9.

1 Stability of Zero-Input Response

LTI System

x  Ax  Bu, x (t0 )
y  Cx  Du
Laplace
Transform
Transfer Function

Y (s)
 C ( sI  A) 1 B  D
U (s)

1
 9.1 Stability of Zero-Input Response

LTI System

x  Ax  Bu, x (t0 ) u(t)

x(t)
y  Cx  Du System
x(t0)

0-input response 0-state response

x(t0) xh(t) u(t) xp(t)

System System

2
9.1 Stability of Zero-Input Response

x  Ax  Bu, x (t0 )
y  Cx  Du

0-input response 0-state response

t
t e
A ( t  t0 ) A(t  )
x (t )  e x (t0 )  Bu( )d
0

y (t )  Cx (t )  Du(t )

3
 9.1 Stability of Zero-Input Response

(Definition) Equilibrium state

Given a system with zero input, if a state xe satisfies the
following condition for t  t0, it is called equilibrium at t = t0.

x  Ax , x (t0 ) x e  0  Axe

• External input is regarded as 0

• A zero vector, i.e. xe = 0, is the equilibrium of all systems

4
 9.1 Stability of Zero-Input Response

Ex: Find equilibrium states.

0 1 0 
x    x    u , x0
 1 2  1 

5
 9.1 Stability of Zero-Input Response

Ex: Find equilibrium states.

0 1  0
x    x    u , x0
 0 2  1 

6
 9.1 Stability of Zero-Input Response
(Definition) Internal stability
The equilibrium is said to be internally stable if, for any x0 in the
region of ||x0 - xe||   < , the trajectory of x starting from x0 stays in
the region of ||x - xe||   <  for t  0.

7
 9.1 Stability of Zero-Input Response

Test of internal stability

The E.S. xe is said to be internally stable if the following
conditions are met:
1. Re(i)  0 for i = 1,2,…,n and
2. For multiple eigenvalues i of ri existing on the
complex axis
• (A -iI) = n - ri
or
• A = P-1AP must have all 1st-order Jordan blocks.

8
 9.1 Stability of Zero-Input Response

Ex: Given a zero-input system.

0 1 
x    x , x0
 0 2 

Check for internal stability

9
 9.1 Stability of Zero-Input Response

(Definition) Asymptotical stability

The equilibrium is said to be asymptotically stable if, for any x0
in the region of ||x0 - xe||   < , the trajectory of x starting from
x0 always converges to xe for t  0.


xe x

10
 9.1 Stability of Zero-Input Response

Test of asymptotical stability

If all the eigenvalues of A have negative real parts, that is,
Re(i) < 0 (i =1,2,...,n), j
then the system is
asymptotically stable

Remark: Be in the shaded 0
area, not on the imaginary
axis.

11
 9.1 Stability of Zero-Input Response

Ex: Asymptotic stability test

0 1  0 
x    x , x0   
 2 2   0.5

12
 9.1 Stability of Zero-Input Response

(Definition) Unstability
The equilibrium is said to be unstable if, for any x0 in the region
of ||x0 - xe||   < , the trajectory of x starting from x0 grows
infinitely for t  0.

13
 9.2 Stability of Zero-State Response

(Definition) BIBS stability

The state x(t) is said to be BIBS (bounded-input bounded-state)
stable if, for bounded input, that is, ||u(t)|| < k, there exists  such
that satisfies ||x(t)||  .

Test of BIBS stability

If all the poles of the TF X(s)/U(s) = (sI - A)-1B have negative real
parts, we say BIBS stability.

14
 9.2 Stability of Zero-State Response

(Definition) BIBO stability

The state x(t) is said to be BIBO (bounded-input bounded-output)
stable if, for bounded input, that is, ||u(t)|| < k, there exists  such
that satisfies ||y(t)||  .

Test of BIBO stability

If all the poles of the TF Y(s)/U(s) = C(sI - A)-1B+D have
negative real parts, we say it is BIBO stabile.

15
 9.2 Stability of Zero-State Response

Ex: Check for BIBS and BIBO stability of a

zero-state system.

1 0  0 
x    x    u , x0  0
1 2  1 
y  1 1 x

16
 9.3 First Lyapunov Method – Linearization

Stability criteria for nonlinear systems

First Lyapunov criterion (reduced method): the stability analysis
of an equilibrium point x0 is done studying the stability of the
corresponding linearized system in the vicinity of the equilibrium
point

Second Lyapunov criterion (direct method): the stability analysis

of an equilibrium point x0 is done using proper scalar functions,
called Lyapunov functions, defined in the state space

1
 9.3 First Lyapunov Method – Linearization

First Lyapunov criterion

Lyapunov “reduced” criterion: Let us consider the following “con
tinuous time” nonlinear system:

In the vicinity of the equilibrium point (x0, u0), let us consider

the corresponding linearized system:

2
 9.3 First Lyapunov Method – Linearization

For the system, the following statements hold:

3
 9.3 First Lyapunov Method – Linearization

Ex: (system linearization)

The system is defined by the set of equations

x1  x12  x2 2  1
x2  x1  x2

4
 9.3 First Lyapunov Method – Linearization

Ex: Unusually damped harmonic oscillator

We have a harmonic oscillator, where the mass m moves in very visco
us medium, where the damping force is proportional to the cube of the
velocity.

