Introduction To Control Engineering: Andy Pomfret and Tim Clarke

Introduction To Control Engineering
Andy Pomfret and Tim Clarke

Last revised: Spring 2009
Corrections made 2011, 2014
Department of Electronics,
University of York, Heslington,
York YO10 5DD, U.K.
tel: (U.K.) 1904 322319
fax: (U.K.) 1904 432335
e-mail address: [email protected]
CONTENTS
Contents
1 INTRODUCTION
2 How is a Control System Developed?
3 Linear Systems
4 The
4.1
4.2
4.3
4.4
Laplace transform
Laplace Transform Pairs . . . .
Laplace Transform Theorems .
Time Responses . . . . . . . . .
Important properties of transfer
. . . . . .
. . . . . .
. . . . . .
function .
5 Modelling Dynamic Systems

5.1 Poles and zeros . . . . . . . . . . . . . .
5.2 Modelling of Linear Electrical Networks
5.2.1 Example Problem . . . . . . . .
5.3 Modelling of Mechanical Systems . . . .
5.3.1 A More Complex Example . . .
5.4 Modelling of Rotational Dynamics . . .
5.5 Some Important Observations . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
6
6
7
9
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
11
11
11
12
13
15
16
6 First Order Systems
16
7 Second Order Systems

18
7.1 Step Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2 Time Response Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8 Analysis for Control System Design
24
9 Steady State Errors

9.1 The Error Transfer Function . . . . . . . . . .
9.2 Error Constants . . . . . . . . . . . . . . . . . .
9.3 System Type . . . . . . . . . . . . . . . . . . .
9.4 Type 0 Systems . . . . . . . . . . . . . . . . . .
9.5 Type 1 Systems . . . . . . . . . . . . . . . . . .
9.6 Type 2 Systems . . . . . . . . . . . . . . . . . .
9.7 System Type, Error Constant and Steady State
9.8 Examples . . . . . . . . . . . . . . . . . . . . .
9.8.1 Type & Order . . . . . . . . . . . . . .
9.8.2 Steady State Errors . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
Errors
. . . .
. . . .
. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
26
27
28
29
30
31
32
32
32
32
10 Stability
33
10.1 Generating a Routh Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
11 The Routh-Hurwitz Stability Criterion
11.1 Some Simple Examples . . . . . . . . . .
11.2 Problem Conditions . . . . . . . . . . .
11.2.1 First Term in a Row is Zero . . .
11.2.2 A Complete Row of Zeroes . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
35
36
37
38
CONTENTS
12 Stability Analysis using the Bode Plot
12.1 Bode Plot Structure . . . . . . . . . . . . .
12.2 Constant Gain (K) . . . . . . . . . . . . . .
12.3 Integration Factors 1s
. . . . . . . . . . .
12.4 Differentiation Factors (s) . . . . . . . . . .
12.5 First Order Lead Terms(1 + Ts) . . . . . .
1
12.6 First Order Lag Terms 1+T
. . . . . . .

s
2
12.7 Second Order Lead Terms 1 + 2n s + s 2 .
!n
12.8 Second Order Lag Terms
1
2
1+ 2 s+ s 2
n
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
39
39
40
41
41
42
42
. . . . . . . . . . . . . . . . . . . . . . 43
. . . . . . . . . . . . . . . . . . . . . . . . 44
12.9 Gain and Phase Margins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

13 Stability Analysis using the Nyquist Plot
13.1 Rapid Sketching of Nyquist Plots . . . . .
13.1.1 Some Nyquist Plot Examples . . .
13.2 Nyquist Stability Criterion . . . . . . . .
13.3 Stability Margins and the Nyquist Plot . .
13.4 More Accurate Sketching of Nyquist Plots
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
46
47
47
48
49
50
14 Root Locus Analysis

51
14.1 How Do We Check the Angle Criterion? . . . . . . . . . . . . . . . . . . . . . . . . 53
14.2 Sketching Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1 INTRODUCTION
INTRODUCTION
Control engineering is about finding solutions to real engineering problems. It is probably one of
the most widely encountered aspects of electronics. Control theory is used in the development of
a huge range of systems, from CD players and hard disk drives to helicopters and space vehicles.
ABS, traction control and engine management systems in cars all employ control techniques;
military jets and modern passenger aircraft use fly-by-wire technology, all based on the principles
of control engineering.
We shall begin the study of control engineering by considering some basic properties of systems.
Analysis tools can be used to gain an understanding of a system and its properties, and the way
in which these properties may be changed using feedback control.
Reference
Input or
Set Point
Error
Signal
+
Controller
Plant
or
Process
Output
Feedback
Signal
Figure 1: A Feedback Control System
Figure 1 shows the basic principle of feedback control in which a measure of the output is fed
back and subtracted from the desired input (hence the term negative feedback). The comparison
signal (or error signal) is amplified and used as a drive signal for the system which is now under
closed-loop control. One of the main objectives of the closed-loop arrangement is to cause the
output variable to follow the input (or demanded) variable (or variables), whilst reducing the
error signal to zero. As the load on the system changes the output variable must also be well
regulated against these changes or disturbances.
The system output is measured by a transducer (or sensor), producing either an analogue (continuous) or digital feedback signal for the controller. The controller performs the operation of
comparison and further amplification and additional signal processing to shape the performance
of the system.
How is a Control System Developed?
All closed-loop controllers are applied to dynamic systems. By this we mean systems which can
be modelled by differential equations whose time-domain solutions describe the response of the
variables of interest to various input demands. Hence, in order to analyse a system with a view
to designing a controller, it is necessary to carry out some mathematical modelling of the process
to be controlled. The degree to which this modelling is done depends on the complexity of the
system and the sophistication of the required controller in terms of e.g. performance, accuracy,
reliability, fault-tolerance, intelligence, etc.
The starting point is a list of performance objectives which are to be followed. Usually we need
to specify ways in which the output variables respond to:
1. Changes in demanded (or input) variables.
3 LINEAR SYSTEMS
2. Effects of noise, disturbances or load variations acting on or within the system.
The performance objectives can be specified in the time-domain or in the frequency-domain (or
a combination of both).
In order to achieve the required performance objectives, it is necessary to choose the right type
of controller and then proceed to design the controller on the basis of a mathematical model
expressed in either the time-domain or the frequency-domain (according to the required specifications).
Therefore, it is important, after choosing the performance objectives, to carry out some tests
on the system to establish if it behaves as predicted. If, at this stage, the system has not
been modelled, then more extensive tests are required to establish the main features of the
mathematical model. It must be stressed that to design a control system, modelling is usually
necessary.
Linear Systems
In performing analysis and designing control systems, the vast majority of design techniques
utilise models which are linear in character. The fundamental properties exhibited by linear
systems, apart from being described by linear, constant coefficient differential equations (that
you know and love so well) can be summarised as follows:
1. Superposition. If two input signals are applied separately to a linear system, and two
separate output signals are recorded, then, if the same input signals are applied simultaneously, the observed output signal will be exactly the same as the individual output signals
summed together.
2. Homogeneity. If an input signal, applied to a linear system, is scaled by some factor, then
the output signal will be scaled by the same amount.
Also, if a sinusoidal input signal is applied to a linear system, the output signal shall have
the same frequency, no matter what input signal amplitude is applied.
Clearly there will be limits to the above. A linear amplifier may be driven into saturation and
then it will exhibit non-linear behaviour.
The Laplace transform
Definitions.
For the Laplace transform, itself:
L{f (t)} = F (s) =
Z
0
f (t)est dt
(1)
4 THE LAPLACE TRANSFORM

where
s = + j
and for the inverse Laplace transform:
L1 {F (s)} =
where
1
2j
+j
F (s)est ds = f (t)u(t)
(2)
u(t) = 1
=0
t>0
t60
Manipulation of the Laplace transform is eased considerably by the use of the Laplace transform
tables. Additionally there is a set of theorems associated with the Laplace transform. For details
of these and their use can be found in any good systems theory text. We present them here for
information.
4.1
Laplace Transform Pairs

f (t)
(t)
u(t)
tu(t)
tn u(t)
eat u(t)
sin tu(t)
cos tu(t)
4.2
F (s)
1
1
s
1
s2
n!
sn+1
1
s+a
s2 + 2
s
s2 + 2
Laplace Transform Theorems
The most important theorem for control theory purposes is the final value theorem. We shall
come back to this later.
Systems theory shows that the response of a system to a forcing function is given by a convolution
integral in the time domain:
Z
y(t) =
g(t )u( )d
(3)
This assumes initial conditions are all zero.

