i

Control Engineering - 1
Martin Braae
Department of Electrical Engineering
University of Cape Town

Licensed to
Control & Instrumentation Laboratory,
Department of Electrical Engineering,
University of Cape Town.
 Published by the author

Copyright: Martin Braae

First published by UCT Press (Pty) Ltd in 1994
Second edition in 2001
First impression in 2001

All rights are reserved. No part of this publication may be

reproduced, stored in a retrieval system, or transmitted in
any form or by any means, electronic, mechanical,
photocopying, recording or otherwise, without the prior
permission of the publisher.

Produced by
MBuct, ElecEng, UCT.
 iv

Contents
Contents .........................................................................................................iii

1. Introduction ...................................................................................... 1
1.1. Classification of Control Systems ...........................................................2
1.1.1. Logic Controllers........................................................................2
1.1.2. Continuous Controllers...............................................................3
1.2. Open Loop Control Systems ...................................................................4
1.2.1. Block Diagram for Open Loop Systems .....................................5
1.2.2. A Disadvantage of the Open Loop Configuration ......................6
1.3. Closed Loop Control Systems.................................................................7
1.3.1. Block Diagram for Closed Loop Systems...................................7
1.3.2. Advantages of Closed Loop Configurations ...............................8
1.3.3. Examples of Closed Loop Systems.............................................10
1.4. Common Features in Feedback Control Systems ....................................13
1.5 Design Procedure .....................................................................................14

2. System Modelling ............................................................................. 15

2.1. Weighting Functions and Convolution Integrals.....................................16
2.1.1. Linearity......................................................................................18
2.2. Practical Use of the Weighting Function Model .....................................20
2.2.1. Predictions Based on the Weighting Function............................21
2.2.2. Cascade Connection ...................................................................21
2.2.3. Parallel Connection ....................................................................21
2.2.4. Feedback Connection .................................................................22
2.3. Differential Equations .............................................................................22
2.3.1. Predictions ..................................................................................23
2.3.2. Cascade Connection -- An example............................................24
2.3.3. Parallel Connection -- An example.............................................24
2.3.4. Feedback Connection -- An example..........................................25
2.4. Transfer Functions ..................................................................................26
2.4.1. Series Connection .......................................................................28
2.4.2. Parallel Connection ....................................................................29
2.4.3. Feedback Connection .................................................................29
2.5. Summary of System Models....................................................................30
 v

3. Common Mathematical Models....................................................... 31

3.1. Electrical Systems ...................................................................................31
3.1.1. Resistor .......................................................................................31
3.1.2. Capacitor.....................................................................................32
3.1.3. Inductor ......................................................................................32
3.1.4. Operational Amplifier.................................................................33
3.1.5. A Transistor Circuit ....................................................................33
3.2. Mechanical Systems ................................................................................35
3.2.1. Spring .........................................................................................35
3.2.2. Damper .......................................................................................36
3.2.3. Inertia..........................................................................................36
3.2.4. Levers .........................................................................................37
3.2.5. Gears...........................................................................................37
3.3. Electro-Mechanical Systems ...................................................................38
3.3.1. Field-controlled dc Motor...........................................................38
3.3.2. Armature-controlled dc Motor....................................................39
3.3.3. dc Generator ...............................................................................40
3.3.4. Motor-Generator Set...................................................................41
3.4. Other Systems .........................................................................................41
3.5. Practical Derivation of Transfer Function Models ..................................42
3.5.1. Model for an Operational Amplifier (Electronic device) ............42
3.5.2. Model for an AC motor (Electro-Mechanical Device) ...............43
3.5.3. Model for a Large Plant (Industrial system) ...............................45
3.6. Block Diagram Algebra...........................................................................46
3.7. Application of Block Diagram Algebra...................................................49

4. Dynamic Variables and Laplace Transforms ................................. 52

4.1. Common Laplace Transforms .................................................................53
4.1.1. Step Function (Constant position) ..............................................53
4.1.2. Ramp Function (Constant velocity) ............................................53
4.1.3. Parabolic Function (Constant acceleration) ................................54
4.1.4. Impulse Function (Dirac Delta Function) ...................................54
4.1.5. Pulse Function (Practical approximation to Dirac Delta) ...........54
4.1.6. Exponential Function..................................................................55
4.1.7. Sinusoidal Functions ..................................................................55
4.1.8. Exponentially Decaying Sinusoids .............................................56
4.2. Useful Properties of Laplace Transforms ................................................56
4.2.1. Real Translation..........................................................................56
4.2.2. Transform of Derivatives ............................................................56
4.2.3. Transform of Integrals ................................................................57
4.2.4. Linearity Theorem ......................................................................57
4.2.5. Final Value Theorem ..................................................................57
4.2.6. Initial Value Theorem.................................................................57
4.3. Laplace Transform Tables .......................................................................58
 vi

4.4. The Inverse Transform ............................................................................58

4.5. Steady State Gain ....................................................................................60

5. Steady State Behaviour and System Type Number....................... 61

5.1. System Type Numbers ............................................................................64
5.2. Steady State Position Error .....................................................................64
5.3. Steady State Velocity Error.....................................................................65
5.4. Steady State Acceleration Error ..............................................................66
5.5. Summary .................................................................................................66
5.6. Controller Design Using the Error Constants..........................................69
5.7. The Internal Model Principle ..................................................................71

6. Prediction of System Response...................................................... 72

6.1. Basic Equations.......................................................................................72
6.2. Calculation of System Responses............................................................73
6.3. Prediction of System Behaviour..............................................................75
6.3.1. Simple Real Roots ......................................................................75
6.3.2. Simple Complex Conjugate Pair of Roots ..................................76
6.3.3. Repeated Roots (For completeness only)....................................77
6.4. Summary .................................................................................................78
6.5. Two Generic Examples ...........................................................................79
6.5.1. A Single Real Root .....................................................................79
6.5.2. A Complex Conjugate Root Pair ................................................80
6.6. Characteristic Function and Poles ...........................................................81
6.7. System Modes .........................................................................................82
6.8. Detailed Analysis of Two Elementary Systems.......................................83
6.8.1. First Order System......................................................................83
6.8.2. Second Order System..................................................................85
6.9. Guide to the s-Plane ................................................................................87
6.10. Pole Dominance ....................................................................................87
6.11. The Effect of System Zeros ...................................................................90
6.12. Graphical Evaluation of Residuals ........................................................93

7. Routh-Hurwitz Stability Criterion .................................................... 95

7.1. Application of Routh-Hurwitz Criterion .................................................98
7.2. Relative Stability from a Routh-Hurwitz Array.......................................100
7.2.1. Further extensions.......................................................................101
 vii

8. Feedback Control Systems ............................................................. 102

8.1. The Closed Loop Control System ...........................................................103
8.2. Non-Unity Feedback ...............................................................................105
8.3. Closed Loop Performance.......................................................................107
8.4. Closed Loop Design................................................................................108

9. Root Locus Design Method ............................................................. 109

9.1. Root Locus Design Methods ...................................................................109
9.2. Sketching Root Loci................................................................................113
9.3. Design Using the Root Locus..................................................................119
9.3.1. Target Region in the s-Plane.......................................................120
9.4. Root Locus for Variations of Parameters ................................................124
9.5. Standard Compensation Elements...........................................................126
9.5.1. The Lag Circuit...........................................................................127
9.5.2. The Lead Circuit .........................................................................128
9.6. General Compensators ............................................................................129

10. Dead Time and the Root Locus....................................................... 130

10.1. Dead-Time in System Responses ..........................................................130
10.2. Pade Approximation of Dead-Time ......................................................132

11. Frequency Domain Design Methods............................................... 133

11.1. Frequency Response Models.................................................................133
11.1.1. Frequency Response Models from Experimentation ................134
11.1.2. Frequency Response Models from Transfer Functions.............134
11.2. Representation of Frequency Response Models....................................137
11.3. Mathematics for Frequency Domain Methods ......................................139
11.3.1. Mapping of s-plane Contours to the q-plane ............................139
11.3.2. Principle of the Argument.........................................................142
11.3.3. Use of Symmetry ......................................................................143
11.3.4. Further simplification ...............................................................143
11.3.5. The Nyquist Plot.......................................................................144
11.4. Closed Loop Stability from Nyquist Plots.............................................144
11.4.1. Number of Encirclements .........................................................149
11.5. Relative Stability from Nyquist Plots....................................................150
11.6. Phase Margins and Gain Margins .........................................................151
11.6.1. C/L Damping Factor from O/L Phase Margin ..........................153
11.6.2. C/L Overshoot from O/L Phase Margin....................................154
11.6.3. General Systems .......................................................................155
11.7. Constant M Circles in the Nyquist Plot.................................................156
11.8. Constant N Circles in the Nyquist Plot .................................................158
 viii

12. The Nyquist Plot Design Method..................................................... 159

12.1. Standard Compensators in the Nyquist Plot..........................................162
12.1.1. Lag Circuits ..............................................................................163
12.1.2. Lead Circuits.............................................................................164
12.2. Summary of Lag and Lead Compensators.............................................166

13. Alternative Frequency Response Plots .......................................... 170

13.1. Bode Diagrams......................................................................................170
13.1.1. Typical Bode Diagram..............................................................171
13.1.2. Closed Loop Stability (Bode Diagrams) ...................................172
13.1.3. Gain and Phase Margins (Bode Diagrams)...............................173
13.1.4. Dynamic Compensators (Bode Diagrams)................................174
13.1.5. Design Using the Bode Diagram ..............................................175
13.1.6. Variable Gains, f, in Bode Diagrams ........................................176
13.1.7. Non-Minimal Phase Systems (Bode Diagrams)........................177
13.2. The Nichols Chart .................................................................................179
13.2.1. Closed Loop Stability (Nichols Chart) .....................................180
13.2.2. Gain and Phase Margins (Nichols Charts)................................181
13.2.3. Constant M and N Contours (Nichols Charts)..........................181
13.2.4. Dynamic Compensators (Nichols Chart) ..................................182
13.2.5. Design Using the Nichols Chart ...............................................183
13.3. The Inverse Nyquist Plot.......................................................................185
13.3.1. Closed Loop Stability (Inverse Nyquist Plot) ...........................186
13.3.2. Gain and Phase Margins (Inverse Nyquist Diagram)................188
13.3.3. Constant M and N Contours (Inverse Nyquist Plot) .................188
13.3.4. Dynamic Compensators (Inverse Nyquist Plots) ......................190
13.3.5. Design Using the Inverse Nyquist Plot .....................................192

14. Summary of Linear Design Methods .............................................. 194

14.1. Pole-Zero (Transfer Function) Methods, q(s) .......................................195
14.1.1. Root Locus................................................................................195
14.1.2. Characteristic Locus .................................................................195
14.2. Frequency Response Methods, q(jω) ....................................................195
14.2.1. Bode Diagrams .........................................................................195
14.2.2. Nichols Chart............................................................................196
14.2.3. Nyquist Plot ..............................................................................196
14.2.4. Inverse Nyquist Plot .................................................................196
14.2.5. Arithmetic Plot .........................................................................196
14.2.6. Rutherford-Aikman Plot ...........................................................196
14.3. Comment...............................................................................................197
 ix

15. Compensation Techniques ............................................................. 198

15.1. The PID or Three-term Controller.........................................................198
15.2. Design of PID Compensators ................................................................202
15.3. Minor Loop Compensation ...................................................................206
15.4. Inverse Response Compensator.............................................................208
15.4.1. Multiplicative Cancellation of Dynamics .................................208
15.4.2. Additive Cancellation of Dynamics ..........................................209
15.5. General Compensation ..........................................................................214
15.6. Feedforward Control .............................................................................216
15.7. Full Control Configuration....................................................................219

16. Sensitivity and Disturbance Rejection ........................................... 221

16.1. Sensitivity..............................................................................................221
16.1.1. Algebra for Sensitivity Functions .............................................222
16.1.2. Application to Feedback Control Loops ...................................222
16.1.3. Representation of Sensitivity on Nyquist Plots.........................223
16.1.4. Representation of Sensitivity on Inverse Nyquist Plots ............224
16.2. Sensitivity to Feedback Elements..........................................................225
16.3. An Alternative Approach ......................................................................228
16.4. Sensitivity and the Rejection of Disturbances.......................................229
16.5. Sensitivity of Closed Loop Pole Positions ............................................232
16.6. General Analysis ...................................................................................237
16.6.1. Sensitivity of Closed Loop Poles to Root Locus Gain, α .........238
16.6.2. Sensitivity of Closed Loop Poles to Open Loop Poles .............238
16.6.3. Sensitivity of Closed Loop Poles to Open Loop Zeros.............238

17. Electronic Circuitry from Transfer Functions ................................ 239

17.1. Two Basic Analog Electronic Building Blocks.....................................239
17.1.1. Summation (of Voltages)..........................................................239
17.1.2. Integration (of Voltages)...........................................................240
17.2. Electronic Circuit for a Comparator ......................................................241
17.3. Electronic Circuit for a Transfer Function ............................................243
17.3.1. A Second Order Example .........................................................243
17.4. Analog Simulators.................................................................................251

18. Engineering Applications ................................................................ 252

18.1. A General Process Control Study..........................................................252
18.2. Non-Linear Systems ..............................................................................257
18.3. Control of a Robot Arm Joint................................................................257
 x

Preface

Modern systems theory provides engineers with an extremely powerful set of

mathematical techniques that have found extensive application in the design of
control systems for diverse industries. Unfortunately, newcomers to the field are
often put off by all its mathematics and many a practising engineer has admitted that
Control Theory, Nyquist Plots and more are merely a vague and slightly unpleasant
memory from undergraduate days. This is a great pity since the theory can produce
extremely useful engineering designs that solve tough practical problems, as well as
providing a rich medium for academic activities.

These notes were written for a first course in Control Engineering given to Electrical
Engineering undergraduates (and their enlightening contributions are gratefully
acknowledged). Thus the notes emphasise electro-mechanical systems and try to
show how aspects of Linear Control Theory can be applied to the design of
electronic circuits that control processes with poor or bad dynamics and ensure
acceptable final systems.

The material is presented in a conversational style in that equations and figures flow
within the text. Essential mathematics cannot be avoided but hopefully the powerful
results that it produces can inspire the effort needed to master it.

A computer program was written to produce design plots and time simulations that
appear in this text. It is used extensively by students on the course and can be
ordered if required. (Details are given at the back of this book.)

Prior exposure to complex variables, differential equations, Fourier transforms and

Laplace transforms is assumed.

Martin Braae
Oakridge,
June 1994
 1
Introduction
Many engineering problems are concerned with the application of large forces, or
high power, to achieve some pre-defined result.

For example, a power transistor regulates the flow of large electric currents, an
amplifier pulses high voltages in a radio transmitter, a petrol engine powers a car, a
jet engine propels an aircraft, a mechanical shovel shifts large masses of earth,
conveyor belts move tonnes of material through unit processes in an industrial plant,
electric motors turn drums that reel in cables that hoist lifts up and down shafts in
both buildings and mines, and so on. The list is virtually endless.

Central to such high-power equipment is a low-power control system that allows an

operator to direct the enormous power at his disposal to obtain the best utilisation of
that power. Thus small base currents control transistor operations, low-voltage
signals control output voltages in power amplifiers, buttons direct lift operation, an
accelerator pedal controls engine power, and levers control the mechanical shovel
and thrust delivered by aircraft engines.

Most power sources have become increasingly sophisticated over the years, as have
the associated control systems. In fact, many modern control systems perform
complex tasks which a human is physically unable to carry out since these are too
fast, too slow, too numerous, too monotonous or too dangerous. For example,
modern aircraft include many control systems that compensate automatically for
changes in flight conditions, provide auto-pilot facilities and even land the plane in
bad weather. As a result its performance is consistent under virtually all expected
operating regimes.

Demands upon the performance of these systems has increased significantly,

especially since the advent of low-cost electronics. For example, modern industrial
plants cannot be operated consistently near optimum efficiency without the
continuous attention provided by process control computers.

Thus a major concern of engineers today is the design and development of automatic
control systems. Such systems are found everywhere −− from the largest chemical
2 Control Engineering -1

plant to the smallest electronic circuit. The methods needed for designing control
systems to achieve specified results form the subject of this text.

1.1. Classification of Control Systems

Because of the vast spectrum of applications, control systems can take on many
forms and classification is an important precursor to analysis. In the first instance
control systems are classified according to the type of signals that they use. Thus
voltage control in a simple operational amplifier circuit is lumped in the same
category as altitude control in a supersonic jet aircraft, even though electronic
circuits and aircraft are vastly different. The common factor is simply that signals in
both systems vary continuously over some range.

1.1.1. Logic Controllers

Other systems use LOGIC signals (i.e. two-valued variables that are either On or
Off). Such systems are controllable by logic chips in digital electronic circuits, relay
racks for smaller industrial applications and PROGRAMMABLE LOGIC
CONTROLLERS (PLCs or PCs) for large installations.

Example

Consider a very small industrial system in which water is forced into a storage tank
by a centrifugal pump that is driven by an electric motor. Two level sensors are
positioned within the tank to monitor water level. These level sensors control the
operation of a single electrical switch that closes when the level falls below a
minimum value and opens when the level rises above a maximum value. During
shut-downs for maintenance, the safety of plant personnel is ensured by a lockable
off-switch. Another remote off-switch is provided for emergency stops. Finally a
temperature sensitive switch prevents thermal damage to the equipment.

The functions of the four switches that control the operation of the electric motor are
summarised below.

Switch Operation
S1 Closes when level low and opens when level high
S2 Locked open during maintenance
S3 Opened for emergency stops
S4 Opens when temperature is high
Ch 1: Introduction 3

This equipment and all its switches are interconnected to form a typical logic control
system, shown in the following schematic.

Figure 1.1 Simple Programmable Logic Control System

The design of such logic controllers is often reasonably straight-forward, relies on

an intimate knowledge of the process to be controlled and is carried out on PLCs
using ladder diagrams. Generally the switches are remote from each other and low-
voltage control signals are cabled through the various switches and used to energise
the coils of a relay that switches power to the motor. For large industrial
applications 10000s of logic signals are involved. Their interconnections are
determined by safety and operational criteria, and are programmed in PLCs which
are special industrial micro-computers. (The topic of logic controllers is not
discussed further. For more details refer to the book by Kissel,TE, Understanding
and using programmable controllers, Prentice-Hall, Englewood Cliffs, 1986.)

1.1.2. Continuous Controllers

If signals in a system are CONTINUOUS (e.g. vary over a range, like the industrial
standards of 4 to 20[mA] or 0 to 10[V]) then a continuous controller is needed.

Example

The simple OpAmp circuit shown below:

Figure 1.2 Simple Continuous Control System

4 Control Engineering -1

is driven by a potentiometer that is turned to give a time-varying voltage signal at its

output. Hence this electronic circuit contains signals that vary continuously, and is
classified as a CONTINUOUS SYSTEM.

The design of controllers for continuous systems is relatively straightforward for

OPEN LOOP CONTROL configurations but quite complicated for CLOSED LOOP
CONTROL configurations. Thus it is useful to classify continuous control systems
further to distinguish between these two types of control structures that are in
common use today. It should be noted that closed loop control is also known as
AUTOMATIC FEEDBACK CONTROL or AUTOMATIC CONTROL or simply
FEEDBACK CONTROL. Open Loop and Closed Loop control systems are now
discussed in some detail to illustrate their potential advantages and disadvantages.

1.2. Open Loop Control Systems

Consider an electric machine system in which voltage, Ef, from a power amplifier is
applied to the field coil of a DC motor in order to control its speed of rotation, Ω.

Figure 1.3 Typical Open Loop Control System

This is a typical open loop system. Speed changes are made by altering the field
voltage but there is no check on whether the desired speed is attained or not.

Many other examples of open loop control systems exist in everyday life. For
example:

• accelerator position controls car speed,

• potentiometers control volume on radio, brightness on TV,...,
• tap position controls water flowrate,
• joystick controls plane motion.

Again the list of systems is endless, but all its entries have one thing in common,
namely, that one variable, often low-power, is used to set the value of another
variable, often high-power.
Ch 1: Introduction 5

1.2.1. Block Diagram for Open Loop Systems

All these diverse systems can be represented schematically as a block with an input
and an output. Thus the electric motor (which converts field voltage, Ef, to speed, Ω
) has the block diagram:

Figure 1.4 Motor Speed Control System

Examples of block diagrams for some common open loop systems are:

Figure 1.5 Other Open Loop Control System

In general, industrial control systems are quite complex and typically consist of a
number of blocks that are interconnected to form a larger system. Alternatively,
details of a large system, given as one block, can be defined in a block diagram that
shows its component sub-systems.

As an example, consider an electric motor that drives a generator. A possible block

diagram for the combined Motor-Generator system (or MG set) is:

Figure 1.6 A Complex Open Loop Control System

It is assumed that the interconnection of blocks does NOT alter the characteristics of
the individual blocks in any way. Thus voltages in the system of resistors:
6 Control Engineering -1

Figure 1.7 A Resistor System

are related by the simple equation:

R2
Vout ( t) = Vin (t )
R1 + R 2

Unfortunately this system of resistors CANNOT be defined as a block:

Figure 1.8 Resistor Block Diagram

since its characteristics are very likely to be altered by interconnection with similar
systems. (More information is necessary to turn it into a true block diagram.)

1.2.2. A Disadvantage of the Open Loop Configuration

Open loop configurations are preferred whenever possible due to their simple
structure, lower component count and ease of design. Unfortunately there are
numerous practical instances in which open loop control is not adequate, because its
characteristics are changing, it is subjected to external disturbances or it is unstable.

For example, a common problem with open loop systems is that the relationship
between the output, like rotational speed Ω(t) of an MG set, and the input, like its
field voltage Ef(t), is neither fixed nor predictable. (Note that such uncertainty in an
open loop system is an indication that it should be controlled in closed loop.)

To illustrate the phenomena, consider the motor-generator system described

previously. If the load on the generator is changed then the torque will change and
hence alter system speed. Graphically the relationship between field voltage, Ef, and
speed, Ω, is a set of curves, each corresponding to a constant torque condition.
Ch 1: Introduction 7

Figure 1.9 Steady State Running Characteristics

Thus for a particular field voltage the actual rotational speed of the motor can vary
considerably, depending on the prevailing torque or load conditions. This is
indicated in the above sketch by the open loop operating line. Notice that setting the
field voltage does not ensure firm control over the motor speed.

1.3. Closed Loop Control Systems

One way to overcome this problem is simply to measure motor speed using a
tachometer and then to vary the field voltage in such a way that the motor speed
remains constant. This ensures that the generator maintains its output voltage in
spite of load fluctuations.

Thus, if the load increases and the motor-generator system slows down, then the
field voltage is increased to ensure that the motor speeds up and remains at the
desired speed. Similarly, should the motor speed up, then the field voltage is
reduced to maintain its speed.

In a feedback control system the field voltage is adjusted automatically.

1.3.1. Block Diagram for Closed Loop Systems

The closed loop system for the motor-generator system is more complex than the
equivalent open loop system as can be seen from its block diagram:
8 Control Engineering -1

Figure 1.10 Typical Closed Loop Control System

The components and configuration of the system are chosen so that the actual speed
of the generator is regulated automatically at the desired value. To achieve this
effect, three new elements have been added to the original system. These are:

• the TACHOMETER BLOCK that measures motor-generator speed

and provides a signal, Measured speed, that is proportional to the
actual motor speed, Ω,

• the COMPARATOR that compares the measured speed to the desired

or Set speed to give an Error signal,

• the CONTROLLER that takes the error signal and produces an

appropriate field voltage, Ef.

1.3.2. Advantages of Closed Loop Configurations

Feedback or Closed Loop control ensures greater accuracy for the relationship
between the input and the output variable than is achieved by open loop control.

To illustrate this, consider the following two systems, one in open loop the other in
closed loop.

Figure 1.11 Comparison of Open and Closed Loop Control

Ch 1: Introduction 9

In both instances a field control voltage of 10[V] is applied to a motor system and
ensures a speed of 1000[rpm].

Now assume that a fault occurs on the motor, or that its load changes, so that the
speed halves in the open loop system. In the closed loop system, feedback provided
by the tacho-generator compensates automatically for the changes that have
occurred in the motor characteristics, thereby minimising its drop in speed.

Figure 1.12 Effect of Motor Change

Observe in particular that the control system moves the process input to ensure that
its output remains constant. Graphically the steady state operating line for the closed
loop configuration is vastly different to that of the open loop system. The reason is
very simply that the closed loop reacts to compensate for unexpected effects, such as
changes in load.

Figure 1.13 Closed Loop Running Characteristics

Thus closed loop control gives improved precision for the process being controlled,
at relatively small additional cost. This is a general property of closed loop
feedback control systems and accounts for their popularity in engineering.
10 Control Engineering -1

1.3.3. Examples of Closed Loop Systems

The versatility of the feedback configuration is now illustrated by listing a number

of diverse engineering problems to which it has been applied. It should also be
noted that the control hardware can take a number of forms, though those with
electronic controllers are considered most significant in this text.

Operational Amplifier Circuit (Electronics application)

Figure 1.14 Closed Loop Control of Output Voltage

The output voltage is related accurately to the input voltage with little sensitivity to
changes in the amplifier gain, A. All the components of this closed loop system are
electronic. (This circuit is so well known that its feedback configuration may not be
obvious, and it is dealt with later.)

Sump Level Regulation

Figure 1.15 Level Control

A common method of level control. In more elaborate systems the float can be
designed to control a power source, such as a pneumatic or hydraulic valve, that in
turn controls the main water valve

Steam Turbine (Mechanical application)

 Ch 1: Introduction
11

Control engineers of today owe a great debt to designers who wrestled with the
practical aspects of speed regulation in steam turbines, shown schematically as:

Figure 1.16 Closed Loop Speed Control

This system has great historic significance since it is the first application of feedback
control in the Western World, devised by James Watt in 1769.

When the steam turbine's speed of rotation increases the weights on the governor are
thrown outwards under centrifugal forces. This closes the steam valve and hence the
turbine slows down. Similarly as the speed of rotation slows, the valve is opened
and the turbine speeds up. In this way the governor regulates system speed. The
main components of this control loop are mechanical and consequently more costly
to adjust than simple electronic components used in modern speed regulators.

Missile Control (Avionics application)

Figure 1.17 Missile Control

The radar/telecoms/computer system keeps the missile on target by monitoring and

feeding back relative target/missile positions continuously.
12 Control Engineering -1

Temperature Control (Process control application)

Figure 1.17 Temperature Control

Room temperature is maintained by switching the refrigeration unit on or off

according to the temperature measured by a thermocouple.

Ship Navigation (Maritime application)

Figure 1.19 Automatic Navigation

An on-board control computer compares the course set for the ship with that
measured via a navigational satellite. The error is then used to adjust the rudder
angle and keep the ship on course.

Paper Winder System (Industrial application)

In countless modern applications, electrical engineering devices, both light and

heavy current, play a central role in the operation of many industrial systems. The
paper industry has been studied extensively and is an important example of such
applications.
 Ch 1: Introduction
13

Figure 1.20 Paper Tension Control

Here paper arrives from the production process at a varying rate. The function of the
control system is to roll the paper onto a wind-up reel without tearing or distorting
it. The controller does this by maintaining constant tension.

1.4. Common Features in Feedback Control Systems

All these closed loop control systems are vastly different in detail, but have a
number of important things in common:

• a plant or process whose performance varies appreciably in open loop,

thereby necessitating feedback control,

• an OUTPUT variable, which is a measurement of an important

physical quantity in the process,

• a SETPOINT, that defines a desired value that the measured output

should have,

• an INPUT variable that is manipulated in order to drive the output

variable towards its setpoint,

• a CONTROL LAW that uses the setpoint and the output variable to
compute the best input variable continuously.

Once again the input is usually low-power and the output high-power.

Control Engineering deals with techniques that are used to design control laws
which ensure optimal performance for such closed loop systems. The processes that
can be controlled are diverse but in many instances the sensors that measure the
outputs, the actuators that set the inputs and the devices that implement the control
laws are electronic, sometimes analog but more often digital. In effect the
14 Control Engineering -1

instrumentation converts the process under investigation into a sophisticated

electronic device, thereby expediting its control.

Thus Electrical Engineers today are often involved in the design of control systems
and may even take on the role of Control Engineers who use electronics for process
optimisation.

1.5 Design Procedure

To derive a satisfactory controller for a given process it is necessary to:

• study the process and determine its primary function in terms of its
continuous signals,

• classify its signals as outputs and inputs, and identify failings of the
present process in its open loop configuration,

• derive a mathematical model that describes the manner in which the

process input affects its output,

• design a controller using one of the many design methods available to

the control engineer. (The controller is defined in the same
mathematical form as the process model.)

• convert the controller model to an electronic circuit (in this text, or

other appropriate physical realisation in general),

• implement the controller on the process and commission the closed

loop system. (This step it often glossed over in academic courses but
could require considerable effort and is vital to ensure a successful
product.)

The technical skills required to design suitable electronic circuits to control

processes fitted with electronic instrumentation are dealt with in subsequent
chapters.
 2
System Modelling
To design a controller for a given process, its behavior must be described by a
suitable mathematical model. Such models are chosen to be as simple as possible
and specifically tailored to meet the needs of the control engineer.

For example, in the typical configuration for a feedback or closed loop control
system:

Figure 2.1 Basic Closed Loop Control System

mathematical models for the Plant (relating OUTPUT to INPUT) and the Controller
(relating INPUT to ERROR) are combined to form a mathematical model for the
open loop system (relating OUTPUT to ERROR) and more significantly for the
closed loop system (relating OUTPUT to SETPOINT).

Figure 2.2 Controlled System

The mathematical model chosen for describing the system should ideally:

• predict the Output from a given Input (or Setpoint),

• allow easy analysis of Cascade, Parallel and Feedback sub-systems,
• be useful for designing controllers,
• be as simple as possible without becoming simplistic.

Consider the problem of modelling the electric motor system:

16 ControlEngineering - 1

Figure 2.3 Servo-motor

Here the field voltage, Ef, is the input variable, u(t), and the speed, Ω, is the output
variable, y(t). Variations in Ef produce variations in Ω and the two signals can be
monitored continuously on an oscilloscope or a digital computer. Typically traces
might be:

Figure 2.4 Input-Output Time Signals for the Servo-motor

Assuming that the system does not change with time (i.e. is time-invariant) and that
all transients have decayed, then, whenever Ef follows the shape shown above the
speed will respond with the same shape as previously.

Thus one possible, though cumbersome, method of modelling a system is simply to

remember all possible inputs and the corresponding outputs. In principle this forms
the basis for two very important methods of modelling the dynamic behaviour of a
system mathematically.

2.1. Weighting Functions and Convolution Integrals

The practical alternative to recording all responses to all possible inputs is to record
the response to one particular shape of input such as the pulse function.
For example:
 Ch 2: System Modelling 17

Figure 2.5 Pulse Response Time Signals

The energy in the input pulse is normalised by giving it unit area. This pulse input is
then the unit pulse function. If its duration in time, ε, becomes extremely small it is
the UNIT IMPULSE FUNCTION (A function that is mathematically convenient but
unrealisable in practice).

The response of a system to a unit impulse input is known as its UNIT IMPULSE
RESPONSE or WEIGHTING FUNCTION, g(t). Note that the system response, y(t),
and hence its weighting function, g(t), is exactly zero for all time, t, less than zero.
This is essential for CAUSALITY.

For linear systems, if the impulse function is scaled to alter its area then the impulse
response is scaled proportionally. Thus an impulse of area A produces a response
y(t) = A g(t)

Also for linear systems, if two impulses of area A1 and A2 are applied to the system
then its response is the sum of the two impulse response functions:

Figure 2.6 Response of a Linear Systems

Mathematically, the response, y(t), to the given input, u(t), is given by

y(t) = A1 g(t) + A2 g(t-t2)
18 ControlEngineering - 1

2.1.1. Linearity

Most real physical systems are NOT linear but many can be assumed to exhibit
linear behaviour for small deviations about some operating point. Since many
analytical techniques in control engineering are only valid for linear systems
LINEARITY is an important concept.

Formally, a system is classed as LINEAR if its response to the input

u(t) = A1 u1(t) + A2 u2(t)

where Ai are constants, is given by

y(t) = A1 y1(t) + A2 y2(t)

where yi(t) is its response to input ui(t).

Thus the output from a linear system can be predicted from any input by:

• approximating the input as a sum of impulse functions,

• computing the impulse response for each impulse input,

• adding the individual impulse responses to give the output.

Example

Given the impulse response, g(t), for a particular system, compute its response to the
input function:

Figure 2.7a Decomposition of Input into Pulse Functions

By decomposition the input is approximated as a sum of impulse functions:

 Ch 2: System Modelling 19

Figure 2.7b Input Component Functions

The component impulse responses are then easily computed:

Figure 2.7c Component Responses

and combined to yield the system response to the given input function.

Mathematically the system response y(t) to input u(t) is computed from the
summation
y(t ) = u(0) ε g( t) + u( ε ) ε g( t - ε ) + u(2 ε ) ε g( t - 2 ε ) +...
t/ε
= å u( kε ) g(t - kε ) ε
k=0

where it is assumed that u(t) is zero for all time t < 0. (This is arranged easily in
practice by suitable definition of the input function.)

In the limit as ε → 0 the response, y(t), defined by the above summation converges
to the CONVOLUTION INTEGRAL
t

ò
y(t) = u(τ ) g(t − τ ) dτ
0

The primary importance of the convolution integral is in the computation of system

responses. (Digital computers deal easily with such integrals.) It is sometimes useful
to note that the concept of causality allows both the above summation and
integration to proceed to ∞.

Alternatives to the Impulse Input

The impulse response or weighting function is a simple and convenient method of

characterising the input-output relationship for a system. It is by no means unique.
20 ControlEngineering - 1

For example, the response to a UNIT STEP INPUT, which is more readily
implemented in practice, could as easily form the basis of a system model.

Figure 2.8 Unit Step Response

The response of systems to a SINUSOIDAL INPUT of unit amplitude and given

frequency is also a possibility, and is in fact an extremely versatile description of
systems.

Figure 2.9 Sinusoidal Response

2.2. Practical Use of the Weighting Function Model

In engineering applications the system model is needed for two primary reasons:

• prediction of system responses to given inputs,

• derivation of an overall model from the inter-connection of component

models in Cascade, Parallel and Feedback configurations.

(The first reason includes the design of control systems.)

2.2.1. Predictions Based on the Weighting Function

The prediction of system responses to arbitrary input functions is obtained from the
convolution integral
 Ch 2: System Modelling 21

t t

ò ò
y(t) = u(τ ) g(t − τ ) dτ = u(t − τ ) g(τ ) dτ
0 0

2.2.2. Cascade Connection

Consider two system blocks, with weighting function models g1 and g2, connected
in cascade:

Figure 2.10 Cascade Connection

The overall model relating y(t) to u(t) is given by the weighting function
t

ò
g(t) = g1(τ ) g 2 (t − τ ) dτ
0

which is derived by manipulation of the convolution integral.

2.2.3. Parallel Connection

Consider two system blocks connected in parallel:

Figure 2.11 Parallel Connection

The overall model is given by the weighting function

g( t) = g1 ( t) + g 2 ( t)
22 ControlEngineering - 1

which relates y(t) to u(t).

2.2.4. Feedback Connection

Consider the closed loop control system:

Figure 2.12 Feedback Connection

In this situation a model relating the output, y(t), to the setpoint, r(t), defines the
behaviour of the closed loop system. Now since
t

ò
y(t) = [r(τ ) − y(τ )] g(t − τ ) dτ
0

the closed loop system, h(t), can only be modelled implicitly in terms of the
weighting function for the open loop system.

