Home › Guides › PID Control

PID Control Explained: P, I & D Terms, Tuning & Problems

Almost every automatic control loop in industry — a tank level, a furnace temperature, a pump pressure, a motor speed — is run by the same three-term controller, invented nearly a century ago and still dominant today. PID control survives because it is simple, robust and understandable: it looks at the error between where a process is and where it should be, and it adjusts an output using three different views of that error — the present, the past and the trend. This guide explains each term in plain language, shows how to tune a loop without black magic, and covers the classic problems that make even experienced engineers reach for the manual override.

The loop and the error

Every PID loop has the same anatomy. A process variable is measured — level, temperature, pressure, flow, speed — by a transmitter. A setpoint states the desired value. The difference between them is the error. The controller computes an output from that error and sends it to a final element — a control valve, a variable frequency drive, a heater — which changes the process and thereby changes the measured variable, closing the loop. The controller's job is to drive the error to zero and keep it there despite disturbances: a change in feed, a load increase, a valve sticking. The quality of the measurement matters enormously here: a loop fed by a drifting, uncalibrated transmitter will misbehave no matter how well it is tuned, which is why transmitter calibration is the unglamorous precondition of good control.

Control quality is judged by the response to a disturbance or a setpoint change. The ideal response reaches the setpoint quickly, overshoots little, settles without prolonged oscillation, and returns to the setpoint exactly rather than settling slightly off. Those four qualities — speed, overshoot, stability and accuracy — pull against each other, and tuning is the art of balancing them for the particular process. A temperature loop on a large furnace can afford to be slow and gentle; a pressure loop protecting equipment cannot.

The P term: react to the present error

The proportional term multiplies the current error by a gain. Double the error, double the corrective action. It is the workhorse of the controller: simple, immediate, and the first thing to adjust. A higher proportional gain makes the loop respond faster and more aggressively; a lower gain makes it gentler and slower. But proportional-only control has a fundamental limitation: as the error shrinks, so does the corrective action, so the loop settles at a point where the remaining error is just large enough to produce the output the process needs to hold steady. That remaining error is called steady-state offset or droop, and it is the reason proportional-only control is rarely enough on its own. On some loops — a level loop where exact level does not matter much — offset is acceptable and P-only control is fine. On a temperature or pressure loop where the setpoint must be held exactly, it is not.

Pushing proportional gain too high creates the opposite problem: the loop overreacts, overshoots the setpoint, then overcorrects back, and the oscillation grows or persists. The gain at which sustained oscillation begins is a real, measurable property of the loop, and it plays a central role in one of the classic tuning methods below. The practical rule is that proportional gain sets the loop's basic responsiveness, and the other two terms exist to fix its two shortcomings: offset and oscillation.

The I term: learn from the past, remove offset

The integral term accumulates the error over time and adds an output proportional to that accumulated total. As long as any error persists, the integral term keeps growing and pushing the output, until the error is driven to zero — which is exactly what removes steady-state offset. Integral action is patient where proportional action is impatient: it does not care how large the error is right now, only that it has been there for a while. That patience is powerful and dangerous in equal measure. Too little integral action and offset lingers; too much and the loop becomes sluggish and oscillatory, because the integral keeps pushing even after the error has started to shrink, like a driver who keeps accelerating after the traffic ahead has already slowed.

The classic integral problem is windup. Suppose a valve is fully open but the process still cannot reach its setpoint — perhaps a pump has failed, or a setpoint was set beyond what the process can achieve. The error persists, so the integral term keeps growing, far beyond what the output can ever use, because the output is already saturated at its limit. When the situation finally changes and the error reverses, the controller must first "unwind" that huge accumulated integral before its output even begins to fall, causing a large overshoot and a long delay. The standard cures are anti-windup: freeze the integral while the output is saturated, clamp it to a limit, or back-calculate it so it never exceeds what the output can express. Any loop that can saturate — and most can — needs anti-windup designed in, not added as an afterthought.