Now, g(t) is the impulse response of the system. For complex systems it is best to use Laplace
variables since:
Y (s) = G(s)U (s)
(4)
and simple algebraic manipulation replaces the integration. G(s)is the transfer function. So we
Figure 2: Theorems For Use with Laplace Transforms
define a transfer function as the Laplace transform of the impulse response of a linear, timeinvariant system.
4.3
Time Responses
The next problem is, given a transfer function and a specific input, to determine the system
response in the time domain.
Consider the simple system
G(s) =
and apply a step input
s+2
(s + 4)(s + 6)
(5)
1
s
(6)
s+2
s(s + 4)(s + 6)
(7)
U (s) =
Therefore
Y (s) =
To calculate the time response we use partial fraction expansion of the Laplace transformed time
response followed by the transform tables.
A
B
C
s+2
= +
+
s(s + 4)(s + 6)
s
s+4 s+6
(8)
s + 2 = A(s + 4)(s + 6) + Bs(S + 6) + Cs(S + 4)
(9)
This yields
Setting s = 0
2 = 24A
A=
1
12
Setting s = 4
2 = B(4)(2)
B=
1
4
Setting s = 6
4 = C(2)(6)
C=
1
3
So the output is
Y (s) =
1
12
1
4
s+4
1
3
s+6
(10)
5 MODELLING DYNAMIC SYSTEMS
Using the transform tables this becomes
y(t) =
1
1
1
+ e4t e6t
12 4
3
(11)
We can use MATLAB to do most of this donkey-work.

Starting with Equation (7), we use the following syntax:
num = [1 2]
den = conv([1 4 0], [1 6])
[r, p, k] = residue(num,den)
which generates:
1
1 1
]
12 4
3
p = [0 4 6]
r = [
k = []
The terms should be self-explanatory. (If not, go into MATLAB and type help residue.)
4.4
Important properties of transfer function
1. Initial conditions are assumed to be zero.

2. The transfer function represents the Laplace transform of the impulse response of a linear,
time-invariant system.
3. The transfer function description does not include any information about the internal structure of the system.
4. If a system can be described by a linear, constant coefficient differential equation, it can
be converted to a transfer function.
Modelling Dynamic Systems
A major part of the overall control problem is understanding the dynamics of the plant. If a
suitable model for these is not forthcoming, the control system designed will be, at best, minimally
effective. For a single-input single-output (SISO) system, it is possible to define the dynamics in
two ways. A set of simultaneous integro-differential equations may be used, to generate a time
domain representation.
An example could be a simple mass/spring/damper system as shown in Figure 3:
10
x(t)
K
M
B
f(t)
Figure 3: A Simple Mass-Spring-Damper System
K is the linear spring constant

B is the linear damping coefficient
M is the Body Mass
x(t) is the displacement of the Mass from some set point
f (t) is the linear force applied to the mass.
Force balance is used to derive an equation for the system dynamics:
f (t) = M
dx(t)
d2 x(t)
+B
+ Kx(t)
dt2
dt
(12)
Or in mathematical shorthand
f (t) = M x
+ B x + Kx
(13)
However, it is not the most effective way of investigating the response to any forcing function f (t).
By far the best, and most frequent representation of the system involves the transfer function.
By converting the time domain elements of the equation to into the s-domain, using the Laplace
transform, and re-arranging it to give the ratio of the responding variable to the forcing function,
the transfer function is generated.
Continuing with the mass/spring/damper example, Equation (13) becomes
F (s) = M s2 X(s) + BsX(s) + KX(s)
(14)
It is now possible to manipulate this using simple algebraic rules:

[M s2 + Bs + K]X(s) = F (s)
(15)
11

From which
X(s)
1
=
2
F (s)
M s + Bs + K
(16)
It is the convention to represent the denominator in monic polynomial form (where the leading
coefficient is unity), to yield:
X(s)
= 2
F (s)
s +
5.1
1
M
B
K
Ms+ M
= G(s)
(17)
Poles and zeros
The denominator of the transfer function is known as the characteristic polynomial. When
equated to zero, it is the characteristic equation. The roots of this equation are the system
poles. The poles define the way the system behaves. If the poles lie in the left half s-plane, then
the system will be stable. If the poles lies on the imaginary axis then the system will oscillate.
Right half plane poles denote an unstable system. The order of the linear system is equal to the
number of poles.
The system zeros are the roots of any numerator polynomial in the transfer function. If the roots
in lie in the right half s-plane, the system is known as non-minimal phase. We shall consider the
effects of non-minimum-phase zeros in more detail later.
5.2
Modelling of Linear Electrical Networks
This will, for obvious reasons, be a very brief treatment. For a detailed discussion see appropriate
texts on electrical circuit theory.
Component
Capacitor
v(t) =
Resistor
Inductor
V-I
Rt
1
C
Impedance (Z(s) =
i( )d
v(t) = Ri(t)
v(t) = L di(t)
dt
V (s)
)
I(s)
1
sC
R
sL
Table 1: Voltage-current characteristics, and corresponding impedances, for simple electrical components
5.2.1
Example Problem
Problem: Find transfer function,
Vc (s)
V (s) ,
from the circuit of Figure 4.
The voltage across the capacitor Vc (s) is given by the impedance of the capacitor multiplied by
the current flowing through it (this is Ohms law):
12
v(t)
+
-
VC(t)
i(t)
Figure 4: A Simple Linear Electrical Circuit
Vc (s) =
1
I(s)
sC
(18)
and the current flowing in the circuit is easily calculated by Ohms law again, as the applied voltage
V (s) divided by the total impedance to which it is connected, which is the series combination of
the impedances of L, R and C:
I(s) =
V (s)
sL + R +
(19)
1
sC
Substitution of Equation (19) into Equation (18) gives

Vc (s) =
=
V (s)
1
sC sL + R +
s2 LC
Vc (s)
= 2
V (s)
s +
5.3
(20)
1
sC
V (s)
+ sCR + 1
(21)
1
LC
1
R
L s + LC
(22)
Modelling of Mechanical Systems
We have already considered a simple example of mechanical modelling. But we can take the idea
of impedance and apply it to mechanical systems, making the modelling process much simpler.
Component
Spring
Damper
Mass
Force-Displacement
f (t) = Kx(t)
f (t) = B dx(t)
dt
f (t) = M d2dtx(t)
2
Impedance (Zm (s) =

K
Bs
Ms2
F (s)
)
X(s)
Table 2: Force-displacement characteristics, and corresponding impedances, for basic mechanical components
These impedance formulae will now be demonstrated in a more complex example of mechanical
modelling.
13

5.3.1
A More Complex Example
Consider the system of Figure 5.
x1(t)
K1
x2(t)
M2
M1
B1
K3
K2
f(t)
B2
B3
Figure 5: A More Complex Linear Translational System