Conclusion. The weighting function is useful for predicting system output

responses. It can be used to find overall models for sub-systems connected in
cascade (with some difficulty) or in parallel (easily) but cannot deal with sub-
systems connected in a feedback configuration. Also its use in controller design is
limited.

2.3. Differential Equations

An obvious alternative approach to system modelling is that based on differential
equations.

Extrapolating from years of experience in modelling the physical world, it is

assumed that the system input, u(t), and output, y(t), can be related by a differential
equation of the form
 Ch 2: System Modelling 23

d dn d dm
b o y( t) + b1 y( t) + ... + b n n y( t) = a o u( t) + a1 u( t) + ... + a m m u(t )
dt dt dt dt

where ai and bi are real constants and m ≤ n for realisable causal systems.

Example

The voltage and current in a simple RC network:

Figure 2.13 An Electronic Circuit

are related by the differential equation

d d
i( t) + RC i( t) = C v(t )
dt dt

This equation yields a compact parametric model for the electronic system. The
model is linear and can be used to characterise systems connected in cascade,
parallel and closed loop configurations. A useful feature of most differential
equation models is that the coefficients of the differential equations (and hence the
system response) can be related directly to the individual physical components
within the system (R and C here).

2.3.1. Predictions

It is very difficult to predict the response of a plant to an arbitrary input using a plant
model based on differential equations. However numeric integration methods like
Runge-Kutta algorithms or Predictor-corrector methods easily evaluate differential
equations using a digital computer. These may assist in predictions but give little of
the insight which is so essential to engineering design.

2.3.2. Cascade Connection -- An example

24 ControlEngineering - 1

Consider two extremely simple sub-systems connected in cascade

Figure 2.14 Cascade Connection

where each sub-system is modelled by a first-order differential equation

d
b1 y1 ( t) + y1 (t) = a1u( t) ... (2.1)
dt
d
b2 y( t) + y( t) = a 2 y1 (t ) ... (2.2)
dt

To derive a single differential equation for the combined system differentiate

Equation (2.2)

d2 d d
b2 2
y( t) + y(t) = a 2 y1 ( t) ... (2.3)
dt dt dt

Compute the addition: (2.2)+b1(2.3) and then substitute from (2.1)

d2 d d
b1b 2 y( t) + {b1 + b2 } y( t) + y( t) = a 2 {b1 y1 ( t) + y1 (t)}
dt 2 dt dt
= a 2a1u( t)

This result gives the differential model for the combined system.

2.3.3. Parallel Connection -- An example

Consider two sub-systems connected in parallel

Figure 2.15 Parallel Connection

where the sub-system models are

 Ch 2: System Modelling 25

d
b1 y1 ( t) + y1 (t) = a1u( t) ... (2.4)
dt
d
b2 y2 ( t) + y 2 (t ) = a 2 u( t) ... (2.5)
dt

Adding Equations (2.4) and (2.5) gives

d d
b1 y1 (t ) + b 2 y2 ( t) + y(t ) = {a1 + a 2 }u( t) ... (2.6)
dt dt

Differentiating both Equations (2.4) and (2.5)

d2 d d
b1 2
y1 (t) + y1 (t) = a1 u(t) ... (2.7)
dt dt dt

d2 d d
b2 2
y 2 (t) + y 2 (t) = a 2 u(t) ... (2.8)
dt dt dt

Computing b2(2.7)+b1(2.8) gives

d2 d d d
b1b 2 2
y( t) + b2 y1 ( t) + b1 y 2 ( t) = {a1b 2 + a 2 b1} u(t ) ... (2.9)
dt dt dt dt

Finally, adding (2.6) and (2.9) gives

d2 d d
b1b 2 y( t) + {b1 + b 2 } y(t) + y( t) = {a1b2 + a 2 b1} u( t) + {a1 + a 2 }u( t)
dt 2 dt dt

which is the differential equation describing the combined system.

2.3.4. Feedback Connection -- An example

Consider the simple feedback control system:

Figure 2.16 Parallel Connection

Assume that the plant is modelled by the simple first order differential equation
26 ControlEngineering - 1

d
b y( t) + y(t ) = au( t)
dt
= a{r ( t) − y( t)}

The model for the closed loop system is then simply

d
b y( t) + {1 + a}y( t) = ar (t)
dt

Conclusion. Models consisting of differential equations can be used to analyse

connected systems. The procedure however is still quite complex. Responses are
very difficult to predict analytically (a numeric integration algorithm is needed).

2.4. Transfer Functions

Laplace transforms convert differential equations to equivalent algebraic equations
that are much easier to deal with.

Recall that the Laplace transform of a derivative is given by

d
L[ y( t)] = sL[ y( t)] − y(0)
dt
= s y(s) - y(0)

where y(s) is the Laplace transform of variable y(t) and y(0) is the initial condition
for y(t) when t=0.

In most control applications the variables used within a plant model can be chosen
skilfully so that the initial conditions for the variables and their time-derivatives are
zero. This simplifies the analysis considerably, since

dk
L[ y( t)] = sk y(s)
dt k

Thus

Differentiation Multiplication
d/dt is equivalent to by s
in time domain in s-domain

Example
 Ch 2: System Modelling 27

To keep subsequent Laplace transformations as simple as possible, control engineers

typically arrange experiments and define variables in such a way that the initial
conditions of all variables and their higher derivatives are zero.

In practice this is achieved easily as illustrated below. The resulting variables have
simple Laplace transforms and are known as DYNAMIC variables.

To derive a model for any process the input variable is first held at a constant value.
Once the output variable steadies, the input is stepped to another value in a so-called
STEP TEST. The dynamic output variable is then defined on the graph by axes <t>
and <y(t)> as shown:

Figure 2.17 Dynamic Variable

The dynamic input variable is similarly defined. Note that the time scales, t, of the
two dynamic variables, y(t) and u(t), are synchronised and that all initial conditions
of both y(t) and u(t) are zero for t<0.

Figure 2.18 Dynamic Variable

The differential equation for the system is then transformed to its Laplace form

b0 y(s) + b1sy(s) + b2 s2 y(s)+... +b nsn y(s) = a 0u(s) + a1su(s) + a 2s2 u(s) +... +a msm u(s)
28 ControlEngineering - 1

with ai and bi real constants and m ≤ n. By re-arranging this equation, the Laplace
transform for the plant output, y(s), can be expressed as an explicit function of its
input, u(s), so that

a 0 + a1s + a 2 s2 + ... + a msm

y(s) = u(s)
b 0 + b1s + b 2s2 + ... + b nsn

The ratio of the two polynomials in 's' that arises in this equation is the TRANSFER
FUNCTION model, g(s), for the plant. So the plant is modelled by
y(s) = g(s) u(s)

The transfer function, g(s), defines the relationship between the system output and
its input and can be obtained directly from the Weighting Function, g(t), defined
earlier, since
g(s) = L[g(t)]

The transfer function model is linear and can be used to compute overall transfer
function models for systems connected in cascade, parallel and closed loop.

2.4.1. Series Connection

Consider the two sub-systems connected in cascade:

Figure 2.19 Cascade Connection

For the total system

y(s) = g 2 (s) u 2 (s)

= g 2 (s) y1 (s)

= g2 (s)g1 (s) u(s)

So the model for the inter-connected system is simply

g(s) = g1 (s)g 2 (s)
 Ch 2: System Modelling 29

2.4.2. Parallel Connection

Consider the two sub-systems connected in parallel:

Figure 2.20 Cascade Connection

Here
y(s) = y1 (s) + y 2 (s)

= g1 (s)u(s) + g 2 (s) u(s)

So the model for the combined system is

g(s) = g1 (s) + g 2 (s)

2.4.3. Feedback Connection

Consider the simple unity feedback control loop:

Figure 2.21 Cascade Connection

Here
y(s) = q(s) u(s)
= q(s) {r(s) - y(s)}

And so
q (s)
y(s) = r (s)
1 + q(s)

= h(s) r(s)
30 ControlEngineering - 1

where h(s) is the transfer function model for the closed loop system. It relates the
plant output, y(s), to its setpoint, r(s).

2.5. Summary of System Models

Three types of mathematical model for defining the dynamic behaviour of physical
plants have been described and are compared below.

Output Interconnection Form Useful in

predictions Cascade Parallel Feedback design
Weighting Easy but Difficult Easy Very No
Function tedious (+) difficult
Differential Very difficult Difficult Difficult Easy No
Equations (+)
Transfer Easy IFF Easy Easy Easy Yes
Functions transforms (x) (+) (+,x)
exist

Clearly, of the modelling options considered, transfer functions are the easiest and
most convenient analytical tools for dealing with inter-connected systems found in
control engineering loops. Analytical methods for designing control systems rely
extensively on transfer function models, g(s), and these are now studied in
considerable depth.

Additional model types, like Frequency Response Models, State Space Models,
Pulse Transfer Function Models and Discrete Models, exist and are dealt with
later as required.
 3
Common Mathematical
Models
When confronted with the task of deriving a mathematical model for a large inter-
connected industrial system, it is often convenient to start by splitting the original
system into simpler component sub-systems. These are then modelled individually
and re-combined later, using block diagram algebra, to provide the required model
for the entire system.

In most control engineering applications, it is vital to keep the models as simple as

possible, without over-simplification. This is something of an art. A few well known
electrical, mechanical and electro-mechanical components and their corresponding
models are given here.

3.1. Electrical Systems

3.1.1. Resistor

Figure 3.1 An Ideal Resistor

The basic mathematical model for the ideal resistance is

v1 (s) − v2 (s) = R i(s)
32 Control Engineering - 1

It is easily adapted to describe other circuit configurations. For example

1
i(s) = {v1 (s) − v 2 (s)}
R

is another possibility that applies to the resistor block diagram. The particular circuit
that is under investigation dictates which model form is required.

3.1.2. Capacitor

Figure 3.2 An Ideal Capacitor

The mathematical model for the ideal capacitance is

i(s) = sC {v1 (s) − v 2 (s)}

for the given block diagram, or, depending on the analytical requirements,
1
v1 (s) − v2 (s) = i(s)
sC

3.1.3. Inductor

Figure 3.3 An Ideal Inductor

The mathematical model for the ideal inductance is

Common Mathematical Models 33

1
i(s) = {v1 (s) − v2 (s)}
sL

or
v1 (s) − v2 (s) = sL i(s)

3.1.4. Operational Amplifier

An Operational Amplifier (or OpAmp) multiplies its input voltage, eA, by a very
large negative gain, -A, to give the OpAmp output voltage, eo. Its circuit diagram is
given below and shows the supply rails as well as the input and output voltages.

Figure 3.4 An Ideal OpAmp

Its block diagram representation is simply:

Figure 3.5 Block Diagram for OpAmp

3.1.5. A Transistor Circuit

A transistor is a current amplifier. When properly biased, its collector current, iC, is
β times its base current, iB. These two currents combine to form the emitter current,
iE, and the block diagram for the transistor on its own is:

Figure 3.6 Transistor Block Diagram

The popular emitter-follower circuit:

34 Control Engineering - 1

Figure 3.7 A Transistor Circuit

is analysed by first constructing a block diagram based on its components:

Figure 3.8 An Ideal Inductor

The diagram also indicates where a load current, iL, would fit in the loop.

A Passive Electric Circuit

The circuit below is in fact a feedback configuration as shown by its block diagram.

Figure 3.9 A Simple R-L Circuit

Common Mathematical Models 35

Individual blocks in the loop are defined as transfer functions, so it is easy to derive
overall models for the circuit. Assume that voltage v2 is the output and voltage v1 is
the input. The appropriate transfer function model is then
v2 (s) sL
=
v1 (s) R + sL

Another model is possible if the output is taken to be the current, i(s). Its transfer
function is
i(s) 1/ R
=
v1 (s) 1 + sL / R

3.2. Mechanical Systems

Modelling of linear and rotational electric devices, like solenoids and dc motors,
requires knowledge of mechanical systems. Basic models are provided below for
both linear and rotational cases, with corresponding devices shown in parallel.

3.2.1. Spring

Figure 3.10 An Ideal Spring

Mathematical models for the linear and the rotational springs are
F(s) = k{x1 (s) − x 2 (s)} and T(s) = k{θ1 (s) − θ2 (s)}

respectively, where F(s) is the net compression force acting on the spring in the
direction of x1(s) and T(s) is the net torque in the direction of θ1(s).
36 Control Engineering - 1

3.2.2. Damper

Figure 3.11 An Ideal Damper

The mathematical models are

F(s) = cs {x1 (s) − x 2 (s)} and T(s) = Bs {θ1 (s) − θ2 (s)}

where F(s) is the net compression force acting on the damper in the direction of
x1(s) and T(s) is the net torque in the direction of θ1(s).

3.2.3. Inertia

Figure 3.12 An Ideal Inertia

The mathematical models are

F(s) = M s2 x(s) and T(s) = J s2θ (s)

where F(s) is the net compression force acting on the mass in the direction of x(s)
and T(s) is the net torque in the direction of θ(s).
Common Mathematical Models 37

3.2.4. Levers

Figure 3.13 A Lever

For small angles of deflection, the mathematical model for levers is

b a
y(s) = x1 (s) + x 2 (s)
a+b a+b

3.2.5. Gears

Figure 3.14 A Set of Gears

The mathematical relationships between variables in the gear system are based on
the gear ratio, n, defined by
d2
n=
d1
1
Speed ratio: θ2 (s) = θ1 (s)
n

Torque ratio: T2 (s) = nT1 (s)

In practice it is often convenient to map inertia, torque and rotational speed through
a set of gears. Thereby every variable is referred to one shaft, before analysis starts.
38 Control Engineering - 1

3.3. Electro-Mechanical Systems

Electric machines combine electrical and mechanical components into one system
and utilise the concepts introduced above.

3.3.1. Field-controlled dc Motor

Figure 3.15 Field-Controlled dc Motor

For constant armature current, the torque developed by the motor is proportional to
field current and can be approximated by
T(s) = K t i f (s) ... (3.1)

where the current through the field coils is

1
i f (s) = e f (s) ... (3.2)
R f + sL f

The torque produced by the magnetic fields is balanced against the mechanical
damping and inertia so that

T(s) = sB θ (s) + s2J θ (s) = {sB + s2J} θ (s) ... (3.3)

Hence, combining transfer function models implied in equations (3.1), (3.2) and
(3.3), the position of the output shaft, θ(s), is related to the field voltage, ef(s), by
the overall transfer function
θ (s) θ (s) T(s) i f (s)
=
e f (s) T(s) i f (s) e f (s)

1 1
= 2
Kt
sB + s J R f + sL f
Common Mathematical Models 39

K t / (B R f )
=
s{1 + s(J / B)}{1 + s(L f / R f )}

K
=
s{1 + sTm }{1 + sTf }

where K = Kt/BRf is the motor gain in [Volt-seconds]-1,

Tm = J/B is its mechanical time constant in [seconds] and
Tf = Lf/Rf is its field coil time constant in [seconds].

In many instances the field time constant is much faster than the mechanical time
constant (Tf << Tm) and so this third-order transfer function model can be simplified
to an approximate second-order model
θ (s) K
=
e f (s) s{1 + sTm }

for reasons that are explained later.

3.3.2. Armature-controlled dc Motor

Figure 3.16 Armature-Controlled dc Motor

The equation for the armature circuit is derived from its voltages
ea (s) = R a ia (s) + sLa i a (s) + e m (s)

The torque developed by the motor is proportional to armature current

T(s) = K t i a (s)

while the back-emf is proportional to motor speed

e m (s) = Kesθ (s)

By algebraic analysis the transfer function for the motor is

40 Control Engineering - 1

θ(s) a0
=
ea (s) s{1 + b1s + b2s2 }

Its constants are derived by a detailed analysis of the block diagram for the motor,
which comprises the component sub-systems:

Figure 3.17 Block Diagram for the Armature-Controlled dc Motor

Note that the position of a load (or braking) torque in the loop is indicated on the
diagram. Once again there is a mechanical time constant (J/B) and an electric time
constant (La/Ra) associated with the loop, but they are not easily separated as in the
case of a field controlled motor.

3.3.3. dc Generator

Figure 3.18 A dc Generator

The field circuit gives

e f (s) = R f i f (s) + sLf i f (s)

For constant speed of rotation, the generator output voltage eg is given by

eg (s) = Kg i f (s)

where Kg is the constant of proportionality.

Common Mathematical Models 41

From a voltage balance around the generator loop

eg (s) = R g i g (s) + sLg i g (s) + Z L i g (s)

Hence the transfer function model between the field voltage and the load voltage is
e L ( s) i g Z L e g i f Kg Z L
= =
e f (s) eg i f e f (R f + sL f )(R g + Z L + sL g )

3.3.4. Motor-Generator Set

Equations for more complicated, interconnected systems like the following motor-
generator set:

Figure 3.19 A Motor-Generator Set

are easily developed using the previous models. The transfer function for the motor-
generator system is computed from the transfer function relationship

θ( s) θ (s) eg (s)
=
e f (s) eg (s) e f (s)

that combines the previous model for a motor with that for a generator.

3.4. Other Systems

These notes concentrate on systems that are common to electrical engineering. Thus
it must be stressed that transfer functions, similar to those introduced in detail for
electrical, mechanical and electro-mechanical devices, are also found for systems in
many other branches of engineering. For example, thermal, fluid, pneumatic,
hydraulic and aero-space systems, to mention but a few, are also amenable to
transfer function analysis. To carry out a control study of such systems, the
42 Control Engineering - 1

necessary differential equations or transfer function models must be obtained from

relevant texts. (In some instances such models may even be closely guarded
industrial secrets.)

3.5. Practical Derivation of Transfer Function Models

Effective mathematical modelling of real processes that provide simple yet realistic
definitions of plant characteristics requires experience. A few selected examples
illustrate the concepts.

3.5.1. Model for an Operational Amplifier (Electronic device)

Sometimes extraction of a suitable mathematical model merely requires

perseverance, as in the case of the operational amplifier circuit:

Figure 3.20 Simple Electronic Circuit

Voltage eA is kept near zero by the operational amplifier and current iA is very
small. The block diagram for the system is derived directly by inspection of the
original circuit:

Figure 3.21 Corresponding Block Diagram

By means of block diagram algebra (explained later) the overall system becomes:
Common Mathematical Models 43

Figure 3.22 Simplified Block Diagram

3.5.2. Model for an AC motor (Electro-Mechanical Device)

On other occasions a little extra effort is necessary. Consider a low-power two-phase

AC motor driving a constant load. The motor is subjected to an external disturbing
torque, TL. The balance of the torque is used to overcome the total system inertia, J,
and drag forces characterised by coefficient B. Schematically the system is:

Figure 3.23 Two-Phase ac Motor

The block diagram for the motor is simply:

Figure 3.24 Block Diagram for Two-Phase ac Motor

The torque, T(t), delivered by the motor is a complex non-linear function of its
speed of rotation, Ω(t), and the current, i(t), flowing through the control field coils.
44 Control Engineering - 1

Figure 3.25 Non-linear Torque Characteristic

Mathematically these sets of graphs define a non-linear relationship between

Torque, Current and Speed
T = N(Ω,i)

To derive a linear mathematical model for the dynamics of this motor, it must be
assumed that the motor is operating at a particular speed, Ωo, and draws a current of
io to overcome the friction and any load torque. Under these conditions the non-
linear function, N, can be linearized about an operating point, (Ωo, io), as follows
∂N ∂N
∆T(t) = ( Ω0 , i 0 ) ∆Ω (t) + ( Ω0 , i 0 ) ∆i(t)
∂Ω ∂i

∆T(t) = − K Ω ∆Ω(t) + K i ∆i(t)

where KΩ and Ki are approximately constant around a given operating point

( Ω0 , i 0 ). Taking Laplace transforms, changes in torque produced by the motor are a
linear function of the change in its speed and its current
∆T(s) = − K Ω ∆Ω(s) + K i ∆i(s) ... (3.4)

The electro-mechanical torque, T(s), produced by the motor must overcome drag
(B), inertia (J) and load torques (TL) within the system. From a torque balance
T(s) = {B + sJ}Ω(s) + TL (s)

Assuming small deviations from the operating point, this becomes

∆T(s) = {B + sJ}∆Ω(s) + ∆TL (s)

Re-arranging gives an explicit expression for speed

Common Mathematical Models 45

∆T(s) − ∆TL (s)

∆Ω(s) =
B + sJ

Substituting for ∆T(s) from Equation (3.4) yields

− K Ω ∆Ω(s) + K i ∆i(s) − ∆TL (s)
∆Ω(s) = ... (3.5)
B + sJ

Finally the current flowing in the field control coil is a function of the applied
voltage
e(s) = Z i(s)

where Z is the coil impedance, e(s) is r.m.s. voltage level and i(s) is the r.m.s.
current. So for small deviations about some (unspecified) operating point
∆e(s) = Z ∆i(s) ... (3.6)

From Equations (3.5) and (3.6) it is possible to draw a block diagram for the AC
motor system:

Figure 3.26 Block Diagram of Two-Phase Motor

These blocks can be combined, as shown later, to give an overall transfer function
for the electric motor system. (Note that the ∆ symbol is omitted.)

3.5.3. Model for a Large Plant (Industrial system)

Here the processes that are involved can be extremely complex, fundamental laws
cannot usually be applied and a heuristic approach is adopted. By measuring the
response y(t) to some input u(t), chosen to be simple (e.g. a STEP Function) or rich
in higher frequency components (e.g. PSEUDO-RANDOM Codes), the transfer
function is computed from the Laplace transformation
46 Control Engineering - 1

L[ y( t)]
g(s) = (Not always easy)
L[ u( t)]

Example

Consider a crusher used in the mineral extraction industry to break large rocks into
smaller particles.

Figure 3.27 Schematic of a Crusher

In the crusher control scheme, speed of the conveyor belt is used to regulate the
feedrate of rocks into the crushing chamber. This in turn influences the power drawn
by the motor that drives the crusher. Thus the block for the crusher system has
Conveyor Speed as input and Crusher Power as output. The relationship between
these two variables is highly non-linear and once again the mathematical model,
albeit heuristic, is a linearized approximation that is valid only for a limited region
about the operating point (at which the experiments were conducted).

3.6. Block Diagram Algebra

Analysis of most large systems leads to a set of inter-connected blocks, each
representing a relationship between important variables in the original system. (A
simple example being the AC motor discussed above). The primary advantage of
block diagrams is that they allow a one-to-one representation of components within
a system. For analysis of large systems it may be convenient to manipulate the
original blocks in such a way as to form a simpler overall block diagram. This
procedure is known as BLOCK DIAGRAM ALGEBRA. Some useful relationships
between blocks follow:
Common Mathematical Models 47

Rule (a): Cascade connection

Here the output from one block is the input to the next. The combined system block
is determined by eliminating the common variable.

Figure 3.28 Block Diagrams for Rule (a)

Rule (b): Parallel connection

The same input goes to both blocks while the two outputs are combined additively.

Figure 3.29 Block Diagrams for Rule (b)

Rule (c): Feedback connection

The input to the forward block, q(s), is the difference between the setpoint and the
output of q(s) fed back through the feedback element f(s).

Figure 3.30 Block Diagrams for Rule (c)

Rule (d): Interchange of Summing points

48 Control Engineering - 1

Comparators are easily moved but beware of the sign changes that are needed.

Figure 3.31 Block Diagrams for Rule (d)

Rule (e): Interchange of Block and Summing point

Figure 3.32 Block Diagrams for Rule (e)

Rule (f)

Figure 3.33 Block Diagrams for Rule (f)

Common Mathematical Models 49

Rule (g): Interchange of Block and Take-off point

Figure 3.34 Block Diagrams for Rule (g)

Rule (h)

Figure 3.35 Block Diagrams for Rule (h)

3.7. Application of Block Diagram Algebra

Recall that analysis of the two-phase ac motor led to the following block diagram:

Figure 3.36 Block Diagram of Two-Phase Motor

It is possible to apply the rules of block diagram algebra to simplify this diagram in
order to determine its transfer functions, relating speed, Ω, to control voltage, e, and
to load torque, TL. (The steps taken in the analysis are not unique but the final result
is.)
50 Control Engineering - 1

To simplify the expressions the blocks are given transfer function symbols, g1, g2, g3
and g4.

Figure 3.37 Two-Phase ac Motor Block Diagram

Using block diagram algebra these blocks are combined in stages. The rules that are
invoked are indicated within rectangles on the diagrams. It is vital to note that only
the items within the rectangle need be considered while applying the rule. First a
cascade and a comparator sub-section are simplified to yield:

Figure 3.38 Block Diagram after First Simplification

Then the feedback loop is transformed to one closed loop block.

Figure 3.39 Block Diagram after Second Simplification

Common Mathematical Models 51

Finally a block and a summer are interchanged.

Figure 3.40 Final Block Diagram

Substitution of transfer functions, (g1, g2, g3 and g4) for the sub-systems gives the
relationship between shaft position and control voltage as well as the effect of the
load torque.

Note

The requirements of the actual problem dictate the manipulations that need to be
carried out on the block diagram. For example, a feedback design might not need
load torque explicitly, and the last step above would be unnecessary. Similarly an
investigation to re-design block g3 would keep it separate and not lump it with any
other blocks.

When used in a flexible manner block diagrams and their associated algebra can
become invaluable tools in formulating and analysing complex engineering systems.
 4
Dynamic Variables and
Laplace Transforms
Once a transfer function model has been derived for the particular system under
investigation, control engineers are interested in predicting its dynamic performance.
For example, the system might be unstable and thus virtually useless, or it might be
too oscillatory or too slow (or both) to perform its specified task. Experienced
designers quickly and easily obtain such information from a detailed investigation of
the transfer function model itself, without considering its input and output in detail.
These conclusions are then exploited in the design to produce a satisfactory final
system that is stable, damped and fast.

To understand how such predictions are made by inspection of transfer function

models, and when required, to compute the response of a system output to a given
input, it is necessary to deal with dynamic variables (or signals) and their associated
Laplace Transforms in a formal manner.

The aim is to be able to draw sensible conclusions about transfer function models
like
y(s) 2 + 4s
= g(s) = ... (4.1)
u(s) 1.01 + 0.2s + s2

simply by studying the model itself, without computing its output for a given input.
(The model could represent either an open loop or a closed loop system.)

For example, it is possible to rewrite the given model, g(s), as

2 + 4s 4(s + 0.5)
g(s) = =
1.01 + 0.2s + s2 (s + 0.1 + j)(s + 0.1 − j)

and know immediately that the system has a steady state gain of 1.98 and that its
transients oscillate with a frequency of 1[rad/s] and decay away in about 40[s].
Ch 4: Dynamic Variables and Laplace Transforms 53

4.1. Common Laplace Transforms

Comprehensive tables of Laplace transforms are given in most texts on control
theory. To exploit these transformations in practice, it is useful to be familiar with
those most commonly found in engineering applications.

4.1.1. Step Function (Constant position)

In the field of process control, variables are often moved suddenly from one value to
another. The equivalent dynamic variable, which results when all initial conditions
are removed, is a STEP function.

Figure 4.1 A Step Function

Mathematically a step of amplitude a is defined in the time and the Laplace domains
as

Time domain Laplace Transform

f(t) = a for t≥0 ∞

ò
a
f(s) = ae −st dt =
=0 elsewhere s
0

4.1.2. Ramp Function (Constant velocity)

Figure 4.2 A Ramp Function

54 Control Engineering - 1

The RAMP function with slope a is defined as

Time domain Laplace Transform

f(t) = at for t≥0 ∞

ò
a
f(s) = at e −st dt =
=0 elsewhere s2
0

4.1.3. Parabolic Function (Constant acceleration)

The PARABOLIC function is defined as

Time domain Its Laplace Transform is

1 2 a
f(t) = at for t≥0 f(s) =
2 s3
=0 elsewhere

4.1.4. Impulse Function (Dirac Delta Function)

The IMPULSE function is defined by the Dirac Delta Function which is zero
everywhere, infinite at zero and has finite area.

Time domain Laplace Transform

f(t) = δ(t) ∞

=∞ when t=0 ò
f(s) = δ (t) e−st dt
0
=0 elsewhere
=1

4.1.5. Pulse Function (Practical approximation to Dirac Delta)

Ch 4: Dynamic Variables and Laplace Transforms 55

Figure 4.3 A Pulse Function

Time domain Laplace Transform

a ∞
f(t) = when 0 ≤ t ≤ ε
ε
ò
f(s) = f(t) e −st dt
0
=0 elsewhere
sin(φ)
≈ form
φ

4.1.6. Exponential Function

Figure 4.4 An Exponential Function

The EXPONENTIAL function is defined by

Time domain Laplace Transform

f(t) = e -at
for t≥0 f(s) =
1
s+a
=0 elsewhere

4.1.7. Sinusoidal Functions

Time domain Laplace Transforms

f(t) = sin(ωt) for t≥0 ω
f(s) =
s + ω2
2

and and
f(t) = cos(ωt) for t≥0 f(s) =
s
s2 + ω 2
56 Control Engineering - 1

4.1.8. Exponentially Decaying Sinusoids

Time domain Laplace Transforms

f(t) = e − at sin(ω t) for t≥0 ω

f(s) =
(s + a)2 + ω 2

and and
− at
f(t) = e cos(ω t) for t≥0 s+a
f (s) =
(s + a)2 + ω 2

4.2. Useful Properties of Laplace Transforms

Most engineering analyses rely on the above basic transforms and deal with more
complex expressions by simplification that relies on a number of properties of
Laplace transformations. These are:

4.2.1. Real Translation

If a function is shifted in time then only the phase angle of its Laplace Transform is
altered. Mathematically the relevant equation is

L [f (t − τ )] = e-sτ f (s) where f(s) = L[f(t)]

4.2.2. Transform of Derivatives

L [f (1) (t)] = sf (s) − f (0) → sf (s)

L [f (2) (t)] = s2 f (s) − sf (0) − f (1) (0) → s2 f (s)

:
:
L [f (n) (t )] = s n f (s) − sn −1f (0) − . . . − f (n −1) (0) → s n f (s)

d nf
where f (n) (0) = Dn f(0) = (0) . Note the recursive nature of these equations.
dt n
Ch 4: Dynamic Variables and Laplace Transforms 57

4.2.3. Transform of Integrals

1 1 1
L [f (-1) (t)] = f (s) + f ( −1) (0) → f (s)
s s s

1 1 1 1
L [f (-2) (t)] = f (s) + f ( −1) (0) + f ( −2) (0) → f (s)
s2 s2 s s2
:
:
1 1 1 1
L [f (-n) (t)] = f (s) + f (-1) (0)+... + f − n (0) → f (s)
sn sn s sn

It is worthwhile glancing back now at sections 4.1.1 to 4.1.4 (pages 53 to 54) and
observing carefully how, for example, a unit step function with zero initial
conditions is derived by integration of a unit impulse function, while a ramp arises
by differentiating a parabolic function, cosines from sines, etc.

Clearly, mixing basic Laplace transforms with appropriate properties provides the
engineer with a powerful set of analytical tools.

4.2.4. Linearity Theorem

L [ α1f1 (t) + α 2 f2 (t)] = α1f1 (s) + α 2 f2 (s)

where αi are real constants.

4.2.5. Final Value Theorem

lim f (t) = lim sf (s)

t →∞ s→ 0

where the theorem only holds for f(s) with poles in the left-half of the s-plane (i.e.
for stable functions).

4.2.6. Initial Value Theorem

lim f (t) = lim sf (s)

t →0 s→∞
58 Control Engineering - 1

4.3. Laplace Transform Tables

For easy reference the Laplace transforms and properties are summarised below.

Function Time domain Laplace Transform

Impulse f(t)=δ(t) f(s)=1
Step f(t)=u(t) 1
f(s) =
s
Ramp f(t) = t 1
f(s) = 2
s
Parabola f(t) = t 2 1
f(s) = 3
s
Exponential f(t) = e − at f(s) =
1
s+a
nth-order f (t) = t n e − at f (s) =
n!
Exponential (s + a)n+1
Sinusoidal f (t) = sin( ωt) ω
f (s) = 2
(Sin) s + ω2
Sinusoidal f (t) = cos( ωt ) s
f (s) = 2
(Cos) s + ω2
Decaying Sinusoid f (t) = e − atsin( ωt) f (s) = 2
ω
(Sin) (s + a)2 + ω 2
Decaying Sinusoid f (t) = e − atsin( ωt) f (s) = 2
s+a
(Cos) (s + a)2 + ω 2

Property Time domain Laplace Transform

Real Translation f(t-τ) e −sτ f(s)
Differentiation L[f (n) (t)] s n f(s) [ −s n −1f(0)−... − f (n −1) (0)]
L[f (-n) (t)] 1 1 (-1) 1
Integration f(s) [+ f (0)+...+ f − n (0)]
sn sn s
Linearity α1f1 (t) + α 2 f2 (t) α1f1 (s) + α 2 f2 (s)
Final Value Thm lim f(t) lim sf(s)
t→∞ s→ 0
Initial Value Thm lim f(t) lim sf (s)
t→0 s→∞

4.4. The Inverse Transform

Wherever Laplace transforms are used to predict system response to an input
function, an inverse Laplace transform is required to convert the output function,
y(s), to a time response, y(t).
Ch 4: Dynamic Variables and Laplace Transforms 59

Formally the Inverse Laplace Transform is given by the integral

y(t) = L−1 [y(s)]

ò
1
= y(s) est ds
2π j
C

where C is a suitable contour in the s-plane.

Luckily, the inverse transform is linear so this integral conversion is rarely required.
Most practical problems expand the original transform into a set of simpler
transforms using partial fractions, followed by standard inversions from tabulated
results.

Example (Inversion)

Consider inversion of the function

3(s + 1)
y(s) = (which is not in the tables)
s(s + 3)

To find the corresponding time function, y(t), start by expanding y(s) in partial
fractions. This gives
1 2
y(s) = +
s s+3

And so, from tables,

1 1
y(t) = L−1 [ ] + 2 L−1 [ ] (Linearity)
s s+3
= 1 + 2 e −3t (From tables)

Example (Final Value Theorem)

2
Find the value that y(t) reaches when the process model has its input suddenly
s+3
changed by 6 units. Thus the input function is given by the Laplace transform
6
u(s) = (From tables)
s
60 Control Engineering - 1

Hence the output response is

2 6
y(s) = From y(s) = g(s) u(s)
s+3 s

The final value for y(t) is then given by

12
y ∞ = lim sy(s) = lim s =4
s→ 0 s→ 0 s(s + 3)

Example (Initial Value Theorem)

In the previous case, how fast does y(t) change immediately after its input is
stepped?
d
L[ y(t)] = sy(s) (Assuming zero initial conditions)
dt
2 6
=s (Derivative rules)
s+3 s

Hence
d
y(0) = lim s[sy(s)] = lim s2 y(s)
dt s→∞ s→∞

12
= lim s = 12
s→∞ s+3

4.5. Steady State Gain

A very useful application of the final value theorem is in the calculation of steady
state (DC) gain of a transfer function. This is a scalar constant, g(0), given by
g(0) = lim g(s)
s→ 0

Clearly the gain of g(s) is the final value of its response to an input that is a unit step
function.
 5
Steady State Behaviour
and System Type Number
The final values attained by variables in closed loop systems are vital to their
acceptance in practice. For example, a system designed to track a setpoint that is
held at a constant value may never reach a setpoint that is constantly changing,
unless its controller is properly specified.

In fact, exact tracking of setpoints that demand constant position, constant velocity
or constant acceleration in the process output variable, is an important criterion for
many controller designs.

To illustrate the problem of steady state error, consider a two-stand steel rolling mill
in which an automatic feedback system is designed to control the steel thickness,
Y(t), to a set value, R(t).

