INSTRUMENTATION & CONTROL SYSTEMS

Chapter 5

Process Controllers
CLO4 · Weeks 19–23

On-Off · Proportional · Derivative · Integral · PID · Tuning · Digital · Fuzzy Logic · Neural Networks
CHAPTER 5 — OUTLINE
Process Controllers: from simple switching to intelligent control

5.1 Introduction — Control Laws, Dead Time, Capacitance 5.2 On-Off Control — Bimetallic Strip, Wax Thermostat, Relays

5.3 Proportional Control (P) — Gain, Proportional Band, Offset 5.4 Derivative Control (D) — PD Control

5.5 Integral Control (I) — PI Control, Eliminating Offset 5.6 PID Control — Three-Mode Controller, Windup

5.7 Tuning — Ziegler-Nichols Methods, Lambda, Adaptive 5.8 Digital Control Systems — Direct Digital Control

5.9 Fuzzy Logic Control — Membership Functions, Rules 5.10 Neural Networks — Training, Direct & Indirect Control
5.1 INTRODUCTION TO PROCESS CONTROLLERS
Control laws (modes) define how a controller responds to an error signal

A process controller takes an error signal (required − actual) and outputs a corrective signal to the process.

Direct Action Dead Time / Lag Capacitance

Increase in INPUT → Increase in OUTPUT System cannot respond instantly. Delays The larger the 'volume' of a system, the
while accommodating change. longer it takes to respond to changes.
Example: Central heating furnace —
more fuel input = higher temperature Causes: sensor response time, distance- Example: Heating a large room takes
output velocity lag (pipe flow), transfer delays longer than a small room

Reverse Action: Increase in input → DECREASE in output (e.g. air conditioner — more energy = lower temperature)
5.2 ON-OFF CONTROL
Simplest form of control — output is either ON or OFF

Controller Output = ON (100%) when error exists | OFF (0%) when error = 0 → Also called: Two-Step Controller

Principle: Characteristics:
Controller Output
ON ✓ Simple and inexpensive

✓ Good for large capacitance systems

✗ Causes cycling/oscillation of variable

✗ Cannot handle slow fine corrections

OFF → Dead band reduces switching frequency

Error
Switch point
ON-OFF CONTROL: OSCILLATION & DEAD BAND
Why the controlled variable cycles — and how dead band reduces it

Without Dead Band — Oscillation: With Dead Band (Neutral Zone):

Temp

Overshoot Switch OFF (top of dead band)

DEAD BAND

Set value
Switch ON (bottom of dead band)

Example: Set point = 20°C

ON when temperature falls to 19.5°C

OFF when temperature rises to 20.5°C

Undershoot Dead band = 1°C

Time → Large dead band → large fluctuations

• Heater overshoots set value → switches OFF
• Temperature falls below set value → switches ON Small dead band → frequent switching
• Cycle repeats — oscillations around set point
5.2.1 RELAYS — ON-OFF CONTROLLER
Electromagnetic relay: small current controls large current

Relay Principle: Key Points:

Small LOW-VOLTAGE error signal → controls LARGE

current correction
Solenoid Electromagnet
(Low voltage (Attracts
control circuit) armature) When current drops below critical level → spring
pushes armature back → OFF

Used for: heaters, motors, solenoid valves

Switch Armature
Contacts (Spring brass
(High current) strip) Modern equivalent: solid-state relays (SSR) using
thyristors/transistors

Error signal applies small current → solenoid creates magnetic field → Applications: room thermostats, motor starters,
electromagnet attracts armature → switch contacts close → large correction industrial control panels
current flows to actuator
5.3 PROPORTIONAL CONTROL (P)
Controller output ∝ size of error

Controller Output = Kp × Error where Kp = Proportional Gain

How it works: Example — Float-Lever Proportional Controller:

Float
Output%

Slope = Kp

Cistern Lever
(Process) (Pivot)

Larger error → larger output correction

Smaller error → smaller output correction
Zero error → zero output Output movement = (x/y) × error
Error% Gain = x/y (lever ratio)
5.3.1 PROPORTIONAL BAND (PB)
% deviation required to produce 100% change in controller output

% PB = (% Deviation / % Change in Controller Output) × 100 | PB = (1/Kp) × (output span / input span) × 100

PB Chart — different gains: Key Terminology:

% Controller
100% Two extreme operating values (e.g. 4–
Output Range:
50% PB 20 mA)
(High gain)
Difference between extremes (e.g. 0–
Span:
30°C → span = 30°C)

Absolute Deviation: Error = measured − set value (e.g. 3°C)

50%
(deviation/span) × 100 (e.g. 3/30 × 100
% Deviation:
= 10%)

50% error → 100% output change

50% PB:
100% PB (high gain)
(Lower gain)
100% error → 100% output change
100% PB:
(lower gain)
-50% +50% % Deviation
5.3.2 LIMITATION OF PROPORTIONAL CONTROL — STEADY-STATE ERROR
(OFFSET)
P control always has a residual error in steady state

ALL proportional-only controllers have a non-zero STEADY-STATE ERROR (offset). Higher Kp reduces error but risks
oscillation.

Why Steady-State Error Exists: Step Response with P Control:

Set value
For proportional control:
Controller Output = Kp × Error
Offset
(Steady-state
For the plant to deliver required output, a NON-ZERO error signal is error)
Steady state (P only)
needed to keep the controller producing an output.

If error → 0, then controller output → 0, and the plant cannot maintain

the required output.

Therefore: some residual error ALWAYS remains.

5.4 DERIVATIVE CONTROL (D)
Output ∝ RATE OF CHANGE of error — anticipates future error

Derivative Output = Kd × (rate of change of error) Kd has units of TIME (derivative time Td)

Key Characteristics: D Response to linearly increasing error:

Responds to RATE of change, not Error

Anticipates:
magnitude
Time →

Fast Response: Large output when error changes quickly Constant output
Output (because rate of change
of error is constant)

No D-only: D alone gives zero output for constant error

D alone cannot handle constant errors

Reduces oscillation: Damping action slows overshoot (constant error → zero rate of change → zero output)
→ Always combine with P control: PD controller

Noise sensitive: Amplifies high-frequency noise

5.4.1 PD CONTROL — PROPORTIONAL + DERIVATIVE
Fast response + damping, but still has steady-state error

PD Output = Kp × error + Kd × (rate of change of error)

= Kp × [error + Td × (rate of change of error)] where Td = Kd/Kp (Derivative Action Time)

Worked Example: Advantages of PD:

Given: Kp = 4, Td = 0.4 s Handles FAST process changes better than P alone

Error increases from 0 at rate of 2%/s
Derivative action responds immediately when error starts
At t = 0:
changing
Error = 0, rate of change = 2%/s
Output = Kp × Td × rate = 4 × 0.4 × 2 = 3.2%
(Derivative action only) Reduces overshoot and oscillation (damping effect)

At t = 2s:
Error = 4% Used in temperature control, position control
P part: 4 × 4% = 16%
D part: 4 × 0.4 × 2% = 3.2% Limitation: Still has steady-state error — cannot achieve zero offset
Total output = 16 + 3.2 = 19.2% for constant disturbances. Add I mode to fix this.
5.5 INTEGRAL CONTROL (I)
Output ∝ integral (accumulated area) of error over time

I Output = Ki × ∫ error dt (area under error-time graph) Ki has units of s⁻¹

How Integral Works: I Response to step error:

Error Area = Ki × output

The output equals the AREA under the error-time graph.

A constant error → area grows linearly with time → output increases Output
at constant rate.
Output grows
Key insight: linearly (ramp)
Rate of change of output ∝ error

Can provide output even when error = 0

(because past area is remembered)
→ Eliminates steady-state error!

Example: Ki = 0.10 s⁻¹, constant error = 20%. At t=1s: output = 0.10×20 = 2%. At t=2s: output = 0.10×40 = 4%
5.5.1 PI CONTROL — PROPORTIONAL + INTEGRAL
Eliminates steady-state error — most common industrial combination

PI Output = Kp × error + Ki × ∫error dt

= Kp × [error + (1/Ti) × ∫error dt] where Ti = Kp/Ki (Integral Action Time)

PI Response to Step Error: Key Advantage:

Error
PI total
The integral part provides a controller output even when error = 0
I part (remembers past errors). This ELIMINATES steady-state offset.

Use PI when:
• Large changes in process variable occur
P part • Steady-state accuracy is critical
• Changes are relatively slow

Limitation:
Slower response than P alone — takes time to integrate. Fast
changes → possible oscillation.