Problem: Find
X2 (s)
F (s)
There are two degrees of freedom (the masses can each be moved horizontally whilst the other
is held). Therefore, two simultaneous differential equations are required to describe the system
dynamics.
We use superposition by holding onto one mass and moving the other, calculating the total forces
induced.
The forces on each mass are due to:
1. Its own motion.
2. The motion of the other mass, transmitted through K2 and B2 .
So the steps are as follows:
1. Hold M2 and move M1 to the right, observing the effect on M1 . See Figure 6 for the
resulting forces on M1 - they all (except F (s)) oppose the motion, so they act to the left.
K1 X1 (s)
B1 sX1 (s)
M1 s2 X1 (s)
F (s)
K2 X1 (s)
M1
B2 sX1 (s)
Figure 6: Forces on M1 from moving it to the right
2. Hold M1 and move M2 to the right, again observing the effect on M1 . See Figure 7 for the
resulting forces on M1 .
14
K2 X2 (s)
M1
B2 sX2 (s)
Figure 7: Forces on M1 from moving M2 to the right
3. Now hold M1 and move M2 to the right, and this time observe the effect on M2 . See
Figure 8 for the resulting forces on M2 , and again note that they oppose its motion.
K2 X2 (s)
B2 sX2 (s)
K3 X2 (s)
M2
B3 sX2 (s)
M2 s2 X2 (s)
Figure 8: Forces on M2 from moving it to the right
4. Finally hold M2 and move M1 to the right, observing the effect on M2 . See Figure 9 for
the resulting forces on M2 .
K2 X1 (s)
M2
B2 sX1 (s)
Figure 9: Forces on M2 from moving M1 to the right
Combining Figure 6 and Figure 7, we can generate a suitable equation to represent the dynamics:

M1 s2 + (B1 + B2 ) s + K1 + K2 X1 (s) (B2 s + K2 ) X2 (s) = F (s)
(23)

(B2 s + K2 ) X1 (s) + M2 s2 + (B2 + B3 ) s + (K2 + K3 ) X2 (s) = 0
(24)
Figure 8 and Figure 9 yield a second, simultaneous equation for the system dynamics:
15
By algebraic manipulation of Equation (23) and Equation (24), we can generate any appropriate
(s)
transfer function that describes an input-output behaviour, such as XF2(s)
. It is left to the reader
to practice such manipulations...
5.4
Modelling of Rotational Dynamics
Systems which move rotationally, as opposed to translationally, can also be modelled using
impedances. Indeed, they are remarkably similar to their translational counterparts; the dual
of force is torque (T ), and torque acts to produce angular acceleration which leads to angular
velocity (). The dual of mass is inertia (J).
Component
Spring
Damper
Inertia
Torque-Displacement
T (t) = K(t)
T (t) = B d(t)
dt
T (t) = J d2dt(t)
2
Impedance (Zm (s) =

K
Bs
Js2
F (s)
)
X(s)
Table 3: Torque-angular displacement characteristics, and corresponding impedances, for

basic rotational mechanical components
Consider a mass being made to rotate at an angular velocity, by a torque, T, as show in

Figure 10.
K
J
T(t), q(t), w(t)
Figure 10: A Simple Linear Rotational System

Imagine that the applied torque T (s) causes the object to move in the direction of the arrow.
The inertia of the object, and the actions of the spring and damper, will all generate torques in
the opposite direction. We can therefore write down:
T (s) = Js2 (s) + Bs(s) + K(s)

1
(s)
=
2
T (s)
Js + Bs + K
=
s2 +
1
J
B
Js
K
J
(25)
(26)
(27)
16
6 FIRST ORDER SYSTEMS
5.5
Some Important Observations
A cursory look at the transfer function of any dynamic system will tell you how complex the
dynamics are. A combination of the number of poles (system order) plus the number and positions
of any zeroes will determine everything about its behaviour. However, it is possible to break all
complex dynamic system denominators and numerators into simple building blocks. These are:
1. real poles (first order)
2. poles at the origin in the s-plane (integration terms)
3. complex conjugate poles (second order)
4. real zeroes (minimum-phase or not)
5. complex conjugate paired zeroes (minimum-phase or not)
We shall now consider the properties of the possible constituent poles of any dynamic system with
a view to arriving at some important conclusions about how we may deal with complex systems
of high dynamic order.
First Order Systems
We shall use the transfer function
A
(28)
s+a
to represent a typical first order system. Using the final value theorem we find the dc gain to be
G(s) =
lim {G(s)} =
s0
A
a
(29)
For a unit step input (U (s) = 1s ), the output is

y(t) =
A
(1 eat )
a
(30)
We note that it is the position of the pole (s = a) which determines the shape of the response.
A left half s-plane pole means that the system is stable. If the absolute value of a is large, the
exponential decay is fast, the system has a fast time constant and it is highly responsive.
Figure 11 below shows two step responses for fast and slow first order systems both with unity
dc gain.
An alternative form of the same transfer function is
G(s) =
Kdc
1 + s
(31)
where Kdc = Aa is the system dc gain and = a1 is the system time constant. This is very useful
for analysing certain properties of dynamic systems in the frequency domain. But more of that
later!
17
6 FIRST ORDER SYSTEMS

Step Response
1
0.9
0.8
Amplitude
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.2
0.4
0.6
0.8
1.2
Time (sec.)
Figure 11: Fast and Slow First Order Step Responses
Normalising the time axis by division by , and normalising the response amplitude to a unity
maximum output, we achieve the following response plot.
Step Response
1
Amplitude
0.8
0.6
0.4
0.2
0
0
2
3
Time Constants
Figure 12: Normalised First Order System Step Response

From this we note that it takes 4 time constants for the output to reach 98% of the final value.
This we define as the settling time.
So
Ts =
4
a
(32)
We also define the rise time as the time for the response to rise from 10% to 90% of the final
18
7 SECOND ORDER SYSTEMS

value. By suitable manipulation of the exponential term we get:
Tr =
2.2
a
(33)
It is possible to estimate the transfer function of a first order system by examining the step
response. The method is simple. Apply a step input of known amplitude. The dc gain is
calculated from the steady-state output value:
Kdc =
steady state output amplitude

input amplitude
(34)
The time constant ( ) is the time to reach 63.3% of the final output value (by definition). So the
transfer function becomes:
G(s) =
Kdc
s+
(35)
Second Order Systems
Given that a system is second order (has two poles) with transfer function G(s), there are standard
expressions for the time and frequency response in terms special parameters which describe the
system dynamic properties.
The second order system is usually written in a standard form as:
Y (s)
n2
= 2
U (s)
s + 2n s + n2
(36)
where and n are the damping ratio and undamped natural frequency.
The two poles of the second order system can be either a complex-conjugate pair or can be both
real. The real parts are, in every case, negative for a stable second order system.
The poles of the second order system are thus given by:
s2 + 2n s + n2 = (s + )2 + d2
(37)
(s + ) = jd
(38)
s = jd
(39)
The factors of which are:
or
19

where:
= n
(40)
and
d = n
1 2
(41)
The constant is known as the Damping Factor or Damping Ratio of the system. The constant
d is the damped natural frequency (in radians per second). The pair of complex poles, for an
underdamped system, are located on the s-plane as shown below. Note that cos() = . The
angle is known as the damping angle.
Figure 13: Complex Poles in the s-plane

Thus , the damping ratio, is a dimensionless parameter which serves as an indicator of the
amount of damping or energy dissipation in a system. It can also be looked upon as the number
of natural oscillations that occur within one exponential time constant.
1 natural period (seconds)

2 exponential time constant
(42)
For > 1, the poles are purely real and we say that the system is overdamped. For < 1, the
system is underdamped and the poles are complex.
For the case = 1, the system is critically damped; = 0 and the poles are real and identical.
7.1
Step Response
One can now examine the step response properties of the second order system. For a unit step
input:
U (s) =
1
s
(43)
20

and
Y (s) =
n2
s(s2 + 2n s + n2 )
(44)
n2
s((s + )2 + d2 )
(45)
Bs + C
A
+
s
(s + )2 + d2
(46)
Completing the square gives:

Y (s) =
Partial fraction expansion yields:
Y (s) =
Multiplying by s on both sides of Equation (46) gives:

A+s
Bs + C
n2
=
(s + )2
s2 + 2n s + n2
(47)
Letting s = 0 gives A = 1. Multiplying by s((s + )2 + d2 ) on both sides of Equation (47) gives:

n2 = (s + )2 + d2 + s(Bs + C) = s2 (B + 1) + s(C + 2) + 2 + d2
(48)
i.e. B = 1, C = 2.
Hence:
Y (s) =
=
s + 2
1
s (s + )2 + d2
1
s+
2
2
s (s + ) + d
(s + )2 + d2
(49)
(50)
Then by using Laplace Transform Tables:
y(t) = 1 et cos d t
t
e
sin d t
d

en t p
1 2 cos d t + sin d t
=1 p
1 2
en t
=1 p
sin(d t + )
1 2
(51)
(52)
(53)
where
= tan1
1 2
(54)
21
Hence, the transient response of the system to a step input can be determined from a knowledge
of the positions of the system poles on the s-plane diagram.
NOTE. Depending upon which textbook you read, you will see different solutions to the ones
above. Do not be alarmed: they are just variations on trigonometric identities. They are all
correct!
Figure 14: Second Order System Unit Step Response
22
7.2
Time Response Specifications
1. Rise Time (Tr ).

This is normally defined as the time taken for the step response to rise to 100% of its final
value (for underdamped response). Also, it is the time taken for response to pass from 10%
to 90% final value for the overdamped case, but this will not be considered here.
Now,
Hence:
p
en Tr
y(Tr ) = 1 p
sin n 1 2 Tr + tan1
1 2
=
Tr =
d
tan1
n
For 0 < < 1 (underdamped systems),

0 < tan1
1 2
1 2
1 2
<
1 2
(55)
(56)
(57)
2. Time to first overshoot (Tp ).

By differentiating the equation for y(t) with respect to time (if you want to try this, its
most easily done by multiplying the output in the Laplace domain by s, then using the
transform tables), and setting the result equal to zero, we get:

p
n en t
2
p
sin n 1 t = 0
1 2
(58)
for various maxima or minima.

Hence:
p

sin n 1 2 t = 0
(59)
So the times for various peaks are given by:

n
and so for the first overshoot
1 2 Tp = 0, , 2, 3, . . .
Tp =
1 2
(60)
(61)
3. Percentage Overshoot (%O.S.)

For a unit step response, the percentage overshoot is a measure of the size of the first peak
- in other words, of the value of the output signal at the time of the first overshoot.
O.S = y(Tp ) 1
en Tp
= p
1 2
(62)
sin(d Tp + )
(63)
23

Now,
sin(d Tp + ) = sin d Tp cos + cos d Tp sin
and
(64)
= sin cos + cos sin
(65)
= sin
d
=
pn
= 1 2
(66)
n Tp = n
=
Hence
(67)
(68)
p
n 1 2
(69)
(70)
1 2
O.S. = y(Tp ) 1
= e
(71)
1 2
(72)
This expression can always be used to find the % overshoot in the step response of a second
order system which has an underdamped response.
Thus
%O.S. =
1 2
100 e
(73)
The inverse relationship is:

=
%O.S.
100
2 +ln2 %O.S.
100
ln
(74)
Figure 15 shows this relationship graphically.

100
90
80
70
%O.S.
60
50
40
30
20
10
0.1
0.2
0.3
0.4
0.5
zeta
0.6
0.7
0.8
0.9
Figure 15: Equation (74) in graphical form
8 ANALYSIS FOR CONTROL SYSTEM DESIGN
24
4. Settling time (Ts )

For the underdamped case, < 1, the time response is oscillatory. The settling time is the
time taken to reach and stay within 2% of the final value. This is very hard to calculate,
but a cursory glance at Equation (53) shows that the output time response to a unit step
consists of a sinusoidal term, whose value must by definition never exceed 1, multiplied by
an exponential term which causes the sinusoid to decay. Hence an approximate indication
of settling time can be obtained by working out when this exponential decay reaches within
2% of its final value.
Therefore:
en Ts
p
= 0.02
1 2
(75)
Thus en ts 0.02 for low value of , and therefore

Ts
4
n
(76)
Ts
(77)
or
(compare with Equation (32)).

5. Steady-State Error (Ess )
Now,
ess = lim {u(t) y(t)}
t
(78)
For a second order system with a unit step input,

ess = lim {1 o (t)} = 0
t
(79)
NOTE. This results is always true for the system with transfer function given by:
n2
Y (s)
= 2
U (s)
s + 2n s + n2
(80)
This is a little misleading though, as it implies that second-order systems never exhibit a
steady-state error. This is not the case, as we shall see.
Analysis for Control System Design
The design of a controller requires a set of design objectives or specifications to be drawn up.
They define the type of control, the design actions to be taken and, also, they give a set of test
objectives through the question:
Has the closed loop system met the specifications?
25
9 STEADY STATE ERRORS
The specification will, in general pertain to the required behaviour of the closed loop system. This
generally implies observations in the time-domain. However, there is the possibility of setting up
a specification which used frequency-based properties. More of this later.
Typical time domain specifications are:
1. Percentage overshoot.
2. Rise time.
3. Settling time.
4. Steady-state errors.
We have yet to address the last of these items. It is of such importance that it justifies its own
section. But, before doing this, we must ask why the time domain specification is couched in
terms associated with second order systems behaviour. It is not often the case that a complex
system is second order. Well, the reality is that many high order systems will have dominantly
second order dynamics. What does this mean?
Consider the case of a system which has the following poles:
s1 = 0.5 + j0.7
s2 = 0.5 j0.7
s3 = 5.0
The real pole, s3 , is ten times further into the left half s-plane than the complex conjugate (second
order) pair. This means that it will have completed its contribution to the overall response (as
e5t ) well before the second order response component. Remember that Ts , the system settling
time, is defined as
Ts =
=
(81)
4
real part of s
(82)
To all intents and purposes, the total system response, as seen by an operator, will look second
order in character. We say that this system has a dominant second order structure.
Steady State Errors
By way of introduction, consider the control of a robot arm which is involved with an assembly
process, typically an application within the automotive industry. The manipulator is required to
present a component up to a part-assembled engine, for example. If the task is to be properly
performed, the component must be located exactly where intended. From a control viewpoint,
26
the robot arm has been issued with a demanded position signal. Once the response of the robot
arm to the demand is completed, there must be no error signal as seen in the Figure 16:
positional
error
required
arm
position
achieved
arm
position
Robot Arm
Figure 16: Robot Arm Block Diagram
When the robot arm has stopped moving it is in its steady state, all transient motions will have
died away. If there exists a difference between the desired and achieved steady state demands,
then there is a steady state error.
This is an example for a position control system. The same applies to velocity control. For
example, if we are using a large metalwork lathe for machining a component, such as cutting a
screw thread, it is necessary for the lathe to turn with precise angular velocity as the cutting tool
moves down the length of the work-piece. So, if the lathe does not rotate at exactly the right
speed, there will be a steady state velocity error. We assume that the lathe has accelerated up
to the correct speed and any acceleration transients have died away.
So, steady state error performance is a measure of the accuracy of a dynamic system.
There are significant differences between the structures of the transfer functions for position and
velocity control systems. In practice, by observing these differences, it it possible to predict the
steady state error performance. To test these systems, we use standard input types. These are:
1. The unit step to represent a position demand
U (s) =
1
s
2. The unit ramp to represent a velocity demand

U (s) =
1
s2
3. The unit parabola to represent an acceleration demand

U (s) =
9.1
The Error Transfer Function
Consider the system of Figure 17.