Set of differential equations describing the system

x  v
v   x   v 3
5
9.4 Second Lyapunov Method – Lyapunov Direct

Energy-Based Analysis
If the system dissipates energy along any trajectory that starts near
the equilibrium state so that the system energy converges to local
minimum, we expect this to correspond to asymptotic convergence
of the trajectory to the equilibrium, thereby indicating an asymptotic
ally stable equilibrium

We observe that conclusions regarding the stability of an equilibrium

state have been inferred from the time rate of change of the system
energy along trajectories

6
9.4 Second Lyapunov Method – Lyapunov Direct

Energy-Based Analysis
Consider the second-order linear translational
mechanical system,

7
 9.4 Second Lyapunov Method – Lyapunov Direct

We recall that these state variables are related to energy stored in

this system

The spring displacement characterizes the potential energy stored in

the spring, and the mass velocity characterizes the kinetic energy
stored in the mass

We therefore can express the total energy stored in the system by

the function

8
 9.4 Second Lyapunov Method – Lyapunov Direct

We observe that system energy is positive whenever [x1, x2]T ≠ [0, 0]T
and attains the minimum value of zero at equilibrium state
x ~
~
[ 1, x 2]T = [0,0]T

On evaluating the energy function along a system trajectory, we can

compute the time derivative

9
9.4 Second Lyapunov Method – Lyapunov Direct
Case-I: For zero damping (c = 0) we have dE/dt = 0, so the total system en
ergy is constant along any trajectory
For m = 1 kg, k = 10 N/m, and c = 0 N-s/m along with the initial state
X(0) = X0 = [1, 2]T:

(a) State-variable responses (b) phase portrait

10
 9.4 Second Lyapunov Method – Lyapunov Direct
In this case, we see oscillatory state-variable time responses, an
elliptical phase portrait, and constant total energy
The system eigenvalues are
λ1,2 = ± j3.16, purely imaginary
(zero real part)
Since they are distinct, the geom
etric multiplicity equals the
algebraic multiplicity for each

(c) energy response for a marginally

-stable equilibrium
11
 9.4 Second Lyapunov Method – Lyapunov Direct

Case-II: For positive damping (c > 0), we have dE/dt < 0 along any trajectory
for which the mass velocity is not identically zero
With c = 1 N-s/m and all other parameters unchanged:

(a) State-variable responses (b) phase portrait

12
9.4 Second Lyapunov Method – Lyapunov Direct

In this case, each state-variable time

response decays to zero as time tends
to infinity, as does the total energy
response

The phase portrait depicts a state traj

ectory that spirals in toward the equil
ibrium state at the origin

(c) energy response for an asym

ptotically-stable equilibrium

System eigenvalues: λ1,2 = -0.50 ± j3.12, each with negative real part
13
9.4 Second Lyapunov Method – Lyapunov Direct

Case-III: For negative damping (c < 0), we have dE/dt > 0 along any trajecto
ry for which the mass velocity is not identically zero
Finally, with damping coefficient changed to c = -1 N-s/m:

(a) State-variable responses

(b) phase portrait

14
 9.4 Second Lyapunov Method – Lyapunov Direct

Each state variable time response

grows in amplitude and total energy
increases with time

The phase portrait depicts a state

trajectory that spirals away from
the equilibrium state at the origin

(c) energy response for

an unstable equilibrium

System eigenvalues: λ1,2 = +0.50 ± j3.12, each with positive real part
15
9.4 Second Lyapunov Method – Lyapunov Direct

(Theorem) Lyapunov stability

A system
x  f ( x , t ), x0

is said to be Lyapunov stable if a scalar Lyapunov function V(x)

satisfying the following conditions exists:

1. V(x)  0 for x, and V(x) = 0 for x = 0

2. V(x) < 0 for x  0

16
9.4 Second Lyapunov Method – Lyapunov Direct

Ex: Consider the system

The total energy of a harmonic oscillator

x  v
v   x   v 3

Consider:
m = 1 and k = 1 1 2 1 2 1 2 2
E  kx  mv E  (x  v )
2 2 2
17
9.4 Second Lyapunov Method – Lyapunov Direct

(Definition) Definiteness
For xTAx defined by x  0 and a real symmetric matrix A,
1. If xTAx >0, A is a positive definite(PD) matrix or shortly PD
2. If xTAx 0, A is positive semi-definite(PSD)
3. If xTAx <0, A is negative definite(ND)
4. If xTAx 0, A is negative semi-definite (NSD)
5. If xTAx is negative or positive, A is indefinite.