The D term: anticipate the trend

The derivative term reacts to the rate of change of the error: if the error is shrinking rapidly, derivative action brakes the output; if the error is growing rapidly, it boosts it. In effect, derivative action looks ahead, damping the overshoot that proportional action alone would cause and allowing a higher proportional gain to be used safely. On paper it is the term that makes a loop fast and stable at the same time. In practice it is the term most often left off, for a good reason: derivative action amplifies noise. A measurement with electrical noise or a jittery transmitter produces a rate of change that swings wildly, and the derivative term turns that noise into output chatter that wears out valves and agitates the process. Derivative action therefore needs a clean measurement — proper filtering, a well-installed sensor, a stable transmitter — and it is usually applied to the process variable rather than to the error, so that a sudden setpoint change does not cause a violent derivative kick.

When derivative action is appropriate — typically on slow, laggy processes such as temperature loops, where anticipation genuinely helps — it is used in moderation, with a filter on the derivative calculation. On fast, noisy loops such as flow control, it is usually omitted entirely, and the loop runs as PI. There is no shame in a PI loop: most industrial loops are PI, and a well-tuned PI loop beats a poorly tuned PID loop every time. The derivative term is a specialist tool, not a badge of sophistication.

Tuning: manual method and Ziegler-Nichols

Manual tuning follows a simple, safe sequence. Start with integral and derivative off, and a low proportional gain. Put the loop in automatic, make a small setpoint step, and watch the response. Increase the gain gradually until the response is acceptably fast without sustained oscillation. Then add integral action slowly — a small integral effect first — until the offset disappears without introducing oscillation. Add derivative last, only if needed, in small amounts, watching for noise amplification. At every step, change one thing at a time and let the loop settle before judging; changing two terms at once makes it impossible to know which change helped. Record the final settings and the response that justified them: a tuned loop without a record is a loop that will be re-tuned from scratch after the next maintenance visit.

The Ziegler-Nichols method gives a structured starting point instead of pure trial and error. In its closed-loop form, integral and derivative are switched off, proportional gain is increased until the loop oscillates steadily, and that ultimate gain and the oscillation period are recorded. Simple formulas then convert those two numbers into suggested P, I and D settings. The result is a reasonable starting point — often slightly aggressive, with noticeable overshoot — which the engineer then softens by trial. The method's value is that it replaces guessing with measurement: two numbers obtained from the real process, not from a table. Its danger is that driving a real process into sustained oscillation can be unsafe on some loops, so it should only be used where oscillation is tolerable, with the loop watched throughout and a manual override ready. On loops where oscillation is not tolerable, open-loop step-response methods or the manufacturer's auto-tuning function are safer alternatives, always followed by observation and refinement.

Auto-tuning deserves a word of caution. Many modern controllers offer an auto-tune button that perturbs the loop and calculates settings. It is a useful tool, but it is not a substitute for understanding: auto-tune on a noisy measurement, a sticking valve or a mis-scaled transmitter produces confident-looking settings for a broken loop. Always check what auto-tune produced, watch the loop's response afterwards, and be ready to override it. The same caution applies to cascade and ratio loops, where one controller's output becomes another's setpoint — tuning the outer loop before the inner loop is stable is a reliable way to produce oscillation in both.

Common problems and how to recognise them

Oscillation that never settles usually means too much gain or too much integral action, or a mechanical problem masquerading as a tuning problem: a sticking valve, a loose linkage, a transmitter with a dead band. Before re-tuning, check the field equipment — a loop cannot be tuned around a sticking valve, and the attempt will only make the oscillation worse. Offset that never disappears means integral action is too weak, switched off, or wound up against a limit. Slow, sluggish response means gains too low, or a process that is genuinely slow — a large tank or furnace cannot be hurried by tuning alone. Output chatter or a valve that hunts constantly suggests derivative action on a noisy signal, or a measurement problem. And a loop that was fine yesterday and wild today has almost never "lost its tuning" on its own: something changed — a transmitter drifted, a valve started sticking, a load changed, a bypass was left open. Find what changed before touching the tuning. That diagnostic habit, more than any tuning formula, is what separates a good loop engineer from a frustrated one. The wider automation context — how the loop sits inside a PLC program or a DCS — is covered in the companion guides.

Frequently asked questions

Educational content only — always follow your project specifications, manufacturer instructions and local safety regulations.