E d u c a tio n a n d C u ltu r e

Pilot Project No: CZ/02/B/F/PP/134001

Course on Dynamics of multidisplicinary and

controlled Systems

Part III
System Control
Module Version 3.2.0
November 4, 2004

Chr. Schmid
Lehrstuhl f
ur Automatisierungstechnik und Prozessinformatik
Ruhr-Universit
at Bochum

ii

Preface
This document contains a course both on the basics of control and on extensions
to get deeper insight into this area. The text contains many examples to illustrate the subjects. Demonstration examples, problems, interactive questions and
demonstration examples with on-line simulations and models visualised in virtual
reality are linked to the Internet. A CD version completes the set.
All the ideas, material and the facilities of this course would not have been implemented and completed without the hard work of the following persons, who
have directly or indirectly contributed to this course (in alphabetical order):
Abid Ali
Derek Atherton
Carsten Fritsch
Arnd Grosse-Frintrop
Christoph Hackstein
Norman Markgraf
Andrea Marschall
Tom Robert
Heinz Unbehauen
Bochum, October 31, 2004
Christian Schmid

iii

Contents
1 Introduction into System Control

1.1

Control objectives . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Open loop vs closed loop . . . . . . . . . . . . . . . . . . . . . . . .

1.3

The basic structure of closed-loop systems . . . . . . . . . . . . . .

2 Basic Control Elements

2.1

2.2

2.3

The Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . .

2.1.1

Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.1.2

Correspondences of the Laplace transform . . . . . . . . . . 10

2.1.3

Main theorems of the Laplace transform . . . . . . . . . . . 11

2.1.4

The inverse Laplace transform . . . . . . . . . . . . . . . . 14

2.1.5

Solving linear dierential equations using the Laplace transform 19

2.1.6

Laplace transform of the impulse (t) and step (t)

. . . . 24

Transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1

Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.2.2

Interpretation of the transfer function . . . . . . . . . . . . 27

2.2.3

Realisability and properness of transfer functions . . . . . . 28

2.2.4

Transfer functions with dead time . . . . . . . . . . . . . . 28

2.2.5

Poles and zeros of the transfer function . . . . . . . . . . . 28

2.2.6

Using transfer functions for calculations . . . . . . . . . . . 30

Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.1

Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.3.2

Nyquist plot of a frequency response . . . . . . . . . . . . . 37

2.3.3

Bode plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.3.4

Some important transfer function elements . . . . . . . . . 41

iv

Contents

2.4

2.5

2.3.4.1

The proportional element (P element) . . . . . . . 41

2.3.4.2

The integrator (I element) . . . . . . . . . . . . . 41

2.3.4.3

The derivative element (D element) . . . . . . . . 42

2.3.4.4

The 1st-order lag element (PT1 element) . . . . . 43

2.3.4.5

The proportional plus derivative element (PD element) 48

2.3.4.6

The derivative lag element (DT1 element) . . . . . 48

2.3.4.7

The 2nd-order lag element (PT2 element and PT2 S element) 50

2.3.4.8

Bandwidth of a system . . . . . . . . . . . . . . . 56

2.3.4.9

Example for the construction of a Bode plot . . . 56

2.3.5

Systems with minimum and non-minimum phase behaviour

58

2.3.6

Systems with dead time . . . . . . . . . . . . . . . . . . . . 61

Stability of linear control systems . . . . . . . . . . . . . . . . . . . 62

2.4.1

Stable and unstable systems . . . . . . . . . . . . . . . . . . 62

2.4.2

Denition of stability and stability conditions . . . . . . . . 62

2.4.3

Algebraic stability criteria . . . . . . . . . . . . . . . . . . . 64

2.4.3.1

The Hurwitz criterion . . . . . . . . . . . . . . . . 65

2.4.3.2

Routh criterion . . . . . . . . . . . . . . . . . . . . 68

2.4.3.3

Nyquist criterion . . . . . . . . . . . . . . . . . . . 70

2.4.3.4

Nyquist criterion using Nyquist plots . . . . . . . 70

2.4.3.5

Simplied forms of the Nyquist criterion

2.4.3.6

The Nyquist criterion using Bode plots . . . . . . 75

. . . . . 74

The root-locus method . . . . . . . . . . . . . . . . . . . . . . . . . 81

2.5.1

Introduction and basic ideas . . . . . . . . . . . . . . . . . . 81

2.5.2

General rules for constructing root loci . . . . . . . . . . . . 86

2.5.3

Example of an application of the root-locus method . . . . 92

3 Behaviour of linear continuous-time control systems

95

3.1

Dynamical behaviour of a closed loop system . . . . . . . . . . . . 95

3.2

Static properties of the closed loop . . . . . . . . . . . . . . . . . . 100

3.3

3.4

3.5

3.6

3.2.1

Transfer function G0 (s) with delayed P behaviour . . . . . 102

3.2.2

Transfer function G0 (s) with delayed I behaviour . . . . . . 103

3.2.3

Transfer function G0 (s) with delayed I2 behaviour . . . . . 104

Performance indices . . . . . . . . . . . . . . . . . . . . . . . . . . 104

3.3.1

Time-response specications . . . . . . . . . . . . . . . . . . 104

3.3.2

Integral performance indices . . . . . . . . . . . . . . . . . . 107

3.3.3

Determination of quadratic performance indices . . . . . . . 109

PID control and associated controller types . . . . . . . . . . . . . 111

3.4.1

The classical three-term PID controller

. . . . . . . . . . . 111

3.4.2

Optimal tuning of PID controllers . . . . . . . . . . . . . . 113

3.4.3

Advantages and disadvantages of the dierent types of controllers116

3.4.4

Empirical tuning rules according to Ziegler and Nichols . . 117

Design of controllers using pole-zero compensators . . . . . . . . . 120

3.5.1

Characteristics in frequency and time domain . . . . . . . . 120

3.5.2

Controller design using frequency domain characteristics . . 131

3.5.3

Application of the design using frequency domain characteristics138

3.5.4

Controller design using the root-locus method . . . . . . . . 142

Compensator design methods . . . . . . . . . . . . . . . . . . . . . 148

3.6.1

Basic ideas of compensator design . . . . . . . . . . . . . . 148

3.6.2

Design by specifying the closed-loop transfer function . . . 149

3.6.3

The method of Truxal and Guillemin . . . . . . . . . . . . . 150

3.6.4

Generalised compensator design method . . . . . . . . . . . 156

3.6.5

3.6.4.1

The basic idea . . . . . . . . . . . . . . . . . . . . 156

3.6.4.2

Zeros of the closed loop . . . . . . . . . . . . . . . 158

3.6.4.3

The synthesis equations . . . . . . . . . . . . . . . 161

3.6.4.4

Application of the method . . . . . . . . . . . . . 163

Compensator design for reference and disturbances . . . . . 168

3.6.5.1

Structure of the closed loop . . . . . . . . . . . . . 168

3.6.5.2

The design procedure . . . . . . . . . . . . . . . . 169

vi

Contents

3.7

3.6.5.3

Design of the pre-lter

3.6.5.4

Application of the design method . . . . . . . . . 179

Improving the control behaviour by more complex loop structures 184

3.7.1

Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

3.7.2

Disturbance feed-forward control . . . . . . . . . . . . . . . 184

3.7.2.1

Disturbance feed-forward on the controller . . . . 184

3.7.2.2

Disturbance feed-forward on the manipulated variable186

3.7.3

Control systems with an auxiliary manipulated variable . . 188

3.7.4

Cascade control systems . . . . . . . . . . . . . . . . . . . . 190

3.7.5

Control system with auxiliary manipulated variable . . . . 192

3.7.6

Control system with anti-windup measure . . . . . . . . . . 193

4 State variable techniques

4.1

4.2

. . . . . . . . . . . . . . . 175

197

State-space representation . . . . . . . . . . . . . . . . . . . . . . . 197

4.1.1

State-space representation of single-input-single-output systems197

4.1.2

State-space representation of multi-input-multi-output systems199

4.1.3

The relationship between transfer functions and the state-space representation199

4.1.4

State-space vs transfer function approach . . . . . . . . . . 201

4.1.5

Uniqueness of the state variables . . . . . . . . . . . . . . . 201

4.1.6

Controllability and observability . . . . . . . . . . . . . . . 203

Design of state-feedback control systems . . . . . . . . . . . . . . . 204

4.2.1

Structures and properties of state-feedback control systems 204

4.2.1.1

State-feedback control in the frequency domain . . 205

4.2.1.2

Steady state of the closed-loop system . . . . . . . 207

4.2.2

State-feedback control with integrator . . . . . . . . . . . . 207

4.2.3

Design of state-feedback controllers by pole placement . . . 209

4.2.3.1

Design of a system in controller canonical form . . 210

4.2.3.2

Design of a system not in a canonical form . . . . 211

4.2.3.3

Design using Ackermanns formula . . . . . . . . . 213

vii
4.2.4

State reconstruction using observers . . . . . . . . . . . . . 214

4.2.4.1

Structure of an observer . . . . . . . . . . . . . . . 215

4.2.4.2

Design of observers . . . . . . . . . . . . . . . . . 216

4.2.5

Combined observer-controllers

4.2.6

Example of a state-feedback control system . . . . . . . . . 219

5 Fuzzy control systems

5.1

5.2

5.3

5.4

. . . . . . . . . . . . . . . . 217

225

Introduction to fuzzy techniques . . . . . . . . . . . . . . . . . . . 225

5.1.1

Crisp and fuzzy logic . . . . . . . . . . . . . . . . . . . . . . 225

5.1.2

Why use fuzzy logic for control ? . . . . . . . . . . . . . . . 227

5.1.3

Ideas of the fuzzy control methodology . . . . . . . . . . . . 229

Basics of fuzzy sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

5.2.1

Fuzzy sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

5.2.2

Membership functions . . . . . . . . . . . . . . . . . . . . . 232

5.2.3

Elementary operators for fuzzy sets . . . . . . . . . . . . . . 235

5.2.4

Fuzzy relations . . . . . . . . . . . . . . . . . . . . . . . . . 237

5.2.5

Fuzzy composition . . . . . . . . . . . . . . . . . . . . . . . 240

Fuzzy systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

5.3.1

Fuzzication . . . . . . . . . . . . . . . . . . . . . . . . . . 243

5.3.2

Fuzzy inference machine . . . . . . . . . . . . . . . . . . . . 243

5.3.3

Defuzzication . . . . . . . . . . . . . . . . . . . . . . . . . 247
5.3.3.1

Centre of gravity method (COG) . . . . . . . . . . 249

5.3.3.2

Centre of singleton method (COS) . . . . . . . . . 249

5.3.3.3

Maximum methods . . . . . . . . . . . . . . . . . 250

5.3.3.4

Margin properties of the centroid methods . . . . 250

5.3.4

The Takagi-Sugeno fuzzy system . . . . . . . . . . . . . . . 250

5.3.5

The components of a fuzzy system . . . . . . . . . . . . . . 251

Fuzzy control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

5.4.1

Basic structure of a fuzzy controller . . . . . . . . . . . . . 252

viii

Contents
5.4.2

5.4.3

5.4.4

Transfer behaviour of fuzzy controllers . . . . . . . . . . . . 253

5.4.2.1

Representation using 2D characteristics . . . . . . 254

5.4.2.2

Inuence of the membership functions and rule base on the characteristic256

5.4.2.3

Representation using 3D characteristics . . . . . . 260

Example of a fuzzy control system . . . . . . . . . . . . . . 261

5.4.3.1

Loading crane plant model . . . . . . . . . . . . . 261

5.4.3.2

Fuzzy control system design . . . . . . . . . . . . 263

Contribution of fuzzy control . . . . . . . . . . . . . . . . . 265

A Mathematical and table appendix

269

A.1 The Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . . 269

A.1.1 Convergence

. . . . . . . . . . . . . . . . . . . . . . . . . . 269

A.1.2 The inverse Laplace transform . . . . . . . . . . . . . . . . 270

A.1.3 Main theorems of the Laplace transform . . . . . . . . . . . 270
A.1.3.1

Derivative theorem

A.1.3.2

Integral theorem . . . . . . . . . . . . . . . . . . . 271

A.1.3.3

Convolution in the time domain . . . . . . . . . . 271

A.1.3.4

Convolution in the frequency domain . . . . . . . 272

A.1.3.5

Initial value theorem . . . . . . . . . . . . . . . . . 273

A.1.3.6

Final value theorem . . . . . . . . . . . . . . . . . 274

A.2 The complex G-plane

. . . . . . . . . . . . . . . . . 270

. . . . . . . . . . . . . . . . . . . . . . . . . 275

A.3 Detailed analysis of 2nd-order lag elements . . . . . . . . . . . . . 277

A.3.1 Determining resonances of 2nd-order lag elements

. . . . . 277

A.3.2 Poles and step responses of 2nd-order lag elements . . . . . 278

A.4 The law of Bode and the Hilbert transformation . . . . . . . . . . 281
A.5 Stability considerations using the weighting function . . . . . . . . 281
A.6 Equivalence of the Hurwitz and Routh criteria . . . . . . . . . . . 282
A.7 Determination of JISE using determinants . . . . . . . . . . . . . . 283
A.8 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

1
Index

295

List of Tables

308

List of Figures

309

Introduction into System Control

1.1

Control objectives

The use of automatic control systems permeates life in all advanced societies
today. Such systems act as a catalyst for promoting progress and development.
Control systems are an integral component of any industrial society and are
necessary for the production of goods. Technological developments have made
it possible to travel to the moon and outer space. The successful production of
chemical components depends on the proper functioning of a large number of
control systems used in lines for their production. As this fact is seldom apparent
control engineering is often called a hidden technology.
Control engineering deals with the task of aecting a temporally changing process
in such a way that the process behaves in a given way. Such tasks are not only
found in technology, but also in daily life in very large number. For example
the ambient temperature in a room must be held between given limits, despite
temporal changes due to sun exposure and other inuences. The grip arm of a
robot must move along the edge of a workpiece or be led as fast as possible from
one point to another in order to grip a workpiece. The same applies to the grip
arm of a crane, which is to carry bricks to a certain place on the building site.
In all of these cases, a manipulated variable must be selected in such a way
that the given goal is achieved. As this selection depends on how well the goal is
reached, a control loop arises that consists of the given process and a new feature,
the controller. In the rst example, the room was the process and the thermal
valve the automatic controller, which measures the current air temperature and
lets more or less heat into the heater depending on the deviation from the target
temperature. In the robot example the control equipment has the task of steering
the grip arm on a given course and/or to a given point, whereby the control is
based on information that is supplied by the sensors installed on the grip arm. In
the third example, the automatic controller is the crane operator, who determines
the current grip arm position by sight and steers the crane.

1.2 Open loop vs closed loop

1.2

Open loop vs closed loop

The terms open-loop control and closed-loop control are often not clearly distinguished. Therefore, the dierence between open-loop control and closed-loop
control is demonstrated in the following example of a room heating system. In the
case of open-loop control of the room temperature R according to Figure 1.2.1
the outdoor temperature A will be measured by a temperature sensor and fed

z 2 = J
A

c o n tro l
d e v ic e

z 1

y = J R

( o p e n w in d o w )

Q
u

M
V

Figure 1.2.1. Open-loop control of a room heating system

into a control device. In the case of changes in the outdoor temperature A (=

disturbance z2
) the control device adjusts the heating ow Q according to the
characteristic Q = f (A ) of Figure 1.2.2 using the motor M and the valve V.
The slope of this characteristic can be tuned at the control device. If the room
disturbance z1
) this will not
temperature R is changed by opening a window (=
inuence the position of the valve, because only the outdoor temperature will
inuence the heating ow. This control principle will not compensate the eects
of all disturbances.
Q
1
2
3

2 0

- 2 0

J A

Figure 1.2.2. Characteristic of a heating control device for three dierent tuning sets
(1, 2, 3)

In the case of closed-loop control of the room temperature as shown in Figure 1.2.3

1 Introduction into System Control

the room temperature R is measured and compared with the set-point value w,
(e.g. w = 20 C). If the room temperature deviates from the given set-point value,
a controller (C) alters the heat ow Q. All changes of the room temperature R ,
e.g. caused by opening the window or by solar radiation, are detected by the
controller and removed.
ro o m

se n so r
w

z 1

y = J R

( o p e n w in d o w )

u
M

Q
V

Figure 1.2.3. Closed-loop control of a room heating system

The block diagrams of the open-loop and the closed-loop temperature control systems are shown in Figures 1.2.4 and 1.2.5, and from these the dierence between
open- and closed-loop control is readily apparent.
z

z 1
1

z 2
2
c o n t r o l u
d e v ic e

+
+

r o o m

Figure 1.2.4. Block diagram of the open-loop control of the heating system

The order of events to organise a closed-loop control is characterised by the

following steps:
Measurement of the controlled variable y,
Calculation of the control error e = w y (comparison of the controlled
variable y with the set-point value w),
Processing of the control error such that by changing the manipulated variable u the control error is reduced or removed.

1.2 Open loop vs closed loop

5
z

z 1
1

z 2
w

+
_

c o n t r o lle r

ro o m
1

+
+

Figure 1.2.5. Block diagram of the closed-loop control of the heating system

Comparing open-loop control with closed-loop control the following dierences

are seen:
Closed-loop control
shows a closed-loop action (closed control loop);
can counteract against disturbances (negative feedback);
can become unstable, i.e. the controlled variable does not fade away, but
grows (theoretically) to an innite value.
Open-loop control
shows an open-loop action (controlled chain);
can only counteract against disturbances, for which it has been designed;
other disturbances cannot be removed;
cannot become unstable as long as the controlled object is stable.
Summarising these properties we can dene:
Systems in which the output quantity has no eect upon the process input quantity are called open-loop control systems.
Systems in which the output has an eect upon the process input quantity in such
a manner as to maintain the desired output value are called closed-loop control
systems.

1 Introduction into System Control

1.3

The basic structure of closed-loop systems

In this section the general structure of control systems having a closed loop will be
analysed in more detail. According to Figure 1.3.1 a closed-loop system consists
of the following four main components:
plant, measurement device, controller and actuator.
The signals in the closed loop will be denoted by symbols. It means:
y
w
e

controlled variable (actual value)

command variable (set point),
control error (deviation)

z
e
w
+
_

c o n tro lle r

u
C

a c tu a to r

'

u
z

manipulated variable
disturbance.

d is tu rb a n c e
b e h a v io u r

c o n tro l
b e h a v io u r +

y
C

m e a su re m e n t
d e v ic e

p la n t

Figure 1.3.1. Basic block diagram of a control system

From this block diagram it can be realised that the task of controlling a process
(plant) consists of holding the controlled value y(t), acquired by the measurement
device, either on a constant set point w(t) = const (xed command control)
or tracking a time-varying reference variable w(t)
= const (variable command
control), independent of external disturbances z(t). This task is performed by a
controller. The controller processes the control error e(t) = w(t) y(t), which is
the dierence between the set point w(t) and the actual value y(t) of the controlled
variable. The control signal uC (t) generated by the controller will act via the
actuator as the manipulated variable u(t) on the plant, such that it counteracts
in the case of xed command control against the disturbance z(t). A closed-loop
control system is characterised by this closed signal path, whereby the controller
function consists in cancelling the occurring control error e(t) or at least holding
it very small.
Closed-loop control problems can be reduced to this basic structure. In most cases
it is not possible to identify all the basic functions clearly. It is therefore proper
to aggregate a control loop only into two blocks. Hereby we distinguish between

1.3 The basic structure of closed-loop systems

z
w
+
_

c o n t r o llin g
sy ste m

7
z

d is t u r b a n c e
b e h a v io u r
c o n tro l
b e h a v+ i o u r

y
+

p la n t

Figure 1.3.2. Simplied block diagram of a closed-loop control system

the plant, which may also aggregate the measurement device, and the controlling
system that usually contains the actuator, as shown in Figure 1.3.2.
From Figures 1.3.1 and 1.3.2 it becomes obvious that the comparison of the setpoint value w and the actual value y of the controlled variable for generating
the control error e will become possible just through the negative feedback of the
controlled variable y. Only because of this negative sign at the summing point
of both signals is the control error generated, which is used by the controller to
build the control signal u using special mathematical functions (e.g. proportional,
integrating, dierentiating). The principle of negative feedback, shortly also called
the feedback principle, is a characteristic for every control loop.
The principles of closed-loop control are demonstrated by the following examples.
Click on the links to start the animation with your Web browser.
Demonstration Example 1.3.1
Water tank level without control and with disturbances
Demonstration Example 1.3.2
Water tank level without control and with set point
Demonstration Example 1.3.3
Water tank level manual control
Demonstration Example 1.3.4
Water tank level min/max control
Demonstration Example 1.3.5
Water tank level closed-loop control
Demonstration Example 1.3.6
Main components of a control system

1 Introduction into System Control

Basic Control Elements

2.1

The Laplace transform

2.1.1

Denition

The Laplace transform is an important tool for solving systems of linear dierential equations with constant coecients. The dierential equations to be solved
for control tasks normally full the conditions that must be met for taking the
Laplace transform. The Laplace transform is an integral transformation, which
maps a large class of original functions f (t) in the time domain unambiguously
reversible into image functions F (s) in the s domain. This mapping is performed
via the Laplace integral of f (t), that is


F (s) =

f (t) est dt ,

(2.1.1)

where in the argument of the Laplace transform F (s) the complex variable s =
+ j appears. For the application of Eq. (2.1.1) to causal systems considered
here the following two conditions for the time function f (t) must be met:
1. f (t) = 0 for t < 0 ;
2. the integral in Eq. (2.1.1) must converge.
To show the correspondence between the original and mapped functions it is
useful to use the operator notation
F (s) =  [f (t)] .
Another possibility of correspondence is to use the sign in the following way:
F (s) f (t) .
During the treatment of control systems usually the original function f (t) is
a function of time. As the complex variable s contains the frequency , the
image function F (s) will often be called a frequency function. Therefore, the
Laplace transform allows one to make a transition from the time domain into
the frequency domain according to Eq. (2.1.1).

10

2 Basic Control Elements

2.1.2

Correspondences of the Laplace transform

The so called back transformation or inverse Laplace transformation, i.e. the

determination of the original function from the mapped function, is described by
the inverse integral shown in section A.1.2. For this inverse Laplace transform an
operator notation in the form
f (t) = 

[F (s)]

can be used.
The Laplace transformation is an unambiguously reversible mapping between
the original function and the mapped function. f (t) and F (s) are referred to as
transform pairs and have a unique correspondence. This is the reason why in
most cases one does not need to use the inverse integral. A Correspondence table,
as shown in Table 2.1.1, will suce. For the inverse transformation case ones
goes from the right column to the left column. In addition some theorems on the
Laplace transform given in the next section may be useful.
Table 2.1.1: Corresponding elements of the Laplace transform
Nr.

time response f (t), f (t) = 0 for t < 0

Laplace transformed F (s)

pulse (t)

unit step (t)

1
s

1
s2

t2

2
s3

tn
n!

eat

1
s+a

teat

1
(s + a)2

t2 eat

2
(s + a)3

tn eat

n!
(s + a)n+1

1
sn+1

2.1 The Laplace transform

11
Table 2.1.1 continued
a
s(s + a)

10

1 eat

11

1 at
(e
1 + at)
a2

12

(1 at) eat

13

sin 0 t

14

cos 0 t

15

eat sin 0 t

0
(s + a)2 + 02

16

eat cos 0 t

s+a
(s + a)2 + 02

17

1
f
a

1
+ a)

s2 (s

s
(s + a)2

 
t
a

s2

0
+ 02

s2

s
+ 02

F (as) (a > 0)

18

eat f (t)

F (s a)

19

f (t a) for t > a 0
0 for t < a

eas F (s)

20

t f (t)

d F (s)
ds

21

(t)n f (t)

dn F (s)
dsn

22

f1 (t) f2 (t)

c+j

1
2j

F1 (p) F2 (s p) dp
cj

2.1.3

Main theorems of the Laplace transform

a) Superposition theorem:
For arbitrary constants a1 and a2 it follows that
 {a1 f1 (t) + a2 f2 (t)} = a1 F1 (s) + a2 F2 (s) .

(2.1.2)

12

2 Basic Control Elements

The Laplace transformation is a linear integral transformation.

b) Similarity theorem:
For an arbitrary constant a > 0
 {f (at)} =

1 s
F
a
a

(2.1.3)

is valid. This follows from Eq. (2.1.1) by the substitution of = at.

c) Real Shifting theorem:
For an arbitrary constant a > 0
 {f (t a)} = eas F (s)

(2.1.4)

is valid. This follows directly from Eq. (2.1.1) by the substitution of = ta.
d) Complex Shifting theorem:
For an arbitrary constant a > 0


 eat f (t) = F (s + a)

(2.1.5)

is valid. This follows directly from Eq. (2.1.1).

e) Derivative theorem:
For a causal function of time, f (t), for which the derivative for t > 0 exists,
then as shown in section A.1.3.1, one obtains


d f (t)
= s F (s) f (0+) ,
(2.1.6)

dt
and in the case of multiple dierentiation



dn f (t)
d tn

= sn F (s)

i=1


(i1) f (t) 
d

sni

(i1)

dt

(2.1.7)

t=0+

f) Complex dierentiation theorem:

This theorem shows that a dierentiation of the mapped function F (s) corresponds to a multiplication with the time t in the time domain:


dk F (s)
.
tk f (t) = (1)k
dsk

(2.1.8)

2.1 The Laplace transform

13

g) Integral theorem:
The integral of a function is mapped by


t


f ( ) d

1
F (s) .
s

(2.1.9)

as shown in section A.1.3.2.

h) Convolution in the time domain:
The convolution of two functions of time f1 (t) and f2 (t), presented by the
symbolic notation f1 (t) f2 (t), is dened as

t
f1 (t) f2 (t) =

f1 ( ) f2 (t ) d .

(2.1.10)

In section A.1.3.3 it is shown that the convolution of the two original functions corresponds to the multiplication of the related mapped functions, that
is
 {f1 (t) f2 (t)} = F1 (s) F2 (s) .
(2.1.11)
i) Convolution in the frequency domain:
Whereas in h) the convolution of two functions of time was given, a similar
result for the convolution of two functions in the frequency domain exists
and is given by
1
 {f1 (t) f2 (t)} =
2j

c+j

F1 (p) F2 (s p) dp .

(2.1.12)

cj

Here F1 (s) f1 (t) and F2 (s) f2 (t) is valid. Furthermore, p is the complex
variable of integration. According to this theorem the Laplace transform of
the product of two functions of time is equal to the convolution of F1 (s) and
F2 (s) in the mapped domain. This is shown in detail in section A.1.3.4.
j) Initial and nal value theorems:
The theorem of the initial condition allows the direct calculation of the function value f (0+) of a causal function of time f (t) from the Laplace transform
F (s). If the Laplace transform of f (t) and f(t) exist, then
f (0+) = lim f (t) = lim s F (s)
t0+

is valid if the lim f (t) exists, see section A.1.3.5.

t0

(2.1.13)

14

2 Basic Control Elements

Using the theorem of the nal value the value of f (t) for t can be
determined from F (s), if the Laplace transform of f (t) and f(t) exist and
the limit lim f (t) also exists. Then it follows from section A.1.3.6 that
t

f () = lim f (t) = lim s F (s) .

t

s0

(2.1.14)

One has to observe that

lim f (t)

or

lim f (t)

t0

can be calculated only from the corresponding Laplace transform  {f (t)}

by application of the theorems of the initial or nal value, if the existence of
the related limit in the time domain is a priori assured. The following two
examples should explain this:
Example 2.1.1
f (t) = et ( > 0) F (s) =

1
s

The limit lim et does not exist so that the nal value theorem may not be
t
applied.


Example 2.1.2
f (t) = cos 0 t F (s) =

s2

s
+ 02

The limit lim cos 0 t does not exist and therefore the nal value theorem may
t

not be applied.

It can be concluded from the last two examples that the following general statement is valid: If the Laplace transform F (s) has, apart from a single pole at the
origin s = 0, poles on the imaginary axis or in the right-half s plane, then the
initial or nal value theorems cannot be applied.

2.1.4

The inverse Laplace transform

The inverse Laplace transform is described by Eq. (A.1.2). As already mentioned

in section 2.1.2 in many cases a direct evaluation of the complex inverse integral is not necessary, as the most important elementary functions are given in
Table 2.1.1. A complicated function, F (s), not given in Table 2.1.1 must be decomposed into a sum of simple functions of s that is

2.1 The Laplace transform

15

F (s) = F1 (s) + F2 (s) + . . . + Fn (s) ,

(2.1.15)

which have a known inverse Laplace transform:



{F (s)} =  1 {F1 (s)} +  1 {F2 (s)} + . . . + 

= f1 (t) + f2 (t) + . . . + fn (t) = f (t) .

{F3 (s)}

(2.1.16)

For many problems in control the function F (s) is a ratio of polynomials in s,

known as rational fraction, that is
F s) =

n 0 + n1 s + . . . nm sm
N (s)
,
=
n
d0 + d1 s + . . . + s
D(s)

(2.1.17)

where N (s) and D(s) are the numerator and the denominator, respectively.
If m > n, then N (s) is divided by D(s), where a polynomial in s and a ratio of
polynomials are obtained. The numerator of the fraction N1 (s) has a lower order
than n. E.g, if m = n + 2, then
N (s)
N1 (s)
= k2 s2 + k1 s + k0 +
,
D(s)
D(s)

(2.1.18)

whereby degree N1 (s) < n and k0 , k1 and k2 are constants.

A rational fraction F (s) given in Eq. (2.1.17) can be decomposed into more
simple functions by application of partial fraction decomposition, as shown in
Eq. (2.1.15). In order to perform this decomposition the denominator polynomial
D(s) must be factorised into the form
F (s) =

N (s)
.
(s s 1) (s s2 ) . . . (s sn )

(2.1.19)

For a denominator polynomial of n-th order one obtains n roots or zeros s =

s1 ,s2 , . . . ,sn . The zeros of D(s) are also known as the poles of F (s), since they
dene where F (s) is innite. The partial fraction decomposition for dierent types
of poles is shown in the following.
Case 1: F (s) has only single poles.
Here F (s) can be expanded into the form
F (s) =

k=1

ck
,
s sk

(2.1.20)

where the residuals ck are real or complex constants. Using the table of correspondences one immediately can obtain the corresponding function of time
f (t) =

k=1

ck esk t

for

t>0.

(2.1.21)

16

2 Basic Control Elements

The values ck can be determined either by comparing the coecients or by using

the theorem of residuals from the theory of functions according to
ck =

N (sk )
N (s) 
)
=
(s

s

k
D
(sk )
D(s) s=sk

(2.1.22)

for k = 1,2, . . . ,n with D
(sk ) = d D/ds |s=sk .

Case 2: F (s) has multiple poles.
For multiple poles of F (s) each with multiplicity rk (k = 1,2, . . . ,l) the corresponding partial fraction decomposition is
F (s) =

rk
l

k=1

ck
(s sk )
=1

with

n=

rk .

(2.1.23)

k=1

The back transformation of Eq. (2.1.23) into the time domain is

f (t) =

k=1

esk t

rk

ck t1
( 1)!

for

t>0.

(2.1.24)

=1

The real or complex coecients ck for = 1,2, . . . ,rk determined by the theorem
of residuals are


d(rk )
1
[F (s) (s sk )rk ]
.
(2.1.25)
ck =
(rk )! ds(rk )!
s=sk

This general relation also contains the case of single poles of F (s). The poles may
be real or complex.
Case 3: F (s) has also conjugate complex poles.
As both, the numerator N (s) and the denominator D(s) of the function F (s) are
rational algebraic functions, complex factors always arise as conjugate complex
pairs. If F (s) has a conjugate complex pair of poles s1,2 = 1 j1 , then for the
function F1,2 (s) in the partial fraction decomposition of
F (s) =

N (s)
= F1,2 (s) + F3 (s) + . . . + Fn (s)
D(s)

Eq. (2.1.20) can be applied to give

F1,2 (s) =

c2
c1
+
,
s (1 + j1 ) s (1 j1 )

where the residuals

c1,2 = 1 j1

(2.1.26)

2.1 The Laplace transform

17

are also a conjugate complex pair. Therefore, both fractions of F1,2 (s) can be
combined, and one obtains
F1,2 (s) =

0 + 1 s
0 + 1 s + s2

with the real coecients

; 1 = 21
0 = 12 + 12
0 = 2(2 1 + 1 1 ) ; 1 = 21

(2.1.27)

(2.1.28)

The determination of the coecients 0 und 1 is performed again using the

theorem of residuals by


N (s) 

.
(2.1.29)
(0 + 1 s) s=s1 = (s s1 ) (s s2 )
D(s) s=s1
As s1 is complex, both sides of this equation are complex. Comparing the real
and imaginary parts of both sides one gets two equations for the calculation of
0 und 1 . This procedure is demonstrated now using the following example.
Example 2.1.3
Find the inverse Laplace transform f (t) of
F (s) =

(s2

1
.
+ 2s + 2) (s + 2)

The partial fraction decomposition of F (s) is

F (s) = F1,2 (s) + F3 (s) =

c3
0 + 1 s
+
,
+ 2s + 2 s + 2

s2

where the function F1,2 (s) contains the conjugate pair of poles
s1,2 = 1 j .
In addition the third pole of F (s) is
s3 = 2 .
For the coecients 0 and 1 it follows from Eq. (2.1.29)


1 

(0 + 1 s) s=s1 =
s + 2 s=s1
(0 1 ) + j1 =

1
1
1
= j .
1 + j + 2
2
2

18

2 Basic Control Elements

Comparing the real and imaginary parts on both sides one obtains
0 1 =

1
2

and

1 =

1
2

and from this nally 0 = 0.

Using Eq. (2.1.22) the residual is


1
1

= .
c3 = (s + 2) 2

(s + 2s + 2) (s + 2) s=s3
2
The partial fraction decomposition of F (s) is thus


s
1 1
1
+
,
F (s) =
2 s2 + 2s + 2
2 s+2
which can be rearranged in the form


s+1
1
1
1

F (s) =
2 (s + 1)2 + 1 (s + 1)2 + 1 s + 2
such that correspondences given in Table 2.1.1 can be directly applied to nd the
inverse transformation. Using the correspondences 16, 15 and 6 of this table, it
follows that
1
f (t) = [et cos t et sin t e2t ]
2

for

t>0,

which can be rearranged as

f (t) =

1 t t
e [e + sin t cos t]
2

for

t>0.

The graphical representation of f (t) is shown in Figure 2.1.1a. Figure 2.1.1b shows
the corresponding poles, marked by a x, for this F (s) in the complex s plane.

It can be seen from this example that the position of the poles s1 ,s2 and s3 aects
the shape of the graph of f (t). In this case all poles of F (s) have negative real
parts, therefore the graph of f (t) shows a damped behaviour, i.e. it decreases to
zero for t . If the real part of one pole be positive, then the graph of f (t)
would be innitely large for t .
Since in control problems the original function f (t) always represents the time
behaviour of a system variable, the behaviour of this system variable f (t) can
be judged to a large extent by investigation of the positions of the poles of the
corresponding mapped function F (s). This will be further commented on in later
sections.

2.1 The Laplace transform

19

f(t)

s p la n e

(a )

0 ,1 5

jw

p o le s

0 ,1 0
s

4
5

- 1
s

- 2

0 ,0 5

(b )

- j

t
8

Figure 2.1.1. (a) Graph of the original function f (t) (function in the time domain)
and (b) position of the poles of F (s) in the s plane

2.1.5

Solving linear dierential equations using the Laplace transform

The Laplace transform, the basics of which have been introduced in the sections
above, is an elegant way for fast and schematic solving of linear dierential equations with constant coecients. In the following the importance of this approach
is demonstrated. Instead of solving the dierential equation with the initial conditions directly in the original domain, the detour via a mapping into the frequency
domain is taken, where only an algebraic equation has to be solved. Thus solving
dierential equations is performed according to Figure 2.1.2 in the following three
steps:
1. Transformation of the dierential equation into the mapped space ,
2. Solving the algebraic equation in the mapped space,
3. Back transformation of the solution into the original space.
o r ig in a l:

d iffe r e n t ia l e q u a t io n
L t r a n s fo r m a t io n

m a p p e d :

a lg e b r a ic e q u a t io n

s o lu t io n
- 1

t r a n s fo r m a t io n
s o lu t io n

Figure 2.1.2. Schema for solving dierential equations using the Laplace transformation

20

2 Basic Control Elements

Demonstration Example 2.1.1

Here the same in animated form
Whereas the rst two steps are trivial, the third step usually demands more eort.
The procedure will be demonstrated by the following two examples.
Example 2.1.4
Consider the dierential equation
f(t) + 3f(t) + 2f (t) = et
with the initial conditions f (0+) = f(0+) = 0.
Proceeding using the steps given above one has
Step 1:
s2 F (s) + 3sF (s) + 2F (s) =

1
s+1

Step 2:
F (s) =

1
1
2
s + 1 s + 3s + 2

Step 3:
The complex function F (s) must be decomposed into partial fractions in
order to use the tables of correspondences. This gives
F (s) =

1
1
1

+
.
s + 2 s + 1 (s + 1)2

By means of the correspondences 6 and 7 of Table 2.1.1 it follows from

the inverse Laplace transformation that the solution of the given dierential
equation is
f (t) = e2t et + t et .


Example 2.1.5
Given the dierential equation
x
+ a1 x + a0 x = 0 ,

(2.1.30)

and x(0+) are

where a0 and a1 are constants and the initial conditions x(0+)
known. Then
Step 1:

+ a1 [s X(s) x(0+)] + a0 X(s) = 0

s2 X(s) s x(0+) x(0+)

2.1 The Laplace transform

21

Step 2:
X(s) =

1
s + a1
x(0+) + 2
x(0+)

,
s2 + a1 s + a0
s + a1 s + a0

(2.1.31)

X(s) = L0 (s) x(0+) + L(s) x(0+)

with the abbreviation
L0 (s) =

s + a1
N0 (s)
= 2
D(s)
s + a1 s + a0

1
N (s)
= 2
.
D(s)
s + a1 s + a0

and L(s) =

Step 3:
Case a): two single real zeros of the denominator :
This means
D(s) = s2 + a1 s + a0 = (s 1 ) (s 2 ) .
For both rational expressions L0 (s) and L(s) it follows by partial fraction
decomposition that
L0 (s) =

A1
A2
+
s 1 s 2

and

L(s) =

B1
B2
+
.
s 1 s 2

The coecients Ai and Bi can now be determined by comparing coecients

or by applying Eq. (2.1.22):
Ai =

N0 (i )
;
D
(i )

Bi =

N (i )
D
(i )

for

i = 1, 2 .

Thus for Eq. (2.1.31) follows





A2
B2
B1
A1
+
+
x(0+) +
x(0+)

,
X(s) =
s 1 s 2
s 1 s 2
and by applying the correspondence 6 from Table 2.1.1 the solution of the
dierential equation is





x(t) = A1 e1 t + A2 e2 t x(0+) + B1 e1 t + B2 e2 t x(0+)

e1 t + [A2 x(0+) + B2 x(0+)]

e2 t .
= [A1 x(0+) + B1 x(0+)]

(2.1.32)

Case b): One double real zero of the denominator :

Here is
D(s) = (s )2 .
For the two rational expressions L0 (s) and L(s) of Eq. (2.1.31) the partial
fraction decomposition is

22

2 Basic Control Elements

L0 (s) =

A2
A1
+
s (s )2

and

L(s) =

B2
B1
+
.
s (s )2

The coecients Ai and Bi are determined by comparing both sides or by

evaluation of Eq. (2.1.25):
 



d N0 (s)
N0 (s)
2
2
(s )
(s )
= 1, A2 =
= + a1
A1 =
ds D(s)
D(s)
s=
s=
and

B1 =



d N (s)
2
(s )
= 0,
ds D(s)
s=

N (s)
(s )2
B2 =
D(s)


=1.
s=

From these results one obtains the solution

X(s) =

x(0+) ( + a1 ) x(0+) + x(0+)

+
s
(s )2

in the mapped space. By applying the inverse Laplace transformation the

required solution of the dierential equation is

t et .
x(t) = x(0+) et + [( + a1 ) x(0+) + x(0+)]

(2.1.33)

Case c): Two conjugate complex zeroes of the denominator :

Here
D(s) = (s 1 ) (s 2 )

with 1,2 = 1 j1 .

Introducing the values of 1 and 2 and after multiplication of this expression

one obtains
D(s) = (s 1 )2 + 12 .
Comparison with the denominator of the original relation, Eq. (2.1.31), gives
according to Eq. (2.1.28)
a0 = 12 + 12

and

a1 = 21 .

With these coecients Eq. (2.1.31) is in the form

1
s 21
x(0+) +
x(0+)

2
2
(s 1 ) + 1
(s 1 )2 + 12


1
1
s 1

x(0+)
=
(s 1 )2 + 12 1 (s 1 )2 + 12
1
1
x(0+)

,
+
1 (s 1 )2 + 12

X(s) =

2.1 The Laplace transform

23

and from this one gets by applying the correspondences 15 and 16 of Table 2.1.1 to X(s) the corresponding time function


1
1
1 t
sin 1 t x(0+) + e1 t x(0+)

sin 1 t
cos 1 t
x(t) = e
1
1
or rearranged
1 t

x(t) = e





1
1
x(0+)

x(0+) sin 1 t .
x(0+) cos 1 t +
1
1

(2.1.34)

Also from Eq. (2.1.31) of this example the importance of the position of the zeros
of D(s), the poles of X(s), on the solution is clear. For all three cases the solution
of the dierential equation according to Eqs. (2.1.32), (2.1.33) and (2.1.34) is
mainly inuenced by the position of the poles of X(s). These poles of X(s)
are as one can see from the two examples only depend on the left side of
the corresponding dierential equation, i.e. the homogeneous part of it. As is
generally known the solution of the homogeneous dierential equation describes
the modes of the system, that is the behaviour, which depends only on the initial
conditions. Therefore, consider for the general case only the homogeneous part
of an nth-order ordinary homogeneous linear dierential equation with constant
coecients that is
n

di xa (t)
ai
=0
(2.1.35)
dti
i=0

with all n initial conditions

(di xa (t)/dti ) |t=0+

for

i = 0,1, . . . ,n 1 .

One obtains by Laplace transformation

 n
  n

i
1


i
i d
ai s
ai
s
x (t) t=0+ = 0
Xa (s)
1 a
dt
=1
i=0

i=1

and a form according to Eq. (2.1.31)

n


Xa (s) =

i=1

ai

i

=1


d1

x
a(t)
t=0+
dt1
N (s)
,
=
n

D(s)
i
ai s

si

(2.1.36)

i=0

where N (s) and D(s) are polynomials in s and the initial conditions are only in
the numerator polynomial N (s). The poles sk (k = 1,2, . . . ,n) of Xa (s) can be
determined directly from the solution of the equation

24

2 Basic Control Elements

n

ai si = 0 .

(2.1.37)

i=0

After factorisation of this equation one obtains

an (s s1 ) (s s2 ) . . . (s sn ) = 0 .

(2.1.38)

The poles sk of Xa (s) make it possible to perform a partial fraction decomposition

of Xa (s), e.g. for the case of single poles according to Eq. (2.1.20). For this case
one obtains following Eq. (2.1.21) the solution of the homogeneous dierential
equation, Eq. (2.1.35), in the form
xa (t) =

ck esk t

for

t>0.

k=1

From this one can realise that the position of the poles sk of Xa (s) in the s plane
completely characterises the modes or inherent behaviour of the system described
by Eq. (2.1.35). Thus one obtains for Re sk < 0 (left-half s plane) a decreasing
and for Re sk > 0 (right-half s plane) an increasing behaviour of xa (t), while for
pairs of poles with Re sk = 0 permanent oscillations occur. Therefore, Eq. (2.1.37)
or equivalently Eq. (2.1.38), is called the characteristic equation and the poles sk
of Xa (s) are often called eigenvalues of the equation. Therefore investigation of
the characteristic equation provides the most important information about the
oscillating behaviour of a system.

2.1.6

Laplace transform of the impulse (t) and step (t)

The impulse function (t) is not a function in the sense of classical analysis,
but a distribution (pseudo-function). Therefore without entering the theory of
distributions the integral


 {(t)} =

(t) est dt

(2.1.39)

is not dened. The singularity exactly matches with the lower integration limit.
The impulse function can be approximately described by the limit
(t) = lim r (t)
0

with the rectangular impulse function


1/ for 0 t
r =
0
otherwise .

(2.1.40)

2.1 The Laplace transform

25

Strictly speaking this representation of (t) is not a distribution, as r (t) for

0 t is not arbitrarily often dierentiable. Because of the simple description
compared with other functions (e.g. Gaussian functions) this approach is preferred
here. From Eq. (2.1.39) it follows that
 {(t)} =




lim r (t) est dt .

(2.1.41)

As Eq. (2.1.40) can also be represented in the form

1
r (t) = [(t) (t )] ,

(2.1.42)

where (t) is the unit step. Since the integration is independent of , the limit
and integration can be permuted so that



1
st
 {(t)} = lim
[(t) (t )] e dt
0
0



1 1
1 es
 {(t)} = lim
.
0 s
By applying l
Hospitals rule one obtains
s es
=1.
0
s

 {(t)} = lim

(2.1.43)

As the impulse (t) has an area of unity it is also called unit impulse.
Example 2.1.6
Given the dierential equation
dy
= (t) .
dt
Find solution y(t).
Remark: The derivative theorem according to Eq. (2.1.6) is as mentioned in
section A.1.3.1 valid only for classical functions. If, however, a signal consists
of a function at t = 0, then the lower integration limit of Eq. (2.1.1) must be
chosen equal to t = 0 and also in Eq. (A.1.11) the left-hand initial condition to
y(0). According to the denition of Eq. (2.1.1) all left-hand initial conditions
are always zero.
The solution can be determined in the following three steps:

26

2 Basic Control Elements

Step 1:
The Laplace transform of the given dierential equation is:
s Y (s) y(0) = 1

withy(0) = 0 .

Step 2:
The solution of the algebraic equation is:
1
Y (s) = .
s
Step 3:
From the back transformation the solution follows as
y(t) = (t) ,

where (t) is the unit step function.

2.2
2.2.1

Transfer functions
Denition

Linear, continuous-time time-invariant systems with lumped parameters as initially a dead time is not being taken into account will be described by the
ordinary dierential equation
n
m

di xa (t)
dj xe (t)
ai
=
bj
.
(2.2.1)
dti
dtj
i=0

j=0

If all initial conditions are set to zero and the Laplace transformation is applied
to both sides of this equation, one obtains
n
m

ai si = Xe (s)
bj sj ,
Xa (s)
i=0

j=0

or reordering
b0 + b1 s + . . . + bm sm
N (s)
Xa (s)
=
,
(2.2.2)
= G(s) =
n
Xe (s)
a0 + a1 a + . . . + an s
D(s)
where N (s) and D(s) describe the numerator and denominator polynomials, respectively. The quotient of the Laplace-transformed output and input of such
a type of system is a rational fraction. The coecients of this fraction depend
only on the structure and parameters of the system. Such a type of function
G(s), which describes completely the transfer behaviour of a system, is called the
transfer function of the system. With such a transfer function the output
Xa (s) = G(s) Xe (s)

(2.2.3)

can be immediately calculated for a known input signal xe (t), and therefore Xe (s).

2.2 Transfer functions

2.2.2

27

Interpretation of the transfer function

Comparing Eq. (2.2.3) with the convolution theorem from section 2.1.3, Eqs. (2.1.10)
and (2.1.11), it then follows for the representation of Eq. (2.2.3) in the time domain that

t
xa (t) = g(t ) xe ( ) d ,
(2.2.4)
0

where obviously the inverse Laplace transform of G(s) is the function g(t). This
function is generally known as the weighting function of the system. In other
words, the transfer function is the Laplace-transformed weighting function according to
(2.2.5)
G(s) =  {g(t)} .
If the unit impulse (t) is taken as the input signal xa (t) for a system described
by the transfer function G(s), one obtains according to Eq. (2.2.3)
Xa (s) = G(s)  {(t)}
and after using the Laplace transform of the unit impulse (t) from Eq. (2.1.43)
Xa (s) = G(s)
or from Eq. (2.2.5)
xa (t) = g(t) .
This shows that the response to an unit impulse (t) is the weighting function.
Therefore the weighting function is also called the impulse response.
Another interpretation is when the system is excited by the input signal
e et sin(t) .
xe (t) = x
For the steady-state case one obtains for the output signal
e et sin(t + ( + j)) .
xas (t) = |G( + j)| x

(2.2.6)

This shows that the modulus |G( + j)| of the transfer function describes the
gain, and that ( + j) = arg G( + j) describes the phase shift of a sinusoidal function with the frequency and with increasing or decreasing amplitude
according to et .
Interactive Questions 2.2.1
Test your knowledge about impulse and step response
Interactive Questions 2.2.2
Test your knowledge about convolution

28

2.2.3

2 Basic Control Elements

Realisability and properness of transfer functions

It must be mentioned that a transfer function with m > n is physically not

realisable. The transfer function of an ideal dierentiator is described by G(s) = s
according to Eq. (2.1.6). Any transfer function with m > n can be decomposed
into
N (s)
N1 (s)
G(s) =
=
+ k0 + k1 s + . . . + kmm smn ,
D(s)
D(s)
where degree N1 (s) = n 1 and terms in s with positive powers also occur. Such
derivative elements would deliver for input signals of arbitrary high frequency
corresponding output signals of arbitrary high amplitude, which are physically
not realisable. The condition of realisability of the transfer function according to
Eq. (2.2.2) is
degree N (s) degree D(s) or m n .
(2.2.7)
This condition is also called as the condition of properness. If a transfer function
does not follow this condition, it is called an improper transfer function. It has
the property that G(s) as s . A realisable transfer function is called
as proper and it always follows G(s) k0 as s . A transfer function with
k0 = 0 is called as strictly proper.

2.2.4

Transfer functions with dead time

If a time delay or dead time Tt is introduced in the input signal xe (t) , one obtains
instead of Eq. (2.2.1) the dierential equation
n

i=0

di xa (t)
dj xe (t Tt )
=
bj
.
dti
dtj
m

ai

(2.2.8)

j=0

In this case taking the Laplace transformation gives the transcendental transfer
function
N (s) sTt
e
.
(2.2.9)
G(s) =
D(s)

2.2.5

Poles and zeros of the transfer function

In some cases (e.g. stability analysis) it is expedient to represent the rational

transfer function G(s) according to Eq. (2.2.2) in the factorised form
G(s) =

(s sZ1 ) (s sZ2 ) . . . (s sZm )

N (s)
= k0
.
D(s)
(s sP1 ) (s sP2 ) . . . (s sPn )

(2.2.10)

2.2 Transfer functions

29

For physical reasons only real coecients ai ,bj occur. Therefore the poles sPi and
the zeros sZj of G(s), respectively, can be real or complex conjugate pairs. The
terms zeros and poles are chosen, because the transfer function is zero at sZj and
innite at sPi . Zeros and poles can be graphically represented in the complex s
plane as shown in Figure 2.2.1. A linear time-invariant system without dead time
is described completely by the distribution of its poles and zeros and the gain
factor k0 .
p o le s
z e r o s

jw

Figure 2.2.1. Example of the pole and zero distribution of a rational transfer function
in the complex s plane

Moreover, the poles and zeros of a transfer function have a further signicance.
Observing a system without input (xe (t) 0) according to Eq. (2.2.1) and determining the time response xa (t) for the given n initial conditions, one has to
solve the associated homogeneous dierential equation
n

di xa (t)
ai
=0,
(2.2.11)
dti
i=0

which corresponds exactly to Eq. (2.1.35). For the approach xa (t) = est of
Eq. (2.2.11) one obtains for the solution in s the characteristic equation
n

ai si = 0 ,
(2.2.12)
P (s) =
i=0

which was already mentioned in Eq. (2.1.37). This relation can be directly determined by setting the denominator of G(s) to zero (D(s) = 0), as long as D(s)
and N (s) have no common factor. The zeros sk of the characteristic equation
are the poles sPi of the transfer function. As already shown in section 2.1.5 the
modes (i.e. xe (t) 0) are described by the characteristic equation, so that the
poles sPi of a transfer function contain all of this information.
The zeros of a transfer function are those values s = sZj for which |G(sZj )| = 0.
This means that the output signal Xa (s) does not contain any components which
depend on sZj . In order to explain this in more detail a stable system with a
transfer function according to Eq. (2.2.10) is excited by the input signal

30

2 Basic Control Elements

s Zj t

xe (t) = e

First for simplication the zero sZj = Zj is assumed to be real. For this case the
t
input signal is xe (t) = e Zj . Because |G(sZj )| = 0 one obtains from Eq. (2.2.6)
the steady-state output signal as
xas (t) = 0.
In the case of complex conjugate pairs of zeros sZj , sZj+1 = sZj both zeros have
to be taken into consideration in the input signal
s Zj t

xe (t) = e

sZ t

+e

Zj t

= 2e

cos t

).
2
Eq. (2.2.6) leads also to the result xas (t) = 0. This shows that a zero sZj of a
t
system blocks the transmission of the input signal e Zj .
Zj t

= 2e

sin(t +

Example 2.2.1
The mass-spring-damper mechanical system in Figure 2.2.2 with the mechanical
constants c1 =1, c2 =2, d = 1.5, m1 =1 and m2 =4 is excited by the force xe . The
transfer function between the force xe and the position xa can be shown to be
s2 + 1
s4 + 0.5s3 + 1.75s2 + 0.5s + 0.5
= j. If this system is excited by the sinusoidal input

G(s) =
which has the zeros sZ1,2
signal

Xe (s) =

1
1
= 2
(s sZ1 )(s sZ2 )
s +1

 1 {Xe (s)} = sin t ,

which is derived from this pair of zeros, the output signal xa (t) decays to zero
as shown in Figure 2.2.3b even though the input signal is a sinusoidal signal and
the mass m1 shows an undamped oscillation (see Figure 2.2.3c). The system does
not pass this oscillation to the mass m2 when the frequency matches the zeros


sZ1,2 .

2.2.6

Using transfer functions for calculations

For combinations of transfer functions simple rules for determining the resulting
transfer function can be derived. The combinations are of the type that transfer
function blocks are connected. In making any transfer function block connection
it is assumed that the connection does not load the block to which the connection
is being made.

2.2 Transfer functions

31

x
2

d
e

1
1

Figure 2.2.2. Mass-spring-damper mechanical system used for the interpretation of

zeros

a) Series connection
From the diagram in Figure 2.2.4 it follows that
Y (s) = G2 (s)Xe2 (s)
Xe2 (s) = Xa1 (s) = G1 (s) U (s)
Y (s) = G2 (s) G1 (s) U (s) .
The total transfer function of this series connection is
G(s) =

Y (s)
= G1 (s) G2 (s) .
U (s)

(2.2.13)

b) Parallel connection
For the output of both functions it follows from Figure 2.2.5 that
Xa1 (s) = G1 (s) U (s)
Xa2 (s) = G2 (s) U (s) .
The output of the total system is
Y (s) = Xa (s) = Xa1 (s) + Xa2 (s) = [G1 (s) + G2 (s)] U (s),
and from this the transfer function of a parallel connection is
G(s) =

Y (s)
= G1 (s) + G2 (s) .
U (s)

(2.2.14)

32

2 Basic Control Elements

x e(t)

(a )
1

0 .5
0

-0 .5

x a(t)

-1

(b )
1

0 .5
0

1 0 0

5 0

-0 .5

-1

x 1(t)

(c )
0

-5

Figure 2.2.3. Response to the sinusoidal input signal xa (t) = sin t (a), (b) position of
the mass m2 and (c) position of the mass m1

U = X
e

X
G 1(s)

a
1

= X
e

X
G 2(s)

a
2

= Y

Figure 2.2.4. Series connection of two transfer functions

c) Feedback loop
From Figure 2.2.6 the output is

Xa2 (s)] G1 (s) .

Y (s) = Xa (s) = [U (s) (+)

Since
Xa2 (s) = G2 (s) Y (s)
one obtains

G2 (s) Y (s)] G1 (s) ,

Y (s) = [U (s) (+)

and from this

Y (s) =

+
()

G1 (s)
U (s) .
G1 (s) G2 (s)

2.2 Transfer functions

33

U = X

(s )
1

a 1

X
+

G
2

= Y
a

(s )
X

a 2

Figure 2.2.5. Parallel connection of two transfer functions

The total transfer function of the feedback loop is

G(s) =

U = X
e

G1 (s)
Y (s)
= +
.
U (s)
1 () G1 (s) G2 (s)

(s )

G
2

(s )

X
a

(2.2.15)

= Y

(+ )

a 2

Figure 2.2.6. Feedback using two elements

As the output of G1 (s) is fed back via G2 (s) to the input, this is called feedback.
One has to distinguish between a positive feedback for positive adding of Xa2 (s)
and a negative feedback for negative adding of Xa2 (s).
Example 2.2.2
For the special case of G1 (s) being a pure amplier with a high gain K ,
one obtains for negative feedback
G(s) =

1
1
K

=
.
1
1 + K G2 (s)
G2 (s)
+ G2 (s)
K

The entire technique of operational ampliers is based on this principle. In a

feedback loop for which G1 (s) is an amplier with K then an element


G2 (s) can be used to realise any transfer function within certain limits.
Interactive Questions 2.2.3
Test your basic knowledge about transfer functions

34

2 Basic Control Elements

Interactive Questions 2.2.4

Test your knowledge about transfer functions and stability
Interactive Questions 2.2.5
Test your knowledge about transfer functions and electrical circuits
Interactive Questions 2.2.6
Test your knowledge about transfer functions and electrical elements
Interactive Questions 2.2.7
Test your knowledge about simple combinations of transfer functions
Interactive Questions 2.2.8
Test your knowledge about complicated combinations of transfer functions
Interactive Questions 2.2.9
Test your knowledge about transfer functions with dead time

2.3
2.3.1

Frequency Response
Denitions

In order to represent a transfer function G(s) in graphical form, there are dierent
possibilities. Representing |G(s)| and arg G(s) over the complex plane s = + j
two three-dimensional diagrams are needed. Figure 2.3.1 shows an example for
the magnitude of the transfer function
G(s) =

s2

s1
s1
=
.
+ s + 1.25
(s + 0.5 j)(s + 0.5 + j)

The two yellow peaks are at the pole positions where |G(0.5 j)| and
the blue negative peak is at the position of the zero s = 1 where |G(1)| = 0
or |G(1)|dB , respectively. This three-dimensional representation clearly
shows the inuence of the poles and zeros on the magnitude of the transfer function. Such diagrams are dicult to handle and are fortunately not necessary for
the analysis and design of control systems, because many properties can be investigated by using the frequency response G(j). The frequency response is just
the cut through the three-dimensional diagrams of G(s) at the imaginary axis
( = 0). For the magnitude diagram in Figure 2.3.1 the intersection of this cut
and the surface is at the red curve, which represents the magnitude of the frequency response. This curve is symmetric with respect to and therefore the
frequency response G(j) is only considered for positive frequencies .

2.3 Frequency Response

35

Figure 2.3.1. Magnitude in dB of a transfer function over the complex plane s = +j

Another interesting possibility to represent a transfer function G(s) in graphical

form, is to map the s plane into the complex G plane as shown in section A.2.
The result is a cartesian representation of a band of curves G( + j) as shown
in Figure A.2.1. For = 0, i.e. the special case s = j, the transfer function G(s)
migrates into the frequency response G(j) and we obtain the Nyquist plot of
the frequency response, which will be discussed later in section 2.3.2.
The frequency response G(j) is an immediately physically interpretable and
measurable quantity. To show this, the frequency response will be represented by
the complex entity
G(j) = R() + jI()
(2.3.1)
with the real part R() and the imaginary part I(), and by the amplitude
response A() and phase response () in polar notation as
G(j) = A() ej() .

