Overview: The Failure Pattern
A retrofitted machine arrived with four Type K thermocouples controlling a silicone-heater-driven aluminium platen. Three of the four sensors were built into the assembly so they cannot be withdrawn without destroying the silicone patch bonded over them. Two independent design faults were present at once:
- Wrong sensor technology for the range. The controlled process band is 90-120 °C. Type K thermocouples were used where a Pt100 RTD gives materially better resolution, repeatability and drift behaviour.
- Wrong sensor location. The measuring junctions sit on the back face of the silicone heater, not on the workpiece. The loop therefore regulates heater surface temperature, not the temperature of the object being heated.
The heated object is a duralumin plate roughly 15 mm thick and about 1 m long. The junctions themselves were hand-twisted wire ends rather than a proper welded bead. This article covers why each choice is wrong, how to quantify the resulting error, and how to retrofit sensors correctly on an assembly that was never designed for sensor replacement.
Sensor Technology: Why Pt100 Wins at 90-120 °C
Thermocouples earn their place above roughly 400-500 °C, in fast-response probes, or where the sensor is consumable. None of those conditions apply to a 90-120 °C platen. The comparison that matters for this loop:
| Attribute | Type K thermocouple | Pt100 RTD (IEC 60751 class) |
|---|---|---|
| Output type | Low-level DC voltage from a temperature difference | Resistance, absolute measurement |
| Typical sensitivity near 100 °C | Tens of microvolts per °C | Fractions of an ohm per °C, readable with a 3- or 4-wire bridge |
| Cold-junction compensation | Required; CJC error adds directly to reading | Not applicable |
| Extension wiring | Must use matched K-type compensating cable end to end | Ordinary copper; 3-wire cancels lead resistance |
| Interference susceptibility | High — microvolt signal near a mains-powered heater | Low — higher signal level, current-excited |
| Interchangeability without calibration | Poor at low span | Good; standard curve applies |
| Failure mode on open circuit | Often reads ambient or drifts, depending on input | Reads full-scale high / open detected cleanly |
For this application, Pt100 sensors were specified as the replacement. Use 3-wire minimum; use 4-wire if the transmitter or module supports it and cable runs exceed a few metres.
The Junction Problem: Twisted vs Welded
A twisted-pair thermocouple junction works — the Seebeck EMF is generated in the wire along the thermal gradient, not "in the bead" — but it is unreliable in service. Twisted junctions oxidise, loosen, and can create multiple electrical contact points at slightly different temperatures. In a humid or vibrating machine they drift and go intermittent.
A serviceable low-cost welding method for K-type wire, using shop-floor parts:
- Transformer: 36 V AC secondary, 50 Hz, 100 W rating.
- Electrode: a recess filled with compacted carbon dust taken from a motor carbon brush, sitting on a metal backing plate.
- Connections: one lead (with crocodile clip) to the backing plate; the other clipped to the thermocouple wire roughly 10 mm above the twisted ends.
- Action: dip the twisted end into the carbon powder; the arc forms the bead in 1-2 seconds.
Note the intent: this technique is documented here as a legitimate field repair for thermocouples that must stay thermocouples. It does not rescue a wrong-technology or wrong-location installation.
Placement Error: Measuring the Heater, Not the Part
Mounting the sensor on the reverse face of a silicone heater turns the loop into a heater-surface regulator. In steady state, heat flows from the heater through the plate to ambient, so the heater face is always hotter than the plate face. Two error components exist:
| Error component | Cause | Behaviour |
|---|---|---|
| Static offset | Thermal resistance of bond line + heater backing between sensor and plate | Roughly proportional to delivered power; changes when duty cycle changes |
| Dynamic lag | Thermal mass of a 15 mm duralumin plate ~1 m long | Sensor leads the part on heat-up, lags on cool-down; encourages overshoot |
| Ambient sensitivity | Sensor on the outer (unclamped) side sees room air and airflow | Setpoint drifts with ambient and enclosure ventilation |
| Longitudinal gradient | 1 m long plate, edge losses at both ends | Single sensor cannot represent the whole plate |
Consequence for a 90-120 °C band: a static offset of even 10-15 °C consumes most of the allowable window, and because the offset scales with power, a fixed software bias only calibrates one operating point. If the machine runs at different loads, the calibration is wrong at every other point.
Retrofit Procedure
The sensors are captive: pulling a probe tears the silicone patch. Plan the intervention as a rework, not a swap.
- Map the existing loops. Record which of the four sensors are removable and which are bonded. Log actual polarity of each K-type pair at the input module before disturbing anything — reversed K polarity is common and produces a reading that falls as temperature rises.
