14.3 PID Control, Loop Tuning & Process Optimization
Key Takeaways
- Proportional action alone always leaves a steady-state offset, because it produces an output only in response to an error; integral action is what eliminates the offset by accumulating that error over time.
- Proportional band and controller gain are reciprocal: proportional band in percent equals 100 divided by the gain, so a wide 200 percent band is a gentle gain of 0.5 and a narrow 20 percent band is an aggressive gain of 5.
- Derivative action responds to the rate of change of the error and therefore amplifies measurement noise, which is why derivative is used on slow thermal loops and almost never on flow or pressure loops.
- Ziegler-Nichols ultimate-sensitivity tuning uses the ultimate gain Ku and the ultimate period Pu: for a PID controller, gain is 0.6 Ku, integral time is Pu divided by 2, and derivative time is Pu divided by 8.
- A loop that oscillates with a square or sawtooth waveform at constant amplitude while the controller output ramps smoothly is showing valve stiction, not a tuning problem — retuning will never fix it.
14.3 PID Control, Loop Tuning & Process Optimization
Quick Answer: Red Seal sub-task F-30.04 asks the industrial electrician to optimize system performance, and in a process plant that means closed-loop control. P responds to how big the error is and always leaves an offset. I responds to how long the error has persisted and removes the offset. D responds to how fast the error is changing and amplifies noise while doing it. Flow loops get P and I with low gain. Level loops get mostly P. Temperature loops are where D earns its keep. And a loop that oscillates in a square wave while the output ramps smoothly has a sticking valve, not a tuning problem.
1. The Vocabulary of a Control Loop
SETPOINT (SP)
│
▼
(+)(−)◄──────────────── PROCESS VARIABLE (PV)
│ ▲
ERROR (e = SP − PV) │
│ │
▼ │
┌──────────────┐ CONTROLLER ┌───┴────────┐
│ P + I + D ├──► OUTPUT (CO)──►│ FINAL │
└──────────────┘ 4–20 mA │ CONTROL │──► PROCESS ──► TRANSMITTER
│ ELEMENT │
└────────────┘
| Term | Meaning |
|---|---|
| PV — process variable | The measured condition: temperature, flow, level, pressure |
| SP — setpoint | The value the process is supposed to hold |
| Error | $e = SP - PV$ (or $PV - SP$, depending on controller action) |
| CO — controller output | The 4–20 mA or digital signal to the final control element |
| Final control element | Usually a control valve, sometimes a VFD, a damper or an SCR power controller |
Direct versus reverse acting
Getting this wrong makes the loop run away to a limit, and it is a five-second check that saves hours.
- Reverse acting — as PV rises, output falls. A steam heating loop with an air-to-open valve: hotter tank, close the steam valve.
- Direct acting — as PV rises, output rises. A cooling-water loop: hotter tank, open the cooling valve further.
The action must account for the valve's fail position too. A reverse-acting controller paired with an air-to-close valve behaves like a direct-acting controller on an air-to-open valve. Work the whole chain — controller action, positioner action, valve fail state — before concluding the controller is misconfigured.
Open loop versus closed loop
- Open loop — the output is set and there is no measurement feeding back. A manual valve position, or a controller in MANUAL.
- Closed loop — the measurement continuously corrects the output. A controller in AUTO.
2. Proportional Action (P)
Output is proportional to the present error:
Gain ($K_c$) and proportional band (PB) are two ways of saying the same thing:
| Proportional band | Gain | Behaviour |
|---|---|---|
| 500% | 0.2 | Very gentle, very sluggish, large offset |
| 200% | 0.5 | Gentle |
| 100% | 1.0 | Full output span for full error span |
| 50% | 2.0 | Aggressive |
| 20% | 5.0 | Very aggressive; likely to oscillate |
Proportional band has a useful physical meaning: it is the percentage of measurement span over which the error will drive the output through its full 0 to 100% range. A 20% band on a 0–200 °C transmitter means the output swings from fully closed to fully open over a 40 °C error.
Offset — the built-in limitation
A pure proportional controller must have an error to produce an output. Consider a heater tank at setpoint with the valve 40% open. Increase the cold feed rate and the tank cools; the controller opens the valve — but it can only do that because the PV has fallen below SP. The loop settles at a new steady state with a permanent offset between PV and SP.
Increasing gain shrinks the offset but never removes it, and eventually causes oscillation. Offset is the reason integral action exists.
3. Integral Action (I) — Reset
Integral action accumulates error over time and keeps moving the output for as long as any error remains. Because it only stops when the error is exactly zero, it drives the offset out completely.