(2.3.2)

e and freIf the system is excited by a sinusoidal input xe (t) with amplitude x
quency , i.e.
e sin t ,
(2.3.3)
xe (t) = x
then in the case of a linear continuous-time system the output signal will oscillate
in the steady state with the same frequency , but with another amplitude x
a

36

2 Basic Control Elements

x e(t)
x e

x e ( t ) = x^ e s i n w t
x e
t

(a )
x a(t)
x a

(b )

x a ( t ) = x^
a

s in ( w t- j )

j
x a

t
t j =

j
w

Figure 2.3.2. (a) Sinusoidal input signal xe (t) and corresponding output signal xa (t)
of a linear element in steady state (b) vector representation of both signals

and with a certain phase shift = t (Figure 2.3.2a). Both oscillations xe (t) and
xa (t) can be represented by two vectors of length proportional to their amplitudes
a according to Figure 2.3.2b. They are rotating with the phase shift
x
e and x
and the same speed . For the system output one gets
a sin(t + ) .
xa (t) = x

(2.3.4)

If this experiment is conducted for dierent frequencies ( = 0,1,2, . . .) with

a and the phase of the output
x
e = const, then one notices that the amplitude x
signal depend on the frequency . Therefore for each frequency one gets
a ( )
x
a, = x

and

= ( ) .

a () the amplitude of the frequency

Now from the ratio of the values x
e and x
response

x
a ()
= |G(j)| = R2 () + I 2 ()
(2.3.5)
A() =
x
e
can be represented as a function of frequency. Furthermore, the phase shift ()
will be represented as the phase of the frequency response. For the phase
() = arg G(j) = tan1

I()
R()

(2.3.6)

is valid, where the value of the tan1 function in the range 0 to 360 must be
found through the signs of R() and I().

2.3 Frequency Response

37

From this experiment it is obvious that the amplitude response A() and phase
response () of the frequency response G(j) can be directly measured by applying sinusoidal input signals xe (t) of dierent frequencies. The total frequency
response G(j) for all frequencies = 0 to describes completely the
behaviour of a linear continuous-time system, like the the transfer function G(s)
or the step response h(t). In some cases only some partial information is sucient
in order to reconstruct the total frequency response. As will be shown later only
a knowledge of the real part R() or of the amplitude response A() may be
necessary.
Between the representations of a linear system in the time domain and frequency
domain there are some general and simple relationships. E.g. on the basis of
the initial and nal value theorems two important relations between the transfer
function G(s), the Frequency response G(j) and the step response h(t) or H(s)
are valid:
lim h(t) = lim s H(s) = lim G(s) = lim G(j) ,

(2.3.7a)

lim h(t) = lim s H(s) = lim G(s) = lim G(j) .

(2.3.7b)

t0+

s
s0

s
s0

j
j0

The suppositions for the application of the initial and nal value theorems are
the existence of the related limits in the time domain according to section 2.1.3.

2.3.2

Nyquist plot of a frequency response

If in the experiment described above for each value of the value of

G(j ) = A( ) ej( )
is plotted in the complex G plane, one obtains the Nyquist plot of the frequency response, which is parameterised by the frequency . Figure 2.3.3 shows
such a curve drawn from measurements of eight frequencies. Using Eqs. (2.3.7a)
and (2.3.7b) the initial and nal values of the step response h(t) can be estimated from a Nyquist plot based on such measurements (Figure 2.3.4). If G(j)
is known analytically the step response can be calculated from G(s) by using


1 1
G(s) .
(2.3.8)
h(t) = 
s
The graphical representation of a frequency response using a Nyquist plot has
among other things the advantage that the frequency responses, both of a series
connection and a parallel connection of two systems, can be simply constructed
graphically. Here the vectors of the Nyquist plots for the same values of must
be taken. For the case of a parallel connection the vectors will be added, and for
the case of a series connection the vectors will be multiplied by multiplying the
lengths of the vectors and adding their angles according to Figure 2.3.5.

38

2 Basic Control Elements

j Im [G ]

p la n e

w = 0

R e [G ]

j (w n)
w n
w

Figure 2.3.3. Example of a Nyquist plot drawn from measurements

j Im [G ]

h (t)
b

G (jw )
R e [G ]

w = 0
a

Figure 2.3.4. Relations between initial and nal value of the frequency response G(j)
and the step response h(t)

Interactive Questions 2.3.1

Test your knowledge about frequency response and Nyquist plots
Interactive Questions 2.3.2
Test your knowledge about frequency response and Nyquist plots with other questions
Interactive Questions 2.3.3
Test your knowledge about frequency response and Nyquist plots with more demanding questions

2.3.3

Bode plot

If the absolute value A() and the phase () of the frequency responseG(j) =
A() ej() are separately plotted over the frequency according to Figure 2.3.6,
one obtains the amplitude response and the phase response. Both together

2.3 Frequency Response

X

(jw )

G
e

39

X
a

(jw )

G
1

(jw )
G

(a )
j Im [G ]
G
1

(jw )
w

w
G
1

w
1

(jw )+ G

j Im [G ]
G

(jw )
w

R e [G ]
1

(b )

(jw )
2

(jw )

(jw )

(jw )
R e [G ]

w
1

G
1

(jw ) G
2

(jw )

Figure 2.3.5. Addition (a) and multiplication (b) of frequency responses in Nyquist
plots

A (w )
3 0

A (w )

(a )

1 0

2 0

1 0
w
0
j (w )
0
-9 0

(b )

d B

2 0

4
2

6 [s
6 [s

-1

] 8

-1

] 8

1 0
1 0
w

w
0 ,1

1 [s

-1

j ( w ) 0 ,1
0

1 [s

-1

-1 0
-2 0

- 9 0

- 1 8 0

- 1 8 0

- 2 7 0

- 2 7 0

1 0

1 0
w

Figure 2.3.6. Plot of a frequency response: (a) linear, (b) logarithmic presentation (
on a logarithmic scale) (Bode plot)

are the frequency response characteristics. A() and are normally drawn with
a logarithm and () with a linear scale. This representation is called a Bode
diagram or Bode plot. Usually A() will be specied in decibels [dB]. By denition
this is
(2.3.9)
A()dB = 20 log10 A() [dB] .
The logarithmic representation of the amplitude response A()dB has consequently a linear scale in this diagram and is called the magnitude.

40

2 Basic Control Elements

The logarithmic representation has some advantages for series connections of

transfer functions. For complicated frequency responses, e.g. with
G(s) = K

(s sZ1 ) . . . (s sZm )
(s sP1 ) . . . (s sPn )

(2.3.10)

for s = j, it can be represented as series connections of the frequency responses

of simple elements of the form

K = const for i = 0
(2.3.11)
Gi (j) =
(j sZi ) for i = 1,2, . . . ,m
and
Gi (j) =

1
j sP

for

i = m + 1,m + 2, . . . ,m + n ,
= 1,2, . . . ,n .

(2.3.12)

From this it follows that

G(j) = KG1 (j) . . . Gm+n (j) ,
with

Gi (j) = Ai () eji ()

(2.3.13)

for i = 1, . . . ,m + n .

From the representation

G(j) = KA1 () A2 () . . . Am+n () ej[0 () + 1 () + 2 () + . . . + m+n ()]
(2.3.14)
and
A() = |K| |G1 (j)| |G2 (j)| . . . |Gm+n (j)| = |K| A1 () A2 () . . . Am+n () ,
respectively, one obtains the logarithmic characteristic of the magnitude
A()dB = 20 log10 [|K| A1 () A2 () . . . Am+n ()]
= |K|dB + A1 ()dB + A2 ()dB + . . . + Am+n ()dB

(2.3.15)

and the phase characteristic

() = 0 () + 1 () + 2 () + . . . + m+n ()

(2.3.16)

with 0 () = 0 for K > 0 and 0 () = 180 for K < 0. Thus, the frequency
response of a series connection is obtained by addition of the individual frequency
response characteristics.
A further advantage of this logarithmic representation is for the determination
of the inverse of a frequency response, that is for 1/G(j) = G1 (j). Here


20 log10 |G(j)|1 = 20 log10 |G(j)| = 20 log10 A()

2.3 Frequency Response

and

41
arg[G1 (j)] = arg[G(j)]

(2.3.17)

are valid, the curves of A() and () need only to be mirrored at the axes
20 log10 A = 0 (0-dB line) and = 0 .
Because of the double logarithmic and of the single logarithmic scale of A() and
(), respectively, the curve of A() and that of () can be approximated by line
segments. This approximation by lines allows the analysis and synthesis of control
systems using simple geometric constructions. They are important concepts for
the control engineer.

2.3.4

Some important transfer function elements

In the following the transfer function G(s), frequency response G(j), Nyquist
plot and the Bode diagram for some important elements will be derived and
shown.

2.3.4.1

The proportional element (P element)

The P element describes a pure proportional relationship between input and

output:
(2.3.18)
xa (t) = K xe (t) ,
with the arbitrary positive or negative constant K. K is called the gain of the P
element. The transfer function of this system is
G(s) = K .

(2.3.19)

The locus of the frequency response

G(j) = K

(2.3.20)

is for all frequencies a point on the real axis with distance K from the origin.
This means that the phase response () is zero for K > 0 or 180 for K < 0.
The characteristic of the magnitude is
A()dB = 20 log10 K = KdB = const .

2.3.4.2

The integrator (I element)

The dynamical behaviour of this element is described in the time domain by

42

2 Basic Control Elements

1
xa (t) =
TI

t
xe ( ) d + xa (0)

(2.3.21)

with input and output signals xe (t) and xa (t), respectively, and with the time constant TI , which has the dimension time. This element integrates the input signal.
Setting xa (0) = 0 one obtains by taking the Laplace transform of Eq. (2.3.21)
the transfer function of the I element as
1
,
s TI

(2.3.22)

1
1 j
=
e 2 .
j TI
TI

(2.3.23)

G(s) =
and with s = j the frequency response
G(j) =

From this the amplitude and phase responses

A() =

1
TI

and () =

follow. For the magnitude characteristic one obtains

A()dB = +20 log10

1
= 20 log10 TI .
TI

(2.3.24)

The graphical representation of Eq. (2.3.24) gives a line with a slope of -20dB/decade,
or equivalently -6dB/octave, in the Bode diagram, Figure 2.3.7a. The phase response is independent of the frequency. The Nyquist plot of the frequency response
1
G(j) = j
TI
coincides with the negative imaginary axis, as shown in Figure 2.3.7b.

2.3.4.3

The derivative element (D element)

The relationship between the input xe (t) and output xa (t) of a derivative element
is described by
d
(2.3.25)
xa (t) = TD xe (t) .
dt
This element dierentiates the input signal xe (t) and therefore is called a derivative element, or in short D element. The associated transfer function is
G(s) = s TD ,

(2.3.26)

2.3 Frequency Response

A (w )

43

(a )

d B

2 0

j Im [G ]

(b )

1 0
0
- 1 0
- 2 0

0 ,1
T I

1
T I

j (w )
0

w [s

-1

w
]

R e [G ]

G (jw )

w [s

-1

w
]

=
0

- 9 0
Figure 2.3.7. (a) Magnitude and phase response (b) Nyquist plot of the frequency
response of an integrator

and with s = j it follows that the frequency response

G(j) = j TD = TD ej 2 ,

(2.3.27)

A()dB = 20 log10 T

(2.3.28)

from which the magnitude

and the phase response

(2.3.29)
2
follow. It can easily be seen, that the I and D elements are related by an inversion.
Therefore the curves of the magnitude and phase response of the D element can
be found as shown above by mirroring those of the I element at the 0-dB
line and = 0 line, respectively. This is obvious from Eqs. (2.3.28) and (2.3.29).
Figure 2.3.8 shows the Bode diagram and the Nyquist plot of the frequency
response of the D element. The slope of the line A() is +20dB/decade and the
phase response is independent of the frequency.
() =

The D element discussed here is as already mentioned in section 2.2.3 an idealisation and therefore not a physically realisable element. For practical applications
the D element will be approximated by the DT1 element (see section 2.3.4.6).

2.3.4.4

The 1st-order lag element (PT1 element)

The 1st-order lag element or in short PT1 element is an element with an output
signal xa (t) that for a step input xe (t) has a certain initial slope and approaches

44

2 Basic Control Elements

A (w )

(a )

d B

2 0

0 ,1
T I

1 0
0
- 1 0
- 2 0

w
1

w [s

-1

(b )

j Im [G ]

G (jw )
]

T I

w =

R e [G ]

j (w )
+ 9 0
0

w [s

-1

Figure 2.3.8. (a) Magnitude and phase response (b) Nyquist plot of the frequency
response of a D element

asymptotically the nal value. An example of such an element is the simple RC

lag circuit shown in Figure 2.3.9. When at time t = 0 a voltage ue = 2V is
applied, the voltage ua at the output will approach exponentially with the time
constant T = RC the nal value ua = 2V.
i

u
e

x e(t)

ia =

u
a

x a(t)

Figure 2.3.9. Simple RC lag as an example of a 1st-order lag element

The circuit involves a single energy storage element, the capacitor C. The dierential equation of this RC lag is
xa (t) + RC x a (t) = xe (t) .

(2.3.30)

For the general notation of a PT1 element one obtains the dierential equation
xa (t) + T x a (t) = K xe (t) .

(2.3.31)

If the initial condition xa is set to zero, on taking the Laplace transform the
transfer function is
K
,
(2.3.32)
G(s) =
1+ sT

2.3 Frequency Response

45

and it follows that with s = j the frequency response is

G(j) = K

1
.
1 + j T

(2.3.33)

1
one obtains
T

With the breakpoint frequency B =

1j
1
B
G(j) = K
 2 .
=K

1+j
1+
B
B

(2.3.34)

The amplitude response is

1


A() = |G(j)| = K 

2

(2.3.35)

I()

= tan1
.
R()
B

(2.3.36)

1+

and the phase response is

() = tan1

The magnitude characteristic derived from Eq. (2.3.35) is


 2

.
A()dB = 20 log 10 K 20 log10 1 +
B
Eq. (2.3.37) can be asymptotically approximated by lines for:
a)

<< 1 by
B
A()dB 20 log 10 K = KdB

(initial asymptote) ,

with
() 0 ;
b)

>> 1 by
B
A()dB 20 log 10 K 20 log 10

()

.
2

with

(nal asymptote) ,

(2.3.37)

46

2 Basic Control Elements

In the Bode diagram A()dB can be consequently approximated by two lines. The
progression of the initial asymptote is horizontal, whereas the nal asymptote
shows a slope of -20dB/decade. The intersection of both lines (breakpoint) can
be determined from

20 log 10 K = 20 log10 K 20 log10

B
and provides the frequency
= B .
Therefore = B is called the breakpoint frequency. As can be easily seen from
Figure 2.3.10a, the deviation between A()dB and the asymptotes has a maximum
A ( w )

3 d B

d B

3 0
2 0
1 0

j ( w )
0

K d B

(a )
a sy m p to te s
-2 0 d B / d e c a d e

j Im [G ]

w
1

0 .1

(b )

w
1 0

- 4 5

K
B

R e [G ]
w = 0
G (jw )
w

w
w

w = w
B

- 9 0
Figure 2.3.10. (a) Magnitude and phase response (b) Nyquist plot of the frequency
response of a PT1 element

at the breakpoint for B . The exact values are

1
A(B ) = K
2

and

(B ) =

.
4

The deviation of the magnitude characteristic from the asymptotes for = B is

A(B )dB = 20 log10 2 dB 3 dB .

The deviations at the other frequencies are symmetrical from the breakpoint on
the logarithmic scale, as can be seen directly from Table 2.3.1. This is the reason
why the magnitude and phase characteristics can be easily constructed on a Bode
diagram. The phase curve is approximately zero up to one tenth of the breakpoint
frequency, and 90 beyond 10 time the breakpoint frequency. In between, it is
approximately a straight line with slope 45 per decade through 45 at the
breakpoint frequency.
As already shown in section A.2 the locus of the frequency response of a PT1
element is a semicircle, which starts for = 0 at K on the real axis and stops for
at the origin, as shown in Figure 2.3.10b.

2.3 Frequency Response

47

Table 2.3.1 Magnitude, phase response and deviation A() of the exact magnitude
from the asymptotes for a PT1 element with K = 1

B
0.03
0.1
0.25
0.5
0.76
1.0
1.31
2.0
4.0
10.0
30.0

A()dB

()

A()dB

0.0
0.04
0.26
0.97
2.00
3.00
4.35
6.99
12
20
30

- 2
- 6
- 14
- 27
- 37
- 45
- 53
- 63
- 76
- 84
- 88

0.00
0.04
0.26
0.97
2.00
3.00
2.00
0.97
0.26
0.04
0.00

The constant T = 1/B in the transfer function and on the frequency response,
respectively, is usually called the time constant of the PT1 element. It can be
determined also by the point of intersection of the line with the initial slope and
the horizontal line of the nal asymptote, h(), of the step response h(t) as
shown in Figure 2.3.11. This time constant can also be physically interpreted. It
h ( t )
h ( )
0 .6 3 h (

0
T

Figure 2.3.11. Graphical representation of the step response, h(t), of a PT1 element

is the time when the step response has reached approx. 63% of the nal value,
h(). K is similar to the P element called the gain of the PT1 element. It is
dened as the value of the frequency response at = 0.
DYNAST study example 2.3.1
Simple automobile model
DYNAST study example 2.3.2
D.C. motor - open loop

48

2 Basic Control Elements

2.3.4.5

The proportional plus derivative element (PD element)

The PD element shows both proportional and derivative behaviour and is described by the transfer function
G(s) = K(1 + s T ) .

(2.3.38)

Apart from the gain factor K this element is the inverse of the PT1 element.
Hence for K = 1 one obtains the magnitude and phase response by mirroring at
the 0-dB axis and the () = 0 line, respectively (compare Figure 2.3.12 with
2.3.10). The locus of the frequency response
G(j) = K(1 + j T )

(2.3.39)

is a semi-line, which starts for = 0 on the real axis at K and progresses parallel
to the imaginary axis for increasing values of .

(a )

A ( w )

(b )

d B

0
-1 0
-2 0

K d B

j Im [G ]
w

1
2 0 d B / d e c a d e

w B

G (jw )

j ( w )
9 0
K

R e [G ]

4 5
0

w =
0 .1

1 0
1

w
w B

Figure 2.3.12. (a) Magnitude and phase response (b) Nyquist plot of the frequency
response of a PD element

2.3.4.6

The derivative lag element (DT1 element)

This element has a step response which initially contains a step and then decreases
exponential to zero with a characteristic time constant as shown in Figure 2.3.13.
Figure 2.3.14 shows an example of such a system; a simple RC high pass lter.
The dierential equation of this circuit is
C

ua
d(ue ua )
=
,
dt
R

2.3 Frequency Response

49

h ( t )
K

t
T

Figure 2.3.13. Graphical representation of the step response, h(t), of a DT1 element

ie

ia =
C

u e= x e( t)

u
a

x a(t)

Figure 2.3.14. Simple RC high-pass circuit as an example of a DT1 element

which can be written as

ua + RC

due
dua
= RC
.
dt
R

Applying the Laplace transform one obtains the transfer function

G(s) =

RCs
Ua (s)
=
.
Ue (s)
1 + RCs

(2.3.40)

The generalised notation of the DT1 element is

G(s) = K

Ts
.
1 + Ts

(2.3.41)

For constructing the Bode plot one starts with the frequency response
G(j) = K
which on substituting B =

j T
,
1 + j T

(2.3.42)

1
gives
T

+j

B
=
K
G(j) = Kj

 .
B 1 + j
B
2
1+
B
B

(2.3.43)

50

2 Basic Control Elements

From Eq. (2.3.43) it follows that

A() = |G(j)| =

1


K
B
1+

and

2 ,


A()dB = 20 log10

+ 20 log10 K 20 log 10
B


1+

2

After some calculations the phase response can be shown to be

 

1
.
() = tan
2
e

(2.3.44)

(2.3.45)

Comparing Eq. (2.3.44) with Eqs. (2.3.37) and (2.3.40) shows that the magnitude
characteristic of the DT1 element can be obtained by adding the corresponding
curves of a PT1 element and a D element. The same also holds for the phase
response (). With this information the curves of the frequency response characteristics and of the Nyquist diagram can be simply constructed according to
Figure 2.3.15.

2.3.4.7

The 2nd-order lag element (PT2 element and PT2 S element)

A 2nd-order lag element is characterised by two independent energy storages.

Depending on the damping properties and the position of the poles of G(s),
respectively, one distinguishes between oscillating and aperiodic behaviour. If a
2nd-order lag element has a conjugate complex pair of poles, then it shows an
oscillating behaviour (PT2 S behaviour). If both poles are on the negative real
axis, then the element has a lag behaviour (PT2 behaviour).
The RLC network of Figure 2.3.16 is an example of such an element. From the
equation of the mesh
die
+ ua = ue
(2.3.46)
ie R + L
dt
with
dua
(2.3.47)
ie = C
dt
the dierential equation becomes
LC

dua
d2 ua
+ ua (t) = ue (t) .
+ RC
2
dt
dt

(2.3.48)

2.3 Frequency Response

A ( w )

51
a s y m p t o t e s

d B

(a )

3 d B

2 0
1 0

D T 1
0

d B

w
w B

- 1 0
- 2 0

P T 1
D

- 3 0

j Im [G ]

j ( w )

= 1 /

9 0

w =

4 5
0

w = w

0 .1

1 0

w
w
B

Figure 2.3.15. (a) Magnitude and phase responses (b) Nyquist plot of the frequency
response of a DT1 element

Interactive Questions 2.3.4

Test using other example
The transfer function is
G(s) =

1
Ua (s)
=
.
Ue (s)
1 + RC s + LC s2

(2.3.49)

For the 2nd-order lag element the general notation of the transfer function
ie

u e= x e(t)

ia =

u
a

x a(t)

Figure 2.3.16. Simple RLC network as an example of a 2nd-order lag element

52

2 Basic Control Elements

G(s) =

K
1 + T1 s + T22 s2

(2.3.50)

is chosen. Introducing terms which characterise the time behaviour, that is the
damping ratio
1 T1
(2.3.51)
=
2 T2
and the natural frequency (frequency of the undamped oscillation)
0 =

1
,
T2

(2.3.52)

one obtains from Eq. (2.3.50)

G(s) =

K
K
.
=
2
1 2
D(s)
1+
s+ 2 s
0
0

(2.3.53)

For s = j the frequency response



 
2

1
j2

0
0
K
=K
G(j) =
2 
2

2 .

2

1 + j2

+ 2
1
0 02
0
0


(2.3.54)

results. The amplitude response is

A() = 


1

K
2 2

+ 2
0

2

(2.3.55)

and the phase response

() = tan1

2
0
 2 .

1
0

(2.3.56)

Here the ambivalence of the tan1 function has to be observed. For the magnitude
characteristic one has from Eq. (2.3.55)

 2 2 

2

+ 2
.
(2.3.57)
1
A()dB = 20 log10 K 20 log 10
0
0
The progression of A()dB can be approximated by the following asymptotes:

2.3 Frequency Response

a) For

53

<< 1 by
0
A()dB 20 log 10 K

(initial asymptote) ,

with
() 0 .
b) For

>> 1 by
0


A()dB 20 log10 K 20 log10

2


20 log10 K 40 log10

(nal asymptote) ,

with
() .
In the Bode diagram the nal asymptote is a line with a slope of -40dB/decade.
The point of intersection of both asymptotes follows from
 

20 log 10 K = 20 log10 K 40 log10

0
to be the normalised frequency 0 . The exact value of A()dB may deviate considerably from the point of intersection at = 0 , because it is
A(0 )dB = 20 log 10 K 20 log 10 2
according to Eq. (2.3.55). For < 0.5 this value is above, for > 0.5 below the
asymptotes.
Figure 2.3.17 shows for 0 < 2.5 and K = 1 the magnitude and phase responses
in a Bode diagram. This graphical representation contains the cases of PT2 S and
PT2 behaviour, which will be discussed in section A.3.2. From Figure 2.3.17 it
can be seen that a maximum magnitude exists for some values of the damping
ratio . This maximum occurs at the so called resonant peak frequency p . A
detailed analysis of this resonance can be found in section A.3.1.
Figure 2.3.18 shows the Nyquist plots of 2nd-order lag elements with high and
low damping. From Eq. (2.3.54) it follows that for = 0 the real part of G(j) is
zero. Therefore the locus of the frequency response intersects with the imaginary
axis at = 0 independent of the values of , as can be seen in Figure 2.3.18.
The modes of a dynamical system are determined according to Eq. (2.2.12) by
the roots of the characteristic equation or by the poles of the transfer function,
respectively. From the characteristic equation of the 2nd-order lag element
P (s) D(s) = 1 +

1
2
s + 2 s2 = 0
0
0

(2.3.58)

54

2 Basic Control Elements

A (w )

d B

5 0

z =

4 0
3 0
2 0
1 0
0

- 1 0

0 .0
0 .0
0 .1
0 .2
0 .5
1 .0
1 .5
2 .0
2 .5
=

0 5
5
2 5
5

- 2 0
- 3 0
- 4 0

0 .1

0 .2

0 .3

0 .4

0 .6

0 .8

1 0

w
w 0

j (w )
0
- 3 0

z =

- 6 0
z =

- 9 0

- 1 2 0

- 1 5 0
- 1 8 0

0 .1

1 .0
1 .5
2 .0
2 .5
0 .2

0 .3

0 .4

0 .6

0 .8

0 .0
0 .0
0 .1
0 .2
0 .5

0 5
5
2 5
5

1 0

w
w 0

Figure 2.3.17.
Bode diagram of a 2nd-order lag element with the transfer function

G(s) = 1/ 1 + s2/0 + (s/0 )2

one obtains the poles of the transfer function as


s1,2 = 0 0 2 1 .

(2.3.59)

The oscillating behaviour of a 2nd-order lag element is dependent on the position

of the poles in the s plane. For a graphical analysis the step response h(t) is
a useful tool. Table 2.3.2 shows step responses and the corresponding poles for
dierent values of . These ve cases will be analysed and discussed in more detail
in section A.3.2.

2.3 Frequency Response

55

Table 2.3.2 Pole positions

 in the s plane and step responses for elements with the
transfer function G(s) = 1/ 1 + s2/0 + (s/0 )2
d a m p in g

p o le p o s it io n s
s

j
1

jw
d

z =
2

0 < z <

1 - z
0

1
s

s
2

s 1= s

1
s

=
0

w
0

z =
0

z =
2

5
t

1 0
w
0

h (t)
2

jw

+ j w 0
1

h (t)

s
2

1 0

0
s
s
2

- jw 0

0
0

w
jw
w

z <

- 1 <

1
w 0z

1 0

z =

=
A

s
1

0 .1

h (t)

jw

z >

n + 1 / 2

jw

z =

z =
n

h
0

0 .5

w 0z =

h (t)

ste p re sp o n se
h (t)

1 - z

w
0

h (t)
2

1 0
5

s
s
w 0z

0
0

1 0
w

56

2 Basic Control Elements

j Im [G ]

j Im [G ]

(a )

w =

w = 0

R e [G ]

R e [G ]
M

(b )

G (jw )

G (jw )
w
0

w
p

Figure 2.3.18. Nyquist plots of 2nd-order lag elements, (a) PT2 element, (b) PT2 S
element

2.3.4.8

Bandwidth of a system

An important term that has not been dened so far is the bandwidth of a system.
Lag elements with a proportional behaviour, e.g. PT1 , PT2 and PT2 S elements as
well as PTn elements (n PT1 elements in series connection), show a so-called lowpass property. This means that they pass low frequencies whereas high frequencies
in signals are attenuated by the strongly decreasing amplitude response of the
frequency response. In order to describe this behaviour the concept of bandwidth
is introduced. This is the frequency b at which the magnitude of the frequency
response is decreased by 3dB from the value of the initial horizontal asymptote,
see Figure 2.3.19.
A ( w )

d B

- 3 d B

w
p

w
0

w
b

Figure 2.3.19. Denition of the bandwidth b of systems with lags (p resonant peak
frequency, 0 natural frequency of the undamped oscillation; on logarithmic scale)

2.3.4.9

Example for the construction of a Bode plot

In section 2.3.3 it has been shown how Bode plots of a system with rational transfer functions can be generated by decomposing the system into smaller elements,

2.3 Frequency Response

57

such as those shown in Table A.8.3. In this section a detailed example will be
given.
A Bode plot is to be drawn for a system, which has only real poles and zeros.
The given transfer function
G(s) = K1

(s + 0.1) (s + 2)
,
s(s + 5) (s + 20)

with K1 = 890

is rearranged into the form

 s

 s
+1
+1
G(s) = K2 0.1   2
 ;
s
2
+1
+1
s
5
20

K2 = K1 /500 = 1.78 .

This system can now be decomposed into an integrator, two PD and two PT1
elements, that is
 s

1
1
K2  s
.
+1
+1 s
G(s) =
s
s
0.1
2
+1
+1
5
20
   

 
 
= G1 (s)
G2 (s)
G3 (s) G4 (s) G5 (s)
From this simple analysis the Bode plot can be determined by adding the Bode
plots of the elements G1 to G5 according to Figure 2.3.20a. In this gure the variable quantities Bi in the terms (s/Bi + 1)1 for i = 1,2,3 and 4 are the breakpoint frequencies in the Bode plot. Also shown in Figure 2.3.20 is the Nyquist
plot of the frequency response. Both representations of Figure 2.3.20 basically
contain the same information about the system.
Based on the example given above the procedure for constructing a Bode plot of
a given system can be recapitulated:
a) The given transfer function must be put into the form

G(s) = K

(s sZ1 ) . . . (s sZm )
=K
(s sP1 ) . . . (s sPn )

m 

!
sZ

=1
n
!
=1

(sP )

sP =0

m
!

1
sk

=1
n
!
=1

sP =0


s
1+
sZ


s
1+
sP

with k = 0,1,2, . . . ,
where possible poles of G(s) at sP = 0 will be specially considered according
to their multiplicity k.
b) Then for s = j the asymptotes of the elements will be used to approximate
A()dB and ().
c) If necessary corrections of the approximations can be performed.

58

2 Basic Control Elements

A (w )

d B

e x a c t

4 0

(a )

3 0
2 0
K

2 d B

1 0
0

0 .1

- 1 0

j Im [G ]
2

G
5

1 0 2 0

- 3 0

e1

e 2

e 3

G
e 4

1 0 0 [s
G

1 0 j

- 2 0

(b )

-1

] 1 0 0 0
0

- 1 0 j

j (w )

- 2 0 j

9 0
G

4 5
0
- 4 5
- 9 0

G
2

1 0

0 ,1

e x a c t
1 0 0 [s

1
G
G

-1

] 1 0 0 0

- 3 0 j

w
1 0

3
2

0 .5

R e [G ]

3 0

1 0

1 0 0
5 0

w [s -1 ]

0 ,1
2 6

2 0

G (jw )

0 .0 6

4
1

G
5

Figure 2.3.20. Representation of a dynamic system by two frequency response diagrams: (a) Bode plot, (b) Nyquist plot

2.3.5

Systems with minimum and non-minimum phase behaviour

Stable systems without dead time, which are described by the transfer function
G(s) =

N (s)
D(s)

and which do not have zeros in the right half s plane, are called minimum phase
systems. They are characterised by the fact that for a known amplitude response
A() = |G(j)| in the range of 0 < the corresponding phase response
() can be calculated from A() and that the value of () determined has its
minimum modulus for the given A().
If a transfer function has poles and/or zeros in the right half s plane then this
system shows non-minimum phase behaviour. The modulus of the phase response
is then always larger than that for a system with minimum phase behaviour, which
has the same amplitude response.
In order to illustrate the non-minimum phase behaviour, two systems will be
considered, which have in fact the same amplitude response A() but dier considerably in the phase response. The transfer functions of the two systems are
Ga (s) =

1 + sT
1 + sT1

and

Gb (s) =

1 sT
1 + sT1

2.3 Frequency Response

59

with 0 < T < T1 . The distributions of the poles and zeros of Ga (s) and Gb (s) in
the s plane is shown in Figure 2.3.21. The amplitude response of the corresponding

(a )

1
-

(b )

s p la n e

jw

s
1

T 1

s p la n e

jw

1
-

T 1

Figure 2.3.21. Distribution of poles and zeros in the s plane (a) Ga (s) and (b) Gb (s)

frequency responses Ga (j) (minimum phase system) and Gb (j) (non-minimum

phase system) is in both cases the same, as

1 + (T )2
.
Aa () = Ab () =
1 + (T1 )2
For the phase responses one obtains
a () = tan1

(T1 T )
1 + 2 T1 T

and

(T1 + T )
1 2 T1 T
a dierent result, which is shown in Figure 2.3.22. For b () the ambivalence
of the tan1 function has to be observed. Here the minimum phase response of
a () can be clearly seen.
b () = tan1

j (w )
0 .1
0
1
j

- 9 0
- 1 8 0

j
b

(w )

1 0
(w )

m im im u m

[s

-1

] w

p h a se

Figure 2.3.22. Phase response of two transfer functions with identical amplitude, but
with minimum and non-minimum phase behaviour: |a | < |b |

The transfer functions of non-minimum phase systems, like Gb (s), can always
been composed by a series connection of a minimum phase system and a pure
phase shift element, which are described by the transfer functions Ga (s) and
GA (s):

60

2 Basic Control Elements

Gb (s) = GA (s) Ga (s) .

(2.3.60)

A pure phase shift element, also called all-pass element, is characterised by the
fact, that the modulus of its frequency response GA (j) has a value of unity at
all frequencies. For this example one obtains
Gb (s) = GA (s) Ga (s)
or

1 sT 1 + sT
1 sT
=
.
1 + sT1
1 + sT 1 + sT1

From this, the transfer function of the all-pass element (1st order) is
GA (s) =

1 sT
,
1 + sT

which has the amplitude response

AA () = 1

(2.3.61)

and the phase response

A () = tan1

2T
= 2 tan1 T .
1 (T )2

This all-pass element has a phase response A () from 0 to -180 . The condition
of Eq. (2.3.61) is only fullled by systems for which the distribution of zeros of
the transfer function GA (s) in the s plane is symmetric to the j axis. This is
shown in Figure 2.3.23 for a stable 4th-order all-pass system.

jw
s

s
2

s
3

s 1

s p la n e
s 2

s 4
s

s 3

Figure 2.3.23. Poles (x) and zeros (o) of a 4th-order all-pass system

The transfer function of an nth-order all-pass system is

GA (s) =

(s s1 ) (s s2 ) . . . (s sn )
,
(s s1 ) (s s2 ) . . . (s sn )

(2.3.62)

where the corresponding poles si and zeros si for i = 1,2, . . . ,n only dier in the
signs of their real parts (Re si = Re si ).

2.3 Frequency Response

61

For minimum phase systems, as already mentioned and shown in section A.4,
the phase response () can be unambiguously determined from the amplitude
response A(). This is not valid for non-minimum phase systems. The verication
of whether a system is non-minimum or minimum phase can be easily estimated
from the progression of () and A()dB for high frequencies. For a minimum
phase system, which is described by the transfer function
G(s) =

N (s)
D(s)

with m = degree N (s) and n = degree D(s) one obtains for the phase
() = 90 (n m) .

(2.3.63)

For a low-pass system with non-minimum phase behaviour the modulus of this
phase value is always larger than that given by Eq. (2.3.63). In both cases the
magnitude response has a slope of
20(n m) dB /decade
for .

2.3.6

Systems with dead time

A typical system with non-minimum phase behaviour is the dead time element
(PTt element), which is described by the transfer function
G(s) = esTt .
The frequency response is

(2.3.64)

G(j) = ejTt

with the amplitude response

A() = |G(j)| = 1
and the phase response (in radians)
() = Tt .
The locus of G(j) is a circle around the origin, starting with = 0 on the real
axis at R() = 1 and perpetually traversing the circle with increasing values of
as shown in Figure 2.3.24.
DYNAST study example 2.3.3
Explanation of plotting frequency responses in DYNAST by way of an example
Problem 2.3.1
Frequency response - seven problems

62

2 Basic Control Elements

(a )

j (w )

j Im [G ]

1
w = 0

(b )
0 .1
1

1 0
w T t

- 1 0 0
- 2 0 0

R e [G ]

G (jw )

- 3 0 0
- 4 0 0

Figure 2.3.24. (a) Nyquist plot and (b) phase of a dead time element

2.4
2.4.1

Stability of linear control systems

Stable and unstable systems

The stability of control systems is an important property. Considering any bounded

input signal xe (t) of a system, and if the output signal xa (t) of the system to such
a signal is also bounded, then the system is called bounded-input-bounded-output
stable. If the output signal does not show this property, the system is unstable.
For illustration see Figure 2.4.1.

(a )

x a

(b )

x a

0
0

Figure 2.4.1. (a) Stable and (b) unstable system response xa (t) to a bounded input
signal xe (t)

2.4.2

Denition of stability and stability conditions

Because of its feedback structure a control system can become unstable, e.g. oscillations with increasing amplitudes in the signals can occur. In section 2.4.1 a
signal-based denition of stability is established, which relies on the boundedness
of the input-output signals. In this section we focus on a denition of stability for

2.4 Stability of linear control systems

63

linear systems that is independent of the input-output signals. First the following
denition is introduced:
A linear time-invariant system according to Eq. (2.2.3) is called (asymptotically) stable, if its weighting function decays to zero, i.e. if
lim g(t) = 0

(2.4.1)

is valid. If the modulus of the weighting function increases with increasing t to innity, the system is called unstable.
A special case is a system where the modulus of the weighting function
does not exceed a nite value as t or for which it approaches
a nite value. Such systems are called critically stable. Examples are
undamped PT2 S and I elements, see sections 2.3.4.2 and 2.3.4.7.
This denition shows that stability is a system property for linear systems. If
Eq. (2.4.1) is valid, then there exists no initial condition and no bounded input
signal which drives the output to innity. This denition can be directly applied to
the stability analysis of linear systems by determining the value of the weighting
function for t . If this value exists, and if it is zero, the system is stable.
However, in most cases the weighting function is not given in an explicit analytic
form and therefore it is costly to determine the nal value. The transfer function
G(s) of a system is often known and as it is the Laplace transform of the weighting
function g(t), there is an equivalent stability condition for G(s) according to
Eq. (2.4.1). The analysis of this condition see section A.5 shows that for the
stability analysis it is sucient to check the poles of the transfer function G(s)
of the system, that is the roots si of its characteristic equation
P (s) D(s) = a0 + a1 s + s2 s2 + . . . + an sn = 0 .

(2.4.2)

Now the following necessary and sucient stability conditions can be formulated:
a) Asymptotic stability
A linear system is only asymptotically stable, if for the roots si of its characteristic equation
Re si < 0

for all

si (i = 1,2, . . . ,n)

is valid, or in other words, if all poles of its transfer function lie in the left-half
s plane.
b) Instability
A linear system is only unstable, if at least one pole of its transfer function
lies in the right-half s plane, or, if at least one multiple pole (multiplicity
r 2) is on the imaginary axis of the s plane.

64

2 Basic Control Elements

c) Critical stability
A linear system is critically stable, if at least one single pole exists on the
imaginary axis, no pole of the transfer function lies in the right-half s plane,
and in addition no multiple poles lie on the imaginary axis.
It has been shown above that the stability of linear systems can be assessed
by the distribution of the roots of the characteristic equation in the s plane
(Figure 2.4.2). For control problems there is often no need know these root with
high precision. For a stability analysis it is interesting to know whether all roots
of the characteristic equation lie in the left-half s plane or not. Therefore simple
criteria are available for easily checking stability, called stability criteria. These
are partly in algebraic, partly in graphical form.
jw

s p la n e

jw

s p la n e

s
c ritic a lly s ta b le

s ta b le
s p la n e

jw

s
s 2
u n s ta b le

jw

s p la n e

s
u n s ta b le

Figure 2.4.2. Stability of a linear system discussed by the distribution of the roots of
the characteristic equation in the s plane

Interactive Questions 2.4.1

Test your knowledge about stability

2.4.3

Algebraic stability criteria

The algebraic stability criteria are based on the characteristic equation, Eq. (2.4.2),
of the system to be analysed. They contain algebraic conditions as inequalities
between coecients ai , which are only valid if all roots of the polynomial lie in
the left-half s plane.

2.4 Stability of linear control systems

2.4.3.1

65

The Hurwitz criterion

A polynomial
P (s) = a0 + a1 s + . . . + an sn

(2.4.3)

with k complex conjugate pairs of roots and (n 2k) real roots can always be
represented as
P (s) = an (s2 2d1 +d21 +12 ) . . . (s2 2dk +d2k +k2 )(ss2k+1 ) . . . (ssn ) . (2.4.4)
If all roots of the polynomial of P (s) are in the left-half s plane then for an > 0
all constants dj and sj in Eq. (2.4.4) are positive. From this follows that
all coecients aj of the polynomial P (s), which are products and sums of positive numbers, are also positive. This result is formulated in the so-called Stodola
criterion:
For the polynomial to have all roots with negative real parts it is necessary that
(2.4.5)
sgn a0 = sgn a1 = . . . = sgn an .
These conditions are also sucient for n = 1 and n = 2 as can be easily veried
by calculating the roots. However, for n 3 this is no longer the case.
Example 2.4.1
The polynomial with positive coecients
P (s) = s3 + s2 + 4s + 30 = (s + 3)(s2 2s + 10)
fullls the Stodola criterion, but not all the roots s1 = 1, s2,3 = 1 j3 have
negative real parts.


A polynomial for which all roots si (i = 1,2, . . . ,n) have negative real parts is
called Hurwitzian. Therefore, according to the stability conditions introduced in
section 2.4.2 a linear system is only asymptotically stable, if its characteristic
polynomial is Hurwitzian. The Hurwitz criterion for the coecients of a Hurwitz
polynomial is as follows:
A polynomial P (s) is Hurwitzian, if and only if for an > 0 all determinants
D1 = an1 > 0


a
an 
>0
D2 =  n1
an3 an2 


an1
an
0 

D3 = an3 an2 an1  > 0
an5 an4 an3 

66

2 Basic Control Elements

until

Dn1


an1

an3

=  .
 .

 0

an
an2
.
.
0


. . . 0 
. . . . 
. . . .  > 0
. . . . 
. . . a1 

(2.4.6)

Dn = a0 Dn1 > 0 .
are positive.
The following schema of the coecients can be used to build the Hurwitz determinants:
an
0
0
0
D1 an1
D2

an3

an2

an1

an

D3

an5

an4

an3

an2

an1

...

D4

an7

an6

an5

an4

an3

...

The Hurwitz determinants D are characterised by the diagonal coecients an1 ,

an2 , . . . ,an ( = 1,2, . . . ,n) and by the increasing indices from left to right.
Coecients with indices larger than n are set to zero. For applying this criterion all determinants until Dn1 have to be calculated. Calculation of the last
determinant Dn is trivial.
While for a 2nd-order system the conditions of the determinants are automatically
fullled as soon as the coecients a0 ,a1 ,a2 are positive, for a 3rd-order system
one obtains the Hurwitz conditions
D1 = a2 > 0


a2 a3 


D2 = 
a0 a1 

a2 a3

D3 = a0 a1
0 0

= a1 a2 a0 a3 > 0

0 
a2  = a0 D2 > 0 .
a0 

It goes without saying that the determinant conditions will be only applied if
the easily checkable conditions of Eq. (2.4.5) are fullled. The Hurwitz criterion
is not only practical for the stability analysis of a system with given coecients
ai , but also of a system with free parameters. This is the task when the range
of parameters must be determined for which the system is asymptotically stable.
Therefore the following example is given.

2.4 Stability of linear control systems

67

Example 2.4.2
Figure 2.4.3 shows a control loop, for which the range of K0 must be determined
such that the closed loop is asymptotically stable. The time constants T1 and T2
W
+

K 0
s
_

1 + T 1s

1 + T 2 s

Figure 2.4.3. Stability analysis of a simple control loop

of both lag elements are known and positive. With the transfer function of the
open loop
K0
s(1 + T1 s) (1 + T2 s)
K0
=
s + (T1 + T2 ) s2 + T1 T2 s3

G0 (s) =

one obtains for the closed-loop transfer function

GW (s) =

G0 (s)
Y (s)
=
W (s)
1 + G0 (s)

and by substituting G0 (s)

GW (s) =

K0
.
K0 + s + (T1 + T2 ) s2 + T1 T2 s3

The characteristic equation of the closed loop is

P (s) = K0 + s + (T1 + T2 ) s2 + T1 T2 s3 = 0 .
According to the Stodola and Hurwitz criteria the following conditions must be
met for asymptotic stability:
a) Coecients a0 = K0 , a1 = 1, a2 = (T1 + T2 ) and a3 = T1 T2 must be positive.
From this the lower limit K0 > 0 follows.
b) Furthermore
(a1 a2 a3 a0 ) > 0
must be valid.
With the coecients given above it follows that
T1 + T2 T1 T2 K0 > 0

68

2 Basic Control Elements

and for the upper limit of K0

K0 <

T1 + T2
.
T1 T2

The closed loop is asymptotically stable for the range

0 < K0 <

T1 + T2
.
T1 T2

2.4.3.2

Routh criterion

For given coecients ai of the characteristic equation the method of Routh, which
is an alternative to the method of Hurwitz, can be applied, see section A.6. Here
the coecients ai (i = 0,1, . . . ,n) will be arranged in the rst two rows of the
Routh schema, which contains n + 1 rows:
n

an

an2

an4

an6

...

n1

an1

an3

an5

an7

...

n2

bn1

bn2

bn3

bn4

... 0

n3
..
.

cn1
..
.

cn2

cn3

cn4

... 0

dn1

dn2

en1

en2

fn1

gn1

The coecients bn1 ,bn2 ,bn3 , . . . in the third row are the results from cross
multiplication the rst two rows according to
an1 an2 an an3
an1
an1 an4 an an5
=
an1
an1 an6 an an7
=
an1

bn1 =
bn2
bn3

..
..

2.4 Stability of linear control systems

69

Building the cross products one starts with the elements of the rst row. The
calculation of these b values will be continued until all remaining elements become
zero. The calculation of the c values are performed accordingly from the two rows
above as follows:
bn1 an3 an1 bn2
bn1
bn1 an5 an1 bn3
=
bn1
bn1 an7 an1 bn4
=
bn1

cn1 =
cn2
cn3
..
.

From these new rows further rows will be built in the same way, where for the
last two rows nally
en1 dn2 dn1 en2
fn1 =
en1
and
gn1 = en2
follows. Now the Routh criterion is:
A polynomial P (s) is Hurwitzian, if and only if the following three
conditions are valid:
a) all coecients ai (i = 0,1, . . . ,n) are positive,
b) all coecients bn1 , cn1 in the rst column of the Routh schema
are positive.
Example 2.4.3
P (s) = 240 + 110s + 50s2 + 30s3 + 2s4 + s5 .
The Routh schema is:
5

30

110

50

240

10

54

240

32.22

240

As in the rst row of the Routh schema a coecient is negative the system is
unstable.

70

2 Basic Control Elements

For proving instability it is sucient to build the Routh schema only until negative or zero value occurs in the rst column. In the example given above the
schema could have been stopped at the 5th row.
Another interesting property of the Routh scheme says, that the number of roots
with positive real parts is equal to the number of changes of sign of the values in
the rst column.

2.4.3.3

Nyquist criterion

This graphical method, which was originally developed for the stability analysis of
feedback ampliers, is especially suitable for dierent control applications. With
this method the closed-loop stability analysis is based on the locus of the openloop frequency response G0 (j). Since only knowledge of the frequency response
G0 (j) is necessary, it is a versatile practical approach for the following cases:
a) For many cases G0 (j) can be determined by series connection of elements
whose parameters are known.
b) Frequency responses of the loop elements determined by experiments or
G0 (j) can be considered directly.
c) Systems with dead time can be investigated.
d) Using the frequency response characteristic of G0 (j) not only the stability
analysis, but also the design of stable control systems can be easily performed.

2.4.3.4

Nyquist criterion using Nyquist plots

To derive this criterion one starts with the rational transfer function of the open
loop
N0 (s)
(2.4.7)
G0 (s) =
D0 (s)
and makes the following assumptions:
1. The polynomials N0 (s) and D0 (s) are relatively prime.
2.
degree N0 (s) = m n = degree D0 (s) ,
which is always valid for realisable systems, see section 2.2.3.

(2.4.8)

2.4 Stability of linear control systems

71

The poles i of the open loop are the roots of the characteristic equation
D0 (s) = 0.

(2.4.9)

For stability analysis just the poles i of the closed loop are of interest, i.e. the
roots of the characteristic equation, which are determined by setting the denominator of the closed-loop transfer function to zero. From this condition
1 + G0 (s) =

Ng (s)
N0 (s) + D0 (s)
=
=0
D0 (s)
D0 (s)

(2.4.10a)

and
P (s) Ng (s) = N0 (s) + D0 (s) = 0

(2.4.10b)

follows. Because of Eq. (2.4.8) degree Ng (s) = n is valid. Thus the function
G
(s) = 1 + G0 (s) must be investigated in more detail. The zeros of this function
match the poles of the closed loop and its poles match the poles of the open loop.
Therefore this function can be represented by
n
!

G (s) = 1 + G0 (s) =

(s i )

k0
i=1
n
!

(2.4.11)

(s i )

i=1

where i are the poles of the closed loop and i the poles of the open loop. With
respect to the position of the poles it is assumed according to Figure 2.4.4 that
s p la n e

jw

jw

s p la n e

n - P - m

m
P

n - N - n

o p e n lo o p

c lo s e d lo o p

Figure 2.4.4. Poles of the open and closed loop in the s plane (multiple poles are
counted according to their multiplicity)

a) from the n poles i of the closed loop

N are lying in the right-half s plane,
are on the imaginary axis, and

72

2 Basic Control Elements

(n N ) in the left-half s plane.

Accordingly,
b) from the n open-loop poles i
P are lying in the right-half s plane,
on the imaginary axis, and
(n P ) in the left-half s plane.
P and are assumed to be known. Then N and will be determined from the
knowledge about the frequency response locus of G0 (j). Therefore with s = j
the frequency response
G
(j) = 1 + G0 (j) =

Ng (j)
D0 (j)

(2.4.12)

is calculated, for which the phase response is given by

() = arg[G
(j)] = arg[Ng (j)] arg[D0 (j)] .
When traverses the range of 0 , the change in the phase =
() (0) consists of the parts from the polynomials Ng (j) and D0 (j) and
is given by
= g 0 .
Each root of the polynomials Ng (s) and D0 (s), respectively, provides for g
and 0 , respectively, an amount of +/2, if they lie in the left-half s plane,
and each root on the right-hand side of the imaginary axis provides an amount
of /2. These phase changes are continuous with respect to .
Each root j on the imaginary axis for > 0 causes during the traverse of j at
j a stepwise change of in the phase. This discontinuous part of the phase will
not be considered in what follows.
Using the terms given above, for the continuous part S of the phase change
one obtains
S = [(n N ) N ] /2 [(n P ) P ] /2
= (n 2N ) /2 (n 2P ) /2 ,
or
S = [2(P N ) + ] /2 .

(2.4.13)

2.4 Stability of linear control systems

73

If besides P and , s is also known, then from Eq. (2.4.13) it can be determined, whether N > 0 or/and > 0 is valid, i.e. whether and how many
closed-loop poles are in the right-half s plane and on the imaginary axis.
To determine S , the locus G
(j) = 1 + G0 (j) can be drawn on the Nyquist
diagram and the phase angle checked. Expediently one moves this curve by 1 to
the left in the G0 (j) plane. Thus for stability analysis of the closed loop the
locus G0 (j) of the open loop according to Figure 2.4.5 has to be drawn. Here
j Im [G ']

j Im [G

G ' p la n e

1 w

w = 0

R e [G ']

- 1

G 0 p la n e

w = 0

R e [G

j
G
0

(jw )

w
w

G '(jw )

Figure 2.4.5. Nyquist diagrams of G
(j) and G0 (j)

S is the continuous change in the angle of the vector from the so called critical
point (-1,j0) to the moving point on the locus of G0 (j) for 0 . Points
where the locus passes through the point (-1,j0) or where it has points at innity
correspond to the zeros and poles of G
(s) on the imaginary axis, respectively.
These discontinuities are not taken into account for the derivation of Eq. (2.4.13).
Figure 2.4.6 shows an example of a G0 (j) where two discontinuous changes of
j Im [G
0

G 0 p la n e

w
D

- 1
ta n g e n t

j 2
j

w
0

A
j 1

= 0
R e [G

B
Figure 2.4.6. Determination of continuous changes in the angle S

74

2 Basic Control Elements

the angle occur. Thereby the continuous change of the angle consists of three
parts
S = AB + CD + DO
= 1 (2 1 2 ) 2 = 2 .
The rotation is counter clockwise positive.
As the closed loop is only asymptotically stable for N = = 0, then from
Eq. (2.4.13) the general case of the Nyquist criterion follows:
The closed loop is asymptotically stable, if and only if the continuous
change in the angle of the vector from the critical point (-1,j0) to the
moving point of the locus G0 (j) of the open loop is
S = (P + /2) .

(2.4.14)

For the case with a negative gain K0 of the open loop the locus is rotated by 180
relative to the case with a positive K0 . The Nyquist criterion remains valid also
in the case of a dead time in the open loop.

2.4.3.5

Simplified forms of the Nyquist criterion

It follows from Eq. (2.4.14) that for an open-loop stable system, that is P = 0
and = 0, then S = 0. Therefore the Nyquist criterion can be reformulated
as follows:
If the open loop is asymptotically stable, then the closed loop is only
asymptotically stable, if the frequency response locus of the open loop
does neither revolve around or pass through the critical point (-1,j0).
Another form of the simplied Nyquist criterion for G0 (s) with poles at s = 0 is
the so called left-hand rule:
The open loop has only poles in the left-half s plane with the exception
of a single or double pole at s = 0 (P, I or I2 behaviour). In this case
the closed loop is only stable, if the critical point (-1,j0) is on the left
hand-side of the locus G0 (j) in the direction of increasing values of .
This form of the Nyquist criterion is sucient for most cases. The part of the locus
that is signicant is that closest to the critical point. For very complicated curves
one should go back to the general case. The left-hand rule can be graphically
derived from the generalised locus according to section A.2. The orthogonal (,)net is observed and asymptotic stability of the closed loop is given, if a curve with
< 0 passes through the critical point (-1,j0). Such a curve is always on the lefthand side of G0 (j).

2.4 Stability of linear control systems

2.4.3.6

75

The Nyquist criterion using Bode plots

Because of the simplicity of the graphical construction of the frequency response

characteristics of a given transfer function the application of the Nyquist criterion
is often more simple using Bode plots. The continuous change of the angle S of
the vector from the critical point (-1,j0) to the locus of G0 (j) must be expressed
by the amplitude and phase response of G0 (j). From Figure 2.4.7 it can be seen
C

j Im [G

- 1
+

w
R e [G

C
]
C

=
1

j Im [G

2
+

- 1

R e [G
0

w
0

(jw )
G
0

(jw )

Figure 2.4.7. Positive (+) and negative () intersections of the locus G0 (j) with the
real axis on the left-hand side of the critical point

that this change of the angle is directly related to the count of intersections of
the locus with the real axis on the left-hand side of the critical point between
(, 1). The Nyquist criterion can therefore also represented by the count of
these intersections if the gain of the open loop is positive.
Regarding the intersections of the locus of G0 (j) with the real axis in the range
(, 1), the transfer from the upper to the lower half plane in the direction of
increasing values are treated as positive intersections while the reverse transfer
are negative intersections (Figure 2.4.7). The change of the angle is zero if the
count of positive intersections S + is equal to the count of negative intersections
S . The change of the angle S depends also on the number of positive and
negative intersections and if the open loop does not have poles on the imaginary
axis, the change of the angle is
S = 2(C + C ) .
In the case of an open loop containing an integrator, i.e. a single pole in the
origin of the complex plane ( = 1), the locus starts for = 0 at j, where
an additional +/2 is added to the change of the angle. For proportional and
integral behaviour of the open loop
S = 2(C + C ) + /2

= 0,1

(2.4.15)

76

2 Basic Control Elements

w

j Im [G

- 1

d > C
-

1 / 2

j Im [G

R e [G
0

d < C
-

1 / 2

- 1

R e [G
0

Figure 2.4.8. Count of the intersections on the left-hand side of the critical point for
I2 behaviour of the open loop

is valid. In principle this relation is also valid for = 2, but the locus starts for
= 0 at + j (Figure 2.4.8), and this intersection would be counted as a
negative one if > 0, i.e. if the locus for small is in the upper half plane of the
real axis. But de facto there is for > 0 (and accordingly < 0) no intersection.
This follows from the detailed investigation of the discontinuous change of the
angle, which occurs at = 0. As only a continuous change of the angle is taken
into account and because of reason of symmetry the start of the locus at = 0
is counted as a half intersection, positive for < 0 and negative for > 0, which
is analogous to the denition given above (Figure 2.4.8). For continuous changes
of the angle
( = 2)
(2.4.16)
S = 2(C + C )
is valid. Comparing Eqs. (2.4.15) and (2.4.16), respectively, with Eq. (2.4.14) then
the Nyquist criterion can be formulated as:
The open loop with the transfer function G0 (s) has P poles in the lefthalf s plane and possibly a single ( = 1) or double pole ( = 2) at s = 0.
If the locus of G0 (j) has C + positive and C negative intersections
with the real axis to the left of the critical point, then the closed loop
is only asymptotically stable, if

for = 0,1

(2.4.17)
D = C C = 2

P + 1 for = 2
2
is valid. For the special case, that the open loop is stable (P = 0, = 0),
the number of positive and negative intersections must be equal.
From this it follows that the dierence of the number of positive and negative
intersections in the case of = 0,1 is an integer and for = 2 not an integer.
From this follows immediately, that for = 0,1 the number P is even, for = 2
the number P + 1 is uneven and therefore in all cases P is an even number, such
that the closed loop is asymptotically stable. This is only valid if D 1.

2.4 Stability of linear control systems

77

The Nyquist criterion can now be transferred directly into the representation
using frequency response characteristics. The magnitude response A0 ()dB , which
corresponds to the locus G0 (j), is always positive at the intersections of the
locus with the real axis in the range of (, 1). These points of intersection
correspond to the crossings of the phase response 0 () with lines 180 , 540
etc., i.e. a uneven multiple of 180 . In the case of a positive intersection of the
locus, the phase response at the (2k + 1) 180 lines crosses from below to top
and reverse from top to below on a negative intersection as shown in Figure 2.4.9.
In the following these crossings will be dened as positive (+) and negative (-)
A

(w )
0

d B

0
w

j 0(w )
0

- 1 8 0

_
+

Figure 2.4.9. Frequency response characteristics of G0 (j) = A0 () ej0 () and denition of positive (+) and negative (-) crossings of the phase response 0 () with the -180
line

crossings of the phase response 0 () over the particular (2k + 1) 180 lines,
where k = 0,1,2, . . . may be valid. If the phase response starts at -180 this point
is counted as a half crossing with the corresponding sign. Based on the discussions
above the Nyquist criterion can be formulated in a form suitable for frequency
response characteristics:
The open loop with the transfer function G0 (s) has P poles in the
right-half s plane, and possibly a single or double pole at s = 0. C +
are the number of positive and C of negative crossings of the phase
response 0 () over the (2k + 1) 180 lines in the frequency range
where A0 ()dB > 0 is valid. The closed loop is only asymptotically
stable, if

for = 0,1

D =C C = 2

P + 1 for = 2
2
is valid. For the special case of an open-loop stable system (P = 0,
= 0)
D = C + C = 0

78

2 Basic Control Elements

must be valid.