Figure 5.1 Schematic of a Rolling Mill

Assume that the system is modelled adequately by the transfer function

y(s)
g(s) =
u(s)
L [Change in steel thickness Y(t )]
= (Initial conditions zero)
L[Change in roller force U(t)]
62 Control Engineering - 1

−10
= [µm/N] (Note negative gain)
1 + 6s

and that a simple P (Proportional) controller is used in the loop

k(s) = −6 [N/µm] (For negative feedback)

The open loop transfer function is then

60
q(s) = g(s) k(s) = [µm/µm]
1 + 6s

and the closed loop transfer function is

q(s) 0. 98
h(s) = = [µm/µm]
1 + q(s) 1 + 0.1s

Note that the controlled system, h(s), has a gain close to unity and is 60 times faster
than the original plant, g(s), in open loop. (It is also more robust.)

In spite of these improvements, the control is not satisfactory since it suffers from a
steady state error when the setpoint is held at a constant value (or position). This is
shown by analysis of steady state conditions in the loop:

Figure 5.2 The Closed Loop Configuration

The transfer function that relates the error signal to the system setpoint is derived
from block diagram algebra to be
1
e(s) = r(s)
1 + q(s)
1 + 6s
= r(s)
61 + 6s

10
To roll steel 10 [mm] in thickness requires a constant step input r(s) =
[mm].
s
The final value for the error signal is computed simply from the Final Value
Theorem and is
 Ch 5: Steady State Behaviour and System Type Number 63

1 + 6s 10 10
lim se(s) = lim s = [mm] = 160 [ µm]
s→ 0 s→ 0 1 + 60s s 61

Thus, although the closed loop system improves on the plant dynamics, there is a
steady discrepancy between the specified setpoint, R(t), and the actual steel
thickness, Y(t). This error is a linear function of the setpoint, as shown:

Figure 5.3 Closed Loop Steady State Error

Such steady state error is undesirable and can be reduced by increasing the open
loop gain of q(s) at low frequencies (i.e. where s → 0).

An integration circuit would eliminate the error completely at steady state, giving
steel of the exact thickness required. The reason for this is simply that an integrator
in k(s) gives q(s) infinite gain at s = jω = j0 (i.e. at steady state). Intuitively, any non
zero error will change the process input, until the error becomes exactly zero.

For example, a PI (Proportional-plus-Integral) controller

1
k(s) = −6 (1 + ) [N/µm]
10s

introduces the required integration into the open loop transfer function. It gives the
open loop transfer function
60(1 + 10s) 1 6(1 + 10s)
q(s) = = [µm/µm]
10s(1 + 6s) s (1 + 6s)

and the error signal for the new design becomes:

1 s(1 + 6s)
e(s) = r(s) = r(s)
1 + q(s) s(1 + 6s) + 6(1 + 10s)

For a thickness setpoint of 10[mm], the steady state error attains the final value

s + 6s2 10
lim se(s) = lim s [mm]
s→ 0 s→ 0 6 + 61s + 6s2 s
64 Control Engineering - 1

=0

This example illustrates how a steady state error can exist in a closed loop system
with acceptable dynamics, unless the design of k(s) ensures that the error is exactly
zero.

5.1. System Type Numbers

The steady state error that arises from a constant setpoint is known as POSITION
ERROR. To ensure zero steady-state error for this class of inputs, one integrator is
required in the open loop transfer function q(s) and the resulting system is known as
a TYPE 1 SYSTEM.

For constant velocity inputs the setpoint is a ramp. Steady state error in this case is
known as VELOCITY ERROR and is zero only if two integrators are included in
q(s). The system is then known as a TYPE 2 SYSTEM.

For constant acceleration inputs the setpoint is a parabolic function of time, the
steady state error is known as ACCELERATION ERROR and is zero only if q(s)
contains three integrators. The system is then a TYPE 3 SYSTEM.

In general, a system q(s) that contains M integrators is known as a TYPE M

SYSTEM and has an open loop transfer function of the form

a 0 + a1s + a 2s2 +... +a ms m

q(s) = with m ≤ n
s (b 0 + b1s + b 2s2 +... +b n − Ms n − M )
M

Each of the three most important cases is now considered in some detail.

5.2. Steady State Position Error

The position error that exists in a closed loop system when the setpoint is a step
function is computed from the POSITION ERROR CONSTANT. It is defined as:
K p = lim q(s)
s→ 0

In the case of the rolling mill, the position error constant is computed from
60
K p = lim = 60 under P control
s→ 0 (1 + 6s)
and
 Ch 5: Steady State Behaviour and System Type Number 65

60(1 + 10s)
K p = lim =∞ under PI control
s→ 0 10 s(1 + 6s)

The Position Error is computed directly from the position error constant
1 1 1 1
e p = lim s = lim =
s→ 0 1 + q(s) s s→ 0 1 + q(s) 1 + K p

For the rolling mill it is

1
ep = = 16 [ µm] under P control
1 + 60
and
1
ep = = 0 [ µm] under PI control
1+ ∞

Note that the results are derived for a unit step change of 1[mm] in the setpoint and
must be scaled for a 10[mm] change.

5.3. Steady State Velocity Error

In more sophisticated control systems, commonly found in robotics and aerospace
applications, the steady state error in tracking a setpoint of constant velocity
becomes an important feature of the closed loop system. This error is known as
VELOCITY ERROR and is computed from the VELOCITY ERROR CONSTANT:
K v = lim sq(s)
s→ 0

The Velocity Error is then simply

1
ev =
Kv

which is derived using the Final Value Theorem on e(s) when r(s)=1/s2 (a ramp).

5.4. Steady State Acceleration Error

In high-performance target-tracking control systems, such as radar scanners for air-
defence, the steady state error in tracking a setpoint of constant acceleration is
important. This error is known as the ACCELERATION ERROR and is computed
from the ACCELERATION ERROR CONSTANT:
66 Control Engineering - 1

K a = lim s2 q(s)
s→ 0

The Acceleration Error is then simply

1
ea =
Ka

5.5. Summary
An open loop system of the form

a 0 + a1s + a 2s2 +... +a ms m

q(s) =
sM (b 0 + b1s + b 2s2 +... +b n − Ms n − M )

where constants a0 and b0 are non zero, is known as a TYPE M SYSTEM since it
contains M integrators. The error at steady state for such systems is dependent on:

(1) the class of setpoint applied to the closed loop system and

(2) the system Type Number.

The effects of type number on final error are tabulated for the most important cases.

Error
System Position, ep Velocity, ev Acceleration, ea
Type 0 Finite Infinite Infinite
Type 1 Zero Finite Infinite
Type 2 Zero Zero Finite
Type 3 Zero Zero Zero

Example 1

Consider the performance of a Type 0 system in closed loop:

 Ch 5: Steady State Behaviour and System Type Number 67

Figure 5.4 A Type 0 Loop

Its open loop transfer function is

10
q(s) =
1+ s

Thus the steady state position, velocity and acceleration constants are
Kp = 10 Kv = 0 Ka = 0

The response of the closed loop system to unit step and unit ramp changes to the
setpoint are:

Figure 5.5 Step and Ramp Responses

Clearly the position, velocity and acceleration errors in this example are
1 1 1 1
ep = = ev = =∞ ea = =∞
1 + K p 11 Kv Ka

These results can be checked by appropriate analyses of the transfer function

e(s) 1 1+ s
= =
r(s) 1 + q(s) 11 + s

with a setpoint, r(s), that is a Step, then a Ramp, and finally a Parabola.
68 Control Engineering - 1

Example 2

Consider the performance of a Type 1 system in closed loop:

Figure 5.6 A Type 1 System

Here the open loop transfer function is

10
q(s) =
s

Thus the steady state position, velocity and acceleration constants are
Kp = ∞ Kv = 10 Ka = 0

And its responses to step and ramp changes in the setpoint are:

Figure 5.7 Step and Ramp Responses

Clearly the position, velocity and acceleration errors are

1 1 1
ep = =0 ev = = 0. 1 ea = =∞
1 + Kp Kv Ka

It should be noted that the velocity error is a constant error in position between the
actual output, y(t), and its setpoint, r(t).

Once again the results can be verified from the error transfer function
e(s) 1 s
= =
r(s) 1 + q(s) 10 + s
 Ch 5: Steady State Behaviour and System Type Number 69

Example 3 Phase Lock Loop

The majority of all Phase Lock Loop (PLL) design problems are amenable to
transfer function analysis. In control terminology the typical phase lock loop:

Figure 5.8 Schematic of a Phase-Lock Loop

can be represented by the block diagram:

Figure 5.9 Phase Lock Loop Block Diagram

The control block k(s) fits into the continuous signal that drives the VCO (voltage
controlled oscillator). The form taken by the transfer function fq(s) is typically
K k
Type 1 fq = k(s) =
s(s + a) s+a
K(s + a ) k(s + a)
Type 2 fq = k(s) =
s2 s
K(s + a )(s + b) k(s + a)(s + b)
Type 3 fq = k(s) =
s3 s2

Clearly k(s) provides the required type number, stabilises the loop and compensates
the system to give acceptable dynamic response times.

5.6. Controller Design Using the Error Constants

In specifications for the performance of a closed loop system, it is usual to include
70 Control Engineering - 1

some reference to its ability to track setpoints of a particular type with either zero
error, or alternatively, some small predefined steady state error. By computing the
relevant error constant it is possible to convert this requirement to a minimum type
number and a minimum gain for the open loop system.

Example

Consider the typical control loop configuration:

Figure 5.10 Typical Closed Loop Control Configuration

where the open loop transfer function is

q(s) = g(s) k(s) = gk(s)

To track a unit parabolic setpoint, r(t), with a steady state error of less than five
[units] error between the setpoint and output, y(t), implies that
ea ≤ 5 [units] when r(t) = 0.5 t2

In turn this requires that the acceleration error constant be

1 1
Ka = (by definition, ea = )
ea Ka

≥ 0.2 (since ea ≤ 5[units])

Thus, from the theory of error constants, the original specification implies that

K a = lim s2 q(s) ≥ 0. 2
s→ 0

which puts a minimum requirement on:

(1) the system type number (Needs M ≥ 2) and

(2) the gain of the open loop transfer function q(s), if M = 2.

 Ch 5: Steady State Behaviour and System Type Number 71

5.7. The Internal Model Principle

The discussion of System Type Numbers is a special case of the INTERNAL
MODEL PRINCIPLE, which states that a closed loop control system will track a
setpoint r(s) with zero error if the loop transfer function, q(s), contains a term equal
to r(s). In many instances this translates to a constraint on the loop controller k(s).

Examples (Assume g(s) is Type 0)

1 1
For Type 1 the setpoint r(s) = and the appropriate controller is k(s) =
s s

1 1
For Type 2 the setpoint r(s) = and the appropriate controller is k(s) =
s2 s2

1 1
For Type 3 the setpoint r(s) = 3
and the appropriate controller is k(s) =
s s3

Conclusion

Extrapolating from these results, it seems likely that the controller form
1
k(s) =
s+a

is best for tracking exponential setpoints with decay rate a, while

1
k(s) =
s2 + ω 2

is best for tracking sinusoidal setpoints of frequency ω.

Speculation

Perhaps the best feedback controller would simply be

k(s) = r(s) giving u(s) = r2(s) - r(s) y(s)
 6
Prediction of System
Response
Asymptotic setpoint tracking is extremely important for closed loop control, but is
only the first step in the successful design of a feedback controller. It is also
necessary to determine the manner in which the system output moves from one
value to another in response to its setpoint. (In fact the Final Value Theorem on
which steady state error analysis is based is only valid for stable systems.)

The response, y(t), of a particular system to a given input, u(t), or a setpoint, r(t), is
readily determined by Convolution or Laplace methods. However it is generally
quicker and easier to predict its performance directly from the relevant transfer
functions. This requires an in-depth understanding of the s-plane and is essential for
the design of successful control schemes.

6.1. Basic Equations

In a transfer function model, g(s), for a system (that represents either an open loop
or a closed loop configuration)

Figure 6.1 A Control System

the input or setpoint, u(s), and output, y(s), are related by the Laplace equation
y(s) = g(s) u (s)

a 0 + a1s + a 2s2 +... +a m sm

= u(s)
b 0 + b1s + b 2s2 +... +b ns n
 Ch 6: Prediction of System Response 73

N(s) N u (s) N y (s)

= =
D(s) D u (s) D y (s)

where the transfer function, g(s), the input, u(s), and the output, y(s), are given as
rational functions in the complex Laplace variable, s.

Specifically
N(s) N u (s) N y (s)
g(s) = u(s) = and y(s) =
D(s) D u (s) D y (s)

where N(s), D(s), Nu(s), Du(s), Ny(s) and Dy(s) are polynomials in s.

6.2. Calculation of System Responses

To invert the expression for y(s), and give the time domain function y(t), the
polynomial ratio Ny(s)/Dy(s) is first expanded in partial fractions, each term
containing a first or second order denominator polynomial. These individual terms
are then inverted separately using standard tables of Laplace transforms.

Expansion of y(s) into partial fractions requires evaluation of the roots of equation
Dy(s) = 0 ... (6.1)

in the s-plane.

Depending on the polynomial coefficients, these roots can be real, complex or both.
For roots that are simple (i.e. distinct) the partial fraction expansion is easy to
perform. An added complication arises when roots are repeated.

Example

To illustrate the approach, consider the transfer function

6 6
g(s) = =
2
20 + 14s + 4s + s 3 (2 + s)(1 + 3j + s)(1 − 3j + s)

that has been subjected to a step input function

3
u(s) =
s
Its response, output y(s), is thus defined by the Laplace transform
74 Control Engineering - 1

y(s) = g(s) u (s)

6 3
=
(2 + s)(1 + 3 j + s)(1 − 3j + s) s
18
= ... (6.2)
20 s + 14 s2 + 4s3 + s 4

Computation of the system response, output y(t), starts with the expansion of
Equation (6.2) into partial fractions. The four roots of Equation (6.1) for this
denominator are
r1 = 0 , r2 = −2 , r3 = −1 + 3 j and r4 = −1 − 3 j

The roots are simple. Two are real and two are complex. The latter form a complex
conjugate pair. Thus the partial fraction expansion of y(s) is
k1 k2 k3 k4
y(s) = + + +
s (2 + s) (1 + 3j + s) (1 − 3j + s)

where ki are constants, known as RESIDUALS and computed from the limits
18
k1 = lim sy (s) = lim = 0. 9
s→ 0 s→ 0 (2 + s)(1 + 3j + s)(1 − 3 j + s)

18
k 2 = lim (s + 2)y(s) = lim = −0 . 9
s→−2 s→−2 s(1 + 3j + s)(1 − 3 j + s)
18
k3 = lim (s + 1 + 3j)y(s) = lim = −0 . 3 j
s→−1− 3j s→−1− 3j s(2 + s)(1 − 3j + s)

18
k4 = lim (s + 1 − 3j)y(s) = lim = 0. 3 j
s→−1+ 3j s→−1+ 3j s(2 + s)(1 + 3j + s)

(The residuals from complex conjugate pairs are themselves complex conjugates.)
Hence the expanded Laplace transform for the output is
0. 9 −0. 9 −0.3 j 0.3 j
y(s) = + + +
s (2 + s) (1 + 3j + s) (1 − 3j + s)

Finally its response is easily found from Laplace transform tables to be

y(t) = 0. 9 − 0. 9 e −2t − 0 . 3 je − t e − j3t + 0 . 3 je − t e j3t ... (6.3)

= 0. 9 − 0. 9 e −2t − 0 . 6 e − t sin(3t)
 Ch 6: Prediction of System Response 75

Note

This numeric example shows that:

(1) each root of Equation (6.1) leads to a corresponding exponential term

in y(t) as indicated in Equation (6.3),

(2) complex roots arise in conjugate pairs and produce oscillatory

behaviour of frequency given by the imaginary part of the root,

(3) the real part of real and complex roots sets its transient decay rate,

(4) if the real part of any root is positive (>0) then its corresponding term
in the response increases exponentially with time and the overall
response of the system is unstable,

(5) the final response contains components that can be traced back to
either the transfer function g(s) or its input u(s).

6.3. Prediction of System Behaviour

Three cases need to be considered to provide the necessary formulae for inverting
y(s) for the most common transfer function models. These are Simple Real Roots,
Simple Complex Conjugate Pair of Roots and Repeated Real Roots.

6.3.1. Simple Real Roots

Assume that Equation 6.1 yields η simple real roots ri so that

η
D y (s) = (s − r1 )(s − r2 )...(s − rη ) = ∏ (s − ri )
i=1

The expansion of y(s) into partial fractions gives

kη η
k1 k2 ki
y(s) = + +... + =å ... (6.4)
(s − r1 ) (s − r2 ) (s − rη ) i=1 (s − ri )

The η RESIDUALS, ki, are easily computed from the limits

k i = lim [(s − ri ) y(s)]
s→ ri
76 Control Engineering - 1

Using tables of Laplace transforms, Equation (6.4) is readily inverted to yield

η
r t
y(t) = k1e r1t + k 2 e r2 t +... +k ηe η = å k ie ri t
i=1

Observations

(1) Each root of Dy(s) = D(s) Du(s) = 0 yields an exponential term, ki

exp(rit), in the time function y(t).

(2) Each root, ri, must be zero or negative for y(t) to be a constant or a
decaying function of time respectively (i.e. ri≤0).

(3) If any root ri is positive (i.e. ri>0) then that term, and hence y(t),
increases without bound.

(4) The roots arising from the D(s) factor in Dy(s) are of greatest interest,
since the control engineer can design these to be in optimal positions,
while the roots of the Du(s) factor are caused by the input u(s) and are
usually unpredictable.

6.3.2. Simple Complex Conjugate Pair of Roots

For real systems, complex roots of Dy(s)=0 ALWAYS occur in pairs that are
complex conjugates. In essence these roots are handled in the same way as real
roots. (The only difference is that the roots and their associated residuals are
complex numbers as opposed to real numbers.)

Consider a response containing one complex conjugate pair of roots so that

η
D y (s) = (s − rc )(s − rc* ) ∏ (s − ri )
i=3

where root rc is the complex root and rc* is its complex conjugate. Then
η
kc k *c ki
y(s) = + +å
(s − rc ) (s − rc ) i=3 (s − ri )
*

The residuals kc and kc* are complex conjugate constants computed from
 Ch 6: Prediction of System Response 77

k c = lim [(s − rc ) y (s)] = ρ cos ( φ ) + jρ sin( φ )

s→ rc

where ρ = k c and φ = ∠k c give kc in the polar form.

Inverse Laplace transformation yields

* η
y(t) = k c e rc t + k *c e rc t + å k i e ri t
i=3
η
= Aeat cos (bt − φ ) + å k i e ri t
i=1

where the complex root rc = a + jb and the amplitude A = 2ρ.

Observation

(1) The complex conjugate pair of roots introduces an oscillatory term

into y(t).

(2) Its frequency of oscillation is given by the imaginary part, 'b', of the
complex root, rc.

(3) This oscillation decays at a rate determined by the real part, 'a', of the
complex root, rc, and 'a' must be zero or negative for a bounded
response in y(t) (i.e. a≤0).

6.3.3. Repeated Roots (For completeness only)

Assume that Dy(s) has one root repeated q times

D y (s) = (s − r)q (s − rq+1 )(s − rq+2 )...(s − rη )

then by partial fraction expansion

k q −1 kq η
k1 k2 ki
y(s) = + +... + −
+ + å
(s − r) (s − r) 2
(s − r) q 1
(s − r ) q
i=q+1 − ri )
(s

The Residuals for the repeated root are computed by a series of differentiations and
limits:
78 Control Engineering - 1

1 dµ
k q − µ = lim [(s − r )q y(s)] for µ=0, 1, 2, ... ,q-2
s→ r µ! d s µ

The inverse transform is then

k 3t 2 k q −1t q − 2 k q t q −1 rt η
y(t) = [k1 + k 2 t + +... + + ]e + å k i e ri t
2! (q − 2)! (q − 1)! i=q+1

Observation

(1) The repeated root must be negative only (i.e. r<0) for a bounded
response.

(2) All the other roots must be zero or negative to ensure a bounded
response in y(t), as before (i.e. ri≤0).

6.4. Summary
The time-domain response y(t) of a system where y(s) is defined by
N y (s) N(s) N u (s)
y(s) = =
D y (s) D(s) D u (s)

is dependent on the positions that roots of the equation

Dy(s) =D(s) Du(s) = 0

take in the s-plane.

This HIGHLY SIGNIFICANT RESULT is central to ALL linear methods of

analysis and design used in control engineering. The importance of the s-plane in
design is simply that with experience, it is possible to use these roots to predict the
time domain performance of a system WITHOUT having to invert the expressions
obtained for y(s)  a skill that saves considerable computational effort.

With the insight that the Laplace variable gives into the behaviour of differential
equations, it is possible to modify the positions of some roots of Dy(s), by altering
D(s), until the final system performs as required by its specifications.
 Ch 6: Prediction of System Response 79

6.5. Two Generic Examples

Since the response y(t) of a general set of differential equations can be split into the
sum of simpler first and second order differential equations, it is important to know
exactly how such elemental differential equations behave.

6.5.1. A Single Real Root

Consider the case where y(s) is given by a function containing a single real root, r,
1
y(s) =
s−r

The correspondence between the position of this root in the s-plane and the time
domain response of y(t) is illustrated by graphs showing both root position and
system impulse response.

Note that the response, y(t), is STABLE (i.e. bounded) when r≤0 and UNSTABLE
(i.e. unbounded) when r>0.

Root-position Impulse response

Figure 6.2 Single Stable Root, r<0

Root-position Impulse response

Figure 6.3 Single Integrating Root, r=0

80 Control Engineering - 1

Root-position Impulse response

Figure 6.4 Single Unstable Root, r>0

Note

The slope of the curve <ln(y)> versus <t> gives the value of 'r'.

6.5.2. A Complex Conjugate Root Pair

Consider the response containing a complex conjugate pair of roots, (a±jb)

1 1
y(s) = =
(s − a − jb)(s − a + jb) (s − a ± jb)

The correspondence between positions of these roots in the s-plane and the time
domain response of y(t) is illustrated by graphs showing the root positions and
corresponding system impulse response.

Note that the response, y(t), is STABLE (i.e. bounded) when a≤0 and UNSTABLE
(i.e. unbounded) when a>0. Also the frequency at which the response oscillates is
given by the value of 'b' (often in [radians/second])

Figure 6.5 Stable Complex Root Pair, a<0

 Ch 6: Prediction of System Response 81

Figure 6.6 Oscillatory Complex Pair, a=0

Figure 6.7 Unstable Complex Pair, a>0

Notes

(1) The period of these oscillations gives 2π/b and hence 'b'.

(2) A graph of the envelope <ln(y)> versus <t> gives 'a'.

6.6. Characteristic Function and Poles

In the expression for y(s), the roots of Dy(s)=0 can be separated into those arising
from D(s), the denominator of the system transfer function g(s), and those due to
Du(s), the input function u(s).

Control engineers design the system transfer function, g(s), to meet specifications
set for the system, but have little or no control over the input function u(s). Thus the
polynomial D(s) associated with the system transfer function, g(s), is of greatest
interest in the analysis and design of control systems. Its importance is reflected in
its name, since D(s) is known as the CHARACTERISTIC FUNCTION of the
system.
82 Control Engineering - 1

The polynomial equation

D(s) = 0

is known as the CHARACTERISTIC EQUATION and the positions of its roots in

the s-plane define a basic property of the system, namely, its performance in terms
of stability and transient response.

The roots of the Characteristic Equation are known as THE POLES of the system
and form the basis for linear control engineering theory.

Roots of the equation Du(s)=0 on the other hand merely characterise the system
input, u(s), and have little bearing on system performance (i.e. stability and transient
response). These roots influence the choice of system type number and are important
in some applications but the designer has no direct control over them and usually
concentrates on D(s).

6.7. System Modes

The various elementary components (single-pole or single complex pole-pair) of the
response y(t) due to the system poles, define its MODES.

When dealing with a system, it is convenient to visualise its behaviour as a sum of

its modes. Thus (conceptually) any transfer function, g(s), can be expanded by
partial fractions to an equivalent system comprising a parallel combination of sub-
blocks, each block containing a mode of the original system.

Figure 6.8 Modal Decomposition

Each mode, gi(s), is defined by either a single pole or a single pair of complex
conjugate poles. Analysis of these modes gives invaluable insight into the
performance of the original system.

Example
 Ch 6: Prediction of System Response 83

The system transfer function

1− s
g(s) =
2 + 3s + s2

can be split into the two parallel systems

2 −3
g(s) = +
1+ s 2+s

These two single-pole sub-systems are clearly stable and so g(s) itself must be
stable. Further deductions can be made, if required.

6.8. Detailed Analysis of Two Elementary Systems

In industry, the dynamic performance of most systems can be approximated by a
dominant first order or a dominant second order differential equation. More
complex systems, where this dominant behaviour is not valid, can be thought of as
parallel models comprising first and second order sub-systems. For effective
application of control engineering theory to practical situations, it is important to be
familiar with the performance of these two basic modes.

6.8.1. First Order System

A
Its transfer function is g(s) =
1 + sT

where A is the gain

T is the time constant
(1+sT) is the Characteristic function

The system pole is at s = −1 / T

When perturbed by an input the resulting transients in the system response decay
away within a time period determined by T. Specifically its responses to Step and
Impulse perturbations of the input u(t) are:
84 Control Engineering - 1

B
Step Response, u(s) = Impulse Response, u(s) = B
s

BA − t / T
y(t) = BA (1 − e − t / T ) y(t) = e
T

Figure 6.9 First Order Responses

Final value y(∞)=BA Final value y(∞) = 0

Initial value y(0)=0 Initial value y(0)=BA/T

Within ±5% band of its Within band about zero of ±5%

final value for t>3T. of initial value for t>3T.
Within ±2% band of its Within band about zero of ±2%
final value for t>4T. of initial value for t>4T.

Practical Application

This information is useful in a number of practical situations. For example, should a

plant respond in the exponential manner shown above when its input is stepped,
then its dynamics can be approximated by a first order model. Also the gain of its
transfer function, g(s), can be deduced from the final value that the output attained.
In addition the time constant can be computed by measuring the time that the system
took to reach 63,21% of its final value (Remember that y(t) must be zero for all t<0).

Another example of its use is in the specification of pole positions in the s-plane.
Should the system under consideration be required to settle to within the ±2% band
in less than 10 seconds, then this criterion translates to the equivalent requirement
that the system have a pole that lies to the left of s = −0 . 4 in the s-plane.
 Ch 6: Prediction of System Response 85

6.8.2. Second Order System

Aω 2n
The transfer function is g(s) =
s + 2 ζω ns + ω 2n
2

where A is the gain

ωn is the natural frequency of oscillation
ζ (Zeta) is the damping factor
(s2 + 2 ζω ns + ω 2n ) is the Characteristic function

The system poles are at s = [ − ζ ± ζ 2 − 1 ] ω n = a ± jb and are

Real if ζ ≥ 1 (real pole case)

Complex if ζ < 1

The pole positions for a second order system with ζ<1 lie along the following lines
in the s-plane

Figure 6.10 Pole Positions in the s-plane

86 Control Engineering - 1

Note

When the damping factor ζ=0.7071 the complex poles lie at 45° to the σ-axes and
the system gives a good response that is Fast with Some overshoot but No
oscillation. This is generally considered to be ideal for mechanical systems where
excessive oscillations lead to accelerated wear-and-tear or metal fatigue, but for
chemical systems a lower damping factor (ζ=0.3) can often be tolerated and gives a
faster response at the cost of more oscillation.

The response of a second order system (with A=1 and ζ<1) to Unit Step and
Impulse perturbations of the input u(t) are

1
Unit Step Response, u (s) = Unit Impulse Response, u (s) = 1
s

ωn σ t ω 2n σ t
y(t) = 1 − e sin ( ω d t + φ ) y(t) = e sin( ω d t)
ωd ωd

where ω d = ω n 1 − ζ2 = b, imaginary part of the root (> 0)

σ = − ω nζ = a, real part of root ( ≤ 0)
−1 b
φ = cos ( ζ ) = − tan ( )
a

Final value y(∞)=1 Final value y(∞)=0

Initial value y(0)=0 Initial value y(0)=0

Within ±5% band for t>3/ζωn.

Within ±2% band for t>4/ζωn.

Practical Application

Once again this information can be used to estimate a model, g(s), for a plant that
responds to a step input in an oscillatory manner. Also the time domain
specifications of, say, no oscillation and a settling time of 12 [s] to the 5% band will
impose the restriction on the system poles that

ζ > 0.7071
a < −0. 25
 Ch 6: Prediction of System Response 87

6.9. Guide to the s-Plane

In general it is possible to make some comments about a system with a transfer
function, g(s), by studying the positions that its poles take in the s-plane. The
significant areas and trends are summarised graphically as:

Figure 6.11 Guide to the s-plane

For any given design it is possible to define an area in the s-plane within which the
poles of the final system must lie in order to meet required specifications.

As an example, poles positioned within the shaded region above indicate a system
that is slightly under damped with ζ>0.7071 and has transients that decay to within
the ±2% band in less than 5 [s].

6.10. Pole Dominance

In an actual engineering problem the control system under consideration is likely to
contain many (hundreds, even thousands) of sub-systems. If an exact scientific
88 Control Engineering - 1

analysis were undertaken, the problem could very easily become extremely complex
and completely insoluble. Thus it is important to exercise some engineering
judgement when dealing with real-world problems. In control systems this usually
translates to knowing which sub-systems need to be modelled as dynamic elements
(with poles) and which can be approximated by constants. The following example,
though simple, illustrates this point.

Example (Control of reactor temperature)

Discuss the performance of a system for regulating temperature in a nuclear reactor

by manipulating its control rods:

Figure 6.12 Controlling a Reactor Pile

The complicated relationship that exists between the position of the control rods and
the temperature of the reactor is approximated by the transfer function
y(s)
g(s) =
u(s)
5
=
(s + 10)(s + 0.1)(s2 + 0. 2s + 1)

The system model has three modes defined by pole positions at

s = −10 ascribed to the temperature sensor,

s = −0.1 the motor and
s = −0.1 ± j0. 99 the hoist system.

On the s-plane, these poles lie in the following positions:

 Ch 6: Prediction of System Response 89

Figure 6.13 Reactor Pile Pole Positions

And the positions of the poles suggest that

(1) Pole No 1 is so fast that it can be approximated by its steady state

gain. (The temperature sensor can be approximated by a gain),

(2) Pole pair No 3 introduces oscillatory responses with frequency of

approximately 1 [rad/sec],

(3) Pole No 2 is closer to the origin than pole pair No 3 and its
characteristic response will dominate the system response to step,
ramp and parabolic inputs functions.

Such observations and predictions are easily made by inspection of system pole
positions in the s-plane. The conclusions are confirmed by comparing step responses
of the original system
5
g(s) =
(s + 10)(s + 0.1)(s2 + 0. 2s + 1)

and it approximation
5
g(s) =
10(s + 0.1)

where the temperature sensor and the hoist sub-systems have been replaced by two
constants.
90 Control Engineering - 1

Figure 6.14 Step Responses

Clearly for step inputs, u(s) = a/s, the system response is dominated by poles that lie
closest to the origin. In general, inputs excite those modes of the system that lie
closest to the poles of the input function (i.e. to the roots of Du(s)=0).

For this example, should the input to the above system be a sinusoidal function with
pole-positions close to Pole pair No 3 then the system response would be dominated
by a sinusoid.

6.11. The Effect of System Zeros

Transfer functions, g(s), are defined by the ratio of two co-prime polynomials
N(s)
g(s) =
D(s)

a 0 + a1s + a 2s2 +... +a ms m

= with m ≤ n.
ba 0 + b1s + b 2s2 +... +b ns n

As mentioned previously roots of the characteristic equation

D(s) = 0

are known as the system poles. The positions that these poles take in the s-plane
determine the performance of the system (i.e. whether it is stable or unstable,
oscillatory or damped and fast or slow)

Roots of the equation

N(s) = 0

are known as the SYSTEM ZEROS. The positions of zeros in the s-plane are also
significant, though not as characteristic as the pole positions. Zeros in fact tend to
cancel the influence of poles.
 Ch 6: Prediction of System Response 91

To illustrate this effect, consider the simple system

1 s+a
g(s) =
a s +1

which has one pole at s = −1 and one zero at s = −a . In the s-plane, the pole and
zero are represented by:

Figure 6.15 Pole-Zero in the s-Plane

The time response of this system to a step input, for different zero positions is:

Figure 6.16 Pole-Zero Step Responses

Algebraically the response is given by the Laplace transform

1 1 s+a
y(s) = F.V: y(∞)=1 I.V: y(0)=1/a
s a s +1
92 Control Engineering - 1

Note

(1) As the zero approaches the pole position from the left, it tends to
cancel the effect of the pole. In the limit, when a=1, the transfer
function becomes g(s) = 1. The zero then cancels the pole exactly and
the system has no dynamics.

(2) For step, ramp and parabolic inputs the actual response is dominated
by the pole or the zero, depending on which is nearest the origin.

(3) When the zero moves into the right-half s-plane the initial jump in the
step response is negative. System stability is determined only by the
pole position, so its response remains stable for all 'a' values.

An intuitively appealing way of viewing the effect of zeros is to expand the original
system model by partial fractions to give an equivalent parallel model:

Figure 6.17 System Decomposition into Parallel Sub-Systems

Thus
g(s) = g1 (s) + g 2 (s)+... +g n (s)

where each sub-system gi(s) in the parallel model consists of a single real pole or a
single pair of complex conjugate poles. The system response is
y(s) = y1 (s) + y 2 (s)+... +y n (s)

Each sub-system contributes to the overall response in direct proportion to the size
of its gain relative to the other sub-systems. And these gains are determined by the
positions of the system zeros.
 Ch 6: Prediction of System Response 93

Example

Consider the simple system

1 s+a
g(s) =
a s +1

or, in block diagram form:

Figure 6.18 Pole-Zero Sub-Systems

Clearly the position of the zero affects the gain of the lower block and hence the size
of its contribution to the output y(t). The mathematical reasons for this are now
discussed.

6.12. Graphical Evaluation of Residuals

For system transfer functions with distinct poles (at s=pi), the expansion of g(s) into
partial fractions is given by the formula
m n
g(s) = γ ∏ (s − z i ) / ∏ (s − p i )
i=1 i=1

n k n
= å (s − ip ==> y(t) = å k ie p i t
i=1 i) i=1

where γ is a constant, zi are the system zeros and pi are the system poles.

The residuals kj are calculated from the limit

k j = lim [(s − p j )g(s)]
s→ p j

m n
= γ ∏ (p j − z i ) / ∏ (p j − pi )
i=1 i≠ j

In geometric terms this equation can be interpreted as:

94 Control Engineering - 1

Product of all vectors from all zeros to pole no j

kj = γ
Product of all vectors from all other poles to pole no j

In the s-plane these vectors for a simple system (in which complex conjugate pairs
have been omitted for clarity) are:

Figure 6.19 Residual Vectors

Note (Engineering interpretation of the geometry)

(1) The closer the zero moves to pole j the shorter vector a becomes and
hence the residual kj associated with pole j decreases.

(2) The closer the other poles move to pole j the shorter vector b becomes
and hence the associated residual kj increases.
 7
Routh-Hurwitz Stability
Criterion
Clearly it is important that the poles of a system lie in the left-half s-plane since this
ensures that it has a stable response. One method for determining how many
unstable poles a given system has is the ROUTH-HURWITZ CRITERION.

Consider a system defined by the transfer function

N(s)
g(s) =
D(s)

where N(s) and D(s) are co-prime polynomials in s. (Co-Prime means that all factors
common to N(s) and D(s) have been cancelled. It is an important constraint on the
polynomials of g(s), even though it is not often stated.) Once again g(s) can
represent either an open loop or a closed loop configuration as required.