We can write down
1
s3
27
U(s)
E(s)
+
G(s)
Y(s)
Figure 17: Typical Closed Loop System
E(s) = U (s) Y (s)
(83)
= U (s) E(s)G(s)
(84)
U (s)
1 + G(s)
(85)
e() = lim sE(s)
(86)
so
E(s) =
The Final Value Theorem states that
s0
sU (s)
s0 1 + G(s)
= lim
(87)
where e() is the steady state error.
9.2
Error Constants
For a step input, the steady state error is:
s 1s
s0 1 + G(s)
1
=
1 + lim G(s)
e() = lim
s0
We define lim G(s) , Kp where Kp is called the position error constant.

s0
For a ramp input, the steady state error is
(88)
(89)
28
s s12
s0 1 + G(s)
1
= lim
s0 s + sG(s)
1
=
lim sG(s)
e() = lim
(90)
(91)
(92)
s0
We define lim sG(s) , Kv where Kv is called the velocity error constant.

s0
For a parabolic input, the steady state error is
s s13
s0 1 + G(s)
1
= lim 2
s0 s + s2 G(s)
1
=
lim s2 G(s)
(93)
e() = lim
(94)
(95)
s0
We define lim s2 G(s) , Ka where Ka is called the acceleration error constant.

s0
9.3
System Type
We shall soon see that the steady state properties of a dynamic system are strictly dependent
upon the presence of integerator terms on the open loop transfer function. So important is this
property that one can define a system by it. The property is the system type and it is defined as
follows.
Given a system of the form shown in Figure 18:
U(s)
G(s)
Y(s)
Figure 18: A Unity Negative Feedback Closed Loop System

If we assume that the open loop transfer function is of the form:
G(s) ,
K(1 + Ta s)(1 + Tb s)
sm (1 + T1 s)(1 + T2 s)
(96)
Then the magnitude of m represents the number of pure integrations in the feedforward path.
The system is then Type m.
29
9.4
Type 0 Systems
If there are no integrations, the system is Type 0. For a step input the steady state error is:
s 1s
s0 1 + G(s)
1
=
1 + lim G(s)
e() = lim
step
(97)
(98)
s0
1
1 + Kp
1
1+K
,
=
(99)
(100)
So the position error constant for a Type 0 system is finite, as is the steady state error.
For a ramp input, the steady state error is:
e() =
ramp
1
lim sG(s)
(101)
s0
1
Kv
1
=
0
=
,
(102)
(103)
(104)
So the velocity error constant for a Type 0 system is zero, and the steady state error is infinite.
For a parabolic input, the steady state error is:
e()
parabola
1
lim s2 G(s)
(105)
s0
1
Ka
1
=
0
=
,
(106)
(107)
(108)
So the acceleration error constant for a Type 0 system is zero, and the steady state error is
infinite.
30
9.5
Type 1 Systems
If there is one integration in the feedforward path, the system is Type 1. For a step input the
steady state error is:
s 1s
s0 1 + G(s)
1
=
1 + lim G(s)
e() = lim
step
(109)
(110)
s0
1
,
1 + Kp
1
=
1+
= 0
(111)
(112)
(113)
So the position error constant for a Type 1 system is infinite, with a consequent zero steady state
error.
e() =
ramp
1
lim sG(s)
(114)
s0
,
=
1
Kv
1
K
(115)
(116)
So the velocity error constant for a Type 1 system is finite, as is the steady state error.
e()
parabola
1
lim s2 G(s)
(117)
s0
1
Ka
1
=
0
=
,
(118)
(119)
(120)
infinite.
31
9.6
Type 2 Systems
If there are two integrations in the feedforward path, the system is Type 2. For a step input the
steady state error is:
s 1s
s0 1 + G(s)
1
=
1 + lim G(s)
e() = lim
step
(121)
(122)
s0
1
,
1 + Kp
1
=
1+
= 0
(123)
(124)
(125)
So the position error constant for a Type 2 system is infinite, with a consequent zero steady state
error.
e() =
ramp
1
lim sG(s)
(126)
s0
1
Kv
1
=
= 0
,
(127)
(128)
(129)
So the velocity error constant for a Type 2 system is infinite, with a consequent zero steady state
error.
e()
parabola
1
lim s2 G(s)
(130)
s0
,
=
1
Ka
1
K
(131)
(132)
infinite.
32
9.7
System Type, Error Constant and Steady State Errors
In summary, we present the following table:

Input Type
Steady State Error
System Type
0
1
2
m13
9.8
9.8.1
1
s
1
1+Kp
1
s2
1
Kv
1
s3
1
Ka
1
1+K
0
0
0
1
1+K
0
0
1
1+K
Examples
Type & Order
a
G(s) =
s+3
s(s + 2)
A type 1, 2nd order system.

b
G(s) =
s2 (s
(s + 4)(s + 6)
+ 3)(s + 4)(s + 5)(s + 7)
A type 2, 5th order system (note the pole/zero cancellation).
9.8.2
Steady State Errors
For this example system:
G(s) =
K
s(1 + T s)
Position Error constant is:

Kp = lim G(s)
s0
= lim
s0
K
s(1 + T s)
K
0
=
=
The steady state error is:

1
1+
= 0
e() =
step
33
10 STABILITY
A type 1 system has zero steady state position error.
Velocity Error constant is:
Kv = lim sG(s)
s0
= lim
s0
sK
s(1 + T s)
= K
e() =
ramp
1
K
A type 1 system has a finite steady state velocity error.

Acceleration Error constant is:
Ka = lim s2 G(s)
s0
= lim
s0
s2 K
s(1 + T s)
= 0
e()
parabola
1
0
=
=
A type 1 system has an infinite steady state acceleration error.
10
Stability
An important attribute of most controlled systems is the stability. We say that a system is unstable
if, for a bounded input, the output is divergent (unbounded) with time. The one exception to this
is the pure integrator, for which the notion of stability is inappropriate. A system is marginally
stable if, for a bounded input, it oscillates with constant amplitude.
The location of the poles in the s-plane defines the stability in an analytic sense. Figure 19 shows
this graphically.
We call this the absolute stability. Another term, relative stability describes how the transient
response of a system behaves. It is normally associated with the level of damping.
For the moment, we shall concern ourselves with absolute stability and how to check it. We ask
the question: For a closed loop system, how can we check that the real part of every closed loop
pole is negative, i.e. lies in the left half s-plane?
The answer is to generate a Routh Array and then apply the Routh-Hurwitz Stability Criterion
34
10 STABILITY
Figure 19: The Relation between Pole Position and Absolute Stability
to it.
10.1
Generating a Routh Array
Begin by writing down the denominator of the closed loop transfer function - the characteristic
polynomial.
a0 sn + a1 sn1 + a2 sn2 + + an2 s2 + an1 s + an
All of the coefficients {ai } must be the same sign (normally positive) but that is not a guarantee
of stability. It is a necessary but not sufficient condition.
Now form a Routh array.
sn
sn1
sn2
sn3
sn4
..
.
s0
a0
a1
b1
c1
d1
..
.
..
.
a2
a3
b2
c2
a4
a5
b3
a6
a7
The complete array has n + 1 rows, padded out with zeroes to the right as elements disappear.
the bi , ci , di , etc. elements are completed according to the following formulae:
35
11 THE ROUTH-HURWITZ STABILITY CRITERION
a1 a2 a0 a3
a1
a1 a4 a0 a5
a1
b1 =
b2 =
..
.
b1 a3 a1 b2
b1
b1 a5 a1 b3
b1
c1 =
c2 =
..
.
c1 b2 b1 c2
c1
d1 =
..
.
11
The Routh-Hurwitz Stability Criterion
Having completed the Routh array, inspect the left hand column:

a0
a1
b1
c1
d1
..
.
Firstly, there should be no zero elements. For this case, see later.
The criterion states that for a stable system, there will be no sign changes down this column. For
every instance of a sign change, there will be one right half s-plane (unstable) pole.
11.1
Some Simple Examples
Example 1.
s4 + 2s3 + 6s2 + 7s + 5 = 0
6
5
s4 1
7
0
s3 2
2510
=
2.5
=
5
0
s2 2617
2
2
=3
s1 2.5725
2.5
s0 5
0
0
0
0
It is allowable to scale, by a positive integer, any row, in order to facilitate subsequent arithmetic:
36

s4 1
6
7
s3 2
2510
=
2.5
=5
s2 2617
2
2
2
s 5
10
s1
57210
5
=3
s0 5
5
0
0
0
This system is stable because there are no sign changes in the left-hand column.
Example 2.
+ -
1
s ( s + s + 1)
Figure 20: Example 2

For the system of Figure 20, what is the range of values of K for which the closed loop system is
stable?
The closed loop transfer function is:
G(s) =
=
K
s(s2 +s+1)
K
+ s(s2 +s+1)
s3
K
+s+K
s2
The closed loop characteristic equation is

s3 + s2 + s + K = 0
and the Routh array is:
s3
s2
s1
s0
1
1
1K
K
1
K
0
0
By inspection of the left column, K must be positive (s0 element) and K must be less than one
(s1 element). i.e. 0 < K < 1.
11.2
Problem Conditions
Problems arise if there is a zero element in the left column of the Routh array. To compound this,
the whole row could be comprised of zero elements. We shall consider each condition in turn.