Table 2.4.1 shows some examples of the Nyquist criterion in the representation
using frequency response characteristics.
Finally the left-hand rule will be given using Bode diagrams, because this version
is for the most cases sucient and simple to apply.
The open loop has only poles in the left-half s plane with the exception
of possibly one single or one multiple pole at s = 0 (P, I or I2 behaviour).
In this case the closed loop is only asymptotically stable, if G0 (j) has
a phase of 0 > 180 for the crossover frequency C at A0 (C )dB = 0.
This stability criterion oers the possibility of a practical assessment of the quality of stability of a control loop. The larger the distance of the locus from the
critical point the farther is the closed loop from the stability margin. As a measure of this distance the terms gain margin and phase margin are introduced
according to Figure 2.4.10. The phase margin
1 - A

- 1
w

j Im [G
P

j
P

]
0

(a )
A

(w )

(b )

d B

w
0

1
R e [G
C

w
C

(jw )

A
C

j 0(w )
0
- 1 8 0

P
P

d B

w
C

Figure 2.4.10. Phase and gain margin C and AP or APdB , respectively, in the (a)
Nyquist diagram and (b) Bode diagram

C = 180 + 0 (C )

(2.4.18)

is the distance of the phase response from the -180 line at the crossover frequency
C , i.e. at the crossing of the magnitude response with the 0dB line (|G0 | = 1).
The gain margin
(2.4.19)
APdB = A0 (P )dB
is the distance of the magnitude response from the 0dB line at the phase of
0 = 180 .
A well damped control system should yield the following characteristics:

2.4 Stability of linear control systems

79

Table 2.4.1 Examples of stability analysis using the Nyquist criterion with frequency
response characteristics

No.

Bode Diagram
( w )

A
0

Stability Analysis

d B

1
j

( w )
0
0

( w )
0

d B

- 1 8 0

(w )
0

S + = 3/2
S = 1

D = 1/2
P =0
w

( w )
0
0

_
1

j
0

( w )
0

(w )
0
w

j
0

P +1
: stable if
2
2 poles
in the
origin

S+ = 0
S = 1

D = 1
P =0

D
= P/2: unstable

d B

D =

d B

- 1 8 0
A

D
= P/2: unstable

- 1 8 0
A

S+ = 1
S = 2

D = 1
P =2

(w )
0

- 1 8 0

S+ = 0
S = 0

D = 0
P =0

D = P/2: stable

80

2 Basic Control Elements


APdB =
C =

12 dB to 20 dB for command response

3.5 dB to 9.5 dB for disturbance response
40
20

to
to

60
50

for command response

for disturbance response .

The crossover frequency C is a measure of the dynamical quality of the control

loop. The higher C the higher the bandwidth of the closed loop, and the faster
the reaction on command inputs or disturbances. As the bandwidth that frequency is understood, at which the magnitude A() of the closed-loop frequency
response has fallen o approximately to zero.
Interactive Questions 2.4.2
Test your knowledge about gain and phase margin
Problem 2.4.1
Stability - three questions

2.5 The root-locus method

2.5
2.5.1

81

The root-locus method

Introduction and basic ideas

A designer can determine whether his design for a control system meets the
specications if he knows the desired time response of the controlled variable. By
deriving the dierential equations for the control system and solving them, an
accurate solution of the systems performance can be obtained, but this approach
is not feasible for other than simple systems. It is not easy to determine from this
solution just what parameters in the system should be changed to improve the
response. A designer wishes to be able to predict the performance by an analysis
that does not require the actual solution of the dierential equations.
The rst thing that a designer wants to know about a given system is whether
or not it is stable. This can be determined by examining the roots obtained from
the characteristic equation
(2.5.1)
1 + G0 (s) = 0
of the closed loop. The work involved in determining the roots of this equation can
be avoided by applying the Hurwitz or Routh criterion as shown in section 2.4.3.
Determining in this way whether the system is stable or unstable does not satisfy
the designer, because it does not indicate the degree of stability of the system,
i.e., the amount of overshoot and the settling time of the controlled variable for
a step input. Not only must the system be stable, but the overshoot must be
maintained within prescribed bounds and transients must die out in a suciently
short time.
The root-locus method described in this section not only indicates whether a
system is stable or unstable but, for a stable system, also shows the degree of
stability. The root locus is a plot of the roots of the characteristic equation of
the closed loop as a function of the gain. This graphical approach yields a clear
indication of the eect of gain adjustment with relatively small eort.
With this method one determines the closed-loop poles in the s plane these are
the roots of Eq.(2.5.1) by using the known distribution of the poles and zeros
of the open-loop transfer function G0 (s). If for instance a parameter is varied,
the roots of the characteristic equation will move on certain curves in the s plane
as shown by the example in Figure 2.5.1. On these curves lie all possible roots
of the characteristic equation for all values of the varied parameter from zero to
innity. These curves are dened as the root-locus plot of the closed loop. Once
this plot is obtained, the roots that best t the system performance specications
can be selected. Corresponding to the selected roots there is a required value of
the parameter which can be determined from the plot. When the roots have been
selected, the time response can be obtained. Since the process of nding the root
locus by calculating the roots for various values of a parameter becomes tedious,

82

2 Basic Control Elements

jw

s p la n e

j2
p
3
2

0 .5

0 .5

-1

-2

0
s
-j

2
3
p

-j2

Figure 2.5.1. Plot of all roots of the characteristic equation s2 + 2s + p = 0 for 0

p < . Values of p are red and underlined.

a simpler method of obtaining the root locus is desired. The graphical method
for determining the root-locus plot is shown in the following.
An open-loop transfer function with k poles at the origin of the s plane is often
described by
G0 (s) =

K0 1 + 1 s + . . . + m sm
sk 1 + 1 s + . . . + nk snk

mn,

(2.5.2)

where K0 is the gain of the open loop. In order to represent this transfer function
in terms of the open-loop poles and zeros it is rewritten as
m
!

G0 (s) =

=1
k0 !
n

(s sZ )

(s sP )

= k0 G(s)

(2.5.3a)

=1

or

m
!

G0 (s) = k0

=1

(sZ )

nk
!

=1
sP
= 0

(sP )

m
!

1
sk

=1
nk
!


s
1+
sZ


s
1+
sP

(2.5.3b)

=1
sP
= 0

with k0 > 0 and sZ
= sP . The relationship between the factor k0 and the
open-loop gain K0 is

2.5 The root-locus method

83
m
!

K0 = k0

=1

(sZ )

nk
!

=1
sP
= 0

(sP )

1
.
sk

(2.5.4)

The characteristic equation of the closed loop using Eq. (2.5.3a) is

1 + k0 G(s) = 0
or
G(s) =

(2.5.5a)

1
.
k0

(2.5.5b)

All complex numbers si = si (k0 ), which full this condition for 0 k0 ,

represent the root locus.
From the above it can be concluded that the magnitude of k0 G(s) must always
be unity and its phase angle must be an odd multiple of . Consequently, the
following two conditions are formalised for the root locus for all positive values
of k0 from zero to innity:
a) Magnitude condition:
|G(s)| =

1
k0

(2.5.6)

b) Angle condition
(s) = arg G(s) = 180 (2 k + 1)

for

k = 0,1,2, . . .
k0 0

(2.5.7)

In a similar manner, the conditions for negative values of k0 ( k0 < 0) can

be determined. The magnitude conditions is the same, but the angle must satisfy
the
c) Angle condition
(s) = arg G(s) = k 360

for

k = 0,1,2, . . .
k0 < 0 .

(2.5.8)

Apparently the angle condition is independent of k0 . All points of the s plane

that full the angle condition are the loci of the poles of the closed loop by
varying k0 . The calibration of the curves by the values of k0 is obtained by the
magnitude condition according to Eq. (2.5.6). Based upon this interpretation of
the conditions the root locus can constructed in a graphical/numerical way.
Once the open-loop transfer function G0 (s) has been determined and put into
the proper form, the poles and zeros of this function are plotted in the s plane.

84

2 Basic Control Elements

Example 2.5.1
Consider the example
G0 (s) =

k0
K0
=
s(s + 2)
(s sP1 ) (s sP2 )

with sP1 = 0, sP2 = 2 and k0 = K0 . The poles of the closed-loop transfer

function
K0
GW (s) = 2
s + 2s + K0
are the roots s1 and s2 of the characteristic equation
P (s) = s2 + 2s + K0 = 0
and are given by
s1,2 = 1

1 K0 .

As s1 = sP1 = 0 and s2 = sP2 = 2 it can be seen that for K0 = 0 the poles of

the closed loop transfer function are identical with those of the open-loop transfer
function G0 (s). For other values K0 the following two cases are considered:
a) K0 1: Both roots s1 and s2 are real and lie on the real axis in the range
of 2 1 and 1 0;
b) K0 > 1: The roots s1 and s2 are conjugate complex with the real part
Res1,2 = 1, which does not depend on K0 , and the imaginary part Im s1,2 =
K0 1.
The curve has two branches as shown in Figure 2.5.2. At (sP1 +sP2 )/2 = 1 is the
jw
2

0
s
P
2

-2

j
2

j
j0 .5
j 1

1
-1

s p la n e

0
P

0
s

-j0 .5

Figure 2.5.2. Root locus of a simple second-order system

breakaway point of the two branches. Checking the angle condition the condition

2.5 The root-locus method

85


(s) = arg{G(s)} = arg

1
s(s + 2)

= arg s arg(s + 2) = 180 (2k + 1)

!

must be valid. The complex numbers s and (s + 2) have the angles 1 and 2 and
the magnitudes |s| and |s + 2|. The triangle (2, 0, 1 + j) in Figure 2.5.2 yields
the angle condition. Evaluating the magnitude condition according to Eq. (2.5.6)




1
= 1

|G(s)| = 
s(s + 2)  K0
one obtains the value K0 on the root locus. E.g. for s = 1 + j the gain of the
open loop is
K0 = |s(s + 2)|s = 1 + j = 2 .
The value of K0 at the breakaway point sB = 1 is
K0 = | 1(1 + 2)| = 1 .


Table 2.5.1 shows further examples of some 1st- and 2nd-order systems.
Table 2.5.1 Root loci of 1st- and 2nd-order systems

G0 (s)

root locus

G0 (s)

root locus

jw

k0
s

jw

k0
(s + 1 )2 + 12

jw 1

- s
1

jw

jw
k0
(s sP1 )(s sP2 )

k0
s2
s

jw

k0
s sP1

k0
s2 + 12

s
P

s
1

jw 1
- jw 1

- jw 1

jw
s

k0 (s sZ1 )
(s sP1 )
|sZ1 | > |sP1 |
k0 (s sZ1 )
(s sP1 )
|sZ1 | < |sP1 |

s
P

s
P

s
1

jw
s
Z

s
1

s
1

jw
P
1

s
Z

86

2 Basic Control Elements

2.5.2

General rules for constructing root loci

To facilitate the application of the root-locus method for systems of higher order
than 2nd, rules can be established. These rules are based upon the interpretation
of the angle condition and the analysis of the characteristic equation. The rules
presented aid in obtaining the root locus by expediting the manual plotting of the
locus. But for automatic plotting using a computer these rules provide checkpoints
to ensure that the solution is correct.
Though the angle and magnitude conditions can also be applied to systems having
dead time, in the following we restrict to the case of the open-loop rational transfer
functions according to Eq. (2.5.3a)
(s sZ1 ) (s sZ2 ) . . . (s sZm )
, k0 0
(s sP1 ) (s sP2 ) . . . (s sPn )

(2.5.9a)

b0 + b1 a + . . . + bm1 sm1 + sm
N0 (s)
.
= k0
n1
n
a0 + a1 s + . . . + an1 s
+s
D0 (s)

(2.5.9b)

G0 (s) = k0
or
G0 (s) = k0

As this transfer function can be written in terms of poles and zeros sP and sZ
( = 1,2, . . . n; = 1,2, . . . ,m) G0 (s) can be represented by their magnitudes and
angles
|s sZ1 | ejZ1 |s sZ2 | ejZ2 . . . |s sZm | ejZm
G0 (s) = k0
|s sP1 | ejP1 |s sP2 | ejP2 . . . |s sPn | ejPn
or

n
m

m
!

Z
P
|s sZ | j

G0 (s) = k0

=1
n
!
=1

=1

=1

(2.5.10)

|s sP |

From Eq. (2.5.6) the magnitude condition

m
!
=1
n
!
=1

|s sZ |
=
|s sP |

1
k0

(2.5.11)

and from Eq. (2.5.7) the angle condition

(s) =

=1

=1

P = 180 (2 k + 1)

for

k = 0,1,2, . . .

(2.5.12)

2.5 The root-locus method

87
s - s
Z

j
s

s
P

jw

Z
1

j
P
2

j
s
Z

s p la n e

P
1

s
2

Figure 2.5.3. Pole-zero diagram for construction of the root locus

follows. Here Z and P denote the angles of the complex values (s sZ )

and (s sP ), respectively. All angles are considered positive, measured in the
counterclockwise sense. If for each point the sum of these angles in the s plane
is calculated, just those particular points that full the condition in Eq. (2.5.12)
are points on the root locus. This principle of constructing a root-locus curve
as shown in Figure 2.5.3 is mostly used for automatic root-locus plotting.
In the following the most important rules for the construction of root loci for
k0 > 0 are listed:
Rule 1 Symmetry As all roots are either real or complex conjugate pairs so
that the root locus is symmetrical to the real axis.
Rule 2 Number of branches The number of branches of the root locus is
equal to the number of poles n of the open-loop transfer function.
Rule 3 Locus start and end points The locus starting points (k0 = 0) are
at the open-loop poles and the locus ending points (k0 = ) are at the openloop zeros. (nm) branches end at innity. The number of starting branches
from a pole and ending branches at a zero is equal to the multiplicity of the
poles and zeros, respectively. A point at innity is considered as an equivalent
zero of multiplicity equal to n m.
Rule 4 Real axis locus If the total number of poles and zeros to the right of
a point on the real axis is odd, this point lies on the locus.
Rule 5 Asymptotes
slope of

There are n m asymptotes of the root locus with a

k = arg s =

180 (2k + 1)
.
nm

(2.5.13)

88

2 Basic Control Elements

For (nm) = 1,2,3 and 4 one obtains the asymptote congurations as shown
in Figure 2.5.4.
jw

jw
s

jw

- m

s
- m

jw

s a
n

s
- m

s a
- m

Figure 2.5.4. Asymptote congurations of the root locus

Rule 6 Real axis intercept of the asymptotes The real axis crossing (a , j0)
of the asymptotes is at

n
m

Re sP
Re sZ .
(2.5.14)
a =

nm
=1

=1

Rule 7 Breakaway and break-in points on the real axis At least one breakaway or break-in point (B , j0) exists if a branch of the root locus is on the
real axis between two poles or zeros, respectively. Conditions to nd such
real points are based on the fact that they represent multiple real roots. In
addition to the characteristic equation (2.5.1) for multiple roots the condition
d
d
[1 + G0 (s)] =
G0 (s) = 0 .
(2.5.15)
ds
ds
must be fullled, which is equivalent to
n
m

1
1
=
(2.5.16)
s sP
s sZ
=1

=1

for s = B . If there are no poles or zeros, the corresponding sum is zero.

Rule 8 Complex pole/zero angle of departure/entry The angle of departure of pairs of poles with multiplicity rP is

n
m

P +
Z 180 (2k + 1)
(2.5.17)
P ,D =

rP

=1
=1

=

and the angle of entry of the pairs of zeros with multiplicity rZ

m
n

Z +
P 180 (2k + 1) .
Z ,E =

rZ

=1
=1

=

(2.5.18)

2.5 The root-locus method

89

Rule 9 Root-locus calibration

mined by using

The labels of the values of k0 can be deter-

k0 =

n
!
=1
m
!
=1

|s sP |
.

(2.5.19)

|s sZ |

For m = 0 the denominator is equal to one.

Rule 10 Asymptotic stability The closed loop system is asymptotically stable for all values of k0 for which the locus lies in the left-half s plane. From the
imaginary-axis crossing points the critical values k0crit can be determined.
The rules shown above are for positive values of k0 . According to the angle condition of Eq. (2.5.8) for negative values of k0 some rules have to be modied. In
the following these rules are numbered as above but labelled by a *.
Rule 3* Locus start and end points The locus starting points (k0 = 0)
are at the open-loop poles and the locus ending points (k0 = ) are
at the open-loop zeros. (n m) branches end at innity. The number of
starting branches from a pole and ending branches at a zero is equal to
the multiplicity of the poles and zeros, respectively. A point at innity is
considered as an equivalent zero of multiplicity equal to n m.
Rule 4* Real axis locus If the total number of poles and zeros to the right
of a point on the real axis is even including zero, this point lies on the locus.
Rule 5* Asymptotes
slope of

There are n m asymptotes of the root locus with a

k = arg s =

360 k
.
nm

Rule 8* Complex pole/zero angle of departure/entry

parture of pairs of poles with multiplicity rP is

n
m

P +
Z 360 k
P ,D =

rP

=1
=1

(2.5.20)
The angle of de-

(2.5.21)

=

and the angle of entry of the pairs of zeros with multiplicity rZ

m
n

Z +
P 360 k .
Z ,E =

rZ

=1
=1

=

(2.5.22)

90

2 Basic Control Elements

The root-locus method can also be applied for other cases than varying k0 . This
is possible as long as G0 (s) can be rewritten such that the angle condition according to Eq. (2.5.12) and the rules given above can be applied. This will be
demonstrated in the following two examples.
Example 2.5.2
Given the closed-loop characteristic equation
a0 + a1 s + . . . + an1 sn1 + sn = 0 ,
the root locus for varying the parameter a1 is required. The characteristic equation is therefore rewritten as
s
=0.
1 + a1
a0 + a2 s2 + . . . + sn
This form then correspondents to the standard form
1 + G0 (s) = 1 + a1

N0 (s)
=0
D0 (s)

to which the rules can be applied.

Example 2.5.3
Given the closed-loop characteristic equation
s3 + (3 + ) s2 + 2s + 4 = 0 ,
it is required to nd the eect of the parameter on the position of the closedloop poles. The equation is rewritten into the desired form
1+

s2
=0.
s3 + 3s2 + 2s + 4

Using the rules 1 to 10 one can easily predict the geometrical form of the root
locus based on the distribution of the open-loop poles and zeros. Table 2.5.2
shows some typical distributions of open-loop poles and zeros and their root loci.
For the qualitative assessment of the root locus one can use a physical analogy.
If all open-loop poles are substituted by a negative electrical charge and all zeros
by a commensurate positive one and if a massless negative charged particle is
put onto a point of the root locus, a movement is observed. The path that the
particle takes because of the interplay between the repulsion of the poles and
the attraction of the zeros lies just on the root locus. Comparing the root locus
examples 3 and 9 of Table 2.5.2 the repulsive eect of the additional pole can
be clearly seen.

2.5 The root-locus method

91

Table 2.5.2 Typical distributions of open-loop poles and zeros and the root loci
N o .

r o o t lo c u s

N o .

r o o t lo c u s

jw

jw

9
s

jw

jw

2
s

1 0
s

jw

jw

1 1
s

jw

jw

1 2
s

jw

jw
5

1 3
s

jw
6

jw
1 4
s

jw

jw
7
s

1 5

jw
8

jw
1 6
s

92

2.5.3

2 Basic Control Elements

Example of an application of the root-locus method

The systematic application of the rules from section 2.5.2 for the construction
of a root locus is shown in the following non-trivial example for the open-loop
transfer function
k0 (s + 1)
G0 (s) =
.
(2.5.23)
s(s + 2) (s2 + 12s + 40)
The degree of the numerator polynomial is m = 1. This means that the transfer
function has one zero (sZ1 = 1). The degree of the denominator polynomial
is n = 4 and we have the four poles (sP1 = 0, sP2 = 2, sP3 = 6 + j2,
sP4 = 6 j2). First the poles (x) and the zeros (o) of the open loop are drawn
on the s plane as shown in Figure 2.5.5. According to rule 3 these poles are
just those points of the root locus where k0 = 0 and the zeros where k0 .
We have (n m) = 3 branches that go to innity and the asymptotes of these
three branches are lines which intercept the real axis according to rule 6. From
Eq. (2.5.14) the crossing is at
a =

13
(0 2 6 6) (1)
= = 4.33
3
3

(2.5.24)

and the slopes of the asymptotes are according to Eq. (2.5.13)

k =

180 (2k + 1)
= 60 (2k + 1) k = 0,1,2, . . .
3
i.e.

0 = 60 ,

1 = +180 ,

(2.5.25)

2 = 60 .

The asymptotes are shown in Figure 2.5.5 as blue lines. Using Rule 4 it can be
checked which points on the real axis are points on the root locus. The points
with 1 < < 0 and < 2 belong to the root locus, because to the right of
them the number of poles and zeros is odd. According to rule 7 breakaway and
break-in points can only occur pairwise on the real axis to the left of -2. These
points are real solutions of the Eq. (2.5.16). Here we have
1
1
1
1
1
+
+
+
=
s s + 2 s + 6 j2 s + 6 + j2
s+1

(2.5.26)

or
3s4 + 32s3 + 106s2 + 128s + 80 = 0 .
This equation has the solutions sB1 = 3.68, sB2 = 5.47 and sB3.4 = 0.76
j0.866. The real roots sB1 = 3.68 and sB2 = 5.47 are the positions of the
breakaway and the break-in point. The angle of departure P3 ,D of the root locus
from the complex pole at sP3 = 6 + j2 can be determined from Figure 2.5.6
according to Eq. (2.5.17):
P3 ,D = 90 153.4 161.6 + 158.2 180 (2k + 1)
= 246.8 + 180 = 66.8 .

(2.5.27)

2.5 The root-locus method

93
jw
6 4 4 .4

j7
j6

j4

j P
P

1 0 0
3 ,D

j2
3

- 6 6 .8
s

1 0 0
- 6

s
s
P

s a
B

P
B

1 0 0
2

- 2

- 1

P
Z

- j2

1 0 0
- j4

- j6
6 4 4 .4

Figure 2.5.5. Root locus of G0 (s) =

and underlined.

- j7

k0 (s+1)
s(s+2) (s+6+j2) (s+6j2) .

Values of k0 are in red

With this specications the root locus can be sketched. Using rule 9 the value
of k0 can be determined for some selected points. The value at the intersection
with the imaginary axis is
k0,crit =

7.2 7.4 7.9 11.1

= 644.4 .
7.25

94

2 Basic Control Elements

s
j
j

j
j

P 1
P 2
P 4
Z 1

= 1
= 1
= 9
= 1

jw

P
3

j2

6 1 .6
5 3 .4
0
5 8 .2

j
j
j
s
P

P
2

Z 1

P
1

Figure 2.5.6. Calculating the angle of departure P3 ,D of the complex pole sP3 =
6 + j2

95

Behaviour of linear continuous-time

control systems

3.1

Dynamical behaviour of a closed loop system

Prepared with an understanding of models, transfer functions and basics about

control loops, we now consider control systems and the types of feedback design
principles available. Figure 3.1.1 shows a block diagram of a closed loop system
with the four classical components: controller, actuator, plant and measurement

z
e
w
+
_

c o n tro lle r

u
C

a c tu a to r

'

d is tu rb a n c e
b e h a v io u r

c o n tro l
b e h a v io u r +

y
C

m e a su re m e n t
d e v ic e

p la n t

Figure 3.1.1. The basic components of a control loop

device. It is often convenient to combine the controller and actuator into one
controller component, while the measurement device is often assigned to the
plant. Usually a set of disturbances zi
(i = 1,2, . . .) may occur, each of them can
enter the plant at dierent locations. The transition behaviour of the plant and of
the parts of the plant between disturbance input and plant output, respectively,
is denoted by Gpzi (s). From this a block diagram of the closed loop system is
obtained according to Figure 3.1.2.
For linear plants all the disturbances Zi
can be combined into one single cumulative disturbance
n

Zi
(s)GPzi (s)

Z(s) =
i=1

according to Figure 3.1.2. This cumulative disturbance will act at the plant output
(see Figure 3.1.3). Furthermore, by a suitable choice of GPzi (s) it can be shown
that the structure from Figure 3.1.2 is also valid for disturbances zi
(t) entering
at other locations in the closed loop.

96

3 Behaviour of linear continuous-time control systems

Z 3

Z 2
Z 1

(s )

E (s )

(s )

G
P

z
3

G
P

z
2

z
1

U (s )
G

(s )
(s )
(s )
+

(s )

Y (s )

Figure 3.1.2. Block diagram of the closed-loop system

W (s )
+

E (s )
_

G
C

(s )

U (s )
G

(s )

Z (s )
+
Y (s )
+

Figure 3.1.3. Block diagram of the closed-loop system with cumulative disturbance
Z(s)

The transition behaviour of this control loop is specied according to the two
inputs (command and disturbance) either command behaviour or disturbance
behaviour. The transfer function of the controller elements briey called in
the following only controller is GC (s) and those of the plant GP (s). From
Figure 3.1.3 the controlled variable of the closed loop is
Y (s) = Z(s) + [W (s) Y (s)] GC (s) GP (s) .
Rearranging, then it follows
Y (s) =

GC (s) GP (s)
1
Z(s) +
W (s) .
1 + GC (s) GP (s)
1 + GC (s) GP (s)

(3.1.1)

Using this equation, the control system tasks already mentioned in section 1.3
can be formulated more precisely as follows:
a) For W (s) = 0 the transfer function of the closed loop for disturbance behaviour the disturbance transfer function
GZ (s) =
is obtained.

1
Y (s)
=
Z(s)
1 + GC (s) GP (s)

(3.1.2)

3.1 Dynamical behaviour of a closed loop system

97

b) Similarly for Z(s) = 0 the transfer function of the closed loop for command
behaviour is the command transfer function
GC (s) GP (s)
Y (s)
=
.
(3.1.3)
GW (s) =
W (s)
1 + GC (s) GP (s)
Both transfer functions GZ (s) and GW (s) contain the dynamical control factor
1
(3.1.4)
R(s) =
1 + G0 (s)
with
(3.1.5)
G0 (s) = GC (s) GP (s) .
Opening the closed loop for W (s) = 0 and Z(s) = 0 according to Figure 3.1.4
at an arbitrary location and dening with respect to the route of the transfer
elements the input as xe (t) and the output as xa (t), the transfer function of the
open loop
Xa (s)
= GC (s) GP (s) = G0 (s)
(3.1.6)
Gopen (s) =
Xe (s)
is obtained.

G
_

G
C

(s )

(s )

X e( s ) X
a

G
P

(s )

(s )

Figure 3.1.4. Open control loop

If G0 (s) can be described by a rational fraction, by setting the denominator of

Eq. (3.1.2) or Eq. (3.1.3) to zero one obtains for the closed loop the condition
1 + G0 (s) = 0

(3.1.7)

analogous to Eq. (2.2.12) for the characteristic equation in the form

P (s) = a0 + a1 s + a2 s2 + . . . + an sn = 0 .

(3.1.8)

The overall goal in designing a control system is to use the principle of feedback
to cause the controlled variable to follow a desired command variable accurately
regardless of the command variables path and to minimise the eect of any
external disturbances or changes in the dynamics of the plant. Reaching this goal
economically the standard structure of Figure 3.1.3 is a relatively complex task
if one must meet the basic requirements listed below:

98

3 Behaviour of linear continuous-time control systems

a) The minimum requirement is that the closed loop is stable.

b) Disturbances z(t) should be rejected or they must have a small inuence on
the controlled variable y(t).
c) The controlled variable y(t) should track the command input w(t) as precisely and as fast as possible.
d) The closed loop should be as insensitive as possible with respect to changes
in the plant parameters.
In order to full the requirements in the ideal case, the command transfer function
must be according to requirement c)
GW (s) =

G0 (s)
Y (s)
=
=1,
W (s)
1 + G0 (s)

(3.1.9)

and the disturbance transfer function according to requirement b)

GZ (s) =

1
Y (s)
=
=0.
Z(s)
1 + G0 (s)

(3.1.10)

A rigorous realisation of these requirements is not possible for physical and technical reasons. The problem will be illustrated using the following simple example.
Example 3.1.1
A common actuator in control systems is the DC motor. It provides rotary motion
for a current input. The dynamical behaviour between current u(t) and speed y(t)
is described by the simplied transfer function
GP (s) =

KP
Y (s)
=
.
U (s)
1 + Ts

(3.1.11)

In order to compensate the plant dynamics, a candidate controller may be

GC (s) = KC (1 + T s) .

(3.1.12)

The open-loop transfer function is

G0 (s) = GC (s) GP (s)
= KC (1 + T s)

KP
= KC KP ,
1 + Ts

(3.1.13)

3.1 Dynamical behaviour of a closed loop system

y (t)= h
W

(t)
K

1 .0

K
C

99
(a )

= 5 0
P

0 .8
K

K
C

= 1 0
P

0 .6
0 .4
K
C

K
P

= 1

0 .2
0
0

0 .0 5

0 .1

0 .1 5

0 .2

0 .2 5

0 .3

t/ T

u (t)
1 0

(b )
8
K
6

K
C

K
4

= 5 0

K
P

= 1 0
K

2
0
0

0 .0 5

0 .1

K
P

= 1

0 .1 5

0 .2

0 .2 5

0 .3

t/ T

Figure 3.1.5. Step response of the closed loop, (a) speed and (b) current for dierent
open-loop gains

which shows a proportional behaviour. On step inputs to the controller the speed
will jump, which is physically not possible due to the inertia of the motor. According to section 2.2.3 the controller in Eq.(3.1.12) is not realisable. Adding a
pole in the controller transfer function to the left in the s plane at sC = 10/T
will cure this problem, but with a delayed speed response. Figure 3.1.5a shows
the controlled speed for a unit step in the command input. The time constant
of the closed loop system changes as the feedback gain increases. Increasing the
controller gain KC will speed-up the behaviour and reduce the steady-state error,
but will also increase the control eort as shown in Figure 3.1.5b. As the current of the motor is limited for physical reasons the manipulated variable u(t) is
also limited. Increasing the controller gain KC to an arbitrary high value is not
suitable. During the design of a controller such limitations have to be taken into
account.
It is often true that closed-loop systems have a faster response as the feedback
gain is increased, and if there are no other eects, this is generally desirable.
However, systems typically also become less well damped and even unstable as
the gain increases. This is shown when we mount the same DC motor on a robotic
manipulator and control the speed of the manipulator arm using the same type of

100

3 Behaviour of linear continuous-time control systems

controller. In this case the speed of the arm movement is the controlled variable
y(t). The transfer function between the current of the DC motor and the speed
of the arm is
Y (s)
KP
.
(3.1.14)
=
GP (s) =
U (s)
(1 + T s)(1 + 2 s + ( s )2 )
0

Figure 3.1.6 shows the step response of this control system.

y (t)= h
W

(t)
1 .5
K
C

K
P

= 5 0

1
K
C

K
P

= 1 0

0 .5
K
0
0

= 1
P

6 t/ T
5

Figure 3.1.6. Step response of the closed loop for dierent open-loop gains for = 4
and 0 = 2/T

From the example given above it can be seen that a denite limit exists on how
high we can make the gain. But there is a design tradeo between gain and steadystate error. Attempts to resolve the conict between small steady-state errors and
good transient or dynamic responses must be undertaken. These two essential
aspects of performance are considered when a control system is designed: the
transient performance and the steady-state performance. The following sections
deal with these aspects in more detail.

3.2

Static properties of the closed loop

Frequently the behaviour of an open loop (according to Figure 3.1.4 and Eq. (3.1.5))
can be described by a generalised transfer function of the form
G0 (s) =

K0 1 + 1 s + . . . + m sm
eTt s
sk 1 + 1 s + . . . + nk snk

mn,

(3.2.1)

where the constant k = 0,1,2, . . . denotes the type of transfer function G0 (s). K0
is the gain of the open loop. Therefore G0 (s) shows for

3.2 Static properties of the closed loop

101

k = 0; delayed proportional behaviour

(delayed P behaviour)
k = 1; delayed integral behaviour
(delayed I behaviour)
k = 2; delayed double integral behaviour (delayed I2 behaviour) .
We assume now that the term of the rational fraction in Eq. (3.2.1) contains only
poles in the left half s plane. For the dierent types of transfer functions G0 (s)
with dierent forms of the command signal w(t) or of the disturbance z(t) the
steady state of the closed loop for t can be analysed.
With
E(s) = W (s) Y (s)

(3.2.2)

from Eqs. (3.1.1) and (3.1.6) it follows for the control error
E(s) =

1
[W (s) Z(s)] .
1 + G0 (s)

(3.2.3)

Under the assumption, that the limit of the control error e(t) for t exists, one obtains by using the nal value theorem of the Laplace transform (see
section 2.1.3) the steady-state value of the control error
lim e(t) = lim s E(s) .

s0

(3.2.4)

For the case of all disturbances being related to the plant output from Eq. (3.2.3)
it follows that sign apart both types of inputs, command or disturbance,
can be treated equally. Hence in the following to represent both types of input
signals the term Xe (s) is chosen as the input signal. Using both Eqs. (3.2.3)
and (3.2.4) the steady-state values of the control error for the dierent signal
types of xe (t) and for dierent types of transfer functions G0 (s) of the open loop
can be obtained. These values characterise the behaviour of the control loop.
They are obtained consecutively for the most important cases.
For further treatment the following test signals according to Figure 3.2.1 are used:
a) Step input signal:
Xe (s) =

xe0
,
s

(3.2.5)

where xe0 is the height of the step.

b) Ramp input signal:

xe1
,
s2
describes the slope of the ramp signal xe (t) .
Xe (s) =

where xe1

c) Parabolic input signal:

xe2
,
(3.2.7)
s3
is a measure of the acceleration of the parabolic signal xe (t) .
Xe (s) =

where xe2

(3.2.6)

102

3 Behaviour of linear continuous-time control systems

(a )

x e(t)

(b )

x e(t)

(c )

x e(t)

e 0

x e(t)= x

e 0

s (t)

x e(t)= x

e 1

s (t)t

0
t

x e(t)= x
t

e 2

s (t)t

Figure 3.2.1. Dierent input signals xe (t), which are frequently used for the disturbance z(t) and command input w(t): (a) step, (b) ramp and (c) parabolic input signal

Following Eq. (3.2.3) the control error is obtained by

1
Xe (s) ,
1 + G0 (s)

E(s) =

(3.2.8)

where the dierence between command and disturbance behaviour is only in the
sign of Xe (s) (disturbance: Xe (s) = Z(s); command: Xe (s) = W (s)). Inserting
this relation into Eqs. (3.2.5) to (3.2.7) the corresponding control error can be
obtained for dierent types of transfer functions G0 (s). This will be demonstrated
in the following.

3.2.1

Transfer function G0 (s) with delayed P behaviour

For this case from Eq. (3.2.1) the transfer function is

G0 (s) = K0

1 + 1 s + . . . + m sm Tt s
e
.
1 + 1 s + . . . + an sn

(3.2.9)

This transfer function describes an open control loop with delayed P behaviour.
The variable K0 is the gain of this open control loop. In this case it is composed
of the gain of the controller KC and of the plant KP in multiplicative form
K0 = KC KP .

(3.2.10)

With Eq. (3.2.4) one obtains for the steady-state error of the closed control loop
lim e(t) = lim s

s0

1
Xe (s)
1 + G0 (s)

(3.2.11)

1
xe .
1 + K0 0

(3.2.12)

and for the step signal with Eq. (3.2.5)

e = lim e(t) =
t

3.2 Static properties of the closed loop

103

It can be shown for a ramp input signal according to Eq. (3.2.6) that the double
pole in Eq. (3.2.8) in the time domain corresponds to e(t) = const t (t) such
that
(3.2.13)
e = lim e(t) ,
t

which is not nite. A similar situation results for a parabolic input signal using
Eq. (3.2.7).

3.2.2

Transfer function G0 (s) with delayed I behaviour

From Eq. (3.2.1) it follows for this case that

G0 (s) =

1 + 1 s + . . . m sm
K0
eTt s .
s 1 + 1 s + . . . + n1 sn1

(3.2.14)

The inherent open control loop shows delayed I behaviour. With Eq. (3.2.11) one
obtains for the steady-state error of the closed loop for the case of a step input
signal
(3.2.15)
e = lim e(t) = 0 ,
t

and in the case of a ramp input signal

e = lim e(t) =
t

1
xe .
K0 1

(3.2.16)

Furthermore in the case of parabolic input signal we have the same result as
Eq. (3.2.13).
DYNAST study example 3.2.1
Cruise control of a car
DYNAST study example 3.2.2
D.C. motor position control
DYNAST study example 3.2.3
Double integrator plant
DYNAST study example 3.2.4
2nd-order plant
DYNAST study example 3.2.5
Load disturbance - 2nd-order plant

104

3.2.3

3 Behaviour of linear continuous-time control systems

Transfer function G0 (s) with delayed I2 behaviour

The transfer function for this particular case is

K0 1 + 1 s + . . . + m sm
eTt s
G0 (s) = 2
s 1 + 1 s + . . . + n2 sn2

(3.2.17)

and it describes a system with delayed I2 behaviour. For the steady-state error of
the closed loop system it follows in the case of a stable loop for a step and ramp
input signal
(3.2.18)
e = lim e(t) = 0 ,
t

and for a parabolic input signal

e = lim e(t) =
t

1
xe .
K0 2

(3.2.19)

From these results, especially from Eqs. (3.2.12), (3.2.16) and (3.2.19) and from
Table 3.2.1 it follows, that the steady-state error e , which characterises the
static behaviour of the control loop is in all cases smaller the larger the loop gain
K0 given by Eq. (3.2.10). In the case of delayed P behaviour of the open loop the
steady-state error e is much smaller the smaller the value of the static control
factor
1
.
(3.2.20)
R=
1 + K0
Often a large loop gain K0 rapidly leads to instability of the closed loop. For
setting K0 usually a compromise is made unless selecting an appropriate type
of controller such that the steady-state error vanishes. Both, the dynamical and
especially the static behaviour depend strongly on the choice of the controller.
Therefore, in the following the most important types of standard controllers will
be introduced.

3.3
3.3.1

Performance indices
Time-response specications

Specications for a control system design often involve certain requirements associated with the time response of the closed-loop system. The requirements are
specied by the behaviour of the controlled variable y(t) or by the control error
e(t) on well dened test signals. The most important test signal is a unit step
on the input of the control system and requirements are placed on the behaviour
of the controlled variable y(t) = hW (t), as shown in Figure 3.3.1. The requirements for a unit step response are expressed in terms of the following standard
quantities:

3.3 Performance indices

105

Table 3.2.1 Steady-state error for dierent type of systems of G0 (s) and dierent
input signals xe (t) (command and disturbance signals if all disturbances will act at the
plant output)

Type of system of
G0 (s)
according to Eq. (3.2.1)

Input signal
Xe (s)

Steady-state
error e

k=0
(delayed

xe0
s
xe1
s2
xe2
s3

1
xe
1 + K0 0

P behaviour)

k=1
(delayed
I behaviour)

k=2
(delayed
I2 behaviour)

xe0
s
xe1
s2
xe2
s3
xe0
s
xe1
s2
xe2
s3

0
1
xe
K0 1

0
0
1
xe
K0 2

The maximum overshoot Mp is the magnitude of the overshoot after the rst
crossing of the steady-state value (100%). This value is normally expressed
as a percentage of the steady-state value of the controlled variable.
The peak time tp is the time required to reach the maximum overshoot.
The settling time t is the time for the controlled variable rst to reach
and thereafter remain within a prescribed percentage of the steady-state
value. Common values of are 2%, 3% or 5%.
The rise time tr is the time required to reach rst the steady-state value
(100%). It may also be dened as the time to reach the vicinity of the
steady-state value particularly for a response with no overshoot, e.g. the
time between 10% and 90%. The 50% rise time tr,50 is dened as the time

106

3 Behaviour of linear continuous-time control systems

to rst reach 50% of the nal value.

y ( t) = h

( t)
W

2 e
p

1 0 0 %

5 0 %
0 %
0
t

t
5 0

t
r

t
p

t
e

Figure 3.3.1. Typical under-damped response of a control system to step command

inputs

Similarly, the behaviour on step disturbances can be characterised as shown in

Figure 3.3.2. Here likewise the terms maximum overshoot and settling time are
dened.
y ( t)= h
Z

(t )

1 0 0 %
u n c o n t r o lle d
c o n t r o lle d
M
0 %

2 e
p

t
p

t
e

Figure 3.3.2. Typical under-damped response of a control system to step disturbances

These standard quantities are measures of some properties of the control system.
Mp and t essentially characterise the damping and tr and tp the speed, i.e. the

3.3 Performance indices

107

dynamics of the control behaviour. The steady-state error e as described in

section 3.2 is a typical characteristic of the static behaviour.
These quantities describe the deviation of the step response from the ideal case
described in section 3.1 and the goal of the design of a control system is to hold
them as small as possible. In most cases one can restrict the values of the three
quantities tr , t and MP .

3.3.2

Integral performance indices

It generally happens that a control system design problem reaches the point where
one or more parameters are to be selected to give the best performance. If a
measure or index of performance can be expressed mathematically, the problem
can be solved for the best choice of the adjustable parameters. The resulting
system is termed optimal with respect to the selected criterion.
Introducing a performance index
Ja = w1 tr + w2 t + w3 MP ,

(3.3.1)

composed of the quantities as discussed in section 3.3.1 and minimising Ja is

unsuitable. This is because one has to subjectively specify the weights w1 , w2 and
w3 , and a straight mathematical approach to solve the minimisation problem is
not available.
The selection of an appropriate performance index is as much a part of the design
process as calculating the nal system. An optimal value of an inappropriate performance measure may result in poor performance. Commonly used performance
indices are based on integral performance measures. From Figure 3.3.1 one can
see that the area between the 100% line and the step response hW (t) is surely a
measure for the deviation of the control loop from the ideal response for a step
command input. Likewise for the deviation of the disturbance rejection from the
ideal case Figure 3.3.2 shows the area between the 0% line and hZ (t) to be a good
measure. In both cases the total area below the control error e(t) = w(t) y(t)
is involved. It is obvious to introduce the integral


Jk =

fk [e(t)] dt

(3.3.2)

as a performance measure, where fk [e(t)] is one of the functions given in Table 3.3.1, e.g. e(t), |e(t)|t and e2 (t). These types of integral performance indices
can also include derivatives of the control error or terms of the manipulated variable u(t) u that adds a penalty for control eort. Using such performance
indices one can now dene the integral criterion as follows:

108

3 Behaviour of linear continuous-time control systems

Table 3.3.1 The most common integral performance indices

properties

performance index
JIE =

Integral of error: suitable for highly

damped or monotonic responses; simple
mathematical treatment.

e(t) dt

JIAE =

Integral of absolute value of error: suitable for non-monotonic responses; awkward analysis.

|e(t)| dt

JISE =

Integral of squared error: highly penalising

large control errors; settling time is larger
than for JIAE ; good analytical treatment.

e2 (t) dt

JITAE =

Integral of time multiplied by the absolute

value of error: eect like JIAE ; regards the
permanence of the control error.

|e(t)| t dt

Integral of the squared time to the q times

error: eect like JISE , but adds a heavy
penalty for errors that do not die out
rapidly; does not try to eliminate too
quickly the inevitable initial error for a
step input; less overshoot MP and shorter
settling time t ; good analytical treatment.

/
[e(t) tq ]2 dt

JISTq E =

JGISE

/
= [e2 (t) + e2 (t)] dt
0

JISESC =

[e2 (t) + (u(t) u )2 ] dt

Generalised integral of squared error: responses more favourable than JISE ; choice
of the weight subjective.
Integral of squared error and squared control eort: Larger overshoot MP , but essentially shorter settling time t ; choice of
the weight subjective.

A control system is better the smaller the value of Jk of the selected

performance index. Therefore the integral criterion


Jk =

fk [e(t)] dt = Jk (c1 ,c2 , . . .) = Min

(3.3.3)

always requires minimising Jk by adjusting the controller parameters

3.3 Performance indices

109

c1 ,c2 , . . ..

3.3.3

Determination of quadratic performance indices

The integral performance indices JIAE and JITAE from Table 3.3.1 have the disadvantage that they have to be evaluated in the time domain, either by laborious
computation or by simulation. All the other squared error type of indices are more
pleasant because of calculating its value in the s-domain rather than in the time
domain. Therefore the analysis is simpler. This will be shown in the following for
JISE .
The performance index is given by


e2 (t) dt

JISE =

(3.3.4)

Applying the convolution theorem in the frequency domain from Eq. (2.1.12) for
s = c = 0 and f1 (t) = f2 (t) = e(t), one obtains Parsevals theorem

1
e (t) dt =
2j

+j

JISE =
0

E(s) E(s) ds .

(3.3.5)

When E(s) is expressed as a ratio of polynomials

E(s) =

b0 + b1 s + . . . + bn1 sn1
,
a0 + a1 s + . . . + an sn

(3.3.6)

dierent methods are available to evaluate the integral. For calculation of the
integral a recursion formula is available. Its solution has also been tabulated up
to quite high values of n in terms of the coecients of the polynomials. Table 3.3.2
below gives a short list. For a detailed analysis, e.g. when the integral depends
on some parameters, a general algebraic approach using determinants is more
suitable as shown in section A.7.
The more general form
JISTq E

= [e(t) tq ]2 dt

(3.3.7)

of a squared performance index can be easily evaluated. Since the Laplace transform of te(t) according to the complex dierentiation theorem Eq. (2.1.8) is equal
to dE(s)
ds , one obtains

110

3 Behaviour of linear continuous-time control systems

Table 3.3.2 Values for the integral

JISE

b20
2a0 a1

2
3

JISTq E

b21 a0 + b20 a2
2a0 a1 a2


b22 a0 a1 + b21 2b0 b2 a0 a3 + b20 a2 a3
2a0 a3 (a0 a3 + a1 a2 )

+j




1
dq
dq
q 2
= [e(t) t ] dt =
E(s)
E(s) ds .
2j
dsq
dsq

(3.3.8)

Since it is easy to compute the derivatives of a polynomial the above integral can
be computed using simple algebra and the aforementioned recursive formula.
Example 3.3.1
Determining the best damping ratio for a second-order system is a simple example
of the use of performance indices. Let us assume that the command transfer
function of a control system is described by Eq. (2.3.53) with K = 1 as
GW (s) =

02
.
02 + 20 s + s2

The control error for a unit step input is



02
20 + s
1
1 2
.
= 2
E(s) =
2
s
0 + 20 s + s
0 + 20 s + s2
From Table 3.3.2 we have
JISE =

(3.3.9)

1 + 4 2
.
40

As this function is a parabola in , with minimum given by

4 2 1
dJISE
=
=0
d
4 2 0
the minimum square error to a step input occurs for = 0.5 .
Demonstration Example 3.3.1
A virtual experiment using manual control

3.4 PID control and associated controller types

3.4
3.4.1

111

PID control and associated controller types

The classical three-term PID controller

We have seen in section 3.1 that proportional feedback control can reduce error
responses but that it still allows a non-zero steady-state error for a proportional
system. In addition, proportional feedback increases the speed of response but has
a much larger transient overshoot. When the controller includes a term proportional to the integral of the error, then the steady-state error can be eliminated,
as shown in section 3.2. But this comes at the expense of further deterioration in
the dynamic response. Addition of a term proportional to the derivative of the
error can damp the dynamic response. Combined, these three kinds of actions
form the classical PID controller, which is widely used in industry.
This principle mode of action of the PID controller can be explained by the
parallel connection of the P, I and D elements shown in Figure 3.4.1. From this
diagram the transfer function of the PID controller is
GC (s) =

UC (s)
KI
= Kp +
+ KD s .
E(s)
s

(3.4.1)

K p
K I
s

E
K

U
+
+

(U )

o r

1
C

T I s

T
s

+ U
C
+
+

(U )

Figure 3.4.1. Block diagram of the PID controller

The controller variables are

KC = Kp
Kp
TI =
KI
KD
TD =
Kp

gain
integral action time
derivative action time

Eq. (3.4.1) can be rearranged to give



1
+ TD s .
GC (s) = KC 1 +
TI s

(3.4.2)

112

3 Behaviour of linear continuous-time control systems

These three variables KC , TI and TD are usually tuned within given ranges.
Therefore, they are often called the tuning parameters of the controller. By proper
choice of these tuning parameters a controller can be adapted for a specic plant
to obtain a good behaviour of the controlled system.
If follows from Eq. (3.4.2) that the time response of the controller output is

t

KC
uC (t) = KC e(t) +
TI

e( ) d + KC TD

de(t)
.
dt

(3.4.3)

Using this relationship for a step input of e(t), i.e. e(t) = (t), the step response
h(t) of the PID controller can be easily determined. The result is shown in Figure 3.4.2a. One has to observe that the length of the arrow KC TD of the D action
is only a measure of the weight of the impulse.
h (t)
K
C

(1 + T D )
D a c t io n
K C

D
1 +
T V
D a c t io n
K

I a c t io n

- T

h (t)

(a )

(b )

I a c t io n
C

a c t io n

- T
t

0
I

a c t io n

Figure 3.4.2. Step responses (a) of the ideal and (b) of the real PID controller

In the previous considerations it has been assumed that a D behaviour can be

realised by the PID controller. This is an ideal assumption and in reality the
ideal D element cannot be realised (see section 2.2.3). In real PID controllers a
lag is included in the D behaviour. Instead of a D element in the block diagram
of Figure 3.4.1 a DT1 element with the transfer function
GD (s) = KD

TV s
1 + TV s

(3.4.4)

is introduced. From this the transfer function of the real PID controller or more
precisely of the PIDT1 controller follows as
GC (s) = Kp +

KI
TV s
+ KD
.
s
1 + TV s

Introducing the controller tuning parameters

KC = Kp , TI =

KC
KI

and

TD =

KD TV
KC

(3.4.5)

3.4 PID control and associated controller types

it follows



s
1
+ TD
.
GC (s) = KC 1 +
TI s
1 + TV s

113

(3.4.6)

The step response h(t) of the PIDT1 controller is shown in Figure 3.4.2b. This
response from t = 0 gives a large rise, which declines fast to a value close to the
P action, and then migrates into the slower I action. The P, I and D behaviour
can be tuned independently. In commercial controllers the D step at t = 0
can often be tuned 5 to 25 times larger than the P step. A strongly weighted
D action may cause the actuator to reach its maximum value, i.e. it reaches its
limits.
As special cases of PID controllers one obtains for:
a) TD = 0 the PI controller with transfer function


1
;
GC (s) = KC 1 +
TI s

(3.4.7)

b) TI the ideal PD controller with the transfer function

GC (s) = KC (1 + TD s)
and the PDT1 controller with the transfer function


s
;
GC (s) = KC 1 + TD
1 + TV s

(3.4.8)

(3.4.9)

c) TD = 0 and TI the P controller with the transfer function

GC (s) = KC .

(3.4.10)

The step responses of these types of controllers are compiled in Figure 3.4.3. A
pure I controller may also be applied and this has the transfer function
GC (s) = KI

3.4.2

KC
1
=
.
s
TI s

(3.4.11)

Optimal tuning of PID controllers

The measure of the quality of the transient response of a PID controlled system
can be performed by calculating an integral performance index as shown in section 3.3.2. The best controller is one that has the minimum performance index.
When this performance index is a minimum for a specied input, the system performance is said to be optimal. When the input signal is specied the quadratic
performance index JISE can be calculated for a given plant transfer function as a
function of the tuning parameters, e.g. KC , TI , TD and TV .

114

3 Behaviour of linear continuous-time control systems

h (t)

h (t)
K

c o n t r o lle r
K

h (t)

P I c o n t r o lle r
K

h (t)
T
(1 + D)
T
K

P D c o n t r o lle r
( id e a l)
)

K
0

- T

(1 + T
C

P D T 1 c o n t r o lle r
( r e a l P D c o n t r o lle r )

t
0

Figure 3.4.3. Step responses of the PID controller family

The mathematical calculation of this performance index for given values of the
tuning parameters is simple as shown in section 3.3.3. But getting the optimal
parameters is a non-trivial task. Though computerised optimisation algorithms
are available to calculate the optimal parameter setting, for the case of quadratic
performance indices a mathematical analysis is possible. The approach shown in
section A.7 gives more insight into the controller settings and can be applied to
all types of plants and PID controllers.
In the following the command and disturbance behaviour of a control system
with a real PID controller and a plant with the transfer function
GP (s) =

KP
.
(1 + T s)4

(3.4.12)

will be investigated. The response of the control error to step changes w(t) =
w0 (t) in the command input and z
(t) = z0 (t) in the plant input is
E(s) =

w0 z0 GC 1
.
1 + GC GP s

For the plant (Eq. (3.4.12)) and the real PID controller (Eq. (3.4.6)) one obtains
w0 TI (1 + T s)4 (1 + TV s) z0 KP TI (1 + TV s)
,
TI (1 + T s)4 (1 + TV s)s + KC KP [1 + (TI + TV )s + (TD + TV )s2 ]
(3.4.13)
which is in the form of Eq. (3.3.6) or Eq. (A.7.5) for k = KC KP .
E(s) =

3.4 PID control and associated controller types

115

Applying the analysis shown in section A.7 to the JISE performance index one
gets the diagrams in Figure 3.4.4, separately for the command and disturbance
TI
. These
inputs. The integral action time constant is normalised by TIN = 4T
8
diagrams are shown for the optimal value TD = TDopt = 3 T of the derivative
T

IN

0 .7 5

IS E

(a )
=

4 .2

0 .6 2 5

d J

IS E

/ d k = 0

0 .5
J

0 .3 7 5
T

IN o p t

0 .2 5

2 .1

= 1 .3 3

1 .4

1 .6

2 .9

0 .1 2 5

u n s t a b le
0
0

= 0 .4 2

IS E o p t

0 .5

IN

IS E

k
1

= 1 .5 9

1 .5

2 .5
2

(b )

2 .3

0 .7 5

o p t

d J

0 .5 5

0 .6 2 5

IS E

/ d k = 0

0 .2 4

0 .5

0 .1 4

0 .3 7 5
T

0 .2 5

IN o p t

= 0 .2 2

IS E o p t

= 0 .1

0 .1 2 5
u n s t a b le
0
0

0 .5

k
1 .5

o p t

= 2 .7 4
2 .5

3
k

Figure 3.4.4. Stability and performance diagram for step changes (a) in the command
input (w0 = 1, z0 = 0) and (b) in the plant input (z0 = 1, w0 = 0)

116

3 Behaviour of linear continuous-time control systems

action time constant. The lter time constant is TV = 0.1TD . The diagrams show
a rather rectangular stability area that makes tuning of KC and TI for a xed
TD easy from the stability point of view. But the performance characteristics are
quite dierent. The optimal parameters for the two cases dier by about a factor
of two. Therefore, an optimal tuned controller is in general never optimally tuned
for command and disturbance inputs.

3.4.3

Advantages and disadvantages of the dierent types of controllers

In the following the disturbance behaviour is investigated using the controllers

introduced in section 3.4.1. Their parameters are tuned optimally according to
the performance index JISE from section 3.3.2. The plant is given by Eq. (3.4.12).
Figure 3.4.5 shows for the dierent types of controller the responses to a step
disturbance z0 (t) of the controlled variable y, which is normalised by KP z0 .
These curves indicate that because w(t) 0 the relation e(t) = y(t) is valid.
For discussing these curves the term settling time t3% according to section 3.3.1
is used, which is related to the steady state of the uncontrolled case
y,without = KP z0 .

(3.4.14)

In addition, the dierent cases should be compared with respect to the normalised
maximum overshoot Mp /(KP z0 ).
The dierent cases are discussed below:
a) The P controller shows a relatively high maximum overshoot Mp /(KP z0 ), a
long settling time t3% as well as a steady-state error e .
b) The I controller has a higher maximum overshoot than the P controller due
to the slowly starting I behaviour, but no steady-state error.
c) The PI controller fuses the properties of the P and I controllers. It shows
a maximum overshoot and settling time similar to the P controller but no
steady-state error.
d) The real PD controller according to Eq. (3.4.9) with TV = TD /10 has a
smaller maximum overshoot due to the faster D action compared with the
controller types mentioned under a) to c). Also in this case a steady-state
error is visible, which is smaller than in the case of the P controller. This
is because the PD controller generally is tuned to have a larger gain KC
due to the positive phase shift of the D action. For the results shown in
Figure 3.4.5 the gain for the P controller is KC = 2.68 and for the PD
controller KC = 4.74. The plant has a gain of KP = 1.

3.4 PID control and associated controller types

117

e) The PID controller according to Eq. (3.4.6) with TV = TD /10 fuses the
properties of a PI and PD controller. It shows a smaller maximum overshoot
than the PD controller and has no steady state error due to the I action.
The qualitative concepts of this example are also relevant to other type of plants
with delayed proportional behaviour. This discussion has given some rst insights
into the static and dynamic behaviour of control loops.

3.4.4

Empirical tuning rules according to Ziegler and Nichols

Many industrial processes show step responses with pure aperiodic behaviour
according to Figure 3.4.6. This S-shape curve is characteristic of many highorder systems and such plant transfer functions may be approximated by the
mathematical model
KP Tt s
e
,
(3.4.15)
GP (s) =
1 + Ts
which contains a 1st-order delay element and a dead time. Figure 3.4.6 shows the
approximation by a PT1 Tt element.
Here the step response is characterised by constructing the tangent at the turning
point T with the following three values: KP ( gain of the plant), Tr (rise time)
and Tu (delay time). Then a rough approximation according to Eq. (3.4.15) is to
set Tt = Tu and T = Tr .
For a plant of the type described above a lot of tuning rules for standard controllers have been developed. These have been mostly developed empirically from
simulation studies. The most famous empirical tuning rules are those of Ziegler
and Nichols. These tuning rules have been derived to provide step responses for
the closed loop, where the response shows a decrease of the amplitude of approx. 25% per period. For the application of these rules according to Ziegler and
Nichols two dierent approaches can be used:
a) Method of the stability margin(I): Here, the following steps are used:
1. The controller is switched to pure P action.
2. The gain KC of the P controller is continuously increased until the
closed loop shows permanent oscillations. The value of the gain KC at
this state is denoted as the critical controller gain KC crit .
3. The length of period Tcrit (critical period) of the oscillations is measured.
4. From KC crit and Tcrit one determines the controller tuning values KC ,
TI and TD using the formulas given in Table 3.4.1.

P ID

z 0K P

,w it h o u t

P I

1 0

P
P D

w it h o u t c o n t r o lle r

2 0

P ID
P I
P D
P
I

C o n t r o lle r

=
+

z 0 K P
0 .2 0
0 .4 4
0 .2 2
0 .4 5
0 .7 6

3 0

t3 %
T
1 9
3 5
3 1
2 2
4 0

(s )

U
+

e
z0 K P
- 0 .1 7
- 0 .2 7
+

Z '

2 .7
2 .4
4 .7
2 .6
-

K C

o p t

o p t

4 0

G P (s )

4 T
0 .2 2
1 .3 8
0 .8 8 K

T I

t T

T D opt
4 T
0 .6 7
0 .3 2
-

Figure 3.4.5. Behaviour of the normalised controlled variable y/(z0 KP ) for step disturbance z
= z0 (t) at the input to the plant
[GP (s) = KP /(1 + T s)4 ; KP = 1 ] for dierent types of controllers

- 0 .2

0 .0

0 .2

0 .4

0 .6

0 .8

1 .0

z 0 K P

118
3 Behaviour of linear continuous-time control systems

3.4 PID control and associated controller types

119

h ( t)
K P

0 .6 3 K P

P T 1T t
a p p r o x im a tio n
T

t
u

T
r

Figure 3.4.6. Describing the step response of a process by the three characteristic
values KP (gain of the plant), Tr (rise time) and Tu (delay time)

b) Method of the step response (II): In the case of an industrial plant it is

often not possible, suitable or allowed to drive the plant into permanent
oscillations for determining KC crit and Tcrit . Measuring the step response of
the plant does not generally cause diculties. Therefore, in many cases the
second form of the Ziegler-Nichols approach is more expedient. The rules are
based directly on the slope KP /Tr of the tangent at the turning point and on
the delay Tu of the step response. One has to observe that the measurement
of the step response needs only to be taken at the turning point T, as the
slope of the tangent already describes the ratio KP /Tr . Using the measured
data Tu and KP /Tr as well as the formula given in Table 3.4.1 the controller
tuning parameters can be determined by simple calculations.
Demonstration Example 3.4.1
A virtual experiment using PID control for tracking
Demonstration Example 3.4.2
A virtual experiment using PID control for high-precision positioning
DYNAST study example 3.4.1
PI control of a PT1 Tt plant
DYNAST study example 3.4.2
PID control of a PT1 Tt plant
DYNAST study example 3.4.3
Disturbance response for PI control of a PT1 Tt plant

120

3 Behaviour of linear continuous-time control systems

Table 3.4.1 Controller tuning parameters according to Ziegler and Nichols

Controller parameters
Type of controller

KC

TI

TD

0.5 KC crit

PI

0.45 KC crit

0.85 Tcrit

PID

0.6 KC crit

0.5 Tcrit

0.12 Tcrit

1 Tr
KP Tu

PI

0.9 Tr
KP Tu

3.33 Tu

PID

1.2 Tr
KP Tu

2 Tu

0.5 Tu

Method I

Method II

3.5

3.5.1

Design of controllers using pole-zero compensators

Characteristics in frequency and time domain

In the following the most important characteristics in the frequency domain of

the open and closed loop for command inputs for a closed loop having a transfer
function with two complex poles will be given. This section is based on the relationship between the frequency characteristics and the performance indices in
the time domain for the closed loop introduced in section 3.3.1.
A closed loop showing a step response hW (t) according to Figure 3.3.1 has a
frequency response GW (j) with a peak as qualitatively shown in Figure 3.5.1.
For describing this behaviour the following characteristics mentioned earlier can
be used:
resonant peak frequency p ,
resonant peak Mr ,
bandwidth b ,
phase angle b = (b ).
These characteristics are shown in Figure 3.5.1.
The closed loop for a PT2 S element has the transfer function

3.5 Design of controllers using pole-zero compensators

G
W

121

d B

M
0
w

w
r

w
p

- 3 d B
j

Figure 3.5.1. Bode plot of the closed-loop frequency response

GW (s) =

02
G0 (s)
= 2
1 + G0 (s)
s + 20 s + 02

(3.5.1)

according to Eq. (2.3.53) with K = 1.

