PID control is how machines hold a value steady against a world that keeps disturbing it — the oven at temperature as the door opens, the motor at speed as the load bites, the pressure constant as demand swings. The algorithm inside virtually every process controller, drive and 3D printer firmware computes its output from three views of the same error: the Proportional term answers the present, the Integral answers the accumulated past, and the Derivative anticipates the immediate future.
This guide explains PID without the transfer functions: why simple on/off control isn't enough, what each of the three terms contributes and costs, how to tune a loop from symptoms rather than theory, and the sensor and output hardware a working loop actually needs.
Why isn't on/off control enough?
On/off — heater full on below setpoint, off above — is honest, cheap and everywhere: it runs fridges, kettles and immersion heaters perfectly well. Its signature is permanent oscillation. The process has momentum, so by the time the sensor crosses the setpoint and the output switches, heat already in flight carries the temperature past it; the result is a sawtooth swinging either side of the target, its amplitude set by the process, not the controller. Where a few degrees of swing don't matter, on/off wins on simplicity. Where they do — a reflow profile, an incubator, an extruder, a process with quality riding on it — the fix is an output that can be partly on, steered by an algorithm that eases off before arrival. That's the whole motivation for PID.
What do P, I and D each do?
| Term |
Looks at |
Contributes |
Too much causes |
| Proportional | The error right now | The muscle — output in proportion to how far off you are | Oscillation |
| Integral | Error accumulated over time | The memory — trims out the persistent offset P alone leaves | Overshoot and windup |
| Derivative | How fast the error is changing | The brakes — damps the approach, resists sudden change | Nervous, noise-chasing output |
Two of the table's corners deserve expansion. P alone can never quite arrive: holding any output requires some error to be proportional to, so a P-only loop settles slightly short of the setpoint — the steady-state offset that the integral term exists to erase, by accumulating that small persistent error until the output drifts up to close it. And the integral's memory is also its vice: if the output saturates — heater already at 100% during a long climb — the integral keeps accumulating a debt it can't spend, then overspends it as a large overshoot on arrival. That's integral windup, and decent controllers include anti-windup precisely because long heat-ups make it inevitable otherwise.
How do you tune a loop without a control theory degree?
Start with autotune. Virtually every modern controller has it: the controller deliberately provokes a controlled oscillation, measures how the process responds, and computes workable parameters — the same information classical hand methods like Ziegler–Nichols extract, gathered automatically. Run it under realistic conditions (typical load, typical setpoint) and it gets most processes most of the way. Refinement, where needed, is symptom-driven:
- Sustained oscillation around setpoint — too much gain: reduce the proportional action
- Crawls to setpoint, or sags under load — too conservative: increase proportional action, or strengthen the integral if it arrives but sits offset
- Overshoots on setpoint changes or after long heat-ups — integral doing too much, or windup: weaken the integral, confirm anti-windup is active
- Output twitches and hunts on a steady process — derivative amplifying sensor noise: reduce it, filter the input, or switch it off
Change one thing at a time, let the process settle between changes, and log the response — a data logger or the controller's own trend display turns tuning from folklore into observation. And an honest industry fact: a large share of working loops run as PI, with derivative set to zero — on noisy processes D costs more jitter than its damping is worth, and leaving it off is practice, not laziness.
What hardware does a PID loop need?
Three parts: a sensor, a brain and an output stage. The sensor is the ceiling on everything downstream — a loop can't control what it can't measure, so a heat process gets a properly installed thermocouple or PT100 with the extension-cable and placement discipline that measurement deserves. The brain is a spectrum: a panel-mount process controller for a standalone loop, PLC function blocks where the loop is one of many in a machine, or firmware where the product is the controller — a 3D printer's hotend autotune being the most democratised PID exercise ever shipped. Panel meters keep the process visible either way.
The output stage is where selection gets consequential. A mechanical relay is cheap and robust but mortal: PID holds temperature by switching in a slow rhythm (time-proportional control — a ten-second cycle whose on-fraction is the output percentage), and that rhythm running day and night consumes a mechanical relay's contact life in months. A solid state relay switches silently, indefinitely and fast — the natural partner for PID-driven heaters — with one non-negotiable clause: an SSR dissipates real heat in service (roughly a watt or more per amp switched) and must be mounted on an adequate heatsink, because SSRs don't fail gracefully. For continuously variable processes, analogue outputs (4–20 mA, 0–10 V) drive valves and motor drives directly, with no switching rhythm at all.
Where does PID show up?
Everywhere a value must hold: ovens, furnaces, reflow and extrusion temperatures; motor speed loops inside every drive and servo; flow, level and pressure in process plant; even the cruise control holding speed up a hill. The pattern recognition is worth having — once the oscillation, offset and overshoot signatures are familiar from one loop, they're diagnosable in every other, because the algorithm and its failure modes are the same whether the plant is a kiln or a quadcopter.
Featured process control
Frequently asked questions
What does PID stand for?
Proportional, Integral, Derivative — the three calculations combined into the controller's output. Proportional responds to the present error, integral to its accumulated history, derivative to its rate of change.
Why does my temperature overshoot the setpoint?
Usually the integral term — either tuned too strong, or wound up during a long full-power heat-up and spending its accumulated debt on arrival. Weaken the integral action, enable anti-windup, and re-run autotune under realistic conditions.
What does autotune actually do?
It provokes a deliberate, controlled oscillation, measures the process's response — how fast, how much lag, how much momentum — and calculates PID parameters from it. Run under normal operating conditions, it replaces manual tuning for most loops.
Should I switch a heater with a relay or an SSR?
For PID control, an SSR: time-proportional output switches every few seconds indefinitely, which wears out mechanical contacts in short order. The SSR's condition is a proper heatsink — it dissipates real heat per amp switched and fails hard without one.
What is integral windup?
The integral term accumulating error while the output is saturated and can't respond — during a long heat-up, say — then discharging that accumulation as overshoot once the setpoint nears. Anti-windup, standard in decent controllers, freezes the accumulation while the output is pinned.
Do I need the derivative term?
Often not — many industrial loops run PI with derivative at zero, because on noisy measurements D amplifies the noise into a jittery output. It earns its place on clean signals and processes that need a damped, decisive approach to setpoint.
Hold the line
Controllers, sensors, solid state relays and the logging to prove the loop — the full chain from measurement to muscle.
Browse process and instrumentation
View solid state relays