11.2.1
37
First Term in a Row is Zero
There are two solutions:

substitute a small positive number, for the zero and then carry on with the array evaluation.
replace s by 1q in the characteristic equation. The number of q-roots in the right half s-plane
is the same as s-roots.
An Example
s5 + s4 + 2s3 + 2s2 + 3s + 5 = 0
The Routh Array becomes:
s5 1
2 3
s4 1
2 5
3
s 0 2 0
Replace the s3 zero element with and carry on:
s3
s2
s1
s0
As 0 the s2 term
2+2
2+2
4452
2+2
2 0
5
is positive.

2
As 0 the s1 term 445
has a limiting value of 2.
2+2
Therefore, there are two sign changes down the left hand column the system is unstable with
two right half s-plane poles.
Alternatively, rewrite the characteristic equation in terms of 1q . . .
1
2
2
3
1
+ 4 + 3 + 2 + +5 = 0
5
q
q
q
q
q
5q 5 + 3q 4 + 2q 3 + 2q 2 + q + 1 = 0
The Routh array becomes:

q5
q4
q3
q2
q1
q0
5
3
43
2
2
32
1
0
0
1
2
2
1
38
1
1
0
0
0
0
So, with two sign changes, this forces the same conclusion as before: the system is unstable with
two right half s-plane poles.
11.2.2
A Complete Row of Zeroes
This occurs because there are symmetrically-placed roots, either side of the imaginary axis, or
conjugate pairs of roots aligned along the imaginary axis.
Consider the following:
s6 + 2s5 + 8s4 + 12s3 + 20s2 + 16s + 16 = 0
The Routh Array for this is:
s6
s5
s4
s3
1
2
2
0
8
12
12
0
20
16
16
0
16
0
0
0
If we take the s4 row, above the zero row, and generate the auxiliary polynomial, A(s), where
A(s) = 2s4 + 0s3 + 12s2 + 0s + 16
Simplifying this, through division by 2, yields:
A(s) = s4 + 6s2 + 8
We then calculate the derivative of this polynomial,
d A(s)
= 4s3 + 12s
ds
and replace the zero elements in the s3 row of the Routh array with the appropriate coefficients
from d A(s)
ds , the Routh array becomes:
s4
s3
s2
s1
s0
2
4
6
4
3
16
12 16 0
12 0 0
16 0
0
39
12 STABILITY ANALYSIS USING THE BODE PLOT

Clearly, there are no sign changes. However, the solution to A(s) = 0 yields roots at:
s12 = 2j
s34 = 2j
These two pairs of roots are also roots of the original characteristic equation. Thus we conclude
that the system is marginally stable, as none of the other roots lie in the right half s-plane.
12
Stability Analysis using the Bode Plot
You will now be very familiar with the plotting of the frequency response of a system (probably
an electrical network or filter) using the Bode plot. Control engineers make good use of the Bode
plot to characterise and aid in control design.
Here we plot the open loop transfer function steady state gain and phase responses to sinusoidal
signals over a range of frequencies. The open loop transfer function form is very important. It
must be that transfer function which, if unity negative feedback is applied, generates the desired
closed loop transfer function.
U(s)
G(s)
Y(s)
Figure 21: A Unity Negative Feedback Closed Loop System

here we would plot the frequency response of G(s).
U(s)
G(s)
Y(s)
H(s)
Figure 22: A Non-Unity Negative Feedback Closed Loop System
But for the system of Figure 22, we have to generate the equivalent unity negative feedback form,
which is shown below in Figure 23.
12.1
Bode Plot Structure
To remind you, there are two elements to a Bode plot:
40
U(s)
1
H(s)
G(s)H(s)
Y(s)
Figure 23: The Equivalent Unity Negative Feedback System
1. Gain in dBs vs frequency in radians per second on a logarithmic scale

2. Phase in degrees vs frequency in radians per second on a logarithmic scale
We can decompose the open loop system into simple elemental blocks made up of:
constant gain (K)
integration factors
1
s
differentiation factors (s)

first order lead terms (1 + T s)

1
first order lag terms 1+T
s

second order lead terms 1 +
second order lag terms
2
s2
2
n s + n
1
2
1+ 2 s+ s 2
n
We plot each constituent part individually and then add all curves (for gain and then phase) to
form a composite Bode plot for the open loop system.
Often, control engineers use a simple approximation to the constituent curves to generate a
sufficiently useful composite shape to enable first approximation analysis of system stability (or
otherwise). The approximation is called the asymptotic approximation.
We shall now consider each constituent part, and its true and approximate response shape.
12.2
Constant Gain (K)
The gain factor is simple, as it is constant for all frequencies in both gain and phase. For the
gain plot it is
20log10 (K)
There is no phase shift at any frequency. Hence the Bode plot becomes:
The asymptotic approximation is the same as the actual Bode plot!
(133)
41
Phase (deg); Magnitude (dB)
20 log10(K) dB
10
1
0.5
0.5
1
1
10
10
10
10
Frequency (rad/sec)
Figure 24: A Simple Gain Bode Plot
12.3
Integration Factors
1
s
The gain factor is frequency-dependant according to:

1
20log10 = 20log10
j
(134)
whilst the phase response is a constant 90 degree lag at all frequencies. Hence:
Bode Diagrams
20
10
10
20
89
89.5
90
90.5
91
1
10
10
10
Frequency (rad/sec)
Figure 25: Bode Plot for an Integration Term

Again, the asymptotic approximation is the same as the actual Bode plot.
12.4
Differentiation Factors (s)
The gain factor is frequency-dependant according to:

20log10 |j| = 20log10
(135)
42
whilst the phase response is a constant 90 degree phase advance (lead) at all frequencies. Hence,
Figure 26:
20
10
10
20
91
90.5
90
89.5
89
1
10
10
10
Frequency (rad/sec)
Figure 26: Bode Plot for a Differentiation Term

Yet again, the asymptotic approximation is the same as the actual Bode plot.
12.5
First Order Lead Terms (1 + T s)
These are frequency dependant in both gain and phase. For the gain factor, the relationship is:
20log10 |(1 + sT )| = 20log10
1 2 T 2 = 10log10 1 2 T 2
and for the phase, the relationship is
phase = tan1 (T )
(136)
(137)
The plot of the actual response plus the straight-line asymptotic approximations are shown at
Figure 27 overleaf.
12.6
First Order Lag Terms
1
1+T s
Like the lead terms, these are frequency dependant in both gain and phase. For the gain factor,
the relationship is:

p

1
= 20log10 1 2 T 2 = 10log10 1 2 T 2
20log10

(1 + sT )
(138)
and for the phase, the relationship is
phase = tan1 (T )
(139)
43