The natural frequency 0 and the damping ratio characterise the control behaviour completely. This can be used as a good approximation for other transfer
functions if they contain a dominant pair of poles according to Figure 3.5.2.
j
w
0

jw
d

s p la n e

1 - z 2

d o m in a n t
p a ir o f p o le s

s
z w
0

Figure 3.5.2. Distribution of poles of an element with a dominant pair of poles

This pair of poles is assumed to be the closest pair to the j axis in the s domain and therefore it describes the slowest mode and inuences the dynamical
behaviour of the system very strongly provided the other poles are suciently
far away on the left-hand side of the s plane.
The step response for the transfer function of Eq. (3.5.1) is

122

3 Behaviour of linear continuous-time control systems



hW (t) =


0 t

1e

cos



1 2 0 t + 

1 2




sin
1 2 0 t
(t)

(3.5.2a)
and according to Eq. (A.3.8) it can be put into the more suitable form





e0 t
cos
1 2 0 t d
(t) ,
(3.5.2b)
hW (t) = 1 
1 2
where for d = sin1 or = sin d is valid. From Eq. (3.5.2b) the weighting
function, which follows by dierentiation, is


0
e0 t sin
1 2 0 t (t) .
(3.5.3)
gW (t) = h W (t) = 
1 2
Therewith the conditions are accomplished in order to determine the maximum
overshoot, rise time and settling time that depends on the characteristics in the
frequency domain, e.g. natural frequency 0 and damping ratio . With 0
and the interesting items AW max dB and r can be calculated directly by the
Eqs. (A.3.1) and (A.3.2).
a) Determination of the maximum overshoot Mp :
For calculation of Mp the time t = tp > 0 will be determined at which h W (t)
will be rst zero according to Eq. (3.5.3). This is when the the sin function
in Eq. (3.5.3) has

1 2 0 t = .
This gives
tp =

1 2

(3.5.4)

From Eq. (3.5.2) it follows that the maximum overshoot is



Mp = hW (tp ) 1 = e

1 2 = f () .
1

(3.5.5)

The maximum overshoot is therefore only a function of the damping ratio

as shown in Figure 3.5.3.
b) Determination of the rise time tr,50 :
In the following the rise time will not be calculated by the tangent at the
turning point, but by the tangent at time t = t50 (see Figure 3.3.1), where
hW (t) reaches 50% of the stationary value hW = 1. So the time t50 must be
determined, for which according to Eq. (3.5.2) hW (t50 ) = 0.5 is valid. From
Eq. (3.5.2a) it follows that

3.5 Design of controllers using pole-zero compensators

M

123

1 0 0 %
8 0 %
6 0 %
4 0 %
2 0 %
0 %
0

0 .1

0 .2

0 .3

0 .5

0 .4

0 .6

0 .7

0 .8

0 .9

1 .0

Figure 3.5.3. Maximum overshoot Mp = f1 () (in %) relative to hW = 100% as

function of the damping ratio


0.5 = 1 e0 t50 cos



1 2 0 t50 + 

1 2




sin
1 2 0 t50
.

This equation for the product 0 t50 must be evaluated numerically. One gets
a function of the form
(3.5.6)
0 t50 = f2 () .
From Eq. (3.5.3) it follows that
tr,50


1
1 2

 ,
=
=
h W (t50 )
0 e0 t50 sin
1 2 0 t50

and from this together with Eq. (3.5.6) the normalised rise time is
0 tr,50


1 2

 = f2 () ,
=

ef2 () sin
1 2 f2 ()

(3.5.7)

which also only depends on the damping ratio . This relationship is shown
in Figure 3.5.4.
c) Determination of the settling time t :
Using Eq. (3.5.2b) the decay of the amplitude to a value less than for t t
can be estimated from the envelope of the response
e0 t

.
1 2
From this the normalised settling time

124

3 Behaviour of linear continuous-time control systems

w
0

r ,5 0

( z )
3
2
1
0
0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1 .0

Figure 3.5.4. The product 0 tr,50 = f2 () (normalised rise time) as a function of the
damping ratio

0 t

1
1 1
ln 
1 2

(3.5.8)

follows. If = 3%(=0.03)

is chosen, one gets

0 t3%


1
3.5 0.5 ln(1 2 ) = f3 () .

(3.5.9)

This relationship is shown in Figure 3.5.5 together with the normalised rise
time 0 tr that will be shown later in Figure 3.5.9.
Comparing the results from Figures 3.5.3 to 3.5.5 one can summarise as follows:
The maximum overshoot Mp depends only on the damping ratio .
A change of the damping ratio in the range of approximately < 0.9
behaves contrary to the settling time t compared to the rise time tr,50 ,
i.e. an increase of the damping ratio in order to obtain a smaller settling
time t increases the rise time tr,50 .
For a xed damping ratio the parameter 0 determines the speed of the
control loop. A large value of 0 shows a small settling and rise time.
For the practical application of the diagrams in Figures 3.5.3 to 3.5.5 the following
example is given.
Example 3.5.1
The response on step changes in the set-point value hW (t) of a closed loop with
a dominant pair of poles should show a maximum overshoot of Mp 10%, a rise
time of tr,50 1 s and a settling time of t3% 4 s. How must the damping ratio
and the natural frequency 0 be chosen?

3.5 Design of controllers using pole-zero compensators

125

w 0tr( z )
1 5
w 0t3% ( z )
w 0t

5 %

( z )
w 0t
1 0

w 0t
w 0t

0 .1

0 .2

3 %

w 0t
r

5 %

w 0t
5

3 % a

5 % a

0 .3

0 .5

0 .4

0 .6

0 .7

0 .8

0 .9

1 .0

Figure 3.5.5. Normalised settling time 0 t3% f3 () and normalised rise time 0 tr as
functions of the damping ratio

With the given value of Mp one obtains from Figure 3.5.3 the damping ratio
= 0.58 .
For this value of with tr,50 = 1 s the natural frequency
0 =

f2 (0.58)
= 2.05 s1
1s

follows from Figure 3.5.4. But from Figure 3.5.5 for t3% = 4 s the required natural
frequency is
f3 (0.58)
= 1.6 s1 .
0 =
4s
The rise time of tr,50 = 1 s is the sharper requirement. Therefore, 0 = 2.05 s1
must be chosen. For the pair (0 ,) from Eq. (A.3.1) the resonant peak frequency

p = 0 1 2 2 = 1.17 s1
and from Eq. (A.3.2) the resonant peak
Mr =

1
1 2

= 0.06

giving
Mr = 0.49 dB ,
respectively, can be determined.

126

3 Behaviour of linear continuous-time control systems

In order to estimate the bandwidth b for a given damping ratio , the relationship between these two parameters is often needed. Based on the bandwidth b
as dened in Figure 2.3.19, that is
1
|GW (jb )| = |GW (0)| ,
2
it follows after a short calculation using Eq. (3.5.1) for s = j and = b that
0

b
= 1 2 2 + (1 2 2 )2 + 1 = f4 ()
(3.5.10)
0
0

and
b = tan1


(1 2 2 ) + (1 2 2 )2 + 1

= f5 () .
2 2 (1 2 2 )2 + 1

(3.5.11)

Furthermore, one obtains using Eqs. (3.5.7) and Eq. (3.5.10)

b tr,50 = f2 () f4 () = f6 () .

(3.5.12)

The graphs of the functions f4 (), f5 () and f6 () are shown in Figure 3.5.6.
w b
w 0
1 .5

w b
=
w 0

f4( z )
w
b

r ,5 0

r ,5 0

= f6( z )

1 .0

2
|j
b

|= f5( z )

0 .5

0
0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1 .0

Figure 3.5.6. Characteristics b /0 = f4 (), b = f5 () and b tr,50 = f6 () depending on the damping ratio of the closed loop with PT2 S behaviour

By approximation of f4 (), f5 () and f6 () the following rules of thumb can be

determined:
b
1.84 1.21
for 0.3 < < 1.0 ,
(3.5.13)
1.
0
for
0 0.8 ,
(3.5.14)
2. |b | 2.23
3.

b tr,50 2.3

for

0.3 < < 0.8 .

(3.5.15)

3.5 Design of controllers using pole-zero compensators

127

Applying these rules to Example 3.5.1 with 0 = 2.05 s1 and = 0.58, the
bandwidth b can be determined either from Eq. (3.5.13) as
b 2.05 (1.84 1.21 0.58) = 2.33 s1
or with tt,50 = 1 s from Eq. (3.5.15) as
b 2.3 s1 .
The Bode plot of a typical corresponding open-loop frequency response G0 (j)
is shown in Figure 3.5.7. From this and from Eqs. (2.4.18) and (2.4.19) one can
use the characteristics:
crossover frequency C ,
phase margin C = 180 + (C ),
gain margin AP dB = |G0 (P )|dB .
|G

( w ) |
0

d B

0
j
0

w
w

( w )
0

A
P

d B

w
j

- 1 8 0

Figure 3.5.7. Bode plot of the open loop

Since the closed-loop transfer function has been assumed to be approximated by

Eq. (3.5.1), the corresponding open-loop transfer function is
G0 (s) =

02
GW (s)
=
1 GW (s)
s(s + 20 )

(3.5.16a)

1
K0
s 1 + Ts

(3.5.16b)

or
G0 (s) =

128

3 Behaviour of linear continuous-time control systems

G

0 d B

0
j

T
C

- 4 0 d B / d e c a d e
w

- 9 0
- 1 8 0

Figure 3.5.8. Bode plotf the open loop with G0 (s) according to Eq. (3.5.16b)

with K0 = 0 /(2) and T = 1/(20 ). The frequency response of Eq. (3.5.16)

is shown in Figure 3.5.8. This Bode plot is considerably dierent from that of
Figure 3.5.7. The system in Figure 3.5.7 does not have an integrator. Furthermore,
it is of order higher than two, as the phase characteristic exceeds the value of 180 . But close to the crossover frequency C , both Bode plots show a similar
behaviour. If for the magnitude response of a given system |G0 (j)| >> 1 is
valid for << D and |G0 (j)| 0 for >> D , then G0 (s) can often be
approximated in the vicinity of the crossover frequency C by Eqs. (3.5.16a) and
(3.5.16b). The associated transfer function GW (s) contains a dominant conjugate
complex pair of poles. In order to transfer the known performance indices of
a second-order system to control systems of higher order, the design must be
performed such that the magnitude response |G0 (j)| decreases by 20dB/decade
in the vicinity of C . For Eq. (3.5.16b) this is only possible if C < 1/T is valid
(compare with Figure 3.5.8). From Eq. (3.5.16a) one obtains under the condition
|G0 (jC )| = 1
after a short calculation
C
=
0

0

4 4 + 1 2 2 = f7 () .

(3.5.17)

With T = 1/(20 ) for C < 1/T from

0

4 4 + 1 2 2 < 2

the condition > 0.42 follows. When for the damping ratio a value of > 0.42 is
chosen, then it is guaranteed that the magnitude response |G0 |dB of the open loop
falls o in the vicinity of the crossover frequency C by 20dB/decade. Figure 3.5.9
shows that only the interval 0.5 < < 0.7 is a range of suitable damping ratios,
since both, the rise time and the maximum overshoot, show acceptable values

3.5 Design of controllers using pole-zero compensators

h

129

z = 0 .2
W

0 .3
0 .4
0 .5

1 .4
1 .2
1 .0

0 .7
0 .9

0 .8
0 .6
0 .4
0 .2
0
0
1

w 0t

Figure 3.5.9. Step response hW (t) of the closed loop with PT2 S behaviour according
to the transfer function GW (s) of Eq. (3.5.1)

from the performance index point of view. This also means that the phase and
gain margin C and AP dB show proper values.
From these considerations one can conclude that for control systems with minimumphase behaviour, which can be approximately described by a PT2 S element, the
magnitude response |G0 (j)|dB of the open loop must decrease by 20dB/decade in
the vicinity of the crossover frequency C if a good performance is to be achieved,
i.e. a sucient large phase margin C .
As already mentioned in section 2.4.3.6, the crossover frequency C is an important performance index of the dynamical behaviour of the closed loop. The larger
C , the larger is the bandwidth b of GW (j) in general, and the faster is the
reaction to set-point changes. For the frequency response for set-point changes
one gets approximately

1
for |G0 (j)| >> 1
G0 (j)
(3.5.18)

GW (j) =
1 + G0 (j)
G0 (j) for |G0 (j)| << 1 .
From this, the asymptote of the magnitude response of GW (j) can be determined
(Figure 3.5.10). If |G0 (j)|dB decreases in the vicinity of C by 20dB/decade, then
for this range
C
G0 (j)
j

130

3 Behaviour of linear continuous-time control systems

is valid, and thus it follows that

1

GW (j)

G
G

0 d B
W

|G
0

( jw ) |d

1+j

.
b

- 2 0 d B / d e c a d e

d B

0
|G
W

( jw ) |d
B

lo w e r

w
C

m id d le

u p p e r

fre q u e n c y ra n g e

Figure 3.5.10. Piecewise determination of |GW (j)|dB from |G0 (j)|dB in the Bode
diagram

GW (j) behaves in this range as a PT1 element. As generally known, the magnitude response of a PT1 element decreases by 3dB at the breakpoint frequency
(here B = C ). Therefore, the crossover frequency C of the open loop is just
the bandwidth b of the closed loop, i.e. C = b . From this it follows that
for minimum phase systems the frequency response of GW (j) can be determined piecewise from G0 (j) according to Figure 3.5.10. Thereby for fullling
Eq. (3.5.18) in the lower frequency range the value of |G0 (j)| and therefore also
the loop gain K0 must be large to hold the steady-state error as small as possible. This lower frequency range of |G0 (j)| is responsible for the steady-state
behaviour of the closed loop, whereas the middle frequency range is essential for
the transient behaviour and is characteristic for the damping. In order to avoid
non-suppressable high-frequency disturbances of the set point w(t) in the closed
loop, |G0 (j)| and therefore also |GW (j)| must decrease quickly in the upper
frequency range.
From these ideas it is now possible to specify besides Eq. (3.5.17) additional
important relationships between the characteristics of the time response of the
closed loop and the characteristics of the frequency response of the open and
partly of the closed loop. Using Eq. (3.5.7) and Eq. (3.5.17) it follows immediately
that
(3.5.19)
C tr,50 = f2 () f7 () = f8 () .
Figure 3.5.11 shows the graphical representation of f8 (). It is easy to check that
this curve can be described in the range of 0 < < 1 by the approximation

3.5 Design of controllers using pole-zero compensators

C tr,50 1.5

131

Mp [%]
250

(3.5.20a)

or
C tr,50 1.5

Mp 20%

for

or

> 0.5 .

(3.5.20b)

A further relationship may be determined from the crossover frequency C for

the phase margin as
C = 180 + arg G0 (jC )

= 180 90 tan

1 C
2 0


,

which yields




0
1
= f9 () .
= tan1 2
C = tan1 2
C
f7 ()

(3.5.21)

Figure 3.5.11 also shows this function. By superposition of f9 () with f1 () one

can show that in the range of the mainly interesting values of the damping 0.3
0.8 the approximation
C [ ] + Mp [%] 70

(3.5.22)

is valid. This rule of thumb can only be applied for values of the variables with
the given dimensions in squared brackets.

3.5.2

Controller design using frequency domain characteristics

The relations fi ()(i = 1,2, . . . ,9) derived in the previous section for the closed
loop behaviour of a PT2 S element, can be applied also to higher-order systems as
long as these systems have a dominant pair of poles. For this class of systems an
ecient synthesis method exists as shown in the following. The starting point of
this method is the representation of the frequency response G0 (j) of the open
loop on a Bode diagram. The specications of the closed loop that must be met
are rst given as characteristics of the open loop according to the above section.
The synthesis requires in the choice of a controller transfer function GC (s), which
modies the open-loop transfer function such that the required characteristics are
met. The method consists of the following steps:
Step 1: In general, the characteristics of the time response of the closed loop,
Mp , tr,50 and e , are given. On the basis of these values from Table 3.2.1 the
gain K0 , from the rule of thumb for C tr,50 1.5 according to Eq. (3.5.20b),
the crossover frequency C and from C [ ] 70 Mp [%] the phase margin
C will be determined, and from f1 () the damping ratio .

132

3 Behaviour of linear continuous-time control systems

w C
w 0

w
w C
= f7( z )
w 0

1 .0
j
C

0 .9

8 0 0 .8
w

7 0 0 .7

r ,5 0

r ,5 0

1 .5

= f8( z )

1 .4

6 0 0 .6
5 0 0 .5

1 .3

4 0 0 .4

|j

3 0 0 .3

|= f9( z )

1 .2

2 0 0 .2
1 0 0 .1
0
0
0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1 .0

1 .1
z

Figure 3.5.11. Frequency domain characteristics of the open loop, C and C , depending on the damping ratio of the closed loop with PT2 S behaviour

Step 2: First a P element will be chosen as controller such that the gain K0
determined during step 1 will be met. By inserting additional elements in
series (often called compensator or correction elements) G0 will be changed
such that the other values from step 1, C and C , can be achieved while
the amplitude response |G0 (j)|dB decreases by 20dB/decade in the vicinity
of the crossover frequency C .
Step 3: It must be checked whether the response meets the required specications. This can be performed directly by determining Mp , tr,50 and e
by simulation, or indirectly by using the formula in section 3.5.1 for the resonant peak Mr according to Eq. (A.3.2) and the bandwidth b according to
Eq. (3.5.15). These values must be veried by calculation of the closed-loop
frequency domain characteristic
GW (j) =

G0 (j)
1 + G0 (j)

from the open-loop characteristic. In the case of too large deviations from
the approximations of Mr and b , step 2 must be repeated in a modied
form.
This method does not inevitably deliver a proper controller during the rst run
and it is a trial-and-error method that leads generally to satisfactory results after
some recursions.

3.5 Design of controllers using pole-zero compensators

133

For the design of this controller the methods given in section 3.4 for a standard
controller are usually not sucient. The controller must be composed of dierent
elements as shown above in step 2. In this procedure two special elements are
of important interest, which have to perform a phase shift as shown below:
a) The lead element
The increasing phase shift element is used to increase the phase in a certain
frequency range. The transfer function of this element is
s
1 + Ts
1/T
, 0<<1.
=
GC (s) =
s
1 + T s
1+
1/(T )
1+

(3.5.23)

For s = j the frequency response

Z
GC (j) =

1+j
N
1+j

(3.5.24)

follows with the two breakpoint frequencies

Z =

1
T

(3.5.25a)

and

1
.
T
A further characteristic is the frequency ratio
N =

mh =

N
1
= >1.
Z

(3.5.25b)

(3.5.26)

From Eq. (3.5.23) the frequency response

GC (j) = A() ej()
1
1 + T 2 2 j(tan1 T tan1 T )
=
e
1 + 2 T 2 2

(3.5.27)

follows. The Nyquist plot is shown in Figure 3.5.12 and it is a semicircle. The
maximum phase shift
() = tan1 T tan1 T
can be determined from the condition d()/d = 0 for
max =

1
1
Z N = N
= Z mh =
mh .
mh
T

(3.5.28)

134

3 Behaviour of linear continuous-time control systems

j Im [G
C

]
w m a x
G

j m a x
1

( jw )

w = 0

1
a

R e [G
C

Figure 3.5.12. Nyquist plot of a lead element

G
C

d B

(a )d B
1

- 2 0 d B / d e c a d e

0
w

D G C
d B
w

j
j

m a x

w m a x
w

Figure 3.5.13. Bode diagram of the lead element

As shown by the Bode plot in Figure 3.5.13 the lead element has at high frequencies an undesirable increase in the magnitude response of
|GC |dB = 20 log 10 (1/) = 20 log10 mh .
If Eq. (3.5.23) is broken down as
1 + T s (1 ) T s
+
1 + T s
1 + T s


T s
1
1
,
=1+

1 + T s

GC (s) =

(3.5.29)

the lead element consists of a parallel connection of a P element with gain 1 and
a DT1 element, which is a special PDT1 controller (compare Eq. (3.4.9)). For the
step response one obtains

t



1
1 e T ,
(3.5.30)
hC (t) = (t) 1 +

3.5 Design of controllers using pole-zero compensators

135

h C (t)
m
h

=
a

a T
0

Figure 3.5.14. Step response of the lead element

which is shown in Figure 3.5.14. For the practical design of lead elements the normalised phase diagram in Figure 3.5.15 is helpful. If the frequency max is known,
from this diagram the frequency ratio mh can be determined. The lower breakpoint frequency Z can be either read from the diagram directly or calculated
from Eq. (3.5.28).
j
6 0

m = 1 5

1 2
1 0
8

5 5
5 0

4 5

4 0
4

3 5
3 0
3

2 5
2 0
2

1 5
1 0
5
0

1 0 -1
2

8 1 0 0
2

8 1 0 1
2

Figure 3.5.15. Normalised phase responses of the lead element:

= ;
== /Z ; m
= mh ;
Z = lower breakpoint frequency; mh Z = upper breakpoint frequency;
normalised phase responses of the lag element:
= ;
= /N ; m
= ms ;
N = lower breakpoint frequency; ms N = upper breakpoint frequency

8 1 0 2

2 1 0 2
w

136

3 Behaviour of linear continuous-time control systems

Example 3.5.2
The phase response of a transfer function must be shifted by = 30 at =
max = 4 s1 . The maximum of the phase shift of = 30 is from Figure 3.5.15
for
= /Z 1.7 and mh = 3. With = max = 4 s1 follows for the lower

breakpoint
frequency Z max /1.7 or from Eq. (3.5.28) Z = max / mh =

4/ 3 = 2.31 s1 and with Eq. (3.5.26) for the upper breakpoint frequency N =


mh Z = 6.93 s1 .
b) The lag element
The lag element is used to decrease the magnitude response above a certain
frequency. Hereby a undesirable decrease of the phase response occurs in a certain
frequency range. The transfer function of this lag element is
s
1 + Ts
(1/T )
=
GC (s) =
s
1 + T s
1+
(1/T )
1+

with

>1.

(3.5.31)

1
1
and N =
the frequency
For s = j and the breakpoint frequencies Z =
T
T
response is

1+j
Z
(3.5.32)
GC (j) =
.
1+j
N
Also in this case a frequency ratio can be dened as
ms =

Z
=>1.
N

(3.5.33)

The decrease of the amplitude response at high frequencies is

|GC |dB = 20 log10

1
= 20 log10 ms .

(3.5.34)

Figure 3.5.16 shows the Nyquist plot and Figure 3.5.17 the Bode diagram of the
lag element. The rearrangement of Eq. (3.5.31) into
1
1
1

GC (s) = +
1 + T s

(3.5.35)

shows that the lag element consists of a parallel connection of a P element with
gain 1/ and a PT1 element with gain (1 1/) and time constant T . The step
response of this lag element follows from Eq. (3.5.35) as

t



1
1
1 e T
(3.5.36)
hC (t) = (t) + 1

3.5 Design of controllers using pole-zero compensators

j Im [G

]
C

j m a x

R e [G
1

137

]
C

w = 0
w

w m a x
G

(jw )

Figure 3.5.16. Nyquist plot of the lag element

R d B

w
0

w
N

w
Z

D G
-

- 2 0 d B / d e c a d e

( a1 ) d B
j

w m a x

d B

j m a x

Figure 3.5.17. Bode diagram of the lag element

h
C

(t)
1

a T

Figure 3.5.18. Step response of the lag element

and is shown in Figure 3.5.18. It is easy to see that this relation is equal to
Eq. (3.5.30) but with > 1. For practical working the phase diagram of Figure 3.5.15 can be used, which is in this case in principle the same as that for the
lead element but with dierent parameters and ipped over.

138

3 Behaviour of linear continuous-time control systems

Example 3.5.3
The magnitude response |G0 |dB of an open-loop system should be decreased at
= 10 s1 by 20dB, whereby the maximum phase shift must be 10 . From
Eq. (3.5.34) it follows that |GC |dB = 20 dB = 20 log10 ms and from this ms = 10.
With = 10 and ms = 10 one obtains from the phase response
= /N
50, and with = 10 s1 for the breakpoint frequencies N = 10/50 s1 = 0.2 s1
and Z = ms N = 2 s1 .

3.5.3

Application of the design using frequency domain characteristics

In the following the design using the frequency domain characteristics will be
demonstrated by an example. The plant is given with the transfer function
GP (s) =

1

s .
(1 + s) 1 +
3

(3.5.37)

For the step response of set-point changes hW (t) of the closed loop the rise time
tr,50 = 0.7 s

(3.5.38a)

Mp = 25%

(3.5.38b)

and the maximum overshoot

is specied. Furthermore, for a ramp command input
w(t) = w1 (t) t
of the set point, the closed loop should show a steady-state error of
e =

1
.
20

(3.5.38c)

The design method will be performed according to section 3.5.2, assuming that
the resultant system will have a dominant pair of complex poles, using the following steps:
Step 1:
One obtains from Eq. (3.5.20) with the specication according to Eq. (3.5.38a)
the approximation of the crossover frequency as


1
1
Mp [%]
=
(1.5 0.1) 2 s1
(3.5.39a)
1.5
C
tr,50
250
0.7

3.5 Design of controllers using pole-zero compensators

139

and from Eq. (3.5.22) with Eq. (3.5.38b) the phase margin as
C [ ] 70 Mp [%] = 45 .

(3.5.39b)

From the specication Eq. (3.5.38c) one obtains for a ramp command input
according to Table 3.2.1 (case k = 1) for xe1 w1 = 1 the open-loop gain as
K0 = KC KP = 20 .

(3.5.39c)

Step 2:
A) First, an I controller is chosen, GC1 (s) = KC /s, with the gain KC such that
Eq. (3.5.39c) is fullled, here KC = 20. The Bode diagram of the open loop
transfer function
G01 (s) = GC1 (s) GP (s) =

20

s
s(1 + s) 1 +
3

(3.5.40)

is plotted in Figure 3.5.19. From this it can be seen that in order to achieve
the requirements of Eqs. (3.5.39a) and (3.5.39b)
1. the phase of G01 (j) must be increased at = C by 53 , and
2. the magnitude of G01 (j) decreased at = C by 11dB .
B) In order to fulll the rst requirement, the I controller will be extended by
a lead element, whose phase response at = C = 2 s1 has a maximum of
(53 + 6 ). Here in this case a 6 larger value is used, as by the usage of a
lag element in step 3 a small unavoidable decrease of the phase occurs. From
the phase diagram (Figure 3.5.15) one reads for max = 59 the frequency
ratio of
mh 12 .
From this, one obtains for = max = C with Eq. (3.5.28) or likewise from
Figure 3.5.15 the breakpoint frequencies
C
0.6 s1
Z =
mh
and

N = Z mh 7.2 s1 .

The transfer function of the extended controller is

s
0.6 .
GC2 (s) = 20
s
)
s(1 +
7.2
1+

(3.5.41)

140

3 Behaviour of linear continuous-time control systems

The open loop has now the transfer function
s
0.6

G02 (s) = GC2 (s) GP (s) = 20
s  .
s 
1+
s(1 + s) 1 +
3
7.2
1+

(3.5.42)

The associated frequency-domain characteristics are shown in Figure 3.5.19.

Because of the additional lead element also the magnitude response of G02 (s)
is slightly changed. Hence, in the following step the magnitude response at
= C must be decreased by 22dB instead of 11dB.
C) In order to reach this decrease of the magnitude at = C = 2 s1 the
controller will be further extended by a lag element. From Eq. (3.5.34) it
follows that
20 log 10 ms = 22 dB
and from this
ms = 12.6 .
2 s1

must not be inuenced too much such that the

The phase at C =
upper breakpoint frequency Z , and therefore also the lower breakpoint frequency N must be suciently far left from C . By the special choice of the
lead element in step 1, the lag element introduced during this step may have
a maximum phase decrease of 6 . From this, one obtains with ms = 12.6
from Figure 3.5.15

= 125
N
and especially for = C = 2 s1
N =

C
= 0.016 s1 .
125

For the upper breakpoint frequency it follows that

Z = N ms = 0.2 s1 .
The transfer function of the nal controller is given by
s
s
1+
0.6
0.2
GC (s) = 20
s
s .
) 1+
s(1 +
7.2
0.016
1+

(3.5.43)

The transfer function of the open loop is


s 
s 
1+
1+
0.2
 0.6
G0 (s) = 20 
s 
s  .
s 
(1 + s) 1 +
1+
s 1+
0.016
3
7.2
The associated frequency response characteristics are shown in Figure 3.5.19.

3.5 Design of controllers using pole-zero compensators

141

G d B
1 0 0

(a )
G 0
2 d B

5 0
G
0

0 d B

G
0

d B
1

2 2 d B

1 0

1 1 d B

0 .1

1 0 0

[s -1 ] w

w C

- 5 0

- 1 0 0

- 1 5 0

- 2
- 4
- 6
- 8
1 0
1 2
1 4
1 6
1 8
2 0
2 2
2 4
2 6
2 8

0 .1

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

j 0 = a rg G 0
1

1
j 0
2

= a rg G 0

2
w

1 0

1 0 0

[s -1 ] w

(b )
6

j
0

= a rg G
0

5 3

j C =

4 5

Figure 3.5.19. The magnitude (a) and phase (b) response of G01 (j), G02 (j) and
G0 (j)

Step 3:
As the design based on frequency-domain characteristics is an approximate
method, one should verify the results by simulation studies and make sure
of the match of the the results with the initial specications. The simulation

142

3 Behaviour of linear continuous-time control systems

results for the time responses are shown in Figure 3.5.20. The required specications for tr,50 , Mp and e are fullled. Furthermore, the manipulated
h

r
W

1 .0 0
4

1 .2 5
h

0 .7 5

( t)
3
r

0 .5 0

( t)
2

0 .2 5
0

1
1
0

t [s ]
7

r ,5 0

0 ,6 7

Figure 3.5.20. Step response hW (t) and ramp response rW (t) of the designed control
system

variable u(t) must be checked for realisability. If this in not the case, the
initial specications for the control system must be modied.

3.5.4

Controller design using the root-locus method

The design using the root-locus method is directly connected to the considerations
in section 3.5. There the specications of maximum overshoot, rise and settling
time for a closed loop having a dominant pair of poles have been converted into the
conditions for the damping ratio and for the natural frequency 0 of the related
transfer function GW (s). With and 0 the poles of the transfer function GW
are directly tightened as shown in Figure 3.5.2. An open-loop transfer function
G0 (s) must now be determined such that the closed loop has a dominant pair of
poles at the desired position, which is given by the values of 0 and . Such an
approach is also called pole assignment.
The root-locus method is as generally known a graphical method, which is
used to analyse the position of the closed-loop poles. This method oers the
possibility to combine in the complex s plane the desired dominant pair of poles
with the root locus of the xed part of the loop and to deform the root locus by
adding poles and zeros such that two of the branches traverse through the desired
dominant pair of poles at a certain gain K0 . Figure 3.5.21 shows, how the root
locus can be deformed to the right by adding a pole and to the left by adding a
zero.
The principal strategy during the controller design by the root-locus method will

3.5 Design of controllers using pole-zero compensators

143

jw

jw

(a )

jw

(b )

jw
s

Figure 3.5.21. Deforming of the root locus (a) to the right by an additional pole, (b)
to the left by an additional zero in the open loop

be shown in the following using two examples.

Example 3.5.4
Given is a plant described by the transfer function
GP (s) =

1
.
s(s + 3) (s + 5)

(3.5.44)

For this plant a controller must be designed, such that the step response hW (t)
of the closed loop shows the following properties:
Mp = 16%

and

tr,50 = 0.6 s .

First, these specications will be transformed. The conditions for and 0 are
from Figures 3.5.3 and 3.5.4
0.5

and

1.85
= 3.1 s1 .
0.6 s

In order to have a geometrical interpretation one should consider Figure 3.5.22,

where a pair of complex poles

sa,b = 0 j0 1 2
is shown. The distance d of both poles sa,b from the origin is

d =

02 2 + (1 2 ) 02 = 0 .

(3.5.45)

0
=
0

(3.5.46a)

The angle is
cos =

144

3 Behaviour of linear continuous-time control systems

s p la n e

jw
s a

w
-z w

jw

1 - z 2
0

s
-jw
0

1 - z

Figure 3.5.22. Pair of conjugate complex poles in the s plane

or

= cos1 ,

(3.5.46b)

where for the current case of 0.5 the condition 60 is met. The damping
ratio describes the angle , the frequency 0 the distance d of the dominant
pair of poles from the origin.
During the design one will try to increase the damping ratio not unnecessarily high
as this step causes an increase in the rise time tr,50 for a given natural frequency
0 (Figure 3.5.4). Increasing the natural frequency 0 implies an increase of
the speed of the control loop. But this parameter should not be unnecessarily
increased, otherwise the dominance of the pair of poles may be lost.
Figure 3.5.23 shows the root locus of the closed loop using a P controller. Potential
positions for the dominant pair of poles are drawn by the two thick blue lines H1
and H2 . It is obvious that the design using a pure P controller (changes in the
gain K0 ) does not lead to the goal, as the root locus does not traverse the two
lines H1 and H2 . Equally it is clear which steps have to be taken, such that the
two branches of the root locus under consideration traverse the two lines H1 and
H2 . If the two poles s2 = 3 and s3 = 5 are shifted further left, the centre of
gravity of the poles will move left and with it the total curve without changing
the structure of the system. One possibility is to perform this shift by a simple
lead element which compensates the pole s2 = 3 by a zero and a pole s4 = 10.
The resulting controller transfer function is
GC (s) = KC

1 + s/3
s+3
= 3.33 KC
1 + s/10
s + 10

(3.5.47)

and the transfer function of the modied open loop

G0 (s) = 3.33 KC

1
.
s(s + 10) (s + 5)

(3.5.48)

The root locus of the closed loop is shown in Figure 3.5.24. It traverses the line
H1 at sk . The associated gain at this point can be determined via the distances
to the three poles from Eq. (2.5.19)

3.5 Design of controllers using pole-zero compensators

H 1

145
jw
5 j
4 j
3 j
2 j

s
- 1 0

- 5

s
2

- 3

j
1

- j

- 2 j
- 3 j
- 4 j
H 2

- 5 j

Figure 3.5.23. Root locus of GW (s) (plant with P controller) and potential positions
of the dominant pairs of poles (blue thick lines)

k0 = 3.33 KC = |sk 0| |sk + 5| |sk + 10| = 3.3 4.4 8.8

as
KC = 38.4 .
The step response of the closed loop in Figure 3.5.25 shows that the specications
are achieved.
It must be mentioned that a complete compensation (cancelling) of the pole
s2 = 3 cannot be realised exactly, as the plant parameters are not exactly
known or may change within some bounds. Thus the root locus will dier in
the vicinity of the compensated pole from the ideal case of Figure 3.5.24, but


intersection with the two lines H1 and H2 is almost invariant.
One realises by means of this example that the root-locus method is well suited
to provide a fast and rst overview about the principal possibilities of corrections.
Often the knowledge about the asymptotes is already sucient. The root-locus
method is especially suited for stabilising unstable plants. This will be explained
by the second example.
Example 3.5.5
Given is an unstable plant with the transfer function

146

3 Behaviour of linear continuous-time control systems

jw

H 1

5 j

4 j
s

3 j
k

2 j
s

s
4

s
3

- 5

- 1 0

j
1

- j
- 2 j
- 3 j
- 4 j
H 2

- 5 j

Figure 3.5.24. Root locus of GW (s) with modied controller according to Eq. (3.5.47)

GP (s) =

1
.
(s + 1) (s + 5) (s 1)

(3.5.49)

At rst, it is close to compensate the pole s1 = 1 by a corresponding zero.

This would be possible, e.g. using a rst-order all-pass element with the transfer
function
s1
.
GC (s) =
s+1
But because of the reason given above, the compensation of the unstable pole will
be practically never complete. So it must be performed without this procedure
for stability reasons.
Another possibility is to provide feed back around the unstable plant so that
the original unstable pole in the closed loop is shifted into the left-half s plane.
A simple P controller would yield the root locus shown in Figure 3.5.26. This
conguration is not stabilisable, as the two branches on the right-hand side remain
in the right-half s plane for all gain values. But if a controller is used that has
a double zero at s = 1 and a pole at s = 0, the pole at s = s2 = 1 will be
substituted by a zero. The root locus will be deformed to the left as shown in
Figure 3.5.27.
This distribution of poles and zeros can be simply realised by a PID controller
with the transfer function

3.5 Design of controllers using pole-zero compensators

h

147

1 .2
M

= 1 5 .2 %

1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

t
0

Figure 3.5.25.
Eq. (3.5.47)

r ,5 0

= 0 .5 8 s
2

4
3

[s ]
5

Closed-loop step response with modied controller according to

jw
3 j

- 1 0

s
3

- 5

- 1 ,6 7

- 1

s
1

- 3 j

Figure 3.5.26. Root locus of the closed loop consisting of an unstable plant and a P
controller



(s + 1)2
1
= KC 2 + + s
GC (s) = KC
s
s


1
+ 0.5 s .
= 2KC 1 +
2s
From Figure 3.5.27 it can be seen for higher gain values than the critical gain

148

3 Behaviour of linear continuous-time control systems

jw
5 j

K 0
s

s
3

- 5

- 1 .5

c r it

s
4

- 1

- 5 j

Figure 3.5.27. Root locus of the closed loop consisting of an unstable plant and a PID
controller

K0 crit the closed loop is stable, as then all poles remain in the left-half s plane.
As the stability of the closed-loop system is not inuenced by the left-half-plane
poles, a compensation of these poles is possible. Even if this compensation is not
completely possible, the system will be stable. A compensation of the right-halfplane poles as discussed above should not be done.


Demonstration Example 3.5.1
A virtual experiment stabilising a pendulum

3.6

Compensator design methods

3.6.1

Basic ideas of compensator design

In the following some analytical design methods will be discussed, which will
lead directly to the design solution in a strongly systematic way. In contrary
to these direct design methods, the methods hitherto discussed, e.g. the design
method using frequency-domain characteristics or the root-locus method, are
indirect methods based more on systematic trial and error techniques iterating

3.6 Compensator design methods

149

through some design steps. The success depends strongly on the experience and
skill of the designer. The starting point was always the open loop, which was
modied iteratively by adding lead and lag elements until the closed loop shows
the desired behaviour.
Whereas in the direct design methods one will always start from the behaviour
of the closed loop. Mostly a desired transfer function GW (s) KW (s) is given.
In general this follows from the specication of some performance indices, which,
for example are required for the step response hW (t). For a series of appropriate
transfer functions a table of numerator and denominator polynomials of the associated transfer function KW (s) and its distribution of zeros and poles to yield
a specic response are given. Then for a known plant behaviour the required
controller can be directly calculated.
The controllers designed in this way are not always optimal. They guarantee
the compliance of the desired specication, e.g. maximum overshoot and settling
time. A drawback of theses methods is that they cannot be applied directly to
systems with dead time.

3.6.2

Design by specifying the closed-loop transfer function

The desired transfer function of the closed loop is given by

KW (s) =

0 + 1 s + . . . + v sv
(s)
=
, u>v ,
(s)
0 + 1 s + . . . + u su

(3.6.1)

where (s) and (s) are polynomials in s. In the following design methods, the
distribution of poles and zeros of KW (s) will be chosen, such that the performance
indices for the step response hW (t) are fullled. Often a detailed investigation of
the distribution of poles and zeros of the desired transfer function is not necessary,
especially when the transfer function does not contain zeros and when because
of the requirement KW (0) 1 just yields 0 = 0 and in the simplest case
KW (s) =

0
.
0 + 1 s + . . . + u su

(3.6.2)

For a closed loop with the transfer function according to Eq. (3.6.2) dierent possibilities exist, the so called standard forms, which can be used by a table lookup
for the step response hW (t), distribution of poles of KW (s) and the coecients
of the denominator polynomial (s).
A rst possibility is a distribution of poles with a real multiple pole at s = 0 .
Here and in the following sections, the term 0 is a relative frequency, not the
natural frequency. Thus one obtains for the step response of the desired behaviour
KW (s) =

0u
.
(s + 0 )u

(3.6.3)

150

3 Behaviour of linear continuous-time control systems

This is a series connection of u PT1 elements with the same time constant
T = 1/0 . This representation is also called a binomial form. The standard polynomials (s) of dierent order u are given in Table A.8.2. As this table further
shows, the normalised step response hW (0 t) will become slower with increasing order u. A design using this binomial form is only considered when the step
response hW is required to have no overshoot.
A further possibility of a standard form for KW (s) of Eq. (3.6.2) is the Butterworth
form. In this form, all u poles of KW (s) are equally distributed on a semicircle with
radius 0 in the left-half s plane and centred at the origin. Table A.8.2 contains the
standard polynomials (s) and the associated normalised step responses hW (0 t).
Numerous further possibilities for the development of standard forms of Eq. (3.6.2)
can be derived from the integral criteria given in Table 3.3.1. For example, the
minimum performance index JITAE is the basis of a standard form that is also
shown in Table A.8.2. Further, often the minimum settling time t is used as the
criterion. This table contains for = 5% the corresponding standard form.
Furthermore, for pole assignment the Weber method can be used. This species
the desired closed-loop transfer function
5k (1 + 2 ) 0k+2
(3.6.4)
(s + 0 + j0 ) (s + 0 j0 ) (s + 50 )k
by a real pole with multiplicity k = u 2 and a pair of complex poles. Table A.8.1
contains for dierent values of k and the normalised step responses hW (0 t).
By a proper choice of k, and 0 a closed-loop transfer function can be found
that fulls in many instances the desired performance.
KW (s) =

3.6.3

The method of Truxal and Guillemin

For the closed loop shown in Figure 3.6.1 the behaviour is described by the

+
W

G
C

(s)=

B (s)
A (s)
U

G
P

(s)=

D (s)
C (s)

Figure 3.6.1. Block diagram of the closed loop to be designed

transfer function
d0 + d1 s + d2 s2 + . . . + dm sm
D(s)
,
(3.6.5)
=
c0 + c1 s + c2 s2 + . . . + cn sn
C(s)
where the numerator and denominator polynomials D(s) and C(s) must have no
common roots. Furthermore, GP (s) is normalised to cn = 1 and m < n must be
valid.
GP (s) =

3.6 Compensator design methods

151

It is assumed that GP (s) is stable and minimum phase. For the controller to be
designed, the transfer function
GC (s) =

b0 + b1 s + b2 s2 + . . . + bw sw
B(s)
=
2
z
a0 + a1 s + a2 s + . . . + az s
A(s)

(3.6.6)

is chosen and normalised to az = 1. Because of the realisability of the controller

the relation w = degree B(s) degree A(s) = z must be valid. Now, the controller
must be designed such that the closed loop behaves like a given transfer function
for Eq. (3.6.1), whereby KW (s) should be freely chosen under the condition of
the realisability of the controller. From the closed-loop transfer function
GW (s) =

GC (s) GP (s)
!
= KW (s)
1 + GC (s) GP (s)

(3.6.7)

one obtains the controller transfer function

GC (s) =

KW (s)
1
GP (s) 1 KW (s)

(3.6.8)

or with the numerator and denominator polynomials given above

GC (s) =

C(s) (s)
B(s)
=
.
A(s)
D(s)[(s) (s)]

(3.6.9)

The condition of realisability for the controller is

degree B(s) = w = n + v degree A(s) = z = u + m
or
uv nm .

(3.6.10)

The pole excess (u v) of the desired closed-loop transfer function KW (s) must
be larger than or equal to the pole excess (n m) of the plant. Within these
constraints the order of KW (s) is free. According to Eq. (3.6.8) the controller
contains the inverse plant transfer function 1/GP (s). This is a total compensation
of the plant as shown in the block diagram of Figure 3.6.2. For the realisation
of the controller Eq. (3.6.9) is used, not the controller structure as shown in
this gure with the plant inverse 1/GP (s). As the controller implicitly contains
the plant inverse, i.e. the plant zeros are in the set of the controller poles and
the plant poles are in the set of the controller zeros, the plant must be stable
and minimum phase as mentioned at the beginning. Otherwise, the manipulated
variable and/or the controlled variable will show unstable behaviour.
Example 3.6.1
The plant transfer function is given as

152

3 Behaviour of linear continuous-time control systems

c o n t r o lle r

+
+
_

(s )
G

1
P

(s )

U
G

(s )

Figure 3.6.2. Compensation of the plant

GP (s) =

5
.
s(1 + 1.4s + s2 )

(3.6.11)

The pole excess of the plant is n m = 3. According to (3.6.10) the pole excess
of the desired closed-loop transfer function KW (s) must be
uv 3 .


The coecients of the transfer function KW (s) that obeys the realisability condition (3.6.10) are subjected to practical constraints, like the maximum range of
the manipulated variable, plant parameter errors and measurement noise in the
controlled variable, which is disturbing the controller output. The procedure for
the design of GC (s) will be demonstrated by the following example.
Example 3.6.2
For a plant with the transfer function
GP (s) =

1
(1 +

s)2 (1

+ 5s)

1
D(s)
=
2
3
1 + 7s + 11s + 5s
C(s)

(3.6.12)

a controller should be designed such that the closed loop shows optimal behaviour
in the sense of the performance index JITAE and has a rise time of tr,50 = 2.4 s.
First, it follows from the realisability condition Eq. (3.6.10) and from n m =
3 0 = 3 that the pole excess of the desired transfer function KW (s) is
uv 3 .
Inspecting Table A.8.2 one obtains from the JITAE form for u = 3 and v = 0 the
standard polynomial
(s) = s3 + 1.750 s2 + 2.1502 s + 03 .

(3.6.13)

From the associated step response hW (0 t) it follows from Table A.8.2 that the
normalised rise time
0 tr,50 = 2.4 ,

3.6 Compensator design methods

153

and with this value from the specied rise time tr,50 = 2.4 s the relative frequency
is 0 = 1 s1 . Eq. (3.6.13) is now
(s) = s3 + 1.75s2 + 2.15s + 1 .
As for the chosen standard form for KW (s) the numerator polynomial is (s) = 1,
so it follows from Eq. (3.6.9) that the compensator transfer function is
GC (s) =

1 + 7s + 11s2 + 5s3
C(s) (s)
=
D(s) [(s) (s)]
1 + 2.15s + 1.75s2 + s3 1

or
GC (s) =

1 + 7s + 11s2 + 5s3
.
s(2.15 + 1.75s + s2 )

This controller contains an integrator. The time responses are shown in Figure 3.6.3.


h
W

, h
U

, h
P

0 .2 h

h
U

1 .0
0 .8
h
P

0 .6
0 .4
0 .2
0
5

1 0

1 5

2 0

[s ]

2 5

Figure 3.6.3. Closed-loop behaviour for the example 3.6.2: hW (t) step response of
the controlled variable on step in the set point, hU (t) step response of the associated
controlled variable, hP (t) step response of the uncontrolled plant

If as a further example the plant given by Eq. (3.6.11) instead of Eq. (3.6.12) is
taken, then for the same KW (s) the controller is
GC (s) =

1 + 1.4s + s2
1 + 1.4s + s2
=
.
(2.15 + 1.75s + s2 ) 5
10.75 + 8.75s + 5s2

For these two very dierent plants the same closed-loop behaviour for the controlled variable can be achieved.

154

3 Behaviour of linear continuous-time control systems

In the considerations of this section it has been hitherto assumed that GP (s)
is stable and minimum phase. For plants that do not have this properties this
design method cannot be applied in this form. The method must be extended to
the following:
A direct compensation of the plant poles and zeros by the controller must be
avoided, otherwise stability problems would arise. In these cases, the closed-loop
transfer function KW (s) cannot be arbitrary. For a stable non-minimum phase
plant the transfer function KW (s) must be given such that the zeros of KW (s)
contain the right-half-plane zeros of GP (s). Whereas for an unstable plant the
zeros of the transfer function 1 KW (s) must contain the right-half-plane poles
of GP (s). Of course, this restricts the choice of KW (s) as the following examples
demonstrate.
Example 3.6.3
For an all-pass plant with the transfer function
GP (s) =

1 Ts
1 + Ts

a controller is to be designed such that the closed loop has the desired transfer
function
1
.
GW (s) KW (s) =
1 + T1 s
Using Eq. (3.6.9) one gets for the controller transfer function
GC (s) =

1 + Ts 1
.
1 T s T1 s

This controller gives a direct compensation (cancellation) of the plant zero. This
is undesirable as already discussed above, and KW (s) must be selected as
KW (s) =

1 Ts
.
(1 + T1 s)2

With Eq. (3.6.9) one obtains the controller transfer function as

GC (s) =

1 + Ts
.
s[(2T1 + T ) + sT12 ]

Because of this choice of KW (s), the closed loop shows also all-pass behaviour.
This eect is more intense the smaller the time constant T1 . Figure 3.6.4 shows
the time responses of this control system.


Example 3.6.4
The transfer function of the unstable plant
GP (s) =

1
1 sT

3.6 Compensator design methods

h
W

, h
U

, h

155

1 .0
h

0 .5

0 .0
h
W

- 0 .5
- 1 .0
0

[s ]

4
t

Figure 3.6.4. Closed-loop behaviour of the example 3.6.3: hW (t) step response of the
controlled variable on step in the set point, hU (t) step response of the associated controlled variable, hP (t) step response of the uncontrolled plant (T = 1 s; T1 = 0.5 s)

is given and a controller GC (s) is required for which KW (s) fullls the realisability
condition u v 1 and for which the zeros of 1 KW (s) must contain the plant
pole s = +1/T . This is expressed by the approach
1 KW (s) =

(1 sT ) K(s)
(s) (s)
=
,
(s)
(s)

whereby K(s) is chosen such that

degree[(1 sT ) K(s)] = degree (s)
is valid. In the present case KW (s) should be chosen such that degree (s) = u =
2. From this follows degree K(s) = 1. In order to have a stable KW (s) one can
take
K(s) = T1 s .
and obtain
(s) (s) = (1 T s) (T1 s) .
Observing the realisability condition, it follows that
(0 0 ) + (1 1 ) s + 2 s2 = T1 s + T1 T s2 .
Comparing the coecients and taking 2 = 1 one obtains
0 0 = 0, 1 1 = T1

and

T1 T = 1

156

3 Behaviour of linear continuous-time control systems

and nally
1
and 0 = 0 .
T
The parameters 0 and 1 are still free and may now be chosen such that taking an
acceptable behaviour of the manipulated variable into account, a given damping
ratio and natural frequency for KW can be reached. Without going into details,
T1 =

0 = 1

and

1 = 2 T1 =

2
T

will be chosen for the present case and from this it follows that
0 = 1

and

1 = 3 T1 =

3
.
T

The desired closed-loop transfer function will be

KW (s) =

1 + (3/T ) s
1 + (2/T ) s + s2

and
1 KW (s) =

(1 T s) (s/T )
.
1 + (2/T ) s + s2

The conditions for the design are fullled and for the controller transfer function
one obtains from Eq. (3.6.8) or Eq. (3.6.9)


T1
1 + (3/T ) s)
=
3
1
+
.
GC (s) =
(1/T ) s
3s
This design obviously produces a PI controller. The time responses of this control
system are shown in Figure 3.6.5 for T = 1 s. The relatively large maximum
overshoot cannot be avoided with an acceptable behaviour of the manipulated
variable.

3.6.4
3.6.4.1

Generalised compensator design method

The basic idea

With the following method a control system according to Figure 3.6.1 using the
controller given by Eq. (3.6.6) will be designed for a plant described by Eq. (3.6.5)
such that the closed loop behaves like the desired transfer function Eq. (3.6.1).
Hereby the orders of the controller numerator and denominator polynomials are
equal, i.e. w = degree B(s) = degree A(s) = z. The closed-loop poles are the roots
of the characteristic equation, which one obtains from

3.6 Compensator design methods

h
W

, h

157

1 .5
h

1 .0
0 .5

0 .5 h
U

0 .0
- 0 .5
- 1 .0
0

6
4

1 0

[s ]

1 2
t

Figure 3.6.5. Closed-loop behaviour of the example 3.6.4: hW (t) step response of the
controlled variable on step in the set point, hU (t) step response of the associated controlled variable

1 + GC (s) GP (s) = 0 .
With respect to the polynomials dened in Eqs. (3.6.5) and (3.6.6) this gives
P (s) (s) = A(s) C(s) + B(s) D(s) = 0 .

(3.6.14a)

On the other hand it follows from Eq. (3.6.1) that

P (s) (s) = 0 + 1 s + . . . + u su = u

u
6
(s s1 ) = 0 .

(3.6.14b)

i=1

This polynomial has order u = 2 + n, the coecients depend linearly on the plant
and controller parameters. Comparing both equations, the rst coecient is
0 = a0 c0 + b0 d0 ,

(3.6.15a)

and the last because of m < n and az = cn = 1

u = az cn = 1 .

(3.6.15b)

A general representation is given by

i = b0 di + b1 di1 + . . . + bw diw
+ a0 ci + a1 ci1 + . . . + az ciz ,

(3.6.15c)

158

3 Behaviour of linear continuous-time control systems

whereby
dk = 0

for

k < 0 and k > m

ck = 0

for

k < 0 and k > n ,

and w = z. The coecients i are obtained from the poles. For the rst, second
last and last one gets
u
6
(si )
0 =

u1 =

i=1
u

(si )

(3.6.16a)
(3.6.16b)

i=1

u = 1 .

(3.6.16c)

While the coecients i according to Eqs. (3.6.16) are directly given by the
closed-loop poles, the coecients i of Eq. (3.6.15) contain the required controller
parameters. Comparing both sides of the latter equation one obtains the synthesis
equation, which is a system of linear equations for 2z + 1 unknown controller
coecients a0 , . . . ,az1 ,b0 ,b1 , . . . ,bz . The number of equations is u = z + n. A
unique solution exists if z = n 1.
A detailed analysis shows, however, that a controller obtained in this way does not
usually achieve the desired goals. Because of its small gain a nite steady-state
error may occur. This must be taken into consideration during the design. For
plants with integral behaviour an order z = n 1 for the controller is sucient;
for proportional behaviour or when disturbances at the input of an integral plant
are taken into consideration, the gain must be inuenced so that an integral
behaviour of the controller can be obtained. This happens if the order of the
controller is increased by one, i.e. z = n, such that the system of equations is of
lower rank. This gives an additional degree of freedom and allows one to choose
the controller gain KC , which is usually introduced as a reciprocal gain factor:
a0
1
= cC =
.
KC
b0

(3.6.17)

Indeed, the order of the closed loop will be increased; it is now double that of the
plant order.

3.6.4.2

Zeros of the closed loop

In the method presented above, the zeros of the closed loop transfer function for
command changes
!

KW (s) = GW (s) =

B(s) D(s)
A(s) C(s) + B(s) D(s)

(3.6.18)

3.6 Compensator design methods

159

are obtained automatically. In fact, the zeros of the plant, i.e. the roots of D(s),
can be considered during the choice of the pole distribution and may be compensated, but the polynomial B(s) arises not in the design and must possibly be
compensated after this step. This can be done by introducing a pre-lter in the
feed-forward path according to 3.6.6a with a transfer function

(a )
c

W
B

(s)

B (s )
A (s)
_

D (s )
C (s )

c o n t r o lle r

(b )
W
B

c
K
K

(s)

+
_

B (s )
A (s )

D (s)
C (s )

c o n t r o lle r

Figure 3.6.6. Compensation of the plant zeros (a) with a controller in the feed-forward
path and (b) in the feedback path

GK (s) =

cK
.
BK (s)

The zeros of the controller and plant are compensated in this way. For stability reasons, this is only possible for left-half-plane zeros. If B + (s) and D+ (s) are
polynomials with only left-half-plane zeros and B (s) and D (s) the corresponding polynomials with only right-half-plane zeros including the imaginary axis, the
polynomials of B(s) and D(s) can be factorised as
B(s) = B (s) B + (s)

(3.6.19)

D(s) = D (s) D (s)

(3.6.20)

with

B (s) =

i
b
i s

i=0

(3.6.21a)
w = w+ + w

B (s) =

i=0

i
b+
i s

(3.6.21b)

160

3 Behaviour of linear continuous-time control systems

and

D (s) =

i
d
i s

(3.6.22a)

i=0

m = m+ + m

i
D (s) =
d+
i s .
+

(3.6.22b)

i=0

For the case that B(s) and C(s), and A(s) and D(s) do not have common divisors,
i.e. the controller does not compensate plant poles and zeros, the denominator
polynomial of the pre-lter can be determined as
BK (s) = B + (s) D+ (s) .

(3.6.23)

The transfer function for a command input is then

B(s) D(s)
cK
BK (s) A(s) C(s) + B(s) D(s)
cK B (s) D (s)
.
=
A(s) C(s) + B(s) D(s)

GW (s) =

(3.6.24)

If both, the controller and the plant, show minimum phase behaviour and their
transfer functions do not have zeros on the imaginary axis, all zeros of the closed
loop can be compensated, such that one obtains instead of Eq. (3.6.24)
GW (s) =

cK
.
A(s) C(s) + B(s) D(s)

(3.6.25)

If the closed-loop transfer function also contains given zeros, the transfer function
GK (s) should have a corresponding numerator polynomial. The coecient cK in
the numerator is used to make the gain KW of the closed-loop transfer function
GW (s) equal to 1. From Eq. (3.6.24) it therefore follows that
KW = GW (0) = cK

b
0 d0
=1.
a0 c0 + b0 d0

(3.6.26)

The expression in the denominator is the rst coecient 0 of the characteristic

polynomial (s), and therefore with Eq. (3.6.26)
cK =

0

b0 d0

(3.6.27)

For a controller with integral action the coecient a0 is zero and according to
Eq. (3.6.17) cK = 0. From Eqs. (3.6.26) and (3.6.19) to (3.6.22) it follows directly
that

3.6 Compensator design methods

161
+
cK = b+
0 d0 .

