Chapter 1 Systems & Control Theory

1
SCI2110
SYSTEM & CONTROL THEORY
Introduction [B1,B3, K1, K3.3-3.6]
A system is an assemblage of interacting elements (called
state variables).
A control system is a system with clearly defined control
input and observed output. Usually, some objectives or
goals are associated with the control system. Thus the
problem is to find suitable input to a system so as to
achieve desired output.
E.g.: Consider the economy of Hong Kong as a system with
control input being the interest rate, and the observed output
being the rate of inflation. The objective may be to control
the rate of inflation by adjusting the interest rate. This is in
fact how many countries attempt to control their economies.
What do you think are the state variables here?
What other outputs of the economy would be of
interest to you?
Chapter 1 Systems & Control Theory
2
There are two kinds of control systems:
a) open-loop
disturbances
Control
Rule
output
objectives
The
System
input
y(t)
u(t)
In an open-loop control system, the input u(t) is
determined by the objectives and is not influenced by
the output y(t) directly.
b) Closed-loop
disturbances
Feedback
Control
Rule
objectives The
System
y(t) u(t)
Sensors
measurement errors
In a closed-loop system, the input u(t) is modified based
on the information about the system provided by the
output y(t), which is feedback into the system.
Chapter 1 Systems & Control Theory
3
Open-loop linear control systems are easier to analyze and
control (Chapters 2-4) but can go wrong especially if the
systems are subjected to too much external disturbances.
Closed-loop control systems can adapt to changing
circumstances (e.g. caused by external disturbances), but
are more difficult to analyze (see Chapter 6) and can
become unstable (see Chapter 5).
Classification of Systems
Chapter 1 Systems & Control Theory
4
Examples
a) RC-circuit
From circuit theory, we have the governing equation
1 1
( )
dy
y u t
dt RC RC
+
Let us introduce a state variable
1
( ) ( ) x t y t
Then
1 1
1 1
( ) ( ) ( ) x t y t u t x
RC RC
! !
or
1 1
1
, (state equation)
, (output equation)
x Ax Bu
y Cx Du
+
+
!
where
1 1
, ,
1, 0.
A B
RC RC
C D


Chapter 1 Systems & Control Theory
5
b) Dam-control
Consider a pair of dams below:
Q
0
depends on the weather and other geological properties
and is usually only measured at some regular intervals (i.e.
this is a discrete-time system).
Suppose we measure the water depths
1
x and
2
x at times
1, 2, 3, , t " then by conservation of flow and assuming
that the volumes of water contained by the dams are
proportional to
1
x and
2
x respectively, we have
[ ]
[ ]
1 1 0 1
2 2 1 2
( 1) ( ) ( ) ( )
( 1) ( ) ( ) ( )
x k x k Q k Q k
x k x k Q k Q k

+ +
+ +
Note that we have difference equations for discrete system
as opposed to the usual differential equations for
continuous systems.
Chapter 1 Systems & Control Theory
6
c) Driven Pendulum
By Newtons laws:
sin ,
g
u
l
+
!!
This is a non-linear system, since sin is non-linear.
However, for small oscillations 0 , we can linearize: i.e.
we make the linear approximation sin . The above
equation can now be written in the standard form:
Let
1 2
, and , x x
!
then
2
1 1
2 2 1
1 1
2 2
0 1
0
1 0
x
x x
d
g
x x dt x u
l
x x
u A Bu
g
x x
l

_
_ _ _




+
, , ,
,
_
_ _ _

+ +

, , ,
,
!
!
!!
!
Chapter 1 Systems & Control Theory
7
d) Population Dynamics
!Study of the population of northern spotted owls in 1990
!There owls were reported threatened with extinction due
to logging in the old-growth forests in the Pacific
Northwest
!Let j
k
denote population of juvenile female owl (up to 1
year old) in year k, s
k
denote population of sub-adult
female owl (1 to 2 years) in year k, and a
k
denote
population of adult female owl (over 2 years) in year k
!Lamberson et al. carried out demographic studies and
field data were modelled by the discrete time state
equation (called Stage-matrix Model):
1
1
1
0 0 0.33
0.18 0 0
0 0.71 0.94
k k
k k
k k
j j
s s
a a
+
+
+
1 1 1
1 1 1