Units — the reciprocal trap
Two conventions are in common use and they are inverses of each other:
| Unit | Meaning | Which direction is more action? |
|---|---|---|
| Minutes per repeat (integral time, $T_i$) | Time for integral action to repeat the proportional contribution once | Smaller number = more integral action |
| Repeats per minute (reset rate) | How many times per minute integral repeats proportional action | Larger number = more integral action |
A technician who "increases reset from 0.5 to 2.0" has added integral action if the controller is in repeats per minute and removed three quarters of it if the controller is in minutes per repeat. Read the units on the controller before touching the number.
Reset windup
If the PV cannot reach setpoint — a valve already wide open, a process shut down, a controller left in AUTO with the equipment off — the integral term keeps accumulating error indefinitely and drives its contribution far outside the output range. When the process finally responds, the loop massively overshoots because the integral term must unwind before the output can come back.
Every industrial controller has anti-reset windup, which freezes or limits integral accumulation once the output saturates. Verifying that anti-windup is enabled and correctly limited is a commissioning step, and the classic symptom of a missing or misconfigured one is a loop that behaves beautifully in normal operation and overshoots wildly after every startup.
4. Derivative Action (D) — Rate
Derivative responds to how fast the error is changing, so it acts before the error becomes large. It is measured in minutes (derivative time, $T_d$), sometimes called rate.
Its value and its danger are the same property. A measurement signal carries noise, and the derivative of a noisy signal is very noisy — a small amount of measurement jitter produces a large, rapid kick in the controller output, which hammers the valve positioner and wears the packing.
| Loop type | Derivative? | Reason |
|---|---|---|
| Flow | No | Fast and inherently noisy; derivative makes it violent |
| Pressure (liquid) | Rarely | Fast dynamics |
| Level | Rarely | Usually an integrating process; derivative destabilizes it |
| Temperature | Yes | Slow, with large thermal lag; derivative anticipates the lag and reduces overshoot |
| Composition / analyzer | Sometimes | Long deadtime; derivative helps only if the signal is well filtered |
Where a process signal is noisy but derivative is genuinely needed, the answer is signal filtering and derivative-on-PV (rather than derivative-on-error), which also eliminates the output spike a setpoint change would otherwise produce.
5. Tuning by Loop Type
Tuning is not one recipe. The loop's own dynamics dictate the answer.
| Loop | Speed | Noise | Typical tuning |
|---|---|---|---|
| Flow | Very fast (seconds) | High | Low gain (0.2 to 0.5), fast integral (0.05 to 0.2 min/repeat), no derivative |
| Pressure (liquid) | Fast | Moderate | Similar to flow |
| Pressure (gas) | Moderate | Low | Moderate gain, moderate integral |
| Level (integrating) | Slow, self-accumulating | Low to moderate | Mostly proportional, gain around 1 to 10 depending on whether tight or averaging control is wanted, very little integral, no derivative |
| Temperature | Very slow, large lag | Low | Full PID — moderate gain, slow integral (minutes to tens of minutes), derivative used |
Why level is different
Level in a tank is an integrating process: with inflow and outflow mismatched, level does not settle at a new value — it ramps until something overflows or empties. That integration is already in the process, so adding much integral action in the controller produces a loop that oscillates slowly and persistently. Many industrial level loops are deliberately tuned as averaging level control: a low gain that lets the level wander across a wide band while passing a smooth, nearly constant flow downstream, because surge capacity is the tank's purpose.
6. Ziegler-Nichols Tuning
Closed-loop (ultimate sensitivity) method
- Put the controller in AUTO with integral and derivative switched off (integral time at maximum, derivative at zero).
- Raise the proportional gain in steps, disturbing the setpoint slightly after each change.
- Find the gain at which the loop produces sustained oscillation of constant amplitude — neither growing nor decaying. That gain is the ultimate gain, $K_u$.
- Measure the period of one complete oscillation cycle. That is the ultimate period, $P_u$.
- Apply the settings:
| Controller | Gain $K_c$ | Integral time $T_i$ | Derivative time $T_d$ |
|---|---|---|---|
| P only | $0.5,K_u$ | — | — |
| PI | $0.45,K_u$ | $P_u / 1.2$ | — |
| PID | $0.6,K_u$ | $P_u / 2$ | $P_u / 8$ |
Worked example. A temperature loop is driven to sustained oscillation at a gain of 4.0 with a cycle period of 8 minutes. For PID control:
The caution that must accompany the method: finding $K_u$ means deliberately driving a live process into sustained oscillation. On a reactor, a fired heater, a compressor or anything with a safety interlock nearby, that is unacceptable. The closed-loop method is used on tolerant loops, on a simulator, or not at all.
Open-loop (process reaction curve) method
Safer, because the process is never destabilized:
- Place the controller in MANUAL.
- Make a step change in output, typically 5 to 10%.