25
20
15
10
5
+20dB/decade slope
0
1/T
5
90
75
60
45
+45deg/decade slope
30
15
0
Frequency (rad/sec)
Figure 27: Bode Plot for a 1st Order Lead Term
Figure 28:
5
1/T
0
10
15
20dB/decade slope
20
0
15
30
=45dB/decade slope
45
60
75
90
Frequency (rad/sec)
Figure 28: Bode Plot for a 1st Order Lag Term
12.7
Second Order Lead Terms 1 +
2
n s
s2
n2
In both gain and phase, these are frequency dependant according to the following relationships.
For the gain:
and for phase:
s

2

2 2
2 2
2
s

1 2
+
20log10 1 +
s + 2 = 20log10
n
n
n
n
phase = tan
2
n
2
1 2
n
(140)
(141)
44
Figure 29:
40
30
+40dB/decade slope
20
10
n
10
180
150
120
90
60
30
0
Frequency (rad/sec)
Figure 29: Bode Plot for a 2nd Order Lead Term
12.8
Second Order Lag Terms
1
2
1+ 2n s+ s 2
In both gain and phase, as with the equivalent lead terms, these are frequency dependent, according to the following relationships. For the gain:
and for phase:

20log10
1+

s

2 2
2 2
1

= 20log10
+
1 2
2
s2
n
n
s
+
2
n
(142)
phase = tan
2
n
2
2
n
(143)
Figure 30:
12.9
Gain and Phase Margins
We can use the Bode plot to inspect the closed loop stability. Remembering that the Bode plot
is the open loop frequency response plot, assuming unity negative feedback is to be applied. If
the loop is closed around the open loop system with no extra gain, then the closed loop stability
properties can be predicted from the open loop Bode plot using the stability margins.
These are defined diagrammatically as follows:
So the gain margin is the amount of extra gain necessary to provide an open loop gain of unity
at the frequency where there is 180 deg of lag. Similarly, the phase margin is the amount of extra
phase lag necessary at the frequency of unity gain, such that the total phase shift is 180 deg of
45
Figure 30: Bode Plot for a 2nd Order Lag Term
GAIN
log w
0dB
gain margin
phase margin
log w
180 deg
lag
PHASE
Figure 31: The Stability Margins
lag. Provided that there is a positive gain margin and a positive phase margin, the closed loop
system will be stable.
The stability margins provide information about not just absolute but also relative stability.
There is a mathematical relationship between the phase margin of the open loop system and the
damping ratio of the closed loop system after unity negative feedback is applied. It is:
phase margin = tan1
It can be represented graphically as follows:
2
q
p
2 2 + 1 + 4 4
(144)
46
13 STABILITY ANALYSIS USING THE NYQUIST PLOT

90
80
70
phase margin
60
50
40
30
20
10
0.2
0.4
0.6
0.8
1
1.2
damping ratio
1.4
1.6
1.8
Figure 32: Phase Margin vs Damping Ratio
We note from Figure 32 that, as a reasonable approximation, for every 10 deg of phase margin,
this corresponds to a 0.1 contribution to the damping ratio of the closed loop system, up to a
maximum phase margin of 60 deg.
13
Stability Analysis using the Nyquist Plot
An alternative method of presenting the same frequency response information as in the Bode
plot (ie open loop and assuming unity negative feedback) is to use the Nyquist plot. This is a
polar plot where the length of a line extending from the origin of the polar plot provides gain
information. The phase information is depicted as a rotational displacement of this line from
the positive horizontal axis, with a lag generating a clockwise rotation. By joining up the tips of
these vectors, as frequency increases, a contour is generated. This is the Nyquist contour or plot.
As an example, consider the system with the following open loop transfer function:
G(s) =
10
s + 10
(145)
We use the same equations to calculate the frequency response data. Namely, for gain:
and for the phase:

1
1

(1 + sT ) = 1 2 T 2
phase = tan1 (T )
(146)
(147)
and the plot is:

All of the previous information on generating frequency response data still holds. The shape of
the plot is the only difference. Also, just as there is a method for approximating Bode plots, so
there is a quick way to sketch the Nyquist plot.
47
0.2
0.1
low
frequencies
high
frequencies
Imaginary Axis
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.2
0.2
0.4
0.6
0.8
Real Axis
Figure 33: The Nyquist Plot for a Simple Lag
13.1
Rapid Sketching of Nyquist Plots
Firstly, we must reconsider the system type and order, as these set up the basic shape of the
plot. Remember, the system type is the number of free integration terms in the open loop
transfer function, whereas the system order denotes the order of the highest power of s in the
characteristic polynomial of the same transfer function. For every integration term, there will be
a 90 degree phase lag at all frequencies. For each lag term, there will be no lag contribution at
zero rad/sec, but there will be 90 degrees of lag at infinite frequency. Similar arguments hold for
differentiation and lead component, but they will contribute phase advances (leads).
So, for a Type 0 system, there will be no phase shift at zero rad/sec. The plot will commence
on the positive real axis, at a point equivalent to the dc gain of the system (note not in dBs).
If the system is Type 1, then the dc gain is infinite (think about this), and there will be 90
degrees of phase lag. Therefore, the plot will commence at a point infinitely down the negative
imaginary axis. As the frequency increases, the contribution of other lags would move the plot
further clockwise and closer to the origin. Lead terms would move the plot counter-clockwise in
a similar fashion.
13.1.1
Some Nyquist Plot Examples
We start with a pure integrator:

1
s
(148)
10
s(s + 10)
(149)
G(s) =
The plot for this is at Figure 34.
Now a Type 1, 2nd order system:
G(s) =
Figure 35 shows the Nyquist plot.
48

1
0.8
0.6
Imaginary Axis
0.4
0.2
0.2
0.4
0.6
0.8
1
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Real Axis
Figure 34: Nyquist Plot for a Pure Integrator
high frequencies
Imaginary Axis
low frequencies
4
to infinity
5
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Real Axis
Figure 35: Nyquist Plot for a Type 1, 2nd Order System
13.2
Nyquist Stability Criterion
By monitoring the position of the open loop system Nyquist contour, as it moves from low to
high frequency, it is possible to determine the stability of the unity negative feedback closed loop
system. Essentially, if the total open loop gain is less than unity when the total phase lag is 180
degrees or more, the system will be closed loop stable. Similarly, if the phase lag is less than
180 degrees as the gain exceeds unity, the system will be closed loop stable. This translates into
a simplified version of the Nyquist Stability Criterion which states that, as the Nyquist contour
moves from low to high frequency, if it leaves the 1 + j0 point (the Nyquist point to the left,
the system will be closed loop stable.
The full Nyquist stability criterion states that:
for a closed loop system with unity negative feedback to be stable, it is a necessary and sufficient
condition that the contour of the open loop frequency response, plotted as a polar diagram, describes a number of anticlockwise encirclements of the 1 + j0 point as varies from to
+, which is not less than the number of poles of the open loop system with positive real parts.
49
The following Nyquist plots illustrate the possible stability conditions (Figure 36 to Figure 39).
-1
Figure 36: A Stable Condition for All Gain Values
-1
Figure 37: An Unstable Condition
-1
Figure 38: A Conditionally Stable Condition
-1
Figure 39: Another Conditionally Stable Condition
13.3
Stability Margins and the Nyquist Plot
As with the Bode plots, it is possible to define stability margins on the Nyquist plot. These are
depicted graphically below (Figure 40).
50
unit circle
G.M.
|G|
P.M.
Figure 40: Nyquist Stability Margins
From Figure 40, it is possible to calculate the dB equivalent gain margin as:
G.M.(dB) = 20 log 10
= 20 log10
13.4
1
|G|
|G|
(150)
(151)
More Accurate Sketching of Nyquist Plots
It is possible to produce rapid sketches of the Nyquist plot of a system and add in appropriate
critical plot coordinates to enable the establishment of stability margins. We shall illustrate this
with the following example (Figure 41) where:
G(s) =
U(s)
500
(s + 1)(s + 3)(s + 10)
G(s)
(152)
Y(s)
Figure 41:
We begin by defining:
G(j) =
500
(j + 1)(j + 3)(j + 10)
(153)
So
G(j) =
(14 2
500
+ 30) + j (43 3 )
(154)
51
14 ROOT LOCUS ANALYSIS

multiply top and bottom by complex conjugate of denominator above:
G(j) =

500 14 2 + 30 j 43 3
(14 2 + 30)2 + (43 3 )2
(155)
At = 0
G(j) =
500
50
=
30
3
As q, the real part of the numerator remains +ve, whilst the imaginary part goes q
ve. At
30
= 14 , the real part 0 whilst the imaginary part remains ve. Substituting = 30
14 into
the equation yields 8.3600j.
At = 43, the imaginary part 0 and the real part goes negative. The substitution of
= 43 yields 0.874.
As ,
G(j)
500j
3
or zero at 270 deg.