Stability of the system is determined by its pole positions in the s-plane (i.e. by the
location of the roots of its characteristic equation

D(s) = (b o + b1s + b 2s2 +... +b ns n ) = 0

From the theory of polynomials, inspection of the coefficients, bi, of D(s) can
indicate stability of transfer function g(s) since:

Rule No 1: If any coefficient bi is negative then g(s) has unstable

poles.

Rule No 2: If any coefficient bi, except bo, is zero then g(s) has
unstable poles or oscillatory poles.

Further information about the poles of g(s) can be obtained by constructing and
analysing the Routh array:
96 Control Engineering - 1

Power of s Row No Routh Array

sn 1 bn bn-2 bn-4 ...
sn-1 2 bn-1 bn-3 bn-5 ...
sn-2 3 cn-1 cn-3 cn-5 ...
sn-3 4 dn-1 dn-3 dn-5 ...
::: ::: ::: ::: ::: ...
s0 n+1 rn-1 ...

where the first two rows contain coefficients of the original polynomial D(s) and the
subsequent rows (3 to n+1) are computed by the determinants

1 bn b n − k −1
cn − k = −
b n −1 b n −1 b n − k − 2

1 b n −1 b n − k − 2
dn−k = − etcetera.
c n −1 c n −1 c n − k − 2

Two special cases can arise in computing the Routh array, namely:

(1) if a zero appears in the first column and there are non-zero elements in
that row then replace the zero with an algebraic variable, ε (>0), and
continue;

(2) if a complete row of zeros appears then this row must be replaced by
the derivative of the previous row (as shown in the examples).

The completed table is then analysed according to the Routh criterion which states
that:

Rule No 3: The system, g(s), is stable iff all elements in the first
column are strictly positive (>0).

Rule No 4: The number of sign changes in the first column is

equal to the number of unstable poles.

Example 1

Determine stability of an open loop system in which the characteristic function is

D(s) = 6s 4 + 3s3 + 2s2 + s + 7 (n=4 Þ 5 rows in its array)

Initial conclusions
Ch 7: Routh-Hurwitz Stability Criterion 97

(1) No negative coefficients, bi, Þ No obvious unstable poles.

(2) No zero coefficients, bi, Þ No obvious unstable or

oscillatory poles.

The Routh table is

s4 1 6 2 7
s3 2 3 1 (0) ← Pad row with a zero
s2 3 ε 7 ← Replace 0 with ε
s 4 (ε-21)/ε (0) ← Pad row with a zero
1 5 7

The first column is thus [6 3 0 -∞ 7]T as ε → 0+ where 0+ means that ε

approaches zero from the positive side.

Conclusion. There are two sign changes so the system has two unstable poles. (This
is confirmed since D(s) has roots at s = −0.82 ± j0. 75 and s = 0 . 57 ± j0.80 )

Example 2

If all the elements in a row are zero, then the polynomial d(s) has an even factor with
coefficients given in the row above. The table may be continued by replacing the
row of zeros with the coefficients of the derivative of the even factor.

For example, consider

d(s) = s5 + 4s4 + 3s3 + 12s2 + 2s + 8 (n=5 Þ 6 rows)

The start of the Routh array gives

s5 1 1 3 2
s4 2 4 12 8
s3 3 0 0 ← Entire row of zeros

The row full of zeros implies an even factor in the previous row

a(s) = 4s 4 + 12s2 + 8

Replace the row of zeros with coefficients of the derivative of the even factor:
98 Control Engineering - 1

d
a(s) = 16s3 + 24s
ds

and complete the table

s5 1 1 3 2
s4 2 4 12 8
s3 3 16 24 (0) ← Pad with zero
s2 4 6 8
s 5 8/3 (0) ← Pad with zero
1 6 8

The first column is [1 4 16 6 8/3 8]T. No sign changes indicate that there are no
unstable poles in this characteristic function. (Its poles are at s = −4 , s = ± j1. 4 and
s = ± j1. 0 )

7.1. Application of Routh-Hurwitz Criterion

The preceeding examples have been numeric to illustrate the Routh-Hurwitz rules.
With modern computational tools, actual roots are easily computed in such cases.

In practice the R-H Array is more useful when it is applied to characteristic

functions that contain some unknown coefficients. For example, consider the
feedback control loop:

Figure 7.1 Closed Loop Control System

Assume that the controller constants, K and T, are set by potentiometers that plant
personnel can tune. Use the Routh array to find ranges of K and T that ensure a
stable system response.

The transfer function model, q(s), for the open loop is:
Ch 7: Routh-Hurwitz Stability Criterion 99

1 K(1 + sT) y(s)

q(s) = g(s) k(s) = =
s(s + 1) sT e(s)

so the transfer function model for the closed loop system is

q(s) K(1 + sT) y(s)
h(s) = = 2
=
1 + q(s) K(1 + sT) + s T(s + 1) r(s)

The characteristic function for h(s) is thus

Ts3 + Ts2 + KTs + K

By inspection of the coefficients, stability requires that

T>0 KT > 0 K>0

Also the Routh Array

s3 1 T KT
s2 2 T K
s 3 K(T-1) (0) ← Pad with zero
1 4 K

requires that T > 1 to avoid sign changes in the first column.

This illustrates how the Routh Table can be used to determine the range of
parameter values over which a system will remain stable. Such information is
conveniently presented in a DESIGN DIAGRAM that summarises the control
engineering conclusions in a general format that is readily understood by others.

Figure 7.2 Design Diagram for the PI Controller

100 Control Engineering - 1

7.2. Relative Stability from a Routh-Hurwitz Array

For engineering applications, stability is merely one of many criteria applied to a
design. Speed of response is another and the Routh Criterion can be applied to
determine whether system poles (i.e. roots of the Characteristic Equation) lie far
enough to the left in the s-plane to ensure a fast speed of response.

To illustrate this, consider a process with the characteristic function

d(s) = s3 + 10.1s2 + 21s + 2 (which is stable)

Use the modified Routh Array to determine whether the system will settle to within
±2% of its final value in less than 20 [s].

Start by defining a new variable, v, as

v = s+0.2

which shifts the Y-axis 0.2 units (from 4/20) to the left in the s-plane. Clearly
instability in the v-plane translates directly to a system that is too slow (or unstable)
in the s-plane. Graphically:

Figure 7.3 Relationship between the v and the s-plane

Substitution of s = v − 0. 2 into the characteristic function yields a new polynomial

d v (v) = (v − 0 . 2)3 + 10.1(v − 0 . 2)2 + 21(v − 0 . 2) + 2

= v3 + 9.5v 2 + 17. 2v − 1.80

Since one coefficient of this polynomial, dv(v), is negative there is at least one pole
that is too slow (i.e. to the right of the s = −0. 2 line)
 Ch 7: Routh-Hurwitz Stability Criterion
101

The modified Routh Array for the system is

v3 1 1 17.2
v2 2 9.5 -1.80
v 3 17.4 (0)
1 4 -1.80

There is one sign change in the first column so only a single root of the original
characteristic equation is too slow. (The roots of D(s), found by numeric methods,
are at s = −7. 2 , s = −2 . 8 and s = −0 .10 )

Once again the Routh-Hurwitz extension is probably most useful when the
coefficients of the original D(s) are unknowns. Unfortunately the resulting algebraic
manipulations tend to be extremely laborious.

7.2.1. Further extensions

The simple translation of the jω axis in the s-plane to form the v-plane can be
generalised using Mobius transformations. As an example, a circular region in the s-
plane that includes damping and transient decay rate is mapped to the left-half of a
v-plane by the transformation pair
γ +v β s + αβ − 1
s= −α and v=γ
β ( γ − v) β s + αβ + 1

Graphically the mapping is:

Figure 7.4 Relationship between the v and the s-plane

 8
Feedback Control
Systems
Control engineers often connect various sub-systems together to form one large
closed loop control configuration. The final feedback system has specified dynamic
behaviour and is insensitive to process changes, disturbances and measurement
noise. To predict its performance, adequate models of the system dynamics are
essential. Such models are usually linear, for ease of manipulation and diversity of
analytical tools, even though the real-world processes are generally non-linear and
time-varying. In practice the assumption of linearity is often justified.

So far transfer function models have been dealt with in some depth to illustrate:

• the Open loop and Closed loop configurations, with particular

emphasis on the advantages of the latter,
• various types of mathematical models for approximating system
dynamics,
• derivation of models for individual components within a large and
complex system,
• characterisation of system components as blocks defined by transfer
function models,
• schematic representation of large and complex systems in a block
diagram, and manipulation of such block diagrams to determine the
characteristics of a block for the overall system,
• correlation between pole-zero positions in the s-plane and system
response in the time domain,
• determination of the number of unstable roots in a polynomial
(particularly for poles, but also applicable to zeros).

The ultimate aim of control engineering design is to produce a final (feedback)

system that meets its specifications. Thorough investigation, and hence
Ch 8: Feedback Control Systems 103

understanding of the problem at hand, is essential in addressing a set of key issues.

For example, should a closed loop configuration be used or is open loop control
adequate? Should the loop be made particularly insensitive to process changes or is
it to be optimal for tracking a particular form of setpoint? Is measurement noise on
the output excessive? Are the limits on the process input adequate?

Experience dictates how many of these criteria are met. Common closed loop
configurations and design aims are described below to provide some guidance.

8.1. The Closed Loop Control System

As demonstrated previously, feedback or closed loop control can improve
dramatically on the performance of a given system. For example, the basic closed
loop configuration is the unity feedback loop:

Figure 8.1 Basic Closed Loop Control Configuration

in which the input, u, to a given plant is manipulated automatically in such a manner

as to ensure that its output, y, tracks the setpoint, r, accurately. The (transfer
function) model for the closed loop system is derived easily using block diagram
algebra, or by eliminating internal variables from its component models.

The specific equations relating to this configuration are:

y(s) = g(s) u(s) The plant model
e(s) = r(s) - y(s) The comparator model
u(s) = k(s) e(s) The controller model

Combining these equations to eliminate the internal variables, e(s) and u(s), gives
y(s) = g(s) k(s) e(s)
= g(s) k(s) [r(s) - y(s)]
104 Control Engineering - 1

and so the model for the closed loop system becomes

g(s) k(s)
y(s) = r(s)
1 + g(s) k(s)

For convenience the following transfer functions are defined:

q(s) = g(s) k(s) Open loop model: y(s)=q(s) e(s)

and
h(s) = [1 + q(s)]−1 q(s) Closed loop model: y(s) = h(s) r(s)

The rationale behind the closed loop configuration is simply that:

As k(s) increases in size, so does q(s) and

if q(s) >> 1 then h(s) → [q(s)]−1 q(s) = 1

Thus setpoints are tracked accurately whenever the open loop model, q(s), is large,
irrespective of the actual plant model, g(s). Pause and take careful note of the
design requirements that are contained in this closed loop analysis:

ENSURE THAT THE GAIN OF q(s) IS LARGE.

This is a clear goal that is elegant, practical and easy to strive for, provided it is
remembered.

Large open loop gains can usually be realised exactly at low frequencies but fall off
at higher frequencies. Obviously instability will limit the size that k(s) can attain in
practice, as will signal amplitudes, actuator saturation, noise levels, etc. (Large gains
also ensure that the closed loop is insensitive to disturbances and changes in process
behaviour.)

The role of the control engineer is to balance these factors to produce an acceptable
design for the given situation.
Ch 8: Feedback Control Systems 105

Example

A simple numeric case illustrates the practical importance of high gain in feedback.
Consider a non-linear, time-varying plant that has been modelled simply by a gain
g(s) = 5 (±4)

where the nominal model, 5, is roughly equal to its uncertainty, 4, and represents a
±80% variation in gain. A unity feedback configuration for the process, which
includes a controller, k(s), that is also merely a gain, ensures that variations in the
closed loop gain are reduced significantly as indicated:

k(s) h(s) ∆h(s) Variation

0.5 0.71 -0.38 to +0.10 -53% to +15%
1 0.83 -0.33 to +0.07 -40% to +8%
10 0.98 -0.07 to +0.01 -7% to +1%

Clearly the closed loop configuration itself ensures considerable improvement over
the open loop process, and the gain of h(s) becomes more consistent as loop gain is
increased. (This phenomena holds for more complex transfer function models, but is
less obvious. The manner in which loop gain is limited by instability and constraints
on the process input, does however require a more elaborate transfer function
approach, and is not pursued here.)

8.2. Non-Unity Feedback

In some cases, notably operational amplifiers, a closed loop system may have non-
unity feedback. The closed loop configuration is then:

Figure 7.2 Non-Unity Feedback Configuration

Here the model for the closed loop system is derived from the equations
106 Control Engineering - 1

y(s) = g(s) u(s) The plant

e(s) = r(s) - f(s) y(s) Feedback & the comparator
u(s) = k(s) e(s) The controller

And so, eliminating the internal variables u(s) and e(s), the equations become
q(s)
y(s) = r(s)
1 + q(s) f (s)
= h(s) r(s)
1
→ r(s) when qf(s) >> 1
f (s)

Note that feedback can be used to invert functions (e.g. Convert capacitors to
equivalent inductors, differentiators to integrators, multiplication to division, etc.).

Example

Consider a process that is modelled by a gain

g(s) = 2

When placed in a non-unity feedback configuration in which f(s) is a gain F and

k(s) is a gain K, the closed loop model, h(s), as a function of (K,F) becomes:

F
K 0.1 1 10
0.1 0.20 0.17 0.07
1 1.67 0.67 0.10
10 6.67 0.95 0.10
Table of h(s)

Observe that the closed loop model approaches 1/F as gkf(s)=qf(s) increases above
one. In practical applications, it is vital to determine where exactly in the loop the
gain is situated since this has far reaching implications for the system components.
The theory presented here merely indicates the necessary goals.
Ch 8: Feedback Control Systems 107

8.3. Closed Loop Performance

As with any dynamic system, the response of a closed loop system is dependent on
the input, r(s), and the system model, h(s). Thus the inherent characteristics of a
closed loop system are determined by the positions of the poles of its transfer
function, h(s), in the s-plane. If the combined transfer function
N(s) N q (s) N f (s)
q(s)f (s) = =
D(s) D q (s) D f (s)

is a ratio of two polynomials, N(s) and D(s), then the closed loop system h(s) has the
Characteristic equation
D(s) + N(s) = 0 ... (8.1)

N q (s)D f (s)
since h(s) = [1 + q(s)f (s)]−1 q(s) = .
D(s) + N(s)

The roots of Equation (8.1) define the closed loop pole positions and hence the
performance of the feedback system.

Note

(1) The three complex-variable equations

1 + f(s) q(s) = 0
and
1
+ q(s) = f (s) + q(s) = 0
f (s)
and
1
f (s) + 
= f (s) + q(s) =0
q(s)

all have roots in the s-plane in the same positions as the Characteristic
equation for the closed loop system. (This fact is exploited later.)

(2) For unity feedback systems, the closed loop zeros of h(s) are the same
as the open loop zeros of q(s). Thus unity feedback only alters pole
positions.

(3) In addition to modules k(s) and f(s), the designer may want to include
a pre-filter module, p(s), on the closed loop setpoint. This has no
108 Control Engineering - 1

severe design implications as p(s) is merely in open loop with h(s).

8.4. Closed Loop Design

To design a feedback control loop, the control engineer first derives a transfer
function model that adequately characterises the plant dynamics in open loop. This
model, g(s), is then used in one or more of the numerous design methods that are
presently available to produce a suitable compensation k(s) in the typical feedback
control loop:

Figure 8.3 Non-Unity Feedback Control System

The feedback element, f(s), is often designed to meet other non-control

requirements. For example, it might be a model that the designer wants to invert, or
it might simply filter out high-frequency measurement noise.

The design of k(s) involves the informed manipulation of differential equations and,
without suitable theoretical tools, presents a difficult, if not impossible problem to
be solved. Many design techniques have been developed over the years to deal with
various classes of design problems and the control engineer needs to know which of
these methods is most appropriate in any given situation, what its advantages and
disadvantages are, and how the methods can be mixed to deal successfully with
difficult practical problems.

In the following chapters two distinctly different approaches to the design of control
schemes are considered:

• a Pole-Zero method (Root Locus) and

• a number of equivalent Frequency domain methods (Nyquist, Bode,

Nichol and Inverse Nyquist).
 9
Root Locus Design
Method
Once it has been ascertained that a particular system needs to be controlled in a
feedback configuration, then the problem of designing a suitable compensator k(s)
arises. Module k(s) must ensure that the closed loop system, h(s), tracks its setpoint
adequately by setting the system Type Number, as discussed earlier. At the same
time the dynamics of the resulting h(s) must be damped and fast.

9.1. Root Locus Design Methods

Positions of system poles and zeros in the s-plane give a good indication of how that
system will perform in the time domain. Control engineers are particularly interested
in predicting the performance of closed loop systems, since these have distinct
practical advantages over open loop systems.

The ROOT LOCUS design method (proposed by WR.Evans in 1948) analyses

dynamic systems on the basis of their pole-zero positions in the s-plane. It provides
considerable guidance as to where the poles and zeros of the compensating element
k(s) should be placed for improving the dynamic behaviour of the overall system.

Consider the closed loop system:

Figure 9.1 Control Loop for Root Locus Analysis

The dynamic part, q*(s), of the open loop transfer function is written as
110 Control Engineering - 1

N* (s)
q * (s) =
D* (s)

s m + a m −1s m −1 +... +a1s + a 0

= with m ≤ n
sn + b n −1sn −1 +... +b1s + b 0

where the coefficients of the highest power of s in both numerator and denominator
polynomials are forced to unity. The ROOT LOCUS GAIN, γ, is a constant gain
factor that is taken to be positive. (Root loci can also be drawn for negative gains γ,
but these are not considered here.)

Note

(1) The open loop transfer function q(s)=g(s)k(s) models the cascade
connection of the plant model, g(s), and the controller model, k(s).

(2) The open loop transfer function, q*(s), used in the Root Locus is a
scaled version of the actual open loop transfer function, q(s).

(3) The Root Locus Gain is not actually a gain in the traditional sense,
though it is proportional to the open loop gain. It is very useful for
reading gain values directly from Root Locus diagrams.

The transfer function for the closed loop system is

γN* (s)
h(s) =
γN (s) + D* (s)
*

Hence the Characteristic Equation for the closed loop system is

γ N*(s) + D*(s) = 0 ... (9.1)

The roots of this equation give the positions in the s-plane of the closed loop poles.

Also the closed loop zeros of h(s) are exactly the same as the open loop zeros of
q(s). i.e. Unity closure of the feedback loop in single-variable systems only alters the
pole positions of the overall system.

Now, for any given value of the constant γ, the closed loop poles of h(s) can be
computed from its Characteristic Equation (Equation 9.1) and plotted in the s-plane.
As the root locus gain, γ, is changed from 0+ to +∞, these closed loop poles move
around in the s-plane, from the positions of the open loop poles to the positions of
the open loop zeros, thereby tracing out the ROOT LOCI. (Thus the root locus
Ch 9: Root Locus Design Method 111

diagram is a plot of closed pole positions, parametrised by γ.)

An Illustrative Example

Consider the second-order process model

2
q(s) =
(1 + 0. 2s)(1 + 0.1s)

which has a gain of 2 and two time constants 0.2 and 0.1 [seconds]. The open loop
transfer function q(s) has two poles, one at s = −5 and one at s = −10 . It also has no
zeros (which is equivalent mathematically to having zeros at infinity).

The scaled transfer function, q*(s) required for drawing the Root Locus is derived as
follows
100 1
q(s) = = 100 = 100 q* (s) = γ q * (s)
(5 + s)(10 + s) (s + 5)(s + 10)

In a unity feedback configuration this open loop model generates the closed loop
characteristic equation

s2 + 15s + (50 + γ ) = 0

For this simple example the closed loop poles can be computed analytically and are
15 1
s=− ± 152 − 4(50 + γ )
2 2

Clearly the closed loop pole positions are a non-linear function of the root locus
gain γ
s = s( γ )

As γ is varied, this function sketches the root loci in the s-plane for this process
model
112 Control Engineering - 1

Figure 9.2 Control Loop Pole Positions Parametrized by γ

Notice that the closed loop poles start at the positions of the open loop poles (i.e. at
s= −5 and s= −10 ) when γ=0. Also the closed loop poles move off to the zeros at
infinity as γ tends to infinity.

Knowing the significance of pole positions in the s-plane, the Root Locus, once
drawn, is studied to predict the dynamic performance of the closed loop system.

(1) The Closed Loop (C/L) is stable for all values of γ ≥ 0.

(2) Its response is dominated by a slow C/L pole when γ << 6.25.

(3) The C/L transients decay faster as γ is increased from 0.

(4) When γ = 6.25 the C/L system is critically damped.

(5) The fastest transient decay rate is achieved at γ=6.25 (The ±5% band
is entered in 0.40 [sec] and the ±2% band in 0.53 [sec]).

(6) The transient decay rate does not change for gains γ > 6.25.

(7) The response becomes more oscillatory as γ increases above 62.5.

These predictions are confirmed by its response to a step change in the setpoint for
different gains:
Ch 9: Root Locus Design Method 113

Figure 9.3 Confirmation of Predictions (γ=6.25, 100 and 200)

Note

As expected the error between the setpoint and output reduces when γ is increased.
Unfortunately its closed loop performance deteriorates since the controlled system
also becomes more oscillatory. The input also increases in amplitude.

9.2. Sketching Root Loci

Root Loci for realistic practical problems can become very complex indeed and for
any given design problem a number of possible pole-zero configurations need to be
considered to develop an effective compensator k(s). Typically the necessary loci
would be computed and plotted quickly using a digital computer with a graphics
facility. However it is possible (and instructive) to start any design by sketching the
loci roughly using a few simple rules of construction (These are valid for γ≥0.
Similar rules exist for γ≤0 but are not dealt with here). Such sketches give an
indication of how the control problem should be tackled.

(1) Symmetry

The root locus diagram is symmetric about the real axis. (Obviously)
114 Control Engineering - 1

(2) Origin

Each locus originates at an open loop pole when γ=0. Thus there are <n>
loci where 'n' is the order of the open loop system q(s).

(3) Terminus

When γ → ∞, <m> of the <n> loci terminate at the <m> zeros of q(s),
while <n-m> loci approach infinity along asymptotes.

(4) Asymptotes

The <n-m> asymptotes are at angles

(180 ± 360k)
where k is an integer.
(n − m)

These asymptotes intersect the real axis at

(Sum of Open Loop poles − Sum of zeros)

σc =
(n − m)

(5) Loci on the Real Axis

Complex poles and zeros of q(s) have no effect on the loci that follow the
real axis. Segments of the real axis are part of the root locus iff the total
number of real poles plus real zeros to their right is odd.

(6) Angle of Departure from O/L Poles

The locus leaves an open loop pole at an angle determined by a trial point,
close to the pole, satisfying the angle condition:

Sum of angles to O/L poles - Sum of angles to zeros = 180±360k

This rule is deduced from the fact that the root locus gain, γ, is a real
positive constant given by
Ch 9: Root Locus Design Method 115

D* (s)
γ =−
N* (s)

and that the angle, ϕ(γ), of the root locus gain, γ, must be zero. This
condition yields the equation

ϕ[ γ ] = 180 o + ϕ[D* (s)] − ϕ[N* (s)] = 0

(7) Angle of Arrival at Zeros

The locus approaches a zero at an angle determined by a trial point, close

to the zero, satisfying the angle condition given in rule 6 above.

(8) Break-Away and Break-In Points for Real Axis Loci

These are also found either (a) by using the angle condition (Rule 6) for a
trial point chosen close to the real axis, or (b) from the differential
d
γ (s) = 0
ds
so that
d d
N(s) D(s) − D(s) N(s) = 0
ds ds

Example

To illustrate how the rules for sketching Root Locus are used in practice, find the
value for the controller gain K that give pole positions which ensure a damping
factor of 0.7071 for the closed loop system:

Figure 9.4 Type 1 Control Scheme

For this system the open loop transfer function is:

116 Control Engineering - 1

2K
q(s) =
s(1 + 0. 2s)(1 + 0.1s)

The scaled transfer function needed for the root locus is given by the relationship
100K 1
q(s) = = 100K = γ q * (s)
s(s + 5)(s + 10) s(s + 5)(s + 10)

so the Root Locus Gain γ = 100K, where K is the controller gain.

Here
<n>=3 as there are 3 open loop poles ( s = 0 , s = −5 and s = −10 ).
<m>=0 as there are no zeros (or all zeros at infinity).
so
<n-m>=3.

The open loop poles, marked as X, are plotted on the s-plane. These give the starting
points on the root loci (where γ = 0). Any zeros, marked as O, are also plotted
and indicate the termination points (Where γ→+∞). In this example there are only
poles and the loci move along asymptotes to terminate at infinity.

The Rules that are activated to plot the root locus are:

Rule 2
There are <n> or 3 loci, starting at s = 0 , s = −5 and s = −10

Rule 3
The <n-m> or 3 loci approach infinity along 3 asymptotes

Rule 4
The asymptotes are angled at (180±360k)/3 = 60° ± 120k = ±60°, 180°
The intersection point σ c = {(0) + ( −5) + ( −10)} / 3 = −5
(These can now be drawn in the s-plane)

Rule 5
The loci on the real axis are simply drawn as shown:
Ch 9: Root Locus Design Method 117

Figure 9.5 Root Locus Sketch

Rule 8
The locus break-away point on the real axis is computed from a trial point
near the axis, as shown in the following sketch.

Figure 9.6 Geometry for Trial Point

Rule 6
The angle criterion for the trial point is
ϕ1 + ϕ 2 + ϕ 3 = π ± 2kπ

From the geometry of the trial point, the angle conditions translates to
δ δ δ
[ π − tan−1 ( )] + [ tan−1 ( )] + [ tan−1 ( )] = π ± 2 kπ
−σb σ b − [ −5] σ b − [ −10 ]

For small values of δ this equation reduces to

118 Control Engineering - 1

δ δ δ
π+( )+( )+( ) = π ± 2 kπ
σb σb + 5 σ b + 10

since tan−1 ( ϕ ) → ϕ for small angles.

And so
δ δ δ
( )+( )+( )=0
σb σb + 5 σ b + 10

which implies that

3 σ 2b + 30 σ b + 50 = 0
or
σ b = −2.11 (σ b = −7. 89 is discarded. It holds for γ ≤ 0)

The remaining complex portion of the Root Loci are now sketched (requiring a little
artistic talent) to give the complete root locus diagram for the system.

The closed loop pole positions (marked as small boxes) satisfy the specification of a
damping factor of 0.7071 for the controlled system h(s). The Root Locus Gain, γ, at
this point, s0, is computed from the magnitude relationship

D* (s0 )
γ=
N* (s0 )

n
∏ (s0 − pi )
= i=1
m
∏ (s0 − zi )
i=1

Product of distances from s0 to all poles

=
Product of distances from s0 to all zeros

= (2.8)(3. 6)(8. 3) = 83. 7

= 100 K (By definition of γ for this example)

And so the controller gain that gives a closed loop system with a damping factor of
0.7071 is:
Ch 9: Root Locus Design Method 119

83. 7
K= = 0.837
100

Note

From the Root Locus diagram for a closed loop system, it is possible to find values
of γ that ensure particular closed loop pole positions on the given loci.

The engineering problem is how to position poles and zeros of a controller, k(s), in
the s-plane to improve on the root loci (i.e. on the closed loop response).

9.3. Design Using the Root Locus

Design is something of an art, perfected by experience. However a few guidelines
ensure that each step in the design nudges the closed loop system towards its
required specifications.

To design a compensated closed loop system, start by plotting the root loci for the
plant transfer function (that includes a component to ensure the correct type
number). These loci show closed loop pole positions for the uncompensated system
and also present the problem in an appropriate form that gives engineering insight
into (a) the dynamic performance of the closed loop system and (b) how to alter the
open loop system to improve on its closed loop performance.

The following rules of thumb are useful in getting started (though not always
correct):

(1) open Loop zeros attract loci of Closed Loop poles,

(2) open Loop poles repel loci of Closed Loop poles,

(3) closed Loop poles close to the origin dominate the response (unless
cancelled by a nearby zero),

(4) closed Loop zeros are identical to the Open Loop zeros (for unity
feedback),

(5) ONLY STABLE poles may be cancelled by nearby zeros (otherwise

the control loop ends up being INTERNALLY UNSTABLE).
120 Control Engineering - 1

By a detailed understanding of the s-plane and judicious application of these five

guidelines to the Root Locus diagram for g(s), a sensible proposal can be made for a
compensation block, k(s), that will improve on the Root Locus of the resulting open
loop system q(s) = g(s) k(s). In practice a number of carefully selected trials may be
required before the design converges to a satisfactory result.

9.3.1. Target Region in the s-Plane

It is convenient to convert specifications for a closed loop system into a region in

the s-plane within which the closed loop poles should lie.

Example 1

Assume that specifications for a closed loop system require

(1) A damping factor exceeding 0.3, and

(2) Transients that decay to the ±2% band within 2.5 [seconds].

The region in the s-plane within which the closed loop poles should lie is given by:

Figure 9.7 Typical Target Region

Ch 9: Root Locus Design Method 121

Example 2

Design a feedback control scheme to stabilise the process modelled by

1
g(s) =
(s + 1)(s − 2)

which has a stable pole at s = −1 and an unstable pole at s = +2 .

The root locus for the open loop system, g(s), shows where the poles of the closed
loop system move as the open loop gain is increased.

Figure 9.8 Root Locus Diagram

The root locus diagram shows quite clearly that closing the loop on the plant g(s),
with unity feedback and no compensation, results in a closed loop system that is
unstable for all values of Root Locus Gain, γ. (Because there is always one pole and
sometimes two poles of the closed loop transfer function
h(s) = g*(s) / [1+γg*(s)]

in the right half s-plane.)

Putting a stable open loop zero near the origin will attract the loci towards (and
hopefully into) the stable left-half s-plane. For example, a cascade element
s + 0. 5
k(s) =
s+3

inserted into the system as follows might suffice.

122 Control Engineering - 1

Figure 9.9 Compensated System

This compensator is a LEAD CIRCUIT. It does not violate the implicit properties
of a satisfactory control element since it is

(1) Stable in open loop (as it has a pole at s = −3. 0 ).

(2) Realisable (as m≤n).

(3) Has a dominant zero at s = −0 . 5 which should attract the loci towards
the left-half s-plane.

The root locus for the new system

s + 0.5
q * (s) =
(s + 3)(s + 1)(s − 2)

is easily computed and drawn

Figure 9.10 Root Locus for Compensated System

Ch 9: Root Locus Design Method 123

As the loop gain is increased, the three closed loop poles move along the root loci.
The unstable closed loop pole on the loci starting at s=+2 is attracted towards the
zero of the compensator, k(s), and crosses into the stable left-half s-plane when the
root locus gain, γ=12. This value for γ is computed from the product formula:
Product of distances to all poles
γ =
Product of distances to all zeros
(3)(1)(2)
= = 12
(0. 5)

Obviously the highest possible loop gain will merely result in a transient decay rate
determined by the dominant and complex conjugate poles at s = −0. 75 ± jb so there
are going to be limits on the transient decay rate that can be attained with this
control system. (This can be improved by further additions to the compensator.)

Check (For interest only)

The closed loop transfer function h(s) is given by

q(s)
h(s) = where q(s) = g(s) k(s)
1 + q(s)
12(s + 0. 5)
=
12(s + 0. 5) + (s + 3)(s + 1)(s - 2)
12s + 6
= 2
when γ=12
s(s + 2s + 7)

As expected, the zeros of h(s) are the same as those of q(s). The closed loop pole
positions are given by the Characteristic Function of h(s)

s(s2 + 2s + 7) when γ=12

or, explicitly,

s=0 and s = −1 ± j 6

From the root locus it is obvious that a slightly higher loop gain is needed to
produce a stable closed loop system. The step response of the original unstable open
loop plant and a stabilised closed loop for γ=20 is shown below, together with the
input required to achieve stabilisation.
124 Control Engineering - 1

Figure 9.11 Step Response of Open and Closed Loop System

Observation

Clearly the root locus diagram for g(s) has guided the designer towards a dynamic
compensator k(s) that stabilizes the system in a closed loop configuration. Further
refinement in the constants of the present k(s) or perhaps a few more zeros (and
poles) should be tested before coming to a final decision.

(Clearly the Type number could do with some improvement, but will destabilise the
loop considerably.)

9.4. Root Locus for Variations of Parameters

The technique for drawing Root Loci based on changes in the gain factor, γ, can be
extended to investigate the effect on closed loop pole positions of changing other
parameters in the loop. These loci are usually known as CHARACTERISTIC LOCI,
rather than Root Loci.

Consider the problem of designing a range of I values that can be used to trim the
following closed loop system without causing instability.
Ch 9: Root Locus Design Method 125

Figure 9.12 Closed Loop System

For this loop

q(s) = g(s) k(s)
2 2(1 + sI)
=
1 + 10s sI

and so
4(1 + sI)
h(s) =
4(1 + sI) + sI(1 + 10s)

The Characteristic equation for the closed loop system, h(s), can be written as
0 = 4(1 + sI) + sI(1 + 10s) = 4 + Is(5 + 10s) = 4 + 10Is(s + 0. 5)
1
= + (s2 + 0. 5s) (Dividing by 10I)
2. 5I
= γ + s(s + 0. 5)

This equation has the polynomial form required for drawing Root Loci. It has Open
Loop Poles at s = 0 and s = −0 . 5 and no zeros.

It root locus is readily sketched to be:

126 Control Engineering - 1

Figure 9.13 Characteristic Loci

From the Characteristic Loci it is concluded that:

(1) The closed loop system is stable for all values of γ, and hence I since
γ=1/2.5I.

(2) At a damping factor of 0.7071, the Root Locus Gain is measured from
the diagram and found to be
γ=(0.354)2=0.125

Thus the ideal value for I is 3.2

9.5. Standard Compensation Elements

In the typical closed loop configuration:

Figure 9.14 Unity Feedback Control Loop

Ch 9: Root Locus Design Method 127

the compensator k(s) is often chosen from a standard range of compensation

elements:

• Lag circuits (Improves stability and Steady State Gain)

• Lead circuits (Improves stability)

• Combination of Lead and Lag circuits

9.5.1. The Lag Circuit

A lag circuit is characterised by a transfer function of the form

1 + sT
k(s) = with 1 < α ≤ 10
1 + sαT

1 1
It has one zero at s = − and one pole at s = − . In the s-plane it provides the
T αT
pole-zero pair:

Figure 9.15 Lag Circuit Pole-Zero

If the input and output of the compensator are in volts then the lag function is
implemented by the passive electronic circuit:

Figure 9.16 Lag Circuit Passive Electronics

128 Control Engineering - 1

The transfer function for this circuit is

E o (s)
k(s) = [V]/[V]
E i (s)

1 + sR 2 C
= [V]/[V]
1 + s(R1 + R 2 )C

In Root Locus designs the dominant pole is used to repel loci in such a way that the
response is improved while the zero helps retain closed loop stability.

9.5.2. The Lead Circuit

Lead circuits are defined by transfer functions of the form

1 + sαT
k(s) = with 1 < α ≤ 10
1 + sT

1 1
It has one zero at s = − and one pole at s = − . It provides the pole-zero pair:
αT T

Figure 9.17 Lead Circuit Pole-Zero

The lead function can be implemented by the simple electronic circuit:

Figure 9.18 Lead Circuit Passive Electronics

Ch 9: Root Locus Design Method 129

Its transfer function model is

E o (s)
k(s) = [V]/[V]
E i (s)

R2 1 + sR1C
= [V]/[V]
R1 + R 2 1 + s R 2 R1 C
R1 + R 2

In Root Locus designs, the dominant zero is used to attract loci in such a way that
the closed loop is stabilized. The pole prevents excessive noise transmission.