1 1 1
1 1 1
] ] ]
!Using this model, researchers can study the effect of
different protection startegies (e.g. should we concentrate
on saving the eggs from predators or protect the adults
from poachers?)
Chapter 1 Systems & Control Theory
8
State variable description
From the above examples, we see that the system can be
written in the form of state equation & output equation
, (state equation)
, (output equation)
x Ax Bu
y Cx Du
+
+
!
where u is the control input vector, y is the output vector, and
x is the state variable vector. In general,
1
n
x
x
x
_

,
# is a n-
dimensional vector, where each state variable
i
x represents the
internal mutually interacting elements of the system, and the
state equation describes their dynamics.
The matrices [ ] , , , A B C D form a state model for the system.
The matrix A is called the state matrix, B is the control or
input matrix, C is the observation or output matrix, and D is
called direct transmission matrix.
If , , , and A B C D are constant matrices, the state model is said
to be time-invariant. Time-invariant systems are much easier
to analyze and we shall concentrate on such systems in this
subject.
Chapter 1 Systems & Control Theory
9
Non-uniqueness of the state variable description
Given any system with the state model [ ] , , , A B C D , we can
always find another equivalent state variable description
, , , 0 A B C
1
]
$ $ $
by carrying out the following transformations:
(i) re-define observation output
, y y Du
so that the output equation now becomes
y Cx .
[from now on, we shall always eliminate the direct
transmission part this way]
(ii) Let P be an n n invertible matrix. We can re-define new
state variables
1
x P x

$
The state equation now becomes
1 1
1 1 1
( ) x P x P Ax Bu
P APP x P Bu
Ax Bu


+
+
+
!
$ !
$ $
$
Chapter 1 Systems & Control Theory
10
And output equation is now
1
y y Cx CPP x Cx

$
$ $
where
1
, A P AP

$
1
, B P B

$
and C CP
$
.
Hence the same system can now be represented by the
new state model
1 1
, , , 0 , , , 0 A B C P AP P B CP

1 1
] ]
$ $ $
! Physically, the non-uniqueness of the state variable
description means that systems with different
interacting elements (state variables) may actually
have the same state model and their observed outputs
will behave in the same way.
! Mathematically, we can choose to represent the same
system in different ways by using different invertible
matrices P.
! Usually choose P to diagonalize the state matrix A,
i.e.
1
P AP

becomes a diagonal matrix much easier

state equation to solve!
Chapter 1 Systems & Control Theory
11
State (or Simulation) Diagrams
The state space description can be put into graphical form by
using the following symbols:
1. multiplier
a
2
x
1 2
x ax
2. adder
2
x
1 2 3
x x x +
3
x

sometimes, a negative sign at the end of the arrow is used

to represent a negation (or inversion) operation e.g.
2
x
1 2 3
x x x
3
x

3. Delay (for discrete time system)

2
x
1 2
( ) ( 1) x t x t
Chapter 1 Systems & Control Theory
12
4. integrator

2
x
0
1 2 1 0
( ) ( ) ( )
t
t
x t x d x t +

1 0
( ) x t

initial condition
For convenience, the integrator is often simplified as follows:

1 0
( ) x t

All these units can be realized in hardware (e.g. RC circuit for

integrator) thus state models can be implemented in analogue
computers (e.g. once used in Mercedes Benz cars!)
The software Simulink can accept state diagrams (plus other
non-linear functions such as conditional switches) and
simulate the behaviour of the system. Output can be viewed as
graphs etc. (see the subject SCI3210 Simulation).
Chapter 1 Systems & Control Theory
13
RC-circuit: (Single input-single output example)
1 1
, x x u y x
RC RC
+ !
Driven Pendulum: (Single input-multiple output example)
[Let
2 g
l
. Suppose we can measure both
1
x and
2
x
!
.]
1 1
2
2 2
0 1 0
,
0 1
x x
u
x x
_ _ _ _
+

, , , ,
!
!
1 1
2 2
y x
y
y x
_ _


, ,
Chapter 1 Systems & Control Theory
14
Exercise: write down the state and output equations
for the 3-dimensional (i.e. there are 3 state variables)
2-input-2-output system below:
Note that since the system is 3-dimensional, the state matrix
will be 3-by-3, since there are 2 input, the input matrix will be
3-by-2, and since there are also 2 output, the output matrix is
2-by-3.
1 1 1 1
1
2 2 2 2
2
3 2 2 3 3
3 3 0 0 1 0
0 1 0 0 1
2 0 1 2 0 1
x u x x
u
x u x x
u
x u x x x
_ _ _ _ _
_