18
9.4 Second Lyapunov Method – Lyapunov Direct

Test for matrix definiteness

Determinants, i = |Ai| (i = 1,2,..,n)

a11 a12
1  a11 , 2  ,..., n  | A |
a21 a22

1. If 1>0, 2>0,..., n>0 A is PD

2. If 10, 20,..., n0 A is PSD
3. If 1<0, 2>0, 3<0, 4>0... A is ND
4. If 10, 20, 30, 40... A is NSD
5. Otherwise A is ID
19
9.4 Second Lyapunov Method – Lyapunov Direct

Test for matrix definiteness

Eigenvalues of A, i(1 i  n)
1. If i > 0  A is a PD matrix or shortly PD
2. If i  0  A is PSD
3. If i < 0  A is ND
4. If i  0  A is NSD
5. Otherwise A is ID

20
9.4 Second Lyapunov Method – Lyapunov Direct

(Theorem) Asymptotical stability

A LTI system x  Ax , x0 having a point of equilibrium at
xe= 0 is asymptotically stable if, for any PD matrix Q, the
Lyapunov equation,
ATP+PA = -Q
has a unique solution P and P is a PD matrix.

(Proof)
Choose V(x) = xTPx as the Lyapunov function, where P is a PD
symmetric matrix. Differentiating V(x) gives

21
9.4 Second Lyapunov Method – Lyapunov Direct

  T T
 T T
V ( x )  x Px  x Px  ( Ax ) Px  x P ( Ax )
 x T ( AT P  PA) x 
In order for the system to be asymptotically stable, it
must be V(x) < 0 for a PD matrix Q.
V ( x )   x T Qx 
Comparing equations  and  yields
ATP+PA = -Q
Therefore, if a PD matrix P that satisfies the above
equation exists, the system becomes asymptotically stable.

22
9.4 Second Lyapunov Method – Lyapunov Direct

Ex: Show that the system is stable using the Lyapunov direct
method.

 0 1
x    x AT P  PA  Q
-1 -1

23
9.4 Second Lyapunov Method – Lyapunov Direct

Ex: The following is an unstable open-loop system.

When it has a feedback with u = -[17 9]x, check the stability of
the feedback system using the Lyapunov direct method.

 2 1 0 
x    x    u, x0
 0 3 1 

24
 10. Controllability and Observability
10.3 Controllability
Controllability concept
Situations:
 We want to design a feedback system with the
desired performance.
 We want to control unstable systems.

Questions:
 Is it possible to move the initial state x(t0) to the
desired state x(t1) after a finite time by applying
the appropriate input u(t)?

1
 10.3 Controllability

Observability concept
Situations:
 Some variables are difficult to measure for
feedback.
 Some sensors are too expensive. So they can not
be used.
Questions:
 Is it possible to know the initial state x(t0) or the
state x(t) from both the input u(t) and the output
y(t) observed for a finite time?

2
 10.3 Controllability

(Definition) Controllability
A system is said to be controllable at t = t0 if there is a control
input u(t), t0 < t  t1 that can move the initial state x(t0) to the
desired state x(t1) at t = t1.

3
 10.3 Controllability

(Definition) Controllability of a LTI system

A LTI system is said to be controllable at t = t0 if u(t), t0< t  t1
exists to move x(t0) to the desired state x(t1) at t = t1. Especially if
one of the following conditions is true, then it is controllable:
1. The rank of the n  nm controllability matrix is n.
U = [B AB  An-1B]
2. All rows of the transfer matrix are linearly independent.
(sI - A)-1B
3. The controllable Gramian matrix WC is nonsingular.
t
e
A T AT 
Wc  BB e d , t  0
0
4
10.3 Controllability

Ex: Given a system

0 1 1 0 
x    x  u, x (t0 )  x0
 1 2  0 1 

Controllability matrix is given by

1 0 0 1  Since U = 2 = n,
U  [ B AB ]    controllable.
0 1 1 2 

5
 10.3 Controllability

Example: Check the controllability using the Matlab functions.

 1 1 0  1 
x   0 1 0  x   2  u, x (t0 )  x0
 0 0 0  1 
Program:
>>A= [-1 1 0;0 -1 0;0 0 0];
>> B= [1;2;1];
>> U= ctrb(A,B); % Controllability matrix
>> r= rank(U) % Rank
r= 3

6
 10.4 Observability

Why do we need to check it?

Designing and implementing state feedback control
require the feedback of all state variables.
When some state variables are unmeasurable, we
need to check whether they can be estimated from the
model.
If the system is unobservable, state observer design is
impossible
Firstly introduced by R.E. Kalman

7
 10.4 Observability

(Definition) Observability
A LTI system is said to be observable if there is enough t = t1>0
that u and y determine x0 uniquely. Especially it is observable if
one of the following conditions is true:
1. The rank of the np  m observability matrix V is n.

 C 
 CA 
V  
  
 n-1 
CA 

8
 10.4 Observability

2. All columns of the transfer matrix are linearly independent.

C(sI - A)-1

3. The observable Gramian matrix WO is nonsingular.

t

AT  T A
Wo  e C Ce d , t  0
0

9
 10.4 Observability

Example: 0 1 0 
Consider a SISO LTI system x    x    u, x (t0 )  x0
 1 2  1 
y  1 0 x

Observability matrix yields

 C  1 0 
V   
CA 0 1 

 Since V = 2 = n, the system is observable.