Note

The Lead circuit tends to amplify high frequencies, so its response is generally quite
noisy. The problem is aggravated by large α-values, hence the reason for its upper
limit of 10. In practice it may be better to use two lead circuits in cascade, with
small α-values, than one lead with a larger α-value.

9.6. General Compensators

Combinations of (buffered) lead and lag circuits are possible for dealing with
extremely difficult processes. It should be noted that practical implementation of
such compensators may require sophisticated hardware.

For large industrial systems a process control computer is often available so the
compensator k(s) can take the form of any stable transfer function

a 0 + a1s + a 2s+... +a ms m
k(s) = with m ≤ n
b 0 + b1s + b 2s+... +b nsn

Analog electronic circuits that realise such transfer functions can also be designed as
discussed in Chapter 17.

Whatever the requirements, the final choice should be biased towards the simplest
possible k(s) that will do the job (as the ultimate goal in engineering is simplicity).
 10
Dead Time and the Root
Locus
The ROOT LOCUS diagram is a very powerful analytical method for dealing with
dynamic systems. Unfortunately it is not universally applicable and is deficient in
some important features. For example:

(1) It is not able to deal easily with all types of dynamic systems that are
found in practice, particularly in industrial application of control
engineering.

(2) Non-dominant modes of the system behaviour, which need to be

ignored by the designer while using root locus diagrams, are not
hidden by the method and tend to confuse the overall picture.

(3) It does not give a direct indication of system sensitivity to changes in

its open loop model, g(s). This is important in practical applications
since any model of a large process is often only a very rough
approximation of its dynamic characteristics and hence highly
variable.

(4) It has not yet been extended to deal effectively with complex
multivariable processes commonly encountered in industrial
applications of advanced microcomputer control schemes.

The first point is now dealt with in detail to motivate a different approach to system
modelling and controller design.

10.1. Dead-Time in System Responses

Many physical systems exhibit dead-times in their responses to inputs. For example,
a system to control density by the addition of water, defined by the schematic
diagram:
Ch 10: Dead Time and the Root Locus 131

Figure 10.1 Density Control System

will respond to a step in the water valve as follows:

Figure 10.2 Density Control System

The time lapse between the instant when the water valve is stepped and the time
when the density starts to respond is known as DEAD-TIME, τ (Tau). (Its value is
clearly dependent on flowrate but is assumed to be constant during analysis.)

From the Real Translation law for Laplace transforms, the response of the system
can be modelled by
y(s)
= g(s) e − sτ [kg / m 3 ] / [o ]
u(s)

where transfer function g(s), given by

a 0 + a1s+... +a ms m
g(s) = with m ≤ n
b 0 + b1s+... +b nsn

describes the dynamic response of the process, while the exponential term, e − sτ ,
accounts for its dead-time.
132 Control Engineering - 1

The dead-time term is extremely difficult to handle by Pole-Zero methods, like Root
Locus, since it effectively introduces an infinite number of poles and zeros into the
system transfer function.

10.2. Pade Approximation of Dead-Time

To illustrate this, the exponential term can be expanded as the ratio of two
polynomials of infinite order

e −sτ /2 e−θ sτ
e − sτ = +sτ /2
= +θ
defining θ =
e e 2
1 1 1 1
1 − θ + θ2 − θ3 + θ 4 − ... +(-1)n θ n +...
= 2! 3! 4! n!
1 1 1 1
1 + θ + θ2 + θ3 + θ 4 +... + θ n +...
2! 3! 4! n!

Note

(1) If both the numerator and denominator polynomials are truncated to n

terms, this ratio expansion of e − sτ is known as the n'th order Pade
approximation to dead-time, and contains n poles and n zeros. For
example the first order Pade Approximation is:
sτ
1−
− sτ 2
e =
sτ
1+
2

(2) Historically the Pade approximations were, and probably still are, very
important for analog electronic implementations of dead-time terms.

(3) In digital computers it is simple to simulate dead-times, whether

integral or non-intergal multiples of the sampling interval.
 11
Frequency Domain Design
Methods
Although dead-time terms, e −sτ , complicate the Root Locus design method
excessively, they are handled easily by frequency domain methods. These are based
on Frequency Response Models of a plant. Such models can be derived directly by
experimentation on stable process, or from transfer function models, g(s).

The frequency response models contain neither poles nor zeros explicitly, yet their
mathematical descriptions can predict implicitly pole/zero positions. In this way
frequency response methods are used for the design of feedback controllers.

11.1. Frequency Response Models

The frequency response model defines how a system output responds to a sinusoidal
input, once all transients have decayed. For linear systems the eventual output is
also a sinusoid of the same frequency as the input, but of different amplitude and
phase. For example:

Figure 11.1 Sinusoidal Input and Output

where u(t) = uosin(ωt) is the input and y(t) = yosin(ωt+φ) is the output response
(after all transients have decayed).
134 Control Engineering - 1

y0
The AMPLITUDE RATIO (ω )
u0

and PHASE ANGLE φ (ω )

are both scalar functions of frequency, ω, and define the FREQUENCY

RESPONSE MODEL for the system.

11.1.1. Frequency Response Models from Experimentation

Direct experimentation on stable plants can produce frequency response models.

Generally a number of frequencies need to be investigated for satisfactory results.
For high-speed electronic circuits the experimental approach is quite attractive as it
merely needs the injection of sinusoidal test signals and the measurement of
responses. Unfortunately for large plants with long time constants, transients decay
very slowly and such tests, which need ultra-low frequencies, are no longer viable.

11.1.2. Frequency Response Models from Transfer Functions

Alternatively, transfer function models are readily derived from an analysis of step
test data that is easily obtained, and frequency response models can be derived from
transfer function models by a change of variable. The relevant theory follows.

Consider the output response of a linear plant model

y(s) = g(s) u(s)

When the input, u(t), is a sinusoid its Laplace transform is

u(s) = L[u(t)] = L[u 0 sin( ω t)]

ω u0
= (From Laplace Tables)
s + ω2
2

the output response is then simply

ω u0 ω u0
y(s) = g(s) = g(s)
s2 + ω 2 s2 + ω 2

After expansion as partial fractions, the response becomes

kc k *c
y(s) = + + Transients due to g(s)
(s − jω ) (s + jω )
Ch 11: Frequency Domain Design Methods 135

where the residuals k c and k *c are complex conjugate constants, evaluated from the
limits
ωu 0
k c ( jω ) = lim [(s − jω ) g(s) ]
s→ jω s2 + ω 2
ωu 0 u
= lim [ g(s) ] = g( jω ) [ − j 0 ]
s→ jω (s + jω ) 2
and
ωu 0 u u
k *c ( jω ) = lim [ g(s) ] = g( − jω ) [ j 0 ] = g* ( jω ) [ j 0 ]
s→− jω (s − jω ) 2 2

Define g(jω) in polar co-ordinates as

g( jω ) = ρe jθ

Then, after some algebraic manipulations, the response is

s ω
y(s) = u 0 ρ sin( θ ) + u 0 ρ cos ( θ ) + Transients
s2 + ω 2 s2 + ω 2

By inverse Laplace transformation this yields the time response

y(t) = u 0 ρ sin( θ ) cos ( ωt) + u 0 ρ cos ( θ ) sin( ωt) + Transients
= u 0 ρ sin( ωt + θ ) + Transients

Hence the frequency response model is simply

y0
( jω ) = ρ = Magnitude of g( jω ) = g( jω )
u0
and
φ ( jω ) = θ = Argument of g( jω ) = ∠ [g( jω )]

which can be computed directly from the transfer function model, g(s).

Thus the Frequency response model, g(jω), is given EXACTLY by the transfer
function model, g(s), evaluated at s=jω.
136 Control Engineering - 1

Example

Consider the simple R-C low-pass filter:

Figure 11.2 An R-C Circuit

The differential equation for the system is

d
RC y(t) + y(t) = u(t)
dt

where y(t) is the output voltage across C in [Volts] and u(t) is the driving voltage in
[Volts]. Taking Laplace transforms, ensuring that all initial conditions are zero,
gives the transfer function model
y(s)
g(s) =
u(s)
1
=
1 + sRC

Thus the frequency response model is

1
g( jω ) =
1 + jω RC

Its amplitude ratio is

yo 1
= g( jω ) = where T=RC
uo 1 + ω 2T2

and the phase angle is

φ = Angle[g( jω )] = − tan-1 ( ω T)

Thus if the input is

u(t) = 12 sin( ω t) [Volts]
Ch 11: Frequency Domain Design Methods 137

then its output response is

y(t) = 12 g( jω ) sin( ω t + ∠g( jω ) )

12
= sin( ω t − tan−1 ( ωT) )
1 + ω 2 T2

11.2. Representation of Frequency Response Models

The frequency response model, g(jω), is a set of complex numbers, parametrised by
frequency. Graphically the model can be represented in many ways, each with its
particular advantages (and disadvantages) for designing closed loop control systems.
For example, the model can be shown as:

(1) Two Functions of Frequency

Figure 11.3 Frequency Response Model on an Arithmetic Plot

These are its ARITHMETIC PLOT.

(2) A Single Graph or Polar Plot

This graph is parametrised by frequency, ω. Each point on the plot is given in polar
co-ordinates by the vector (<y0/u0> , <φ>).
138 Control Engineering - 1

Figure 11.4 Frequency Response Model on a Nyquist Plot

This example is known as the NYQUIST PLOT.

Note that the frequency response model is easily derived from either the original
differential equations or the Laplace transforms by direct substitution of (jω) for
d
( ), (D) or (s). As before, it is assumed that all initial conditions are zero.
dt

Example 1

If the input-output relationship in a plant is modelled by the differential equation

(1 + 4 D + D2 ) y(t) = (2 + 3 D ) u(t)

then its frequency response model is simply

2 + j3ω
g( jω ) =
1 − ω 2 + j4 ω

where (jω) has been substituted for the derivative operator (D).

Example 2

If a plant with dead-time is modelled by the transfer function

e −3s
y(s) = u(s)
1 + 7s

then its frequency response model is

Ch 11: Frequency Domain Design Methods 139

e − j3ω
g( jω ) =
1 + j7 ω
1
= g1 ( jω ) e − j3ω where g1 ( jω ) =
1 + j7 ω

Note

− j3ω
(1) The dead-time term, e , is merely the complex number

cos (3ω ) − j sin(3ω )

so the frequency response model has no difficulty representing dead-

time terms.

(2) The dead-time term, e − j3ω , does not change the magnitude of the
frequency response model, g1(jω). It only increases its phase angle by
−3ω [radians].

11.3. Mathematics for Frequency Domain Methods

Frequency response models form the basis for a number of extremely versatile
methods that have been developed to design feedback control systems. These
techniques rely on a few simple results from complex variable theory.

11.3.1. Mapping of s-plane Contours to the q-plane

Recall that the stability of a closed loop control system is determined by its pole
positions in the s-plane. Although the frequency response model no longer contains
explicit information on system poles, particularly when dead-times are involved, it is
possible to study a polar plot of the model of an open loop system, q(jω), and to
predict whether the poles of its corresponding closed loop system, h(s), are stable or
not. (The analysis is based on the Principle of the Argument.)

The polar plot of a frequency response model, q(jω), is simply its complex value
plotted on an Argand diagram as a function of frequency, ω. (The Argand diagram
for q(jω) is called the q-plane in these notes.)

For example, consider the open loop transfer function:

140 Control Engineering - 1

2
q(s) =
1 + 5s

Its frequency response model is

2
q( jω ) =
1 + 5 jω

At s = σ+jω = j2 [radians/second] this function takes the value

2
q( j2) =
1 + j10

Or, in polar co-ordinates,

q(j2) = [r, φ] = [0.20 , -84.3°]

Graphically the complex numbers 's' and 'q(s)' can be shown as two 2-Dimensional
vectors on separate Argand diagrams, or complex planes

Figure 11.5 Conformal Mapping

Clearly, as indicated on the figure, function q(s) can be considered to have mapped
the point 's = 0+j2' in the s-plane to the point 'q = 0. 2 e − j84.3 ' in the q-plane.
0

Extending this concept further, the function q(s) can be evaluated and plotted in the
q-plane as the variable 's' traverses its imaginary jω-axis in the s-plane, from ω=0
to ω→∞ . This operation can be viewed as mapping the entire jω-axis in the s-plane
to a contour in the q-plane. Obviously the points that q(s) maps out as 's' traverses its
imaginary axis are actually the points q(jω) and correspond exactly to the points
drawn out by its frequency response model.

Thus the complex function q(jω) can be viewed as a mapping of the jω-axis in the s-
plane onto a contour in the q-plane. A typical result is the polar plot:
Ch 11: Frequency Domain Design Methods 141

Figure 11.6 Typical Polar Plot of q(jω)

It is important to notice that the vector from the point (-1/f , 0) to the point q(jω) in
the polar plot represents the complex number
1
+ q( jω ) where f is a real-valued constant.
f

And, if a new set of axes were constructed to pass through the (-1/f , 0) point, then
the given polar plot also represents the frequency response of the complex function
1
X(s) = + q(s)
f

Now, let 's' take on more values by moving clockwise along a large semi-circle of
infinite radius until it reaches the s = − jω axis. Finally let 's' move up this negative
axis until it returns to its origin s = 0. In this way 's' traverses a huge semi-circle that
encloses the whole of the right half of the s-plane. The path that 's' followed is
important for control and is referred to as the NYQUIST CONTOUR or the D
CONTOUR.

Figure 11.7 The Nyquist Contour

142 Control Engineering - 1

Contour D is designed to enclose all the right-half of the s-plane and to avoid any
singularities on the jω axis due to poles of X(s) by indenting into the right-half s-
plane as shown, when necessary. (These poles are same as those of q(s).)

11.3.2. Principle of the Argument

A plot of X(s) itself, as s traverses the Nyquist contour, traces out a contour in its
own plane. Note that the shape of the contour of X(s) is identical to that of q(s).
Also the origin of the X-plane is the so-called CRITICAL POINT at (-1/f, 0) in the
q-plane. A typical X contour and its q-plane equivalent are:

(a) X(s) on X-Plane (b) q(s) on q-Plane

Figure 11.8 Typical Plots of X(s) and q(s)

If X(s) is a complex function of 's' with both zeros and poles in the s-plane, and

<Z> = The number of zeros of X(s) that lie inside the D contour
<P> = The number of poles of X(s) that lie inside the D contour
<N> = The net number of encirclements that X makes of its origin

then the PRINCIPLE OF THE ARGUMENTS states that

<N> = <Z>−<P>

Example

In the above plots, it is obvious that <N>=2 since X(s) encircles its origin twice.
Exactly the same is concluded by studying q(s), which encircles its critical point at
(-1/f , 0) twice. This means that
2= <Z>−<P>

and X(s) has two more zeros than poles in the right-half s-plane. (The precise
Ch 11: Frequency Domain Design Methods 143

number of zeros and number of poles is not given by the principle of the argument.)

11.3.3. Use of Symmetry

Note that the contour X(s) and q(s), as 's' traverses the full Nyquist contour, is
symmetric about its real axis. Thus it is common practice only to evaluate and plot
q(s) as s traverses from point 'a' to point 'c' through point 'b' in the D contour

(a) Half D Contour (b) Half of q(s) on q-Plane

Figure 11.9 Symmetry in the Polar Plot

11.3.4. Further simplification

Also when 's' traverses from point 'b' to point 'c' the open loop transfer function

a 0 + a1s + a 2s2 +... +a ms m

q(s) = with m ≤ n
b0 + b1s + b 2s2 +... +b ns n

tends to a (constant) limit

a ms m
→ →0 if m < n
b ns n

am
→ if m = n
bn

In both cases, moving 's' from point 'b' to point 'c' has no effect on q(s) near its
critical point, and can be ignored (unless a n = − b n ).
Hence the polar plot used in practice shows only q(jω) as 's' moves from point 'a' to
point 'b' on the Nyquist contour. This is simply the Frequency Response model
144 Control Engineering - 1

plotted on polar co-ordinates and is known as the NYQUIST PLOT of q(s). The
contour X(s) is immediately deduced from q(s).

11.3.5. The Nyquist Plot

The Nyquist Plot is an extremely versatile representation of the frequency response

model of an OPEN LOOP system, q(jω), and is used extensively in the design of
feedback control systems. It can be interpreted to reveal a variety of properties of the
CLOSED LOOP system, h(jω), like:

• Closed loop stability

• Closed loop damping factor and overshoot

• Closed loop frequency response

• Closed loop bandwidth

• Sensitivity to modelling errors, particularly in g(s)

• Disturbance and noise rejection

It also gives direct guidance on what compensation k(s) is required to improve on

the system performance in closed loop.

11.4. Closed Loop Stability from Nyquist Plots

Consider the closed loop system:

Figure 11.10 Closed Loop System

in which the feedback element, f, is assumed to be a convenient scaling factor that is

not a function of 's'. The open loop transfer function is
Ch 11: Frequency Domain Design Methods 145

N(s)
q(s) = g(s) k(s) =
D(s)

A useful complex function X(s) is then defined as

1
X(s) = + q(s) = f + q(s)
f
D(s) + f N(s)
=
f D(s)
φ c (s)
=
f φ o (s)

where φc (s) is the closed loop characteristic function and φo (s) the open loop
characteristic function of the feedback system:

Figure 11.11 Closed Loop System

Element f is a constant gain and the origin of X(s) is at (-1/f , 0) in the q-plane.

The principle of the argument, applied to this X(s), holds that the net number of
likewise encirclements that q(s) makes of its critical point (-1/f , 0) is given by

<N> = <Z> - <P>

= <No of unstable C/L poles> - <No of unstable O/L poles>

since the Nyquist contour encircles the entire right-half s-plane.

The Nyquist Stability Criterion

By inspection of the Nyquist plot of any open loop system model, q(s), it is possible
146 Control Engineering - 1

to count the number of likewise encirclements, <N>, that q(jω) makes of its critical
point (-1/f , 0) as 's' traverses contour D. Then its closed loop system, h(s), is stable
if and only if

< N > = − p0

where p 0 is the <No of unstable O/L poles>. This criterion ensures that the number
of unstable closed loop poles, <Z>, is zero (i.e. h(s) is stable).

Example 1

Find all values of the feedback gain, f, that ensure stability of the closed loop
system:

Figure 11.12 Closed Loop System

Plot the Nyquist diagram for

4
q(s) =
1 + 2s

as 's' traverses the Nyquist contour. The result is:

Ch 11: Frequency Domain Design Methods 147

Figure 11.13 Nyquist Contour and its Mapping

From q(s) it is clear that p 0 = 0 because there are no unstable poles in the open
loop. (Determined from the Routh-Hurwitz Array, if necessary.) Thus for C/L
stability the number of encirclements of the critical (-1/f, 0) point, <N>, must be
zero.

By inspection of the Nyquist plot this is true for

1
f>0 (from − < 0 with f > 0)
f

which means that the closed loop system is going to be stable for all gains in a
negative feedback configuration. However it is interesting to note that the closed
loop is also going to be stable for a limited range of positive feedback gains, namely,
for
1 1
− <f<0 (from − > 4 with f < 0)
4 f

(In this case positive feedback is not recommended, even though it is stable, since
the resulting closed loop system will be slower than the open loop.)

Example 2

Find values of controller gain, k, that stabilise the open loop system q(s) in the
feedback control loop:

Figure 11.14 Control of an Unstable System

In analysing this loop it is generally useful to move the controller gain, k, into the
feedback path by block diagram algebra and to study the modified system which has
a closed loop response given by
k q(s)
y(s) = r(s)
1 + k q(s)
148 Control Engineering - 1

q(s)
= f r(s) where f = k
1 + f q(s)
= h(s) r(s)

Movement of the controller gain into the feedback path allows the theory to
incorporate a mobile critical point ( −1 / f , 0) into the Nyquist plot. This critical
point is readily moved along the real q(jω) axis until a satisfactory gain is achieved.
Once the design is complete, the controller gain is then merely moved back into the
forward path for controller implementation.

The Nyquist plot for q(jω) in this example is:

Figure 11.15 Nyquist Plot of the Unstable System

Clearly the <No of likewise encirclements of critical point> = 0 for negative

feedback gains. Moving the critical point anywhere along the real axis there is no
point at which q(jω) encircles the critical point once in an anti-likewise direction.
Thus there is no controller gain that will stabilise the loop.

It is possible to compute the number of closed loop unstable poles since

<No of unstable C/L poles> = <Z>
= <N> + p0 = 1 for f>-1/3.5 (including f>0)
= <N> + p0 = 2 for f<-1/3.5

11.4.1. Number of Encirclements

To determine easily the number of encirclements that a particular Nyquist diagram

for q(jω) makes of the critical (-1/f , 0) point, draw a straight line through the critical
point to beyond the diagram. Then at each point where the Nyquist diagram crosses
Ch 11: Frequency Domain Design Methods 149

this line write (+1) if the crossing is in a likewise direction to that being traversed by
's' along the D contour or (-1) if it is in an anti-likewise direction. To ascertain the
value for <N>, total the (+1) and (-1) of the diagram.

For example, consider the complicated (but NOT atypical) Nyquist plot:

Figure 11.16 Encirclements by a Nyquist Plot

Here <N> = (+1) + (-1) + (-1) + (+1) = 0

150 Control Engineering - 1

11.5. Relative Stability from Nyquist Plots

In engineering applications, relative stability of the feedback control system:

Figure 11.17 Feedback Control Loop

is generally of more interest that absolute stability. This means that designers like to
know whether or not the poles of the closed loop system lie within the following
region shown shaded in the s-plane where Zeta, ζ, is the damping factor for second
order systems.

Figure 11.18 Region of Damped Response

Ch 11: Frequency Domain Design Methods 151

11.6. Phase Margins and Gain Margins

Two important measures of relative stability for frequency domain methods are:

(1) PHASE MARGIN, that lets Nyquist plots of the open loop model, q(j
ω), indicate whether the poles of the closed loop system, h(s), are
damped or not,

(2) GAIN MARGIN, that indicates how sensitive a given system is to

changes in its process model g(s).

These two parameters measure the amount of extra phase angle and extra gain for
q(jω) that would result in marginal stability of the closed loop system (i.e. that
would give closed loop poles along the imaginary axis in the s-plane).

A few examples that illustrate how the Phase Margin and Gain Margin are measured
from particular Nyquist Plots are now considered.

Example 1 – Positive Phase Margin

Assume that frequency response experimentation on a plant produced the Nyquist

Plot:

Figure 11.19 Nyquist Plot

By definition of the phase margin, it is measured directly from the Nyquist diagram
and found to be
PM = +24o

where the positive sign indicates that the system is stable.

152 Control Engineering - 1

Example 2 – Negative Phase Margin

Figure 11.20 Nyquist Plot

The phase margin here is

PM = -27o (and the C/L system is unstable)

Example 3 –Gain Margin

Figure 11.21 Nyquist Plot

From the Nyquist Plot the Gain Margin (GM) is defined by:
1
GM = − with f > 0 always, and a < 0 here.
af
1
= 20Log10 [ − ] in [dB]
af
Ch 11: Frequency Domain Design Methods 153

Observe that multiplying the open loop transfer function by the gain margin forces
the Nyquist plot of q(jω) to pass directly through the critical point and results in a
closed loop system that oscillates forever. Any further increase in gain would make
the closed loop system unstable.

Example 4 – Gain Margin

Figure 11.22 Nyquist Plot

The plot does not cross the negative real axis in the q-plane and so
GM = ∞

This is unlikely to be an accurate prediction since any practical system is extremely

likely to have non-dominant poles that would produce a finite gain margin.

11.6.1. Closed Loop Damping Factor from Open Loop Phase Margin

A number of useful relationships between the phase margin of an open loop system,
q(s), and the performance of its closed loop system, h(s), are now derived.

Assume that the closed loop system is approximated by a dominant second order
response with unity gain. Its transfer function is

ω 2n q(s)
h(s) = =
s 2
+ 2 ζω ns + ω 2n 1 + q(s)

The corresponding open loop transfer function, q(s), would be

h(s)
q(s) =
1 − h(s)
154 Control Engineering - 1

ω 2n
=
s(s + 2 ζω n )

A plot of q(jω) on a Nyquist diagram reveals that

2ζ
PhaseMargin of q(s) in degrees = tan−1 [ ]
4ζ 4 + 1 − 2ζ2
≈ 100ζ

where ζ is the damping factor of h(s) and the Phase Margin of q(s) is given in
degrees [°]. The approximation holds good for the range
0 < ζ < 0.7071
0 < PM < 64°

which covers most useful situations.

Example

For a closed loop system to have a damping factor of 0.6 or more, the phase margin
of q(jω) must exceed 60°. This provides a design constraint on the Nyquist plot.

11.6.2. C/L Overshoot from O/L Phase Margin

For a step response in the closed loop system:

Figure 11.23 Overshoot in the Closed Loop Step Response

the maximum percentage overshoot of y(t) above its steady state value is given by:
Ch 11: Frequency Domain Design Methods 155

− πζ
PercentageOvershoot, M p , of h(s) = 100 exp ( )
1 − ζ2

≈ 75 − PhaseMargin of q(s) in degrees

Example

A closed loop system, h(s), in which the open loop system, q(s), has a phase margin
of 27° will have an overshoot of approximately 48% when its setpoint is stepped.

11.6.3. General Systems

For higher order systems only the general trends hold true, unless a dominant pole-
pair exists. Many industrial engineering applications exhibit first or second order
behaviour, so these approximate relationships are extremely practical.

Acceptable closed loop systems result if

PM > 30° (Chemical systems) or > 70° (Mechanical systems)
and
GM > 2 (= 6 [dB])

In the Nyquist diagram this identifies the graphical criteria that frequency response
models of open loop plants should lie to the right of the points indicated in the
following diagram:

Figure 11.24 Design Specifications in the Nyquist Plot

156 Control Engineering - 1

11.7. Constant M Circles in the Nyquist Plot

More detailed information about closed loop performance can be derived from its
open loop frequency response plot. For example, M-Circles show the amplitude of
the closed loop frequency response model, h(s) in the Nyquist Plots of the
corresponding open loop system, q(s).

A Nyquist diagram for the open loop frequency response model q(jω) can be used to
predict exactly the gain |h(jω)| of the frequency response model of its closed loop
system, where:

Figure 11.25 Closed Loop Configuration

and
q(s)
h(s) =
1 + q(s)

For a given fixed gain, M, of the closed loop system, h(s),

yo < Output Amplitude >
M = h( jω ) = =
ro < Setpoint Amplitude >

x + jy
= if q(jω) = x + jy
1 + x + jy

This equation defines a circle in the Nyquist plot

M2 M2
(x + )2 + y 2 =
M2 − 1 (M 2 − 1)2

M2
with Centre (− , 0)
M2 − 1
M
and Radius
M2 − 1
Ch 11: Frequency Domain Design Methods 157

On the Nyquist diagram a number of constant M-circles can be drawn for different
values of closed loop gain M.

Figure 11.26 Constant M Circles

Thus it is possible to determine the amplitude ratio of the frequency response model
h(jω) of a closed loop system by inspection of the Nyquist plot of its open loop
frequency response model q(jω), on which constant M-circles have been drawn.

Example

To predict the bandwidth of the CLOSED LOOP system, h(s), from a Nyquist
diagram of its OPEN LOOP model, q(jω) observe the frequency at which the plot of
q(jω) crosses into the M=0.7071 circle which marks the -3dB point of the closed
loop frequency response. (It is assumed that h(j0)=1, or equivalently that q(j0) lies
on an M≈1 circle.)

When the Nyquist plot is used to design a compensator k(s) for a process model g(s)
the maximum M-value that q(jω) reaches should be limited to the range
1.1 < Mmax < 1.5 or 1 dB < Mmax < 3 dB

to avoid excessive oscillations in the closed loop system response. This provides an
alternative specification for designs using Nyquist Plots.
158 Control Engineering - 1

11.8. Constant N Circles in the Nyquist Plot

Similar loci can be sketched in the Nyquist plot along which the phase angle of the
closed loop system, h(jω), remains constant, φ. Again it is easy to show that these
loci are circles, defined by

1 1 N2 + 1
(x + )2 + (y − N)2 =
2 2 4N 2

where N = tan(φ).

Figure 11.27 Constant N Circles

 12
The Nyquist Plot
Design Method
Consider the typical feedback control loop:

Figure 12.1 A Control Loop

A design session starts by plotting the frequency response model, g(jω), for the plant
(plus any type number adjustment) on a Nyquist diagram. This assumes that k(s)=1
and presents the problem, g(s), in an appropriate form.

To continue, the following design information needs to be given in the same form
(i.e. on a Nyquist plot):

(1) the closed loop specifications (Phase Margin, Gain Margin, M circle,
C/L bandwidth, C/L damping factor, etc.),

(2) various possible compensators (Lead, Lag and other standard

compensators).

The design engineer studies the Nyquist plot for g(jω) and then selects a suitable
compensator, or combination of compensators, k(s), that is likely to improve on the
original system g(jω). The new open loop transfer function q(jω) = g(jω) k(jω) is
then computed and plotted on the Nyquist diagram. After a few well-chosen
iterations a good compensator that meets the closed loop specifications is usually
found.
160 Control Engineering - 1

Example

Consider a plant:

Figure 12.2 The Process Model

approximated by the following transfer function model

10 e − s
g(s) = [kg/m3]/[°]
1+ s

Design a compensator for this process that results in a C/L system h(s) with a
damping factor of 0.3 or more. Also ensure an adequate Gain Margin (> 2).

First it should be noted that the dead-time in the process is of the same order of
magnitude as the time constant. Thus it cannot be neglected and the Root Locus
method is not suitable for analysing this plant.

The Nyquist plot of the frequency response model g(jω) is:

Figure 12.3 Nyquist Plot for Model g(s)

The design calls for a damping factor of 0.3 or more in h(s). This criterion translates
to a requirement that the open loop system q(jω) have a Phase margin of 30° or
Ch 12: The Nyquist Plot Design Method 161

more, which in turn can be represented on the Nyquist plot by a line drawn through
the origin at an angle of -150°.

From the plot of g(jω) it is clear that without a compensator element, k(s), in the
feed forward path:

(1) The system g(jω) in closed loop is unstable for feedback gains, f,
greater than 0.23 (from 1/4.4).

(2) For the specified closed loop damping factor a Phase Margin of 30° is
required for the open loop model q(jω). To achieve this, the gain, f,
should be 0.19 (from 1/5.36).

(3) At a gain of f=0.19 the Gain Margin is a mere 1.22 (from 5.36/4.4) or
1.7dB and needs to be increased to exceed the recommended value of
2 or 6dB. This could be achieved by decreasing f further to 0.12 (from
1/8.8), but so small a gain gives a large damping factor, a very slow
closed loop response and worse steady state error.

(4) The steady state error at f=1 would be 0.091 (from 1/(1+10) for a
unity step setpoint), but the closed loop is unstable. At f=0.19 the
steady state error increases to 0.34 (from 1.9/2.9).

(5) The closed loop bandwidth with f=1 is in the range 3 to 4 [rad/s]
(from the frequency at which the Nyquist plot of q(jω) enters the
M=0.7071 circle). It drops to approximately 1 [rad/s] when the gain is
reduced to f=0.19.

As mentioned previously, two possible compensator types, namely LEAD or LAG,

are available. In this case a Lag circuit for k(s) would decrease the open loop gain of
q(jω) at high frequencies and is likely to improve on the Gain Margin. For example
the lag circuit
1 + 0. 6s
k(s) =
1 + 2s

has a low-frequency gain of 1.0 (from 1/1) and a high-frequency gain of 0.3 (from
0.6/2) for ω >> 1.4 (from 1/0.6) [rad/s].

Note that steady state error is not given much attention in this example, but cannot
be ignored in a practical situation.

The Nyquist plot of the compensated open loop system q(jω)=g(jω)k(jω) is:
162 Control Engineering - 1

Figure 12.4 Nyquist Diagram for the O/L Model q(s)

At a Phase Margin of 30[°] this compensator gives an improved Gain Margin of

1.54 (or 3.74 dB) which is much closer to the recommended value of 2 (or 6dB).

The parameters '0.6' and '2' of k(s) can be trimmed, or further lag circuits could be
installed, in order to provide the necessary Gain Margin and the designer would
embark on a few more trials, each evaluated by a study of the resulting Nyquist plot
for Gain Margin and Phase Margin. (In practice Nyquist plots are computed and
displayed with ease by modern digital computers, so the design engineer's primary
task is to evaluate these plots carefully and to suggest modifications to k(s) that
improve on the characteristics of the closed loop system).

12.1. Standard Compensators in the Nyquist Plot

Compensation elements, k(s), in the feedback control loop:

Figure 12.5 Typical Feedback Control Loop

generally take the form of Lead or Lag circuits, or combinations of both. It is thus
important to know how these will affect the frequency response model, g(jω), of a
Ch 12: The Nyquist Plot Design Method 163

process in order to improve on its open loop model, q(jω).

12.1.1. Lag Circuits

In the context of Nyquist plots, the function of Lag circuits is to:

(1) attenuate high-frequencies, which improves gain margins, and

(2) amplify low-frequencies, which boosts loop gain and reduces steady
state errors.

It should be noted that the phase lag introduced by a lag circuit swings the Nyquist
plot clockwise towards the critical point. This is an undesirable side-effect, that is
partially minimised at high frequencies by the presence of its zero (which explains
why a simple R-C circuit having only one pole is avoided in control loops).

The transfer function for a lag circuit is simply

1 + Ts
k(s) = K
1 + αTs

with 1 < α ≤ 10 and its pole being dominant.

On a Nyquist diagram k(jω) has the following trace:

Figure 12.6 Nyquist Plot for a Lag Circuit

The maximum attenuation provided by the circuit is K/α and occurs at high
frequencies (ω > 1/T). For low-frequency amplification (at ω < 1/αT) a gain term,
K, is included in k(s). (Remember that K=1 is usually assumed since it is moved into
the feedback path to become feedback gain, f=K, giving critical point -1/f in the
164 Control Engineering - 1

Nyquist plot).

The maximum phase lag introduced by the lag circuit depends only on α and is
given by the trigonometric function

1/ α − α
φcm = tan−1 ( )<0
2

and occurs at a frequency

1
ω cm =
αT

Note

(1) The phase lag is an undesirable effect introduced by the lag circuit.

(2) A larger α value gives a larger phase lag.

(3) The frequency at which φcm occurs is set by T.

In designing a lag compensator for a given plant, the initial Nyquist plot for g(jω)
would suggest what low-frequency gain is required to boost the gain of q(jω)
sufficiently to ensure that the desired closed loop gain of h(jω), indicated by M
circles, is achieved. The lag circuit gain, K, is then selected to provide this low-
frequency gain; at the same time constant α is chosen to reduce the loop gain at
high-frequencies in order to avoid instability and provide adequate gain margin. The
frequency at which the dominant effects are needed is finally set by T.

12.1.2. Lead Circuits

The primary function of lead circuits is to improve on the Phase Margin by

providing additional phase advance in the circuit. Unfortunately the circuit also
introduces extra gain at high frequency as an undesirable side effect.

The transfer function for a lead circuit is simply

1 + αTs
k(s) = K
1 + Ts

with 1 < α ≤ 10 and its zero is dominant.

Ch 12: The Nyquist Plot Design Method 165

On a Nyquist diagram k(jω) has the following trace:

Figure 12.7 Nyquist Plot for a Lead Circuit

The maximum phase advance provided by the lead circuit is simply

α − 1/ α
φcm = tan−1 ( )>0
2

which occurs at a frequency

1
ω cm =
αT

The low-frequency gain provided by the lead circuit is K, while the undesired high
frequency gain is Kα. (Once again K=1 as this term is usually included in the
critical point (-1/f , 0) in the Nyquist plot).

