Control loop glossary

Anti-reset windup

Anti-reset windup stops integral action accumulating while the output is at a limit. Check it before you blame the constants: a saturated PID overshoots on recovery when this is off.

Bad actor loop

A loop that consistently performs poorly and consumes disproportionate attention.

Bump test

A bump test is a deliberate step in manual (or a small setpoint change in automatic) so you can see how the loop actually responds. Without it, a quiet trend tells you almost nothing.

Bumpless transfer

Bumpless transfer is switching AUTO/MANUAL without the output jumping. Without it the mode change is an upset, and the manual test that splits hunting from a disturbance becomes unusable.

Cascade control

Two controllers in series, where the outer loop sets the setpoint of a faster inner loop.

Cohen-Coon tuning

Cohen-Coon is an open-loop reaction-curve method from the same era as Ziegler-Nichols, fitted for setpoint response rather than load. A defensible start; a poor gospel on a running plant.

Dead time

Dead time is the delay between a change at the output and the first sign of a response in the measurement. It is the number that sets how fast reset can safely repeat.

Deadband

A range of signal change that produces no movement at the valve at all. On a trend that is the flat spot on the PV at every output reversal: lost motion, not a constant you retune.

Derivative on the process variable

Taking derivative action on the measurement rather than the error, so a setpoint change does not produce a derivative kick.

Feedforward control

Acting on a measured disturbance directly, rather than waiting for the process to respond to it.

First-order plus dead time (FOPDT)

The first-order plus dead time model: three numbers — process gain, time constant and dead time — that most tuning rules start from. How to read them off a bump test, and where the fit breaks.

Gain scheduling

Changing the controller tuning automatically as a function of operating point.

Hysteresis

Hysteresis is a different valve position for the same signal depending on whether it was approached from above or below. On a plant it is used with deadband and backlash for the same lost motion.

Integral windup

Integral action keeps accumulating while the output sits at a limit, causing a large overshoot when the loop recovers. Why detuning it makes it worse.

Internal model control (IMC)

Internal model control reduces PID design to one knob — lambda, the closed-loop time constant. Lambda tuning is IMC on an FOPDT model; a poor model wraps a confident method around a fiction.

Lambda tuning

Lambda tuning sets the closed-loop response time to a value you choose, rather than pushing the loop as fast as a formula will go. Robust on a plant; it needs a process model, not a trend screenshot.

Limit cycle

A limit cycle is a constant-amplitude oscillation from a nonlinearity — stiction, backlash, a travel stop — not from the constants. Detuning stretches the period and leaves the amplitude.

Measurement filtering

Smoothing the process variable before the controller sees it, trading noise for added lag.

Override (selector) control

Two or more controllers sharing one output through a selector, so the limit that matters most wins.

PID hunting

PID hunting is a loop that will not settle around setpoint. If the cycle decays, it is tuning; if the amplitude holds, the valve is sticking and the faceplate will not fix it.

PID offset

PID offset is a steady gap between PV and setpoint that proportional-only action leaves behind. The three reasons a loop with integral still holds offset.

Process gain

The measurement change per unit change in controller output — the K in FOPDT. How it differs from controller gain, and why the units set the tuning.

Proportional band

What is proportional band: the error, as a percent of range, that drives the output across its span. PB% = 100 / gain; raising the band weakens proportional action — the opposite of raising gain.

Quarter amplitude damping

Quarter amplitude damping: each peak a quarter of the one before. Classical Ziegler-Nichols aims at it; most operating plants want a loop that settles inside two peaks instead.

Ratio control

Two flows held in a fixed proportion, where one flow's setpoint is a ratio times the other flow's measurement.

Sample time

Sample time is how often the controller executes. Between scans the output holds, so a slow sample time is extra dead time. Fast loops on a slow scan cannot be retuned out of it.

Shrink and swell

Drum level moving opposite to the boiler as pressure flashes or collapses the steam bubbles. Why it forces three-element control on a drum level loop.

Smith predictor

A dead-time compensation structure that removes long dead time from the loop by predicting the measurement.

Split-range control

One controller output divided between two valves that share the range between them.

Stiction

Stiction is static friction that stops a valve stem until the signal change is large enough to break it free. The loop then cycles at constant amplitude — a valve problem, not a tuning one.

Three-element control

Three-element control is the boiler drum level scheme: drum level trims a feedwater flow controller, and steam flow feedforwards load so the valve does not wait on a swelling drum.

Time constant

The time constant is lag after dead time: roughly how long the PV takes to reach 63% of the new value once it starts moving. Dead time is the wait; this is the climb.

Ultimate gain

Ultimate gain is Ku: the controller gain that just holds a constant-amplitude cycle. Finding it on a running plant means driving the loop to the edge of stability. PIDSnap never asks for it.

Ultimate period

Ultimate period is Pu: the time for one complete cycle at the ultimate gain. Closed-loop Ziegler-Nichols reset and derivative are fractions of it. Do not find it on a production loop.

Velocity (incremental) PID

A PID form whose output is a change in output, not the output position itself.