10
 10.5 Similarity Transformation

A LTI system can be represented as a combination of subsystems

with appropriate similarity transformation

 Controllable and
observable system
 Controllable but
unobservable system
 Observable but
uncontrollable system
 Uncontrollable and
unobservable system

11
 10.5 Similarity Transformation

Controllable subsystem

x  Ax  Bu, x (t0 )  x0
y  Cx  Du

If U = n1< n, via the following similarity transformation

x  Qx

The original system can be represented by a combination of

controllable subsystems and uncontrollable subsystems.

12
10.5 Similarity Transformation

(n - n1) vectors randomly chosen

such that Q is nonsingular

where Q = [q1 q2 … qn1 qn1+1 … qn]

n1 independent linear
vectors selected from U

13
 10.5 Similarity Transformation

Original system
x  Ax  Bu, x (t0 )  x0
y  Cx  Du

Similarity transformation x  Qx

x  Ax  Bu
y  Cx  Du
where
A  Q 1 AQ, B  Q 1 B, C = CQ, D  D

14
 10.5 Similarity Transformation
Transformed system
 x c   Ac A12   xc   Bc 
    u

 c   0
x Ac   xc   0 
 xc 
y  Cc Cc     Du
 xc 
Controllable subsystem

x c  Ac xc  Bc u
y  Cc xc  Du

15
 10.5 Similarity Transformation

Observable subsystem

x  Ax  Bu, x (t0 )  x0
y  Cx  Du

If V = n2 < n, via the similarity transformation

x  Q 1 x

the original system can be represented by a combination of

observable subsystems and unobservable subsystems.

16
 10.5 Similarity Transformation

 q1 
 q 
 2  n2 independent vectors
   selected from V
 
where Q   qn2 
q 
 n2 1  (n- n2) vectors
   randomly chosen such
  that Q is nonsingular
 qn 

17
 10.5 Similarity Transformation

Original system

x  Ax  Bu, x (t0 )  x0
y  Cx  Du
1
Similarity transformation x  Q x

x  Ax  Bu
y  Cx  Du
where A  QAQ 1 , B  QB, C  CQ 1 , D  D

18
 10.5 Similarity Transformation
Transformed system
 x o   Ao A12   xo   Bo 
    u
 x o   0 Ao   xo   0 
 xo 
y  Co Co     Du
 xo 
Observable subsystem
x o  Ao xo  Bo u
y  Co xo  Du

19
 10.6 Controllable Canonical Form

Plant
x  Ax  Bu, x (t0 )  x0
y  Cx  Du

where A  nn , B  n1 , C  1n ,D  

Assumption
The plant is a SISO LTI system(u and y are scalars)
It is controllable & observable (U = n, V = n)

20
 10.6 Controllable Canonical Form

Similarity transformation
Before
x  Ax  Bu, x (t0 )  x0
y  Cx  Du
x  Qx
After
Q is
x  Ax  Bu, x (t0 )  x0 nonsingular
y  Cx  Du
1 1
where A  Q AQ, B  Q B, C  CQ, D =D
21
 10.6 Controllable Canonical Form
Controllability
n 1
Before U  [ B AB  A B ], U  n
After U  [ B AB  Ai 1 B  An 1 B ]
The ith term in the above equation
i 1 1 1 1 1 1 i 1
A B  (Q AQ )(Q AQ ) (Q AQ )Q B  Q A B

Thus (i-1) times

U  Q 1[ B AB  Ai 1 B  An 1 B ]  Q 1U , U  n

If Q is nonsingular, controllability is kept after transformation.

22
 10.6 Controllable Canonical Form

Observability
 C   C 
 CA   
 CA 
Before V   , V  n After V 
    
 n 1   
CA  CAn1 

As the previous case

CAi 1  CQ (Q 1 AQ )(Q 1 AQ ) (Q 1 AQ )  CAi 1Q

(i-1) times

23
 10.6 Controllable Canonical Form

After transformation

 C   C 
   
     
V   CAi 1    CAi 1  Q  VQ
   
     
 n 1   n 1 
CA  CA 

Thus,
V  n
 Observability is kept after transformation.

24
 10.6 Controllable Canonical Form

Conversion to controllable canonical form

System
x  Ax  Bu, x (t0 )  x0
y  Cx  Du

Controllable canonical form

x  Ax  Bu, x (t0 )  x0
y  Cx  Du
where A  Q 1 AQ, B  Q 1 B, C  CQ, D =D

25
 10.6 Controllable Canonical Form
Controllable canonical form(SISO)
 0 1 0  0  0
 0 0 1  0  0
   
x         x     u, x (t0 )  x0
   
 0 0 0  1  0
 a a  1 
 n n 1  an  2   a1 

y  (bn  b0 an ) (bn 1  b0 an 1 )  (b1  b0 a1 )  x  b0 u

Transfer function
(b1  b0 a1 )s n 1  (b2  b0 a2 )s n 1    (bn  b0 an )
G( s)  n n 1
 b0
s  a1s    an 1s  an

26
 10.6 Controllable Canonical Form

Conversion procedure

Step 1: Check the controllability of the system (U = n).

U  [ B AB  An1 B ]

Step 2: Find the characteristic polynomial from A.

