Heater Relay Autotuning: Validate the Oscillation Before Using Its Gain Estimate

System boundary

A customer-owned heater, load, sensor, actuator and temporary relay-feedback identification mode. The element remains within its reviewed operating boundary; the method does not supply a universal PID tuning rule or equipment safety approval.

Integration interfaces

Integration risks

Separate temporary identification from normal regulation

A relay-feedback test uses a switching rule to perturb the process and observe its dynamics. It is not the same task as selecting an on/off thermostat band for ordinary operation. The experiment should end with an identification record that a control designer can assess. Normal control then uses separately chosen parameters and validated response requirements.

Identify the complete installed plant: command-to-power conversion, heater, mounting, load, heat losses and sensor. A different sensor filter or fixture can alter the observed oscillation even when the ceramic element is unchanged. Record the hardware and software state before the test. The estimate belongs to that configuration and operating region, not to the heater material as a universal property.

Choose two levels within an authorized envelope

For a heating-only system, represent the test command as a bias u0 plus or minus an amplitude h. The lower level can still be positive heating power. A negative perturbation relative to the bias does not mean that the heater supplies active cooling. Confirm that natural heat loss and the permitted levels can produce the intended response without exceeding the accepted temperature range.

The controls and safety owners must define maximum temperature, minimum and maximum applied power, allowable rate of change, timeout and abort response before testing. Keep independent protective functions active. If the process cannot tolerate deliberate oscillation, use an approved alternative identification method rather than forcing this experiment to run. This guide does not authorize an unattended thermal excursion.

Use the describing-function approximation explicitly

For an ideal symmetric relay with negligible hysteresis, its output perturbation is approximately a square wave of amplitude h. Its fundamental component has amplitude four h divided by pi. If the resulting process oscillation is close to sinusoidal with amplitude a, a first-harmonic approximation gives an equivalent gain of four h divided by pi a.

Under the applicable relay-loop conditions this is used as an estimate of critical gain, with the observed oscillation period providing the corresponding time scale. It is an approximation, not a complete plant model. Strong harmonics, hysteresis, drift or nonlinear thermal behaviour can weaken the interpretation. Keep those limitations attached to the estimate instead of reporting extra decimal places as evidence of precision.

Keep output units and half amplitudes visible

Assume the applied command switches between 30 and 50 percent of a defined full-scale power command. The bias is 40 percent and h is ten percentage points. Suppose a justified sinusoidal fit gives temperature amplitude of 2 kelvins and period of 80 seconds. The approximate critical gain is 6.366 percentage points per kelvin and angular frequency approximately 0.07854 radian per second.

Using the full twenty-percentage-point swing as h would incorrectly double the gain estimate. Using a four-kelvin peak-to-peak temperature span as a would incorrectly halve it. Expressing output as a zero-to-one fraction instead changes the numeric gain to 0.06366 per kelvin. That is a unit conversion, not a different process. Preserve the convention expected by the controller interface.

Establish a repeatable cycle before extracting a number

Retain several cycles and compare their centers, amplitudes and periods. Early warm-up may move the mean temperature while the oscillation is still forming. A changing load, fan or inlet condition can produce similar drift. Select an identification interval only after the behavior satisfies an approved stationarity criterion; do not pick the visually neatest cycle from an otherwise inconsistent trace.

Use the actual applied command or power record to check symmetry and timing. A supply limit, shared-power allocation or minimum pulse constraint can make the delivered perturbation different from the requested two levels. Strongly unequal high and low durations or distorted temperature waves require an appropriate extended model. The simple calculation is not a way to erase those observations.

Do not treat noise suppression as a free modification

Measurement noise near the switching condition can create rapid unwanted relay changes. Practical methods may include hysteresis or other controlled measures. Such modifications change the test's mathematical behaviour. In particular, finite relay hysteresis introduces phase effects that are absent from the ideal real-valued expression used above. Record the actual rule and use the corresponding analysis.

Sensor filtering also participates in the identified dynamics. If a filter is changed after tuning, the controller no longer sees exactly the tested plant. Preserve the raw sensor signal where practical and distinguish measurement noise from real temperature variation. A smoothed sinusoid can look suitable while hiding switching artifacts or a process harmonic that matters to the estimate.

Use the estimate as input, not final acceptance

A critical-point estimate supplies limited frequency-response information. It does not uniquely determine every useful PID setting or prove an overshoot, settling-time or disturbance-rejection requirement. The control designer must select a suitable tuning method and controller form, including output limits, integral handling and derivative filtering where used. Do not apply a remembered coefficient table without checking those definitions.

Validate the resulting normal-control response with the permitted load and disturbances, including startup and relevant operating points. Keep setpoint tracking and load-disturbance behavior as separate observations. A controller that behaves well near the autotune bias can behave differently elsewhere. If conditions change substantially, assess whether new identification or a planned gain schedule is required rather than transferring the original coefficients automatically.

Keep the oscillation evidence with the heater configuration

The handoff should contain actual high and low commands, power mapping, raw temperature and event timestamps, accepted cycle interval, amplitude extraction and units. Record the fixture, load, sensors, filters and safety envelope. Include the reason the simple approximation was accepted or rejected. An autotune status code without this context gives a later engineer little basis for evaluating a changed assembly.

ChipSimple can review the drawing-defined heater and electrical stress associated with the proposed operating sequence. The equipment owner remains responsible for identification, controller design and complete thermal performance. Keeping those records separate makes a heater revision traceable without implying that one resistor measurement establishes the stability or quality of the closed-loop system.