Now sketching the Nyquist plot and adding the details, we get (Figure 42):
50/3
-0.874
w =0
w = 43
-0.0167j
w=
30
14
Figure 42: A More Accurate Nyquist Plot Sketching
14
Root Locus Analysis
In summary, poles must lie in the left-half s-plane for absolute stability. If they are, their position
in the left-half s-plane tells us about relative stability: how fast the system oscillates and how
quickly any transients die away..
Take the simple system below:
52
+
R(s)
G(s)
C(s)
Figure 43: A Unity Negative Feedback Control System
We know that
C(s)
K G(s)
=
R(s)
1 + K G(s)
(156)
and that the roots of 1 + K G(s) = 0 give us the information we need.

If G(s) is given, then it is K that determines the closed loop stability. If we vary K from
0 , where do the closed loop poles lie? We could calculate pole positions by trial and error.
Alternatively, we could use the Root Locus technique.
Using simple rules, we construct the paths (loci ) that the closed loop poles take as K varies. Not
only can we use the Root Locus approach to establish stability and transient response, but also
we can design controllers that shift the subsequent closed loop poles where we want them.
There are two criteria that are used to establish the rules for determining the closed loop root
loci. We know that 1 + K G(s) = 0 must be satisfied by all values of s that correspond to closed
loop system poles. WE can rewrite this as:
K G(s) = 1
(157)
K G(s) = ej
(158)
which in polar form is:
or, more rigourously,

K G(s) = ej(2n+1)
s.t. n {0, 1, 2, . . .}
(159)
This last can be further subdivided into two criteria:

1. Gain Criterion (G.C.).
|K G(s)| = 1
(160)
2. Angle Criterion (A.C.).

K G(s) = (2n + 1)
n Z+
(161)
53
One way to move forward would be to select poles randomly and test them against A.C. and
G.C.
14.1
How Do We Check the Angle Criterion?
In the following example:
+jw
-2 + 2j
33.7o
o x
-s
-5
45
-4
116.6
x
-1
Figure 44:
at the point s = 2 + 2j the angle given by
K G(s) = +33.7o 45o 116.6o = 127.9o
(162)
which does not satisfy the A.C. Therefore, 2 + 2j is not a pole of the closed loop system
corresponding to a particular value of K.
Clearly, this trial-and-error approach does not work. We need a systematic method for sketching
the loci. The Root Locus approach does this by using some simple sketching rules.
14.2
Sketching Rules
1. Loci start at open loop poles (@ k = 0) and end at open loop zeros (@ K = ).
2. If the number of open loop poles exceeds the number of open loop zeros, then the excess
of poles migrate to zeros at and do so along asymptotes.
3. The locus lies along the real axis wherever the sum of poles and zeros to the right is odd.
4. The root locus diagram is symmetrical about the real axis.
5. The loci terminating at zeros at tend towards asymptotes having directions according
to
A =
(2n + 1)180o
excess of poles
(n = 0, 1, 2, . . .)
6. The asymptotes intersect on the real axis at a centre of gravity, given by
(163)
Sc.g. =
poles zeros
excess of poles
(164)
7. Dual (complex) root locus branches depart from or arrive at the real axis at 90o .
8. The breakaway/ break-in points for complex branches on the real axis are found by solving
for s in the following equation:
m
X
i=1
1
s + zi
n
X
i=1
1
s + pi
(165)
9. To find the angle of departure of the locus from a complex pole, consider the total angle
contributions from each open loop zero and the remaining open loop poles as follows:
p1 = z1 + z2 + z3 p2 p3 180o
(166)
where
s
qp 1
x
qz3
qz2
qp 2
qz1
o
jw
qp 3
x
10. To find angles of arrival at a finite complex zero, a similar relationship must be invoked:
z1 = p1 + p2 + p3 z2 z3 180o
(167)
where
s
o qz1
x o x
qp 2 qz3 qp 3
qp 1
jw
qz2
11. Imaginary axis crossings are found using the Routh array and the Routh-Hurwitz stability
criterion.
12. The gain, K, for any point on the locus is found using:
Q
finite pole lengths
Q
finite zero lengths
54
(168)

Introduction To Control Engineering: Andy Pomfret and Tim Clarke

Uploaded by

Copyright:

Available Formats

Introduction To Control Engineering: Andy Pomfret and Tim Clarke

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Control Engineering: Andy Pomfret and Tim Clarke

Uploaded by

Copyright:

Available Formats

What are the main topics covered in the document?

What are the main topics covered in the document?

How are dynamic systems modeled mathematically?

How are dynamic systems modeled mathematically?

Introduction To Control Engineering

Andy Pomfret and Tim Clarke

2 How is a Control System Developed?

5 Modelling Dynamic Systems

6 First Order Systems

7 Second Order Systems

9 Steady State Errors

12.9 Gain and Phase Margins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

14 Root Locus Analysis

How is a Control System Developed?

The Laplace transform

L{f (t)} = F (s) =

4 THE LAPLACE TRANSFORM

Laplace Transform Pairs

Laplace Transform Theorems

This assumes initial conditions are all zero.

4 THE LAPLACE TRANSFORM

Figure 2: Theorems For Use with Laplace Transforms

4 THE LAPLACE TRANSFORM

s + 2 = A(s + 4)(s + 6) + Bs(S + 6) + Cs(S + 4)

5 MODELLING DYNAMIC SYSTEMS

Using the transform tables this becomes

We can use MATLAB to do most of this donkey-work.

Important properties of transfer function

1. Initial conditions are assumed to be zero.

Modelling Dynamic Systems

5 MODELLING DYNAMIC SYSTEMS

Figure 3: A Simple Mass-Spring-Damper System

K is the linear spring constant

It is now possible to manipulate this using simple algebraic rules:

5 MODELLING DYNAMIC SYSTEMS

Poles and zeros

Modelling of Linear Electrical Networks

Problem: Find transfer function,

from the circuit of Figure 4.

5 MODELLING DYNAMIC SYSTEMS

Substitution of Equation (19) into Equation (18) gives

Modelling of Mechanical Systems

Impedance (Zm (s) =

5 MODELLING DYNAMIC SYSTEMS

A More Complex Example

Consider the system of Figure 5.

Figure 5: A More Complex Linear Translational System

Figure 6: Forces on M1 from moving it to the right

5 MODELLING DYNAMIC SYSTEMS

Figure 7: Forces on M1 from moving M2 to the right

Figure 8: Forces on M2 from moving it to the right

Figure 9: Forces on M2 from moving M1 to the right

5 MODELLING DYNAMIC SYSTEMS

Modelling of Rotational Dynamics

Impedance (Zm (s) =

Table 3: Torque-angular displacement characteristics, and corresponding impedances, for

Consider a mass being made to rotate at an angular velocity, by a torque, T, as show in

Figure 10: A Simple Linear Rotational System

T (s) = Js2 (s) + Bs(s) + K(s)