(3.6.28)

When the controller is inserted into the feedback path according to Figure 3.6.6b
the inherent closed-loop dynamics will not be changed compared with the conguration according to Figure 3.6.6a, because the denominator polynomial of the
transfer function, and therefore the characteristic equation of the closed loop,
are preserved. Indeed, the zeros of the controller transfer function do no longer
arise, but their poles as zeros in the closed-loop transfer function. Analogous
considerations for AK (s) lead to
AK (s) = A+ (s) D+ (s) ,

(3.6.29)

whereby the polynomial A+ (s) contains the poles of the controller and D+ (s) the
plant zeros in the left-half plane. The transfer function
GW (s) =

cK A (s) D (s)
A(s) C(s) + B(s) D(s)

(3.6.30)

is the same as for the case of a stable controller and a minimum-phase plant
according to Eq. (3.6.25).
The constant cK for a proportional controller is
cK =

0

a0 d0

(3.6.31)

For an integral controller in the feedback loop a feed-forward path is not realisable.

3.6.4.3

The synthesis equations

The system of equations described by Eq. (3.6.15c) can be rewritten in matrix notation. Thus the required controller parameters are combined into one parameter
vector. The matrix of the plant parameters applies for both the cases, controller
order z = n 1 and z = n.
For integral plants (c0 = 0) with controller order z = n 1 and normalised cn = 1
the system is:

162

3 Behaviour of linear continuous-time control systems

d0

d1

d2

.
..

dn1

..
.

0
d0

c1

d1
..
.

d0
.. ..
.
.

c2
..
.

c1
..
.

cn2 cn3
dn2

d1

d0

dn1 dn2 d1
dn1 d2
.
..
. ..
0

0
0
..

cn1 cn2
1

..

c1
c2

cn1 cn2
1

cn1
.. ..
.
.
..

dn1

0
0

b 1

1
0

b2 2 0

.
.
.
.

. . ..

bn2 n2 0

0
c1

n1
bn1 =
c

c2 a0 n

c3
a

1 n+1 .

. ..
..
..
.

. . .

n2

cn1 an3
2n3
cn1
an2
1
2n2
(3.6.32a)
0

and
an1 = u .

(3.6.32b)

For proportional plants or in the case of disturbances at the input of integral

plants, where the order of the controller is increased by one to z = n, with
Eqs. (3.6.17) and (3.6.15a) it follows that
a0 = cR b0
0
.
b0 =
d0 + cR c0

(3.6.33)
(3.6.34)

Here the system is:

1
b1
d0
c0
d1 + cR c1
0

b2 2
0

d1 d0
0
c1 c0
0
d2 + cR c2

d d d

b3
c2 c1 c0

2
1
0

. ..
..
..

..
.. . . . .
..
.. . . . .

.
.

.
.
. .
. .
.
.
.
. .

+
c
c
c
d
0
n1
R n1

c0

bn = n b0

cR

dn1dn2 d1 d0 cn1cn2 c2 c1

1
0

0 d d d
a

1
n+1

1 cn1cn2 c2

n1 n2
1

a
2

1 cn1 c3

..

.
.
..
.
.
..
. . ..
..
.. ..

..
.
.
.
.

0
0
cn1

c
n1
0
dn1
1
0
an1

2n1

(3.6.35a)

3.6 Compensator design methods

163

and
an = u .

(3.6.35b)

The (2n 1) (2n 1) matrices on the left side of Eqs. (3.6.32a) and (3.6.35a)
are equal for c = 0. This matrix is always regular and therefore the solution is
always unique.

3.6.4.4

Application of the method

Example 3.6.5
An integral plant has the transfer function
GP (s) = 0.25

1 + 5s
1 + 5s
=
.
s(1 + 0.25s)
4s + s2

As we do not consider a disturbance at the input of the plant, the controller

coecients can be calculated by Eqs. (3.6.32a) and (3.6.32b).
According to section 3.6.4.1 one obtains for this second-order plant the order
z = n 1 = 1 of the controller. The closed loop has therefore order u = z + n = 3.
The step response of this loop should comply with the binomial form given in
Table A.8.2, and tr,50 2.5 s should be met. This corresponds to a value of
0 1s1 . The associated characteristic polynomial is
(s) = (1 + s)3 = 1 + 3s + 3s2 + s3 ,
and Eqs. (3.6.32a) and (3.6.32b) provide the synthesis equation
    

b0
0
1 0 0
1
5 1 4 b1 = 3 0
0 5 1
3
4
a0
with
a1 = 1,
from which the controller coecients follow as
a0 =
b0 =

9
19

a1 =

1 ,

b1 =

2
19

The transfer function of the controller is

2
s
19 .
GC (s) =
9
+s
19
1

164

3 Behaviour of linear continuous-time control systems

It can be seen that this design leads to an unstable controller, which is in addition
non-minimum phase. The closed-loop transfer function is


2
1 s (1 + 5s)
19
,
GW1 (s) =
(1 + s)3
and it contains in the numerator polynomial, besides the zero of the plant, the
right-half-plane zero of the controller. According to the considerations in section 3.6.4.2 this zero cannot be compensated by a pre-lter for stability reasons.
Also this would not be necessary, as the plant zero is dominant and has a stronger
inuence on the closed-loop step response (see Figure 3.6.7). The denominator
polynomial BK (s) for the transfer function GK (s) of the pre-lter is determined
u ( t), y ( t)
1 .8
h

1 .6

W
1

( t)

1 .4
u 1( t)

1 .2
1 .0
0 .8
h

0 .6

( t)
h

( t)

u ( t)

0 .4
0 .2
0
-0 .2
0

4
8

1 2

1 6

[s ]

2 0
t

Figure 3.6.7. Step responses of the controlled variable y(t) for the case: (a) without
pre-lter hW1 (t), (b) with pre-lter hW (t), and the associated manipulated variable u1 (t)
and u(t), and the step response hK (t) of the given closed-loop transfer function KW (s).

as
BK (s) = D+ (s) B + (s) = 1 + 5s ,
and the numerator is
cK = 1 .
The closed-loop transfer function including the pre-lter is thus
2
s
19
.
GW (s) =
(1 + s)3
1

3.6 Compensator design methods

165

This in fact still contains in the numerator polynomial the controller zero, but as
can be seen from Figure 3.6.7, the corresponding step response hW (t) does not
show a large deviation from the step response hK (t) of the given transfer function
KW (s) =

1
.
(1 + s)3

Conspicuous is the fact that the step response hW1 (t) of the closed loop without
pre-lter has a large overshoot and does not show any similarity with the other
step responses, though all three other systems have the same inherent behaviour.
Here, the dominant behaviour of the plant zero sN = 0.2 has a noticeable eect.


Example 3.6.6
Given the third-order plant transfer function
GP (s) =

1
(1 +

s)2 (1

+ 5s)

0.2
d0
.
=
2
3
0.2 + 1.4s + 2.2s + s
C(s)

The step response hP (t) of this plant is shown in Figure 3.6.8. A controller should
be designed such that the step response hW (t) of the closed loop has a desired
standard form chosen from Table A.8.1. In the current case of a PTn plant the
order m of the controller must be chosen equal to the order n of the plant, that
is
z=n.
Thus one obtains a sixth-order closed-loop transfer function GW (s). The desired
transfer function according to Eq. (3.6.4), which is the basis for Table A.8.1, has
then exactly the total order z + n = 6, if
k=4
is chosen. For this case the step response with
=4
(see Table A.8.1) will be desired. The last unspecied parameter of the transfer
function KW (s) is the relative frequency 0 . All step responses in Table A.8.1 are
normalised by this value such that the time scale can still be chosen. Consequently,
by a proper choice of 0 the normalised step response from Table A.8.1 can be
scaled to the desired time scale. E.g., for a value of 0 = 0.4 s1 a rise time
tr,50 1.6 s would follow, for 0 = 2 s1 this would be tr,50 0.32 s. If a large
value is taken for 0 to obtain a small rise time, this would result in controller
coecients of very dierent order such that this controller would not be realisable
from a numerical point of view. Therefore here
0 = 0.4 s1

166

3 Behaviour of linear continuous-time control systems

is a good choice. For the determined values of k, and 0 one obtains for the
desired closed-loop transfer function from Eq. (3.6.4)
KW (s) =

43.52
0
.
=
43.52 + 99.84s + 106.88s2 + 72.96s3 + 33.12s4 + 8.9s5 + s6
(s)

The Eqs. (3.6.33) and (3.6.34) deliver at rst the absolute coecients of the
controller transfer function GC (s)


b0 =

d0 1 + cC

c0
d0

 = 217.6

and
a0 = cC b0 = 217.6

1
1 + cC

cC
.
1 + cC

The remaining coecients are obtained from Eq. (3.6.35a) by solving

b1
0
0.2 0 0 0.,2 0
1.4cC
99.84

2.2cC 0
0 0.2 0 1.4 0.2
b2

106.88

1
c 0.2
0 0 0.2 2.2 1.4 72.96
b3 =
217.6
.
C

1 + cC
0 1.4
0 0 0 1 2.2 a1 33.12
0 0 0 0 1
8.8
0
2.2
a
2

The solution is
cC
1 + cC
cC
= 407.4 2393.6
1 + cC
cC
= 128.4 1088
1 + cC
= 17.2
= 6.6 .

b1 = 482 1523.2
b2
b3
a1
a2

If the reciprocal gain is chosen according to Eq. (3.6.17) as

cC = 0.2

for

KC = 5 ,

then the controller coecients have the same order of magnitude and the controller transfer function is
181.33 + 228.13s + 8.467s2 52.93s3
=
36.26 + 17.2s + 6.6s2 + s3
52.93(s 2.468) (s + 1.154 + j0.236) (s + 1.154 j0.236)
.
=
(s + 4.573) (s + 1.013 + j2.627) (s + 1.013 j2.627)

GC (s) =

3.6 Compensator design methods

167

From Eq. (3.6.27) one obtains for the coecient cK of the pre-lter
cK =

0

b0 d0

43.52
= 0.33 .
2.468 52.93 1

The transfer function of the pre-lter is

GK =

0.33
1.67
.
=
0.2 (s + 1.154 + j0.236) (s + 1.154 j0.236)
1.39 + 2.31s + s2

The step response of the controlled variable hW1 (t) of the closed loop is shown
y ( t)
1 .5
h
W
2

(a )

( t)
h
W

1 .0

0 .5

( t)

h P ( t)

0 .0
5

u ( t)
1 0

1 0

1 5

[s ]

2 0

2 0

(b )
u 1( t)

u 2( t)

1 0

1 5

[s ]

Figure 3.6.8. (a) Step responses of the controlled variable y(t) for the plant in the
uncontrolled case hP (t) and in the controlled case for the designed compensator hW1 (t)
and for the optimal PI controller hW2 (t); (b) Step responses of the associated manipulated
variables u1 (t) and u2 (t)

in Figure 3.6.8. For comparison this gure also shows the corresponding step
response hW2 (t) of the closed loop using a PI controller, which is optimal in
the sense of the performance index JISE according to section 3.3.3. Both, the
maximum overshoot and the rise time of this control system with a PI controller
are clearly worse than for the controller designed here. From the behaviour of the
manipulated variables u1 (t) and u2 (t), respectively, it can be seen that in general

168

3 Behaviour of linear continuous-time control systems

a smaller rise time must be bought by a larger amplitude of the controller output.
Because of the always existing limitation on the value of the manipulated variable,
too demanding specications for the transfer function KW (s) cannot be realised.

3.6.5

Compensator design for reference and disturbances

3.6.5.1

Structure of the closed loop

In the previous sections dierent design methods have been described for the
design of controllers for a given plant transfer function GP (s) and for a desired
closed-loop transfer function GW (s) = KW (s). If in addition a disturbance is
considered, which always exists in real control systems, and if one wants to exert
an inuence on the disturbance behaviour, the pre-lter GK (s) is often necessary.
Figure 3.6.9 shows this structure. Furthermore, dierent entry points for the
disturbances must be taken into account, because disturbances at the plant input
or output act dierently on the controlled variable.
p la n t
Z

K (s )
L (s )
d is t u r b a n c e G P Z ( s )

M ( s )
N ( s )
+

p r e -filt e r G K ( s )

B (s )
A ( s )

D ( s )
C ( s )

c o n t r o lle r G C ( s )

p la n t G P ( s )

Figure 3.6.9. Control system designed for reference and disturbance behaviour

For the design of the controller and pre-lter the following transfer functions are
introduced:
M (s) m0 + m1 s + . . . + mx sx
=
, yx;
(3.6.36)
GK (s) =
N (s)
n0 + n1 s + . . . + ny sy
b0 + b1 s + . . . + bw sw
B(s)
=
, zw ;
(3.6.37)
GC (s) =
A(s)
a0 + a1 s + . . . + az sz
d0 + d1 s + . . . + dm sm
D(s)
=
, nm.
(3.6.38)
GP (s) =
C(s)
c0 + c1 s + . . . + cn sn
If for the control system the desired closed-loop transfer function for the command
input

3.6 Compensator design methods

KW (s) =

169

0 + 1 s + . . . + v sv
(s)
=
, uv
(s)
1 + 1 s + . . . + u su

(3.6.39)

and the desired disturbance transfer function

0 + 1 s + . . . + q sq
(s)
=
, pq
(s)
1 + 1 s + . . . + p s p

KZ (s) =

(3.6.40)

are given, then on step changes in the command variable and in the disturbance
a steady-state error must not occur. Therefore, the following must be valid:
(s)
= 1,
s0 (s)

KW (0) = lim
and

(s)
= 0,
s0 (s)

KZ (0) = lim

i.e. 0 = 1

(3.6.41)

i.e. 0 = 0 .

(3.6.42)

For the synthesis of the closed loop it is additionally required to choose the
numerator and denominator polynomials of KW (s) and KZ (s) of as minimum an
order as possible and according to given criteria.

3.6.5.2

The design procedure

The closed-loop transfer function is given by

GW (s) =

GV (s) GC (s) GP (s) !

Y (s)
=
= KW (s) .
W (s)
1 + GC (s) GP (s)

(3.6.43)

and the disturbance transfer function can be obtained from

GZ (s) =

GPZ (s)
Y (s)
!
=
= KZ (s) .
Z(s)
1 + GC (s) GP (s)

(3.6.44)

Then from Eq. (3.6.44) it follows that the controller transfer function is
GC (s) =

GPZ (s) KZ (s)

.
GP (s) KZ (s)

(3.6.45)

For disturbances at the input of the plant, GPZ (s) = GP (s) is valid and therefore
one obtains with Eq. (3.6.45) the controller transfer function as
GC (s) =

D(s) (s) (s) C(s)

.
D(s) (s)

(3.6.46)

Disturbances at the plant output are taken into account by GPZ = 1 in Eq. (3.6.45).
In this case the controller transfer function is

170

3 Behaviour of linear continuous-time control systems

GC (s) =

1
(s) (s)
C(s) [(s) (s)]
=
.
D(s) (s)
GP (s)
(s)

(3.6.47)

Design for disturbances at the plant input

Contemplating Eq. (3.6.46), the degree w of the numerator polynomial is given
either by m + p or by q + n. The degree z of the denominator polynomial is xed
by m + q. Because of the realisability condition z w of the controller
qp

and

mn

should be valid. But, as this realisability condition cannot hold because of Eqs. (3.6.38)
and (3.6.40), the higher-order terms in s of the numerator polynomial of Eq. (3.6.46)
D(s) (s) (s) C(s)
must disappear. This is only possible with a choice of
m+p = q+n

or

nm=pq .

(3.6.48)

Eq. (3.6.48) tells us that the pole excess both of the plant and the disturbance
transfer function must be the same. Furthermore, it can be seen that in the
numerator polynomial of Eq. (3.6.46) exactly
(m + p) (m + q) = p q = n m
higher-order terms in s must disappear, i.e. they must be compensated (cancelled). From this it follows that the degree of the plant numerator and denominator polynomials is
w = z = (m + p) (p q) = m + q .
As n m higher-order terms in the numerator polynomial of Eq. (3.6.46) must
be compensated, and the degree of the polynomial (s) must be minimal, the
degree of this polynomial must be chosen as
q =nm .

(3.6.49)

Therefore it follows with Eq. (3.6.48) that

p = q + n m = 2(n m) .

(3.6.50)

The previous determined degree of the numerator and denominator polynomials

of GC (s) can be nally obtained using Eq. (3.6.49) as
w = z = m+q = n .

(3.6.51)

In the following the design will be demonstrated by some simple examples.

3.6 Compensator design methods

171

Example 3.6.7
The plant transfer function
GP (s) =

D(s)
d0 + d1 s
=
C(s)
c0 + c1 s + c2 s2

with n = 2 and m = 1 is given. According to the Eqs. (3.6.49) and (3.6.50) both
polynomials (s) and (s) have the degrees
p = 2 (2 1) = 2
and
q =21 = 1 .
For the disturbance transfer function one gets
KZ (s) =

1 (s)
(s)
.
=
2
1 + 1 s + 2 s
(s)

If the polynomials (s) and (s) are now inserted into Eq. (3.6.46), the controller
transfer function
GC (s) =

d0 + (d0 1 + d1 1 c0 )s + (d0 2 + d1 1 1 c1 )s2 + (d1 2 1 c2 )s3

d0 1 s + d1 1 s2

follows. With 1 = 2 d1 /c2 the highest-order term in the numerator polynomial

of GC (s) disappears, and one obtains the realisable controller transfer function
of
b0 + b1 s + b2 s2
GC (s) =
a1 s + a2 s2
and the disturbance transfer function of
2 d1
s
c2
.
KZ (s) =
1 + 1 s + 2 s 2
This example clearly shows that the disturbance behaviour cannot be freely designed, but the inherent dynamics inuences the disturbance response, i.e. the
poles of the disturbance transfer function. In this case the coecients of the numerator polynomial (s) depend on those of the denominator polynomial (s).


Example 3.6.8
Similarly to the previous example a second-order plant is again considred, but
with the pole excess n m = 2. The transfer function is

172

3 Behaviour of linear continuous-time control systems

GP (s) =

d0
D(s)
=
.
C(s)
c0 + c1 s + c2 s2

The degree of the numerator and denominator polynomials of the disturbance

transfer function can be obtained from Eqs. (3.6.49) and (3.6.50) as
q =nm=2

and

p = 2(n m) = 4 .

Therefore, the disturbance transfer function has the form

KZ (s) =

1 s + 2 s2
.
1 + 1 s + 2 s 2 + 3 s 3 + 4 s 4

According to Eq. (3.6.51) the degree of the numerator and denominator polynomials of the controller transfer function will be
w =z =m+q =n=2 .
On the other hand one obtains formally from Eq. (3.6.46) the controller transfer
function as
GC (s) =

b0 + b1 s + b2 s2 + (d0 3 1 c2 2 c1 ) s3 + (d0 4 2 c2 ) s4
.
d0 1 s + d0 2 s2

However, for realisability reasons n m = 2 higher-order terms in s in the numerator polynomial must disappear. From this it follows for the disturbance transfer
function that
d0 4
2 =
c2
and
d0 3 c2 d0 4 c1
,.
1 =
c22


Design for disturbances at the plant output
Starting from Eq. (3.6.47) similar results as in the previous section will be obtained. The degree of the controller numerator polynomial in this case is given
either by n + p or by q + n. For realisability reasons a compensation of the supernumerary terms in the numerator of Eq. (3.6.47) is necessary. This is possible
if
p=q
(3.6.52)
is valid. That p must equal q is also understandable from the fact that the disturbance acts directly at the output of the plant and therefore the disturbance
transfer function has always a feed-through term in the transfer function. The
(n + p) (m + q) = n m higher-order terms in the numerator polynomial of
Eq. (3.6.47) must be compensated. Therefore, the degree of the numerator and
denominator polynomials of GC (s) is given by

3.6 Compensator design methods

173

w = z = (q + n) (n m) = m + q .
As in the numerator polynomial of Eq. (3.6.47) nm terms must be compensated
and the degree of the the polynomial (s) must be minimal, its degree is chosen
as
q =nm .
(3.6.53)
From this it follows with Eq. (3.6.52) that
p=nm .

(3.6.54)

The previous given degree of the polynomials of the controller transfer function
GC (s) will now be with Eq. (3.6.53)
w =z =m+q =n .

(3.6.55)

It can be seen from Eq. (3.6.47) that the poles of the plant transfer function are
cancelled by the zeros of the controller transfer function. If the plant parameters
are varying slightly, in the case of an unstable plant the closed loop will also be
unstable. For unstable plants, (s) will therefore chosen such that it contains the
unstable terms C (s) of degree n , that is
(s) = C (s) (s) .
Consequently the controller transfer function GC (s) has the form
GC (s) =

C + (s) [(s) C (s) (s)]

,
D(s) (s)

whereby C + (s) is that part of the polynomial C(s) which has zeros in the left-half
s plane.
The polynomial (s) of degree must be chosen such that the polynomial
(s) C (s) (s) itself has only zeros in the left-half s plane. The order of the
disturbance transfer function is given by p = n + , as for realisability reasons
in the numerator polynomial those n n + p m higher-order terms in s
must be compensated. The controller transfer function is then of order m + .
A detailed example of the design for an unstable plant can be found in section
3.6.5.4. The procedure of the design will be demonstrated now in the following
using some simple examples with stable plants.
Example 3.6.9
For the plant with the transfer function
GP (s) =

d0 + d1 s
c0 + c1 s + c2 s2

174

3 Behaviour of linear continuous-time control systems

and the pole excess n m = 1, the degree will be using Eqs. (3.6.53) and (3.6.54)
q = p = 1. For the disturbance transfer function KZ (s) follows according to
Eqs. (3.6.40) and (3.6.42)
1 s
KZ (s) =
.
1 + 1 s
Inserted in Eq. (3.6.47) the controller transfer function arises formally as
GC (s) =

c0 + [c1 + c0 (1 1 )] s + [c2 + c1 (1 1 )] s2 + c2 (1 1 )s3

.
d0 1 s + d1 1 s2

With the realisability condition 1 = 1 one obtains the realisable controller

transfer function as
c0 + c1 s + c2 s2
GC (s) =
d0 1 s + d1 1 s2
and for the disturbance transfer function
1 s
.
KZ (s) =
1 + 1 s


Example 3.6.10
For the plant transfer function
d0
c0 + c1 s + c2 s2
is n m = 2 and therefore p = q = 2. From this it follows that the disturbance
transfer function is
1 s + 2 s2
.
KZ (s) =
1 + 1 s + 2 s 2
The realisability condition of the controller transfer function
GP (s) =

GC (s) =

b0 + b1 s + b2 s2 + [c1 (2 2 ) + c2 (1 1 )] s3 + (2 2 ) s4
d0 1 s + d0 2 s2

is
2 = 2
and
1 = 1 .
The realisable controller transfer function is
b0 + b1 s + b2 s2
.
GC (s) =
d0 1 s + d0 2 s2


Recapitulating, we can ascertain that the order for the disturbance transfer function depends on the entry point of the disturbance and on the pole excess (n m)
of the plant transfer function. If the inherent dynamics of KZ (s) is chosen, the
zeros of KZ (s) are given by the realisability conditions for the controller.

3.6 Compensator design methods

3.6.5.3

175

Design of the pre-filter

The starting point for the synthesis of the pre-lter transfer function GK (s) are
the Eqs. (3.6.43) and (3.6.44). From these equations it follows directly that
GK (s) =

GPZ (s) KW (s)

.
GC (s) GP (s) KZ (s)

(3.6.56)

If the disturbance acts at the input of the plant, GPZ (s) = GP (s) is valid and
therefore the pre-lter transfer function is
GK (s) =

KW (s)
.
GC (s) KZ (s)

(3.6.57)

In the case of disturbances at the plant output the pre-lter will be, because
GPZ (s) = 1, given by
GK (s) =

KW (s)
.
GC (s) GP (s) KZ (s)

(3.6.58)

Design of the pre-lter for disturbances at the plant input

With Eq. (3.6.57) one obtains the pre-lter transfer function as
GK (s) = KW (s)

1
(s) A(s) (s)
1
=
.
GC (s) KZ (s)
(s) B(s) (s)

Taking with Eqs. (3.6.37) and (3.6.46)

A(s) = D(s) (s)
into consideration, the pre-lter transfer function can also be written as
GK (s) =

M (s)
(s) D(s) (s)
=
.
(s) B(s)
N (s)

(3.6.59)

For the determination of the degrees v and u of the polynomials (s) and (s),
respectively, the realisability condition for the pre-lter transfer function GK (s)
will be investigated. Accordingly, one obtains with the results from section 3.6.5.2
with p = 2(n m) and w = n the condition u + n v + m + 2(n m) or
unm+v .

(3.6.60)

According to Eq. (3.6.59) the degree of the numerator polynomial M (s) and of
the denominator polynomial N (s) is given by
x = m + v + 2(n m) = 2n m + v

(3.6.61)

176

3 Behaviour of linear continuous-time control systems

and
y =n+u ,

(3.6.62)

as far as M (s) and N (s) have no common divisor. For example, if the degree of
the numerator of the closed-loop transfer function for command input is given by
v = 0, the following is valid:
u nm .
For v > 0 the polynomial (s) can be used to cancel the unavoidable zeros of the
controller transfer function in the right-half s plane. Otherwise this would lead
to an unstable pre-lter. But the non-minimum-phase behaviour caused by the
controller still remains.
Example 3.6.11
The transfer functions GP (s), KZ (s) and GC (s) are taken from example 3.6.7.
According to Eq. (3.6.59) the pre-lter can be obtained as
GK (s) =

(s) D(s)
(s) .
(s) B(s)

With v = 0 and u = 1 the closed-loop transfer function is chosen as

KW (s) =

1
(s)
=
.
(s)
1 + 1 s

Herewith, the degree of the pre-lter numerator and denominator polynomial

is given by x = y = 3. After substitution of the results from example 3.6.7 in
Eq. (3.6.59) the pre-lter transfer function in general form is
GK (s) =

m0 + m1 s + m2 s 2 + m3 s 3
.
n0 + n1 s + n2 s2 + n3 s3

Example 3.6.12
If we have as in example 3.6.8 a pole excess of the plant transfer function of
n m = 2, and if the results for KZ (s) and GC (s) are considered, then for v = 0
the closed-loop transfer function can be chosen as
KW (s) =

1
.
1 + 1 s + 2 s2

From Eqs. (3.6.59), (3.6.61) and (3.6.62) the pre-lter transfer function
GK (s) =
follows.

m0 + m1 s + m2 s 2 + m3 s 3 + m4 s 4
n0 + n1 s + n2 s2 + n3 s3 + n4 s4

3.6 Compensator design methods

177

Design of the pre-lter for disturbances at the plant output

Similarly as in section 3.6.5.1 one obtains here from Eq. (3.6.58)
GK (s) =

A(s) C(s) (s) (s)

M (s)
=
.
B(s) D(s) (s) (s)
N (s)

If the Eqs. (3.6.37) and (3.6.46) are considered, one obtains with
A(s) = D(s) (s)
from the above equation
GK (s) =

C(s) (s) (s)

.
B(s) (s)

(3.6.63)

The rst task is again to specify the degrees u and v of the polynomials (s) and
(s) such that the pre-lter becomes realisable. For the polynomials M (s) and
N (s) previous results from section 3.6.5.2 can be considered and from this follows
n+u n+nm+v
and
unm+v .

(3.6.64)

From Eq. (3.6.63) one obtains nally for the degree of the numerator and denominator polynomials M (s) and N (s) of GK (s) that
x = 2n m + v

(3.6.65a)

y =n+u

(3.6.65b)

and

as far as M (s) and N (s) have no common divisor. If during the design of N (s)
unstable poles arise, then one can proceed as in the case of disturbances at the
plant input. The degree v of the numerator polynomial of KW (s) will be increased
by the number of compensated terms.
For the case of unstable plants with C(s) = C + (s) C (s) and the unstable part
C (s) and with the denominator polynomial (s) = C (s) (s) of the disturbance transfer function KZ (s), the pre-lter transfer function can be obtained
from the previous section by using the controller transfer function
GC (s) =

B(s)
C + (s) [(s) C (s) (s)]
=
D(s) (s)
A(s)

as
GK (s) =

C + (s) (s) a(s)

.
B(s) (s)

(3.6.66)

178

3 Behaviour of linear continuous-time control systems

Example 3.6.13
Again as in example 3.6.9 the plant transfer function
GP (s) =

D(s)
d0 + d1 s
=
C(s)
c0 + c1 s + c2 s2

is given. Considering earlier results

KZ (s) =
and
GC (s) =

1 s
(s)
=
(s)
1 + 1 s

b0 + b1 s + b2 s2
B(s)
=
,
A(s)
a1 s + a2 s2

one obtains for v = 0 the closed-loop transfer function for command input using
Eq. (3.6.64) as
1
(s)
=
.
KW (s) =
(s)
1 + 1 s
Substituted in Eq. (3.6.63) the pre-lter transfer function follows as
GK (s) =

m0 + m1 s + m2 s 2 + m3 s 3
.
n0 + n1 s + n2 s2 + n3 s3

Example 3.6.14
Similarly as in example 3.6.10 the pole excess of the plant transfer function is
given as n m = 2. With
KZ (s) =

1 s + 2 s2
1 + 1 s + 2 s 2

GC (s) =

b0 + b1 s + b2 s2
a1 s + a2 s2

and

follows with the choice of v = 0 and applying Eq. (3.6.64) the pre-lter transfer
function is
m0 + m1 s + m2 s 2 + m3 s 3 + m4 s 4
.
GV (s) =
n0 + n1 s + n2 s2 + n3 s3 + n4 s4

3.6 Compensator design methods

3.6.5.4

179

Application of the design method

The design procedure for compensators described in the previous sections will
be applied in this concluding section to an example already introduced in section 3.6.3. It is the unstable plant with the transfer function
GP (s) =

1
D(s)
=
,
C(s)
1 sT

T = 1s

(see example 3.6.4).

Disturbance at the plant input
If a disturbance at the plant input is assumed, the degree of the numerator and
denominator polynomial of the free selectable disturbance transfer function according to Eqs. (3.6.49) and (3.6.50) is given by q = 1 and p = 2. The disturbance
transfer function will be obtained as
KZ (s) =

1 s
(s)
=
(s)
1 + 1 s + 2 s 2

and the the controller transfer function formally according to Eq. (3.6.46) as
GC (s) =

1 + (1 1 ) s + (2 + 1 T ) s2
B(s)
=
.
A(s)
1 s

For the realisability condition of the controller it follows that

1 =

2
.
T

The nal realisable controller transfer function is then


2 
1 + 1 +
s
T
.
GC (s) =
2
s
T
Furthermore, the above equation for 1 provides for the disturbance transfer
function
2
s
T
.
KZ (s) =
1 + 1 s + 2 s 2
The procedure for the design of the pre-lter for disturbances at the plant input
starts with the given order of the transfer function for command input according
to Eq. (3.6.60). For the command behaviour u = 2 and v = 0 is chosen. According
to that the transfer function for command input is
KW (s) =

1
(s)
=
.
(s)
1 + 1 s + 2 s2

180

3 Behaviour of linear continuous-time control systems

The synthesis of the controller is required to be performed such that the behaviour for a command input will be characterised by approximately 10% maximum overshoot and t3% 3 s settling time. In this case of a second-order system
these demands are met by a natural frequency 0 = 2 s1 and by a damping ratio
= 0.6 according to Figures 3.5.3 and 3.5.5. From this the transfer function for
command input follows as
KW (s) =

1
.
1 + 0.6s + 0.25s2

If for the inherent behaviour of the closed loop the same coecients are taken for
the disturbance case, then
(s) = (s) ,
and therefore the disturbance transfer function is
KZ (s) =

0.25s
1 + 0.6s + 0.25s2

and the controller transfer function is

GC (s) =

1 + 0.85s
.
0.25s

The pre-lter design, which is given by

GK (s) =

D(s) (s) (s)

,
B(s) (s)

is simplied by (s) = (s) to

GK (s) =

1
D(s) (s)
=
.
B(s)
1 + 0.85s

Step responses for the disturbance and command inputs are shown in Figure 3.6.10.
Disturbance at the plant output
If the closed loop is designed for a disturbance at the plant output, for the unstable
plant the controller transfer function
GC (s) =

C + (s) [(s) C (s) (s)]

D(s) (s)

from section 3.6.5.2 is signicant. With

C + (s) = 1 ,
and with the approach

C (s) = 1 sT

3.6 Compensator design methods

181

h ( t)
0
h

( t)

Z Y

(a )

- 0 .5
h

( t)

Z U

- 1 .0
- 1 .5
0
h ( t)
1 .0

2
h
W

[s ]

( t)
Y

0 .5
h

(b )

( t)

0
- 0 .5
0
h ( t)
1 .0
h

Z W

[s ]

8
t

( t)

0 .5
0
- 0 .5
h

- 1 .0

Z W

(c )

( t)

- 1 .5
0

[s ]

8
t

Figure 3.6.10. Step responses for the design for a step disturbance z = (t) at the
plant input:
(a) controlled variable hZY (t) and manipulated variable hZU (t) for z = (t)
(b) controlled variable hWY (t) and manipulated variable hWU (t) for w = (t)
(c) controlled variable hZWY (t) and manipulated variable hZWU (t) for the simultaneous
application of z = (t) and w = (t)

182

3 Behaviour of linear continuous-time control systems

(s) = a + bs

it follows as in the previous case for

(s) = (s)
that

1 a + (1 b + aT ) s + (2 + bT ) s2
.
a + bs
The realisability condition for the controller is thus
GC (s) =

b=

2
,
T

and if for simplicity a is set equal to zero, then the controller transfer function is


2
1 + 1 +
s
1 + 0.85s
T
=
,
GC (s) =
2
0.25s
s
T
as before. The corresponding disturbance transfer function is
2
s + 2 s2
(s) C (s)
T
=
.
KZ (s) =
(s)
1 + 1 s + 2 s2
The transfer function of the pre-lter is calculated according to Eq. (3.6.58), that
is
A(s) C(s) (s) (s)
.
GK (s) =
B(s) D(s) (s) (s)
With the assumptions
1
(s)
=
,
(s)
1 + 1 s + 2 s2
(s) = (s) ,

KW (s) =

(s) = C (s) (s) ,

A(s) = D(s) (s)
the pre-lter is simplied to
GK (s) =

1
=
B(s)

1 + 1 +

2
T

=
s

1
,
1 + 0.85s

again as before. The step responses for this case are shown in Figure 3.6.11.
Problem 3.6.1
Compensator design - two questions

3.6 Compensator design methods

183

h ( t)
3 .0
h

2 .0

Z U

(a )

( t)

1 .0
h
0

( t)

Z Y

- 1 .0
0
h ( t)
1 .0

2
h
W

[s ]

8
t

( t)

0 .5
h

(b )

( t)

0
- 0 .5
0

[s ]

8
t

h ( t)
3 .0
2 .0

(c )
h

Z W

( t)

1 .0
0
h

- 1 .0
0

Z W

( t)
4

[s ]
8

Figure 3.6.11. Step responses for the design for a step disturbance z = (t) at the
plant output:
(a) controlled variable hZY (t) and manipulated variable hZU (t) for z = (t)
(b) controlled variable hWY (t) and manipulated variable hWU (t) for w = (t)
(c) controlled variable hZWY (t) and manipulated variable hZWU (t) for the simultaneous
application of z = (t) and w = (t)

184

3.7

3 Behaviour of linear continuous-time control systems

Improving the control behaviour by more complex loop structures

3.7.1

Problem

The control systems discussed hitherto are single-loop control systems. These
control systems may not meet extra high requirements even in an optimal design
case for higher-order plants and plants with dead time concerning the maximum
overshoot Mp , rise time tr and settling time t . This is apparent especially in
the case of large disturbances and when large delays occur between the actuator and measurement device. An improvement of the control behaviour can be
obtained if the signal paths between actuator and disturbance are shortened, or
if disturbances are already compensated by a separate pre-controller before they
enter the plant. In this case the disturbance must be measurable and controllable
via an actuator. Shortening the signal paths within a control system leads to a
structural expansion of the basic control loop and therefore to more complex loop
structures. In the following, the most important basic structures of these complex
control systems will be discussed.

3.7.2

Disturbance feed-forward control

The disturbance feed-forward control corresponds to the basic control loop, which
is superimposed by an open-loop control scheme with the goal to compensate the
disturbance by a control element GFFi (s)(i = 1,2, . . .) as far as possible before
it fully acts on the controlled variable y. This conguration is of course only
realisable, if the disturbance is measurable at the plant input. With regard to
a feed-forward conguration, the following two dierent cases are distinguished,
whereby the following transfer functions, see Figure 3.7.1, will be used:
B(s)
;
A(s)
D(s)
;
GP (s) =
C(s)

GC (s) =

3.7.2.1

BFFi (s)
AFFi (s)
DZ (s)
GPZ (s) =
.
CZ (s)

GFFi (s)=

(i = 1,2)

Disturbance feed-forward on the controller

According to Figure 3.7.1 the disturbance z
will feed via the transfer function
GFF1 (s) to the controller, which will compensate the inuence of the disturbance.
From this diagram the controlled variable directly follows as

3.7 Improving the control behaviour by more complex loop structures

185

Z
G

F F 1

(s )
G

P Z

( s )

fe e d -fo r w a r d e le m e n t
W

_
+
_

G C (s )

G P (s )

c o n t r o lle r

+
Y

p la n t

Figure 3.7.1. Block diagram of the feed-forward on the controller

Y (s) = [W (s) Y (s) Z
(s) GFF1 (s)] GC (s) GP (s) + Z
(s) GPZ (s) .

(3.7.1)

With some manipulations one obtains from this

Y =

GC GP
GPZ GFF1 GC GP

Z +
W ,
1 + GC GP
1 + GC GP

(3.7.2a)

which gives
Y =

BD
AFF1 A C DZ BFF1 B D CZ

Z +
W ,
AFF1 CZ (A C + B D)
AC + B D

(3.7.2b)

where for brevity the argument s is omitted. From the transfer functions of
Eq. (3.7.2b) one can see that the characteristic equation is
AFF1 CZ (A C + B D) = 0

(3.7.3a)

with regard to disturbance behaviour and

AC + B D = 0

(3.7.3b)

with regard to the reference behaviour. The disturbance will be fully compensated
if
(3.7.4)
GPZ = GFF1 GC GP ,
from which the required transfer function for the feed-forward element is
GFF1 =

GPZ
A C DZ
=
.
GC GP
B D CZ

(3.7.5)

This approach can only be realised by a controller if the pole excess of GP is

not larger than that of GPZ . Otherwise a total compensation is not possible.
Moreover, the polynomial B D CZ must be Hurwitzian.

186

3 Behaviour of linear continuous-time control systems

For the frequent case that the disturbance and control behaviour are equal,
i.e. the case of GPZ = GP , the transfer function of the feed-forward elements
is
1
A
GFF1 =
.
(3.7.6)
=
GC
B
As the total compensation of a disturbance in a plant with P behaviour is only
possible by a controller with I behaviour, the transfer function of the feed-forward
element, according to Eq. (3.7.6), should thus show ideal D behaviour. If there
is a PI controller in the loop, the feed-forward element must be designed as a
DT1 element.
Often the feed-forward element cannot be realised as ideally designed according
to Eqs. (3.7.5) or (3.7.6), because GC , besides pure I behaviour, normally contains
delay elements. Also in these cases a DT1 element is recommended.

3.7.2.2

Disturbance feed-forward on the manipulated variable

The conguration with feed-forward on the manipulated variable or on the actuator, respectively, is shown in Figure 3.7.2. From this for the controlled variable
it follows that
Y = [(W Y ) GC Z
GFF2 ] GP + Z
GPZ
and after rearranging
Y =
or
Y =

GC GP
GPZ GFF2 GP

Z +
W
1 + GC GP
1 + GC GP

BD
A(AFF2 C DZ BFF2 D CZ )

Z +
W .
AFF2 CZ (A C + B D)
AC + B D

(3.7.7a)

(3.7.7b)

The characteristic equations are the same as in the previous case with feedforward to the controller. For the ideal compensation of the disturbances it follows
from Eq. (3.7.7) that
(3.7.8)
GPZ = GFF2 GP ,
from which the transfer function of the feed-forward element follows is
GFF2 =

GPZ
C DZ
=
.
GP
D CZ

(3.7.9)

For the special case of GPZ = GP , where the disturbance z acts directly at the
plant input, the compensation is performed by GFF2 = 1 directly at the plant
input..
Similarly as in the case of Eq. (3.7.5) the realisation of the feed-forward element
according to Eq. (3.7.9) is not possible if

3.7 Improving the control behaviour by more complex loop structures

Z

fe e d -fo r w a r d e le m e n t
G

( s )
_

E
_

F F 2

G C ( s )
+

187

P Z

( s )

+
+

G P ( s )

c o n t r o lle r

p la n t

Figure 3.7.2. Block diagram of the feed-forward on the manipulated variable

degree DZ + degree C > degree CZ + degree D

(3.7.10)

with GPZ = DZ /CZ and GP = D/C is valid, as GFF2 must be realised by PD

elements. Also in the case of a non-minimum phase behaviour of GP or of instability of GPZ the Eq. (3.7.9) cannot be realised, as the required feed-forward
element is unstable. In those cases in which a dynamical compensation according
to Eq. (3.7.9) is not possible one must be content with a static compensation
using a P element
KPZ
,
(3.7.11)
GFF2 =
KP
where KPZ and KP are the gains of the transfer functions GPZ and GP .
Figure 3.7.3 shows disturbance feed-forward congurations on (a) the controller
s t e a m
flo w

m  = z

(a )
F F

(b )
w

G C

S H

G C

S H

m  = z
J = y

J = y

s t e a m
flo w

C
v a lv e

c o o lin g
w a t e r

F F

u
M

M
C
v a lv e

c o o lin g
w a t e r

Figure 3.7.3. Examples of disturbance feed-forward congurations (a) on the controller

and (b) on the actuator of a steam superheater temperature control system

and (b) the manipulated variable, for the case of a temperature control system

188

3 Behaviour of linear continuous-time control systems

of a steam superheater (SH) in a power station. The steam temperature at

the superheater outlet is the controlled variable. The manipulated variable is the
cooling water ow in the spray-water cooler (C). Fluctuations of the steam ow
m
have an inuence on the steam temperature and are treated as disturbances.
The steam ow (disturbance z) is measured and fed via GFF1 or GFF2 to the
controller or to the manipulated variable, respectively.

3.7.3

Control systems with an auxiliary manipulated variable

For plants with a distinctive delayed behaviour, besides the actual controlled
variable y, a secondary variable can often be measured and used as an auxiliary
variable yA . The auxiliary control loop consists, as shown in Figure 3.7.4, of
the rst part of the plant with the transfer function GP1 (s) and the auxiliary
controller with the transfer function GCA (s). The controlled variable then follows
Z

W
+

E
G

C A

(s )

(s )

+
_

P Z

(s )

(s )
1

p la n t

+
G

(s )

Figure 3.7.4. Block diagram of a control system with auxiliary manipulated variable
yA

directly as shown in Figure 3.7.4. Now




Y


GCA GP1 + Z GPZ GP2
Y =
(W Y ) GC
GP2

(3.7.12)

or on rearranging
Y =

GC GP1 GP2
GPZ GP2
Z
+
W
1 + (GC GP2 + GCA ) GP1
1 + (GC GP2 + GCA ) GP1

(3.7.13a)

giwing
Y =

A C1 D2 DZ
Z

CZ [A C2 (C1 AA + D1 BA ) + D1 D2 B AA ]
B D1 D2 AA
+
W
A C2 (C1 AA + D1 BA ) + D1 D2 B AA

(3.7.13b)

3.7 Improving the control behaviour by more complex loop structures

189

with

BA
D1
D2
; GP1 =
; GP2 =
.
AA
C1
C2
The characteristic equation with regard to the disturbance behaviour is
GCA =

CZ [A C2 (C1 AA + D1 BA ) + D1 D2 B AA ] = 0

(3.7.14a)

and with regard to the reference behaviour is

A C2 (C1 AA + D1 BA ) + D1 D2 B AA = 0 .

(3.7.14b)

From this it is obvious that the introduction of the auxiliary controlled variable
has an inuence on the stability of the control system.
By a proper choice of GCA , on the one hand a reduction of the eects of the
disturbance on the second section of the plant (GP2 ) can be achieved and on the
other an improvement of the behaviour of the main control loop. The location
where the auxiliary measurement yA is taken should be after the entry of the
disturbance but as close as possible to the plant input. If the rst section of the
plant contains only short delays then a P controller is sucient for GCA . Often
the need for an auxiliary controller can be avoided if the auxiliary variable is
connected directly to the input of the main controller via a PT1 element.
Figure 3.7.5 again shows the example of the temperature control of a steam supers t e a m
flo w

J 2 = y
w
G C

S H

J
1

= y

G C A

+
_

M
C

u
c o o lin g
w a t e r

Figure 3.7.5. Example of a of a steam superheater temperature control system with

auxiliary controlled variable yA

heater system, where here it is congured with an auxiliary controller that uses
the steam temperature measurement 1 at the superheater inlet as an auxiliary
controlled variable.
DYNAST study example 3.7.1
PID and PI-D control of a PT1 Tt plant
DYNAST study example 3.7.2
PID and PI-D control of a 3rd-order plant

190

3 Behaviour of linear continuous-time control systems

3.7.4

Cascade control systems

Cascade control systems are special cases of control systems with auxiliary controlled variables. Here, as shown in Figure 3.7.6, the main controller with the
transfer function GC2 does not directly eect the actuator, but provides the reference value for the underlying auxiliary controller with the transfer function
GC1 . This auxiliary controller forms together with the rst plant section GP1 the
auxiliary control loop, which is inside the main control loop. Disturbances in the
rst plant section will be already controlled by the auxiliary controller such that
they have less inuence on the second section. The main controller has then only
to act slightly.
Z
W

+
E
_

G
C

(s)

+
_

G
C

(s) U

P Z

G
P

p la n t

(s)

(s )

+
G

(s)

Figure 3.7.6. Block diagram of a cascade control system

When multiple auxiliary variables are measured, multiple cascade control systems
can be built. For the cascade control system of Figure 3.7.6 the controlled variable
y is given by



Y


(3.7.15)
GC1 GP1 + Z GPZ GP2 ,
Y =
(W Y ) GC2
GC2
which gives
Y =

GC1 GC2 GP1 GP2

GPZ GP2
Z
+
W
1 + GC1 GP1 (1 + GC2 GP2 )
1 + GC1 GP1 (1 + GC2 GP2 )

(3.7.16a)

and can be written as

Y =

with

A1 A2 C1 D2 DZ
Z

CZ [A1 A2 C1 C2 + B1 D1 (A2 C2 + B2 D2 )]
B1 B2 D1 D2
W
+
A1 A2 C1 C2 + B1 D1 (A2 C2 + B2 D2 )

(3.7.16b)

3.7 Improving the control behaviour by more complex loop structures

191

B1
B2
; GC2 =
.
A1
A2
The characteristic equation with regard to the disturbance behaviour is
GC1 =

CZ [A1 A2 C1 C2 + B1 D1 (A2 C2 + B2 D2 )] = 0

(3.7.17a)

and with regard to the reference behaviour

A1 A2 C1 C2 + B1 D1 (A2 C2 + B2 D2 ) = 0 .

(3.7.17b)

It is obvious that the stability is inuenced by the underlying auxiliary control

loop. If the reference behaviour of the auxiliary control loop in Figure 3.7.6 is
summarised as
GC1 GP1
,
(3.7.18)
GA =
1 + GC1 GP1
then Eq. (3.7.16a) can be represented by
Y =

GC2 GA GP2
GPZ
GP2
Z
+
W
1 + GC2 GA GP2 1 + GC1 GP1
1 + GC2 GA GP2

(3.7.19)

From this equation the block diagram of a single-loop control system can be
drawn as shown in Figure 3.7.7, which describes the same system as Figure 3.7.6.
The auxiliary control system with the transfer function GA is an element of the
G

+
_

G
C

(s)

P Z

1 + G

1
C 1

1 + G

G
P

1
P

C 1

+
1
P

+
G

(s)

Figure 3.7.7. Rearranged block diagram of a cascade control system

basic control loop. Therefore, it is possible to design a cascade control system in

the following steps:
1. Design of the auxiliary control system, i.e. parametrisation of the controller
transfer function GC1 for the given section GP1 of the plant for disturbances.
The auxiliary control system must be fast (high bandwidth) and therefore
mostly a P or PD controller is chosen for GC1 .
2. Design of the main controller transfer function GC2 for the plant transfer
function GA GP2 . GC2 has the task to remove steady-state errors in the controlled variable. Therefore, it is expedient to use a PI controller as long as
the plant has PTn behaviour.

192

3 Behaviour of linear continuous-time control systems

(a )
J = y

v a lv e

c o o lin g
w a te r

M
w
+

+
_

+
_

g o v e rn o r
+
G C
2

c o o lin g
w a te r

p u m p

m 1) y A

s t ir r e d
ta n k
re a c to r

u
G

c u rre n t
r e g u la t o r
G C
y

c o n tro l
a n g le
a
c u rre n t
c o n v e rte r

(b )
F

u a

T
n

ia

y = n
Figure 3.7.8. Examples of cascade control systems: (a) temperature control of a stirred
tank reactor, (b) governor of a DC motor with motor current control

Figure 3.7.8 shows two examples of cascade control systems.

Demonstration Example 3.7.1
A virtual experiment using cascade control

3.7.5

Control system with auxiliary manipulated variable

Disturbances can also be counteracted by a conguration where an additional

auxiliary control signal uA is inserted between the actuator of the main control loop and the controlled variable. This auxiliary manipulated variable will
be generated by an additional auxiliary controller GCA . Figure 3.7.9 shows the
corresponding block diagram.
From this diagram it directly follows that the controlled variable is
Y = [(W Y ) GC GP1 + Z
GPZ + (W Y ) GCA ] GP2

(3.7.20)

3.7 Improving the control behaviour by more complex loop structures

Z

W
+

E
_
G

C A

(s)
(s)

G
G

P Z

(s)

(s)
1

193

p la n t
+
+

U A

+
G

(s )

Figure 3.7.9. Block diagram of a control system with an auxiliary manipulated variable
uA

or rearranged
Y =

GP2 (GCA + GC GP1 )

GPZ GP2
Z
+
W ,
1 + (GC GP1 + GCA ) GP2
1 + (GC GP1 + GCA ) GP2

(3.7.21a)

which becomes
Y =

A C1 D2 AA DZ
Z

CZ [A C1 (C2 AA + D2 BA ) + B D1 D2 AA ]
D2 (B D1 AA + A C1 BA )
+
W .
A C1 (C2 AA + D2 BA ) + B D1 D2 AA

(3.7.21b)

The characteristic equation with regard to the disturbance behaviour is

CZ [A C1 (C2 AA + D2 BA ) + B D1 D2 AA ] = 0

(3.7.22a)

and with regard to the reference behaviour is

A C1 (C2 AA + D2 BA ) + B D1 D2 AA = 0 .

(3.7.22b)

The stability of the main control loop is inuenced by adding the auxiliary manipulated variable uA . During the choice of the auxiliary manipulated variable it
must be observed that the second plant section GP2 should have an as small as
possible delay, because the auxiliary controller can cancel disturbances faster. In
the steady state uA must be zero, if the steady state must be inuenced only by
u. This is possible when for GCA a DT1 element is used.

3.7.6

Control system with anti-windup measure

In practical applications the manipulated variable u must not exceed given extreme values. This is the case due to either the bounded power of the actuator
or to the physical constraints of the plant. In most cases these hard limitations
of the manipulated variable must be respected. This means that the modulus of
the manipulated variable must not exceed given bounds

194

3 Behaviour of linear continuous-time control systems

|u(t)|  umax .

(3.7.23)

For control system design using linear methods it is dicult to cope with this
problem and to abide by the bounds on the manipulated variable. When the
design is performed such that the amplitudes of the manipulated variable are
small and do not reach the bounds, the actuator is not fully exploited and, thus,
the control response is slow. On the other hand, when the bounds are exceeded
for a reasonable period of time undesirable control behaviour may be obtained.
In order to discuss the problem, the bounds are described by a saturation element,
as shown in Figure 3.7.10. The variable uC is the manipulated variable obtained

e
w
+
_

c o n tro lle r

u
C

p la n t

s a tu ra tio n

Figure 3.7.10. Block diagram of a control system with a bounded manipulated variable

from the controller and u the manipulated variable acting on the plant, which is
determined from

umax for uC (t) < umax

(3.7.24)
u(t) = uC (t)
for |uC (t)|  umax .

umax
for uC (t) > umax
When the bounds are exceeded the nonlinear saturation characteristic will take
eect and inuence the dynamical behaviour. In some cases the closed-loop system may also become unstable or show an oscillating behaviour.
This undesired phenomenon, called the windup eect, occurs in all control systems
where an integrator is used in the controller. This integrator is necessary to have
a zero steady-state control error. In order to demonstrate this eect, the example
from section 3.5.3 is taken. The plant is given by Eq. (3.5.37) and the controller
by Eq. (3.5.43). The step response of the closed loop without saturation is shown
in Figure 3.7.11, where the response of the controlled variable is the same as
in Figure 3.5.20. If the manipulated variable is bounded by umax = 1.5 (with
saturation) the rise time increases due to the smaller values of the manipulated
value in the period from 0.1 s to 2 s. The increased maximum overshoot and
settling time reect a worse control behaviour. The reason for this is the following:
From the beginning, the control error decreases and changes sign at 1.6 s. As uC
is very large at this time ( 2.1), the manipulated variable u cannot be reduced
despite the negative control error. This will only occur when uC falls below umax
at 2 s. The problem is obviously that the controller continues to integrate though
the manipulated variable has already reached its bound. As the controller output
uC further grows unnecessarily, this is called the windup eect.

3.7 Improving the control behaviour by more complex loop structures

195

The goal of an anti-windup measure is to counteract the integration of the controller. This can be performed by feeding back the dierence u uC to the controller. Figure 3.7.12 shows a simple approach for an anti-windup measure, where
the dierence is weighted by the factor and fed into the controller. Figure 3.7.11
shows the improvement of the behaviour for = 1. The settling time is close to
the case without saturation, but the maximum overshoot is half of that without
saturation.
u ( t) ,u
C

(t)
2 .5

w it h o u t s a t u r a t io n
u

2 .0

(a )

(t)
w it h s a t u r a t io n

1 .5
1 .0
0 .5
0

w it h a n t i-w in d u p
1
0

y (t)

t [s ]

(b )

w it h s a t u r a t io n

1 .2 5

1 .0 0
0 .7 5

w it h a n t i-w in d u p
w it h o u t s a t u r a t io n

0 .5 0
0 .2 5
0
0
1

t [s ]
7

Figure 3.7.11. Step response of the closed loop system with and without anti-windup
measure, (a) manipulated variable u and controller output uC , (b) controlled variable y

w
_

c o n tro lle r

u
C

p la n t

s a tu ra tio n
_
g

Figure 3.7.12. Block diagram of a control system with an anti-windup measure

196

3 Behaviour of linear continuous-time control systems

197

State variable techniques

4.1

State-space representation

4.1.1

State-space representation of single-input-single-output systems

In the following a short introduction into the representation of systems using

state-variable techniques is given. For this purpose, the example of the RLC
network from Figure 2.3.16 is used. The dynamical behaviour of this network is
completely dened for t t0 , if the
initial conditions ua (t0 ), ie (t0 )
and the
input variable ue (t) for t t0
are known. For these specications the variables ie (t) and ua (t) can be determined
for all t t0 . The variables ie (t) and ua (t) characterise the state of the network
and are therefore called state variables of the network.
The dierential equations of Eqs. (2.3.46) and (2.3.47) describe the dynamical
behaviour of this network. Inserting Eq. (2.3.47) into Eq. (2.3.46) one obtains
the 2nd-order dierential equation according to Eq. (2.3.48), which completely
describes the system with respect to the input-output behaviour. But one can also
use the two original dierential equations and can write them in vector notation
so that the 1st-order vector dierential equation

die (t)
R
1 
 1

i
dt L
e (t)
L
+ L ue (t)
(4.1.1)

du (t) = 1
ua (t)
a
0
0
C
dt
with the initial condition


ie (t0 )
ua (t0 )

198

4 State variable techniques

is obtained. This linear 1st-order vector dierential equation describes the connection between the input variable and the state variables. To complete a state-space
system, one needs an additional equation that describes the dependence of the
output variable on the state variables. In this example, it is the direct relationship
y(t) = ua (t) .
Introducing the state vector
 

ie (t)
x1 (t)
=
x(t) =
x2 (t)
ua (t)


into Eq. (4.1.1), with the vectors



i (t )
x0 = x(t0 ) = e 0
ua (t0 )


,

1
b= L
0
with the matrix

and

cT = [0

1

L
0

R
L
A= 1
C

1] ,

and with the scalar variables

u(t) = ue (t)

and

d=0

one obtains the general state-space representation of a linear time-invariant singleinput-single-output system:

x(t)
= A x(t) + b u(t)

x(t0 ) initial condition

y(t) = cT x(t) + d u(t) .

(4.1.2)
(4.1.3)

The Eq. (4.1.2) is the state equation, and in the general case it is a linear system
of 1st-order dierential equations of n state variables x1 , x2 , . . . , xn , which are
combined n the state vector x = [x1 x2 . . . xn ]T . Eq. (4.1.3) is the output equation,
which maps the states and inputs linearly to the output. This is an algebraic
equation, whereas the state equation is a dierential equation.

4.1 State-space representation

4.1.2

199

State-space representation of multi-input-multi-output systems

The Eqs. (4.1.2) and (4.1.3) describe an nth-order linear time-invariant singleinput-single-output system. For linear multi-input-multi-output systems of order
n with r inputs and m outputs these equations become

x(t)
= A x(t) + B u(t)

with the initial condition

y(t) = C x(t) + D u(t) ,

where the following notation is used:

x1 (t)

state vector
x(t) = ...
xn (t)

u1 (t)

u(t) = ...

x(t0 )

(4.1.4)
(4.1.5)

(n 1) vector

input vector

(r 1) vector

ur (t)

y1 (t)

y(t) = ...

output vector

(m 1) vector

ym (t)
system matrix
input matrix
output matrix
feedthrough matrix

A
B
C
D

(n n) matrix
(n r) matrix
(m n) matrix
(m r) matrix

It goes without saying that the general representation of Eqs. (4.1.4) and (4.1.5)
also includes the single-input-single-output case. The matrices A, B, C and D
have constants elements. If these elements are time-varying, the matrices of the
corresponding time-varying system are substituted by matrix functions of time,
e.g. A(t).

4.1.3

The relationship between transfer functions and the statespace representation

In the following, the Eqs. (4.1.4) and (4.1.5) will be transformed into the s domain
using the Laplace transform, which will be done analogously to the scalar case in
section 2.1.5. For this, the operator notation

200

4 State variable techniques

F (s) =  [f (t)]

from section 2.1.1 is adopted and when applying it to Eq. (4.1.4), one obtains
sX(s) x(t0 ) = A X(s) + B U (s) ,
or rearranged
(sI A) X(s) = x(t0 ) + B U (s) .
The solution of the state equation in the s domain is then given by

with

X(s) = (s) x(t0 ) + (s) B U (s)

(4.1.6)

(s) = (sI A)1 .

(4.1.7)

Similarly, for Eq. (4.1.5) yields

Y (s) = C X(s) + D U (s) .
Substituting X(s) from Eq. (4.1.6), the system output in the s domain is
Y (s) = C (s) x(t0 ) + [C (s) B + D] U (s) .
To obtain the relationship with transfer functions, the initial condition x(t0 ) has
to be set to zero. For a single-input-single-output system according to Eqs. (4.1.2)
and (4.1.3) the system output is


Y (s) = cT (s) b + d U (s) .
Comparing this equation with Eq. (2.2.3) the transfer function is given by
G(s) = cT (s) b + d .

(4.1.8)

The matrix (s) from Eq. (4.1.7) is a matrix of rational functions of s, which
can always be represented by
(s) =

1
(s) ,
det(sI A)

(4.1.9)

where (s) is a matrix with polynomial elements in s. From Eq. (2.2.2) it is

obvious that
(4.1.10)
N (s) = cT (s) b + d D(s)
and
D(s) = det(sI A) ,

(4.1.11)

which is the characteristic polynomial of the system. The zeros of this polynomial are the poles of the transfer function and at the same time eigenvalues of
the system matrix A. If the system in the state-space representation is fully controllable and observable (see section 4.1.6), then the number of poles are equal
to the number of eigenvalues.