Solution: Add D mode → PID controller

5.6 PID CONTROL — THREE-MODE CONTROLLER
The most widely used industrial controller — combines P, I, D

PID Output = Kc × [error + (1/Ti) × ∫error dt + Td × (d error/dt)]

Proportional (P) Integral (I) Derivative (D)

Kp × error Ki × ∫error dt Kd × d(error)/dt

Provides initial correction proportional to Accumulates past errors — eliminates Responds to rate of change — anticipates &
error offset damps

✓ Fast response ✓ Zero steady-state error ✓ Reduces overshoot

✗ Steady-state error ✗ Slower, can oscillate ✗ Noise sensitive

PID = P (fast response) + I (eliminate offset) + D (reduce oscillation) → Best overall performance
5.6 PID CONFIGURATIONS & INTEGRATOR WINDUP
Parallel, Interactive, Non-Interactive — and the windup problem

Non-Interactive (C)
Parallel (A) Interactive (B)
(Ideal Form)

P, I, D act independently. Controller gain Kc affects all three modes. Controller gain Kc affects all modes.
Kp does NOT affect I or D terms. Derivative acts on output of integrator. I and D act directly on proportional output.
Rarely used in practice. Commonly used. Most common industrial form.

INTEGRATOR WINDUP — A Critical Problem with PID Controllers

Cause: When actuator reaches its saturation limit (e.g. valve fully open, motor max
speed), it cannot produce more output — but the integrator keeps integrating. The Solution: Anti-Windup Circuit
integral term becomes very large.
• Detects when actuator reaches saturation
• Automatically switches off integral action
Effect: Large overshoot when system finally reaches set point. Long recovery time. • All commercial PID controllers include this
CONTROL MODE COMPARISON — STEP RESPONSE
How P, PI, and PID each respond to a sudden change in set point

(A) Proportional (P) (B) PI Control (C) PID Control

Set value Set value Set value

Offset

Fast response No steady-state error No steady-state error

Oscillations decay More oscillations before settling Less oscillation than PI
Steady-state ERROR remains Slower than P alone D mode damps oscillations
✗ Has offset ✓ No offset, more oscillation ✓ Best overall performance
5.7 TUNING — SELECTING OPTIMAL PID SETTINGS
Finding the right Kp, Ki, Kd for best system performance

Tuning Goal: fast response + rapidly decaying oscillations + minimum overshoot + no steady-state error

Process Reaction Ultimate Cycle Quarter Amplitude Auto-Tuning

Lambda Tuning
Method (Z-N) Method (Z-N) Decay (Software/Adaptive)

Open-loop test. Closed-loop test. Controller injects small

Step change in manual
Apply step change to Increase Kp until Adjust P until oscillation disturbance.
mode.
correction element. sustained oscillations amplitude decays by ¼ per Measures system
Measure dead time D,
Measure lag L and begin. cycle. response.
time constant τ, process
gradient M of reaction Record critical gain Kpc Then set Ti = T/1.5, Td = Adjusts parameters
gain Kp.
curve. and period Tc. T/6. automatically.
Non-oscillatory response.
Calculate Kp, Ti, Td from Apply Z-N tables for Trial and error approach. Available in all modern
No overshoot design.
tables. settings. PID controllers.
5.7.1 PROCESS REACTION METHOD (ZIEGLER-NICHOLS)
Open-loop step test — measure lag L and gradient M

Z-N Settings from Process Reaction Curve (P = % step change in correction element, R = M/P = normalised slope)

Controller Kp Ti (integral time) Td (derivative time)

P only P / (M × L) — —

PI 0.9 × P / (M × L) 3.3 × L —

PID 1.2 × P / (M × L) 2×L 0.5 × L

WORKED EXAMPLE

Process reaction curve from 10% step: Lag L = 5 min, gradient M = 0.8%/min
→ Kp = 1.2 × 10 / (0.8 × 5) = 3.0
→ Ti = 2 × 5 = 10 min → Ki = Kp/Ti = 3/10 = 0.3 min⁻¹
→ Td = 0.5 × 5 = 2.5 min → Kd = Kp × Td = 3 × 2.5 = 7.5 min
5.7.2 ULTIMATE CYCLE METHOD (ZIEGLER-NICHOLS)
Closed-loop: increase Kp until sustained oscillations begin

Procedure: Z-N Settings (Quarter Amplitude Decay)