+


,

+ +
, , , , ,
!
!
!
1
1 2 1
2
3 2
3
1 1 0
0 0
x
x x y
x
kx y k
x
_
+ _ _ _




, , ,

,
!
Chapter 1 Systems & Control Theory
15
Signal Flow Graphs
An alternative graphical representation makes use of the
following symbols (note the hat above the state variables
serve to remind users that this is just a representation):
1. multiplier
2
x
1 2
x a x
a
2. adder
2
x
1 2 3
x x x +
3
x
3. Delay (for discrete time system)
2
x
2
1

x
x
z

1
z

Thus time delay is now represented by multiplying by

1
z

.
Chapter 1 Systems & Control Theory
16
4. integrator
1
1 0
( ) x t s

1
s

2
x
1 1
2 1 0
( ) x s x t s

+
We say a system is relaxed if the initial conditions are all
zero. Thus for relaxed systems, integration is now
represented by multipling by
1
s

:
1
s

2
x
1
2
x s

In Chapter 4, we shall learn how these signal flow graphs

can be manipulated to help us solve problems without
doing any mathematics. For the time being, just treat them
as another graphical representation of state models.
!
Exercise: draw the signal flow graphs for the RC-
circuit system and the Driven Pendulum system
given on page 13.
Chapter 1 Systems & Control Theory
17
Local Linearization of Non-linear Models
In our previous example of a driven pendulum, we
converted a non-linear model into a linear model by
approximating sin by the first term in its Taylor series
expansion. This technique can be used for all non-linear
models as long as the non-linear functions have Taylor
series at the nominal point (for the pendulum case, the
nominal point was 0 ) that we choose.
Taylor series (revision)
" Let x x x
nom
+ , u u u
nom
+ ,
where x and u are small changes from the nominal
points
nom
x and
nom
u respectively. Then the Taylor
series to first order gives:
x
dx
df
x f x f
nom
x
nom
+ ) ( ) (
" Example: consider
2
) ( x x f , and let the nominal point
be 1
nom
x , then
x
x x f x f
nom
x

2 1
2 ) 1 ( ) 1 (
+
+ +
Chapter 1 Systems & Control Theory
18
This means that locally, the function is a straight line
of the form x y 2 1+
f(x)
x
1
Similarly, a function of u may also be expanded in the
same way
u
du
df
u f u u f
nom
u
nom nom
+ + ) ( ) (
" If we have a function of several state variables, say,
3 2 1
, , x x x and several control input, say,
2 1
,u u , then the
Taylor series generalizes to
1 2 3
1 2 3
1 2
1 2
( , )
( , )
nom nom nom
nom nom
nom nom
nom nom
f x x u u
f f f
f x u x x x
x x x
f f
u u
u u



+ +

+ + +


+ +

Chapter 1 Systems & Control Theory
19
" By defining the small changes , x u to be the new
state variables and control input respectively, we would
then obtain a linearized state model, operating near the
chosen nominal point.
" Example: consider the non-linear system
1 5 . 0 2 ) 1 (
2 3
+ + + u y y y y y ! ! !
Let us introduce state variables as before:
1 5 . 0 ) 1 ( 2
1 5 . 0 ) 1 ( 2
2 3
1 2 1 1 2
2 3
2 2
2 1 1
+ +
+ +

u x x x x x
u y y y y y x y x
x y x y x
!
! ! ! ! !
! !
The last state equation is non-linear. Let us linearize
this equation about the nominal point 1 , 0
2 1
u x x .
I.e., we consider small changes about the nominal
point,
1 1 2 2
0 , 0 , 1 x x x x u u + + +
1 5 . 0 ) 1 ( 2 ) , , (
2 3
1 2 1 1 2 1 2
+ + u x x x x u x x f x!
2 1 2
1 2
2
2 1 1 1 2
2 1 2
(0, 0,1)
0 (2 1.5 ) (1 ) 2
2 2
nom
nom nom
nom nom
nom
f f f
x f x x u
x x u
x x x x x u u
x x x u