4.1 State-space representation

4.1.4

201

State-space vs transfer function approach

The key advantage of transfer functions is in their compactness, which makes

them suitable for frequency-domain analysis and stability studies. However, the
transfer function approach suers from neglecting the initial conditions. Not only
does state-space representation serve as an alternative to transfer functions, but
also it is not limited to linear and time-invariant systems and it has the following
advantages:
1. Single-input-single-output and multi-input-multi-output systems can be formally treated equal.
2. The state-space representation is best suited both for the theoretical treatment of control systems (analytical solutions, optimisation) and for numerical calculations.
3. The determination of the system response in the homogeneous case with the
initial condition x(t0 ) is very simple.
4. This representation gives a better insight into the inner system behaviour.
General system properties, for example, the system controllability or observability can be dened and determined, see section 4.1.6.

4.1.5

Uniqueness of the state variables

Initially it sounds paradoxical that the choice of the state variables is not unique.
This means that for one and the same system with the input u, the output y and
n state variables, there exist an innite number of state-space representations.
For each value of time t one gets the state in the n-dimensional state space. The
n values are the cartesian coordinates of the state

x1
x2

x = .. = x1 e1 + x2 e2 + . . . + xn en ,
.
xn
where the unit vectors ei are n-dimensional linear independent vectors. Their
elements are besides the ith element, which has the value of 1 all zero. For
describing the state also other basis vectors can be used. Candidates are all n
linear independent and n-dimensional vectors ti . Therefore, it is always possible,
to write the state as
(4.1.12)
x = z1 t1 + z2 t2 + . . . + zn tn .
After introducing the new state

202

4 State variable techniques

z1
z2

z = ..
.
zn

and the quadratic matrix



T = t 1 , t 2 , . . . , tn

one can rewrite Eq. (4.1.12) as

x=Tz .

(4.1.13)

The constant and regular matrix T is a so-called transformation matrix. Now,

instead of the state x the new state z can be used and its behaviour analysed.
Describing the system in the new coordinates the state x from Eq. (4.1.13) is
inserted into Eqs. (4.1.4) and (4.1.5) and one obtains

z(t)
= (T 1 A T ) z(t) + (T 1 B) u(t)

(4.1.14)

y(t) = (C T ) z(t) + D u(t)

(4.1.15)

with the initial condition

z(t0 ) = T 1 x(t0 ) .

The benets of the transformation of systems into dierent state-space representations are:
Most system properties do not depend on the choice of the state variables.
They remain unchanged after a regular transformation and may be analysed
in an appropriate representation form.
The computational determination and analysis of system properties can be
tremendously simplied if the representation form is specically selected. In
particular certain canonical forms are of interest.
Example 4.1.1
In order to demonstrate a transformation, the example from Eq. (4.1.1) is used
with the system parameters R = 3, L = 1 and C = 0.5. The initial condition
is assumed to be zero and therefore omitted. With these values one obtains the
state equation as


 
3 1
1

x(t)
=
x(t) +
u(t)
2
0
0
and the output equation as


y(t) = 0 1 x(t) .
For the regular transformation matrix

4.1 State-space representation

203



1 1
T =
1
2
one obtains the matrices for Eqs. (4.1.14) and (4.1.15) as

1 

 

1 1
3 1
1 1
2
0
1
=
T AT =
1
2
2
0 1
2
0 1

   
2 1 1
2
=
T 1 B =
1 1 0
1



 1 1


CT = 0 1
= 1 2 .
1
2
The new representation


 
2
0
2

z(t)
=
z(t) +
u(t)
0 1
1


y(t) = 1 2 z(t)

consists of two decoupled dierential equations with respect to the state variables


z1 and z2 .
The analysis and treatment of a system in such a structured representation form,
as shown in the example above, is doubtless more simple. As the representation
form must be specically selected depending on the type of analysis or synthesis
problem, the dierent representation forms, for example, the canonical forms, are
not discussed separately and are introduced in the following sections when they
are needed.

4.1.6

Controllability and observability

Controllability and observability are important structural properties of a dynamic

system. Controllability can be dened as follows:
The system of Eqn. (4.1.2) is controllable if there exists a control signal
u(t) that will take the state of the system from any initial state x(t0 )
to any desired nal state xf in a nite time interval.
This condition is equivalent to the following condition:
The system of Eqn. (4.1.2) is controllable if the controllability matrix


(4.1.16)
S = b Ab . . . An2 b An1 b
has full rank n.

204

4 State variable techniques

The concept of observability is parallel to that of controllability and all can be

transformed to statements about observability by invoking the property of duality,
as discussed later in section 4.2.4.2. The observability denition analogous to
those for controllability are as follows:
The system of Eqs. (4.1.2) and (4.1.3) is observable if, for any x(t0 ),
there is a nite time such that x(t0 ) can be determined from u(t) and
y(t) for 0 t .
This condition is equivalent to the following:
The system of Eqs. (4.1.2) and (4.1.3) is
matrix

cT
cT A

O=
..

observable if the observability

(4.1.17)

cT An1
has full rank n.

4.2
4.2.1

Design of state-feedback control systems

Structures and properties of state-feedback control systems

In the following, the design of state-feedback controllers for single-input-singleoutput systems described by

x(t)
= A x(t) + b u(t)

x(t0 ) initial condition

y(t) = cT x(t)

(4.2.1)
(4.2.2)

is discussed in detail.
The dynamical characteristics, for example, stability, decay of oscillations or sensitivity to disturbances, are determined by the distribution of the eigenvalues of
the system matrix A in the s plane. The goal is to inuence the system specically
so that it shows a desired behaviour. In the sense of command input regulation
the control system is congured as shown in Figure 4.2.1. It is assumed, that all
state variables are measured. A linear combination of the state variables is fed
back by
(4.2.3)
uf (t) = f T x(t)
with

4.2 Design of state-feedback control systems

u v
v

N
f

c o n t r o lle r

N = ) N + >  u
y = ? T N

- u

p la n t
.

u
+

205

Figure 4.2.1. Basic structure of a state-feedback control system



f = f1 , f2 , . . . , fn .

(4.2.4)

While this feedback determines the dynamical behaviour, the feedforward of the
command variable w
(4.2.5)
uv (t) = v w(t)
using the scalar gain v inuences the static behaviour. The manipulated variable
is obtained from
u(t) = uf (t) + uv (t) = f T x(t) + v w(t) .

(4.2.6)

After the substitution of the manipulated variable into Eq.(4.2.1) using Eq. (4.2.6),
one obtains the closed-loop system as



(4.2.7)
x(t0 ) initial condition
x(t)
= A b f T x(t) + b v w(t)
y(t) = cT x(t) .

(4.2.8)

For the dynamical and static behaviour the following specications must be fullled:
The dynamical behaviour of the closed loop should be specied by given
poles. As these closed-loop poles are the eigenvalues of the closed-loop system
matrix (A b f T ), the desired distribution of the eigenvalues in the s plane
for this matrix is specied.
In the steady state the control error must vanish so that the plant output
follows
(4.2.9)
y = lim y(t) = cT lim x(t) = lim w(t) = w .
t

4.2.1.1

State-feedback control in the frequency domain

As the dynamical behaviour of the closed loop is specied in the s domain, it is

appropriate to discuss some properties in the frequency domain. The initial point
of this discussion is the closed-loop transfer function

206

4 State variable techniques

Y (s)
= GW (s)
W (s)

(4.2.10)

Applying the Laplace transform as shown in section 4.1.3 to Eqs. (4.2.7) and
(4.2.8) one obtains the closed-loop transfer function as
GW (s) = cT (sI A + b f T )1 b v .

(4.2.11)

The same procedure for the plant from Eqs. (4.2.1) and (4.2.2) gives
GP (s) = cT (sI A)1 b .

(4.2.12)

With Eq. (4.1.7) the closed-loop and plant transfer functions are rewritten as
GW (s) = cT (1 (s) + b f T )1 b v
and
GP (s) = cT b =

N (s)
.
D(s)

(4.2.13)

(4.2.14)

Applying the matrix inversion lemma to the inner part (1 (s) + b f T )1 of

Eq. (4.2.13), one obtains


1 T

f (s) b v
GW (s) = cT (s) (s) b 1 + f T (s) b


cT (s) bf T (s) b
T
v
= c (s) b
1 + f T (s) b
cT (s) b v
.
(4.2.15)
=
1 + f T (s) b
After inserting Eqs. (4.1.9), (4.1.11) and (4.2.14) into Eq. (4.2.15) the closed-loop
transfer function is given by
cT (s) b v
D(s) + f T (s) b
N (s) v
,
=
D(s) + f T (s) b

GW (s) =

(4.2.16)



where the denominator D(s) + f T (s) b is the characteristic polynomial of
the closed-loop system and D(s) is that of the open-loop system. As only the
plant numerator polynomial N (s) appears in the closed-loop transfer function,
the closed-loop zeros are the same as the open-loop zeros. This means that the
zeros cannot be inuenced by a state-feedback controller; it only moves the poles.

4.2 Design of state-feedback control systems

4.2.1.2

207

Steady state of the closed-loop system

The steady state from Eq. (4.2.9) can only be reached if the state of the closedloop system for t approaches the nal value
x = lim x(t) .
t

The condition

x(t)
= 0 = (A b f T ) x + b v w
is obviously valid, from which the steady state can be obtained as
x = (A b f T )1 b v w ,
which is always possible, because for an asymptotically stable closed-loop the
matrix (A b f T ) has always full rank. For the output, one obtains
y = cT x = cT (A b f T )1 b v w ,

(4.2.17)

where it must be observed that this value must not vanish, i.e.
cT (A b f T )1 b
= 0 .
This condition can be fullled, if the closed-loop transfer function GW (s) from
Eq. (4.2.11) has no zero at s = 0. From the discussion about the zeros in section 4.2.1.1 it is clear that the plant transfer function from Eq. (4.2.12) must not
have a zero at s = 0. This means that
cT A1 b
= 0 .

(4.2.18)

With the condition given in Eq. (4.2.18) all given values y are reachable with
command input signals that have a constant steady-state value of w .
From the conditions in Eqs. (4.2.17) and (4.2.9) the feed-forward gain of the
controller is obtained as
v=

4.2.2

cT (A

1
.
b f T )1 b

(4.2.19)

State-feedback control with integrator

For a given feedback vector f the feedforward gain v can be calculated according
to Eq. (4.2.19) so that the closed-loop system in Figure 4.2.1 shows the desired
static behaviour with a vanishing control error. The control structure used is not
very robust with respect to the control error, because this error is not fed back.

208

4 State variable techniques

e
_

- e
v

u v
+

p la n t
.

N = ) N + >  u
y = ? T N

- u f
B

N


Figure 4.2.2. Basic structure of a cascade state-feedback control system with integrator

Uncertainties in the plant model parameters or disturbances acting on the plant

may cause steady-state control errors. In order to reject such eects, one can
use a similar approach to that shown in section 3.7.4 for cascade control. The
cascade control system of Figure 4.2.2 has the state-feedback conguration from
Figure 4.2.1 as its inner loop. The controlled variable y is compared with the
set-point value w and the control error e is fed back to an integrator. The former
feedforward gain v is now the gain of the integrator. From the closed-loop point
of view with this conguration the gain like f is a feedback parameter. This will
be shown in the following.
From Figure 4.2.2 the equations of the closed-loop system can be directly determined as follows:

x(t)
= A x(t) + b u(t) x(t0 ) initial condition

(4.2.20)

y(t) = cT x(t)

(4.2.21)

(t)
= y(t) w(t)

(4.2.22)

u(t) = uf (t) + uv (t) = f T x(t) v (t) .

(4.2.23)

Combining Eqs. (4.2.20) to (4.2.22) of the open-loop system in matrix notation,

one obtains

  
 

 

A 0 x(t)
b
0
x(t)
+
u(t) +
w(t) .
(4.2.24)
= T
(t)
0
1
(t)

0
c
Eq. (4.2.23) for the manipulated variable can be represented by

 T 
x(t)
f
.
u(t) =
(t)
v
With the abbreviations


 
 


A 0
b
0
x(t)

, b =
, v =
, x (t) =
A = T
0
1
(t)
0
c

(4.2.25)

4.2 Design of state-feedback control systems

209

Eqs. (4.2.24) and (4.2.25) are

x (t) = A x (t) + b u(t) + v w(t)

x (t0 ) initial condition

u(t) = f T x (t)

(4.2.26)
(4.2.27)

and for the closed-loop system after inserting Eq. (4.2.27) into Eq. (4.2.26)


(4.2.28)
x (t) = A b f T x (t) + v w(t) x (t0 ) initial condition ,
which is formally the same system as that with the simple state-feedback controller in Eq. (4.2.7). Here, the controller parameters in f instead of those in f
must be determined for the desired specications. This extended problem is now
reduced to the original problem, and for both cases the same design procedure
can be applied. It means, that for the state-feedback controller with integrator
the same design procedures for the controller parameters can be applied as for the
original system. This simplies the design of the extended problem signicantly.

4.2.3

Design of state-feedback controllers by pole placement

The diculty of this design consists essentially in the determination of the feedback vector f so that the n eigenvalues of the system matrix (A b f T ) have
the desired distribution. After that, the determination of the feedforward gain v
in the control structure without integrator is very simple, see Eq. (4.2.19).


The characteristic polynomial Q(s) = det s I (A b f T ) of the closed loop
system from Eq. (4.2.7) is a monic polynomial of order n. The coecients qi of
it are functions of the controller parameters hi :


det s I (A b f T ) = q0 (f ) + q1 (f )s + q2 (f )s2 + . . . + qn1(f )sn1 + sn .
By a proper choice of the vector f this polynomial Q(s) should be made equal
to the desired polynomial P (s) with n zeros, which are the desired poles or
eigenvalues si , respectively, of the closed-loop system:
P (s) =

n
6

(s si ) .

(4.2.29)

i=1

Multiplying all factors on the right-hand side of Eq. (4.2.29) one obtains this
polynomial as
P (s) = p0 + p1 s + p2 s2 + . . . + pn1 sn1 + sn .

(4.2.30)

For all values of s, the condition Q(s) = P (s) must be fullled. A comparison of
the corresponding terms of both sides gives the coecients as
q0 (f ) = p0 , q1 (f ) = p1 , q2 (f ) = p2 ,

...

qn1 (f ) = pn1 ,

(4.2.31)

from which the controller parameters fi can be obtained. This approach, however,
has the following drawbacks:

210

4 State variable techniques

The Eqs. (4.2.31) for determining the controller parameters are complicated,
in general nonlinear.
For higher-oder systems the computational eort is large.
There is no systematic way to solve the equations.
The determination of the feedback parameters can be signicantly simplied when
the invariance of the eigenvalues of a system under a regular transformation is
observed and used. The idea is to transform the system into a form which is
suitable for the determination of the controller parameters. In the sections below
this idea is used.

4.2.3.1

Design of a system in controller canonical form

A state-space system in controller canonical form has the following structure:

z(t)
= Ac z(t) + bc u(t) z(t0 ) initial condition

(4.2.32)

y(t) = cT
c z(t)

(4.2.33)

with

0
0

Ac =
.
0
a0

1
0
.
0
a1

0
1
.
0
a2

.
0
0
.
0

..
.
.
=
,
b

,
c

0
.
1
1
. an1

c1
c2

cc = .. .
.

(4.2.34)

cn

This canonical form has the following properties:

The characteristic polynomial can be directly determined from the last line
of Ac , which is
Pc (s) = det(s I Ac ) = a0 + a1 s + a2 s2 + . . . + an1 sn1 + sn . (4.2.35)
A system of this structure is always controllable as its controllability matrix
according to Eq. (4.1.16) has always full rank.
The transfer function of the system is immediately given by
G(s) =

c1 + c2 s + c3 s2 + . . . + cn sn1
.
a0 + a1 s + a2 s2 + . . . + an1 sn1 + sn

4.2 Design of state-feedback control systems

211

The feedback is now dened as

u(t) = f T
c z(t)
with

(4.2.36)

fc1
fc2

f c = .. .
.
fcn

For the closed-loop system the system matrix is

0
1
0

0
0
1

.
.
.
Ac b c f T
c =

0
0
0
a0 fc1 a1 fc2 a2 fc3

.
0

.
0

(4.2.37)
.
.

.
1
. an1 fcn

and its characteristic polynomial is

Q(s) = (a0 + fc1 ) + (q1 + fc2 ) s + (a2 + fc3 ) s2 + . . . + (an1 + fcn ) sn1 + sn .
Equating this polynomial with the polynomial with the desired poles from Eqs. (4.2.30)
one obtains by comparison of the corresponding terms of both polynomials the
controller parameters as
fci = pi1 ai1

for

i = 1, 2, . . . , n .

(4.2.38)

In the controller canonical form the calculation of the controller feedback parameters is reduced to the calculation of a simple dierence between the coecients
of two polynomials.

4.2.3.2

Design of a system not in a canonical form

In general, when a system is not given in the controller canonical form, one has
to transform it by a regular transformation
z(t) = T x(t) ,

(4.2.39)

which brings the system into the desired canonical form according to Eqs. (4.2.32)
to (4.2.34). The determination of the controller parameters in f c is performed according to Eq. 4.2.38. The feedback law in the original state is, using Eq. (4.2.39)
given by
T
(4.2.40)
u(t) = (f T
c T )x(t) = f x(t) .

212

4 State variable techniques

Finally, the feedback vector is transformed back to

fT = fT
cT .

(4.2.41)

The main task in the pole-placement design for systems that are not in controller
canonical form, is the determination of the transformation matrix T . When the
original state equation from Eq. (4.2.1) is transformed by Eq. (4.2.39), one obtains
the transformed entities
(4.2.42)
Ac = T A T 1
and
bc = T b .

(4.2.43)

Eq. (4.2.42) will be further analysed in the form by right-multiplying with T

Ac T = T A .
With the row vectors tT
i of the matrix T this equation is
T T

t1
t1
0
1
0
.
0
tT tT
0
0
1
.
0
2 2

.
.
.
.
.
T. = T. A

0
0
0
.
1
tn1
tn1
T
a0 a1 a2 . an1
tn
tT
n

(4.2.44)

and after multiplication:

T
tT
t1 A
2

tT A
tT
3

= . .
.

tT
tT
n
n1 A
T
T
T
T
T
a0 t1 a1 t2 . . . an2 tn1 an1 tn
tn A

(4.2.45)

From the rst n 1 rows of both sides one obtains the recursive relationship
T
tT
i+1 = ti A

for

i = 1, 2, . . . , n 1 ,

and when tT
1 is known the remaining rows of the matrix T are
T i
tT
i+1 = t1 A

for

i = 1, 2, . . . , n 1 .

(4.2.46)

The rst row tT

1 is obtained from Eq. (4.2.43), which is also valid. Using the
results from Eq. (4.2.46), the right-hand side of Eq. (4.2.43) has the form:

tT
t1
1b

tT
tT
2
1 Ab

.
(4.2.47)
bc = T b =

. b=
n2
T
tT

b
t1 A
n1
T An1 b
tT
t
n
1

4.2 Design of state-feedback control systems

213

or in transposed form:


T
n2
b An1 b = tT
bT
c = t1 b Ab . . . A
1 S .

(4.2.48)

The matrix S is the controllability matrix from Eq. (4.1.16), which has full rank
if the system is completely controllable. Under this condition one can obtain the
rst row by
T 1
tT
,
(4.2.49)
1 = bc S
which is the last row of the inverse controllability matrix, because bc is the n-th
unit vector.
Summarising the procedure of the pole-placement design for a system that is not
in a canonical form, the following steps are necessary:
1. Calculation of the controllability matrix S from Eq. (4.1.16), its inverse and
extracting the last row according to Eq. (4.2.49). This is the rst row of the
transformation matrix T .
2. Row-wise calculation of the remaining rows of the transformation matrix T
using Eq. (4.2.46).
3. Choice of the n eigenvalues si .
4. Calculation of the coecients pi of the desired polynomial according to
Eqs. (4.2.29) and (4.2.30).
5. Transformation of the system matrix A according to Eq. (4.2.42) and extracting the coecients ai of the characteristic polynomial of the open-loop
system from the last row of the transformed matrix.
6. Calculation of the coecients of the feedback vector f T
c according to Eq. (4.2.38).
7. Back transformation of the feedback vector according to Eq. (4.2.41).

4.2.3.3

Design using Ackermanns formula

The method described in the previous section 4.2.3.2 needs the computation of
the full transformation matrix T . This is not necessary as shown by the simplied
procedure of Ackermann.
Steps 5 and 6 from the end of the previous section 4.2.3.2 will be combined by
using the relationship for the transformation matrix T from Eqs. (4.2.41) and
(4.2.46):

214

4 State variable techniques

tT
1
T
t1 A
..
.




f T = p0 a0 , p1 a1 , . . . pn1 an1
.

n1
tT
A
1
After multiplying, the feedback vector can be represented by two sums as
f T = tT
1

n1

i=0

pi Ai tT
1

n1

ai Ai .

i=0

n
Adding the term tT
1 A to both sums, one obtains
T
f T = tT
1 P (A) t1 Pc (A) .

(4.2.50)

P (A) is the desired polynomial from Eq. (4.2.30), where the scalar variable s
is substituted by the matrix A. The same holds for the characteristic polynomial Pc (s) of the system matrix A from Eq. (4.2.35). According to the CayleyHamilton theorem a quadratic matrix always satises its own characteristic polynomial, i.e. Pc (A) = 0, Eq. (4.2.50) is simplied to
f T = tT
1 P (A) .

(4.2.51)

1
the calculation of
As tT
1 is the last row of the inverse controllability matrix S
the feedback vector is simpler than in the previous case of section 4.2.3.2. The
following steps are necessary:

1. Calculation of the controllability matrix S from Eq. (4.1.16), its inverse and
extracting the last row.
2. Choice of the n eigenvalues si .
3. Calculation of the coecients pi of the desired polynomial according to
Eqs. (4.2.29) and (4.2.30).
4. Calculation of the term P (A).
5. Calculation of the coecients of the feedback vector f T according to Eq. (4.2.51).

4.2.4

State reconstruction using observers

The state-feedback controllers designed in the previous sections assume that all
state variables can be measured. This may not be the case. That is, certain components of the state vector may correspond to inaccessible internal variables,
which may not be available for measurement. The theory of observers has been

4.2 Design of state-feedback control systems

215

p la n t
.

N = ) N + > u
y = ? T N

o b se rv e r
N
Figure 4.2.3. Principle of an observer

proposed to reconstruct an approximation of the state vector based only on available measurements. Being a dynamical system in itself, the observer is designed
to track the n state variables of the original system. The block diagram in Figure 4.2.3 shows the system structure. The observer has two inputs, u(t) and y(t),
(t) of the plant state x(t). The estimate should
and it provides an estimate x
satisfy the condition
(t) = lim x(t) .
lim x
t

Thus, the error between the plant state and the estimate
(t) = x(t) x
(t)
x
should converge asymptotically to zero, i.e.
(t) = 0 .
lim x

4.2.4.1

(4.2.52)

Structure of an observer

A candidate for a dynamical system in state space form which can full the
condition from Eq. (4.2.52), may have the following form:
x
u(t) + g
(t) + b
y(t) x
(t0 ) initial condition .
x
(t) = A

(4.2.53)

In order to show this, Eq. (4.2.53) is subtracted from Eq. (4.2.1) and using
Eq. (4.2.2) the state equation




u(t) x
x
(t) + b b
(t0 ) initial condition (4.2.54)
cT x(t) A
x
(t) = A g
is to be independent of the plant input u(t),
is obtained. If the observing error x
one should make the choice

216

4 State variable techniques

= b,
b

(4.2.55)

and to have a homogeneous error state equation, one has to set the observer
system matrix to
=Ag
cT .
(4.2.56)
A
Then the error state equation is


(t) x
(t0 ) initial condition .
cT x
x
(t) = A g

(4.2.57)

is the only design parameter of the observer, which has to be

The parameter g
specied so that the error converges asymptotically to zero, independent of the
(t0 ). This is only the case if all eigenvalues of the matrix
unknown initial error x
cT ) have negative real parts.
(A g
In order to show how an observer works and to gain the interpretation of the ob in Eq. (4.2.53) is substituted by Eq. (4.2.56) and introducing
server equations, A
the output error term
(t) ,
y(t) = y(t) cT x
the observer state equation is
(t) + b u(t) + g
y(t)
x
(t) = A x

(t0 ) initial condition .

x

From this equation one can see, that the observer consists of a copy of the plant
model and of a feedback of the error between the plant model output and the
estimated plant output using the state estimates from the observer. The feedback
are the design parameters of the observer.
gains in g

4.2.4.2

Design of observers

has to be specied so that the matrix

The design parameter g
=Ag
cT
A
has n given eigenvalues si in the left-half s plane. This means that the characteristic polynomial is given by
n
 6

cT ) =
(s si ) .
det s I (A g
i=1

The determination of the design parameter is solved in an elegant way by observing that the characteristic polynomial of a matrix is invariant to transposition.
Therefore, the above equations can be rewritten with transposed matrices as:

4.2 Design of state-feedback control systems

217

T = AT c g
T
A
n

 6
T) =
det s I (AT c g
(s si ) .
i=1

A comparison of these equations with those of the state-feedback controller do

not show structural dierences. Instead of A now AT , instead of b now c and
is used. Therefore, an observer design is the same task as
instead of f now g
designing the feedback
uf (t) = g T (t)
for the system

(t)
= AT (t) + c y(t) .
Such a system is called a dual system. For this dual system the same design
procedure as for the state-feedback controller can be applied, i.e. the design of
a state-feedback controller of a dual system is the same as the design of an
observer for the original system. Therefore, designing an observer, one takes the
dual system and designs a state-feedback controller according to section 4.2.3.
of the
The resulting feedback vector is then the required parameter vector g
observer.

4.2.5

Combined observer-controllers

One is led to the observer problem because of the need to obtain the states for
(t) of the
use in the controller. Now, only the asymptotically correct estimates x
states x(t) rather than the states themselves are available. A natural question is
whether the previous result on pole placement via state feedback will continue
to hold when these estimates of the actual states are available. There is no other
option available than to see what happens with the observer and controller.
In the steady state there should clearly be no loss in using the asymptotic observer, since the error in the estimates will be zero. The question for the present
scheme is whether the incorporation of the observer dynamical system into the
feedback loop will aect the stability of the overall system. That is, interconnections of stable subsystems may lead to unstable overall systems.
The following analysis on stability is performed by setting up the joint observercontroller system of Figure 4.2.4. For this scheme the equations for the overall
system can be written down by inspection as
(t) + b v u(t) x(t0 ) initial condition

x(t)
= A x(t) b f T x


T
T
T
x
c bf x
(t) + b v u(t) x
(t0 ) initial condition .
c x(t) + A g
(t) = g

218

4 State variable techniques

u v

u
+

o b s e r v e r -c o n t r o lle r

N = ) N + >  u
y = ? T N

- u

p la n t

.
N ^ = ) ^ N^ + > u + C ^ y
f

B


Figure 4.2.4. Basic structure of a combined observer-controller

(t) the second

To study the overall system in terms of the state x(t) and its error x
equation is subtracted from the rst, and one obtains


+ b v u(t) x(t0 ) initial condition

x(t)
= A b f T x(t) + b f T x


x
(t) x
(t0 )initial condition
cT x
(t) = A g
which in matrix form can be written

  

 

x(t)
x(t)
bv
b fT
A b fT
+
w(t)
=
T
x
(t)
x
0
c
(t)
0
Ag

x(t0 )
(t0 )
x


initial condition .

As the system matrix of the above overall system has a triangular block form,
the characteristic polynomial of the overall system


b f T
s I (A b f T )
det
cT )
0
s I (A g
is just the product of the characteristic polynomial of the observer and the characteristic polynomial of the controlled system




cT )
det s I (A b f T ) det s I (A g
assuming perfect knowledge of the states.
This is nice, because it means that the natural frequencies or modes of the overall
system can always be arranged to be stable. In fact, they can be chosen completely
arbitrarily.
Another useful consequence is that the controller and observer can be designed
independently of each other. Whether the true states are available, or only asymptotically correct estimates of the states, is immaterial to the calculation of the
feedback vector f . Similarly, the dynamics of the observer can be calculated from
knowledge of A and c without consideration of whether the observer is to be combined with a feedback controller or not. This is the so-called separation property
of the observer-controller design procedure.

4.2 Design of state-feedback control systems

4.2.6

219

Example of a state-feedback control system

For a 2nd-order plant, described by


 

  
 
x 1 (t)
1
0
x1 (t)
8
8
=
+
u(t) +
z(t)
0 0.5 x2 (t)
8
8
x 2 (t)


 x1 (t)
y(t) = 2 1
x2 (t)

and

with the disturbance

z(t) = (t 6)



5
x(0) =
5

and the initial condition

a state-feedback controller with and without observer is to be designed so that

the poles of the closed-loop system are s1 = s2 = 3. The eigenvalues of the
observer are given as s1 = s2 = 8.
The plant is unstable and has the eigenvalues at 1.0 and -0.5. The desired characteristic polynomial with the zeros s1 = s2 = 3 as the given poles is from
Eq. (4.2.29)
P (s) = (s + 3) (s + 3) = 9 + 6 s + s2 .
The controllability matrix is


S = b Ab =
and its inverse
S 1 =


0.0417
0.0833


8
8
8 4

0.0833
0.0833


.

The last row of the inverse is


tT
1 = 0.0833


0.0833 .

With this data, the feedback vector using Ackermanns formula Eq. (4.2.51) is
 


 




1
0
1
0
1
0
1
0
T
+6
+
f = 0.0833 0.0833 9
0 0.5
0 0.5
0 0.5 0 0.5



 16


0
= 0.0833 0.0833
= 1.3333 0.5208 .
0 6.25
The feedforward gain according to Eq. (4.2.19) is

220

4 State variable techniques

1
 
  

 1
 1 8
0
8 
2 1
1.3333 0.5208

0 0.5
8
8
1
=

1  


9.6667 4.1667
8
2 1
10.6667 3.6667
8
1

 
=

 0.4074 0.4630 8
2 1
1.1852 1.0741 8


v=

= 0.5625 .
For the observer design the dual system
 

  

1
0
1 (t)
2
1 (t)
=
+
y(t)
0 0.5 2 (t)
1
2 (t)
is used. The desired characteristic polynomial with the zeros s1 = s2 = 8 as the
given eigenvalues of the observer is from Eq. (4.2.29)
P (s) = (s + 8) (s + 8) = 64 + 16 s + s2 .
The controllability matrix of the dual system is




2
2
T
S= c A c =
1 0.5
and its inverse
S

0.1667
=
0.3333

0.6667
0.6667


.

The last row of the inverse is


tT
1 = 0.3333


0.6667 .

With this data, the feedback vector using Ackermanns formula Eq. (4.2.51) is
 


 




1
0
1
0
1
0
1
0
T
= 0.3333 0.6667 64
+ 16
+
g
0 0.5
0 0.5
0 0.5 0 0.5



 81


0
= 0.3333 0.6667
= 27 37.5 .
0 56.25
The observer system matrix is

 





1
0
27
53
27
T
=Ag
2 1 =
c =

A
0 0.5
37.5
75 37

4.2 Design of state-feedback control systems

221

and, nally, the following observer equation is obtained:

 

  


53 27 x
1 (t)
8
27
x
1 (t)
=
+
u(t) +
y(t) .
2 (t)
75 37 x
8
37.5
x
2 (t)

Figure 4.2.5 shows the time responses of the state-feedback control system with
and without observer. The responses are divided into three periods. The control
system is started at t = 0 s and until t = 3 s the decay of the observing error is
demonstrated. At t = 3 s one can see the behaviour of the system following the
change in the set point w(t). Then from t = 6 s the disturbance z(t) is active and
one can see the disturbance behaviour. The observer is started at t = 0 s with
zero initial conditions, whereas the plant state is not zero. The estimated states in
Figure 4.2.5c and 4.2.5d converge asymptotically to the real states and the value
of w(t) = 1 is reached by the controlled variable y(t) as shown in Figure 4.2.5b.
The control system follows the set-point change from from 1 to 4 applied at
t = 3 s. However, due to the proportional behaviour of the open-loop system the
controlled variable y(t) shows a large steady-state error after the disturbance is
applied at t = 6 s.
The control structure used does not show the desired static behaviour with a
vanishing control error under disturbances. Therefore, a state-feedback controller
with an integrator is required to cope with the disturbance problem. For the modied control structure according to Figure 4.2.2 one has to design a state-feedback
controller for the extended system according to Eqs. (4.2.24) and (4.2.25):

1
0
0
x1 (t)
8
0
x 1 (t)
x 2 (t) = 0 0.5 0 x2 (t) + 8 u(t) + 0 w(t) .
(t)

2 1 0
(t)
0
1
The initial condition is now

5
x(0) = 5 .
0

For the eigenvalues of the closed-loop system a third value must be given, as
the additional state (t) of the integrator is introduced. For simplicity the choice
s1 = s2 = s3 = 3 is taken and the desired characteristic polynomial is
P (s) = (s + 3) (s + 3) (s + 3) = 27 + 27 s + 9 s2 + s3 .
The controllability matrix is

S = b

A b

8
8
8

A2 b = 8 4 2
0 8 20

222

4 State variable techniques

and its inverse

S 1

0.0625
= 0.1042
0.0417

The last row of the inverse is


tT
1 = 0.0417

0.0625
0.1042
0.0417

0.0417

0.0313
0.0313 .
0.0625

0.0625 .

With this data, the feedback vector using Ackermanns formula Eq. (4.2.51) is

64
0
0


f T = 0.0417 0.0417 0.0625 0 15.625 0
74 22.75 27


= 1.9583 0.7708 1.6875 .
From this the feedback vector of the state-feedback is


f T = 1.9583 0.7708
and the feedforward gain
v = 1.6875 .
As the state of the integrator is known, the same observer can be used as in the
case without integrator.
Figure 4.2.6 shows the time responses of the same experiment as in Figure 4.2.5.
The responses in the rst and second period are similar, the amplitudes are
somewhat larger and due to the additional integrator the response on set-point
changes is slower, but there is no steady-state error on disturbances. This is for
the case with and without observer, though the observer shows a steady-state
error when the disturbance z(t) is active.
Demonstration Example 4.2.1
A virtual experiment using state-feedback control of a tank system
Demonstration Example 4.2.2
A virtual experiment using state-feedback control of a VTOL

4.2 Design of state-feedback control systems

u ( t)
4
2
0
-2
-4
-6
-8
-1 0

223

w it h o b s e r v e r

(a )

w it h o u t o b s e r v e r

w ( t) ,y ( t) ,
z '( t )

9
t

(b )

1 2

w it h o u t o b s e r v e r
w ( t)
8
4
0

z '( t )

w it h o b s e r v e r

-4
0

9
s

x 1 ( t ) , ^x 1 ( t )

(c )

^x 1 ( t )

1 0

5
0
w it h o u t o b s e r v e r
w it h o b s e r v e r

-5
-1 0
0

x 2 ( t ) , ^x 2 ( t )

(d )

^x 2 ( t )

1 0

0
w it h o u t o b s e r v e r
w it h o b s e r v e r

-1 0
-2 0
0

9
t

Figure 4.2.5. Time responses of the state-feedback control system without integrator,
(a) manipulated variable u(t),
(b) controlled value y(t), set point w(t) and disturbance z(t),
(c) state x1 (t) and its estimate x
1 (t),
(d) state x2 (t) and its estimate x
2 (t)

224

4 State variable techniques

u ( t)
4
2
0
-2
-4
-6
-8
-1 0

w it h o b s e r v e r

(a )

w it h o u t o b s e r v e r

w ( t) ,y ( t) ,
z '( t )

9
t

(b )

1 2

w it h o u t o b s e r v e r
w ( t)
8
4
0

z '( t )

w it h o b s e r v e r

-4
0

9
s

x 1 ( t ) , ^x 1 ( t )

(c )

^x 1 ( t )

1 0

5
0
w it h o u t o b s e r v e r
w it h o b s e r v e r

-5
-1 0
0

x 2 ( t ) , ^x 2 ( t )

(d )

^x 2 ( t )

1 0

0
w it h o u t o b s e r v e r
w it h o b s e r v e r

-1 0
-2 0
0

9
t

Figure 4.2.6. Time responses of the state-feedback control system with integrator,
(a) manipulated variable u(t),
(b) controlled value y(t), set point w(t) and disturbance z(t),
(c) state x1 (t) and its estimate x
1 (t),
(d) state x2 (t) and its estimate x
2 (t)

225

Fuzzy control systems

5.1

Introduction to fuzzy techniques

5.1.1

Crisp and fuzzy logic

Fuzzy logic forms a bridge between the two areas of qualitative and quantitative
modelling. Although the input-output mapping of such a model is integrated
into a system as a quantitative map, internally it can be considered as a set
of qualitative linguistic rules. Since the pioneering work of Zadeh in 1965 and
Mamdani in 1975, the models formed by fuzzy logic have been applied to many
varied types of information processing including control systems.
The term Fuzzy Logic is a misnomer. It implies that in some way the methodology
is vague or ill-dened. This is in fact far from the case. Fuzzy logic just evolved
from the need to model the type of of vague or ill-dened systems that are dicult
to handle using conventional binary valued logic, but the methodology itself is
based on mathematical theory.
We are all familiar with binary valued logic and set theory. An element belongs
to a set of all possible elements and given any specic subset, it can be said
accurately, whether that element is or is not a member of it. For example, a
person belongs to the set of all human beings, and given a specic subset, such
as all males, one can say whether or not each particular person (element) belongs
to this set. This is appealing since it seems to describe the way human reason.
Collecting many elements into sets allows to describe many occurrences with few
rules. For example, the simple statement
IF person is male AND a parent THEN person is a father
applies to many people across the world with complete precision. The rules are
formed using operators. Here, the intersection operator AND is used, which manipulates the sets.
Unfortunately, not everything can be described using binary valued sets. The
classications of persons into males and females is easy, but it is problematic to
classify them as being tall or not tall. The set of tall people is far more dicult
to dene, because there is no distinct cut-o point at which tall begins. This
is not a measurement problem, and measuring the height of all elements more

226

5 Fuzzy control systems

precisely is not helpful. Such a problem is often distorted so that it can be described using the well-known existing methodology. Here, one could simply select
a height, e.g. 1.80 m, at which the set tall begins, see Figure 5.1.1a. The output of
a reasoning system using this denition would not be smooth with respect to the
height of a person. A person of height 1.79 m would produce a dierent output
than a person of 1.81 m. In human reasoning this property is not observed and it
is also undesirable for reasoning systems that are part of a control system.
Fuzzy logic was suggested by Zadeh as a method for mimicking the ability of human reasoning using a small number of rules and still producing a smooth output
via a process of interpolation. It forms rules that are based upon multi-valued
logic and so introduced the concept of set membership. With fuzzy logic an element could partially belong to a set and this is represented by the set membership.
For example, a person of height 1.79 m would belong to both tall and not tall
sets with a particular degree of membership. As the height of a person increases
the membership grade within the tall set would increase whilst the membership
grade within the not tall set would decrease, see Figure 5.1.1b. The output of a
fuzzy reasoning system would produce similar results for similar inputs. Fuzzy
logic is simply the extension of conventional logic to the case where partial set
membership can exist, rule conditions can be satised partially and system outputs are calculated by interpolation and, therefore, have output smoothness over
the equivalent binary-valued rule base. This property is particularly relevant to
control systems.

(a )

g ra d e o f tru th
n o t t a ll

1 .8 0 m

t a ll

g ra d e o f tru th

h e ig h t x

n o t t a ll

(b )

t a ll

1 .8 0 m

h e ig h t x

Figure 5.1.1. The dierence between the grade of truth in (a) binary valued logic {0,1}
and (b) fuzzy logic [0,1]

A fuzzy logic control system is one that has at least one system component that
uses fuzzy logic for its internal knowledge representation. Although it is possible
for fuzzy systems to communicate information using fuzzy sets, most applications
have a single fuzzy system component communicating with conventional system
components via deterministic values. In this case, and also in this chapter, fuzzy
logic is used purely for internal knowledge representation and, externally, can be
considered as any other system component.
Demonstration Example 5.1.1
Tumbler full or empty ?

5.1 Introduction to fuzzy techniques

227

Demonstration Example 5.1.2

The fuzzy set of full and empty tumblers

5.1.2

Why use fuzzy logic for control ?

Controlling a system means that some characteristics of this system are monitored, and, depending on the values of these characteristics, dierent controls are
applied. An algorithm that transforms sensor inputs into corresponding control
values is called a control strategy. The previous chapters deal with the traditional
approach of control systems design that consists of the following:
First, one tries to to describe the behaviour of the system in precise mathematical terms, i.e., one comes up with the exact model of the system.
Second, one tries to describe in precise terms what one wants to achieve.
One wants the control that is the best in the sense of some criterion.
Now that the controlled system is described in precise mathematical terms,
and the objective function is described in the same manner, it can be determined for each control strategy and for each initial state how exactly the
system will change and what the resulting value of the control will be. The
main goal is then to nd the control strategy for which the resulting value of
the objective function is the largest possible one. This is a well-dened mathematical optimisation problem, and traditional control theory has developed
many methods for solving this problem and designing the corresponding
control strategies.
Traditional control theory has many important applications. There are, however,
practical cases when this theory is not applicable. Indeed, to apply the traditional
control theory, one must
know the model of the controlled system,
know the objective function formulated in precise terms, and
be able to solve the corresponding mathematical design problem.
If one of these conditions is not satised, then traditional control methodology is
not applicable, as in the following cases:

228

5 Fuzzy control systems

Sometimes, the model and the objective function is known, but the design
problem cannot be solved. This is when the design problem is very complicated, time consuming or when the problem is new and algorithms for
solving it have not yet been developed. For example, parking a car is an example of a problem that traditional control theory has not considered until
recently.
Sometimes, the model is known, but the objective function is unknown. For
example, if a control system for a vehicle is designed, the intended goal is to
make the ride most comfortable, but there is no well-accepted formalism of
what comfortable means.
Sometimes, one does not even know the model of the controlled system. In
many practical applications one can in principle measure all the possible
variables and determine the model exactly, but this will increase the cost
drastically. In other practical situations, the main goal of the controlled
system is to explore the unknown, e.g., to control a rover over a terrain
of unknown type, or to control surgery instruments. In such situations, the
entire objective of the control is to learn as much about the system, and one
cannot have a precise model of this system before the control is over.
If traditional control methodology cannot be applied, how can one control? Often,
there is an additional expert knowledge available, for example, expert operators
who successfully control the desired system. Expert operators know how to operate a plant. Therefore it is desirable to extract the control rules from the expert
and use this knowledge in an automatic control system. At rst glance, the problem seems very simple. Since the person is a real expert, one simply ask her
multiple questions like suppose that x1 is equal to 1.2, x2 is equal to -2.7, ...,
what is u ? After asking all these questions, one will get many pattern, from
which one will be able to extrapolate the function f (x1 ,...,x2 ) using one of the
known methods. Alas, there are two problems with this idea:
There is a computational problem. Since one needs to ask a question for
each combination of sensor readings, one may end up having to ask too
many questions that takes years.
There is a more serious problem that makes it in most cases impossible
to implement. If one asks a car driver a question like you are driving at
80 km/h when a car which is 20 m in front of you slows down to 50 km/h,
for how many seconds do you hit the brakes?, nobody will give a precise
number.
An expert cannot usually express his knowledge in precise numerical terms, like
hit the brakes for 1.27 s, but he can formulate his knowledge by using words
from natural language. The knowledge, which one can extract from an expert

5.1 Introduction to fuzzy techniques

229

consists of statements like if the velocity is a little bit smaller than maximum,
hit the breaks for a while.
For the fuzzy control methodology one has to
know the experts control rules formulated by words from natural language
and
one wants to produce a precise control strategy.
The methodology that transform the informal expert control rules into a precise
control strategy is called fuzzy control. The idea was rst proposed by Zadeh,
and the methodology itself was rst proposed and applied by Mamdani. In this
chapter it is described exactly how this transformation is done.

5.1.3

Ideas of the fuzzy control methodology

Before coming to the details, the main ideas of fuzzy control methodology on
the example of a situation will be illustrated in which everyone feels himself an
expert. One of the most widely used control systems is the simplest rule-based
system imaginable, a thermostatic temperature controller. This rule-based system
operates with two rules:
(1) IF temperature is below set point THEN heat is on
(2) IF temperature is above set point THEN heat is o
The success of this controller is due to the combination of the properties, that it
is simple, robust and does not require a complex process model. The model is:
when the heat is on, the temperature rises slowly, and when the heat is o, the
temperature falls slowly.
The two IF-THEN clauses above can also be formally rewritten as
IF x is Ar THEN u is Br

for r = 1,2

In the thermostat example there is only one input variable (linguistic variable),
the temperature x. In the general case, there are several input variables x1 ,...,xn ,
so, in addition to the logical connective IF-THEN, another logical connective is
needed, AND. Then the IF-THEN clauses are
IF x1 is Ar1 AND x2 is Ar2 ... AND xn is Arn THEN u is Br .

230

5 Fuzzy control systems

Here, r = 1,...,R is the rule number, and Ari and Br are words from natural
language (linguistic terms), like below set point, on, small, large, approximately 1.5, etc. If the standard mathematical notation for IF-THEN and
AND is used, the above rules can be re-formulated as follows:
Ar1 (x1 ) Ar2 (x2 )... Arn (xn ) Br (u) ,

(5.1.1)

where r = 1,...,R. The set of rules is usually called a rule base. The left-hand
side of a rule is called premise, the right-hand side conclusion and the rule itself
implication.
The general idea is to represent the rule base in a computer. It has a clear
structure. A rule base consists of rules and each rule, in its turn, is obtained
from properties expressed by linguistic variables and terms and using logical
connectives. In view of this structure, it is reasonable to represent the rule base
by rst representing the basic elements of the rule base, premises and conclusions,
and then by extending this representation to the rule base as a whole. It makes
sense to use the following steps in the methodology:
1. Representation of the basic properties Ari (xi ) and Br (u).
2. Representation of the logical connectives.
3. The representations of the basic properties and of the logical connectives is
used to get the representations of all the rules.
4. Combination of the representations of dierent rules into a representation of
a rule base.
As a result of these four steps, one obtains an expert system that can give advise
for a specialist, who has to make a decision. In control, a system to automatically
make a decision based on its own conclusions is wanted. Therefore, for control
situations, a fth follow-up step is needed:
5. Based on facts for xi and on the rule base a reasoning procedure makes a
decision.
In the following section from the very beginning it is described how these ve
steps are implemented.

5.2 Basics of fuzzy sets

5.2
5.2.1

231

Basics of fuzzy sets

Fuzzy sets

In the classical set theory a set can be represented by enumerating all its elements
using
A = {a1 ,a2 ,a3 , ,an } .
If these elements ai (i = 1, ,n) of A are together a subset of the universal base
set X , the set A can be represented for all elements x X by its characteristic
function

1 if x A
(5.2.1)
A (x) =
0 otherwise .
In classical set theory A (x) has only the values 0 (false) and 1 (true), so
two values of truth. Such sets are also called crisp sets.
Non-crisp sets are called fuzzy sets, for which also a characteristic function can be
dened. This function is a generalisations of that in Eq.(5.2.1) and called a membership function. The membership of a fuzzy set is described by this membership
function A (x) of A, which associates to each element x0 X a grade of membership A (x0 ). In contrast to classical set theory a membership function A (x) of
a fuzzy set can have in the normalised closed interval [0,1] an arbitrary grade of
truth, see introductory section 5.1 and Figure 5.1.1. Therefore, each membership
function maps elements of a given universal base set X , which is itself a crisp set,
into real numbers in [0,1]. The notation for the membership function A (x) of a
fuzzy set A
A : X [0,1]
(5.2.2)
is used. Each fuzzy set is completely and uniquely dened by one particular membership function. Consequently symbols of membership functions are also used
as labels of the associated fuzzy sets. That is, each fuzzy set and the associated
membership function are denoted by the same capital letter. Since crisp sets and
the associated characteristic functions may be viewed, respectively, as special
cases of fuzzy sets and membership functions, the same notation is used for crisp
sets as well, see Figure 5.2.1
The base set X is introduced rst above as a universal set. In practical applications, physical or similar quantities are considered that are dened in some
interval. When such quantities are described by sets, a base sets can be generalised seamless to a crisp base set X that exists in a dened interval. This is a
generalisation of fuzzy sets.
Base sets are not always crisp sets. Another generalisation is that the base set
is itself a fuzzy set. This is necessary for multi-dimensional fuzzy sets, which are
discussed later in this chapter.

232

5 Fuzzy control systems

m
m

1 .0
0 .5

( x )
C

m F ( x )

0
x

Figure 5.2.1. Membership functions of a crisp set C and a fuzzy set F

5.2.2

Membership functions

The membership function A (x) describes the membership of the elements x of

the base set X in the fuzzy set A, whereby for A (x) a large class of functions
can be taken. Reasonable functions are often piecewise linear functions, such as
triangular or trapezoidal functions.
The grade of membership A (x0 ) of a membership function A (x) describes for
the special element x = x0 , to which grade it belongs to the fuzzy set A. This
value is in the unit interval [0,1]. Of course, x0 can simultaneously belong to
another fuzzy set B, such that B (x0 ) characterises the grade of membership of
x0 to B. This case is shown in Figure 5.2.2.
m
m

1 .0
m

0 .5
m
0

m
A

( x 0)

( x 0)
x
0

Figure 5.2.2. Membership grades of x0 in the sets A and B: A (x0 ) = 0.75 and
B (x0 ) = 0.25

Demonstration Example 5.2.1

Colour as a fuzzy set
Interactive Questions 5.2.1
Test yourself here

5.2 Basics of fuzzy sets

233

In the following, a set of important properties and characteristics of fuzzy sets

will be described.
Having two fuzzy sets A and B based on X , then both are equal if their
membership functions are equal, i.e.
A = B A (x) = B (x),

xX .

(5.2.3)

The universal set U is dened as

U (x) = 1,

xX .

(5.2.4)

The height of a fuzzy set A is the largest membership grade obtained by any
element in that set, i.e.
hgt(A) = sup A (x) .

(5.2.5)

xX

A fuzzy set A is called normal when hgt(A) = 1, and it is subnormal when

hgt(A) < 1.
The support of a fuzzy set A is the crisp set that contains all the elements
of X that have nonzero membership grades in A, i.e.
supp(A) = {x X | A (x) > 0} .

(5.2.6)

An illustration is shown in Figure 5.2.3.

1

c o re

0
b o u n d a r y
a

b o u n d a r y

s u p p o rt

Figure 5.2.3. Some characteristics of a membership function

The core of a normal fuzzy set A is the crisp set that contains all the elements
of X that have the membership grades of one in A, i.e.
core(A) = {x X | A (x) = 1} .

(5.2.7)

234

5 Fuzzy control systems

The boundary is the crisp set that contains all the elements of X that have
the membership grades of 0 < A (x) < 1 in A, i.e.
bnd(A) = {x X | 0 < A (x) < 1} .

(5.2.8)

Having two fuzzy sets A and B based on X , then both are similar if
core(A) = core(B) and supp(A) = supp(B).

(5.2.9)

If the support of a normal fuzzy set consists of a single element x0 of X ,

which has the property
supp(A) = core(A) = {x0 } ,

(5.2.10)

this set is called a singleton.

The type of representation of the membership function depends on the base
set. If this set consists of many values, or is the base set a continuum, then a
parametric representation is appropriate. For that functions are used that can be
adapted by changing the parameters. Piecewise linear membership functions are
preferred, because of their simplicity and eciency with respect to computability.
Mostly these are trapezoidal or triangular functions, which are dened by four and
three parameters, respectively. Figure 5.2.3 shows a trapezoidal function formally
described by

0,
x < a, x > d

xa , a x b
,
(5.2.11)
(x,a,b,c,d) = ba

1,
b
<
x
<
c

dx
dc , c x d
which migrates for the case b = c into a triangular membership function. For
some applications the modelling requires continuously dierentiable curves and
therefore smooth transitions, which the trapezoids do not have. Here, for example,
three of these functions are mentioned, which are shown in Figure 5.2.4. These
m
1

(a )

m
1

(b )
1

(c )

T
0

s
z

0
z

z
2

0
z

Figure 5.2.4. Membership functions with smooth transitions (Eqs.(5.2.12) to (5.2.14))

are

5.2 Basics of fuzzy sets

235

the normalised Gaussian function (Figure 5.2.4a)

2

(x)
(x,,) = e 22 ,

(5.2.12)

the dierence of two sigmoidal functions (Figure 5.2.4b)


1 
1
1 + e2 (x2 )
(x,1 ,1 ,2 ,2 ) = 1 + e1 (x1 )

(5.2.13)

and
the generalised bell function (Figure 5.2.4c)
 1

 x 2

.
(x,,,) = 1 + 



(5.2.14)

Interactive Questions 5.2.2

What are linguistic terms?

5.2.3

Elementary operators for fuzzy sets

The basic connective operations in classical set theory are those of intersection,
union and complement. These operations on characteristic functions can be generalised to fuzzy sets in more than one way. However, one particular generalisation,
which results in operations that are usually referred to us as standard fuzzy set
operations, has a special signicance in fuzzy set theory. In the following, only
the standard operations are introduced. The following operations can be dened:
The fuzzy intersection operator (fuzzy AND connective) applied to two
fuzzy sets A and B with the membership functions A (x) and B (x) is
AB (x) = min{A (x),B (x)},

xX .

(5.2.15)

The fuzzy union operator (fuzzy OR connective) applied to two fuzzy sets
A and B with the membership functions A (x) and B (x) is
AB (x) = max{A (x),B (x)},

xX .

(5.2.16)

The fuzzy complement (fuzzy NOT operation) applied to the fuzzy set A
with the membership function A (x) is
A (x) = 1 A (x),

xX .

(5.2.17)

236

5 Fuzzy control systems

Whilst the operations according to Eqs. (5.2.15) and (5.2.16) are based on min/max
operations, the complement is an algebraic one. Union and intersection can also
be dened in an algebraic manner but giving dierent results as:
The fuzzy intersection operator (fuzzy AND connective) can be represented
as the algebraic product of two fuzzy sets A and B, which is dened as the
multiplication of their membership functions:
AB (x) = A (x) B (x),

xX .

(5.2.18)

The fuzzy union operator (fuzzy OR connective) can be represented as the

algebraic sum of two fuzzy sets A and B, which is dened as:
AB (x) = A (x) + B (x) A (x) B (x),

xX .

(5.2.19)

Demonstration Example 5.2.2

What is Fuzzy AND ?
Demonstration Example 5.2.3
What is Fuzzy OR ?
Demonstration Example 5.2.4
What is Fuzzy NOT ?
The standard connective operations for fuzzy sets are now dened. As one can
easily see, these operations perform precisely as the corresponding operations for
crisp sets when the range of membership grades is restricted to the set {0,1}.
That is, the standard fuzzy operations are generalisations of the corresponding
classical set operations. However, they are not the only possible generalisation.
As shown above, the fuzzy intersection, union and complement are not unique
operations, contrary to their crisp counterparts. Dierent functions may be appropriate to represent these operations in dierent contexts. The capability to
determine appropriate membership functions and meaningful fuzzy operations in
the context of each particular application is crucial for making fuzzy set theory
practically useful.
When fuzzy operators are later applied within more complex fuzzy logic operations for rules and fuzzy reasoning, one has to take care of the right combinations
of fuzzy operations. For example, in classical set theory, the operations of intersection and union are dual with respect to the complement in the sense that they
satisfy the De Morgan laws
AB =AB

and

AB = AB .

It is desirable that this duality be satised for fuzzy sets as well. Other combinations need equivalences for commutativity, associativity and distributivity. From
Table 5.2.3 the type of operations can be determined for which operations are
valid. Only distributivity is not given in the arithmetic case. Therefore, one has
to be careful in applications where arithmetic operations are performed.

5.2 Basics of fuzzy sets

237

Table 5.2.1 Validity of fuzzy equivalences

commutativity
associativity
distributivity
De Morgan

max/min

arithmetic

AB =BA
AB =BA

A (B C) = (A B) C
A (B C) = (A B) C

A (B C) = (A B) (A C)
A (B C) = (A B) (A C)

(A B) = A B
(A B) = A B

Interactive Questions 5.2.3

What do you know from fuzzy operators ?
Interactive Questions 5.2.4
The fuzzy NOT properties

5.2.4

Fuzzy relations

In the introduction to the fuzzy control methodology, section 5.1.3, rules have
been introduced, which in mathematical notation are connective operations over
fuzzy sets. For example, the operations on premises in Eq.(5.1.1) can be handled
for each rule already by the elementary standard operators introduced in section 5.2.3. The means are now available to handle steps 1 and 2 of the methodology. But to cope with step 3 something more is needed to complete the modelling
of rules. That is now added in this section.
First, relations are explained by a simple example from daily life using discrete
fuzzy sets. Let us describe the relationship between the colour of a fruit x and
the grade of maturity y and characterise the linguistic variable colour by a crisp
set X with three linguistic terms as
X = { green, yellow, red } ,
and similarly the grade of maturity as
Y = { verdant, half-mature, mature } .
One knows that a crisp formulation of a relation X Y between the two crisp
sets would look like this in tabular form:

238

5 Fuzzy control systems

verdant half-mature
green
yellow
red

1
0
0

0
1
0

mature
0
0
1

The zeros and ones describe the grade of membership to this relation. This relation is now a new kind of crisp set that is built from the two crisp base sets X
and Y. This new set is now called R and can be expressed also by the rules:
(1) IF the colour is green THEN the fruit is verdant
(2) IF the colour is yellow THEN the fruit is half-mature
(3) IF the colour is red THEN the fruit is mature
As can be seen from this example, a relation, which is called a rule or rule base,
can be used to provide a model.
Demonstration Example 5.2.5
Test this with your own tomatoes
This crisp relation R represents the presence or absence of association, interaction
or interconnection between the elements of these two sets. This can be generalised
to allow for various degrees of strength of association or interaction between
elements. Degrees of association can be represented by membership grades in a
fuzzy relation in the same way as degrees of the set membership are represented
in a fuzzy set. Applying this to the fruit example, the table can be modied to
verdant half-mature
green
yellow
red

1
0.3
0

0.5
1
0.2

mature
0
0.4
1

where there are now real numbers in [0,1]. This table represents a fuzzy relation
and models the connectives in a fuzzy rule base. It is a two-dimensional fuzzy set
and the question now is, how can this set be determined from its elements.
Demonstration Example 5.2.6
Test this with your own strawberries
In order to do this, the elements are generalised. In the above example, the linguistic terms where treated as crisp terms. For example, when one represents the
colours on a colour spectrum scale, the colours would be described by their spectral distribution curves that can be interpreted as membership functions and then
a particular colour is a fuzzy term. Treating also the grades of maturity as fuzzy
terms, the above relation is a two-dimensional fuzzy set over two fuzzy sets. For

5.2 Basics of fuzzy sets

239

example, taking from the fruit example the relation between the linguistic terms
red and mature, and represent them by the membership functions as shown in
Figure 5.2.5a, a fruit can be characterised by the property red AND mature. This
m
A

0 .5

(a )

r e d
)

m
B

m a tu r e

0 .5

c o lo u r x

m a t u r it y y

(b )

(c )
m
1

0 .8

0 .6

0 .8
0 .6

0 .4

0 .4

0 .2

0 .2

y
u r it y
m a t

y
u r it y
m a t

Figure 5.2.5. Relation between two fuzzy sets: (a) membership functions, (b) 3-D view
of the membership functions, (c) membership function of the relation after applying the
min operation to (b)

expression can be re-written in mathematical form using elementary connective

operators (see Eqs. (5.2.15) or (5.2.18)) for the membership functions by
R (x,y) = min{A (x),B (y)}

(5.2.20)

R (x,y) = A (x) B (y) .