1 Set controller to manual, plant near normal conditions Controller Kp Ti Td

2 Turn off I and D — proportional only

P 0.5 Kpu — —
3 Set Kp to a LOW value (wide proportional band)
PI 0.45 Kpu Tu/1.2 —
4 Switch to automatic — introduce small set-point change

5 Increase Kp until SUSTAINED oscillations (neither grow nor decay) PID 0.6 Kpu Tu/2 Tu/8

6 Record: Kpu = critical gain, Tu = period of oscillation

7 Apply Ziegler-Nichols tables to get optimal settings

Example: Kpu = 3.3, Tu = 500 s → PID quarter-amplitude decay:

Kp = 0.6 × 3.3 = 1.98 Ti = 500/2 = 250 s (Ki = 1.98/250 = 0.0079 s⁻¹) Td = 500/8 = 62.5 s (Kd = 1.98 × 62.5 = 123.75 s)
5.8 DIGITAL CONTROL SYSTEMS
Direct Digital Control (DDC) — microcontroller implements PID

Direct Digital Control: computer/microcontroller samples feedback, calculates control signal, outputs to DAC at each sampling interval

Micro-
Digital DAC Actuator
controller Plant
Set Value → + ADC/DAC
→ (Digital→ → (Correction → (Process)
Output
(Memory) Analog) Element)
+ Control Law

Measurement + ADC (Analog→Digital)

← Sampled digital feedback

Microcontroller Program Cycle for PID:

① Sample ADC inputs (set value + feedback) ② Compute error ③ Calculate P component ④ Calculate I component (error × Δt, add to previous
integral sum) ⑤ Calculate D component (current error − previous error) ⑥ Sum all three → output to DAC ⑦ Wait for next sampling interval ⑧
Repeat
5.9 FUZZY LOGIC CONTROL
Control based on linguistic rules rather than exact mathematics

Traditional: crisp YES/NO decisions vs Fuzzy: degrees of membership (0 to 1 scale — partial truth)

Classical (Crisp) Set Theory: Fuzzy Set Theory:

Person is either in 'TALL' set OR 'SHORT' set Overlapping sets — person can be in both
No in-between — hard boundary at 1.8 m 'TALL' and 'SHORT' sets simultaneously
with different degrees of membership
Membership
Short Tall Short Tall
1

1.8m Height →
At 1.5m: membership 0.251.8m in Tall, 0.75 in Short
(fuzzy
→ 'Not very tall, quite short'
boundary)
5.9.1 FUZZY MEMBERSHIP FUNCTIONS — TEMPERATURE EXAMPLE
A temperature can belong to multiple sets with different degrees

Fuzzy sets: COLD (0–20°C), WARM (10–30°C), HOT (20–40°C) — overlapping ranges allow gradual transitions

1 Membership
COLD Function WARM HOT
(0 to 1)

0.5

0°C 10°C 18°C 20°C 30°C Temperature

40°C(°C) →

At 18°C: COLD membership = 0.2 | WARM membership = 0.8 | HOT membership = 0.0
→ 18°C is mostly WARM with a little COLD — controller decides action based on these weighted memberships
5.9.2–5.9.3 FUZZY LOGIC CONTROL SYSTEM
Fuzzification → Rules → Defuzzification → Crisp Output

2. RULE EVALUATION
1. FUZZIFICATION 3. DEFUZZIFICATION
(IF-THEN Rules)

Apply linguistic rules to fuzzy inputs: Convert fuzzy output back to a single crisp
Convert crisp input values into fuzzy set
value.
memberships using membership functions.
⟹ IF temp is HIGH THEN output is LOW
IF temp is MEDIUM THEN output is MEDIUM
⟹ Mamdani's method: truncate output
Example: Temperature 18°C → COLD(0.2),
IF temp is LOW THEN output is HIGH membership functions at their membership
WARM(0.8)
values, combine (OR), find centre of gravity.
Rules written using common sense — no
Maps sensor inputs to 0–1 scale.
complex maths needed. Output: single crisp control signal.