- Record the PV response and read three values off the curve: the deadtime (delay before the PV moves at all), the time constant (time to reach 63.2% of the total change), and the process gain (change in PV per unit change in output).
- Apply the corresponding tuning rules.
Modern distributed control systems automate this as auto-tune, which performs a step or a relay-feedback test and calculates the constants. Auto-tune gives a good starting point; it does not know what the operator wants the loop to do, which is why an averaging level loop and a tight level loop get different answers from the same process.
Quarter-amplitude damping
The traditional performance target is a response in which each successive overshoot is about one quarter the amplitude of the one before. It is a reasonable compromise between speed and stability, and it is what the Ziegler-Nichols rules are designed to produce.
7. Advanced Strategies an Industrial Electrician Meets
| Strategy | What it does | Typical application |
|---|---|---|
| Cascade | A master (outer) loop sets the setpoint of a slave (inner) loop; the inner loop must be several times faster | A reactor temperature master setting a jacket-water temperature slave; a level master setting a flow slave |
| Ratio | Holds one flow at a fixed proportion of another (the wild flow) | Blending, combustion air-to-fuel ratio |
| Feedforward | Measures a disturbance and corrects before the PV moves; almost always trimmed by a feedback loop | Boiler feedwater corrected for steam flow |
| Split range | One controller output drives two final elements over different portions of the range | 0–50% opens the cooling valve, 50–100% opens the heating valve |
| Override / selector | A high or low selector picks the more conservative of two controller outputs | Compressor capacity control overridden by a minimum-flow surge controller |
Cascade tuning order: always tune the inner (slave) loop first, with the master in manual, and only then tune the master. Tuning the master against a sloppy slave produces settings that are meaningless the moment the slave is fixed.
8. Diagnosing the Loop That Will Not Behave
This is the practical heart of F-30.04, because most "tuning problems" are not tuning problems.
| Observed behaviour | Most likely cause | Action |
|---|---|---|
| Steady offset from setpoint, no oscillation | No integral action, or integral time far too long | Add integral action |
| Fast oscillation, growing amplitude | Gain too high | Reduce gain |
| Slow, persistent oscillation | Integral action too aggressive (integral time too short) | Lengthen integral time |
| Large overshoot only after startup | Reset windup — anti-windup missing or misconfigured | Enable and limit anti-reset windup |
| Output jitters rapidly; valve packing wears fast | Derivative acting on a noisy signal | Remove derivative or filter the PV |
| Square-wave or sawtooth PV oscillation while the controller output ramps smoothly | Valve stiction — the valve sticks, then jumps past the required position | Service the valve; retuning will never fix this |
| Loop sluggish and hard to tune at any settings | Valve grossly oversized, so it works entirely in the first 10% of travel | Resize the valve or trim |
| Loop cycles with a long, fixed period unrelated to gain | Large deadtime (long transport delay) | Reduce gain, lengthen integral, or use a deadtime-compensating strategy |
| PV noisy and erratic but the process is steady | Transmitter, wiring, grounding or a ground loop on the 4–20 mA signal | Chase it as an instrument fault, not a control fault |
The stiction signature is worth memorizing
A sticking valve produces a very distinctive trace: the controller output ramps smoothly while the PV sits still and then jumps, giving a square or sawtooth PV waveform and a triangular output waveform. The loop appears to be oscillating, and a technician who responds by lowering gain simply makes the cycle slower without removing it. The valve needs mechanical attention — packing friction, a corroded stem, a failed positioner — and no set of PID constants will compensate.
Optimization beyond tuning
Sub-task F-30.04 is broader than PID constants. Optimizing an industrial control system also means:
- Scan time — moving high-speed logic into a periodic task or a high-speed counter module so it is not limited by the main program scan;
- Alarm rationalization — eliminating nuisance alarms so operators respond to the real ones;
- Setpoint and sequence review — confirming that setpoints still match current production, because process conditions drift over years while setpoints do not;
- Energy optimization — VFD minimum speeds, pressure setpoints, and duty cycling that were set conservatively at commissioning and never revisited.
A temperature control loop is tuned using the Ziegler-Nichols closed-loop method. With integral and derivative disabled, sustained constant-amplitude oscillation occurs at a proportional gain of 4.0 with a cycle period of 8 minutes. What PID settings do the Ziegler-Nichols rules give?
A flow control loop cycles continuously. The trend shows the controller output ramping up and down in a smooth triangular pattern while the process variable sits still and then jumps, producing a square-wave trace. A technician has already reduced the gain twice, which slowed the cycle but did not stop it. What is the actual fault?
An industrial electrician is commissioning a cascade loop in which a reactor temperature controller sets the setpoint of a jacket-water temperature controller. In what order should the loops be tuned, and why?