In designing lead compensators, the value set for α determines the amount of phase
advance that will be introduced into q(jω) while the choice of T sets the frequency at
which this phase advance is provided. Since α also determines the high-frequency
gain, it should be kept at a minimum. (In practice, two or more lead circuits with
small α-values are more effective than one lead circuit with a large α-value.)
166 Control Engineering - 1

12.2. Summary of Lag and Lead Compensators

The Lag circuit introduces a frequency-dependent loop gain that improves on the
system characteristics at low frequencies by increasing the magnitude of q(s) as
variable s→0. At high frequencies it increases the Gain Margin, thereby improving
stability.

The Lead circuit is intended to improve on Phase Margin thus damping the closed
loop response, usually making the system faster with less oscillatory behaviour. It is
unfortunately sensitive to measurement noise.

In practice a clever combination of well-designed lead-lag circuits may be necessary

to compensate a process adequately.

Example

In a helicopter, lift is provided by the main rotor and manipulated by adjusting the
angle of the blades of this rotor. For the purposes of a control analysis, the dominant
relationship between <altitude> in [m] and <main-rotor pitch> in [°] of a hovering
helicopter:

Figure 12.8 Hovering Helicopter

can be approximated very simply by the transfer function

4
g(s) = [m/o]
s2

This double-integrating system is virtually impossible to control in open loop in the

presence of disturbances.

The Nyquist plot of g(jω) is:

Ch 12: The Nyquist Plot Design Method 167

Figure 12.9 Helicopter Nyquist Plot

This plot of g(jω) shows clearly that the uncompensated helicopter in closed loop is
marginally stable, as the damping factor of h(s) is 0. Also, since the plot of g(jω)
passes through the M=∞ circle at ω=2 the uncompensated helicopter in closed loop
will oscillate with a frequency of 2 [rad/s].

Figure 12.10 Uncompensated Closed Loop Response

Clearly the present system, g(s), needs more phase margin. Thus a Lead circuit is
proposed to introduce phase advance into the loop in order to improve on the
damping factor of the closed loop system h(s).

Setting α=4 (initially compromising between phase advance and high frequency
gain) the Lead circuit will introduce a phase advance of

4 − 1/ 4
φ = tan−1 ( ) = 37. 0 [o ]
2

(Larger values of α would give more phase advance, but aggravate noise problems).
As an initial estimate, the phase advance should occur around ω=2 [rad/sec] (since
168 Control Engineering - 1

this is the frequency at which the magnitude of g(jω) is 1 and so the Lead circuit is
likely to have the most effect on Phase Margin). Thus because
1
ω cm = 2 =
αT

the require choice for T in k(s) is

1
T= = 0. 25 [sec]
2 α

A possible lead compensator for the helicopter is thus

1+ s
k(s) =
1 + 0. 25s

giving the open loop transfer function

4(1 + s)
q(s) =
s2 (1 + 0. 25s)

for the compensated system. Its frequency response model

4(1 + jω )
q( jω ) =
− ω 2 (1 + j0. 25ω )

produces the Nyquist plot:

Figure 12.11 Nyquist Plot of Compensated Helicopter

Ch 12: The Nyquist Plot Design Method 169

which has a Phase Margin of 32° when K=1 and an infinite Gain Margin. The
closed loop damping factor is thus 0.32 with a 43[%] overshoot to step inputs and
its bandwidth is approximately 3-to-6 [rad/s] as determined from M-circles.

These predictions are confirmed by digital simulation of the controlled system.

Figure 12.12 Altitude Step Response

Notice the large initial value (of 4 [°]) for the input. Such input spikes are common
in lead compensation controllers and should be reduced by introducing a low pass
filter into the loop. This is positioned after k(s) on signal u(s) and must have non-
dominant dynamics relative to h(s) so that it does not alter the closed loop
performance.

The Nyquist plot of q(jω) shows that a decrease in the gain, K, of the lead circuit
will increase the phase margin of q(jω) slightly and hence the damping factor of the
closed loop system, h(s). It will also slow the response of h(s) down (which is
deduced from inspection of the M=0.7071 circle).

Clearly the first attempt at designing k(s) is partially successful, but can be improved
by further adjustments that make optimum use of the phase advance provided by a
Lead circuit.
 13
Alternative Frequency
Response Plots
A number of alternative representations of frequency response models have been
derived over the years. Each has its own strong features and the astute control
engineer will choose that representation which best suits the needs of the design
under consideration. Also, some industries have adopted particular representations
as their standards.

Thus the concepts derived for Nyquist Plots are now re-stated for Bode Diagrams,
Nichols Charts and Inverse Nyquist Plots.

13.1. Bode Diagrams

Transfer functions like q(s) can be factored into the product of one or more of the
following standard terms (or their inverses):

s Pure derivative/integral

1+sT A single Zero/Pole

s2 + 2 ζω ns + ω 2n
A pair of Zeros/Poles
ω 2n

Multiplicative combinations of these terms and their inverses can be handled

additively if the frequency response model is given as:
y0
20Log10 ( ) in [dB] and Phase, φ in [°]
u0

For example, the transfer function model:

Ch 13: Alternative Frequency Response Plots 171

s
q(s) =
(1 + sT1 )(1 + sT2 )

has a frequency response model that can be written as

1 1
q( jω ) = jω × ×
1 + jωT1 1 + jωT2

Taking complex Logarithms this becomes

20Log10 q( jω ) = 20Log10 jω − 20Log10 1 + sT1 − 20Log10 1 + sT2

with phase angles

Angle[q( jω )] = Angle[ jω ] − Angle[1 + jωT1 ] − Angle[1 + jωT2 ]

= 90 ° − tan-1 ( ωT1 ) − tan-1 ( ωT2 )

If both {20Log10 ïq(jω)ô} and {Angle[q(jω)] in [°]} are plotted on two separate
Cartesian graphs as functions of {Log10 ω} then the magnitude graph can be
approximated quite accurately by straight line asymptotes.

This graphical representation of the frequency response model is known as a BODE

DIAGRAM. It has a number of advantages, since:

(1) multiplicative combinations of sub-systems in cascade simplify to

additive operations,

(2) linear asymptotes (with slopes 0, ±20, ±40, ±60,... [dB/decade])

approximate the gain curve well,

(3) the open loop gain and bandwidth is shown explicitly,

(4) Gain and Phase Margins are easily determined.

Its main disadvantage is that it does not predict instability in systems that contain
any zero in the right-half s-plane i.e. in Non-Minimal Phase systems. (For such
systems the Nyquist stability criterion must be applied.) There are other
disadvantages as will become clear later.

13.1.1. Typical Bode Diagram

Consider a refrigerated storage room in which the flow of Freon is used to regulate
room temperature:
172 Control Engineering - 1

Figure 13.1 Cold Storage Room

The transfer function model for this thermal system is assumed to be

y(s) 4 e −2s
g(s) = = [°C] / [litre/minute]
u(s) (1 + 5s)(1 + 60s)

Its Bode Diagram consists of two graphs:

Figure 13.2 Bode Plot for Cold Storage Room

13.1.2. Closed Loop Stability (Bode Diagrams)

Stability of the closed loop system can be determined from a Bode Diagram of the
open loop model. Specifically closed loop stability requires that:

(1) q(s) be open loop stable

(2) 20Log10 q( jω ) < 0dB at Angle[q( jω )] = − π

Ch 13: Alternative Frequency Response Plots 173

13.1.3. Gain and Phase Margins (Bode Diagrams)

Recall that Gain and Phase Margins of the open loop provide an indication of
robustness and damping for the closed loop system.

By comparison of the Nyquist plot:

Figure 13.3 Nyquist Plot for Cold Storage Room

and its corresponding Bode Diagrams, the Gain Margin (=10) and Phase Margin
(=80°) are readily obtained from the latter:

Figure 13.4 Bode Plot for Cold Storage Room

174 Control Engineering - 1

13.1.4. Dynamic Compensators (Bode Diagrams)

Once again, when using a particular system representation, it is important to know

what to expect from the standard lead/lag compensation elements.

Lag Circuit

This standard compensator is intended to boost low frequency gains or to reduce

high frequency gains (or both). Unfortunately it also introduces undesirable phase
lag into the loop. Its transfer function is
1 + sT
k(s) = K 1 < α ≤ 10
1 + sαT

and its Bode diagram is:

Figure 13.5 Bode Plot for Lag Circuit

Lead Circuit

The lead circuit improves on the phase margin for q(jω). It does however amplify
high frequency noise and should be used with caution. Its transfer function is
1 + sαT
k(s) = K 1 < α ≤ 10
1 + sT

Its Bode diagram shows this clearly:

Ch 13: Alternative Frequency Response Plots 175

Figure 13.6 Bode Plot for Lead Circuit

13.1.5. Design Using the Bode Diagram

Consider the hovering helicopter example where the transfer function model is given
by
4
g(s) =
s2

The Bode Diagram for this open loop process is:

Figure 13.7 Bode Plot for the Helicopter

To improve on the Phase Margin, the phase plot at 2 [rad/s] must be moved upward.
176 Control Engineering - 1

A lead circuit can do this as shown by its Bode diagram. Set α=4 to give a Phase
advance of 37° and chose T=0.25 so that this phase advance peaks at frequency of ω
=2 [rad/s].

The Bode diagrams for the compensated system, q(jω) = g(jω) k(jω), are then:

Figure 13.8 Bode Plot for Compensated Helicopter

Clearly a smaller value for T (i.e. a larger value for ωcm) might give better results for
the same value of α in the compensator k(s). The design for k(s) would proceed
along these lines until a satisfactory result was obtained.

It is useful to compare this example with the identical problem analysed previously
in the Nyquist plot in terms of the ease with which:

(1) the Bode shows that T should be changed,

(2) the Nyquist plot shows the closed loop gains.

13.1.6. Variable Gains, f, in Bode Diagrams

The effect that a gain factor, f, has on the design can be determined in the Bode
diagram by shifting the horizontal frequency axis in the gain plot vertically
downwards by
20Log10 f [dB]

(This is equivalent to moving the graph up by 20Log10 f as required).

Ch 13: Alternative Frequency Response Plots 177

Thus, by inspection of the Bode diagrams for q(jω) in the above example, it is clear
that reducing the gain will improve the phase margin of q(jω) and hence the
damping factor of the C/L system, though at the cost of reducing the bandwidth.

13.1.7. Non-Minimal Phase Systems (Bode Diagrams)

A system with a zero or a pole in the right-half s-plane is known as a non-minimal

phase system.

Bode Diagrams CANNOT be used to test for stability of such systems as the results
are invalid. For these systems the Nyquist stability criterion MUST be used.

Example

Consider the open loop process model

1
q(s) =
1− s

which has an unstable pole at s=+1 and is thus non-minimal phase. In closed loop
this system is unstable for all gains, f ≥ 0, as can be seen using the Nyquist stability
criterion.

Figure 13.9 Nyquist Plot for O/L Unstable Process

From the open loop system

178 Control Engineering - 1

<P> = po = No of O/L unstable poles

=1

From the Nyquist plot

<N> = No of likewise encirclements of critical point
=0

Hence the Nyquist criterion

<N> = -po

fails and the system in closed loop will be unstable.

On the Bode Diagram however:

Figure 13.10 Bode Plot for O/L Unstable Process

the phase angle never reaches -180° and so the closed loop system appears to be
stable.

Note

As the name implies, non-minimal phase systems have the same amplitude as
minimal phase equivalents but their phase angle is larger. To illustrate this consider
the Bode Diagram:
Ch 13: Alternative Frequency Response Plots 179

Figure 13.11 Bode Plot for O/L Stable Process

1 1
for q(s) = and compare this with the Bode Diagram of q(s) = . The
1+ s 1− s
magnitude is identical and the phase angle is less (minimal, in fact).

13.2. The Nichols Chart

For the closed loop system:

Figure 13.12 Closed Loop Control Configuration

the open loop frequency response model is given by the product

q(jω) = g(jω) k(jω)

These quantities are complex numbers, so this multiplicative combination of g(jω)

and k(jω) can be simplified by expressing the complex numbers in polar co-
ordinates as
<20Log10[Magnitude]> and <Phase angle>

Then the multiplicative relationship between q and g-k is equivalent to an additive

180 Control Engineering - 1

operation. For magnitude

20Log10 q( jω ) = 20Log10 g( jω ) + 20 Log10 k( jω )

and for phase

Angle[q(jω)] = Angle[g(jω)] + Angle[k(jω)]

The NICHOLS CHART plots the open loop frequency response models on the
logarithmic axes for complex numbers:

Figure 13.13 Axes of Nichols Chart

and has its origin at (-180° , 0dB). The Nichols Chart origin defines the critical point
where q(jω) has a magnitude of 1 (i.e. 0dB) and a phase angle of -180°. It is
important for Gain and Phase margin evaluation.

13.2.1. Closed Loop Stability (Nichols Chart)

As with Bode Diagrams, the Nichols Charts can be used to test for stability of the
closed loop system from a plot of the open loop transfer function, PROVIDED the
open loop system is stable. The stability test is simply that the closed loop system is
stable if the plot of q(jω) passes to the right of the origin, while it is unstable if q(jω)
passes to the left. These conditions are illustrated graphically in the following
Nichols Charts.
Ch 13: Alternative Frequency Response Plots 181

Figure 13.14 Nichols Charts for Stable and Unstable Systems

13.2.2. Gain and Phase Margins (Nichols Charts)

These are read directly from the Nichols Chart as indicated:

Figure 13.15 Gain and Phase Margins from Nichols Chart

Recall that Phase Margin of the open loop system q(jω) gives an indication of the
damping factor that will be exhibited by the closed loop system h(jω), while the
Gain Margin gives an indication of its sensitivity to modelling errors. (Such errors
usually occur in the model g(jω) assumed for the process.)

13.2.3. Constant M and N Contours (Nichols Charts)

Contours along which Magnitude[h(jω)] and PhaseAngle[h(jω)] of the closed loop

system are constant can be super-imposed on the Nichols Chart. Unlike the Nyquist
Plot these M contours (for Magnitude) and N contours (for Phase) are quite complex
in the Nichols Chart, necessitating the use of special graph paper, or a digital
computer.
182 Control Engineering - 1

Figure 13.16 M and N Contours on a Nichols Chart

13.2.4. Dynamic Compensators (Nichols Chart)

To select suitable compensation elements, k(s), when using Nichols Charts, it is

convenient to know how these will affect the system plot for g(jω).

Lag Circuit

The lag circuit increases low frequency gain and/or decreases high frequency gain
but introduces more phase lag. Its transfer function is
1 + sT
k(s) = K 1 < α ≤ 10
1 + sαT

On the Nichols Chart it produces the following contour:

Figure 13.17 Nichols Chart for Lag Circuit

Ch 13: Alternative Frequency Response Plots 183

Thus for K=1, the lag circuit moves the process transfer function g(jω) down on the
Nichols Chart at high frequencies (which is desired) but at the same time shifts it to
the left by an amount determined by α, especially at ωcm (which is bad).

Lead Circuit

The lead circuit introduces phase advance into the open loop which tends to stabilise
the closed loop system. Unfortunately it also amplifies high frequency noise. Its
transfer function is
1 + sαT
k(s) = K 1 < α ≤ 10
1 + sT

and has the Nichols Chart profile:

Figure 13.18 Nichols Chart for Lead Circuit

Thus with K=1 the lead circuit moves the process model, g(jω) or q(jω), to the right
(which is desired) but also moves it up at high frequencies (which is bad).

13.2.5. Design Using the Nichols Chart

Recall that a good closed loop system has a gain margin greater than 2dB, a phase
margin in the range 30° to 70° and an Mmax in the range 1dB to 3dB. In the Nichols
Chart this defines the regions:
184 Control Engineering - 1

Figure 13.19 Design Regions in the Nichols Chart

Consider once again the helicopter example where

4
g(s) = [m] / [°]
s2

The plot of its frequency response model on a Nichols Chart is:

Figure 13.20 Nichols Chart of the Helicopter

For closed loop stability and a reasonable phase margin (i.e. closed loop damping
factor) the open loop trace should lie to the right of the (0dB , -180°) origin.

As before, a lead circuit which swings to the right in the Nichols Chart would
achieve this. Setting α=4 gives a maximum phase advance of 37° at ωcm, without
introducing too much gain at high frequency. By inspection of the plot of g(jω) this
Ch 13: Alternative Frequency Response Plots 185

phase advance is required at a frequency, ωcm, of approximately 2 [rad/s] and so

1
T= = 0. 25
αω cm

The plot of the resultant open loop frequency response model

q(jω) = g(jω) k(jω)
4 1 + 2s
= 2
s 1 + 0. 25s

becomes

Figure 13.21 Nichols Chart of the Compensated Helicopter

The compensated open loop system is minimal phase and passes to the right of the
(0dB , -180°) origin so the resulting closed loop system will be stable. Also the new
Phase Margin is 31° so the closed loop response should have a damping factor of
0.3 which is an improvement on the open loop response. Its Gain Margin is ∞ and
the gain could be changed to improve on the Phase Margin.

13.3. The Inverse Nyquist Plot

It is sometimes more convenient to use the Inverse Nyquist Plot than the Direct
Nyquist Plot.
186 Control Engineering - 1

Figure 13.22 Closed Loop Configuration for Inverse Nyquist Plots

Once again stability of the typical closed loop system is determined by its
characteristic function
1 + f q(s)

Dividing through by q(s) gives the complex function

1
X(s) = f + 
= f + q(s)
q(s)

N(s)
If q(s) = is a ratio of polynomials and f is a constant (independent of s) then
D(s)
function X(s) becomes
f N(s) + D(s)
X(s) =
N(s)

Clearly the zeros of X(s) are the poles of the closed loop system while the poles of
X(s) are the zeros of the open loop system.

13.3.1. Closed Loop Stability (Inverse Nyquist Plot)

As with the Nyquist diagram, stability of the closed loop system is determined by
traversing the Nyquist contour, D, in the s-plane, evaluating this function X(s) and
counting the number of likewise encirclements, <N>, of the origin of X(s).

Then from the Principle of the Argument

< N >=< Z > − < P >

where <Z> for this X(s) is the number of C/L unstable poles and
<P> for this X(s) is the number of O/L unstable zeros.

The Nyquist criterion then states that the closed loop system is stable iff <Z>=0, or
Ch 13: Alternative Frequency Response Plots 187

alternatively iff
< N >= − < P >= − p 0

where p0 is the number of unstable open loop zeros.

The INVERSE NYQUIST PLOT is a plot of the inverse frequency response model
q ( jω ) on a polar plot. Hence the critical point (-f , 0) in the q - plane is the origin
of the function X(s) and the encirclements <N> are those that q ( jω ) make of the
critical (-f , 0) point.

Example

Determine the gains, f, that give a stable closed loop system for the open loop model
2
q(s) =
1 + 3s

which has no zeros and one pole at s = −1 / 3 . Plotting the function

q ( jω ) = 0. 5 + j1. 5ω

as 's' moves around the Nyquist contour D gives the Inverse Nyquist Plot:

Figure 13.23 Inverse Nyquist Plot of 2/(s+3)

Since <P> = <No of unstable O/L zeros> = 0 here, stability of the closed loop
system requires that <N> = 0 which is true for all gains −f < 1 / 2 . Thus f > −0. 5
for stability (which includes all gains in a negative feedback configuration).
188 Control Engineering - 1

13.3.2. Gain and Phase Margins (Inverse Nyquist Diagram)

These are obtained in a similar manner to the Nyquist plot. For example:

Figure 13.24 Gain and Phase Margins in Inverse Nyquist Plots

1 1
Gain Margin = = 20 Log10 ( ) in [dB]
q of qof

Phase Margin = φ in [o]

13.3.3. Constant M and N Contours (Inverse Nyquist Plot)

One of the attractions of the Inverse Nyquist plot is the simple relationship that
exists between the inverse models of the open and the closed loop systems. For the
typical control loop:

Figure 13.25 Typical Feedback Control Loop

the transfer function model for the closed loop system is

q(s)
h(s) =
1 + fq(s)

The inverse relationship is very simply

Ch 13: Alternative Frequency Response Plots 189

h −1 (s) = f + q −1 (s)

Or, by definition of q (s),


h(s) 
= f + q(s)


Then an Inverse Nyquist plot of q (s) gives h(s) by a mere shift in the axis since:

Figure 13.26 Open and Closed Loop on the inverse Nyquist Plot

Constant gain circles (M contours) are defined by

Magnitude[h(jω)] = M

so
 1
Magnitude[h(s)] =
M

Since points on the Inverse Nyquist Plot are q ( jω ) = x + jy the equation for the M
contours is simply
1
(1 + x)2 + y 2 = assuming that f=1
M2

This equation defines a circle

Centre at (x,y) = (-1 , 0) or (-f , 0)
Radius r = 1/M or f/M

Similarly for constant N contours

190 Control Engineering - 1

Angle[h(jω)] = N

so that

Angle[h(s)] = −N

This gives straight lines radiating from the critical point at (-f , 0) in the Inverse
Nyquist plot at angles −N :

Figure 13.27 Constant M and N Contours in the Inverse Nyquist Plot

13.3.4. Dynamic Compensators (Inverse Nyquist Plots)

When designing closed loop systems in the Inverse Nyquist Plane, it is important to
know what lag/lead compensation does to the graphs.

Lag Circuit

This standard compensator is intended to boost low frequency gains or to reduce

high frequency gains (or both). Unfortunately it also introduces undesirable phase
lag into the loop. Its transfer function is
1 + sT
k(s) = K 1 < α ≤ 10
1 + sαT

and its Inverse Nyquist plot is:

Ch 13: Alternative Frequency Response Plots 191

Figure 13.28 Inverse Nyquist Plot for a Lag Circuit

Lead Circuit

The lead circuit improves on the phase margin for q(jω). It does however amplify
high frequency noise and should be used with caution. Its transfer function is
1 + sαT
k(s) = K 1 < α ≤ 10
1 + sT

and its contour on an Inverse Nyquist plot is

Figure 13.29 Inverse Nyquist Plot for a Lead Circuit

192 Control Engineering - 1

13.3.5. Design Using the Inverse Nyquist Plot

For the helicopter example, the nominal process transfer function is

4
g(s) =
s2
so its inverse model is

g ( jω ) = −0. 25 ω 2

The Inverse Nyquist diagram for the uncompensated open loop system is:

Figure 13.30 Inverse Nyquist Plot for the Helicopter

Compensating the helicopter using a lead circuit with α=4 will ensure a phase
advance of φcm=37°. Setting T=0.25, gives this phase advance at a reasonable
frequency of ωcm=2 [rad/s].

Thus in feedback, without compensation, it would be marginally stable. The trace

for q ( jω ) passes through the point in the q - plane where M=∞ at ω=2[rad/s] so the
closed loop system is expected to oscillate with a period of 3.14[s].

The new Inverse Nyquist Plot for the compensated open loop system is:
Ch 13: Alternative Frequency Response Plots 193

Figure 13.31 Inverse Nyquist Plot for the Compensated Helicopter

Clearly its closed loop is stable as it does not encircle the critical point (-f , 0). Its
damping factor exceeds 0.3 (as the Phase Margin is over 30°) and is assumed
adequate. The overshoot to a step in the setpoint is expected to be about 45[%].

Further adjustments in the compensator parameters should be made to optimize on

its design values.
 14
Summary of Linear Design
Methods
For the typical unity feedback control loop:

Figure 14.1 Closed Loop Block Diagrams

Control element k(s) affects the open loop model q(s) directly and linearly, since
q(s) = g(s) k(s)

and the closed loop model h(s) indirectly and non-linearly, since

h(s) = {1 + q(s)}−1 q(s)

= {1 + gk(s)}−1 gk(s)

where q(s) = gk(s) = g(s) k(s).

The theoretical methods developed here all represent the open loop system model
q(s) in one form or another, with the express purpose of giving the designer insight
into the design problem. This allows systematic selection of optimal cascade
compensation elements, k(s). In addition the representations of q(s) often give
indirect indications (eg. M-contours) of the effects that the open loop system model
q(s) has on the closed loop system model h(s), thus aiding the design further.

Each design method has its own particular strengths and weaknesses, so it is
important in practice to select that method which relates most closely to the problem
in hand. Features of the various design methods are now summarized.
Ch 14: Summary of Linear Design Methods 195

14.1. Pole-Zero (Transfer Function) Methods, q(s)

14.1.1. Root Locus

Contours of closed loop poles in s-plane as open loop gain is varied from 0 to +∞.

• Good for complicated structures with many poles and zeros.

• Clear indication of closed loop response characteristics.
• Has difficulty dealing with deadtime (e-sT) terms.
• Non-dominant poles not hidden.

14.1.2. Characteristic Locus

Contours of closed loop poles in s-plane as some open loop parameter, other than
gain, is varied from 0 to +∞. (N.B. Rules for drawing root loci and characteristic
loci with gain variations from 0 to -∞ also exist.)

• Check effects of parameter variations.

• Same comments as for Root Locus.

ω)
14.2. Frequency Response Methods, q(jω

14.2.1. Bode Diagrams

Two cartesian plots of: 20Log10 q( jω ) versus Log10 ω and

PhaseAngle[q(jω)] versus Log10 ω

• Easy combination of cascaded elements.

• Gain and Phase margins clearly shown.
• Closed loop stability only for non-minimal phase systems.
• Open loop gain clearly shown.
• No indication of closed loop characteristics.
196 Control Engineering - 1

14.2.2. Nichols Chart

Cartesian plot of: 20Log10 q( jω ) versus PhaseAngle[q(jω)]

• Direct access to Gain and Phase margins.

• Closed loop gain and phase loci possible.
• Easy combination of cascaded elements.

14.2.3. Nyquist Plot

Polar plot of: vector q(jω).

• Unambiquous closed loop stability criterion.

• Easy access to Gain and Phase margins.
• Closed loop gain and phase contours possible.
• Sensitivity of closed loop system indicated, including disturbances.
• Effect of open loop gain from critical (-1/f , 0) point.

14.2.4. Inverse Nyquist Plot

Polar plot of: vector 1/q(jω) = q ( jω ) .

• Comments as for Nyquist plot, but critical point at (-f , 0).

• Direct relationship between open and closed loop systems.

14.2.5. Arithmetic Plot

Two cartesian plots of: q( jω ) versus ω and

PhaseAngle[q(jω)] versus ω

• Arithmetic version of the Bode plot.

14.2.6. Rutherford-Aikman Plot

Two cartesian plots of:  jω ) versus Log10 (2 π / ω ) and

Log10 q(
PhaseAngle[q(jω)] versus Log10 (2 π / ω )

• Inverse Bode diagrams.

Ch 14: Summary of Linear Design Methods 197

14.3. Comment
(1) Not all control loops found in practice fit exactly into the form, like
unity feedback, required by specific design methods. Thus it is
encumbent on the engineer to modify the actual loop configuration for
analysis, if not implementation, until it fits that handled by the
proposed design method. Loop modification is easily achieved by
Block Diagram Algebra. Obviously the practical implications of such
alterations must be considered carefully.

(2) Alternative contours in the s-plane can be used to draw different Polar
plots for q(s) which may highlight other properties of the closed loop
system than merely stability. For example, in engineering applications
it is often desirable to contain the poles of the closed loop system
within the area of the s-plane shown shaded in:

Figure 14.2 An Alternative Contour in the s-plane

A polar plot of q(s) as s traverses this contour is used to determine

whether or not the poles of the closed loop system lie within the
desired shaded region.
 15
Compensation
Techniques
Combinations of simple Lead and Lag circuits mentioned previously may not be
adequate in some situations. Other well-known compensators are now considered:

(1) The Proportional-Integral-Derivative Controller.

(2) Minor-loop compensation.
(3) Inverse response compensator.
(4) General compensator.
(5) Feedforward compensation.

15.1. The PID or Three-term Controller

The Proportional-Integral-Derivative (PID) controller is a very popular form of
Lead-Lag compensator that is used extensively in process control applications as a
cascade element, k(s) in the loop:

Figure 15.1 PID Compensation

Ch 15: Compensation Techniques 199

The controller output, u(t), is given by:

Note

(1) The proportional term is not scaled.

(2) The integral term is scaled by a constant I in [time units]. To remove
or minimize integral action the I-term should be as large as possible.
(3) The derivative term is scaled by a constant D in [time-units]. For no
derivative action the D-term is set to zero.
(4) The gain, K, is applied to the overall result.

In theory the PID controller is defined by the transfer function

1
k(s) = K{1 + + sD}
sI

1 + sI + s2 DI
= K{ }
sI

Thus the PID controller puts a single pole at s=0 and two zeros at arbitrary positions
anywhere in the s-plane. The pole ensures Type 1 behaviour while the zeros attract
undesirable root loci towards the left in the s-plane.

In practice an additional non-dominant pole is included to ensure causality. For

example

1 + sI + s2 DI
k(s) = K{ }
sI(1 + sT)

Wide-spread use of PI controllers in industry implies that numerous processes can

200 Control Engineering - 1

be approximated by a dominant second order differential equation/model, and that

most processes are required to track setpoints that step to constant values.

Four versions of this popular control algorithm are commonly available in electronic
hardware. These are:

(1) Proportional-Only control k(s)=K

(2) Proportional-Integral control k(s)=K[1+{1/sI}]
(3) Proportional-Derivative control k(s)=K[1+sD]
(4) PID control k(s)=K[1+{1/sI}+sD]

Note

(1) Commercial units often provide each term in the controller as an

option, and charge accordingly.

(2) The derivative term is usually implemented approximately as a Lead

circuit. This term is also moved to act on the measurement y(s) alone
rather than the error e(s) in order to avoid Derivative bumps from the
setpoint r(s) on the plant input u(s).

(3) Should a non-zero error persist because the plant input has reached its
limits then the integral term grows without bound. Large values for
integral action cause the plant input to remain at its physical limit long
after the error has changed sign (indicating that a smaller plant input is
required). This induces excessive oscillatory behaviour in the loop that
can easily be avoided. Good PID controllers are designed to prevent
such reset wind-up in the integral term by limiting it to a value that
represents full-scale for the plant input u(s).

(4) The controller signals, <SP, PV and CV>, are often 4-20[mA] or
0-10[V], though many other possibilities exist. The requirements for a
particular application must be assessed to ensure that the correct
hardware is ordered.

The effect that a PID compensator has on a process model g(jω) can be estimated by
the following approximate reasoning based on a few critical points from the
frequency response model k(jω) for the PID controller. (Similarities between the PI
control and a Lag circuit plot at high frequencies, and between the PD control and a
Lead circuit plot at low frequencies should also be noted.)
Ch 15: Compensation Techniques 201

Figure 15.2 Effect of PI, PD and PID Controllers

In industry the PID control algorithm is often installed in a loop and its parameters
found by trial-and-error adjustments (known as TUNING the controller). Such
methods for setting controller parameters appear adequate on the surface but give no
indication of how sensitive the closed loop system is to the inevitable changes in the
process model, g(s), or to measurement noise. Also, should the design call for a
complex compensator then a tuning procedure results in a sub-optimal design, and
in severe cases may never converge to a stable closed loop system.

General comment:

(1) The I-term is introduced to increase the low frequency gain of the
open loop system model q(s) and hence ensure that the closed loop
model h(s) tracks setpoints accurately (Type 1 control).

(2) The D-terms is intended to stabilise the loop by improving its phase
margin.
202 Control Engineering - 1

15.2. Design of PID Compensators

Example 1 PD Compensation

Consider a chemical process modelled by a transfer function

4
g(s) =
s2

Its Nyquist plot is:

Figure 15.3 Nyquist Plot for Chemical Process Model

Notice from Fig 15.2 that a PD controller introduces a 45° phase advance at a
frequency of ω=1/D [rad/s]. Applied to the chemical plant this would improve the
phase margin of the combined open loop system q(s), and hence on the damping
factor of the closed loop system h(s).

To design the PD parameters, start by setting D=1 which would give a phase
advance of 45° at a frequency of ω=1 rad/s, and hence more phase advance at higher
frequencies (specifically at ω=2 rad/s where it is required). The PD compensator
transfer function is
k(s) = K(1+s) In theory
or
1+ s
k(s) = K In practice
1 + 0. 01s
Ch 15: Compensation Techniques 203

The Nyquist plot of the new q(s) is shown in the following figure with both the
theoretical and the practical k(s).

Figure 15.4 Compensated Chemical Process Model

The Phase margin of 75° achieved by this PD compensation is above the value of
70° considered acceptable for mechanical systems, and well above the 30° value
recommended for many chemical processes. Since the D term is prone to noise
problems, it is advisable to use as small a value for D as possible. With this in mind,
the D-value could be reduced thereby minimising any noise problem while still
maintaining an adequate Phase Margin. In an actual design the parameter D would
be varied until a suitable compromise was reached between closed loop damping
factor (indicated by open loop phase margin) and noise amplification (caused by the
D term).

Example 2 PI Compensator

Consider a process modelled simply by

1
g(s) =
1+ s

Its Inverse Nyquist plot is:

204 Control Engineering - 1

Figure 15.5 Inverse Nyquist Plot of Process Model

At steady state (ω=0) the open loop model q(j0) lies on the M=0.5 circle in the
Inverse Nyquist plot. Thus the closed loop system has a gain of h = 0. 5 and the
process output, y, at steady state is given by
y = 0.5 r

Clearly it will not track its setpoint, r, exactly.

A PI controller can improve on this situation. Here the transfer function for the
compensator is
1 + sI
k(s) = K
sI

Its inverse frequency response model is

 jω ) = jω I
k(
K(1 + jω I)

Start by setting I=1 [sec] (Since g(s) has a phase lag of 45° at ω=1 [rad/s] and a PI
controller with I=1[s] introduces another 45°, this gives a reasonable total phase lag
of 90° in the open loop model). The inverse Nyquist plot for q(s) becomes:
Ch 15: Compensation Techniques 205

Figure 15.6 Compensated Process Model


From the plot of q(s) 
= k(s)  , it is clear that:
g(s)

(1) the closed loop system h(s) is stable,

(2) the Phase margin for q(s) is 90° so h(s) is damped,
(3) h = 1 at steady state (ω=0), for exact setpoint tracking.

Note that a slightly smaller value of I would cause q(jω) to bend to the left and so lie
closer to the M=1 circle for low frequencies, thus improving on the design.

Figure 15.7 Second Design for Process Model

(This does however have the disadvantage of increasing the sensitivity of the design
206 Control Engineering - 1

to changes in model g(s) and to external signals, d(t), disturbing the output, y(t) as
discussed in Chapter 16.)

15.3. Minor Loop Compensation

In some situations, notably position servos and process control, minor loops are
used to improve the response of an overall closed loop system. A typical system is:

Figure 15.8 Minor Loop Compensation

From block diagram analysis

y(s)
h(s) =
r(s)
k g1 g2
=
1 + g1f + k g1 g2

In process control the first block g1(s) is usually faster than the second block g2(s).
The function of the minor loop is then to ensure quick elimination of any
disturbances to input u2(s).

In servo-mechanisms, the function of the minor loop is usually to improve on the

dynamic characteristics of the overall closed loop system, h(s). It does this by
shifting one of the poles of the open loop system, which comprises the inner closed
loop in cascade with g2(s).

Example

A well-known extremely useful minor loop compensation in position control servos

is velocity feedback from a tachogenerator.
Ch 15: Compensation Techniques 207

Figure 15.9 Servo Position Control

Here a tachometer measures speed and yields the transfer function model
1 Ω (s)
g1 (s) = =
B + sJ E(s)

while position measurement gives

1 θ (s)
g2 (s) = =
s Ω (s)

where Ω(s) is angular velocity and θ(s) is angular position.