Fuzzy rules for heater ON-time control:

IF error is Large Negative → timer ON very long | IF error is Zero → timer ON medium | IF error is Large Positive → timer ON very short
5.9.3 FUZZY LOGIC APPLICATIONS
Cruise control, washing machines, PID auto-tuning

Automobile Cruise Control Washing Machine Control PID Auto-Tuning

Inputs: Inputs: Inputs:

Velocity error (too fast/zero/too slow) Weight of load, fabric type, Steady-state error, oscillation level,
Acceleration (positive/zero/negative) degree of dirtiness response speed

Rules: Rules: Rules:

• IF dirt low AND fabric delicate AND weight
• IF too fast AND accel negative → minimum
low → short wash, low spin, little water, low • IF steady-state error large → increase Kp
force
temp

• IF dirt high AND fabric cotton → long wash,

• IF zero error AND accel zero → normal force • IF response oscillatory → increase Kd
high spin

• IF too slow AND accel positive → maximum

• IF overshoot too large → decrease Kp
force

MATLAB Fuzzy Logic Toolbox provides software for simulating and analysing fuzzy control systems
5.10 NEURAL NETWORKS — INTRODUCTION
Inspired by the human brain — neurons, connections, weights, learning

ANN: Artificial Neural Network — mathematical model of interconnected neurons that learns patterns from data through
training

ANN Architecture: Key Concepts:

Basic processing unit — sums weighted inputs,

Neuron:
fires if threshold exceeded

Connection strengths — determine influence

Weights:
of each input on output

Input Hidden Output Adjust weights by comparing actual vs desired

Layer Layer(s) Layer
Training:
output (supervised learning)

Extract patterns from data — complexity

Hidden Layer:
determines capability

Threshold/Firin Neuron outputs 1 if weighted sum > threshold,

Output g: else 0
Inputs Weights
5.10.1 NEURAL NETWORKS FOR CONTROL
Direct and indirect neural control — learning from data

Direct Neural Control Indirect Neural Control Advantages over PID

The neural network IS the controller.

Neural network models the PLANT (not the
Replaces the conventional controller. Can Can handle non-linear systems
controller). Predicts plant behaviour based
act like on/off switch AND learn to No mathematical model required
on history.
recognise when conditions change. Learns from data patterns
Adapts to changing conditions
Conventional controller parameters are
Example: Room heater with 4 inputs (room Good for complex multi-variable systems
derived from the model. Especially useful
temp, Δroom temp, outdoor temp, MATLAB Neural Network Toolbox available
when mathematical plant model is
Δoutdoor temp) → 9 hidden neurons → 1 for simulation
unknown.
output (ON/OFF heater)

Comparison: Neural Networks vs Fuzzy Logic vs PID

PID: needs model, works well for linear systems | Fuzzy: rule-based, no model needed, good for uncertain systems | Neural: learns from data, best
for unknown non-linear systems
 SUMMARY — ALL CONTROL MODES COMPARED
Quick reference for all controllers covered in Chapter 5

Eliminates Handles Fast

Controller Output Formula Complexity Typical Use
Offset? Changes?

On-Off ON or OFF only No (cycling) No (lag) Very Low Thermostat, kettle

No (steady-state
P Kp × e Moderate Low Simple processes
error)

PD Kp×e + Kd×ė No (still has offset) Yes (fast, damped) Medium Position control

Not ideal (may

PI Kp×e + Ki×∫e dt Yes Medium Process industry
oscillate)

PID Kp×e + Ki×∫e + Kd×ė Yes Yes (well damped) High Most industrial processes

Fuzzy Logic Rule-based IF-THEN Possible Yes (flexible) Medium-High Appliances, non-linear

Neural
Learned from data Possible Yes (adaptive) Very High Complex/unknown systems
Network
REVIEW QUESTIONS — CHAPTER 5

1. What is meant by 'dead time' and 'capacitance' in a control system? How do they affect controller design?

2. Why does on-off control cause oscillation? How does a dead band reduce this problem?

3. A proportional controller has Kp=5 and needs to maintain 60% output. What is the % steady-state error?

4. Explain why proportional control always has a steady-state error. Which mode eliminates it?

5. A PD controller has Kp=4, Td=0.4s. Error increases at 2%/s. Find output at t=0 and t=2s.

6. Calculate PID settings using Z-N ultimate cycle method if Kpu=3.3, Tu=500s.

7. Describe the three stages of a fuzzy logic controller: fuzzification, rules, defuzzification.

8. What is the difference between direct and indirect neural control?

 END OF CHAPTER 5

Process Controllers
Instrumentation and Control Systems

Next: See accompanying exercise workbook for practice problems with full solutions