( ) |  I  A |  n  a1 n 1    an 1  an

27
 10.6 Controllable Canonical Form

Step 3: Obtain Q = [q1 q2 … qn]

qn  B
qn 1  AB  a1 B
qn  2  A2 B  a1 AB  a2 B

q1  An 1 B  a1 An  2 B   an 1 B

Step 4: Obtain matrices

A  Q 1 AQ, B  Q 1 B, C  CQ, D =D

28
 11. Regulator
11.1 State Feedback Control

r u d y
F + + + Plant
-
K x

Purpose of using state feedback controllers

Setpoint tracking Tracking controller
- Set point changes from time to time
- Controls the output to follow a set point
Disturbance rejection Regulator
- Set point is fixed for a long time
- Controls to keep the output at a set point even when load or
disturbance changes
1
 11.2 P-Type Regulator

Regulator allows the system to quickly eliminate the effects

of load or disturbance and to ensure that the output quickly
matches the set point.
Assumptions:
The system is controllable (U = n).
All state variables can be fed back.
Control input is not limited.
The set point (or reference input) is considered 0 for
convenience.

2
 11.2 P-Type Regulator
(Theorem) Consider a SISO LTI system

x  Ax  Bu , x0

If it is controllable, there exists K that allows the poles of the

closed-loop system to be in the desired position through a
regulator, u = -Kx.

r=0 + u = r - Kx y
Plant
-

K
x
3
 11.2 P-Type Regulator
Plant (open-loop)
Regulator
x  Ax  Bu
u  Kx
y  Cx  Du

Feedback (closed-loop)
x  Ax  Bu  (A  BK )x +0r
y  Cx  Du  (C  DK )x +0r

4
 11.2 P-Type Regulator

Method 1: Pole assignment method

Given poles 1 , 2 , , n of a closed-loop system, find the
feedback gain matrix, K of

u = -Kx.

5
 11.2 P-Type Regulator
1. Brute force method
Step 1: Obtain the characteristic polynomial from the given
poles: 1 , 2 , , n

n n-1
( )  (  1 )(  2 ) (  n )    a1    an 

Step 2: Find the characteristic polynomial of (A - BK)

( )   I  A  BK   n  a1 n -1    an 

Step 3: Compare  with  and determine K.

ai  ai (i  1, 2, , n)
6
 11.2 P-Type Regulator
(Theorem) When the system is a controllable canonical form,
x  Ax  Bu

if the given closed-loop poles are 1 , 2 , , n , then the gain

matrix K of regulator u = -Kx is given by:

K  (an  an ) (an 1  an 1 )  (a1  a1 ) 

where
( )  A   I   n  a1 n 1  ...  an
( )  (  1 )(  2 )...(  n )
  n  a1 n 1  ...  an
7
 11.2 P-Type Regulator
2. Similarity transformation method
Step 1: Obtain the characteristic polynomial from the given poles.
( )  (  1 )(  2 ) (  n )
  n  a1 n -1    an

Step 2: Obtain characteristic polynomial of the open-loop system.

( )   I  A   n  a1 n -1    an

Step 3: obtain u  Kx

K = (an  an ) (an -1  an -1 )  (a1  a1 ) 

8
 11.2 P-Type Regulator

Step 4: Obtain matrix Q = [q1 q2 … qn].

qn  B
qn 1  AB  a1 B
qn  2  A2 B  a1 AB  a2 B

q1  An 1 B  a1 An  2 B   an 1 B

Step 5: Obtain
K  KQ 1

9
 11.2 P-Type Regulator
3. Ackermann formula:
Step 1: Obtain () from the given poles.
( )  (  1 )(  2 ) (  n )   n  a1 n -1    an

Step 2: In (), obtain (A) by substituting A for .

 ( A)  An  a1 An -1    an 1 A  an I

Step 3: Obtain the controllability matrix U.

U  [ B AB  An1 B ]

Step 4: Find K with the following formula

K   0  0 1 U 1 ( A)
10
 11.2 P-Type Regulator

Notes on pole selection

• The closed-loop poles should be not
too far from the open-loop poles.
• The real part of the closed-loop
poles should be not too large (large
K can amplify the noise).
• Typically, select from the yellow
area in the right figure.

11
 11.3 PI-Type Regulator

Type-n plant
x  Ax  Bu G ( s )  C ( sI  A) 1 B  D
y  Cx  Du K (1  T1' s )(1  T2' s ) (1  Tn' s )

s n (1  T1s )(1  T2 s ) (1  Tn s )

Type-0 plant(n= 0)
x  Ax  Bu K (1  T1' s )(1  T2' s ) (1  Tn' s )
G (s) 
y  Cx  Du (1  T1s )(1  T2 s ) (1  Tn s )

12
 11.3 PI-Type Regulator

(Problem) Consider a SISO plant with constant disturbance.

x  Ax  Bu +Wd , x (0)
y  Cx  Du

P-type regulator u = -Kx introduces a steady-state error.

So, let's design a PI-type regulator where d means disturbance.

