerance. Standard-grade Type K wire is allowed ±2.2 °C or ±0.75% of reading under ASTM E230, whichever is larger, and that wire tolerance often dominates the PLC-side error. Take the adapter figure from its manual and add the wire class for the number you can defend.

Does the wiring between the thermocouple and the adapter matter?

Yes. Run Type K extension or compensation cable from the probe to the adapter terminals, keep polarity consistent end to end, and stop using copper after the thermocouple plug, because a copper junction creates a second reference point the compensation cannot see. A quick field check: the positive Type K leg is non-magnetic and the negative leg is slightly magnetic.

Will averaging or filtering the native channel make it usable?

No. Averaging reduces random noise, and filtering smooths the trace, but neither adds information the converter never captured: the input still reports in the same coarse millivolt steps, each worth about 60 °C of Type K signal. The reading would look calmer while being just as coarse. Resolution has to exist at conversion, which is why the fix is hardware.

Keep reading

  • Troubleshooting

    Troubleshooting non-linear 4-20 mA readings on an S7-1200 analog input

    A 4-20 mA loop is linear by construction, so counts swinging between 10 and 30246 mean the current at the terminal is not what the sensor LCD shows. On the SM 1231, 27648 counts is the top of the 4-20 mA range and 30246 decodes to 21.5 mA over-range fault current. Check the loop with a series meter, split sensor from module with a calibrator, and anchor NORM_X to 0 and 27648.

    Controllers and I/O

  • Troubleshooting

    Wiring a 1769-OF2 analog output to a PowerFlex 753 input

    Wire it directly: the 1769-OF2 channel sources the loop current and the PowerFlex 753 analog input set to current mode is a passive 93 Ω receiver, so no isolator is needed. Set the card's power switch to backplane, land only two wires on AI+ and AI-, and hunt repeated input-card failures in a second 24 V source, a shared 24 Vdc/0 Vdc common, or an unverified replacement card.

    Controllers and I/O

  • Selection guide

    Add 4-20 mA analog inputs when the CompactLogix rack is full

    Swap the installed four-channel analog input card for an eight-channel one when channel isolation matches; that adds four inputs against a need of three, with no rack, power-supply or enclosure change. If isolation rules the swap out, reposition the existing power supply to free an end slot. When the cabinet itself is full, add a POINT I/O analog module or IO-Link converters over EtherNet/IP.

    Controllers and I/O