The transfer function for the closed loop system (including the minor loop feedback)
is then
y(s) k
= h(s) = 2
v(s) s J + (B + f )s + k

where compensator k(s) is a constant gain k.

Comparing this to the transfer function of the standard second order system

ω 2n
s + 2 ζω ns + ω 2n
2

it is obvious that the minor loop, f, directly affects the damping factor, ζ, and hence
characteristics of the closed loop response.

Thus minor loop is a very important configuration for motor position control since
damping factor, ζ, determines (closed loop) pole positions in the s-plane as follows:
208 Control Engineering - 1

Figure 15.10 Lines of Constant Damping

15.4. Inverse Response Compensator

Control of chemical processes has provided some interesting control configurations
for system compensation. Two such schemes are given here with no comment on
their practical viability (such as doubtful pole-zero cancellations, etc.).

15.4.1. Multiplicative Cancellation of Dynamics

This configuration is based on minor loop control. Thus in the loop:

Figure 15.11 Multiplicative Cancellation of Dynamics

the inner loop is given by the transfer function:

Ch 15: Compensation Techniques 209

u(s) k 2 (s)
=
v(s) 1 + k 2 (s)m(s)

which tends to the limit

1
→
m(s)

when control module k2(s) becomes very large.

Now, if m(s) is a model of the plant dynamics, g(s), then the transfer function of the
process (or open loop system) for the outer loop is
y(s) k 2 (s)
= g(s)
v(s) 1 + k 2 (s)m(s)

which tends to the limit

g(s)
→ when k2(s)→∞
m(s)
→1 if m(s) ≈ g(s)

Thus the function of the inner loop is to cancel the plant dynamics (exactly in
theory, approximately in practice). Module, k1(s), then introduces any desired
characteristics for the open loop system of the outer loop to give the specified closed
loop performance.

15.4.2. Additive Cancellation of Dynamics

Dead-time, e − sτ , is a common characteristic of industrial processes where material is

transported around a plant in pipes or on conveyor belts. Unfortunately such
processes are very difficult to regulate by feedback control schemes.

It is strongly recommended in practice that every effort is made to reduce the dead
time term, τ, by altering the circuit itself, rather than to correct the problem by
feedback control. However, when this is not possible, the SMITH-PREDICTOR is
an ingenious scheme that alleviates the problem.

Consider the control loop:

210 Control Engineering - 1

Figure 15.12 Additive Cancellation of Dynamics

The transfer function m(s) is a model of the stable function g(s) that defines plant
dynamics without its dead-time. The output from Model 1 is intended to cancel the
process output exactly. The controller k(s) is designed (e.g. by Nyquist plots) to
control Model 2, the plant model m(s). This model does not contain a troublesome
dead-time term and design of k(s) is generally simple. Any discrepancies between
g(s) and m(s) are taken care of by the signal path through Model 1.

The modern INTERNAL MODEL CONTROL configuration is an extension of the

Smith Predictor control in which k(s) and m(s) are combined into one single
controller block, Q, that is easily designed by open loop methods as h(s)=Q(s)g(s).

Example (Multiplicative Cancellation)

Consider the control loop:

Figure 15.13 Multiplicative Cancellation of Dynamics

For the inverse response model

u(s) 10
=
v(s) 1 + 10 / (1 + s)
Ch 15: Compensation Techniques 211

10(1 + s)
=
11 + s
10
= at low frequencies
11

Consequently the response between v(s) and y(s) becomes

y(s) 10
=
v(s) 11 + s

which has a pole at s = −11 and is much faster than the original plant which had a
pole at s = −1 .

The transfer function, h(s), for the complete control system is then
y(s) 10
=
r(s) s2 + 11s + 10

which has poles at s = −1 and s = −10 .

Without the inverse response loop the system reduces to:

Figure 15.14 Without Multiplicative Cancellation of Dynamics

with closed loop transfer function model

y(s) 1
= h(s) = 2
r(s) s + s +1

which has poles at s = −0. 5 ± j0.87 . This closed loop system is slower than the
open loop plant, and slightly underdamped.

A major advantage of this method is that internal variables from the model are
readily available for use in minor loop controls for the inverse response loop. For
example, the model m(s) for a servo system could easily provide an estimate of the
speed of the motor and the inner inverse response loop could then incorporate
velocity and position feedback WITHOUT the need for a tachogenerator. (Such
212 Control Engineering - 1

designs, making full use of the process model, are studied in STATE SPACE
methods with particular reference to STATE OBSERVER systems).

Example (Smith Predictor)

Consider the control loop

Figure 15.15 Smith Predictor Example

The transfer function for this system is derived as follows

e −2s
y(s) = u(s)
1+ s
and
r(s) − w(s)
u(s) =
s

The main feedback signal is

1 e −2s e −2s
w(s) = u(s) + { − }u(s)
1+ s 1+ s 1+ s
1
= u(s) IFF the process is modelled exactly
1+ s

And so
1 1
u(s) = r(s) − u(s)
s s(1 + s)
Ch 15: Compensation Techniques 213

giving
s +1
u(s) = r(s)
s2 + s + 1

Hence the closed response of the entire system is modelled by

e −2s
y(s) = 2
r(s)
s + s +1

with poles at s = −0. 5 ± j0.87 , implying a slightly oscillatory response.

Without the models, the loop, fitted with the same controller, would be:

Figure 15.16 Without the Smith Predictor

The open loop transfer function is

e −2s
q(s) =
s(1 + s)

and is unstable, as shown by its Nyquist Plot:

Figure 15.17 Nyquist Plot of Dead-Time System

214 Control Engineering - 1

Thus use of the Smith predictor allows a tighter control than would otherwise have
been possible.

15.5. General Compensation

The advent of low-cost computing power allows control schemes to be as
sophisticated as necessary with few constraints (primarily limited by computation
time). Thus, in the general case, where simpler compensation techniques have
failed, the compensator k(s) can be designed to be any realisable, stable transfer
function

a 0 + a1s + a 2s2 +... +a ms m

k(s) = with m ≤ n
b 0 + b1s + b 2s2 +... +b ns n

Poles and zeros of k(s) are often chosen to be in the stable left-half s-plane and the
number of poles must equal (seldom) or exceed (generally) the number of zeros.

Such complex compensators are usually designed using pole-zero configurations

(e.g. Root Locus methods).

Example (Control of Large Structures)

Control of roll angle in an aircraft:

Figure 15.18 Aircraft Roll Control

results in an open loop process model containing primarily a dominant exponential

response with some underdamped oscillations due to vibrations in the airframe.
Typically a step input gives the roll angle response:
Ch 15: Compensation Techniques 215

Figure 15.19 Roll Step Response

The real pole is known as the Roll Mode, while the oscillations are known as Dutch
Roll.

Pole locations for the transfer function g(s) are:

Figure 15.20 Poles and Zeros of Roll Control

A suitable compensator k(s) for the aircraft deals with the two complex poles (the
Dutch roll mode) using two zeros in the same vicinity in the s-plane. Two poles then
need to be included in k(s) to make the compensator realisable. The compensator
probably also speeds up the dominant roll mode and eventually ends up being a
third or higher order transfer function.

The coupled drive system used in many teaching laboratories:

216 Control Engineering - 1

Figure 15.21 Coupled-Drive System

exhibits a response similar to that of the aircraft roll system.

15.6. Feedforward Control

Particularly in process control applications, a disturbance on the plant may be
measurable but unavoidable. For example, the block diagram:

Figure 15.22 Process Disturbance

shows a DISTURBANCE signal, d(s), affecting the process output, y(s).

Simple feedback control acts to reduce the effect of the disturbance since
q g
y(s) = r(s) + d d(s)
1+ q 1+ q

and a good design ensures that q(s) is very large (especially at low frequencies) so
that the second transfer function, between the disturbance and the plant output,
tends to zero.
Ch 15: Compensation Techniques 217

Feedforward control can improve disturbance rejection further, particularly at higher

frequencies. In this loop:

Figure 15.23 Feedforward Controller

the transfer function, kd(s), for the feedforward controller is designed from the
equation
g d (s)
k d (s) =
g(s)

IFF this is realisable (i.e. stable and causal). Otherwise kd(s) should be designed to
approximate the transfer function ratio.

Once the loop is closed, the output response becomes

q g − gk d
y(s) = r(s) + d d(s)
1+ q 1+ q
q g d (s)
→ r(s) + 0 iff k d (s) =
1+ q g(s)

In practice it is important to determine the plant model g(s) as accurately as possible.

The success of feedforward control depends on it.

For an improved response over a system with no feedforward compensation it is

necessary that
g d (s) − g(s)k d (s) < g d (s)
218 Control Engineering - 1

Example (Feedforward compensation)

Consider the simple system:

Figure 15.24 Feedforward Control Problem

In this loop, with feedback compensation only, the response is given by

q g
y(s) = r(s) + d d(s)
1+ q 1+ q

1 2s + 2s2
= r(s) + d(s)
1 + s + s2 2 + 3s + 3s2 + s3

The transfer function between d(s) and y(s) in the closed loop system has a Bode
plot:

Figure 15.25 Disturbance Rejection

Ch 15: Compensation Techniques 219

which shows clearly that the closed loop system AMPLIFIES disturbances, d(s),
around a frequency of 1 rad/sec. In practice this means that the plant operator will
notice a deterioration in the performance of his plant under automatic control (and
will immediately condemn the entire control system).

Installation of a feedforward control scheme can alleviate the problem:

Figure 15.26 Disturbance Rejection

For this system the closed loop performs according to the transfer functions
1
y(s) = r(s) + {≈ 0} d(s)
1 + s + s2

so the effect of the disturbance d(s) is eliminated (at least in theory). In practice
there will be some mismatch between the model used for the design of the
feedforward element and the plant. However feedforward is still likely to improve
on the performance of the overall system. Notice that the cost of the extra
feedforward control is an instrument that measures the disturbance d(s).

15.7. Full Control Configuration

In industrial applications, the process under investigation may be subjected to
external disturbances, d(s), and measurement noise, n(s). (Also operator inputs to
the system might be too lively.)

The block diagram of a typical industrial control problem includes components that
model these effects:
220 Control Engineering - 1

Figure 15.27 Practical Control Problem

The controller required to deal with these problems comprises four modules:

k(s)the feedback controller

f(s) the measurement filter
p(s)the pre-filter (on the setpoint)
kd(s) the feedforward controller

The final loop configuration is shown below:

Figure 15.28 Practical Control Configuration

Briefly, the design objectives for each block are:

k(s) Type Number, Stability, Robustness, Disturbance rejection.

kd(s) Disturbance rejection.
f(s) Filter out n(s) (Non-dominant dynamics).
p(s) Filter out lively setpoints, Shape the overall response.
 16
Sensitivity and
Disturbance Rejection
Models used in the design of control systems are usually nominal dynamic models
whose parameters depend on factors like nonlinearities, wear, aging, temperature
fluctuations, etc. Such uncertainties, ∆g, in the transfer function models, g, are
known as PARASITICS, and could include modelling errors.

Figure 16.1 Parasitics in Control Loops

In any practical design, it is important to know how variations in the parameters of

the nominal model, g, affect the dynamics of the closed loop control system, h. This
is ascertained by sensitivity analysis.

16.1. Sensitivity
Mathematically, the sensitivity function, S(s), of a quantity, h(s), with respect to
variations in parameter, K, is defined by

h ∂h/h ∂ ln[h]
SK = =
∂ K / K ∂ ln[K]
Fractional change in h(s)
=
Fractional change in K
222 Control Engineering - 1

Note that the sensitivity function S(s) is frequency-dependent. Obviously a good

robust design would aim to achieve a small value for the sensitivity function. This is
usually possible at low frequencies, but becomes less feasible as frequency
increases.

16.1.1. Algebra for Sensitivity Functions

Sensitivity functions are combined by the chain rule. For example, the sensitivity of
h with respect to K is given by
h
SK = Sqh Sqg SgK

when h is a function of q, q is a function of g and g is a function of K.

16.1.2. Application to Feedback Control Loops

An important application of this theory is in the sensitivity analysis of the closed

loop transfer function, h(s), to changes in the open loop transfer function, q(s). Here
q(s)
h(s) = = h(q, s)
1 + q(s)

and so, differentiating with respect to q,

∂h 1 q(s)
(s) = −
∂q 1 + q(s) {1 + q(s)}2

1
=
{1 + q(s)}2

The sensitivity function for h with respect to q is thus

∂h/h ∂h q
Sqh = =
∂q/q ∂q h
1
=
1 + q(s)
Ch 16: Sensitivity and Disturbance Rejection 223

16.1.3. Representation of Sensitivity on Nyquist Plots

The sensitivity function Sqh depends on the open loop model, q(s), only, so it is
possible to define a region in the q-plane or Nyquist diagram for which

Sqh > 1

indicating a sensitive design. This criterion defines a disc in the Nyquist plot, of
radius (1/f), centred on the critical point (-1/f , 0).

The formula for the disc is easily derived from the definition of Sqh since

1
Sqh = ≤β (where β is a set constant. E.g. β=0.2)
1 + q( jω )

becomes
1
(1 + x)2 + y 2 ≥ when q(jω) = x+jy
ß2

The mathematical boundary between sensitive and insensitive systems occurs when
ß=1. Graphically this defines the following circle (and disk) in the Nyquist plane.

Figure 16.2 Sensitivity Circle in the Nyquist Plot

Obviously good designs will attempt to avoid entering the disc. In practice this is
224 Control Engineering - 1

usually possible for low frequencies, but most designs cannot avoid the disc at high
frequencies.

16.1.4. Representation of Sensitivity on Inverse Nyquist Plots

To indicate regions in the Inverse Nyquist plot where the system is sensitive, the
sensitivity function
1
Sqh = ≤β
1 + q( jω )

can be re-written as

 jω )
q( x + jy
= ≤β
 jω )
1 + q( 1 + x + jy

when q ( jω ) = x + jy . Once again this defines a set of circles on the Inverse Nyquist
Plot. (These are similar to the M-circles found on Nyquist Plots).

In the Inverse Nyquist diagram this inequality for ß<1 defines the region to the right
of the straight line through the point (-f/2 , 0). Thus systems that lie to the right of
this line are robust (i.e. insensitive to modelling errors and disturbances).

Figure 16.3 Sensitivity Region in the Inverse Nyquist Plot

Ch 16: Sensitivity and Disturbance Rejection 225

16.2. Sensitivity to Feedback Elements

In systems with non-unity feedback the closed loop transfer function is
q(s)
h(s) = = h(q, f , s)
1 + f (s)q(s)

Thus sensitivity of the closed loop system, h, to variations in the feedback element,
f, is given by the sensitivity function
fq(s)
Sfh (s) = −
1 + fq(s)
= -1 when fq(s) >> 1

This shows clearly that feedback does not reduce the sensitivity of a closed loop
system to changes in its feedback element f(s). Since f(s) models the measurement of
y(s), it is VITAL to note that the feedback configuration does not enhance the
accuracy of instrumentation (including any mathematical models in this path).

Example (An Application of Sensitivity Analysis)

An electric motor is used to coil steel wire unto a drum. It has a transfer function
Ω(s) 50
= [rad/Volt-sec]
E(s) 1 + 7 s

where Ω(s) is the rotational velocity and E(s) is the input voltage to the motor.

As the steel wire is wound onto the drum, the process time constant changes from
100 milliseconds through 7 seconds to 500 seconds, because it is a function of the
system inertia. (Its nominal value T=7[s] is used extensively in the calculations.)

Using a cascade compensator k(s) = 10, sensitivity of the closed system, h(s), to
variations in the time constant of the open loop model, q(s), is given by
1
Sqh (s) =
1 + q(s)
1 + 7s
= ... (16.1)
501 + 7s

At steady state
226 Control Engineering - 1

1
Sqh ( j0) =
501

so the closed loop response will be minimally affect by changes in the open loop
model q(s), at very low frequencies.

At high frequencies on the other hand

Sqh ( j∞ ) = 1

so changes in q(s) clearly begin to affect the closed loop response.

Closed loop sensitivity to changes in the process time constant are computed from
the sensitivity function

STh = Sqh Sqg SgT

Now, to compute SgT

50
g(s) = with 0.1 ≤ T ≤ 500 [s] and T=7[s] nominally
1 + sT
so
-50s
∂g= ∂T (by differentiation)
(1 + sT)2
and hence
∂g − sT ∂ T
=
g 1 + sT T

Thus, by definition of sensitivity functions,

− sT
SgT = ... (16.2)
1 + sT

For Sqg , the open loop system is

q(s) = g(s) k(s)

so
∂q=k∂g + g∂k=k∂g (since ∂ k = 0)

yielding

Sqg = 1 ... (16.3)

Ch 16: Sensitivity and Disturbance Rejection 227

Combining Equations (16.1), (16.2) and (16.3), the sensitivity of h(s) to variations
in T is computed from the chain rule

STh = Sqh Sqg SgT

1 + sT − sT
= 1
501 + sT 1 + sT
− sT
=
501 + sT

Hence the sensitivity of the closed loop system to changes in the time constant T is
−7 s
STh = for T=7[s], nominally
501 + 7s

The Bode magnitude Plot of this sensitivity function is:

Figure 16.4 Bode Plot for the Sensitivity Function

Note

(1) STh = 0 when ω = 0.

(2) Sensitivity of h(s) to changes in T is reduced for frequencies below the

breakpoint frequency of 71,6 [rad/s] (from 501/7).
(3) Since ∂T/T ≈ 70 (from 500/7 or 7/0.1), the actual changes in h are still
large, as shown by the Bode magnitude plot for the closed loop system
model h(s) at different values of T.
228 Control Engineering - 1

Figure 16.5 Bode Plots for the Closed Loop Transfer Function

Observe that the relative change in the time constant, ∂T/T, is 70. Thus the
sensitivity function STh must be very small (<1/70 or < -36.9[dB]) to ensure that the
relative change in the closed loop system ∂h/h is reasonable (< 1). This occurs for
frequencies below ω=1 [rad/s] as shown in the Bode diagram of the sensitivity
function.

16.3. An Alternative Approach

It is possible to convert the sensitivity analysis problem into an equivalent feedback
stability problem which can be analysed by any of the methods described previously.
The relevant block diagrams are:

(a) Original Loop (b) Re-Arranged Loop

Figure 16.6 Closed Loops for Sensitivity Analysis

If it is assumed that the setpoints in both loops are zero then the equations for the
original loop are
Ch 16: Sensitivity and Disturbance Rejection 229

u = k(s) [r − y] = − k(s) y
y = g(s) u + v
Hence
k(s)
−u = v The Process
1 + kg(s)
Also
v = ∆g(s) u The Controller

Then any design plot (e.g. Nyquist, Root Locus, etc.) of the process shows clearly
the range of controllers (i.e. parasitics, ∆g) that the designed loop can tolerate.

16.4. Sensitivity and the Rejection of Disturbances

In most practical processes, the output, y(t), is dependent on the process input, u(t),
and numerous other variables. The latter are conveniently modelled as a single
disturbance term, d(t), at the output. Thus in Laplace transforms
y(s) = g(s) u(s) + d(s)

Under closed loop control:

Figure 16.7 Closed Loop System with Output Disturbance

the response of the process is given by

q(s) 1
y(s) = r(s) + d(s)
1 + fq(s) 1 + fq(s)

Thus the disturbance d(s) is reduced by the factor:

1
1 + fq(s)

in a feedback configuration.
230 Control Engineering - 1

This factor is exactly the same as the function derived for loop sensitivity, namely,
1
Sqh (s) =
1 + fq(s)

In the Nyquist plot for open loop model q(jω), the region in which the disturbance is
amplified is easily identified by the same circle as before:

Figure 16.8 Disturbance Rejection Disk in the Nyquist Plot

Disturbances are amplified for all frequencies within the shaded disc, so this region
should be avoided.

Example

To illustrate the importance of disturbance rejection consider the control loop:

Ch 16: Sensitivity and Disturbance Rejection 231

Figure 16.9 Process with Disturbance

The Nyquist plot for the open loop transfer function

1
q(s) =
s(1 + s)
is:

Figure 16.10 Nyquist Plot

Assuming that the disturbance is a simple sinusoid of unity amplitude and frequency
d(t) = 1.0 sin t

then the plant response in open loop is

y(t) = 1.0 sin t + terms due to u(t)

Under feedback control

y(t) = 1.4 sin t + terms due to r(t)

Graphically the response to the disturbance would be:

232 Control Engineering - 1

Figure 16.11 Sinusoidal Disturbance

Clearly a disturbance of 1 [rad/s] frequency is amplified by the feedback system.

This deterioration in system performance is noted at the design stage since:

(1) Its Nyquist plot enters the sensitivity disk for frequencies above
0.7[rad/s], and

(2) A disturbance occurs in the process with frequencies in the sensitive

range.

However for ω=0.2 the closed loop improves performance since a disturbance
d(t) = 1.00 sin 0.2t

results in an open loop response of

y(t) = 1.00 sin 0.2t + terms due to u(t)

and a closed loop response of

y(t) = 0.21 sin 0.2t + terms due to r(t)

Once again this information is obtained directly from the Nyquist plot during the
design.

16.5. Sensitivity of Closed Loop Pole Positions

In order to study the sensitivity of closed loop poles to changes in a parameter K, the
sensitivity function
ds
SsK =
dK

is most useful. It defines the rate of change of pole position, s, with respect to the
parameter K and is used in Root Locus studies.

Example

Consider a first order process model

A
g(s) = [m] / [V]
1 + sB
Ch 16: Sensitivity and Disturbance Rejection 233

Its parameters, A and B, were derived from a series of step tests performed on the
plant. The results were
A = 8.1 ± 2.6 [m]/[V] and B = 9.3 ± 1.2 [sec]

So there is some uncertainty in the nominal model

8.1
g(s) = [m] / [V]
1 + 9.3s

Using a PI controller
(1 + s)
k(s) = [V] / [m]
s

the nominal open loop transfer function, q(s), for the process model is
A(1 + s) 8.1(1 + s)
q(s) = = [V] / [V]
s(1 + sB) s(1 + 9.3s)

The characteristic function for the closed loop system is

ϕ c (s) = A + (1 + A)s + Bs2 = 8.1 + 9.1s + 9.3s2 ... (16.4)

and the poles for the nominal closed loop system are at
sp = −0 . 489 ± j0.795 = sp (A, B)

Variations in Parameter A

The sensitivity of these closed loop pole positions, s= s p , to changes in parameter A

is given by the derivative
∂s
(s p )
∂A

It is computed from the closed loop characteristic equation (16.4) to be

∂s ∂s
1 + s + (1 + A) + 2Bs =0 assuming s(A)
∂A ∂A

Re-arranging this equation gives closed loop pole sensitivity with respect to A
234 Control Engineering - 1

∂s − (1 + s)
= SsA =
∂A 1 + A + 2Bs
− (1 + sp )
=
1 + A + 2Bsp

= −0.054 − j0.035 when sp = −0 . 489 − j0. 795

= 0.064 ∠ − 147 ° in polar co-ordinates

Thus the closed loop pole movement is

∂s
∆s p = SsA ∆A = ∆A
∂A
= −0.140 − j0.091 [rad/s] when ∆A = 2.6
= 0.166 ∠ − 147 ° in polar co-ordinates

On the s-plane this uncertainty means that the pole position, p, is given by the range
p = s p ± ∆s p

= −0.629 − j0.886 when ∆A = +2.6

= −0.349 − j0.704 when ∆A = −2.6

Graphically, movement in the nominal pole position at sp = −0 . 489 − j0.795 due to

changes in A is shown as a line of uncertainty in the s-plane

Figure 16.12 Uncertainty in Closed Pole Position due to Parameter A

This indicates that the closed loop system becomes less damped, with a lower
frequency of oscillation, as the open loop gain decreases. The exact extent of the
change can be quantified by the analysis. (Since the poles of a real system are
symmetric about the real axis in the s-plane the results follow immediately for the
other pole s p = −0 . 5 + j0.8 .)
Ch 16: Sensitivity and Disturbance Rejection 235

For interest, actual computation of the closed loop pole positions shows that
s p = −0 . 629 − j0.868 when ∆A = +2.6 (i.e. A = +32% change)
and
s p = −0 . 350 − j0. 685 when ∆A = -2.6 (i.e. A = -32% change)

Variations in Parameter B

Analysis to determine how changes in parameter B affect closed loop pole positions
is similar to the above. Thus only the main results are given below.

From the closed loop characteristic equation

A + (1 + A)s + Bs2 = 8.1 + 9.1s + 9.3s2 = 0

the appropriate sensitivity function is

∂s − s2
=
∂B (1 + A + 2Bs)
= −0.053 − j0.027 when sp = −0 . 489 − j0. 795

= 0.059 ∠ − 153° in polar co-ordinates

And so the uncertainty in the closed loop pole position is

∂s
∆s p = SsB ∆B = ∆B
∂B
= −0 . 063 − j0. 032 [rad/s] when ∆B = 1.2

= 0 . 071 ∠ − 153°

Thus the closed loop pole position moves in the range

p = s p ± ∆s p

= −0.437 − j0.768 when ∆B = +1. 2

= −0.542 − j0.821 when ∆B = −1. 2

Graphically, movement in the nominal pole position at sp = −0 . 489 − j0.795 due to

changes in B is:
236 Control Engineering - 1

Figure 16.13 Uncertainty in Closed Pole Position due to Parameter B

Clearly an increase in the parameter B (∆B = +1.2) produces a response of the

closed loop system that is slower and more oscillatory with a lower frequency of
oscillation than the nominal design. The exact extent of the change can be quantified
by further analysis.

For interest, actual computation of the closed loop pole positions shows that
sp = −0 . 433 − j0.764 when ∆B = +1.2 (i.e. B = +13% change)
and
sp = −0 . 562 − j0.827 when ∆B = -1.2 (i.e. B = -13% change)

Variations in Both Parameters A and B

For simultaneous changes in both parameters, A and B, the expression

∆s p = SsA ∆A + SsB∆B

applies. The magnitude of changes in A and B would identify a region around the
nominal closed loop pole positions in the s-plane:
Ch 16: Sensitivity and Disturbance Rejection 237

Figure 16.14 Uncertainty in Closed Pole Position

These areas indicate the extent to which the closed loop system would be
downgraded by changes in the open loop model.

An Alternative Approach

Recall that the variation of parameters can also be dealt with by Root Locus
methods, as described earlier. Changes in A are nothing more than changes in the
Root Locus Gain, γ, while changes in B define a Characteristic Locus based on the
characteristic equation

1 9. 1
s2 + (A + {1 + A}s) = s2 + (s + 0.11)
B B

Both instances are easily handled by a root locus plotting program.

16.6. General Analysis

The above analysis was carried out on a simple second order example. For the
general transfer function, consider
N(s)
q(s) = γ
D(s)

where γ is the Root Locus gain, and the numerator and denominator polynomials are
defined by:
238 Control Engineering - 1

m
N(s) = ∏ (s − z i ) Open loop zeros, zi
i=1

n
D(s) = ∏ (s − p i ) Open loop poles, pi
i=1

Sensitivity analysis yields three functions. It involves considerable algebraic

manipulation, including differentiation of the above expression with respect to the
variables that change. (Be cautioned that the results can be misleading. For details
refer to the book by Frank,P.M. entitled Introduction to system sensitivity theory.
Academic Press, London, 1978.)

16.6.1. Sensitivity of Closed Loop Poles to Root Locus Gain, α

∂s − N(s)
=
∂γ γ ∂N +∂D
∂s ∂s

16.6.2. Sensitivity of Closed Loop Poles to Open Loop Poles

∂s 1 D(s)
=−
∂ pi s − pi γ ∂ N ∂D
+
∂s ∂s

16.6.3. Sensitivity of Closed Loop Poles to Open Loop Zeros

∂s γ N(s)
=−
∂ zi s − zi γ ∂ N + ∂ D
∂s ∂s

It is left an an exercise in mathematics to check that these formulae are correct.

In practice, with the wide-spread availability of digital computers, it is probably

much easier to use Monte Carlo methods based on a simple root-finding program
that computes closed loop pole positions for random values of A and B in their
given ranges.
 17
Electronic Circuitry from
Transfer Functions
Control engineering theory involves the design of a suitable controller transfer
function, k(s), given a process transfer function model, g(s), and some design
criteria (e.g. Ramp-tracking, Disturbance rejection, Maximum input, etc.).
Extraction of g(s) from experimental data forms a part of control engineering theory,
but implementation of k(s) is not often discussed in depth in control literature,
probably due to the large variety of formats that k(s) can take in practice (e.g.
Pneumatic, Hydraulic, Mechanical, Electro-Mechanical, etc.) In Electrical
Engineering, it is assumed that k(s) is implemented in electronic hardware; either as
an analog voltage circuit (in this text), or as a digital computer program.

17.1. Two Basic Analog Electronic Building Blocks

Any controller transfer function (or process transfer function without a dead-time,
for an analog simulator) of the form

a o + a1s + a 2s2 + ... + a ms m

with m ≤ n
b o + b1s + b 2s2 + ... + b ns n

can be transformed to an (infinite number of) electronic circuit(s, all behaving

identically; at least in theory). Implementations of k(s) in analog electronics are
based on suitable combinations of two basic analog electronic circuits which are
given below.

17.1.1. Summation (of Voltages)

Assuming that all signals remain within a range defined by the supply voltage, Vs ,
the operational amplifier circuit:
240 Control Engineering - 1

Figure 17.1 An OpAmp Summer

behaves in such a way that the output voltage is a negated sum of all input voltages,
individually scaled. (For clarity this circuit shows two inputs, e1 and e2, only.)
Mathematically this means that
R R R
e0 = −[ e1 + e2 + e3 + ... ]
R1 R2 R3

= −[ c1e1 + c2e2 + c3e3 + ... ]

The equivalent block diagram for the (two-input) circuit is thus:

Figure 17.2 Block Diagram for the OpAmp Summer

As many additional inputs can be included in the circuit as necessary (though large
systems must remain electronically sound.). Note that a control study defines only
constants, c1,c2,..., so the exact choice of resistor values, R, R1,R2,... depends on
electronic criteria (e.g. OpAmp input impedance, Leakage current, etc.).

17.1.2. Integration (of Voltages)

The operational amplifier circuit:

Figure 17.3 An OpAmp Integrator

acts as an integrating circuit in that the output voltage is a negated integral of the
sum of all the input voltages, individually scaled. (Once again a two-input circuit is
Ch 17: Electronic Circuitry from Transfer Functions 241

shown, with inputs e1 and e2.) Mathematically this means that

ò ò ò
1 1 1
e0 = − [ e1dt + e2 dt + e3dt + ...]
R1C R 2C R 3C

c1 c c
= −[ e1 + 2 e 2 + 3 e3 + ...]
s s s

The equivalent block diagram for the (two-input) circuit is thus:

Figure 17.4 Block Diagram for an OpAmp Integrator

Control theory provides the constant values c1,c2,..., so the exact choice of capacitor
value, C, and resistor values, R1,R2,... comes from electronic considerations.

This circuit is more difficult to deal with than the summation circuit since it includes
a capacitor that can retain a charge. Practical circuits must take this into
consideration. In control applications, the capacitors need to have correct initial
voltages on them to prevent unforeseen bumps to the system when the controller is
switched in. Thus a practical circuit contains numerous switches in addition to the
basic circuit shown above. In cheaper controllers an extremely large resistor is often
placed across the capacitor to leak off excess charge. This saves on switches, but
may cause a bumpy transfer of control that would not be tolerated on industrial
plants. In fact control theory and electronics are only the beginning of the design
and considerable effort has to go into commissioning the system to make it
acceptable to all concerned under operational conditions.

17.2. Electronic Circuit for a Comparator

To illustrate how control theory (Block Diagram algebra) is used to derive a suitable
OpAmp controller based on the above two circuits, consider the function of a
comparator normally associated with feedback control
e = r − y

Its block diagram is simply:

242 Control Engineering - 1

Figure 17.5 Block Diagram for a Comparator

As it stands this diagram does not link directly with the above electronic circuits (i.e.
There is no one-to-one mapping between this diagram and OpAmp circuits).
However skilful use of block diagram algebra can convert the original comparator
circuit to a block diagram that is more in line with those for Summation and
Integration in OpAmp circuits . (It is not clear exactly what-to-do-when, so a little
experimentation is essential during this block-shuffling exercise. There are many
possibilities so the chances of success are good.)

Figure 17.6 Another Block Diagram for the Comparator

This block diagram uses an OpAmp that sums two signals, +r and -y, to give -e (It
has been devised with extensive reference to the OpAmp summer circuit). To obtain
signal -y from +y and signal +e from -e, two single-input summers are needed. The
final block diagram result has exactly the same transfer characteristics as the original
comparator. Thus the electronic circuit requires three operational amplifiers, as
indicated, to mimic the comparator (providing all the signals remain within the
power rails.).

If desired, block diagram algebra can reduce this further to:

Figure 17.7 Simplified Block Diagram for the Comparator

Ch 17: Electronic Circuitry from Transfer Functions 243

One more simplification is possible in practice. Remove the operational amplifier on

the setpoint signal, +r, and use -r rather than +r as the comparator setpoint (which
might not always be an acceptable alternative).

Such reductions in circuit complexity may not be worthwhile from an economic

point of view, especially for small production runs. (Note that there may be better
electronic designs that do the same job.)

17.3. Electronic Circuit for a Transfer Function

Consider a transfer function

a o + a1s + a 2s2 + ... + a ms m

k(s) =
b o + b1s + b 2s2 + ... + b ns n

with m ≤ n, and real constants ai and bi.

First cancel all common factors (i.e. Equal poles and zeros) between the numerator
and denominator polynomial. Then consider eliminating all factors in which poles
and zeros are nearly equal (Using the principle of Pole-Zero cancellation) and also
non-dominant poles (The principle of Pole dominance). Such preparation will
simplify the electronic circuit and may have little or no effect on the performance of
the final control system; but check that this is a valid assumption and ensure
causality (i.e. m ≤ n) in the final transfer function.

17.3.1. A Second Order Example

It is easiest to illustrate how a transfer function k(s) is transformed to an electronic

circuit by considering a simple second-order example in which m = n = 2. (Larger
more complex systems are handled in exactly the same way. Just expand the recipe.)

Consider

a' o + a' 1 s + a' 2 s2

k(s) =
b o + b1s + b 2s2

and assume that k(s) is in its simplest possible form (i.e. all common factors have
been cancelled, etc.).
244 Control Engineering - 1

(a) If m=n (as it does here) then divide the numerator by the denominator to give
a o + a1s a' 2
k(s) = +
b o + b1s + b 2s2 b2
or
a o + a1s u(s)
k(s) = +α =
b o + b1s + b 2s 2 e(s)

Note that the controller output, u(s), can be written as the sum of two terms
a o + a1s
u(s) = e(s) + α e(s) = u1 (s) + u 2 (s)
b o + b1s + b 2s2

where the second term is readily implemented as a straight-forward gain term in an

OpAmp circuit.

To continue, concentrate on the first term alone, and execute the following
procedure (which converts k(s) from a transfer function to a state-space model, but
can be regarded as a recipe required to find OpAmps circuits. Although it appears
complicated, it does produce the simplest electronics most easily. Also, state space
theory is useful in manipulating the design further, if necessary).