13
 11.3 PI-Type Regulator

Introduce a new state variable to design a PI-type regulator

z   ydt

Differentiating both sides gives

z  y

Original system x  Ax  Bu+Wd 

New state variable  y  Cx  Du
z= 
 x
Augmented vector x   
z
14
 11.3 PI-Type Regulator

x  Ax  Bu +Wd
z  y  Cx  Du

Combining  and  yields

 x   A 0   x   B  W 
 z   C 0   z    D  u +  0  d
        
 x
y  C 0    Du
z
15
 11.3 PI-Type Regulator

That is,
x  Ax
   B u +W d
y  Cx   Du

 x    A 0    B   W 
where x    , A    , B    ,W    ,
z C 0  D 0
C  C 0 , D  D

If the original system (A,B) is controllable, the

augmented system ( A , B ) is also controllable.
16
 11.3 PI-Type Regulator

So, the PI-type regulator becomes

 x
u   Kx    K1 k2      K1 x  k2  ydt

z

K
where can be obtained by one of the previous methods.

17
 12. Tracking Controller
12.1 P-Type Tracking Controller

(Problem) Consider a LTI system

x  Ax  Bu , x (0)
y  Cx  Du

Given closed-loop poles 1 , 2 , , n , find K and F of

the tracking controller

u = Fr - Kx

At this time, the output should follow the set point

without the steady-state error.

1
 12.1 P-Type Tracking Controller

d
r + u y
F Plant
-

K x

• Disturbance d is assumed to be 0 or constant.

• Set point or reference input r is considered as a step signal.
2
 12.1 P-Type Tracking Controller

Plant
x  Ax  Bu , x (0)
y  Cx  Du

Tracking controller

u = Fr - Kx

Closed-loop system
x  ( A  BK ) x  BFr
y  (C  DK )x  DFr
3
 12.1 P-Type Tracking Controller

The design procedure of the tracking controller is summarized

into two steps:
Step 1:
Determine the feedback gain matrix K from the given poles:

1 , 2 , , n

Step 2:
Determine the feedforward gain F so that the output follows the
set point without steady-state error.

4
 12.1 P-Type Tracking Controller

Calculation of K u = Fr-Kx
Use one of the following methods with the closed-loop poles
1 , 2 , , n
- Brute force method
- Similarity transformation method
- Ackermann formula

Calculation of F
x  ( A  BK ) x  BFr
y  (C  DK )x  DFr
5
 12.1 P-Type Tracking Controller

The transfer function is given by

Y(s)/R(s) = G(s)
= (C-DK)(sI-A+BK)-1BF+DF

For the unit step input r = 1, R(s) = 1/s

lim y (t )  lim sY ( s )  lim sG ( s ) R ( s )
t  s0 s0

 lim[(C  DK )( sI  A  BK ) 1 BF  DF ]  1
s0

(C  DK )(  A  BK ) 1 BF  DF  1

Find F from the equation.

6
 12.2 PI-Type Tracking Controller

(Problem) Consider a LTI system with disturbance d:

x  Ax  Bu+Wd , x (0)
y  Cx  Du

Given the closed-loop poles of (n+1) 1 , 2 , , n , n 1 .

Find K= [K1 k2] of the PI-type tracking controller

u = -K1x + k2(r-y)dt

Output should follow the set point without the steady-state error.

7
 12.2 PI-Type Tracking Controller

r + + u y
 k2 Plant
- -

K1 x

• Disturbance d is considered constant

• Set point(or reference) input r is considered a step signal

8
 12.2 PI-Type Tracking Controller
In order to design the PI-type tracking controller without steady-
state error for a constant disturbance or the type-0 plant, the
new state variable is defined as follows:


z  ( y  r )dt

Differentiating both sides once gives,

z = y  r
From the dynamic equation,
x  Ax  Bu  Wd
z  y  r  Cx  Du  r
9
 12.2 PI-Type Tracking Controller

x 
By defining the augmented vector as x    and rewriting the
equations gives z 

   B u  W d  0  r , x (0)
x  Ax
1 
 
y  Cx   Du

where
 A 0    B   W  

A  , B    , W    , C  C 0 , D  D
C 0  D 0

10
 12.2 PI-Type Tracking Controller

If the open-loop system is controllable, the augmented system is

also controllable and a PI-type controller can be written as

 x
    K
u   Kx 1
z

k2      K1 x  k2 (r  y )dt

d
r + + u y
 k2 Plant
- -
K1 x

11
 12.2 PI-Type Tracking Controller
  into
Substituting u  Kx
   B u  W d  0  r
x  Ax
1 
 
the feedback system becomes
  ) x  W d  0  r
x  ( A  BK
1 
 

(Theorem) If K is chosen so that all eigenvalues of ( A  BK   ) have

negative real parts, the feedback system becomes asymptotically stable
without steady-state error for the step input and constant disturbances.

12
 12.2 PI-Type Tracking Controller
Design procedure
Step 1: Write the augmented system from the system.

