What Is PID Control?
Proportional, Integral, Derivative

Back to Blog

The Three Terms

Proportional (P) acts on the present error: the bigger the gap between setpoint and process variable, the harder the controller pushes. P alone is fast but leaves a small permanent offset - the loop never quite reaches setpoint. Integral (I) acts on accumulated error over time; it keeps nudging the output until the offset is gone, driving the process exactly to setpoint. Derivative (D) acts on the rate of change of error, adding damping to reduce overshoot and oscillation on fast or noisy loops.

In practice, most process loops in oil and gas run as PI (proportional-integral) because derivative amplifies measurement noise, which is common on flow and pressure signals. Derivative is reserved for slow loops like temperature where its damping helps and noise is low.

Tuning and Where PID Runs

Each term has a tuning constant - gain (or proportional band), integral time (reset), and derivative time (rate). Tuning sets how aggressively the loop responds: too aggressive and it oscillates or overshoots; too sluggish and it drifts and recovers slowly from upsets. Good tuning balances speed against stability for that specific process.

PID loops almost always execute inside the controller closest to the process - a PLC, a DCS controller, or an RTU or flow computer - not in a remote SCADA host, because control must run fast and survive a communications outage. A SCADA system typically monitors the loop, trends PV against SP, and may write remote setpoints, while the PID math itself stays local.

Direct vs Reverse Action Comes Before Tuning

Before any tuning constant matters, the controller's action has to match the process. A reverse-acting controller decreases its output as the process variable rises; direct-acting does the opposite. Which one is right depends on the whole chain: what the valve does with more signal, whether it fails open or closed, and which direction the process moves in response. Get it backwards and the loop does not oscillate politely - it runs to a limit, because every correction makes the error worse.

This is the first thing to check on a loop that slams to full output or full closed the moment it goes into auto. The action setting, the valve's fail direction, and any signal reversal in a positioner must be reasoned through as one chain, and that reasoning belongs in the loop's documentation so the next tech does not rediscover it during an upset.

Note the interaction with valve selection: swap a fail-closed valve for a fail-open one during a rebuild and the required controller action flips with it. More than one mystery runaway loop has followed a valve change that nobody connected to the controller configuration, which is exactly why the action decision is documented rather than remembered.

Output Saturation and Integral Windup

When the controller output reaches its limit and the error still is not gone, the integral term keeps accumulating - winding up - because that is what integrating a persistent error does. The stored correction then has to unwind before the output comes off the limit, so when the process finally turns, the controller overshoots hard and holds the overshoot. The classic sighting is after a long shutdown: a loop left in auto with the process far from setpoint winds up fully, then overshoots badly on restart.

Modern controllers include anti-windup schemes that pause or back-calculate the integral while the output is saturated, and understanding yours is part of knowing the loop - the variants are described in anti-windup PID control. Saturation itself is also diagnostic: an output living at its limit means the final control element is undersized, the process has moved, or the setpoint is unreachable, and no tuning fixes any of those.

Auto, Manual, and Bumpless Transfer

Every PID controller carries at least two modes: automatic, where the algorithm drives the output, and manual, where the operator sets the output directly and the algorithm stands by. Manual is not a failure state - it is how loops are commissioned, how upsets are ridden out when the measurement is suspect, and how final elements get exercised during testing. On critical services, how long a loop may stay in manual, and who decides, is a site procedure question rather than a technician preference.

The transition between modes is where care is needed. Switching to auto with a stale setpoint or a mismatched output produces a jolt the process feels immediately. Well-configured controllers avoid this with setpoint tracking and output tracking so the changeover is smooth - the behavior explained in bumpless transfer. A loop that bumps on every mode change is misconfigured, and operators will respond by leaving it in manual forever, at which point the automation is decorative.

Reading the Trend: What the Symptoms Point To

Most tuning diagnosis happens by looking at a trend of setpoint, process variable, and output together. The shapes are surprisingly consistent:

Trend symptom	What it usually points to
PV parallels setpoint but never reaches it	Proportional-only behavior; integral action absent or far too slow
Steady, even cycling around setpoint	Tuning too aggressive for the loop's lag and dead time
Sawtooth cycling with a stepping output	Stiction or backlash in the valve, not a tuning problem
Output spikes on every setpoint change	Proportional or derivative acting on setpoint steps
Output flat at a limit while PV drifts	Saturation: sizing, process change, or windup

The habit worth building is to ask which term the shape implicates before touching anything, then change one thing and re-observe. The meanings of the three tuning constants and their common unit conventions are covered in PID tuning gain, reset, and rate - and half of apparent tuning problems turn out to be valve problems wearing a tuning costume.

Frequently Asked Questions