(5.2.21)

or
Figure 5.2.5b shows a 3-dimensional view of these two fuzzy terms and Figure 5.2.5c the result of the connective operation according to Eq.(5.2.20). This
result combines the two fuzzy sets by an operation that is a Cartesian product
R : X Y [0,1] .

(5.2.22)

From this example it is obvious that the connective operation in a rule for the
operation is simply performed by a fuzzy intersection in two dimensions. For
this, both intersection operators from Eqs. (5.2.15) or (5.2.18) can be used.
Combining rules into a rule base the example from above may help when it is
rewritten as

240

5 Fuzzy control systems

(1) IF the colour is green THEN the fruit is verdant
OR
(2) IF the colour is yellow THEN the fruit is half-mature
OR
(3) IF the colour is red THEN the fruit is mature

which describes in a linguistic way a union of three rules. For the complete rule
base R one can combine the relations formed for each individual rule with a fuzzy
union operator, which is the fuzzy OR according to Eqs. (5.2.16) or (5.2.20).
Now, step 4 of the methodology introduced in section 5.1.3 can be specied by
taking the rule base from Eq. (5.1.1) and applying the union operator by writing
the rule base with max/min operators as follows:
R (x1 ,x2 , . . . ,xn ,u) = max{min{Pr (x1 ,x2 , . . . ,xn ),Br (u)}} ,
r

(5.2.23)

where Pr (x1 ,x2 , . . . ,xn ) is the premise of the rth rule. This representation is
the standard max/min representation of a rule base that will be later used for
fuzzy controllers. Instead of the max/min representation a so called max-prod
representation is also usual, where the algebraic product
R (x1 ,x2 , . . . ,xn ,u) = max{Pr (x1 ,x2 , . . . ,xn ) Br (u)}
r

(5.2.24)

is used to build the relation between the premise and the conclusion.

5.2.5

Fuzzy composition

In order to make step 5 of the methodology introduced in section 5.1.3, a reasoning

method is needed. Explaining reasoning by the fruit example, it is assumed that
one has a crisp fact: a green fruit. The decision from the rule base is obvious: the
fruit is verdant, and this is similar for the other facts: yellow and red. But if one
has a fact like: the fruit is orange, one does not know how to determine which rule
res the decision and what the decision is. In the following, a new fuzzy operation
type is introduced, that allows to operate with a given fact and a fuzzy relation
to produce an output that represents the decision in a fuzzy way. This operation
is called fuzzy reasoning, which is a special case of the more general operation
called fuzzy composition.
Two relations of the form given in Eq. (5.2.22)
R : X Y [0,1]
S : Y Z [0,1] ,
can be composed to one relation

(5.2.25)

5.2 Basics of fuzzy sets

241
T : X Z [0,1] .

(5.2.26)

This process is known as composition and, using the max and min operators for
union and intersection, one can express the composition operation T = R S by
the corresponding membership functions as follows:
T (x,z) = max{min{R (x,y),S (y,z)}} .
yY

(5.2.27)

When one takes the above fruit example again with the colour-maturity relation
R
R

verdant

half-mature

mature

green
yellow
red

1
0.3
0

0.5
1
0.2

0
0.4
1

and dene for S a maturity-taste relation

S

sour tasteless sweet

verdant
half-mature
mature

1
0.7
0

0.2
1
0.7

0
0.3
1

then by applying Eq. (5.2.27) to the elements of these two tables, the following
is obtained:
T =R S
green
yellow
red

sour tasteless sweet

1
0.7
0.2

0.5
1
0.7

0.3
0.4
1

When the fuzzy set S is now interpreted as a rule base and the fuzzy set R as
a fact obtained from some measurement data, then the fuzzy set T is the result
from the reasoning process, which is in this case a relation.
Demonstration Example 5.2.7
Before you continue taste your fruits
In the following, rst, a one-dimensional crisp fact is taken. Dene the fruit colour
green as a fact by the singleton
C
= { 1

0} ,

242

5 Fuzzy control systems

where the numbers are the intensity grades of the colours green, yellow and red.
When one calculates the composition T
= C
R by applying Eq. (5.2.27),
where in this case the rst operand has only one dimension, the fuzzy set for the
maturity
T
= { 1 0.5 0 }
is obtained. The result is obvious from the rst rule of the rule base R. When a
dierent colour is taken than included in the rule base entries, say orange as
C
= { 0

0.5

0.5 } ,

then there is no problem to obtain the value for the maturity

T
= { 0.3

0.5

0.5 }

by applying the composition formula. The reasoning process is now solved.

In the same manner as relations can be composed, the one-dimensional facts
can be composed with the rule base to realise the reasoning operation. This can
now be precisely re-formulated for the general case of a rule base according to
Eq. (5.1.1).
If for the rule base
R : X Y [0,1]
its membership function is described by Eq. (5.2.23) or (5.2.24) and if there is a
fact described by the fuzzy set
A
: X [0,1]
and its membership function A (x), the result
B
= A
R : Y [0,1]
of the fuzzy reasoning is represented by the membership function
B (y) = max{min{A (x),R (x,y)}} .
xX

5.3

(5.2.28)

Fuzzy systems

In the previous section, elementary fuzzy terms and fuzzy logic operations have
been introduced. In this section, the application to the treatment of rule-based
knowledge follows. For this a rule-based fuzzy system is needed, containing a rule
base and a reasoning algorithm, which is used to process crisp or fuzzy input
values xi , i = 1, . . . ,n to a crisp or fuzzy output value y, see Figure 5.3.1. Using

5.3 Fuzzy systems

243
x 1
x 2

fu z z y
sy ste m

x n
Figure 5.3.1. Rule-based fuzzy system with n inputs and one output

multiple inputs and one output implies no restriction as a multi-input-multioutput fuzzy system can always be decomposed into multiple systems according to
Figure 5.3.1. Such systems are the basis for the realisation of fuzzy controllers. As
there are mostly crisp input values xi from measurements and for controllers only
a crisp output y, a fuzzy system must contain additional components, fuzzication
and defuzzication.

5.3.1

Fuzzication

The fuzzication comprises the process of transforming crisp values into grades of
membership for linguistic terms of fuzzy sets. The membership function is used
to associate a grade to each linguistic term.
Example 5.3.1
For the fuzzication of the car speed value x0 = 70 km/h the two membership
functions A and B from Figure 5.3.2 can be used, which characterise a low and
a medium speed fuzzy set, respectively. The given speed value of x0 = 70 km/h
belongs with a grade of A (x0 ) = 0.75 to the fuzzy set low and with a grade


of B (x0 ) = 0.25 to the fuzzy set medium.
Demonstration Example 5.3.1
Fuzzication example

5.3.2

Fuzzy inference machine

The core section of a fuzzy system is that part, which combines the facts obtained
from the fuzzication with the rule base and conducts the fuzzy reasoning process.
This is called a fuzzy inference machine. Here rule and composition operations
are applied from sections 5.2.4 and 5.2.5.

244

5 Fuzzy control systems

m

lo w

1 .0 0

m e d iu m
A

m
B

0 .7 5
0 .5 0
0 .2 5
0

2 0

4 0

6 0
8 0 1 0 0
x 0= 7 0 k m / h

1 2 0

1 4 0 k m / h x

Figure 5.3.2. Fuzzication of a car speed

In the following, for simplicity it is assumed that there is only one input x1 = x
and the rule base is described with max/min operators from Eq. (5.2.23) by
R (x,u) = max{min{Pr (x),Br (u)}} ,
r

(5.3.1)

then inserting Eq. (5.3.1) into Eq. (5.2.28) yields

B (u) = max{min{A (x), max{min{Pr (x),Br (u)}}}} .
xX

The max operations can be reordered such that only the relevant operands are
on the right-hand side. Then
B (u) = max{min{max{min{A (x),Pr (x)}} ,Br (u)}}
r
xX


Hr

(5.3.2)

= max{min{Hr ,Br (u)}}

r

is obtained for the reasoning process. The inner term Hr , which combines the
fact with the premise, is a constant and is called degree of relevance of the rule
r. It characterises the relevance of the red rule r and can be treated as a denormalised universal fuzzy set. The following example will help to interpret and
provide understanding of this reasoning operation by applying graphical means.
Example 5.3.2
A simple fuzzy system is given, which models the brake behaviour of a car driver
depending on the car speed. The inference machine should determine the brake
force for a given car speed. The speed is specied by the two linguistic terms
low and medium, and the brake force by moderate and strong. The rule
base includes the two rules

5.3 Fuzzy systems

245

(1) IF the car speed is low THEN the brake force is moderate
(2) IF the car speed is medium THEN the brake force is strong
For the linguistic terms in the premises speed is low and speed is medium
those from Figure 5.3.2 are used. The terms force is moderate and force is
strong are dened in Figure 5.3.3. For these two rules the membership functions
m

m o d e r a t e

1 .0 0

s t r o n g
C

m
D

0 .7 5
0 .5 0
0 .2 5
0

4 0

5 0

6 0

7 0

8 0

9 0

1 0 0
%

Figure 5.3.3. Fuzzy set of a car brake force

for the premises are P1 = A , P2 = B , and for the conclusions B1 = C ,

B2 = D . The car speed is x0 = 70 km/h, which is a crisp value and therefore can
be represented in fuzzy notation by the singleton A (x) = {70 km/h}. Inserting
these values into Eq. (5.3.2), one obtains
B (u) = max{min{H1 , C (u)}, min{H2 , D (u)}}
with
H1 = max{min{A (x),A (x)}}
xX

and
H2 = max{min{A (x),B (x)}} .
xX

In rule 1, there is only a relation between A and C . The degree of relevance

H1 for rule 1 is a horizontal line of height 0.75, which is the value from the
fuzzication as shown in Figure 5.3.4a. The reasoning operation min{H1 ,C (u)}
with rule 1 cuts the membership function C (u) by this line (yellow area). The
rule 2 is evaluated analogously, as shown in Figure 5.3.4b. Figure 5.3.4c shows
the nal result from the union operation over all yellow areas from the inferences
obtained with the individual rules.


Demonstration Example 5.3.2
An inference example with one rule
Demonstration Example 5.3.3
An inference example with more rules

246

5 Fuzzy control systems

m

lo w

1 .0 0
0 .7 5
m

0 .5 0

A '

m o d e r a te

(a )
C

R u le 1

0 .2 5
0

2 0

4 0

6 0

8 0

m
1 .0 0
m

1 0 0

1 2 0

m e d iu m
A '

4 0

1 4 0 k m / h x

5 0

6 0

7 0

8 0

m
m

9 0

s tr o n g

0 .7 5

1 0 0
%

(b )
m
D

R u le 2

0 .5 0
H

0 .2 5
0

2 0

4 0

6 0

8 0 1 0 0
x 0= 7 0 k m / h

1 2 0

1 4 0 k m / h x

4 0

5 0

6 0

7 0

8 0

9 0

1 0 0

(c )
m

4 0

5 0

6 0

7 0

B '

8 0

9 0

1 0 0
%

Figure 5.3.4. Fuzzy inference example: (a) inference with rule 1, (b) with rule 2, and
(c) nal fuzzy set

The inference shown above is based on the max/min representation of a rule

base from Eq. (5.2.23). Taking alternatively the max-prod representation from
Eq. (5.2.24)
B (u) = max{max{min{A (x),Pr (x)}} Br (u)}
r
xX


Hr

(5.3.3)

= max{Hr Br (u)}
r

is obtained for the max-prod inference. The dierence between the max/min
and max-prod inference is that in the rst case the membership function of the
conclusion is cut and in the second case scaled. Figure 5.3.5 illustrates this.

5.3 Fuzzy systems

m

247
m

(a )
1

(b )

0
u

0
u

Figure 5.3.5. Membership function of a conclusion using (a) max/min and (b) maxprod inference

Example 5.3.3
For a fuzzy system with the two inputs x1 and x2 and the output u the inference
of the two fuzzy rules
(1) IF (x1 = P) AND (x2 = M) THEN (u = M)
(2) IF (x1 = N) OR (x2 = S) THEN (u = S) ,
should be evaluated. The linguistic term S stands for small, M for medium,
L for large, N for negative and P for positive.
Each rule contains two premises, which are dierently connected. In rule 1 the
connective operation is the intersection, which can be performed by the min
operation according to Eq. (5.2.15) for A11 (x1 ) and A12 (x2 ), and in rule 2
the premise is a union of the two premises (x1 = N) and (x2 = S), which can
be performed by the max operation according to Eq. (5.2.16) for A21 (x1 ) and
A22 (x2 ). In the rst case the degree of relevance is H1 = A12 (x2 ) and in the
second case H2 = A22 (x2 ). The membership functions of the conclusion of each
rule will be determined using the degree of relevance of the corresponding rule
by applying either the max/min inference method according to Eq. (5.3.2) or the
max-prod inference method according to Eq. (5.3.3). The reasoning process using
both inference methods is visualised in Figure 5.3.6.

5.3.3

Defuzzication

As a result of applying the previous steps, one obtains a fuzzy set B (u) from
the reasoning process that describes, for each possible value u, how reasonable
it is to use this particular value. In other words, for every possible value u,
one gets a grade of membership that describes to what extent this value u is
reasonable to use. Using a fuzzy system as a controller, one wants to transform
this fuzzy information into a single value u
that will actually be applied. This

A 2 1

A 1 1

x 2
0 .0

0 .5

0 .8
x 1

p r e m is e 1
P

1 x

A 2 2

A 1 2

1 x

M I
m in
N

1 x

M
m A aX x

s u p e r p o s it io n o f
r u le 1 a n d 2

p r e m is e 2

B 2

B 1

1 u

1 u

1 u

m a x / m in in fe r e n c e
m

B 2

B 1

m a x -p r o d in fe r e n c e

Figure 5.3.6. Example of the application with two premises with (a) max/min inference and (b) max-prod inference

r u le 2

r u le 1

1 u

1 u

1 u

248
5 Fuzzy control systems

5.3 Fuzzy systems

249

transformation from a fuzzy set to a crisp number is called a defuzzication. It is

not a unique operation as dierent approaches are possible. The most important
ones for control are described in the following.

5.3.3.1

Centre of gravity method (COG)

This approach has its origin in the idea to select a value u
that, on average,
would lead to the smallest error in the sense of a criterion. If u
is chosen, and the
best value is u, then the error is u
u. Thus, to determine u
the least squares
method can be used. As weights for each square (u
u)2 , one can take the grade
of membership B (u) with which u is a reasonable value. As a result one has to
nd




B (u) (u
u)2 du .
(5.3.4)
u = arg min

u

Dierentiating with respect to the unknown u
and equating the derivative to

zero, the formula
/
u B (u)du


(5.3.5)
u = /U

U B (u)du
is obtained, which determines the value of the abscissa of the centre of gravity of
the area below the membership function B (u).

5.3.3.2

Centre of singleton method (COS)

The defuzzication can be strongly simplied if the membership functions Br (u)

of the conclusions are singly defuzzied for each rule such that each function is
reduced to a singleton that has the position u
r of the individual membership
functions centre of gravity. The centre of singletons is calculated by using the
degree of relevance as follows:




R ur Hr
.
(5.3.6)
u = 
R Hr
The simplication consists in that the singletons can be determined already during the design of the fuzzy system and that the membership function B (u) with
its complicated geometry is no longer needed. The defuzzication using this formula is an approximation of the defuzzication by Eq. (5.3.5). Experiences from
control show that there are slight dierences between both approaches, which can
be in most cases neglected.

250

5 Fuzzy control systems

5.3.3.3

Maximum methods

This class of methods determines u
by selecting the membership function with

the maximum value. If the maximum is a range, either the lower, upper or the
middle value is taken for u
depending on the method. Using these methods, the
rule with the maximum activity always determines the value, and therefore they
show discontinuous and step output on continuous input. This is the reason why
these types of method are not attractive for use in controllers.

5.3.3.4

Margin properties of the centroid methods

As the centre of gravity of the area below the membership functions cannot reach
the margins of U, the membership functions, which are at the margins, must be
symmetrically expanded when obtaining the centre of gravity. This is necessary
in order to have the full range of U available. This is shown in Figure 5.3.7. The
same expansion is also necessary for the COS method.
m

(a )
1

0
0

1 0 0 u

(b )

1 0 0 u

Figure 5.3.7. Margin of B
(u) (a) original and (b) expanded

Demonstration Example 5.3.4

Properties of the defuzzication methods

5.3.4

The Takagi-Sugeno fuzzy system

In the rule bases described hitherto with the IF-THEN rules of this chapter fuzzy
sets both in the premises and in the conclusions are used. This kind of inference
is called Mamdani inference. A modied inference scheme, developed by Takagi
and Sugeno, represents the conclusions by functions. A rule of this form will be
IF x1 is Ar1 AND x2 is Ar2 ... AND xn is Arn THEN u = fr (x1 ,x2 , . . . ,xn ) .

5.3 Fuzzy systems

251

The structure of the premises are the same as for the Mamdani inference. However, in the conclusion all linguistic terms Br are substituted by the functions fr ,
and therefore it is not necessary to dene a priori linguistic terms Br (u) for the
conclusions as in Eq. (5.1.1). The function fr represents a direct mapping from
the input space X1 X2 . . . Xn with the input values x1 ,x2 , . . . ,xn to the
output space U.
The connective operation in a rule is in this case performed via the degree of
relevance Hr of the premise of the rule Rr and the function fr in the conclusion.
The nal output is determined as a weighted mean value over all R rules according
to

Hr fr (x1 ,x2 , . . . ,xn )



.
(5.3.7)
u = R
R Hr
The eort of performing a defuzzication is saved, as the crisp value u
is directly
determined by the inference operation and this makes this method attractive.
The Takagi-Sugeno fuzzy system builds an overall combination of functions fr ,
which are valid in some range. If the membership functions of the fuzzy sets
in the premises are overlapping, the transition between the functions is always
continuous. For the special case of linear functions
fr (x1 ,x2 , . . . ,xn ) =

cr x

(5.3.8)

=1

the coecients cr can be determined by some identication procedure.

5.3.5

The components of a fuzzy system

In the introduction to this fuzzy system section, in Figure 5.3.1 a fuzzy system
is drawn as a black box with some inputs and an output. Figure 5.3.8 shows the
contents of a fuzzy system. Now it is clear, what is in this black box. The input
signals combined to the vector x = [x1 , x2 , . . . ,xq ]T are crisp values, which are
transformed into fuzzy sets in the fuzzication block as discussed in section 5.3.1.
The output u comes out directly from the defuzzication block, which transforms
an output fuzzy set back to a crisp value using methods from section 5.3.3. The
set of membership functions responsible for the transforming part and the rule
base as the relational part contain as a whole the modelling information about
the system, which is processed by the inference machine from section 5.3.2. This
rule-based fuzzy system is the basis of a fuzzy controller, which is described in
the following section.

252

5 Fuzzy control systems

fu z z y sy ste m
r u le b a s e

fu z z ific a t io n

in fe r e n c e

d e fu z z ific a t io n
u

m e m b e r s h ip fu n c t io n s

Figure 5.3.8. Components of a fuzzy system

5.4
5.4.1

Fuzzy control
Basic structure of a fuzzy controller

A fuzzy controller can be handled as a system that transmits information like a

conventional controller with inputs containing information about the plant to be
controlled and an output that is the manipulated variable. From outside, there is
no vague information visible, both, the input and output values are crisp values.
The input values of a fuzzy controller consist of measured values from the plant
that are either plant output values or plant states, or control errors derived from
the set-point values and the controlled variables.
A control law represented in the form of a fuzzy system is a static control law.
This means that the fuzzy rule-based representation of a fuzzy controller does not
include any dynamics, which makes a fuzzy controller a static transfer element,
like the standard state-feedback controller. In addition to this, a fuzzy controller
is in general a xed nonlinear static transfer element, which is due to those
computational steps of its computational structure that have nonlinear properties.
In what follows the computational structure of a fuzzy controller will be described
by presenting the computational steps involved. The computational structure of
a fuzzy controller consists of three main steps as illustrated by the three blocks
in Figure 5.4.1:
1. signal conditioning and ltering at the input (input lter),
2. fuzzy system according to Figure 5.3.8, and

5.4 Fuzzy control

253

fu z z y c o n t r o lle r
L

in p u t
filt e r

fu z z y
sy ste m

o u tp u t
filt e r

Figure 5.4.1. Basic structure of a fuzzy controller

3. signal conditioning and ltering at the output (output lter).

The input and output lters are for signal conditioning. The external input signals
v must be scaled such that they can be fed as signals x into the fuzzication part
of the fuzzy system. In many cases, the signals in v are the control error e and
its derivative e.
In this case the input lter contains a dierentiating element.
Also other dynamical elements can be in the input lter, e.g. integrators for the
control error. Additionally auxiliary signals from the plant measurements may be
used that represent plant states or disturbances acting on the plant. The design
of this input lter depends on the application, which will be illustrated later by
an example.
The fuzzy system contains the control strategy and consists of those components
already discussed in section 5.3.5. For example, a linguistic formulation of a proportional control strategy would be expressed by the following rules of the fuzzy
system:
(1) IF (control error positive)
(2) IF (control error zero)
(3) IF (control error negative)

THEN (manipulated variable positive),

THEN (manipulated variable zero),
THEN (manipulated variable negative).

A proper rule base can be found either by asking experts or by evaluation of

measurement data using data mining methods.
The output lter is for the adaptation of the crisp output c from the fuzzy system to the manipulated variable u of the plant. In principle, there are many
dynamical and static operations possible. Often, the output c of the fuzzy system
describes an increment of the manipulated variable, and thus an integration of
this increment must occur.

5.4.2

Transfer behaviour of fuzzy controllers

As already stated, fuzzy systems describe the static behaviour of a fuzzy control
strategy. Therefore it is obvious that the transfer behaviour of such a system

254

5 Fuzzy control systems

can be represented by nonlinear characteristics. This provides a ne opportunity

to compare fuzzy control strategies amongst themselves and with conventional
nonlinear ones.

5.4.2.1

Representation using 2D characteristics

To provide a rst approach for the design of membership functions for a fuzzy
controller component, some prototype membership functions are introduced. For
such prototypes, linguistic terms are introduced. For example, the following seven
terms
NL
NM
NS
AZ

negative large
negative medium
negative small
approximately zero

PL positive large
PM positive medium
PS positive small

can be used to characterise the triangular shaped fuzzy sets according to Figure 5.4.2. It is important to recognise that the fuzzy sets dened in this gure
m ( x )
N L

-1

N M

N S

A Z
1

P S

P M

P L

1
x

Figure 5.4.2. Prototype membership functions for a fuzzy set with seven linguistic
terms

and the seven linguistic terms are only a reasonable example. For various reasons,
emerging from specic applications, other shapes of membership functions might
be used over the given ranges. Moreover, dierent fuzzy sets may be dened for
dierent variables. The prototype membership functions are usually chosen only
as a preliminary candidate. They may later be modied by the designer.
For a fuzzy system using this kind of prototype with membership functions which
overlaps, both for the premise and the conclusion the transfer characteristic is
nonlinear, as shown in the following example.
Example 5.4.1
For a proportional fuzzy controller with the control error e = w y as input,
with the manipulated variable u as output, with the rule base

5.4 Fuzzy control

(1)
(2)
(3)
(4)
(5)
(6)
(7)

IF
IF
IF
IF
IF
IF
IF

e=
e=
e=
e=
e=
e=
e=

255
NL
NM
NS
AZ
PS
PM
PL

THEN
THEN
THEN
THEN
THEN
THEN
THEN

u=
u=
u=
u=
u=
u=
u=

NL
NM
NS
AZ
PS
PM
PL

and with both membership functions of the form shown in Figure 5.4.2, the static
nonlinear characteristic u = u(e) is as shown in Figure 5.4.3 using the assumptions
given below.


m ( e )
N L

N M

N S

A Z

-1

P S

P M

P L

u
1

0
1

-1

- 1

N L

N M

N S

m (u )

A Z

P S

P M

P L

-1

Figure 5.4.3. Nonlinear characteristic of a fuzzy controller

256

5 Fuzzy control systems

5.4.2.2

Influence of the membership functions and rule base

on the characteristic

The above example shows that a fuzzy controller is a nonlinear controller. The
form of the characteristic depends only on the rule base and the membership
functions of e and u. In the following discussions about the inuence of membership functions the following assumptions for the fuzzy controller with the input
signal e and the output signal u will be used:
For the AND connectives the min and for the OR connectives the max
operator will be used.
The max/min inference will be used.
The defuzzication will be performed by the COG method with symmetrical
membership functions at the margins.
Input and output values are normalised to the interval [1,1], and at rst, only
the three linguistic terms NS (negative small), AZ (approximate zero) and PS
(positive small) are considered. The rule base is that of a proportional fuzzy
controller
(1) IF
(2) IF
(3) IF

e = NS
e = AZ
e = PS

THEN
THEN
THEN

u = NS,
u = AZ,
u = PS,

with the membership functions shown in Figure 5.4.4a and b. The static characteristic in Figure 5.4.4c is odd symmetrical about the origin due to the symmetry
of the membership functions. Because of the dierent supports of the membership
function for the fuzzy sets AZ of both functions, the characteristic is approximately piecewise linear and has three distinct levels. The membership functions
of the input e have two overlaps in the intervals [0.6, 0.4] and [0.4,0.6] that
correspond precisely with the ranges with the positive slope of the curve. The
reason for this is just that two rules in these ranges are simultaneously active. On
the other hand, in the non-overlapping ranges only one rule is active. The membership function of the output depends in this case only on the degree of relevance
and thus the centre of gravity of the membership function remains constant.
If the number of linguistic terms for the input and output is increased, the characteristic is similar, but with more sections. The number of sections depends only
on the number of linguistic terms and the width of the sections depends on the
degree of overlapping. In the special case without overlapping in the input one
obtains the characteristic of a three-level controller, as shown in Figure 5.4.5.
In this case only one rule is active such that only the three crisp values -1, 0

5.4 Fuzzy control

N S

257

(a )

m (e )

A Z
1

P S

N S

(b )

m (u )

A Z

(c )
u

P S

a b

- 1

-1
0

1 e

a b

-1

1
0

-1
u

Figure 5.4.4. Membership functions and static characteristic of the fuzzy controller

(a )

m (e )
N S

A Z
1

N S

P S

(b )

m (u )

A Z

P S

-1

-1

0
a

1
u

(c )
1

a
1

-1

Figure 5.4.5. Inuence of the (c) characteristic of a proportional fuzzy controller (a)
without overlapping in the input membership functions and (b) with full overlapping in
the output membership functions

and 1 are generated. Now, consider varying the degree of overlap of the output
membership functions. Figure 5.4.6 shows the case with full overlap on input and
output, where the result is approximately a linear behaviour.
A modication of the output membership functions so that they do not overlap
will cause the characteristic to become close to that of Figure 5.4.6, compare
Figure 5.4.7. Therefore one can establish the fact that the degree of overlap in the
input membership functions has a strong inuence on the static characteristic of a
fuzzy controller. While small overlaps in the input membership functions generate
step characteristics, with a higher degree of overlap the curves become smoother.
The inuence of overlap in the output membership functions has less eect on the
characteristic. For a reduction of the support of the output membership functions
the characteristic of Figure 5.4.8 is obtained which does not dier signicantly
from that of Figure 5.4.6.

258

5 Fuzzy control systems

The size of the individual output membership function has a strong inuence on
the characteristic. Figure 5.4.9 shows the case for a very small support of the
output membership function AZ, which generates an S-type characteristic with
a high gain at the origin. Widening the support of the membership function AZ
inverts the S-curve with a small gain at the origin, as shown in Figure 5.4.10. Thus
the form of the characteristic depends strongly on the support of the individual
output membership function.

N S

(a )

m (e )

P S

A Z
1

(b )

m (u )
N S 1

(c )
u

A Z P S

-1

-1
0

-1
e

1
0

e
-1

Figure 5.4.6. Inuence on the (c) characteristic of a proportional fuzzy controller with
(a) full overlap in the input membership functions and (b) full overlap in the output
membership functions

(a )

m (e )
N S

A Z
1

(b )

m (u )
P S

N S

A Z
1

(c )
u

P S

-1

-1
0

1
e

-1
0

1 u

1 e
-1

Figure 5.4.7. Inuence on the (c) characteristic of a proportional fuzzy controller with
(a) full overlap in the input membership functions and (b) without overlap in the output
membership functions

5.4 Fuzzy control

259

(a )

m (e )
N S

A Z
1

(b )

m (u )
N S

P S

P S

A Z
1

(c )
u
1

-1

1 e
0

-1

1
0

-1

1
e

-1
u

Figure 5.4.8. Inuence on the (c) characteristic of a proportional fuzzy controller with
(a) full overlap in the input membership functions and with (b) reduced support in the
output membership functions

(a )

m (e )
N S

A Z
1

P S

(b )

m (u )
N S

P S

A Z
1

(c )
u
1

-1

-1

-1

0
1

1 e
-1

Figure 5.4.9. Inuence on the (c) characteristic of a proportional fuzzy controller with
(a) full overlap in the input membership functions and with (b) a small support in the
output membership function AZ

The eects of a modied rule base will be demonstrated by an example. The

same full overlapping membership functions are used as in Figure 5.4.6a and b.
A modied rule base of the form
(1) IF
(2) IF
(3) IF

e = NS
e = AZ
e = PS

THEN
THEN
THEN

u = PS,
u = AZ,
u = NS,

will give a modulus-type of characteristic, as shown in Figure 5.4.11.

260

5 Fuzzy control systems

(a )

m (e )
N S

A Z
1

(b )

m (u )
N S

P S

A Z
1

(c )
u

P S

-1

-1

1 e
0

-1

1 e
-1

1 u
0

Figure 5.4.10. Inuence on the (c) characteristic of a proportional fuzzy controller

with (a) full overlap in the input membership functions and with (b) a large support in
the output membership function AZ

(a )

m (e )
N S

-1

A Z
1

m (u )
N S
P S
1 A Z

P S

1 e

-1

(b )

(c )

-1

1
e

Figure 5.4.11. Inuence on the rule base on the (c) characteristic of a proportional
fuzzy controller with (a) full overlap in the input membership functions and (b) full
overlap in the output membership functions

5.4.2.3

Representation using 3D characteristics

Up to now, fuzzy controllers with only one input and one output of the fuzzy
system have been considered. The same arguments with respect to the degrees
of freedom of a fuzzy system are also valid in the case of multiple inputs. A
graphical representation of the characteristic is not as easy as in the 2D cases
of section 5.4.2.2. Moreover, for the case of two inputs, a 3D representation is
possible. For a fuzzy controller with a fuzzy system having two inputs e1 and
e2 and one output u one gets a band of characteristics in a 2D discrete representation, or a 3D representation with the output over the two inputs, as shown
in Figure 5.4.12 for a PD-type of fuzzy controller. For the case with more than
two inputs, projections on the 3D space can be used to generate multiple 3D
diagrams, but in general these representations have only a limited usefulness.

5.4 Fuzzy control

261
e
w
+

(a )
e
1

fu z z y s y s te m

_
d

e
2

d t

p la n t

fu z z y
c o n t r o lle r

(b )
u
1
0

-1
1
e
1

0
-1

-1

1
2

Figure 5.4.12. Control system with PD-type fuzzy controller: (a) block diagram and
(b) 3D representation of the characteristics of the fuzzy system with e1 = e and e2 = e

A fuzzy controller is typically a classical controller using nonlinear characteristics.

But the design and parametrisation is entirely dierent.

5.4.3

Example of a fuzzy control system

In the following the design and functioning of a fuzzy control system will be presented using the example of the portal-type loading crane shown in Figure 5.4.13

5.4.3.1

Loading crane plant model

The schematic diagram in Figure 5.4.14 shows the principle of a loading crane,
which consists of a crab moving on rails. The load (freight) hangs on a rope from
the crab such that the rope and load together can be treated as a pendulum. The
load moves only in the plane that contains the direction of the rails. The force of
the electrical drive, which moves the crab, is proportional to the control signal,
fed into the drive system. The position of the crab and the rope (pendulum)

262

5 Fuzzy control systems

Figure 5.4.13. View of a portal-type loading crane

F u
M

y
m

Figure 5.4.14. Schematic representation of the loading crane

angle are measured. The real manipulated variable to control the drive is limited
to 10 V .
The plant dynamical behaviour can be described by the two coupled nonlinear
dierential equations
1
[m sin (g cos + l 2 ) + F u S sgn(yC ) D y C ],
M + m sin2
1

=
l (M + m sin2 )

yC =

(5.4.1)
(5.4.2)

[m sin (g + l 2 cos ) + M g sin + cos (F u S sgn(y C ) D y C )]

(5.4.3)
y = yC + l sin
with the following parameters and variables of the crane:
position of the load:
position of the crab:

y,
yC ,

5.4 Fuzzy control

pendulum angle:
mass of the crab:
mass of the load:
pendulum length:
acceleration due to gravity:
drive coecient:
static friction force:
dynamical friction coecient:
experimental initial position:
experimental target position:

263
,
M = 1000 kg,
m = 20 . . . 1250 kg,
l = 10 . . . 20 m,
g = 9.80665 m/s2 ,
F = 1000 N/V,
S = 500 N ,
D = 777 kg/s,
y0 = 9 m,
yT = +9 m .

The control task is to move the load from the initial position y0 to the target
position yT such that the load does not swing at the target position and the
transition goes smoothly with a minimum of oscillations and no overshoot.

5.4.3.2

Fuzzy control system design

In general, the design of a fuzzy controller is characterised by the fuzzy methodology as described above. For designing a fuzzy control system from scratch more
or less heuristic methods are available. If no expert or operator is available, one
cannot tackle the design problem without some information, typical a mathematical model, of the plant. The controller of the crane has been designed and tested
using simulation studies by the following generic procedure:
1. Identication of the relevant input and output variables of the controller,
i.e. choice of the linguistic variables,
2. setting of the possible ranges of the input and output values, i.e scaling of
the linguistic variables,
3. denition of meaningful linguistic terms and their membership functions for
each linguistic variable,
4. setting up the rule base, and
5. simulation of the closed loop if possible or testing at the plant site.
As the maximum number of rules R, considering all possible connective operations, increases strongly with the number of inputs n and membership functions
m according to
(5.4.4)
R = mn ,

264

5 Fuzzy control systems

one has to try to keep it low. Hence, as a rst approach a control structure
according to Figure 5.4.12a is taken, and for the linguistic variables the control
error e = w y and its derivative e are chosen, each with the three membership
functions N (negative), Z (zero) and P (positive) including the output. Thus
from Eq. (5.4.4) 9 rules for a full rule base have to be specied. The control
error is directly calculated as the dierence between the set-point value w and
the position of the load y, not of the crab. For determining the derivative e
a dierentiating lter may be used. As the set point is changed only in steps
when the pendulum has settled, the derivative e is equal to the derivative y.

Most crane systems measure the speed of the crab for the internal drive control.
Therefore it is reasonable to avoid the dierentiating lter and use the speed yC
of the crab instead. The input lter is now reduced to gains for scaling signals
such that they t to the range of the linguistic variables. In order to compensate
step disturbances, e.g. caused by static friction S, the output lter is provided
with an integrator in parallel to the proportional channel.
This relatively simple fuzzy control system shows a reasonable behaviour, as
shown in Figure 5.4.19a. The maximum amplitude of the oscillations of the load is
about 7 , which is still relatively large. The control performance can be improved
when additional information about the angle and its speed are exploited in
the new structure shown in Figure 5.4.15. As is not measured, an input lter
(DT1 element) is necessary to generate this signal. Now due to the doubling of
w
+
_

K
2

fu z z y
c o n t r o lle r

fu z z y
sy ste m

D T
1

u
K
P

K
s

y
c ra n e
j

.
C

Figure 5.4.15. Structure of the loading crane fuzzy control system

the number of inputs, the problem of an ecient rule base arises. According to
Eq. (5.4.4), 81 rules are now necessary if all inputs enter into all premises. It
is thus dicult to formulate such a large number of useful rules. Therefore one
can use a simpler approach of using two separate rule bases, the one for position
control that has already been used for the linguistic variables of e and yC and
the other for angle control for the linguistic variables of and .
Both rule bases
are linked by the fuzzy union operation. The maximum number of rules is now
reduced to 18 and can be further reduced to 14 by removing all unred rules.

5.4 Fuzzy control

265

Figure 5.4.16 shows the membership functions for the ve linguistic variables.
The rules are illustrated in tabular form in Figure 5.4.17. In Figure 5.4.18 two
3D characteristics of the fuzzy system are shown. The nonlinear behaviour can
be clearly recognised.
m
N

- 2

1
P

- 1

m
/ s
m

- 6
- 1 5
- 1 0

/ s
V

e
.

1
o

1 5

1 0

e
u

j
j

Figure 5.4.16. Triangular membership functions of the ve linguistic variables

.
N

Z
N
P

j
P

.
N

N
Z

Z
P

Z
Z

Figure 5.4.17. Rule bases for position and angle control

Figure 5.4.19 shows the results of the fuzzy control system when moving the
load from y0 to yT . The input signals are scaled such that they use almost the
full range of the linguistic variables. The scaling factors from Figure 5.4.15 are
K1 = 0.1, K2 = 0.5, K3 = 1 and K4 = 6. The parameters of the output lter
are KP = 2 and KI = 1.4. The additional angle feedback damps the oscillations to
about a half the number with pure position control, as shown in Figure 5.4.19b.
The control system shows robustness when the length of the rope is doubled as
shown in Figure 5.4.19c.

5.4.4

Contribution of fuzzy control

In many real-life situations, there is a need to automatically control a system such

as a car, a chemical reactor or a crane. In some situations, a reasonable model

266

5 Fuzzy control systems

(a )

(b )
6

1
4
2
0
0
-2

-1

-4
-6
1

e
0

-1

-2

-1

4
2

-2

-4

-6

-2

-1

Figure 5.4.18. 3D characteristics of the fuzzy system (a) u = u(e,e),

(b) u = u(e,)

of the controlled object and how it will react to dierent controls are known.
Also one can often describe precisely the objective of the control usually, to
maximise or minimise a certain characteristic such as the plants output, say in
response to a system disturbance can be dened. In such cases, the search for the
optimal control strategy can be reformulated as a precise mathematical problem.
Thus, in many real-life cases one can explicitly solve the optimisation problem
and thus nd the desired control.
In many other situations, however, one does not have a good description of the
controlled system, or a good model is available, but the corresponding optimisation problem is too dicult to solve. In such situations, one may have the
expertise of skilled operators who have the experience of controlling the system. For example, the experience of drivers who control cars, the experience of
chemical engineers who successfully control chemical plants and the experience
of crane operators who successfully operate their cranes. It may be appropriate
to transform this expert experience into an automatic control strategy.
It is often dicult to come up with such a transformation, because expert operators are often unable to describe their experience in precise terms. Instead they
describe their control by using words of natural language, which do not have a
precise meaning and are, in this sense, fuzzy. There is thus a need for a methodology that translates such fuzzy rules into a precise control strategy. Fuzzy control
is such a methodology.
Fuzzy control is successful in many real-life problems in which traditional control
methods fails or at least is not so successful. Does this mean that traditional
control as shown in the other chapters is the thing of the past and only fuzzy
methods should be used? Of course not, fuzzy control has its limitations too.
The main limitation of fuzzy control is that it is applicable only in the situation

5.4 Fuzzy control

267

of uncertainty, when there is the complete knowledge about the controlled system
not available. Fuzzy control is therefore good but not optimal. Often, as one gains
more and more experience of controlling the system, one gets a better and better
understanding of how the system works. Eventually, this understanding leads to
a precise description of the system, which allows to nd an optimal control, which
is better than any other control and in particular, which is better than a fuzzy
control strategy.
From this viewpoint, fuzzy control is a temporary phenomenon. This does not
mean when our knowledge grows that fuzzy control will be used less and less.
As one obtains more and more knowledge about the system that is controlled
for a long time, new systems and objects attract our attention, and one needs
to be able to control them. For example, the car manufacturer nds a precise
description of a certain type of motor and the engineers learn how to optimally
control the motor of this type, but when new improved motors appear, for these
new devices, the engineers do not have the exact model, and thus, they have
to use fuzzy control or similar techniques. As the progress intensies, more and
more new objects and systems appear that have to be controlled, and therefore,
the relative use of fuzzy control increases. Fuzzy control methodology has its
limitations, but it does not have limits.

268

5 Fuzzy control systems

y u j
C

1 0
o

8
6

u
4

(a )
y

2
0
-2

-4
-6
-8
-1 0

y u j
C

2 0

1 0

3 0

4 0

5 0

6 0

1 0
o

y
6

(b )
C

u
4

2
0

-2
-4
-6
-8
-1 0

y u j
C

2 0

1 0

3 0

4 0

5 0

6 0

1 0
o

y
6

u
4

(c )

2
0
-2

-4
-6
-8
-1 0
0

1 0

2 0

3 0

4 0

5 0

6 0

Figure 5.4.19. Responses of the portal-type loading crane with fuzzy control from
initial to target position for load m = 1000 kg: (a) without angle feedback, l = 10 m, (b)
with angle feedback, l = 10 m, (c) with angle feedback, l = 20 m

269

Mathematical and table appendix

A.1

The Laplace transform

A.1.1

Convergence

With respect to the range of convergence of the Laplace integral

F (s) = f (t) est dt,

(A.1.1)

now the following considerations are taken:

If the function to be transformed f (t) is stepwise continuous and if there are real
numbers and
such that for all t 0


|f (t)| < e t
is valid, then the Laplace integral will converge for all s with Re s >
. Particularly, if for
the smallest possible value 0 is taken, the condition Re s > 0 is
the smallest possible convergence range. Consequently the Laplace integral exists
only within some part of the complex s plane. This part is called convergence
area, as shown in Figure A.1.1. The variable 0 is called the abscissa of converjw
s p la n e

c o n v e rg e n c e a re a

Figure A.1.1. Convergence area of the Laplace integral

gence. For values of s with Re s > 0 Eq. (A.1.1) makes no sense. Thus for > 0
the limit value of f (t) et for t must go to zero, but not for < 0 .

270

A Mathematical and table appendix

Example A.1.1
f (t) = tn
The Laplace integral converges for all s with > 0, as et for t decreases
faster than any power of t increases.

A.1.2

The inverse Laplace transform

The so called back transformation or inverse Laplace transformation, i.e. the

determination of the original function from the mapped function, is given by the
inverse integral
c+j


1
F (s) est ds
t > 0,
(A.1.2)
f (t) =
2j
cj

where f (t) = 0 for t < 0 is valid. The variable c must be chosen such that the path
of integration is in the convergence area along a line parallel to the imaginary axis
at distance c from it, where c must be larger than the real parts of all singular
values of F (s).
It must be observed that Eq. (A.1.2) at a step location t = ts delivers the arithmetic mean value of the left and right limits [f (ts +) + f (ts )]/2, particularly for
t = 0 at the origin, as [f (0+) + f (0)]/2 = f (0+)/2.

A.1.3

Main theorems of the Laplace transform

A.1.3.1

Derivative theorem

For a causal function of time f (t), for which the derivative for t > 0 exists, and
take into account any steps at t = 0, the value of 0+ will be chosen for the lower
limit of integration for Eq. (A.1.1). This is necessary to eliminate the case t = 0
from the integration interval. This has no inuence on the value of the integral
as far as we restrict on classical functions (no distributions). So one obtains by
partial integration



d f (t)
dt

=

st

e
0+




 st
 
d f (t)
dt = e f (t)  + s est f (t) dt
dt
0+
0+

or


d f (t)
dt

= s F (s) f (0+).

(A.1.3)

A.1 The Laplace transform

271

In the case of multiple dierentiation it follows that




A.1.3.2

dn f (t)
d tn

= sn F (s)

i=1


(i1) f (t) 
d

sni

d t(i1) 

(A.1.4)

t=0+

Integral theorem

From


t


f ( ) d

t
=

f ( ) d est dt

one obtains by partial integration


t


f ( ) d

1
=
s

1
s


t

st

f ( ) d e

1
+
s

f (t) est dt

f (t) est dt


t


f ( ) d

1
F (s).
s

(A.1.5)

A.1.3.3

Convolution in the time domain

For the convolution of two functions of time f1 (t) and f2 (t)

t
f1 (t) f2 (t) =

f1 ( ) f2 (t ) d.

(A.1.6)

I can easily be shown by permutation of the variables that the convolution is a

symmetrical operation, so that
f1 (t) f2 (t) = f2 (t) f1 (t)
or

t

t
f1 ( ) f2 (t ) d =

f2 ( ) f1 (t ) d.
0

272

A Mathematical and table appendix

In the following it will be shown that the convolution of two original functions
corresponds to multiplication of the related mapped functions, that is
 {f1 (t) f2 (t)} = F1 (s) F2 (s).

(A.1.7)

The Laplace transform of Eq. (A.1.6) is given by


t
 {f1 (t) f2 (t)} = 

f1 ( ) f2 (t ) d
0

t

est f1 ( ) f2 (t ) d dt.

t=0 =0

Substituting = t and d = dt, respectively and using the valid extension of

the upper bounds of integration to yields



 {f1 (t) f2 (t)} =

es( +) f1 ( ) f2 () d d.

= =0

As both functions f1 (t) and f2 (t) have zero values for t < 0, it follows with respect
to the lower limit of integration that


 {f1 (t) f2 (t)} =

es f1 ( ) d

es f2 () d.

The right-hand side of this equation is just the product F1 (s)F2 (s).

A.1.3.4

Convolution in the frequency domain

Whereas in section A.1.3.3 the convolution of two functions of time was the focus
of interest, here the convolution of two functions in the frequency domain is of
concern and it can be shown that
1
 {f1 (t) f2 (t)} =
2j

c+j

F1 (p) F2 (s p) dp.

(A.1.8)

cj

Here F1 (s) f1 (t) and F2 (s) f2 (t) is valid. Furthermore, p is the complex
variable of integration. According to this theorem the Laplace transform of the
product of two functions of time is equal to the convolution of F1 (s) and F2 (s)
in the mapped domain.

A.1 The Laplace transform

273

For the product of two causal functions of time

f (t) = f1 (t) f2 (t)

(A.1.9)

with Laplace transforms F1 (s) and F2 (s) and areas of convergence Re s > 1 and
Re s > 2 , respectively, the expression


 {f (t)} = F (s) =

f1 (t) f2 (t) est dt

(A.1.10)

follows after taking the Laplace transform of f (t). Using the inverse integral
according to Eq. (A.1.2)
1
f1 (t) =
2j

c+j

F1 (p) ept dp

c > 1

(A.1.11)

cj

and by substituting this relationship into Eq. (A.1.10) it follows that

c+j

1
F1 (p) ept dp dt.
F (s) = f2 (t) est
2j
0

(A.1.12)

cj

Permuting the sequence of integration (as far as the integrals full the conditions
of convergence) one obtains
1
F (s) =
2j

c+j

F1 (p) dp

f2 (t) e(sp) t dt,

(A.1.13)

cj

where for the second integral one can make the substitution


F2 (s p) =

f2 (t) e(sp) t dt.

(A.1.14)

This integral converges for Re(s p) > 2 . By substituting Eq. (A.1.14) into
Eq. (A.1.13) the validity of Eq. (A.1.8) is shown.

A.1.3.5

Initial value theorem

It is required to show that

274

A Mathematical and table appendix

f (0+) = lim f (t) = lim s F (s).
s

t0+

One has that



(A.1.15)

f(t) = f(t) est dt = s F (s) f (0+)
0+

which as s can be written

f(t) est dt = lim [s F (s) f (0+)].
lim
s
0+

As the integration is independent of s, the calculation of the limit and the integration can be permuted provided that the integral converges uniformly. If  [f (t)]
exists, then
lim f(t) est = 0
s

is valid. Therefore one gets

lim s F (s) = f (0+).

A.1.3.6

Final value theorem

The nal value theorem states that

f () = lim f (t) = lim s F (s).
t

s0

(A.1.16)

To prove this one evaluates the limit

lim

s0
0+

f(t) est dt = lim [s F (s) f (0+)].

s0

Again one can permute the sequence of determining the limit and the integration
provided the integral converges. The result is


f(t) dt = lim [s F (s) f (0+)],
s0

0+

and after integration it follows that

f () f (0+) = lim [s F (s) f (0+)]
s0

f () = lim s F (s).
s0

A.2 The complex G-plane

A.2

275

The complex G-plane

The complex transfer function G(s) describes a local conformal mapping of the
s plane to the G plane. Because of the preservation of the angles in this transformation the orthogonal grid of lines parallel to the axes = const and = const
of the s plane will be mapped into an orthogonal, but warped net of curves in
the G plane, as shown in Figure A.2.1.

jw

s p la n e
w =

jIm [G ]

G p la n e

co n st
R e[G ]

s
s = co n st
s = co n st

w = co n st

Figure A.2.1. Local conformal mapping of the lines = const and = const of the s
plane into the G plane (generalised Nyquist plot)

This mapping property will be discussed in the following using the simple example
of a 1st-order transfer function
G(s) =

K
.
1 + sT

(A.2.1)

For s = + j one obtains from Eq. (A.2.1)

G( + j) =

1 + T jT
K
=K
.
1 + T + jT
(1 + T )2 + 2 T 2

From this it follows that for the real and imaginary parts of G(s)
Re{G(s)} = K

1 + T
,
(1 + T )2 + 2 T 2

(A.2.2a)

Im{G(s)} = K

T
.
(1 + T )2 + 2 T 2

(A.2.2b)

For mapping the following two cases are treated:

a) Mapping of the lines = const

276

A Mathematical and table appendix

From from Eqs. (A.2.2a) and (A.2.2b) one obtains
T = (1 + T )

Im{G(s)}
Re{G(s)}

(A.2.3)

and substituting Eq. (A.2.3) in Eq. (A.2.2a) shows, after a simple rearrangement, that

2
2

K
K
2
Re{G(s)}
+ Im {G(s)} =
.
(A.2.4)
2(1 + T )
2(1 + T )
This relationship represents for the variables Re{G(s)} and Im{G(s)} the
equation of a band of circles with the parameter . The centres of these
circles are on the real axis of the G plane at K/[2(1 + T )]. The radii are
K/[2(1 + T )]. For 0, K > 0 and T > 0 the lines = const map into
semicircles in the lower G plane, as Figure A.2.2 shows. The semicircles for
= const are parameterised by the values of . The circles start with = 0
on the real G axis and end for at the origin of the G plane.
G

jIm [G ]
w

w =
s >
0

p la n e

0
K

s <
0

w
=

R e [G ]

w
B

s =
s = -

w =

c o n s t

fo r w 0

Figure A.2.2. Conform mapping of the upper s plane ( > 0) into the G plane for the
example G(s) = K/(1 + sT )

A very important case is the semicircle with the parameter = 0. It represents the conformal mapping of the positive imaginary axis of the s plane and
is called the frequency response locus, G(j), for the system. This semicircle
starts for = 0 with a value of K on the positive real axis of the G plane
and it has for | Re{G(j)}| = | Im{G(j)}| the frequency = B = 1/T ,
which is also called the breakpoint frequency.
From Eq. (A.2.4) it is obvious, that for > 0 the radius of the semicircle
will decrease until it becomes zero for and the nal semicircle will
coincide with the origin of the G plane. For < 0 , however, the radius will
increase to innity for = 1/T , and the semicircle will degenerate to the
negative imaginary axis of the G plane.

A.3 Detailed analysis of 2nd-order lag elements

277

Mapping of the lines = const

If Eq. (A.2.3) is solved for T then


Re{G(s)}
T = 1 + T
Im{G(s)}

(A.2.5)

and using this in Eq. (A.2.2b), one obtains by elementary manipulations



K
Im{G(s)} +
2T

2

K
+ Re {G(s)} =
2T
2

2
.

(A.2.6)

This relationship also represents a band of circles, but for the parameter
. The centres of the circles for 0 are on the negative imaginary axis
at K/(2T ) and because of the radii of size K/(2T ) they pass to the
origin of the G plane. For = 0 the radius will be innite and the circle
degenerates to a line that is the real axis of the G plane. For the
radius shrinks to zero and the circle degenerates to the origin of the G plane.
It can be easily shown that both bands of circles according to Eqs. (A.2.4)
and (A.2.6) are orthogonal.
The transfer function G(s) = K/(1 + sT ) belongs to a special class of local
conformal mappings, the so called linear mappings. A mapping, described
by the equation G(s) = (As + B)/(Cs + D) always maps the circles in the
s plane to circles in the G plane. Here lines are treated as a special case of
circles. Introducing the complex G plane one gets for the special case = 0
of G(s) the frequency response locus G(j). The description of systems using
frequency responses G(j) in the G plane is of high importance for practical
applications as the frequency response is a directly measurable description
of a dynamical system.

A.3

Detailed analysis of 2nd-order lag elements

A.3.1

Determining resonances of 2nd-order lag elements

From Eq. (2.3.55) the maximum of the magnitude A()max = A(p ) = Mr and
the resonant peak frequency p can be simply determined. A() is at a maximum,
when the denominator of Eq. (2.3.55), i.e.

a() = 1

2 2

+ 2
0

2
,

is at minimum. Setting the derivative of 1st order to zero one obtains for = p

278

A Mathematical and table appendix


 2
p
da() 
= 0 = 1 +
+ 2 2 .

d p
0

From this the resonant peak frequency is


p = 0 1 2 2

for <

1
2

(A.3.1)

and the maximum of the amplitude for K = 1 is

Mr =

1
1 2

(A.3.2)

From Eq. (A.3.1)

it follows
that a maximum only exists for (1 2 2 ) > 0, which

gives < 1/ 2. For = 1/ 2 = 0.707, p = 0 is valid and Mr = 1 or Mr dB = 0.

For = 0, p = 0 and Mr = is valid.

A.3.2

Poles and step responses of 2nd-order lag elements

From the characteristic equation of the 2nd-order lag element

P (s) D(s) = 1 +

1
2
s + 2 s2 = 0
0
0

one obtains the poles of the transfer function


s1,2 = 0 0 2 1.

(A.3.3)

(A.3.4)

The oscillating behaviour of a 2nd-order lag element depends on the position of

the poles in the s plane and this is discussed in the following sections:
a) Case I: 0 < < 1 (oscillating behaviour: PT2 S element)
For this case Eq. (A.3.4) gives a conjugate complex pair of poles

s1,2 = 0 j0 1 D2 .

(A.3.5)

The corresponding step response is obtained from

H(s) =

1
K

1 2 s
1 + 2 s + 2s
0
0

(A.3.6)

which can be written

H(s) =

K02
.
(s s1 ) (s s2 ) s

(A.3.7)

A.3 Detailed analysis of 2nd-order lag elements

279

Writing in partial fractions and nding the inverse Laplace transform gives




 

 

t0
sin 0 1 2 t
(t).
cos 0 1 2 t + 
h(t) = K 1 e
1 2
(A.3.8)
The decay of the oscillations is inuenced by the quantity TA = 1/(0 ),
which is therefore called the decay time constant. From the position of the
poles s1 and s2 of G(s) the time constant TA can be found. The ratio


 Re(si ) 


for i = 1 or 2,
= 
Im(si ) 
or
= 

1 2

= tan d .

(A.3.9)

is dened as the relative damping ratio of the oscillation.

If a certain value of is given then the angle d between a line from the
origin to a complex pole and the imaginary axis in the s plane is xed. Also,
from the damped oscillation of the step response h(t), which has the damped
natural frequency of

(A.3.10)
d = 0 1 2 < 0 ,
then from Eq. (A.3.8) the ratio of a consecutive overshoot to overshoot or
vice versa (see Table 2.3.2)
hn+ 1/2
hn

=e

1 2

can be determined, and from this one obtains the damping ratio as
ln h hn 1

=

n+ 2

2 .
hn

2 + ln hn+1/2
b) Case 2: = 1 (critical damping: PT2 element)
For Eq. (A.3.4) the two poles of G(s) are
s1,2 = 0 .
This is a double pole on the negative real axis. If a time constant
T =

1
0,
0

(A.3.11)

280

A Mathematical and table appendix

is dened, one obtains the transfer function as
G(s) =

K
,
(1 + T s) (1 + T s)

(A.3.12)

which is a series connection of two 1st-order lag elements with identical time
constants. The step response follows from Eq. (A.3.8) with = 1 as


(A.3.13)
h(t) = K 1 et0 (1 + 0 t) (t).
c) Case 3: > 1 (aperiodic behaviour: PT2 element)
In this case from Eq. (A.3.4) G(s) has two negative real poles

s1,2 = 0 0 2 1.
With the denition of the time constants
1
1
and T2 =
T1 =
s1
s2
one obtains the transfer function as
G(s) =

K
.
(1 + T1 s) (1 + T2 s)

(A.3.14)

Here the system is composed of two PT1 elements with dierent time constants in series connection. This element also shows typical PT2 behaviour.
During the calculation of the step response h(t) according to Eq. (A.3.8)
the arguments of sin and cos will be complex. By applying the hyperbolic
functions
cos jx = cosh x and sin jx = j sinh x
one obtains directly using Eq. (A.3.8)



 


 

sinh 0 2 1t
(t).
h(t) = K 1 et0 cosh 0 2 1t + 
2 1
(A.3.15)
d) Case 4: = 0 (undamped behaviour: oscillating element)
Eq. (A.3.4) gives for this case a pair of imaginary poles of G(s) at
s1,2 = j0 .
With Eq. (2.3.53) and = 0 one obtains the transfer function
G(s) =

2
K
= K 2 0 2.
1
0 + s
1 + 2 s2
0

(A.3.16)

The step response h(t) is an undamped oscillation

h(t) = K(1 cos 0 t) (t)
of frequency 0 , which is usually called the natural frequency.

(A.3.17)

A.4 The law of Bode and the Hilbert transformation

281

e) Case 5: < 0 (unstable element)

In this case both poles of G(s) are in the right half s plane and may be real
or conjugate complex:

(A.3.18)
s1,2 = 0 || 0 2 1.
Eq. (A.3.8) is also valid for this case, but the exponential term has now a
positive sign. Therefore the oscillations will not decay. The amplitudes of the
oscillations in h(t) will increase exponentially with time. This behaviour, for
which h(t) for t increases indenitely, is dened as unstable.

A.4

The law of Bode and the Hilbert transformation

For a given amplitude response A() there exists only one minimum phase system
that realises this amplitude response. The phase () of it can be determined for
discrete frequencies according to the law of Bode


2
ln A() ln A( )
d.
(A.4.1)
( ) =

2 2
0

The connection between the real part R() = Re{G(j)} and the imaginary part
I() = Im{G(j)} of the frequency response, which is valid both for minimum
phase and non-minimum phase systems, is given by the Hilbert transformation


2
I()
d
(A.4.2a)
R( ) = R()

2 2
0

and
2
I( ) =

0

R()
d.
2 2

(A.4.2b)

From this it is obvious that with knowledge of only the real or the imaginary
part, the imaginary part or the real part and therefore the frequency response
G(j) can always be completely reconstructed.

A.5

Stability considerations using the weighting function

As a transfer function G(s) is the Laplace transformed weighting function g(t),

the stability condition according to Eq. (2.4.1) can also be reformulated in terms

282

A Mathematical and table appendix

of G(s). In order to show this, the inverse Laplace transform from section 2.1.4
is used. If G(s) is given as a rational fraction
G(s) =

N (s)
N (s)
=
,
D(s)
a0 + a1 s + . . . + an sn

(A.5.1)

and sk = k + jk are the poles of the transfer function G(s), i.e. the roots of the
denominator polynomial
D(s) = an (s s1 ) (s s2 ) . . . (s sn ) =

ai si ,

(A.5.2)

i=0

then according to Eq. (2.1.21) and Eq. (2.1.24) the weighting function
g(t) =

gj (t)

(A.5.3)

j=1

consists of n terms
gj (t) = cj t esi t , = 0,1,2, . . . , j = 1,2, . . . ,, i = 1,2, . . . ,n.
In general cj is a complex constant, and for multiple poles si of multiplicity r the
exponent is = r 1 > 0. For the modulus of this function one obtains
|gj (t)| = |cj t esi t | = |cj | t ei t .
For i < 0 the exponential function decays to zero for t , and therefore so
also does |gj (t)|, even if > 0, because the exponential function decays faster to
zero than any other nite power of t increases.
By this consideration it is obvious, that Eq. (2.4.1) is only valid, if all poles of
G(s) have a negative real part. If the real part of only one pole is positive or of
a multiple pole is zero, the weighting function grows with t beyond all limits.