(Original) x  Ax  Bu +Wd , x (0)

y  Cx  Du

(Augmented)
   B u  W d  0  r , x (0)
x  Ax
1 
 
y  Cx   Du

where
 A 0    B   W  

A  , B    , W    , C  C 0 , D  D
C 0  D 0
13
12.2 PI-Type Tracking Controller

Step 2: Obtain K= [K1 k2] from the given poles of (n+1)

 , B :
1 , 2 , , n , n 1 and A

- Brute force method

- Similarity transformation method
- Ackermann formula

Step 3: Implement the PI-type tracking controller

u = -K1x+k2(r-y)dt

14
 13. State Estimator
13.2 Full State Observer Design

When state feedback control is used

• Designed on the assumption that all state variables can be

fed back, but some variables (signals) may not be measured
due to the limitation of the measuring device.
• Even if the measurement is possible, we have to consider
the cost of the measuring equipment
• In these cases, a state observer can be used to estimate state
variables.

1
 13.2 Full State Observer Design

SISO LTI system

x  Ax  Bu , x (t0 )
y  Cx

• xn, u, y and A, B, C are nn, n1, 1n real

matrices, respectively.
• D is regarded as 0 for convenience.
• In case of D  0, if y= y-Du, the system is of the form y= Cx .
• The pair (A, C) is observable (V = n).
2
 13.3 Open-loop State Observer
Open-loop state observer
xˆ  Axˆ  Bu , xˆ (t0 )
x

• Estimating the state x

by connecting the
model in parallel with
the system x̂
• Observer outputs x +
with u as input

• x 𝟎 is the initial value of the state observer.

3
 13.3 Open-loop State Observer

Characteristics of open-loop state observer

If the model is accurate and x 0 of the state observer is
synchronized with x 0 of the system, then x is equal
to x(t) at t  t0.
But this method [open-loop state observer] has two drawbacks:
• Firstly, it is not easy to synchronize x
1) 0 with x 0
exactly when the state observer is turned on.
• Secondly, if the system parameters change during
2)
operation, x no longer coincides with x(t).
4
 13.3 Open-loop State Observer
State control with an open-loop observer

d
r + + u y
 k2 Plant
- -
Observer

K1
x

u   K1 xˆ  k2  (r  y )dt
5
 13.4 Asymptotic State Observer
Asymptotic state observer Luenberger observer
xˆ  Axˆ  Bu  L( y  Cxˆ ), xˆ (t0 )

• Improve the disadvantage

of the open-loop state
제어대상 시스템
observer
• Output x using u and y as
input
• L is the gain matrix
multiplied by the error
+
between the system and
the observer output.
6
 13.4 Asymptotic State Observer

• Defining the error between the system state and the observer state as

e  x  xˆ

• Differentiating both sides and substituting the previous expressions

yields

e  x  xˆ  Ax  Bu  Axˆ  Bu  L( y  Cxˆ )

 A( x  xˆ )  L(Cx  Cxˆ )
 A( x  xˆ )  LC ( x  xˆ )
 ( A  LC )e
7
 13.4 Asymptotic State Observer
(Theorem) Consider an error system
e(t )  ( A  LC )e (t ), e (t0 )  x (t0 )  xˆ (t0 )

Its characteristic equation is,

|I  A + LC | = 0.
If L is selected so that all eigenvalues of (A - LC) have negative
real parts, then for any e(t0), e(t)0 as t, that is, x t x(t).

• The solution of the error system is

( A  LC )t
e (t )  e e (t0 )
8
 13.4 Asymptotic State Observer

(Theorem) Consider an asymptotic state observer

xˆ  Axˆ  Bu  L( y  Cxˆ ), xˆ (t0 )

If the matrix pairs (A, C) of the system are

observable, then L can be chosen appropriately for
the observer to have arbitrary poles.

(Proof) Refer

9
 13.4 Asymptotic State Observer

How to design an asymptotic observer

Given n poles 1 , 2 ,  , n , find the gain matrix L

of the following asymptotic observer.

xˆ  Axˆ  Bu  L( y  Cxˆ ), xˆ (t0 )

where (A, C) is observable and the initial vector of the

state observer is generally set to x 𝟎 = 0.

10
 13.4 Asymptotic State Observer

Method 1: Brute force

Step 1: Find the characteristic polynomial of the state observer.

( )   I  A  LC   n  aˆ1 n-1    aˆn 

Step 2: Find the characteristic polynomial from the given

poles of n, 1 , 2 ,  , n .

( )  (  1 )(  2 ) (  n )   n  a1 n-1    an 

Step 3: Compare the coefficients of two polynomials to find L.

aˆi  ai (i  1, 2,  , n)
11
 13.4 Asymptotic State Observer

Method 2: Similarity transformation

Step 1: Set A  AT , B  C T , K  LT

Step 2: Find characteristic polynomial of the open-loop system.

( )   I  A   n  a1 n -1    an 

Step 3: Find the characteristic polynomial from the given poles.

( )  (  1 )(  2 ) (  n )
n n -1 
   a1    an
12
 13.4 Asymptotic State Observer
Step 4: Find K from  and .
K   an  an an -1  an -1  a1  a1 

Step 5: Find Q= [q1 q2 ... qn].

qn  B
  a B
qn -1  Aq n 1

  a B
q1  Aq 2 n -1

Step 6: Find K and L.

K  KQ 1  L  K T
13
 13.4 Asymptotic State Observer
Method 3: Ackermann Formula
Step 1: Find () from the given poles.