- Characterise the current offset before rework. With the machine at steady state, attach a reference probe (clamped, thermally paste-coupled) to the duralumin plate at three points: centre and both ends. Record the delta against the installed sensor at two different power levels. This gives the magnitude of the error you are removing and the plate's longitudinal gradient.
- Choose the new sensing point. Machine a blind pocket into the duralumin plate from the non-working face, sized for a Pt100 element or a mineral-insulated Pt100 probe. Depth should place the element close to the working surface without breaking through.
- Couple the sensor thermally. Use thermal compound rated above the maximum process temperature, and mechanically retain the probe with a spring or grub screw so contact pressure does not relax on thermal cycling.
- Cut the heater patch cleanly. Where an old thermocouple must be extracted, cut the silicone patch with a scalpel rather than pulling the wire. Re-bond with an RTV silicone compatible with the heater and rated for continuous service above 120 °C. Verify the heater element itself was not nicked before re-powering.
- Rewire to copper. Remove the K-type compensating cable entirely. Run screened, twisted 3-wire copper for each Pt100, screen bonded to earth at the panel end only. Keep the cable out of the heater power loom; cross power cabling at 90° where routes must meet.
- Reconfigure the input module. Change the channel from thermocouple K to Pt100 (IEC 60751 / DIN 385 curve), select 3-wire mode, and enable open-circuit detection. Re-scale any raw-to-engineering conversion in PLC logic — do not leave the old thermocouple scaling active.
- Retune the loop. The new sensor location has more thermal lag and less noise. Re-run an autotune or a step test; the old gains were tuned against a fast, offset heater-surface signal and will run sluggish or oscillatory with the new PV.
- Add an over-temperature interlock. If sensing has moved into the plate, the heater itself is now unmonitored. Fit an independent thermal cut-out or a second sensor on the heater to trip on heater-face over-temperature, hard-wired into the heater contactor coil circuit.
Verification
| Check | Method | Acceptance |
|---|---|---|
| Wiring integrity | Measure Pt100 resistance at panel terminals, cold | Near 100 Ω at 0 °C, rising with room temperature; all three leads within a fraction of an ohm of each other |
| Open-circuit alarm | Lift one sense lead at the terminal | Module reports break within its declared response time; PLC drives heater output to 0 |
| Static accuracy | Compare PV against a calibrated reference on the plate at 90 °C and 120 °C setpoints | Deviation within the loop's specified tolerance at both points, not just one |
| Cross-channel agreement | Soak the machine at setpoint with heaters off long enough to equalise | All four channels converge within a few °C of each other |
| Noise immunity | Trend PV at 100 ms while cycling heater contactors and any VFDs | No step in PV correlated with switching events |
| Control performance | Step setpoint 90 -> 120 °C | Overshoot and settling within process spec; no sustained oscillation |
Design Rules to Carry Forward
- Below roughly 400 °C, default to Pt100 unless response time or sensor cost forces a thermocouple.
- Measure the object under control, not the actuator that heats it. If you must sense the actuator, add a second sensor on the object.
- Any sensor that cannot be removed without destructive rework is a maintenance defect. Provide pockets, glands or clamps at design time.
- Never run thermocouple or RTD cable in the same loom as heater power. Screen and earth at one end only.
- On imported or one-off machines with serial number 001, verify sensor type, polarity, location and scaling from the terminals inward before trusting any documentation.
- Keep an independent over-temperature trip in hardware, separate from the control sensor.
FAQ
Can I use a Type K thermocouple below 100 °C?
Yes, it works electrically, but the signal is only tens of microvolts per °C and depends entirely on accurate cold-junction compensation. At 90-120 °C a Pt100 RTD gives better resolution, repeatability and noise immunity, and needs only ordinary copper wiring.
Does a twisted thermocouple junction give a wrong reading?
Not inherently — the EMF is generated along the wire gradient, so a twisted junction can read correctly at first. The problem is reliability: twisted ends oxidise, loosen and go intermittent, so the junction should be welded into a single bead.
How do I weld a K-type thermocouple bead without a dedicated welder?
Use an isolated 36 V AC, 50 Hz, 100 W transformer, a carbon-dust electrode (compacted carbon-brush powder in a recess on a metal backing plate), one lead clipped to the backing plate and the other to the wire about 10 mm above the twisted end. The arc forms the bead in 1-2 seconds. Fuse the primary and use eye protection.
Why is mounting the sensor on the back of a silicone heater wrong?
It regulates heater surface temperature rather than the workpiece. The offset between heater and part scales with delivered power, so a single software correction is only valid at one duty cycle, and the sensor also picks up ambient air on its exposed side.
Should I use 2-, 3- or 4-wire Pt100?
Use 3-wire as the minimum for any panel-mounted installation so lead resistance is cancelled; use 4-wire where the module supports it and the cable run is long or accuracy is critical. Avoid 2-wire except for very short leads.