A.6

Equivalence of the Hurwitz and Routh criteria

The validity of the Routh criterion can be veried easily by the equivalence with
the Hurwitz criterion. From the coecients of the rst row of the Routh schema
the connection with the Hurwitz determinants can be seen directly:
D1 = an1
D2 = an1 bn1 = D1 bn1
D3 = an1 bn1 cn1 = D2 cn1
..
.
Dn = an1 bn1 cn1 . . . dn1 en1 fn1 gn1 = Dn1 gn1

A.7 Determination of JISE using determinants

283

The coecients bn1 ,cn1 . . . in the rst column of the Routh schema are just the
quotients of consecutive Hurwitz determinants. When all Hurwitz determinants
are positive, then their quotients and therefore also the coecients in the rst
column of the Routh schema are positive. When the coecients of the Routh
schema are positive, then also, as an1 = D1 , all Hurwitz determinants are positive. Thus, the Routh criterion is equivalent to the Hurwitz criterion.

A.7

Determination of JISE using determinants

0
an

..

a3

a1

an

an2

an4

a2
.

an3

..

..

an1

..

..

..
.

0
..
.

..

..

a1
..
.

a3
..
.

..

..

..
.

0
..
.
..
.

an3
..
.

..

an1

C even
n

..

and for even n

..

Writing the quadratic performance indices in terms of determinants is a general

algebraic approach that is more suitable for a detailed analysis. Eq. (3.3.5) can
be rewritten as
(1)n+1 det C bn
.
(A.7.1)
JISE =
2an det C n
For odd n the matrix C n is

an
an2
a3
a1
0
0
.
..
..
..
..
..
..

.
.
.
.
.
. ..
0

.
an2
a3 a1 0

an
.

(A.7.2)
=
C odd
n

0
an1 an3 . . . a2 a0
..

.
..
a2 a0 0
an1 an3

..
0
.
a2
a0
0
0
an1 an3 . . .

an2

an4

a2

a0

0
..
.

0
.
a0

..
.
0

(A.7.3)

Both matrices consist in the upper part of a Toeplitz matrix with the odd coecients and in the lower part of a Hankel matrix with the even coecients of the
denominator polynomial A(s) of the control error in Eq. (3.3.4). The determinant
of this matrix is nothing other than the Hurwitz determinant from Eq. (2.4.6).
The other determinant in the numerator of Eq. (A.7.1) is from the same matrix
as C n , but the coecients in the last row in Eq. (A.7.2) and in the rst row in
Eq. (A.7.3) are exchanged for those of the polynomial

284

A Mathematical and table appendix

C(s) = B(s)B(s) = c0 + c1 s2 + + cn1 s2(n1) .

(A.7.4)

Example A.7.1
Example 3.3.1 of determining the best damping ratio for a second-order system
is rewritten using determinants. For the control error from Eq. (3.3.9) we have
the polynomial C(s) from Eq. (A.7.4) as
C(s) = 4 2 02 s2
and from Eqs. (3.3.9) and (A.7.3) follows


1 4 2 02
det
1
02
1
1 + 4 2

 =
.
JISE =
2
40
20 0
det
1
02
As this function is a parabola in , with minimum given by
4 2 1
dJISE
=
=0
d
4 2 0
the minimum square error to a step input occurs for = 0.5 .

The procedure shown in this example can be generalised for all quadratic performance indices. If the control error can be written as
E(s) =

N (s)
,
A(s) + kB(s)

(A.7.5)

where k is a feedback gain to be chosen or for which the optimal value in the
sense of a criterion according to Eq. (3.3.3) must be found, the performance index
can be represented as
JISE =

(1)n+1 det(C + kD)

.
2(an + kbn ) det(A + kB)

(A.7.6)

The matrices A and B correspond to the matrix given by Eq. (A.7.2) or (A.7.3),
respectively, and C to A where the coecients in the upper or lower row are
exchanged for those of the polynomial
C(s) = N (s)N (s) = c0 + c1 s2 + + cn1 s2(n1)

(A.7.7)

and D to B where the coecients in the upper or lower row are zeros. The
determinants in Eq. (A.7.6) are polynomials in k. Therefore the performance
index can be rewritten as
Q(k)
(A.7.8)
JISE =
P (k)

A.7 Determination of JISE using determinants

or
1+

1 Q(k)
= 0.
JISE P (k)

285

(A.7.9)

As Eq. (A.7.9) corresponds directly to the characteristic equation (2.5.5a), the

root-locus method can be applied for the analysis of the performance index in
the k plane.
As we are interested in real values of k, only the behaviour on the real axis of the
complex k plane is needed. The branches are calibrated by the inverse of JISE .
The denominator in Eq. (A.7.6) or (A.7.8) is the highest Hurwitz determinant.
Therefore the real poles (real zeros of P(k)) represent critical values of k, where
the closed loop has poles on the imaginary axis and where JISE . One can
take the stability margins directly from the branches on the real axis in the k
plane.
Now inspecting real breakaway or break-in points. If we apply rule 7 from section 2.5.2 on Eq. (A.7.9), from Eq. (2.5.15) we have the condition
d Q(k)
= 0,
dk P (k)

(A.7.10)

d
JISE = 0 for an extremum of the performance
which is the same as the condition dk
index. Therefore real breakaway or break-in points in the k plane represent a
minimum or maximum in the performance index and can be used to nd optimal
adjustments of parameters.

Example A.7.2
Consider the following plant transfer function:
GP (s) =

4 3s 17s3 + s4
4 + 17s + 28s2 + 21s3 + 7s4 + s5

and the controller transfer function

GC (s) = KC

1 + (1 + TI )s + 2TI s2
,
TI s(1 + s)

which contains two free parameters, the gain KC and the time constant TI . The
control system should be optimised on step changes in the command or in the
plant input using the performance index JISE .
For both inputs the control error can by described by Eq. (A.7.5) for k = KC ,
where it diers only in the numerator polynomial N (s). While determining the
optimal gain by evaluation of the condition in Eq. (A.7.10) the other parameter
TI is held constant. In the same manner the stability margins can be determined
by using P (k) = 0. For constant values of JISE one gets the gains according to
Eq. (A.7.9).

286

A Mathematical and table appendix

Scanning over a given range of TI and plotting the results will generate a combined
stability and performance diagram as shown in Figure A.7.1. From these diagrams
one can see that the optimal controller structure does not dier signicantly for
the two system inputs.


T
I

3
2 .5

J
2
T

1 .5
1
0 .5

Io p t

= 1 .9 7

d J

IS E

/ d K
C

IS E

1 8 1 0

(a )
6

= 0
J

IS E o p t

= 5 .5 5 3 7

u n s t a b le

-0 .5
-1
-1 .5
-2
T

K
-0 .4

-0 .2

0 .2

= 0 .3 6 1 9

C o p t

0 .4

0 .6

0 .8
K

3
J

2 .5
2
T

Io p t

IS E

1 8 1 2

(b )

7 .5

= 2 .1 7

1 .5
1

d J

IS E

/ d K
C

= 0
J

0 .5
0

IS E o p t

= 7 .2 2 4 5

u n s t a b le

-0 .5
-1
-1 .5
-2

K
-0 .4

-0 .2

0 .2

C o p t

0 .4

= 0 .3 3 6 5
0 .6

0 .8 K
C

Figure A.7.1. Stability and performance diagram for step changes (a) in the command
and (b) plant input

A.8 Tables

A.8

287

Tables

Table A.8.1 Step responses for a given pair of complex poles and a real pole with
multiplicity k according to Eq. (3.6.4)

k =

1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

1
k =
0 .6
0 .8
1
2

4
w

6
t

k =

2
w

4
0

1 0

5
k =
0 .6
1
2
3
4
6 8 1 0

k =

1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0
1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

2
k =
0 .6
1
2
3

4
w

6
t

k =

2
w

4
0

1 0

k =
0 .6
1
2
3
4
6 8 1 0

k =

1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0
1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

k =
0 .6
1
2
3
0

6
t

1 0

7
k =
0 .6
1
2
3

2
w

4
0

jw
k w

k m u lt ip le
0

4
w

k =

D is t r ib u t io n o f p o le s

- 5 w

- w
0

1 0

k =

1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0
1 .2
1 .0
0 .8
0 .6
0 .4
0 .2
0 .0

2
w

k =
0 .6
1
2
3
4
6 8 1 0
4

t
0

k =

2
w

s p la n e
0

8
k =
0 .6
1
2
3
4
6 8 1 0

s6 +

06

s + 20 s + 202 s + 03
s4 + 2.60 s3 + 3.402 s2 + 2.603 s + 04
s5 + 3.240 s4 + 5.2402 s3 + 5.2403 s2 + 3.2404 s + 05
3.860 s5 + 7.4602 s4 + 9.1403 s3 + 7.4604 s2 + 3.8605 s

s + 0
s + 1.40 s + 02

s3 + 30 s2 + 302 s + 03

s + 40 s + 602 s2 + 403 s + 04
s5 + 50 s4 + 1002 s3 + 1003 s2 + 504 s + 05
s6 + 60 s5 + 1502 s4 + 2003 s3 + 1504 s2 + 605 s +

s + 0
s + 20 s + 02

Standard polynomials (s) for u = 1, 2,. . .

+ 06

0 .0 0

0 .2 5

0 .5 0

1 .0 0 u =
0 .7 5

h W
1 .2 5

0 .0 0

0 .2 5

0 .5 0

0 .7 5

1 .0 0

h W
1 .2 5
u =

2
4

1
3
4

5
6

1 2

1 6

Step response hW (0 t)

w 0 t 2 0

w 0t 1 0

Table A.8.2 Standard polynomials (s) of Eq. (3.6.2) and corresponding step responses of control systems of dierent order u

Binomial form

Butterworth form

288
A Mathematical and table appendix

JITAE form

t5% settling time form

s + 0
2

s + 1.40 s + 02

06

+ 06

s + 1.550 s + 2.102 s + 03
s4 + 1.60 s3 + 3.1502 s2 + 2.4503 s + 04
s5 + 1.5750 s4 + 4.0502 s3 + 4.103 s2 + 3.02504 s + 05
s6 + 1.450 s5 + 5.102 s4 + 5.303 s3 + 6.2504 s2 + 3.42505 s +

s6 +

s + 1.750 s + 2.1502 s + 03
s4 + 2.10 s3 + 3.402 s2 + 2.703 s + 04
s5 + 2.80 s4 + 502 s3 + 5.503 s2 + 3.404 s + 05
3.250 s5 + 6.602 s4 + 8.603 s3 + 7.4504 s2 + 3.9505 s

s + 0
s + 1.40 s + 02

Standard polynomials (s) for u = 1, 2,. . .

Table A.8.2 continued

0 .0 0

0 .2 5

0 .5 0

0 .7 5

1 .0 0

h W
1 .2 5

0 .0 0

0 .2 5

0 .5 0

0 .7 5

1 .0 0

h W
1 .2 5

u =

u =

2
3

4
5

6
4

1 2

1 6

Step response hW (0 t)

w 0 t 2 0

w 0t 1 0

A.8 Tables
289

Name

PT1

PT2

PT2 S

No.

K R

K R

T
I

Step
response

KR

1 2

sin

(
)



(t)

1 2 0 t

(t)

1 2 0 t

cos

Tt
T1
1
T1 T2 e
t 
T

T2
T1 T2 e

1e0 t

KR 1



t
KR 1 e T (t)

t
TI (t)

KR (t)

h(t) =

<1

KR
1+2 s+

1 2
2s
0

KR
(1+sT1 )(1+sT2 )

KR
1+sT

1
sTI

KR

G(s) =

K R

K
R

R ( w )

R ( w )

R ( w )

j I ( w )

j I ( w )

j I ( w ) K

R ( w )

jI( w )

R ( w )

j I ( w )

Nyquist plot

Table A.8.3: Standard elements in control engineering

d B

0
0

A d B

- 9 0

d B

- 9 0
- 1 8 0

0
- 9 0

d B

d B

j
0
- 9 0
- 1 8 0

w 0

1 1
T 2 T 1

1
I

Bode plot

- w 0z

- 1
T 1

- 1
T

jw

- 1
T 2

jw

1 - z 2w 0

jw

jw

none

Poles x and
zeros o in s
domain

290
A Mathematical and table appendix

IT1 S

PI

DT1

PD

PID

10

11

K R

h
R

T I/ K

T
TI

t
TI

t
T

(t)

TD (t)

KR 1 +

(t)



KR 1 + TtI (t) +
TD (t)

KR [(t) + TD (t)]

t
TI


 t
e T 1 (t)

+s2 TI TD
KR 1+sTIsT
I

KR (1 + sTD )

sT
1+sT

sTD

I
KR 1+sT
sTI

1
TI s(1+sT )

Table A.8.3 continued

jI( w )

j I ( w )

K
R

w
R ( w )

R ( w )

R ( w )

w
R ( w )

w )
R R (

1R ( w )

j I ( w )

j I ( w )

j I ( w )

jI( w )

- T
T I

d B

d B

9 0

d B

A d B
j

9 0

d B

9 0
0

d B

9 0
0
- 9 0

j 0
0
- 4 5
- 9 0

- 1 8 0

0
- 9 0

w 1

1
D

1
T D

1
T I

w 2

1 w
T I

< T

w 1

j w

jw

jw

jw

jw

for 1 and
2 see
No. 12

w 2

4 T

- T 1
D

- 1
T

- 1
T

jw

A.8 Tables
291

PIDT1

PTt

1storder
all
pass

Lead

Lag

12

13

14

15

16

w N
w

w N
w

- 1

K R
T

T t

1+

1+

N
Z

N
Z


(t)

N <Z


1 eN t (t)

N >Z


1 eN t (t)

t
2e T

KR (t Tt )




TD
t
T +
+ T 1 e
KR 1 T +
TI
T
TI

t (t)
TI

s
1+
Z
s
1+

s
1+
Z
s
1+

1sT
1+sT

KR esTt

I +s TI TD
KR 1+sT
sTI (1+sT )

Table A.8.3 continued

j I ( w )
w Z

w N

jI( w )

- 1

j I ( w )

j I ( w )

K R 1 - TT

I
T

w T

j I ( w
)

w Z

w N

R ( w )

R ( w )

R ( w )

+ 1

R ( w )

R ( w )

d B

(w )
w

d B

0
0

d B

- 9 0

d B

- 1 8 0

d B

- 9 0

(
w

w
N

w
N

N Z

w w

N Z

w w

Z d B

w 1 w 2 1 w
T

w 2

< T
w 1

jw

w N w Z

jw

w Z w N

- 1
T

jw
T

1
2T
1 D
B= A2
A=

jw

1
TI TD

2 =AB

1 =A+B

4 T

292
A Mathematical and table appendix

denominator

1+x+

x2
2!

2!

1+x+

x2

1+x+

x3
3!

x2
2!

3!

x3

x4
4!

2 x2
1 x3
4 2! + 4 3!
1 15 x
3 x2
2 x3
1 x4
5 2! + 5 3! + 5 4!

1 + 34 x +
1 + 45 x +

1 x2
3 2!

1 14 x

1 + 23 x +

1 x3
10 3!
1 x2
10 2!
2
1 x3
1 12 x + 15 x2! 20
3!
2
1 x3
1 + 12 x + 15 x2! + 20
3!
3
1 x2
1 x3
1 7 x + 7 2! 35 3!
2
4 x3
1 x4
+ 47 x + 27 x2! + 35
3! + 35 4!

3 x2
10 2!

1 + 25 x +

1 35 x +

1 x2
6 2!
2
1 + 12 x + 16 x2!
1 x2
1 25 x + 10
2!
3 x2
1 x3
1 + 35 x + 10
+
2!
10 3!
1 x2
1 13 x + 15
2!
2
3
1 x4
1 + 23 x + 25 x2! + 15 x3! + 15
4!

1 12 x +

1 13 x

1
1+x

1 x + x2! x3!
1
2
3
3
1 4 x + 24 x2! 14 x3!
1 + 14 x

1 x + x2!
1
2
2
1 3 x + 13 x2!
1 + 13 x

1x
1
1 12 x
1 + 12 x

1
1

numerator
0

Table A.8.4 Pad

e table for ex

A.8 Tables
293

294

A Mathematical and table appendix

Index

295

Index
absolute value of the frequency response,
38
Ackermann, 213
actual value, 6, 7
actuator, 6, 95, 113, 184, 190, 192
addition
frequency response characteristic, 40
using Nyquist plots, 37, 39
algebraic
equation, 19
algebraic product, 236
Algebraic stability criteria, 64
algebraic sum, 236
all-pass element, 60, 146
all-pass plant, 154
amplitude response, 38, 61
denition, 35
logarithmic, 61, 77
of a minimum/non-minimum phase system, 58
angle condition
root-locus method, 83, 86
angle of departure
root-locus method, 88, 89, 92
angle of entry
root-locus method, 88, 89
anti-windup, 193
aperiodic behaviour, 280
approximation by a PT1 Tt element, 117
approximation by lines, 41
asymptote
deviation of the magnitude, 46
nal, 45, 46, 53
initial, 45, 46, 53
point of intersection, 53
slope, 46, 53
asymptotic stability, 63
auxiliary control loop, 188, 190
underlying, 190
auxiliary controlled variable, 189, 190
auxiliary controller, 188, 189
auxiliary manipulated variable, 192, 193
auxiliary variable, 188
bandwidth, 80

closed loop, 120, 126, 132

of a system, 56
basic control loop, 184
behaviour
aperiodic, 50, 117, 280
dynamical, 104, 121
inherent, 180
minimum phase, 58
non-minimum phase, 58, 154, 176, 187
of a continuous-time system, 37
of a minimum/non-minimum phase system, 58
of an open loop, 100
oscillating, 50
oscillating PT2 S element), 278
permanent oscillation, 119
PT2 , 50
PT2 S element, 278
set-point, 129
static, 100, 101, 104
steady state, 130
undamped, 280
unstable, 173
bell function, 235
binary valued logic, 225
binomial form, 150
block diagram
closed loop, 150
closed-loop control, 5
closed-loop control system, 7
closed-loop system, 96
combined observer-controller, 218
compensator, 152
control system, 6
feedback loop, 33
observer principle, 215
of a series connection of transfer function
elements, 32
open-loop control, 4
open-loop system, 97
PID controller, 111
state-feedback control system, 205
state-feedback control system with integrator, 208
Bode diagram, 38, 56, 57

296
lag element, 137
lead element, 134
of a PT2 S element, 54
Bode plot
of the open loop, 127, 128
breakaway point
root-locus method, 84
breakpoint frequency, 46, 57
PT1 element, 130
lag element, 136, 138
lead element, 133
PT1 element, 45, 276
Butterworth form, 150

Index

crossover frequency, 78
gain margin, 78
phase margin, 78
characteristics of the open loop in the frequency domain
crossover frequency, 127
gain margin, 127, 129
phase margin, 127, 129, 131
closed loop
step response, 150
transfer function, 149, 150
bandwidth, 120, 132
basic structure, 6
behaviour, 96, 149
calibration of the root-locus, 83
block diagram, 150
car brake, 244, 245
characteristic equation, 83, 97
cascade control, 208
characteristics in the frequency domain,
cascade control systems, 190
120
causal function of time, 12, 13, 270, 273
command behaviour, 97
causal system, 9
compensator, 152
centre of gravity method, 249
disturbance behaviour, 96
centre of singleton method, 249
instability, 104, 173
centroid methods
order, 158
margins, 250
phase angle, 120
characteristic equation, 29, 67, 68, 83
pole-zero distribution, 149
coecients, 68
poles, 71, 156, 158
denition, 24
resonant peak, 120
of a cascade control system, 191
resonant peak frequency, 120
of a system with auxiliary manipulated response to step disturbance, 106
variable, 193
signals, 6
of a system with auxiliary variable, 189
single loop, 184
of the closed loop, 67, 83, 97
stability, 70
position of roots, 63
standard quantities in time domain, 104
stability analysis, 63
static behaviour, 101
system with feed-forward on the controller, static property, 100
185
structure, 168
system with feed-forward on the manip- time response, 130, 131
ulated variable, 186
transfer function, 67, 96, 97, 205, 288, 289
characteristic function, 231
with cumulative disturbance, 96
characteristics of the closed loop in the fre- zeros, 158, 206
quency domain, 120
closed-loop transfer function, 150, 168
bandwidth, 120, 126, 132
closed-loop transfer function for command
damping ratio, 121
input, 169
natural frequency, 121
desired, 168
phase angle, 120
closed-loop control, 3
resonant peak, 120, 132
denition, 5
resonant peak frequency, 120
closed-loop system, 206
characteristics of the open loop in frequency COG, 249
domain

Index

297

command behaviour, 96, 102, 107

frequency response of the open, 127
command input, 96, 101
resonant peak frequency of the closed
step, 106
loop, 125
command signal, 101
step response of a closed, 129
command variable
control system, 168
step, 169
complex loop structures, 184
common divisor, 176, 177
single-loop, 191
comparison of the dierent types of con- state-feedback, 204
trollers, 116
with auxiliary manipulated variable, 192
compensation, 172
with auxiliary variable, 189
disturbance, 185, 186
controllability, 203
of disturbances, 184
controllability matrix, 203, 213
of poles, 145, 154
controlled chain, 5
of zeros, 154
controlled value, 6
plant, 151
controlled variable, 46, 96, 168, 188, 190,
static, 187
192
compensator, 132, 153
controller, 6, 95, 96, 104
generalised design method, 156
integral, 160
compensator design methods, 148
state-feedback, 204
complement, 235
tuning parameters, 112
complex dierentiation theorem of the Laplacecontroller transfer function, 131, 151, 156,
transform, 12
169, 176, 177, 179, 180, 182
complex shifting theorem of the Laplace controller canonical form, 210, 211
transform, 12
transformation, 212
condition of properness
controller coecients, 158
of a transfer function, 28
controller gain, 158
condition of realisability
critical, 117
controller, 151
controller order, 161
conformal mapping, 275, 276
controller output, 112
continuous-time system, 26, 35, 37
controller parameter, 157, 158
control
controller transfer function, 111
cascade, 190
convergence
control behaviour, 184
Laplace transform, 9, 269
control device, 3
convolution, 27
control error, 4, 6, 7
convolution in the frequency domain, 13,
dynamical, 104
272
steady-state, 107
convolution in the time domain, 13, 271
steady-state value, 101
convolution integral, 27
control factor
convolution of two functions in the frequency
dynamical, 97
domain, 13
static, 104
convolution of two functions of time, 13
control loop
correction elements, 132
auxiliary, 188
correspondences of Laplace transform, 10
bandwidth of the closed loop, 126
COS, 249
block diagram, 6, 7
crisp set, 231
Bode plot of an open, 128
critical controller gain, 117
Bode plot of the open loop, 127
critical damping, 279
closed, 5
critical period, 117

298
critical point, 7375
critical stability, 63, 64
crossover frequency, 78, 127
cumulative disturbance, 95
D action, 113, 116
D behaviour, 112, 113, 186
D element, 42
D step, 113
damped natural frequency, 279
damping ratio, 52, 121, 131, 180
relative, 279
dead time, 26, 117, 184
decay of the amplitude, 123
decay time constant, 279
deforming the root locus, 143
defuzzication, 243, 247, 249251, 256
delay time, 117
delayed behaviour, 188
delta function, 24
demonstration example, 7, 20, 110, 119, 148,
192, 222, 226, 227, 232, 236, 238, 241,
243, 245, 250
derivative action time, 111
derivative element, 42
derivative theorem of the Laplace transform, 12, 270
design
root-locus method, 148
direct, 148
for an unstable plant, 145, 177
for reference and disturbances, 168
frequency-domain characteristics, 148
indirect, 148
state-feedback, 209211, 213
design in controller canonical form, 210,
211, 213
design method
root locus, 142
compensator, 148
complex loop structures, 184
empirical, 117
frequency domain characteristics method,
120
pre-lter, 175
design of observer, 216
desired transfer function, 150
dierential equation

Index
initial condition, 26
of a RC lag, 44
solving using the Laplace transform, 19
distribution, 24, 25
distribution of poles, 149
disturbance, 3, 5, 6, 95, 96, 101, 184, 189,
192
at the plant input, 168170, 175, 179
at the plant output, 168, 169, 172, 175,
180
compensation, 184186
entry point, 168, 174
feed-forward on the controller, 184
feed-forward on the manipulated variable, 186
high-frequency, 130
step, 169
transfer function, 96
disturbance behaviour, 96, 102, 106, 168
closed loop, 96
disturbance control, 6, 96
disturbance feed-forward, 184
disturbance reduction, 189
disturbance rejection, 107
disturbance transfer function, 169, 174, 177,
179, 180, 182
desired, 169
dominant pair of poles, 121, 124, 128, 131,
138, 142
DT1 element, 48, 112, 186, 193
dual system, 217
dynamical behaviour, 205
of the closed loop, 104, 121
dynamical control factor, 97
dynamical quality, 80
dynamics of the control behaviour, 107
DYNAST study example, 47, 61, 103, 119,
189
eigenvalue, 24, 200, 204
empirical tuning rules, 117
equation
algebraic, 19
characteristic, 24, 29, 63, 67, 68, 83, 97,
185, 186
equivalence
Hurwitz and Routh criteria, 282
error

Index
steady state, 130, 169
feed-forward
auxiliary manipulated variable, 188
feed-forward gain
state-feedback control, 207
feedback, 33
negative, 7, 33
feedback loop
of transfer functions, 32, 33
feedback principle, 7
feedforward
state-feedback controller, 205
feedforward gain, 207
feedthrough matrix, 199
nal asymptote, 45, 46, 53
nal value
frequency response, 38
nal value theorem of the Laplace transform, 13, 14, 37, 101, 274
fraction
rational, 282
frequency
breakpoint, 130, 133, 136, 138
crossover, 127
damped natural, 279
resonant, 125, 277
frequency domain, 9, 19, 37
state-feedback controller, 205
frequency domain characteristics method
application example, 138
design method, 120
synthesis of controllers, 131
frequency function, 9
frequency range, 130, 133, 136
frequency ratio
lag element, 136
lead element, 133, 135
frequency response, 61, 130
absolute value, 38
addition, 37, 39
all-pass element, 60
amplitude response, 35, 58, 59, 61, 75
closed loop, piecewise determination from
GW (j), 130
dead time, 61
nal value, 38
imaginary part, 35, 281

299
initial value, 38
locus, 276
logarithmic amplitude response, 38, 61,
77
magnitude, 36
multiplication, 37, 39
Nyquist plot, 37, 57, 61
of the open loop, 127
phase response, 35, 36, 38, 58, 61, 75, 77
real part, 35, 37, 281
frequency response characteristic
addition, 40
Bode diagram, 38
calculation, 38
denition, 38
Nyquist criterion, 70, 74
simple transfer function elements, 41
frequency response characteristics, 39
function
bell, 235
characteristic, 231
Gauss-, 235
hyperbolic, 280
sigmoidal, 235
function of time, 9
causal, 12, 13, 270, 273
fuzzication, 243, 245, 251, 253
fuzzy AND operator, 235, 236
fuzzy composition, 240
fuzzy control, 225, 252, 261
fuzzy controller, 252
fuzzy controller design, 263
fuzzy inference machine, 243
fuzzy logic, 225, 226, 231
fuzzy NOT operator, 235
fuzzy OR operator, 235, 236
fuzzy relations, 237
fuzzy set, 231, 232
fuzzy system, 242
components, 251
gain, 102, 111
controller, 111, 117, 158
of a PT1 element, 47
of a transfer function element, 47
of the controller, 102
of the open loop, 82, 100, 102
P element, 41

300
PID controller, 111
plant, 102, 117, 119, 187
gain margin, 78, 127, 129
Gaussian function, 25, 235
generalised design method
compensator, 156
generalised integral of squared error, 108
GISE, 108
grade of membership, 232
grid
orthogonal, 275
Hilbert transformation, 281
Hurwitz conditions, 66
Hurwitz criterion, 6567, 282
example, 67
Hurwitz determinant, 66
Hurwitz polynomial, 65, 69
hyperbolic function, 280
I action, 113, 117
I behaviour, 101, 103, 105, 113, 116, 186
I controller, 113, 160
I element, 41, 43, 57
I2 behaviour, 101, 104, 105
IAE, 108
IE, 108
image function, 9
impulse
Laplace transform, 24
inherent behaviour, 24, 29, 121, 180
inherent dynamics, 174
initial asymptote, 45, 46, 53
initial condition, 197199, 202, 204, 205,
208210, 215218
of a dierential equation, 26
initial value
frequency response, 38
initial value theorem of the Laplace transform, 13, 37, 273
input
Laplace transform, 26
input matrix, 199
input signal
bounded, 62
sinusoidal, 36, 37
input variable, 197
input vector, 199

Index
input-output behaviour, 197
inputs
multi-input-multi-output system, 199
instability
denition, 63
of the closed loop, 104
proving, 70
integral action time, 111
integral of error, 108
integral of absolute value of error, 108
integral of squared error, 108
integral of squared error and squared control eort, 108
integral of time multiplied by the absolute
value of error, 108
integral performance index, 107
integral plant, 161
integral theorem of the Laplace transform,
13, 271
integral transformation, 9, 12
integrator
state-feedback, 207
intersection operator, 235, 236
intersections
of the locus, 75
inverse frequency response, 40
inverse integral
Laplace transform, 10, 14
inverse Laplace transform, 10, 14, 15, 270,
282
inverse plant transfer function, 151
ISE, 108
ISESC, 108
ISTq E, 108
ITAE, 108
lag, 44
lag element, 136
Laplace integral, 9
Laplace transform, 9, 269
abscissa of convergence, 269
complex dierentiation theorem, 12
complex shifting theorem, 12
convergence, 9, 269
convergency, 269
convergency area, 269, 270
convolution in the frequency domain, 13,
272

Index

301

convolution in the time domain, 13, 271

correspondences, 10
derivative theorem, 12, 270
examples, 270
nal value theorem, 13, 14, 37, 101, 274
impulse, 24
initial value theorem, 13, 37, 273
input, 26
integral theorem, 13, 271
inverse, 10, 14, 15, 270, 282
inverse integral, 10, 14, 270
main theorems, 11, 270
operator notation, 10
output, 26
real shifting theorem, 12
similarity theorem, 12
solving linear dierential equation, 19
state-space representation, 199
superposition theorem, 11
weighting function, 27, 63, 281
law of Bode, 281
lead element, 133
left-hand rule of the Nyquist criterion, 74
for Bode diagram, 78
linear mapping, 277
linear plant, 95
linear system, 26, 29, 35, 37
linguistic term, 230, 235, 238, 243245, 247,
251, 254, 256, 263
linguistic variable, 229, 230, 237, 263
loading crane, 261, 264
logarithmic amplitude response, 38
logarithmic, 38
loop
poles of the open loop, 142
zeros of the open loop, 143
loop gain, 104, 130
low-pass property, 56
lumped parameters, 26

Mamdani fuzzy system, 250

manipulated variable, 4, 6
mapped function, 10, 13, 270
mapped space, 19
mapping
conformal, 275, 276
linear, 277
mathematical model, 117
maximum methods, 250
maximum overshoot, 116, 180
as function of the damping ratio, 123
calculation, 122
denition, 105
measurement device, 6, 95, 184
membership, 226
membership function, 231, 232
boundary, 234
core, 233
height, 233
normal, 233
singleton, 234, 241, 245
subnormal, 233
support, 233
method
frequency domain characteristics, 131
Truxal and Guillemin, 150
Weber, 150, 287
minimum phase behaviour, 58
minimum phase system, 59, 61, 281
mode, 121
modes, 23, 24, 29
multi-input-multi-output system
inputs, 199
outputs, 199
state-space representation, 199
multiple pole
real, 149
multiplication
using Nyquist plots, 37, 39

magnitude condition
root-locus method, 83, 85, 86
magnitude of the frequency response, 36
magnitude response
increase, 134
main control loop, 189, 190, 192
main controller, 189, 190
main theorems of the Laplace transform,
11, 270

natural frequency, 52, 56, 121, 180, 280

damped, 279
negative feedback, 7, 33
non-minimum phase behaviour, 58, 154, 176,
187
non-minimum phase system, 59, 61, 281
nonlinear characteristics, 253, 254, 260, 266
numerator degree

302
transfer function, 28
Nyquist criterion, 70
frequency response characteristic, 70, 75
general case, 74
left-hand rule, 74
left-hand rule for Bode diagram, 78
simplied case, 74
theory, 70
Nyquist criterion using Nyquist plots, 70
of the open loop, 72
Nyquist plot
lag element, 137
lead element, 134
observability, 203
observability matrix, 204
observer, 214, 215
design, 216
state reconstruction, 214
state-feedback controller, 217
structure, 215
observer-controller, 217
open control loop, 97
open loop, 97
frequency response, 131
characteristics, 131
gain, 82, 100, 102
pole and zero distribution, 90, 91
poles, 71, 72
transfer function, 67, 70, 82, 100
zeros, 206
open-loop control, 3
denition, 5
operator
complement, 235
intersection, 235, 236
union, 235, 236
operator notation
Laplace transform, 10
operators
for fuzzy sets, 235
optimal tuning
PID controller, 113
order of controller, 158
original function, 9, 10, 13, 18
original space, 19
orthogonal grid, 275
output

Index
Laplace transform, 26
output equation, 198
output matrix, 199
output signal
bounded, 62
sinusoidal, 36
output vector, 199
outputs
multi-input-multi-output system, 199
overshoot, maximum, 184
P behaviour, 101, 102, 104, 105, 113, 186
P controller, 114, 147, 161
P element, 41, 187
P step, 113
Pad
e table, 293
pair of poles
dominant, 121, 124, 128, 131, 138, 142
imaginary, 280
parabolic input signal, 101104
parallel connection, 37
of transfer function elements, 31, 33, 37
parameter
lumped, 26
Parsevals theorem, 109
partial fraction decomposition, 15, 24, 279
conjugate complex poles, 16
example, 17
multiple poles, 16
PD controller, 114
PD element, 48, 57, 187
PDT1 controller, 114, 115, 134
peak time
denition, 105
performance diagram, 115
performance index, 107
quadratic, 108
performance index in time domain
delay time, 117
generalised integral of squared error, 108
integral of absolute value of error, 108
integral of error, 108
integral of squared error, 108
integral of squared error and squared control eort, 108
integral of time multiplied by the absolute value of error, 108
maximum overshoot, 105, 116, 122

Index
peak time, 105
rise time, 105, 117, 122, 184
settling time, 105, 122, 123, 184
period
critical, 117
permanent oscillations, 24
phase
non-minimum, 154
phase angle
maximum, of a lead element, 133
of the closed loop, 120
phase behaviour
minimum, 58, 59
non-minimum, 58
of a minimum/non-minimum phase system, 58
phase margin, 78, 127, 129, 131, 132
phase response, 38, 61, 77
denition, 35
lead element, 135
logarithmic representation, 39
Nyquist criterion, 77
of a minimum/non-minimum phase system, 58
of a non-minimum phases system, 58
phase shift, 36
continuous, 72
lag element, 136
lead element, 133
step, 72
PI controller, 114, 186
PID controller, 111, 148
block diagram, 111
ideal, 112
optimal tuning, 113
real, 112
plane, complex
s plane, 142, 146, 148, 173, 176
G plane, 37, 73, 275, 276
left-half s plane, 24, 64, 72, 101
right-half s plane, 64, 73
s plane, 19, 24, 59, 60, 63, 71, 72, 83, 269,
275277
plant, 6, 95
all-pass, 154
gain, 119, 187
integral, 158, 161
linear, 95

303

parameter, 145, 157, 173

proportional, 158, 162
step response, 119
unstable, 145, 147, 148, 154, 173, 177,
179, 180
zeros, 159
plant transfer function, 168, 179
inverse, 151
plant output, 101
point of intersection
asymptote, 53
pole
conjugate complex, 16
double, 103
multiple, 16, 63
of the closed loop, 142
plant, 173
single, 15, 64
unstable, 177
pole and zero distribution
of a minimum/non-minimum phase system, 58
of a rational transfer function, 29, 81
of an all-pass element, 59, 60
of an open loop, 90, 91
pole assignment, 142
pole distribution, 63, 64
of a PT2 element, 55
pole excess, 151, 170, 174, 185
pole placement, 209
state-feedback, 209
pole-zero distribution
of a closed loop, 149
poles, 15, 200
of the closed loop, 71
of the open loop, 71, 72, 142
of the transfer function, 28, 29
positive feedback, 33
pre-controller, 184
pre-lter, 159, 168
design, 175, 179, 180
unstable, 176
preservation of angles, 275
problem, 61, 80, 182
product, algebraic, 236
properness
of a transfer function, 28
PT1 element, 43, 46, 47, 56, 57, 130, 189

304
PT1 Tt element, 117
PT2 element, 50, 56
PT2 S element, 56, 278
PTn element, 56
PTt element, 61
quadratic performance index, 108, 109
calculation, 114, 283
quality
dynamical, 80
questionnaire, 27, 33, 34, 38, 51, 64, 80,
232, 235, 237
ramp input signal, 101104
rational fraction, 15, 282
rational transfer function, 28
RC high pass element, 48, 49
RC lag, 44
real part
negative, 282
real shifting theorem of the Laplace transform, 12
realisability, 172
realisability condition, 174, 179
controller, 170, 172, 182
of a transfer function, 28, 71
pre-lter, 175, 177
reference behaviour, 101
reference value, 190
relative damping ratio, 279
residual, 15, 16, 18
theorem, 16
resonant peak, 120, 125, 132
of the closed loop, 120
resonant peak frequency, 278
of a PT2 S element, 53
of the closed loop, 120, 125
rise time, 117, 184
calculation, 122
denition, 105
RLC lag, 50, 51, 54, 56, 197
root-locus, 84
denition, 81
root-locus method
angle condition, 83, 86
angle of departure, 88, 89, 92
angle of entry, 88, 89
application for controller design, 142

Index
breakaway point, 84
examples, 92
magnitude condition, 83, 85, 86
rules for constructing, 86
Routh criterion, 68, 69, 282
example, 69
Routh schema, 68
rule base, 230, 238244, 246, 253, 256, 259,
260, 263, 265
s domain, 121
state equation, 200
scales
double logarithmic, 41
single logarithmic, 41
separation property, 218
series connection, 37
of transfer function elements, 31, 32, 37,
40, 59
set
crisp, 231
fuzzy, 231
set point, 6, 7, 208
constant/not constant, 6
set theory, 231
set-point value, 4
settling time, 180
as function of the damping ratio, 125
calculation, 123
denition, 105
sigmoidal function, 235
signal path, 6
signals in the closed loop, 6
similarity theorem of the Laplace transform, 12
single-input-single-output system
state-space representation, 197, 198, 200
time-invariant, 198
singleton, 234, 249
singularity, 24
specication
time response, 104
spray-water cooler, 188
stabilising of an unstable plant, 145
stability, 62
asymptotic, 63
closed loop, 70
conditions, 62, 63

Index
critical, 63, 64
fenition, 62
stability analysis, 63, 70
stability criteria
algebraic, 64
stability criterion, 65, 68
Nyquist, 70
stability diagram, 115
stable system, 62
standard form of the transfer function, 149
standard forms
Butterworth form, 150
ITAE form, 150
settling time form, 150
state, 197
state equation, 198
s domain, 200
state reconstruction using observers, 214
state variables, 197
uniqueness, 201
state vector, 199
state-feedback, 204
with integrator, 207
design, 209
pole placement, 209
state-feedback control
feed-forward gain, 207
steady state, 207
zeros, 206
state-feedback control system, 204
state-feedback control with integrator, 207
state-feedback controller, 204, 209
frequency domain, 205
observer, 217
structure, 204
state-space transformation, 202
example, 202
state-space representation, 197
controllability, 203
Laplace transform, 199
multi-input-multi-output system, 199
observability, 203
single-input-single-output system, 197, 198,
200
transfer function, 199
transformation, 202
static behaviour, 100
of the closed loop, 101, 104

305
static control factor, 104
statical behaviour, 205
steady state, 116, 193, 205
state-feedback control, 207
steady state behaviour, 130
steady-state error, 158
steam ow, 188
steam superheater, 188, 189
steam temperature, 188, 189
step command input, 106
step disturbance, 116
step input signal, 101104
step response, 180
lag element, 136
lead element, 134
normalised, 150
of a closed loop with PT2 S behaviour,
129
of a PT2 element, 55, 280
of a PT2 S element, 278, 280
of standard forms, 288, 289
PD controller, 113
PDT1 controller, 113
PI controller, 113
PID controller, 112, 113
PIDT1 controller, 113
PT1 Tt element, 119
Stodola criterion, 67
structure
state-feedback controller, 204
structure of an observer, 215
sum, algebraic, 236
summing point, 7
superposition theorem of the Laplace transform, 11
synthesis
for an unstable plant, 173
synthesis equation, 158
synthesis method
for an unstable plant, 180
frequency domain characteristics, 131
pre-lter design, 179, 180
system
causal, 9
continuous-time, 26, 35, 37
invariant, 29
linear, 26, 29, 35, 37
minimum phase, 58

306
multi-input-multi-output, 199
non-minimum phase, 154, 187
output signal, 35
realisable, 70
single-input-single-output, 197, 198
stable, 58, 62
time-invariant, 26, 198
unstable, 62, 63
with dead time, 70
with lumped parameters, 26
without dead time, 29, 58
system matrix, 199
system of dierential equations, 198
system property, 63
Takagi-Sugeno fuzzy system, 250
tangent at the turning point, 117
temperature control, 187, 189
test signals, 101
theorem
Parseval, 109
theorem of residuals, 16, 17
time constant
of a PT1 element, 44, 47
of a PT2 element, 67, 280
of an I element, 42
time domain, 9, 14, 37
time response
of the closed loop, 130, 131
time-invariant system, 26, 29
time-response specications, 104
time-varying system, 199
transcendental transfer function, 28
transfer function
all-pass element, 59
combinations, 30
dead time, 28, 61
denition, 26
desired, 149
disturbance, 169, 177, 179, 180
feedback loop, 32, 33
for calculations, 30
I controller, 113
I element, 42
improper, 28
inverse, 151
lag element, 136
lead element, 133

Index
numerator degree, 28
of a cascade control system, 190
of the closed loop, 67
of the open loop, 67, 70, 82, 97, 100
P controller, 113
P element, 41
parallel connection, 31, 33, 37
PD controller, 113
PD element, 48
PDT1 controller, 113
PI controller, 113
PID controller, 111
PIDT1 controller, 112
poles and zeros, 28, 29
pre-lter, 168, 175, 182
proper, 28
properness, 28
PT1 element, 44
PT2 element, 51, 280
PT2 S element, 52, 280
PTt element, 61
rational, 28, 29, 56, 61
realisability, 28
realisable, 179
series connection, 31, 32, 40
standard forms, 149
state-space representation, 199
stricly proper, 28
transcendental, 28
transfer function elements, 41
series connection, 40
transfer function for command input, 179,
180
transformation
controller canonical form, 212
state-space representation, 202
transformation matrix, 202, 212
tuning parameters of the controller, 112
tuning rules
empirical, 117
turning point
tangent, 117
unambiguously reversible mapping, 10
undamped behaviour, 280
union operator, 235, 236
universal set, 233
unstable

Index
PT2 S element, 281
unstable behaviour, 173
unstable plant, 145, 147, 148, 154, 173, 177,
179, 180
unstable system, 62, 63
value of truth, 231
variation of the plant parameters, 173
weighting function, 27, 282
Laplace transform, 63, 281
windup, 193
windup eect, 193, 194
zeros
controller transfer function, 173, 176
of the closed loop, 158
of the open loop, 142, 143
of the plant, 159
of the transfer function, 28, 29
plant, 173
state-feedback control, 206
Ziegler-Nichols tuning rules, 117
method of the stability margin, 117
method of the step response, 119

307

308

List of Tables
2.1.1 Corresponding elements of the Laplace transform . . . . . . . . . 10

2.3.1 Magnitude, phase response and deviation A() of the exact magnitude from the asymptotes for a

2.3.2 Pole positions in the s plane and step responses for elements with the transfer function G(s) = 1/

2.4.1 Examples of stability analysis using the Nyquist criterion with frequency response characteristics 7
2.5.1 Root loci of 1st- and 2nd-order systems . . . . . . . . . . . . . . 85
2.5.2 Typical distributions of open-loop poles and zeros and the root loci 91

3.2.1 Steady-state error for dierent type of systems of G0 (s) and dierent input signals xe (t) (command
3.3.1 The most common integral performance indices . . . . . . . . . . 108
3.3.2 Values for the integral . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4.1 Controller tuning parameters according to Ziegler and Nichols . . 120
5.2.1 Validity of fuzzy equivalences . . . . . . . . . . . . . . . . . . . . 237

A.8.1 Step responses for a given pair of complex poles and a real pole with multiplicity k according to Eq

A.8.2 Standard polynomials (s) of Eq. (3.6.2) and corresponding step responses of control systems of di
A.8.3 Standard elements in control engineering

. . . . . . . . . . . . . 290

A.8.4 Pad
e table for ex . . . . . . . . . . . . . . . . . . . . . . . . . . 293

309

List of Figures
1.2.1 Open-loop control of a room heating system . . . . . . . . . . . .

1.2.2 Characteristic of a heating control device for three dierent tuning sets (1, 2, 3)
1.2.3 Closed-loop control of a room heating system . . . . . . . . . . .

1.2.4 Block diagram of the open-loop control of the heating system . .

1.2.5 Block diagram of the closed-loop control of the heating system .

1.3.1 Basic block diagram of a control system . . . . . . . . . . . . . .

1.3.2 Simplied block diagram of a closed-loop control system . . . . .

2.1.1 (a) Graph of the original function f (t) (function in the time domain) and (b) positio
2.1.2 Schema for solving dierential equations using the Laplace transformation 19

2.2.1 Example of the pole and zero distribution of a rational transfer function in the comp
2.2.2 Mass-spring-damper mechanical system used for the interpretation of zeros 31
2.2.3 Response to the sinusoidal input signal xa (t) = sin t (a), (b) position of the mass m2
2.2.4 Series connection of two transfer functions . . . . . . . . . . . . . 32
2.2.5 Parallel connection of two transfer functions . . . . . . . . . . . . 33
2.2.6 Feedback using two elements . . . . . . . . . . . . . . . . . . . . . 33
2.3.1 Magnitude in dB of a transfer function over the complex plane s = + j 35

2.3.2 (a) Sinusoidal input signal xe (t) and corresponding output signal xa (t) of a linear ele
2.3.3 Example of a Nyquist plot drawn from measurements . . . . . . . 38

2.3.4 Relations between initial and nal value of the frequency response G(j) and the ste
2.3.5 Addition (a) and multiplication (b) of frequency responses in Nyquist plots 39

2.3.6 Plot of a frequency response: (a) linear, (b) logarithmic presentation ( on a logarith

2.3.7 (a) Magnitude and phase response (b) Nyquist plot of the frequency response of an i

2.3.8 (a) Magnitude and phase response (b) Nyquist plot of the frequency response of a D
2.3.9 Simple RC lag as an example of a 1st-order lag element . . . . . 44

2.3.10 (a) Magnitude and phase response (b) Nyquist plot of the frequency response of a P

310

A List of Figures
2.3.11 Graphical representation of the step response, h(t), of a PT1 element 47
2.3.12 (a) Magnitude and phase response (b) Nyquist plot of the frequency response of a PD element 48
2.3.13 Graphical representation of the step response, h(t), of a DT1 element 49
2.3.14 Simple RC high-pass circuit as an example of a DT1 element . . 49
2.3.15 (a) Magnitude and phase responses (b) Nyquist plot of the frequency response of a DT1 element 5
2.3.16 Simple RLC network as an example of a 2nd-order lag element . 51


2.3.17 Bode diagram of a 2nd-order lag element with the transfer function G(s) = 1/ 1 + s2/0 + (s/0 )

2.3.18 Nyquist plots of 2nd-order lag elements, (a) PT2 element, (b) PT2 S element 56

2.3.19 Denition of the bandwidth b of systems with lags (p resonant peak frequency, 0 natural freque

2.3.20 Representation of a dynamic system by two frequency response diagrams: (a) Bode plot, (b) Nyqui
2.3.21 Distribution of poles and zeros in the s plane (a) Ga (s) and (b) Gb (s) 59

2.3.22 Phase response of two transfer functions with identical amplitude, but with minimum and non-min
2.3.23 Poles (x) and zeros (o) of a 4th-order all-pass system . . . . . . . 60
2.3.24 (a) Nyquist plot and (b) phase of a dead time element . . . . . . 62
2.4.1 (a) Stable and (b) unstable system response xa (t) to a bounded input signal xe (t) 62
2.4.2 Stability of a linear system discussed by the distribution of the roots of the characteristic equation
2.4.3 Stability analysis of a simple control loop . . . . . . . . . . . . . 67

2.4.4 Poles of the open and closed loop in the s plane (multiple poles are counted according to their mult
2.4.5 Nyquist diagrams of G
(j) and G0 (j) . . . . . . . . . . . . . . . 73
2.4.6 Determination of continuous changes in the angle S . . . . . . 73

2.4.7 Positive (+) and negative () intersections of the locus G0 (j) with the real axis on the left-hand si

2.4.8 Count of the intersections on the left-hand side of the critical point for I2 behaviour of the open loo

2.4.9 Frequency response characteristics of G0 (j) = A0 () ej0 () and denition of positive (+) and nega
2.4.10 Phase and gain margin C and AP or APdB , respectively, in the (a) Nyquist diagram and (b) Bode

2.5.1 Plot of all roots of the characteristic equation s2 + 2s + p = 0 for 0 p < . Values of p are red an
2.5.2 Root locus of a simple second-order system . . . . . . . . . . . . 84
2.5.3 Pole-zero diagram for construction of the root locus . . . . . . . . 87
2.5.4 Asymptote congurations of the root locus . . . . . . . . . . . . . 88
2.5.5 Root locus of G0 (s) =

k0 (s+1)
s(s+2) (s+6+j2) (s+6j2) .

Values of k0 are in red and underlined. 93

311
2.5.6 Calculating the angle of departure P3 ,D of the complex pole sP3 = 6 + j2 94
3.1.1 The basic components of a control loop . . . . . . . . . . . . . . . 95
3.1.2 Block diagram of the closed-loop system . . . . . . . . . . . . . . 96
3.1.3 Block diagram of the closed-loop system with cumulative disturbance Z(s) 96
3.1.4 Open control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3.1.5 Step response of the closed loop, (a) speed and (b) current for dierent open-loop ga

3.1.6 Step response of the closed loop for dierent open-loop gains for = 4 and 0 = 2/T

3.2.1 Dierent input signals xe (t), which are frequently used for the disturbance z(t) and c
3.3.1 Typical under-damped response of a control system to step command inputs106
3.3.2 Typical under-damped response of a control system to step disturbances106
3.4.1 Block diagram of the PID controller . . . . . . . . . . . . . . . . 111
3.4.2 Step responses (a) of the ideal and (b) of the real PID controller 112
3.4.3 Step responses of the PID controller family . . . . . . . . . . . . 114

3.4.4 Stability and performance diagram for step changes (a) in the command input (w0 =

3.4.5 Behaviour of the normalised controlled variable y/(z0 KP ) for step disturbance z
= z

3.4.6 Describing the step response of a process by the three characteristic values KP (gain
3.5.1 Bode plot of the closed-loop frequency response . . . . . . . . . . 121
3.5.2 Distribution of poles of an element with a dominant pair of poles 121

3.5.3 Maximum overshoot Mp = f1 () (in %) relative to hW = 100% as function of the d

3.5.4 The product 0 tr,50 = f2 () (normalised rise time) as a function of the damping ratio

3.5.5 Normalised settling time 0 t3% f3 () and normalised rise time 0 tr as functions of

3.5.6 Characteristics b /0 = f4 (), b = f5 () and b tr,50 = f6 () depending on the dam

3.5.7 Bode plot of the open loop . . . . . . . . . . . . . . . . . . . . . . 127
3.5.8 Bode plotf the open loop with G0 (s) according to Eq. (3.5.16b) . 128

3.5.9 Step response hW (t) of the closed loop with PT2 S behaviour according to the transfe
3.5.10 Piecewise determination of |GW (j)|dB from |G0 (j)|dB in the Bode diagram130

3.5.11 Frequency domain characteristics of the open loop, C and C , depending on the da
3.5.12 Nyquist plot of a lead element . . . . . . . . . . . . . . . . . . . . 134

312

A List of Figures
3.5.13 Bode diagram of the lead element . . . . . . . . . . . . . . . . . . 134
3.5.14 Step response of the lead element . . . . . . . . . . . . . . . . . . 135

= mh ;Z = lower breakpoint
3.5.15 Normalised phase responses of the lead element: = ;
== /Z ; m
3.5.16 Nyquist plot of the lag element . . . . . . . . . . . . . . . . . . . 137
3.5.17 Bode diagram of the lag element . . . . . . . . . . . . . . . . . . 137
3.5.18 Step response of the lag element . . . . . . . . . . . . . . . . . . . 137
3.5.19 The magnitude (a) and phase (b) response of G01 (j), G02 (j) and G0 (j)141
3.5.20 Step response hW (t) and ramp response rW (t) of the designed control system142

3.5.21 Deforming of the root locus (a) to the right by an additional pole, (b) to the left by an additional z
3.5.22 Pair of conjugate complex poles in the s plane . . . . . . . . . . . 144

3.5.23 Root locus of GW (s) (plant with P controller) and potential positions of the dominant pairs of pole
3.5.24 Root locus of GW (s) with modied controller according to Eq. (3.5.47)146
3.5.25 Closed-loop step response with modied controller according to Eq. (3.5.47)147
3.5.26 Root locus of the closed loop consisting of an unstable plant and a P controller147
3.5.27 Root locus of the closed loop consisting of an unstable plant and a PID controller148
3.6.1 Block diagram of the closed loop to be designed . . . . . . . . . . 150
3.6.2 Compensation of the plant . . . . . . . . . . . . . . . . . . . . . . 152
3.6.3 Closed-loop behaviour for the example 3.6.2: hW (t) step response of the controlled variable on step

3.6.4 Closed-loop behaviour of the example 3.6.3: hW (t) step response of the controlled variable on step i

3.6.5 Closed-loop behaviour of the example 3.6.4: hW (t) step response of the controlled variable on step i

3.6.6 Compensation of the plant zeros (a) with a controller in the feed-forward path and (b) in the feedb

3.6.7 Step responses of the controlled variable y(t) for the case: (a) without pre-lter hW1 (t), (b) with pr

3.6.8 (a) Step responses of the controlled variable y(t) for the plant in the uncontrolled case hP (t) and in
3.6.9 Control system designed for reference and disturbance behaviour 168

3.6.10 Step responses for the design for a step disturbance z = (t) at the plant input:(a) controlled variab

3.6.11 Step responses for the design for a step disturbance z = (t) at the plant output:(a) controlled varia
3.7.1 Block diagram of the feed-forward on the controller . . . . . . . . 185
3.7.2 Block diagram of the feed-forward on the manipulated variable . 187

3.7.3 Examples of disturbance feed-forward congurations (a) on the controller and (b) on the actuator o

313
3.7.4 Block diagram of a control system with auxiliary manipulated variable yA 188

3.7.5 Example of a of a steam superheater temperature control system with auxiliary cont
3.7.6 Block diagram of a cascade control system . . . . . . . . . . . . . 190
3.7.7 Rearranged block diagram of a cascade control system . . . . . . 191

3.7.8 Examples of cascade control systems: (a) temperature control of a stirred tank react
3.7.9 Block diagram of a control system with an auxiliary manipulated variable uA 193
3.7.10 Block diagram of a control system with a bounded manipulated variable194
3.7.11 Step response of the closed loop system with and without anti-windup measure, (a)
3.7.12 Block diagram of a control system with an anti-windup measure . 195
4.2.1 Basic structure of a state-feedback control system . . . . . . . . . 205
4.2.2 Basic structure of a cascade state-feedback control system with integrator208
4.2.3 Principle of an observer . . . . . . . . . . . . . . . . . . . . . . . 215
4.2.4 Basic structure of a combined observer-controller . . . . . . . . . 218

4.2.5 Time responses of the state-feedback control system without integrator,(a) manipula

4.2.6 Time responses of the state-feedback control system with integrator,(a) manipulated

5.1.1 The dierence between the grade of truth in (a) binary valued logic {0,1} and (b) fu
5.2.1 Membership functions of a crisp set C and a fuzzy set F . . . . . 232
5.2.2 Membership grades of x0 in the sets A and B: A (x0 ) = 0.75 and B (x0 ) = 0.25232
5.2.3 Some characteristics of a membership function . . . . . . . . . . . 233
5.2.4 Membership functions with smooth transitions (Eqs.(5.2.12) to (5.2.14))234

5.2.5 Relation between two fuzzy sets: (a) membership functions, (b) 3-D view of the mem
5.3.1 Rule-based fuzzy system with n inputs and one output . . . . . . 243
5.3.2 Fuzzication of a car speed . . . . . . . . . . . . . . . . . . . . . 244
5.3.3 Fuzzy set of a car brake force . . . . . . . . . . . . . . . . . . . . 245

5.3.4 Fuzzy inference example: (a) inference with rule 1, (b) with rule 2, and (c) nal fuzz

5.3.5 Membership function of a conclusion using (a) max/min and (b) max-prod inference

5.3.6 Example of the application with two premises with (a) max/min inference and (b) m
5.3.7 Margin of B (u) (a) original and (b) expanded . . . . . . . . . . 250

314

A List of Figures
5.3.8 Components of a fuzzy system . . . . . . . . . . . . . . . . . . . . 252
5.4.1 Basic structure of a fuzzy controller . . . . . . . . . . . . . . . . . 253
5.4.2 Prototype membership functions for a fuzzy set with seven linguistic terms254
5.4.3 Nonlinear characteristic of a fuzzy controller . . . . . . . . . . . . 255
5.4.4 Membership functions and static characteristic of the fuzzy controller257

5.4.5 Inuence of the (c) characteristic of a proportional fuzzy controller (a) without overlapping in the i

5.4.6 Inuence on the (c) characteristic of a proportional fuzzy controller with (a) full overlap in the inpu

5.4.7 Inuence on the (c) characteristic of a proportional fuzzy controller with (a) full overlap in the inpu

5.4.8 Inuence on the (c) characteristic of a proportional fuzzy controller with (a) full overlap in the inpu

5.4.9 Inuence on the (c) characteristic of a proportional fuzzy controller with (a) full overlap in the inpu

5.4.10 Inuence on the (c) characteristic of a proportional fuzzy controller with (a) full overlap in the inpu

5.4.11 Inuence on the rule base on the (c) characteristic of a proportional fuzzy controller with (a) full ov

5.4.12 Control system with PD-type fuzzy controller: (a) block diagram and (b) 3D representation of the c
5.4.13 View of a portal-type loading crane . . . . . . . . . . . . . . . . . 262
5.4.14 Schematic representation of the loading crane . . . . . . . . . . . 262
5.4.15 Structure of the loading crane fuzzy control system . . . . . . . . 264
5.4.16 Triangular membership functions of the ve linguistic variables . 265
5.4.17 Rule bases for position and angle control . . . . . . . . . . . . . . 265
5.4.18 3D characteristics of the fuzzy system (a) u = u(e,e),
(b) u = u(e,)266

5.4.19 Responses of the portal-type loading crane with fuzzy control from initial to target position for load
A.1.1 Convergence area of the Laplace integral . . . . . . . . . . . . . . 269

A.2.1 Local conformal mapping of the lines = const and = const of the s plane into the G plane (gen

A.2.2 Conform mapping of the upper s plane ( > 0) into the G plane for the example G(s) = K/(1 + sT
A.7.1 Stability and performance diagram for step changes (a) in the command and (b) plant input286