Why Does My Temperature Controller Reading Jump Near Setpoint?

Why doesn't retuning, filtering, or swapping the mat settle the reading?

The first instinct on a heater-mat loop that reads 90.6°C one scan and 98°C the next is to touch the controller. Each of the usual moves fails for a reason worth understanding before you spend a shutdown window on it.

Re-running auto-tune. Auto-tune drives the output, watches the process response, and writes new P, I, and D terms. It has no path into the measurement side. The loop in question was already in auto-tune and the display jumped anyway; a second tune cycle computes new gains from a corrupted PV and produces worse gains. Tuning does not fix wiring.

Switching to on-off control or widening the proportional band. This changes when the output switches, so the jumping moves to a different band around setpoint or changes its rhythm. It does not remove the mechanism that ties the PV to the output state.

Raising the input digital filter. A heavy filter averages the two alternating readings into a phantom value between them. The display looks calm and the controller regulates the mat to a temperature that does not exist. Filtering hides a measurement fault; it does not correct one.

Replacing the heater mat. This sometimes works, because a mat with degraded insulation between the element and the thermocouple is one of the two credible root causes. Done blind, it is a shotgun: if the coupling is in the cable bundle or the grounding, the new mat installed the same way misbehaves the same way.

Replacing the controller first. Identical controllers with the same configuration on other plants run stable. The controller belongs at the end of the list, not the start. Look at the trend first.

What happens in the signal chain when the loop enters its proportional band?

Follow the chain: a K-type thermocouple produces a few tens of microvolts per degree, the controller cold-junction-compensates and linearizes that into a PV, the PID block compares PV to the 90°C setpoint and issues a time-proportioned output, and a solid-state relay switches mains current into the mat element on that duty cycle.

Below setpoint the output sits at 100% duty. The SSR is continuously on and mat current flows continuously. If that current couples a voltage into the thermocouple loop, the coupled offset is constant, so the PV is wrong by a fixed amount but moves smoothly. Nobody notices a smooth ramp that is a few degrees high.

As PV climbs into the proportional band, the output drops to an intermediate duty and the SSR begins cycling. Now the coupled offset is present during the ON portion of each cycle and absent during the OFF portion. The PV alternates between the true temperature and the true temperature plus the offset. That is exactly the observed pattern: 90.6°C in one state, 98°C in the other, switching in step with the output, appearing only near setpoint. A mat with real thermal mass cannot move 8°C in a second; the thermocouple voltage can.

Scale the offset to see what you are hunting. A K-type junction produces roughly 41 µV per °C in this range, so an 8°C jump corresponds to about 330 µV superimposed on the thermocouple signal (assumption: standard K-type sensitivity, no controller scaling). That is a small voltage, easily produced by any of four paths:

• Capacitive and inductive coupling from the mat power cable to the thermocouple extension, made worse by bundling both in the same sleeving over the run.
• Leakage through the mat insulation from the live element to a grounded or exposed thermocouple junction, which injects mains-referenced current into the measuring circuit.
• A ground loop between a grounded junction at the mat and the controller input common, carrying element leakage or SSR return current through the thermocouple leads.
• Supply modulation by the SSR itself. Burst-fired or integral-cycle controllers drop whole half-cycles from the supply, and the resulting distortion at the distribution board can disturb a controller sharing that supply, particularly one with a poorly filtered input stage.

A second, purely thermal mechanism belongs on the list because it produces a related symptom. If the thermocouple bead sits directly on the heating wire rather than in the mat body, and the mat has little heat capacity, the sensor genuinely sees the element temperature swing with every output pulse. That swing is real, but it shows as a ramp up and a ramp down, not an instantaneous step, and its amplitude shrinks as the duty cycle shortens.

Which signal is wrong, and what does each fault look like on the display?

How do you separate electrical pickup from a real thermal swing?

The signature is in the time domain. Watch the PV against the output indicator over a dozen cycles. Electrical coupling produces a step on both edges: the reading changes within one scan when the SSR fires and within one scan when it drops out. A thermal swing at the bead produces a rise with a rounded start when the element energizes and an exponential decay when it de-energizes. An 8°C step in one second on a heater mat is electrical.

Confirm it with the two tests that do not require touching the wiring. First, put the controller in manual at 0% output with mains still applied to the SSR input side. If the PV settles and cools smoothly, the disturbance comes from the switched load path. Second, put the controller in manual at 100% output. If the PV shifts by a fixed amount and then moves smoothly, that shift is the coupled offset; its magnitude tells you how far the loop has been regulating from the true temperature all along.