(b) If m<n then multiply both denominator and numerator of k(s) by an internal
variable, w(s), as follows
u(s) a o + a1s w(s)
k(s) = = ∗
e(s) b o + b1s + b 2s2 w(s)

From the denominators deduce that

b1 b 1
s[sw(s)] = − [sw(s)] − 0 [w(s)] + e(s) = s2 w(s)
b2 b2 b2

s[w(s)] = [sw(s)] ... (17.1)

Note that there are exactly n (=2) equations here. These define the relationships
between two internal variables, [w(s)] and [sw(s)], and the transfer function input,
e(s). The second equation looks a bit odd but is very simply stating that the
derivative of variable [w(s)] is the variable [sw(s)] (Obviously). The brackets [ ] are
used to indicate an electronic signal, wherever this might be unclear. In this
example, signals [w(s)] and [sw(s)] are two important variables of the electronic
circuit and are known as state variables [x1(s)] and [x2(s)].
Ch 17: Electronic Circuitry from Transfer Functions 245

From the numerator it can be concluded that

u(s) = a1 [sw(s)] + a 0 [w(s)] ... (17.2)

The electronic circuit is then constructed from a set of integrators (and invertors,
where necessary) since

ò
[sw(s)] = s[sw(s)]dt
and

ò
[w(s)] = s[w(s)]dt (mixing mathematical notations)

The resulting block diagram is readily developed directly from the above Equation
17.1. Start with the input [s2w(s)] to an integrating block:

Figure 17.8 Starting the Circuit

Then integrate [s2w] to give [sw] and finally [w]. Finally, connect the signal lines:

Figure 17.9 First Part of the Circuit

To complete the transfer function collect the necessary terms, [sw] and [w] (and if
m=n, then [e] as well), to form the transfer function output, according to Equation
17.2. The required block diagram is:
246 Control Engineering - 1

Figure 17.10 Second Part of the Circuit

In practical circuits it is extremely important to ensure that all the electronic signals,
especially [sw] and [w], remain within the range set by the OpAmp power rails. The
Final Value Theorem, Initial Value Theorem and digital simulation can assist in
finding extreme values for the circuit signals, using the transfer functions:
u(s) a o + a1s
=
e(s) b o + b1s + b 2s2

w(s) 1
=
e(s) b o + b1s + b 2s2

[sw(s)] s
=
e(s) b o + b1s + b 2s2

[s2 w(s)] s2
=
e(s) b o + b1s + b 2s2

and assuming an appropriate e(s). (Note that larger n values need more equations.)

If the signals are too big (or too small) then return to the original equations, 17.1
and 17.2, and scale the variables [w] and [sw] by appropriate constant factors.
Substitute these new variables into the equations to generate a second set of
equations
b1 α b0 α
s[ α sw(s)] = − [ α sw(s)] − [ β w(s)] + e(s)
b2 β b2 b2

β
s[β w(s)] = [ α sw(s)]
α

These equations are then converted to a block diagram and checked. Remember this
is a design exercise, so iterate as required until the perfect circuit emerges.

Sometimes the constants, ci, are enormous (e.g. 20x107) and the gains of individual
blocks within the circuit may need additional reduction. Simple block diagram
Ch 17: Electronic Circuitry from Transfer Functions 247

algebra is useful in distributing such gains amongst the system blocks. (The key to
finding a successful circuit is experimentation at the block diagram stage.)

A Numeric Example

Convert the transfer function

3 + 15 s u(s)
k(s) = =
6 + 4s + 2s2 e(s)

to an OpAmp circuit with input, e(s), and output, u(s). Assume that Laplace variable
(s) is in [rad/s].

Sample Solution

(1) Note (a) that m(=1) < n(=2) so that k(s) is realisable, (b) there will be
n(=2) differential equations to implement so n(=2) integrators are needed,
and (c) the transfer function is in its simplest form already; there are no
cancellations nor dominant roots since:

Zero at s = −0. 2 [rad/s]

Poles at s = −1 ± j1. 4 [rad/s]

(2) Convert transfer function k(s) to an equivalent state space model by the
Direct Programming method as follows
u(s) 3 + 15s w(s)
k(s) = = ∗
e(s) 6 + 4s + 2s w(s)
2

From the denominators two differential equations can be deduced

s[sw(s)] = −2[sw(s)] − 3[w(s)] + 0. 5e(s) = s2 w(s)

s[w(s)] = [sw(s)]

From the numerator it is concluded that

u(s) = 15[sw(s)] + 3[w(s)]

248 Control Engineering - 1

(3) Draw the block diagram corresponding to the denominator (differential)

equations:

Figure 17.11 Integration Section of the Circuit

Note that the basic OpAmp circuits, consisting of two integration units
and one summation unit, are indicated on the diagram as thin-lined
rectangular blocks, but could have been omitted until the final stages of
the conversion exercise.

(4) Add the blocks defined by the numerator equation in terms of [sw] and
[w] to form the transfer function output from the internal variables, [sw]
and [w]. (Include signal [e] when m=n.) The resulting block diagram is:

Figure 17.12 Block Diagram for the Final OpAmp Circuit

Note that the circuit consists of two distinct operations. First input, e, is
mapped to internal variables (or states), [sw] and [w], by a set of
integrations. Secondly the states, [sw] and [w], are mapped to the output,
u, by a set of summations. No matter how complex k(s) becomes this is a
standard form; only the number of integrators and summers change. (By
the way, the states are not uniquely defined.)

(5) Study this block diagram closely and try to simplify it by block diagram
algebra to give the final four OpAmp diagram:
Ch 17: Electronic Circuitry from Transfer Functions 249

Figure 17.13 Block Diagram Simplification

(6) Convert each block to an OpAmp circuit and connect its signals to form
the final electronic circuit corresponding to the transfer function k(s). The
resulting electronic circuit is:

Figure 17.14 Final Electronic Circuit

The control study provides the ratios of resistors and capacitors. The
engineer chooses the final component values. In this design:

Set C = 10[ µF ] and it follows from k(s) that

R1 = 50 [kΩ ], R2 = 200 [kΩ ], R3 = 33 [kΩ ], R4 = 100 [kΩ ]

Set R5 = 100 [kΩ ]

Set R6 = 150 [kΩ ] and k(s) gives

R7 = 10 [kΩ ], R8 = 50 [kΩ ]

clearly indicating which values were selected by the user (on electronic
criteria) and which were dictated by the transfer function k(s).

(8) Obtain step response signals for use in commissioning the electronic
circuit. The relevant transfer functions for this design are:
250 Control Engineering - 1

u(s) 3 + 15s
=
e(s) 6 + 4s + 2s2
w(s) 1
=
e(s) 6 + 4s + 2s2
[sw(s)] s
=
e(s) 6 + 4s + 2s2

[s2 w(s)] s2
=
e(s) 6 + 4s + 2s2

Digital Simulation results provide signal shapes and amplitudes that can
be used to determine circuit adjustments that improve on its internal
signals by optimising their amplitudes (both [w(s)] and [sw(s)] are a bit
small, compared to [u(s)], and could be scaled up). Such traces are also
useful for hardware commissioning and acceptance testing.

Figure 17.15 Electronic Test Signals

(7) Finally build k(s) and test its performance THOROUGHLY.

Initial conditions

The addition of initial capacitor voltages is not considered here. (These

can have a major effect on circuit performance.) Possibilities for setting
these initial conditions include (a) shorting each capacitor with a reset
Ch 17: Electronic Circuitry from Transfer Functions 251

button, (b) providing a high-impedance leakage resistor across each

capacitor, or (c) applying an initial voltage to each capacitor. The process
that is to be controlled will dictate which is best.

17.4. Analog Simulators

For large applications it is essential to evaluate the control strategy, including its
electronic circuit, EXHAUSTIVELY before implementing it on the actual process.

An analog simulator for the process can be built from OpAmps by modelling its
transfer function g(s). It should include non-zero inputs, U(t), and outputs, Y(t), to
test the effect of initial conditions on the performance of the controller design.

This electronic circuit combination of g(s) and k(s) forms a convenient and cost-
effective test-bed on which to experiment with the proposed control strategy and its
hardware implementation. Effects such as switch-over from Manual to Automatic
control, down-grading due to fault conditions, the influence of limited signal
amplitudes on overall performance, long-term drift, thermal stability, etc. all need to
be thoroughly investigated. In this way the final product, the electronic controller, is
ruggedised to survive in harsh industrial environments, and the high cost of on-line
experimentation is minimised.
 18
Engineering Applications
Engineering design should aim for elegant simplicity. Thus it is good practice to
start any controller design project by ensuring that the model derived for its process
contains only essential information (without becoming simplistic). The attention
devoted to modelling has a direct impact on the complexity of the final electronic
circuit produced by control theory and is thus important to its design.

Skilful use of such simple dynamic models provides vast amounts of useful
engineering information about a control problem and its solution. Ultimately the
insight gained from these data could mean the difference between success and
failure of the controller design.

18.1. A General Process Control Study

To illustrate how the design theory, that was discussed in these notes, is exploited in
an engineering application, consider a process modelled by
y(s) 4
= g(s) = [mA]/[V]
u(s) 1 + 7s

Specification for the Closed Loop Performance

The problem is to design a controller, k(s), for the unity feedback configuration:

Figure 18.1 Feedback Control Loop

 Ch 18: Engineering Applications 253

so that:

(1) the closed loop system is stable,

(2) its output, y(t), tracks the setpoint, r(t), exactly for step inputs,

(2) its output settles to its new value within 10[s],

(4) its input, u(t), remains in the range ±4 [V] for demanded setpoint
changes of ±1 [mA].

Design

The design is carried out in a number of well-defined stages, following procedures

described previously.

Asymptotic Setpoint Tracking

To track step setpoints with zero error requires a Type Number of 1. Selecting an
appropriate controller form, the open loop transfer function model is
4
q(s) = g(s) k(s) = [mA] / [mA]
s(1 + 7s)

Decide to use Nyquist Plots as the design tool. (Since there is no dead time in q(s)
the Root Locus would also work.) The Nyquist Plot of q(jω) also shows a plot of the
plant model g(jω) for comparison.

Figure 18.2 Nyquist Plots

254 Control Engineering - 1

From the plot of q(jω) it is possible to predict (roughly, but adequately) the
dynamics of the closed loop system (i.e. Type number, Stability, Likely damping
factor, Closed loop bandwidth, etc.). Specifically it is observed that:

(1) The system is Type 1 since it asymptotes up from the negative

imaginary axis. (The original plant model was Type 0 since it
asymptotes from the positive real axis.)

(2) The closed loop system will be stable (since N=0 and p0=0).

(3) The phase margin for q(jω) is small so the closed loop system will be
oscillatory with

• damping factor, ζ ≈ PM[°]/100 ≈ 0.1 (from PM=10°) and

• frequency of oscillation ≈ 0.7 [rad/s] (from M=∞ circle) or
• period of oscillation = 9[s] (from 2π/0.7).

(4) The closed loop is sensitive to parasitics and disturbances for

frequencies ω>0.5[rad/s].

From transfer function analysis, it is deduced that

1 s(1 + 7s)
e(s) = r(s) = r(s)
1 + q(s) s(1 + 7s) + 4
and
k(s) 1 + 7s
u(s) = r(s) = r(s)
1 + q(s) s(1 + 7s) + 4

The Final Value Theorem shows that the steady state error for step inputs, r=1/s, is
zero
s(1 + 7s) 1
e ∞ = lim e(t) = lim se(s) = lim s =0
t →∞ s→ 0 s→ 0 s(1 + 7s) + 4 s

while the Initial Value theorem shows that the initial input is also zero.

So the closed loop system, with this controller in place, tracks any constant setpoint
exactly. Digital simulation of g(s) and k(s)=1/s produced the following closed loop
step response data, confirming the predictions made from the Nyquist plot about its
oscillatory behaviour with a period of 9[s].
 Ch 18: Engineering Applications 255

Figure 18.3 Step Response of First Design

Note that the improved setpoint tracking has been achieved at the cost of stability,
since the closed loop system is much more oscillatory than the open process.

With a bias towards simplicity, it is possible to add either a pole or a zero to k(s). A
zero tends to swing q(jω) in a counter-clockwise direction. This is good since it will
increase the phase margin (and hence the damping factor). From the Nyquist plot it
is observed that the frequency closest to the critical point (-1, j0) is 0.7[rad/s], so a
zero at s=-0.7[rad/s] is a likely choice to start with. (Remember that experimentation
is a critical component of any design exercise.) Thus try adding the factor (1+2s) to
k(s), giving k(s) = (1+2s)/s where 2 is chosen as it is a round number near 1.4[s]
(which was itself computed from 1/0.7).

The resulting Nyquist plot, showing both the previous and the new q(jω), is then:

Figure 18.4 Improved Nyquist Plot

The observation that can be made from the new design are that:

(1) the closed loop will be stable,

256 Control Engineering - 1

(2) its damping factor is now closer to 0.7,

(3) its response time is approximately 4[s] (computed from 4/1).

These predictions are confirmed by the corresponding closed loop step response:

Figure 18.5 Improved Nyquist Plot

Engineering Conclusions

Once the controller is designed, it is useful to extract pertinent engineering

information by varying the controller gain, K, and zero time constant, B, in the
transfer function

k(s) = K(1+sB)/s

about (near) the design values (K=1[°]/[°/s], B=2[s]). In each case deduce important
engineering criteria, such as:

(1) closed loop settling time, Ts[s],

(2) initial input bump, u0 [°],

(3) percentage overshoot, OS[%].

These are three-dimensional functions of K and B, computed from various sources

(e.g. Initial Value Theorem, Nyquist plots, Digital Simulation, etc.) and finally
summarised in the following Engineering Decision Table (which is not complete).
 Ch 18: Engineering Applications 257

B=2 B=5 B=10

Criterion Ts u0 OS Ts u0 OS Ts u0 OS
K=0.1 20 0.5  20 0.5  15 1.0 
K=0.2    12 1.0  5 2.5 
K=0.5  1.0 18 3 2.5 4 2 5.0 0
K=1.0  2.0 15  5.0 4   
K=2.0  4.0 10      

In this table the design trade-offs are summarised as clearly as possible.

For example, it is obvious that an increase in B decreases the overshoot, but requires
a simultaneous decrease in K to preserve the response time and initial input value.
The selection of K=1 and B=2 gives a reasonable design (but the final choice is
subjective, and K=2 with B=2 is also acceptable).

18.2. Non-Linear Systems

The power of linear theory can be further enhanced in practice by using simple
linearization methods on the instruments and actuators of any non-linear process in
order to linearize its behaviour, before analysis by linear g(s)-to-k(s) design
methods. In this manner the techniques given in these notes can be extended to
include a much wider class of control problem.

Should this simple artefact fail to produce an adequate solution then specialised
non-linear control engineering methods must be consulted and applied, or the
process itself should be altered (which in itself is sometimes a very useful
conclusion resulting from an application of control theory to a practical problem).

18.3. Control of a Robot Arm Joint

However, there are instances in which a pragmatic engineering approach produces
satisfactory results without linearization, even when the original process is highly
non-linear.

A robot arm joint driven directly from a motor shaft and operated in a gravitational
field is such an example. It is discussed here because it has many features that
illustrate the subtle difference between an engineering approach (that, in spite of the
risks, might work well and save design time) and a more fundamental scientific
258 Control Engineering - 1

investigation (with detail that, as it happens in this case, is not necessary to solve the
control problem).

Consider the main (shoulder) joint in the robot arm system:

(a) Stable Region (b) Unstable Region

Figure 18.6 Schematic Diagram of the Robot Arm

Clearly the robot arm is inherently stable in the position shown in Figure 18.6a and
unstable in its normal operating position (Figure 18.6b). Thus it needs feedback
control to produce an acceptable final system.

The aim of the controller design is to hold the robot arm at any position within its
operating circle by manipulating voltage to its motor field coils. Its angular position
is measure by a potentiometer.

Joint Transfer Function Model

The model relating <field coil voltage>, u(t), and <joint angle>, y(t), is based on a
block diagram that is derived from elementary laws of physics:
 Ch 18: Engineering Applications 259

Figure 18.7 Robot Arm Block Diagram

In the diagram, the field voltage produces a torque (the K-block). This must
overcome the effects of gravity (the N-block) and possibly drag (the B-block). Any
excess torque accelerates the arm (the J-block), producing angular velocity, Ω, and
hence angular position, θ, measured by a potentiometer (the A-block).

The non-linear block, N, in the diagram is defined by the function

N( θ ) = MgL sin( θ )

where L is the distance from the arm joint to the centre of gravity of the entire arm
system (as shown in Figure 18.6b), M is the equivalent mass of the robot arm and g
is the gravitational constant 9.8 [m/s2].

This non linearity is readily linearized about an operating point, θ0 , to yield

λ ( θ0 ) = Mg L cos ( θ0 )

Thus the dynamic model for the robot arm is

AK y(s)
g(s) = = [V] / [V]
Js2 + Bs + λ u(s)

where the gain term λ is a constant dependent on θ0 .

System Identification

From the block diagram analysis of the process it was noted that
λ ( θ 0 ± π ) = − MgL cos ( θ0 ) = − λ ( θ0 )

Thus the model parameters (J, B, λ) derived for the robot arm in its open loop stable
position (below the horizontal) immediately give the parameters in its unstable
operating region (above the horizontal).
260 Control Engineering - 1

This observation is an extremely valuable in practice. Experiments conducted on the

robot arm below the horizontal position provide the following step test data:

(a) Step Perturbations (b) Step Responses

Figure 18.8 Step Response Test Data

(Note that upward responses are different to downward ones  the arm is highly
non-linear.) From these graphs a set of linear dynamic models were derived for the
robot arm in its stable position of operation. The nominal model (showing standard
deviations as well) for the robot arm in its stable operating region is
(0. 093 ± 0. 010)
g(s) = ... (18.1)
1 + ( 0 . 076 ± 0 . 013 )s + (0. 0111 ± 0. 0012)s2

In its unstable position the model would be simply

− (0. 093 ± 0. 010)
g(s) = ... (18.2)
1 − ( 0 . 076 ± 0 . 013)s − (0. 0111 ± 0. 0012)s2

Controller Design (Root Locus)

It is now assumed that a linear analysis will produce a reasonable feedback

controller that will allow adequate control of the robot arm in any position. (This
assumption may not be true... The arm is obviously highly non-linear, is both stable
and unstable, and has both positive and negative steady state gains. However, it is
worth trying a simple approach before opting for complex non-linear methods.)

Root Loci for the two linearized cases, Equations (18.1) and (18.2), predict similar
 Ch 18: Engineering Applications 261

behaviour at high open loop gains.

(a) Stable Region (b) Unstable Region

Figure 18.9 Root Locus Plot for Process Model g(s)

From these graphs it is concluded that a high gain proportional controller would
stabilise the robot arm both above and below the horizontal. Also, for positions
above the horizontal, the gain must exceed a minimum to ensure stable behaviour.

Experimental observations confirm this prediction, provided that setpoint changes

are kept sufficiently small to conform to the assumptions of the linear analysis.
Large signals result in instability due to excessive overshoot in the closed loop
response. Such behaviour warns of the dangers of too simplistic a linear approach to
non-linear problems, but is not studied further.

Note however that the linear design is not yet complete since the control system can
be much improved by feedback compensation in the form of a simple lead circuit.
The aim is to increase the damping of the controlled system. After a few attempts, in
which designs for g(s) in Equation (18.2) were tested on g(s) in Equation (18.1), the
final design for a robot arm feedback compensator was
50 + 4s
k(s) = ... (18.3)
1 + 0.02s

Sections of the Root Loci for q(s), with gains around the design value, are now more
damped in both linearized cases Also the differences between the Loci for the stable
and the unstable linear models is no longer very significant:
262 Control Engineering - 1

(a) Stable Region (b) Unstable Region

Figure 18.10 Root Locus Plot for Open Loop Model q(s)

These design predictions are confirmed by digital simulation of both situations:

Figure 18.11 Simulated Closed Loop Step Response

Controller Implementation

The controller transfer function is converted to an active electronic circuit based on

simple operational amplifier modules, though it could as easily have been
implemented as a standard passive R-C circuit. The final electronics allows for a
variable gain:
 Ch 18: Engineering Applications 263

Figure 18.12 Compensator Electronic Circuit

Closed Loop Evaluation

When implemented on the robot arm system, the closed loop results are impressive
in that its behaviour is extremely consistent, especially compared to the uncontrolled
robot:

(a) Control Action (b) Setpoint and Response

Figure 18.13 Closed Loop Step Responses

The controlled robot is also very robust, as shown by its reaction to a variety of
setpoint changes and impulse disturbances during normal operation:
264 Control Engineering - 1

(a) Control Action (b) Setpoint and Response

Figure 18.14 Typical Closed Loop Operation

Conclusion

The obvious difference that a simple electronic circuit, designed by a pragmatic

engineering application of linear theory, makes to the behaviour of the robot arm
illustrates clearly the vital role that control theory and electronics play in industrial
applications.
Index
AC motor, 41, 43, 45, 48 119, 135, 156, 160, 163, 167,
Acceleration error, 62, 63, 67 174, 182, 188, 192, 196, 197,
Altitude, 1, 2, 160, 163 201, 203, 208, 212, 213
Amplitude ratio, 126, 128, 129, 151 Compensator, 107, 115, 116, 120,
Analog electronic, 125, 232 122, 152-156, 158, 162, 167,
Application, 1, 9-12, 33, 48, 59, 78, 169, 192, 194-198, 202, 203,
79, 81, 93, 98, 113, 123, 143, 208, 210, 219, 226
192, 216, 234, 243, 248 Complex roots, 70, 72
Argand diagram, 132, 134 Continuous signals, 13
Arithmetic plot, 130, 191 Control law, 12, 13
Armature, 36, 38 Convolution, 15, 18-20, 52, 69
Asymptotic, 69, 244 Correlation, 97
Automate, 1 Critical point, 138, 141, 143, 157-
Automatic, 1, 11, 59, 213, 242 159, 170, 172, 178, 181, 194,
Avionic, 10 217, 246
Bandwidth, 135, 151, 153, 155, 162, Crusher, 44, 45
165, 169, 244 Current, 22, 32, 33, 36, 38, 41-44,
Block diagram, 4-7, 30, 32, 33, 38, 233
41, 43, 44, 45, 48, 49, 60, 66, Damper, 34
87, 97, 98, 140, 188, 192, 201, Damping factor, 79, 80, 109, 112,
221, 233-236, 238-241 114, 119, 144, 148, 153-155,
Bode, 102, 164-172, 189, 191, 192, 161-163, 169, 173, 176, 177,
213, 220, 221 196, 197, 202, 244, 246
Capacitance, 31, 40 DC machine, 4
Capacitor, 31, 234, 242 Dead time, 123
Cascade, 14, 19-22, 26-28, 46, 48, Decision table, 247
104, 115, 121, 165, 188, 192, Density, 40, 123, 124
201, 219, 226 Derivative, 24, 59, 92, 93, 131, 164,
Causality, 16, 193 192-194, 226, 237
Characteristic Equation, 50, 51, 77, Design, 244-246
85, 90, 95, 96, 100, 101, 104, Differential equation, 21-24, 26, 50,
105, 118, 141, 226, 227 51, 78, 102, 128, 131, 193,
Characteristic Function, 50, 76, 78, 240
80, 91, 93, 94, 95, 117, 178 Dirac, 53, 54
Characteristic loci, 118, 119 Direct programming, 240
Chemical process, 1, 196, 197 Distinct roots, 70
Classification, 2 Disturbance, 135, 160, 201, 210,
Co-prime, 84, 90 211, 213-215, 222-225, 232,
Comparator, 8, 48, 98, 100, 235, 236 244
Compensation, 8, 101, 113, 115, Dynamic compensator, 167
266 Control Engineering - 1

Dynamic variable, 25, 26 Initial value theorem, 58, 59, 238,

Electric motor, 2, 4, 5, 14, 43, 219 245
Electronic, 1, 13, 22, 43, 44, 120- Instrument, 13, 214, 219, 248
122, 125, 193, 232, 233, 235- Integral, 15, 18-20, 52, 58, 61, 125,
239, 241-243 164, 192-194, 234
Exponential function, 54 Integration, 22, 24, 61, 233, 235,
Feedback, 4, 8-10, 12, 14, 19, 21, 240
24, 27, 28, 33, 46, 48, 49, 59, Internal model principle, 67
60, 68, 69, 93, 97, 98, 99, 100, Inverse Nyquist, 102, 177-184, 191,
101, 104, 105, 113-115, 119, 197, 198, 218
135, 139, 140, 143, 144, 153, Inverse response compensat, 192
155, 156, 158, 179, 180, 188, Inverse transform, 58, 72
192, 201, 202, 204, 206, 207, Jet, 1, 2
211, 213, 215, 216, 218, 219, Ladder diagram, 3
221, 223-225, 235, 243 Lag circuit, 119, 120, 155, 157, 158,
Feedforward, 155, 192, 210, 211- 160, 167, 168, 174, 175, 182,
215 192
Field coil, 4 Laplace, 24-26, 42, 44, 50-56, 58,
Field voltage, 4-8, 15, 37, 39 65, 69-73, 86, 98, 124, 127,
Field-controlled, 36 129, 130, 222, 239
Final value theorem, 57, 59, 60, 69, Lead circuit, 115, 120, 121, 158-
238, 245 163, 168, 175, 176, 183, 184,
First order, 24, 78, 79, 125 194
Flow, 1, 40, 124, 165 Lever, 1, 35, 160
Frequency of oscillation, 73, 79, Linear, 16, 17, 22, 26, 33, 42, 58,
227, 245 61, 73, 77, 98, 126, 127, 165,
Frequency response, 28, 126-133, 188, 215, 248
135, 136, 142, 149-154, 157, Linearity, 17, 57, 58
162, 164, 165, 171, 172, 176, Logic control, 2, 3
178, 189, 194, 198 Logic signals, 2, 3
Gain margin, 144, 146, 153-156, M circles, 150, 151, 158
158, 160, 162, 176 M contour, 174
Gear, 36 Mapping, 132, 133, 137, 139, 235
General compensat, 122, 192 Mathematical model, 13, 14, 28, 30,
Generator, 5-8, 39, 40, 201 31, 35, 36, 42, 45, 97
Helicopter, 160-162, 168, 176, 177, Minimal phase, 165, 169-171, 177,
183, 184 184, 185, 190
Impulse, 16-18, 53, 74, 75, 78, 80, Minor-loop compensat, 192
81 Missile, 10
Inductance, 31 Mobius transform, 96
Inductor, 31, 33 Modes, 77, 78, 83, 84, 123
Industrial, 2, 3, 5, 12, 30, 44, 122, Motor-generator, 6-8, 39, 40
123, 204 N circles, 152, 181
Industry, 44, 78, 193 N contour, 133, 174
Inertia, 35-37, 41, 219 Nichols, 102, 171-177, 191
Index 267

Noise, 99, 121, 135, 160, 161, 195, 229, 236

197, 214, 244 Position error, 62
Non-minimal phase, 165, 169-171, Potentiometer, 3, 242
185, 190 Power, 1, 3, 4, 12, 41, 45, 50, 90,
Non-unity feedback, 99-101, 218 103, 123, 208, 235, 238, 248
Nonlinear, 41, 42, 45, 98, 215 Predictor-corrector, 22
Nyquist, 102, 130, 135-144, 146- Principle of the argument, 133, 135,
163, 165-170, 174, 177-185, 137, 138, 178
191, 196-198, 205, 208, 217, Programmable logic control, 2, 3
218, 222-225, 244, 245-247 Proportional, 8, 38, 60, 61, 104, 192,
Operational amplifier, 2, 3, 9, 32, 43, 193, 194
44, 100, 233-244 Pulse, 15-17, 28, 54
Oscillator, 90, 107, 160, 168, 194, Pump, 2
228, 245 Ramp function, 53
Overshoot, 80, 148, 149, 162, 247, Reactor, 82, 83
248 Real translation, 56, 124
Pade approximation, 125 Regulate, 10, 165, 204
Paper Winder, 12 Relative stability, 95, 143, 144
Parabolic function, 53, 62 Relay, 2, 3
Parallel, 14, 19-24, 26-28, 33, 46, Repeated roots, 71
77, 78, 86 Residual, 88
Parasitic, 215, 245 Resistor, 5, 6, 30, 233, 234, 242
Partial fraction, 51, 58, 71, 127 Rolling mill, 59, 62
Performance, 1, 12-14, 50, 63-66, Root Locus, 102-106, 108-117, 119,
69, 73, 77, 78, 83, 85, 97, 98, 121, 123, 125, 126, 154, 189,
100, 101, 103, 106, 107, 113, 209, 225, 229, 244
135, 148, 150, 157, 163, 204, Rotor, 160
213, 214, 225, 236, 242, 243 Routh-Hurwitz, 90, 93, 95
Phase angle, 56, 126, 128, 129, 131, Rudder, 1, 12
144, 152, 165, 171, 172 Runge-Kutta, 22
Phase lock loop, 65, 66 Rutherford-Aikman plot, 192
Phase margin, 144, 147-149, 153- s plane, 4
156, 158, 160-163, 165, 168, Second order, 50, 51, 70, 73, 78-80,
169, 172, 174, 176, 182, 183, 144, 148, 149, 193, 202, 229,
190, 195-197, 199, 245, 246 236
PID, 192-196 Sensitive, 144, 160, 195, 217, 218,
PLC, 2, 3 225, 245
Polar plot, 130, 133, 178, 191, 193 Sensitivity function, 215-221, 225,
Pole, 77-80, 82-90, 96, 97, 101-109, 227
112-121, 125, 127, 134, 141, Settling time, 81, 247
149, 157, 164, 169, 170, 179, Ship navigation, 11
185, 189, 193, 198, 202, 206, Signal, 2, 3, 8, 60, 61, 194, 205, 207,
209, 210, 225-230, 236, 246 233, 235-237, 240, 242, 244
Polynomial, 70, 76, 77, 90-92, 95, Sinusoid, 73, 84, 126, 127, 224
96, 100, 104, 125, 127, 178, Sinusoidal function, 84
268 Control Engineering - 1

Smith-predictor, 204 126, 128, 246

Spring, 34 Transistor, 32
Stability, 77, 86, 90, 91, 94, 95, 99, Transmitter, 1
118-120, 133, 135, 137-145, Tuning, 195
158, 160, 165, 166, 169, 170, Type number, 59, 63, 67, 77, 97,
172, 176-179, 184, 185, 190- 215, 244
192, 215, 221, 244, 246 Unity feedback, 99-101, 105, 113,
Stable, 57, 74, 75, 78, 85, 86, 90, 91, 115, 119, 192, 218
94, 95, 106, 113-117, 119, Unstable, 69, 74, 75, 76, 85, 90-93,
122, 127, 128, 138, 140, 155, 114-117, 136, 138-142, 147,
161, 166, 171-173, 177-179, 155, 170, 173, 178, 179, 208
195, 199, 208, 209, 244-246 Valve, 10, 124
State space model, 28, 240 Velocity error, 62, 63, 65
State variable, 238 Weighting function, 16, 18-21, 26,
Steady state gain, 59, 84, 119 28
Steam turbine, 10 Zero, 16, 18, 25, 44, 51, 53, 61-63,
Step function, 44, 52, 59, 62 66, 67, 71-73, 79, 85-88, 90-
Step test, 25, 127, 226 94, 97, 102, 103, 107, 109,
Summation, 18, 70, 233-235, 240 113, 115, 116, 119-121, 125,
Sump, 11, 106, 236 129, 130, 134, 139, 140, 157,
Switch, 2, 234, 242 159, 164, 165, 169, 185, 189,
Tacho, 7, 201 193, 194, 198, 209, 211, 222,
Tank, 1, 2 236, 240, 242, 244-247
Tanker, 1
Target, 10, 63, 113, 114
Temperature, 1, 2, 11, 40, 82, 83, 84,
165, 215, 242
Thermal, 2, 40, 166
Time constant, 37, 38, 78, 79, 105,
154, 158, 198, 219, 220, 221,
247
Torque, 6, 34-38, 41, 42, 49
Transfer function, 26, 28, 33, 37-41,
43, 44, 59-62, 64-67, 69, 76-
79, 81, 83, 86, 90, 94, 97, 100,
101, 103-105, 107, 109, 113,
115, 117, 119, 122, 124-129,
131, 132, 135, 138, 142, 148,
153, 154, 157, 159, 160, 162,
164, 166-168, 172, 174, 175,
180, 182, 183, 189, 193, 196,
198, 202, 204-213, 215, 216,
218, 219, 221, 224, 226, 229,
232, 236-242, 244, 245, 247
Transient, 77, 96, 106, 107, 116,
 269

Bibliography
Aseltine, J.A. Transform method in linear system analysis. McGraw-Hill, New
York, 1958.
Bissell, C.C. Control engineering. Van Nostrand Reinhold, London, 1988.
Bolton, W. Control engineering. Longman, Harlow, 1992.
Borrie, J.A. Modern control systems. Prentice-Hall, Englewood Cliffs, 1986.
Brogan, W.L. Modern Control Theory. Prentice-Hall, Englewood Cliffs, 1985.
Chen, C-T. System and signal analysis. Holt, Rinehart & Winston, New York, 1989.
Chen, C-T. Linear system theory and design. Holt, Rinehart & Winston, New York,
1984.
D'Azzo, J.J. & Houpis, C.H. Linear control system analysis and design. McGraw-
Hill, New York, 1988. (3rd Edition)
Dorf, R.C. Modern Control Systems. Addison-Wesley, Reading, 1992. (6th Edition)
Franklin, G.F, Powell, J.D. & Emami-Naeini, A. Feedback control of dynamic
systems. Addison-Wesley, Reading, 1986.
Hostetter, G.H., Savant, C.J. & Stefani, R.T. Design of feedback control systems.
Holt, Rinehart & Winston, New York, 1982.
Karni, S. & Byatt, W.J. Mathematical methods in continuous and discrete systems.
Holt, Rinehart & Winston, New York, 1982
Kuo, B.C. Automatic control systems. Prentice-Hall, Englewood Cliffs, 1982. (4th
Edition)
Maciejowski, M. Multivariable feedback design. Addison-Wesley, Wokingham,
1989.
Miron, D.B. Design of feedback control systems. Harcourt Brace Jovanovich, San
Diego, 1989.
Morari, M. & Zafiriou, E. Robust process control. Prentice-Hall, Englewood Cliffs,
1989.
Nise, N.S. Control systems engineering. Benjamin Cummings. Redwood City, 1992.
Phillips, C.L. & Harbor, R.D. Feedback Control Systems. Prentice-Hall, Englewood
Cliffs, 1991. (2nd Edition)
Raven, F.H. Automatic Control Engineering. McGraw-Hill, New York, 1987. (4th
Edition)
Saucedo, R. & Schiring, E.E. Introduction to continuous and digital control
systems. Macmillan, New York, 1968.
Sinha, N.K. Control systems. Holt, Rinehart & Winston, New York, 1986.
Slotine, J-J.E. & Li, W. Applied nonlinear control. Prentice-Hall, Englewood Cliffs,
1991.
Van De Vegte, J. Feedback Control Systems. Prentice-Hall, Englewood Cliffs,
1986.
270 Control Engineering - 1

Notes
 271

Notes (Continued)
272 Control Engineering - 1

Computer-Aided Design Program

Many of the design diagrams and simulation plots shown in these notes were
produced by sCAD, a computer-aided design program written by the author.
The original program was written in Pascal for DOS and has subsequently been
converted to Visual Basic running on Windows 98 and Windows NT. The latest
version of the program is available for download from the website:

www.ci.ee.uct.ac.za

Other computer aided design programs, zCAD and ppCAD, are also available
from this site. All these programs are continuously being upgraded in reaction to
student comment, and include comprehensive on-line help files.
 Property of
Control & Instrumentation Laboratory,
Department of Electrical Engineering,
University of Cape Town.

LicenceExpired.doc, 26 August, 2001

 Licence Details
This product in The MBuct Collection is licenced to:

Full-time students while registered at U.C.T.

for control engineering courses that use:
The Control & Instrumentation Laboratory,
Department of Electrical Engineering,
University of Cape Town,
Private Bag,
Rondebosch 7701,
South Africa.

Should you wish to continue using this product then please enquire about a licence.

MBuct(c)2001. All rights reserved.

LicenceExpired.doc, 26 August, 2001