+ + +

+ + +
+
!
!
Chapter 1 Systems & Control Theory
20
Together with the first state equation, which now
becomes:
2 1 2 1
x x x x ! !
Thus the new linearized state model is
u
x
x
x
x
dt
d
u x x x
x x

,
_

,
_

,
_

,
_

2
0
1 2
1 0
2 2
2
1
2
1
2 1 2
2 1
!
!
" The output equation depends on what we can (or wish)
to measure
" Example 1: suppose we can only measure y , then
the output equation is given by
( )

,
_

2
1
1
0 1
x
x
x y

" Example 2: suppose for some reason, we can

measure
2
) 1 ( y + and u y + ! , then the (non-linear)
output equation is given by

,
_

+
+

u y
y
Y
!
2
) 1 (
,
!Exercise: linearize this output equation.
Matrix Review Systems & Control Theory
21
# Eigenvectors
" Let ) (
jk
a A be a n-by-n matrix, an eigenvector x of A,
with eigenvalue has the property
x Ax
" Let
n
e e e , , ,
2 1
% be eigenvectors of A with corresponding
eigenvalues
n
, , ,
2 1
% . Construct matrix P with each
column of P being one of the eigenvector. I.e.
] | | | [
2 1 n
e e e P %
then
) 1 (
] | | | [
] | | | [ ] | | | [
2 2 1 1
2 1 2 1

P
e e e
Ae Ae Ae e e e A AP
n n
n n
%
% %
where
) , , , (
2 1
2
1
n
n
diag

%
&

,
_

" is an example of a diagonal matrix, which is a
Matrix Review Systems & Control Theory
22
particularly simple form of matrix with very convenient
properties:
" ) , , , (
2 1
k
n
k k k
diag %
" if we define ,
! 0


k
k
A
k
A
e then
) , , , (
)
!
, ,
!
,
!
(
!
) , , , (
!
2 1
0 0
2
0
1
0
2 1
0
n
e e e diag
k k k
diag
k
diag
k
e
k
k
n
k
k
k
k
k
k
n
k k
k
k



%
%
%

# Diagonalization
" From equation (1), we see that if P is invertible (i.e. its
determinant is non-zero), then we have


AP P P AP
1
(2)
" Equation (2) tells us that A can be transformed to a
diagonal matrix (only if P is invertible), which we know
has nice properties. We can use such a similarity
transform to help us evaluate functions of matrices
which can be diagonalized in this way
Matrix Review Systems & Control Theory
23
" In particular,
) , , , ( ) (
) (
2 1
1
1 1 1 1 1
k
n
k k k k
k
diag AP P
P A APP APP APP P P A P
%
"

1
2 1
1
) , , , (

P Pdiag P P A
k
n
k k k k
%
" Similarly,
) , , , (
!
! !
2 1
0
0
1
0
1 1
n
e e e diag e
k
k
P A P
P
k
A
P P e P
k
k
k
k
k
k
A

%

,
_

1 1
) , , , (
2 1

P e e e diag P P Pe e
n
A
%
# Solution to matrix differential equation Ax x !
" We first note that in the scalar case, the solution is an
exponential
At
e x
" To see that this is also true for matrices, we consider
At
p
p p
k
k
k
k
k
k
k
k k
At
Ae
p
t A
A t
k
AA
kt
k
A
k
t A
dt
d
e
dt
d

,
_

,
_

0 1
1
1
1
1
0
! )! 1 (
! !
(substitution p = k1)
Matrix Review Systems & Control Theory
24
" From the above result, we see that a general solution to
Ax x ! is
) 0 ( ) ( x e t x
At
or ) ( ) (
0
) (
0
t x e t x
t t A

# Matrix inversion using Cramers Rule

" The minor
jk
M of a square matrix A is the determinant
of A after the jth row and kth column have been
deleted
" The cofactor
jk
k j
jk
M A
+
) 1 (
" The adjoint of A is defined as ( )
T
jk
A A adj ) (
" It can be proved that (Cramers rule)
"

n
k
jk jk
A a A
1
det
"
A
A adj
A
det
) (
1

, (assuming A is invertible)
" In practice, Cramers rule is cumbersome and better
methods such as Gaussian Elimination (using row
operations) can be used for calculating determinants and
inverses.