When the plant can be stopped, put a millivolt meter across the thermocouple leads at the controller terminals with the leads disconnected from the input. A fast meter or a scope shows whether the abrupt swing is on the leads (wiring, mat, grounding) or only appears once the controller processes it (controller). A second thermocouple with an independent indicator placed on the same mat gives a reference that does not share the suspect cable or supply.

What is the step-by-step procedure?

1. Trend PV and output state together at the controller's fastest available rate. Record whether the PV moves as a step or a ramp and whether every step aligns with an output edge.
2. Set the controller to manual, 0% output. Record whether the PV becomes smooth. Set 100% output and record the fixed offset relative to a reference thermometer on the mat.
3. Inspect the routing. Note where the thermocouple extension and the mat power cable share sleeving, run parallel, or enter the panel through the same gland, and note the run length over which they are bundled.
4. Isolate mat power at the source. Disconnect the thermocouple leads at the controller and measure millivolts with the mat still warm; the reading should decay smoothly. Reconnect power to the mat with the leads still on the meter and look for a step in the millivolt reading as the SSR cycles or as you command it in manual.
5. Check the junction and mat insulation. Determine whether the junction is grounded or ungrounded per the mat's documentation. With the thermocouple disconnected from the controller, measure insulation resistance from the mat element to the thermocouple conductors and from the element to the mat sheath; compare against a known-good mat of the same type.
6. Check bonding. Confirm the mat sheath, the panel, and the controller earth are at a common potential and that the thermocouple shield, if any, is grounded at one end only.
7. Look at the controller supply with a scope while the output cycles. Identify the SSR firing mode (zero-cross, burst, phase-angle) and whether notching or missing half-cycles coincide with the PV steps.
8. Substitute the thermocouple with a second known-good sensor and separate cable routed clear of power. If the PV is now smooth, the fault is in the original sensor, mat, or run. If it still steps, substitute the controller.

Which corrections hold up in service?

Fix what step 4 through 8 identified, in this order of likelihood for this class of installation.

Cable coupling. Pull the thermocouple extension out of the shared sleeving and route it on its own path with as much separation from the mat power cable as the mechanical arrangement allows. Where a parallel run is unavoidable, use twisted, shielded thermocouple extension of the correct K-type alloy, and cross power cables at right angles rather than running alongside. Ground the shield at the controller end only. Keeping thermocouple and power cables sleeved together is common practice because it is tidy, and it survives on most installations because the coupling stays below the controller's noise floor; on the one that does not, separation is the cure.

Mat leakage. A mat whose element-to-thermocouple insulation resistance is markedly lower than its siblings, or that shows a millivolt step on the leads with the extension removed from the equation, is the fault. Replace it. Specify an ungrounded junction if the process allows the slightly slower response, because an ungrounded junction breaks the leakage path into the measurement circuit.

Grounding. Where the junction is grounded and a potential difference exists between the mat and the panel, bond them or move to an ungrounded junction. Do not ground the thermocouple circuit at two points.

Supply modulation. If the scope shows the SSR distorting the controller's supply, feed the controller from a separate, clean circuit or add a line filter at the controller supply terminals, and prefer zero-cross switching over burst or phase-angle firing where the mat allows it.

Controller. Only after the sensor substitution in step 8 still steps: swap the controller with a known-good unit of the same model and configuration. Keep the removed unit for bench verification with a millivolt source before scrapping it.

Once the measurement is clean, revisit tuning. Auto-tune on a low-heat-capacity load with a fast on-off output can produce poor results; if the mat behaves that way, tune manually. Start proportional-only, confirm stable modulation in the band, then add a small amount of integral to remove the offset. Add derivative last, and only if the process needs it.

How do you verify the fix?

Bring the loop up from cold with PV and output trended together. Through the 100% region the PV must track the reference thermometer within the thermocouple's tolerance, with no fixed offset. As the loop enters the proportional band and the SSR begins cycling, the PV must continue as a smooth curve with no step at any output edge, and it must settle at 90°C with the deviation the tuning permits rather than oscillating between two values.

Repeat the manual 0% and 100% tests from the diagnostic phase. The PV must not shift when the output state changes. Put the millivolt meter back across the leads at the controller with the SSR cycling under manual command and confirm the millivolt reading is free of steps. Finally, compare the settled PV against an independent reference on the mat surface; a residual offset here points at a remaining leakage or coupling path that the fix reduced but did not eliminate.

Stop and escalate when the PV still steps with a known-good sensor on a separate run, a known-good controller, and insulation and supply measurements that match the stable plants. At that point, send the manufacturer of the heater mat the insulation resistance figures and the manufacturer of the controller the millivolt and supply traces, using their official technical support channels. Both need the measured data, not the symptom description, to take the case further.

FAQ