What Is Feedback Control?
Correcting Error After It Appears

Back to Blog

How Feedback Control Works

In a feedback loop the controller continuously subtracts the measured process variable from the setpoint to get an error, then drives its output to shrink that error toward zero. This is called negative feedback because the correction opposes the deviation: if level rises above setpoint, the controller acts to bring it back down. PID is the most common algorithm used to compute that correction.

The defining trait of feedback control is that it is reactive - it cannot act until a disturbance has actually moved the process variable. That makes it robust and general-purpose: it does not need a model of every disturbance, because it simply responds to whatever error appears. The trade-off is that some deviation must occur before the controller can respond to it.

Feedback vs Feedforward

Feedforward control is the complement to feedback. Instead of waiting for error, a feedforward scheme measures a known disturbance and pre-emptively adjusts the output before the process variable moves - for example, boosting heater output the moment cold feed flow increases, rather than waiting for the temperature to sag. Feedforward is proactive but needs a good model of the disturbance and cannot catch unmeasured upsets.

In real oil-and-gas control schemes the two are combined: feedforward handles a large, measurable disturbance quickly, while feedback trims out the residual error and everything the feedforward model missed. Feedback remains the foundation because it corrects for anything, measured or not.

The Four Elements Every Loop Shares

Every feedback loop, from a thermostat to a separator level controller, is built from the same four elements: a sensor that measures the process variable, a controller that compares it to setpoint, a final control element - usually a valve or a variable-speed drive - that acts on the process, and the process itself, which turns that action back into a change the sensor can see. The signal travels this circle continuously, and the loop is only as good as its weakest arc.

That framing is useful when a loop misbehaves, because each element fails differently. A drifting or damped sensor feeds the controller fiction, and the controller will faithfully regulate the fiction. A worn valve with stiction turns smooth controller output into jumps, which the loop then chases. And process-side changes - fouling, changed feed composition, a swapped impeller - alter the very relationship the tuning was built around. Before re-tuning a misbehaving loop, walk the circle: instrument, controller, final element, process, in that order. A structured version of that walk is a loop check.

Dead Time and Lag: The Limits of Feedback

Feedback's weakness is time. Lag is the process taking time to respond in a smeared, gradual way, like a big tank warming slowly; dead time is a pure delay in which the controller's action produces no visible effect at all, like transport delay down a long pipeline before a composition change reaches the analyzer. Feedback tolerates lag reasonably well, but dead time is poison: the controller acts, sees nothing, acts harder, and by the time the effect arrives it has overcorrected - the recipe for oscillation.

This is why loop layout matters as much as tuning. Measuring close to where you act shrinks dead time; measuring far downstream inflates it. Where dead time is irreducible, tuning has to be deliberately relaxed, and structures beyond plain feedback start to earn their complexity. It is also why noisy measurements are corrosive: feedback cannot distinguish noise from error, so it dutifully chases both unless the signal is filtered.

A quick field test for which one you face: make a small, deliberate output step with the loop in manual and time what the process variable does. A response that starts immediately but takes a long while to settle is lag. A response that does nothing at all for a stretch and then moves is dead time. The distinction decides how the loop can be tuned and whether feedback alone is even the right structure.

Beyond a Single Loop

When one loop is not enough, the standard escalation is cascade control: an outer loop that owns the goal writes the setpoint of a faster inner loop that owns a handle on the process, such as a level controller setting the setpoint of a flow controller on the outlet. The inner loop absorbs fast disturbances before they ever reach the slower outer variable, which plain single-loop feedback cannot do.

The algorithm inside almost all of these loops is the same: PID control, computing the correction from the size, history, and rate of the error. Feedback is the architecture; PID is the arithmetic. Keeping those two ideas separate makes control conversations much clearer - you can change tuning without changing structure, and change structure without abandoning feedback.

A Worked Example: A Separator Level Loop

Take a symbolic separator level loop. The level transmitter reads the liquid level and reports it to the controller as the process variable. The operator's setpoint is mid-range of the vessel. A slug of liquid arrives from the gathering system: level rises above setpoint, the error goes positive, and the controller opens the dump valve further. Liquid leaves faster, level falls back toward setpoint, the error shrinks, and the valve eases back. Every step opposed the deviation - negative feedback in action.

Now run the failure case in your head, because that is where feedback's honesty matters. If the transmitter sticks at a fixed reading, the controller sees zero error and stops correcting while the real level climbs; nothing in plain feedback detects the lie. That is why critical vessels carry independent high-level protection separate from the control loop, and why unusual quiet from a normally active loop deserves suspicion - with any response to a suspected instrument failure handled by qualified personnel per site procedures.

Frequently Asked Questions