( )  (  1 )(  2 ) (  n )   n  a1 n-1    an

Step 2: Find
 ( A)  An  a1 An -1    an I

Step 3: Find the observability matrix

 C 
 CA 
V  
  
 n1 
CA 
14
 13.4 Asymptotic State Observer

0 
Step 4: 0 
Obtain the gain matrix L. L   ( A)V 1  
 
 
1 

Ex.: Consider the same system.

 2 1  1 0.5
x    x    u, x (t0 )   
 0 3 1 0 
y  1 0 x

Design an asymptotic state observer with two poles of -53j.

15
 14. Optimal Control
14.3 P-Type Optimal Regulator

SISO x  Ax  Bu
Linear quadratic regulator (LQR)
system y  Cx  Du

Quadratic 1 
function J
2 
0
xT Qx  uT Rudt

Find a control law u = -Kx for a dynamical system in terms of

minimizing the objective function where Q and R are user-defined
weighting matrix and coefficient, respectively.

The performance index J can be obtained in terms of the initial

condition x(0) and P
J  x T (0) Px(0)
1
 14.3 P-Type Optimal Regulator

Feedback control to minimize J,

u  Kx
where K is given by,
1 T
KR B P

P is the solution that satisfies the following Riccatti equation:

T 1 T
A P  PA  PBR B P  Q  0

If A is a real matrix and Q is a real symmetric matrix, then P is

a real symmetric matrix and PD (positive definite).
2
 14.3 P-Type Optimal Regulator

Optimal control
Step 1: Find the solution P of Riccatti equation from A, B, Q, and R.

AT P  PA  PBR 1 BT P  Q  0

Step 2: Find the feedback gain matrix K.

K  R 1 BT P

% MATLAB command
[K, P, E]= lqr(A,B,Q,R) % E are the eigenvalues of (A-BK)
3
14.3 P-Type Optimal Regulator

Ex: Consider a LTI system

 1 1  0  1
x    x    u , x (0)   
 0 1 1  1
y  1 0 x

Find the optimal regulator to minimize J.

1 
J
2 0
4 x12  x22  u 2 dt

4
 14.4 PI-Type Optimal Regulator

(Problem) Consider a SISO plant with constant disturbance.

x  Ax  Bu +Wd , x (0)
y  Cx  Du

• P-type regulator introduces a steady-state error for constant

disturbance.
• So, let's design a PI-type regulator where d means disturbance.

The new state variable is defined as,

z   ydt
5
 14.4 PI-Type Optimal Regulator

Consider the augumented system again

x  Ax
   B u +W d
y  Cx   Du


 x    A 0    B   W 
where x    , A    , B    ,W    ,
z C 0  D 0
C  C 0 , D  D

6
 14.4 PI-Type Optimal Regulator

Find a control law for the augumented system in terms of

minimizing the following objective function where Q and R are
user-defined weighting matrices.

1 
J
2  0
   uT Rudt
x T Qx 

Feedback control to minimize J


u  Kx
where K  R 1  T 
B P and P is the solution that
satisfies the following Riccatti equation:

A T P  PA
   PB
  R 1 B T P  Q  0
7
 14.4 PI-Type Optimal Regulator

Ex.: x   x  2u  d , x(0)= 0.5

Consider a scalar system. yx

From the system,  x   a 0   x  b   w 

a = -1, b = 2, w = 1, c =1  z    c       u+   d
0   z  0   0 
  
The augmented system is
 1 0   x   2   w
       u+   d
1 0  z  0  0 
A B

Find the optimal 1 

regulator to minimize J
J
2 0
x 2  4 z 2  u 2 dt
8
 14.5 PI-Type Tracking Controller

(Problem) Consider a SISO LTI system with constant

disturbance d:
x  Ax  Bu+Wd , x (0)
y  Cx  Du

In order to design the PI-type tracking controller without

steady-state error for a constant disturbance or the type-0 plant,
the new state variable is defined as follows:


z  ( y  r )dt

Differentiating z = y  r
both sides once gives 9
 14.5 PI-Type Tracking Controller

Consider the augumented system again

   B u +W d  0  r
x  Ax
1 
 
y  Cx   Du

 x    A 0    B   W 
where x    , A    , B    ,W    ,
z C 0  D 0
C  C 0 , D  D

10
 14.5 PI-Type Tracking Controller
Find a control law for the augumented system in terms of
minimizing the following objective function where Q and R are
user-defined weighting matrices.

1 
J
2 0
   uT Rudt
xT Qx 

Feedback control to minimize J


u  Kx
where K  R 1 B T P and P is the solution that
satisfies the following Riccatti equation:
A P  PA  PBR B P  Q  0
 T       1  T 
11
 14.5 PI-Type Tracking Controller

The PI-type tracking controller can be written as,

 x
    K
u   Kx 1
z

k2      K1 x  k2 (r  y )dt

d
r + + u y
 k2 Plant
- -
K1 x

• Disturbance d is considered constant

• Set point(or reference) input r is considered a step signal 12
 14.5 PI-Type Tracking Controller
Design procedure
Step 1: Write the augmented system from the system.
x  Ax  Bu +Wd , x (0)
(Original)
y  Cx  Du

   B u  W d  0  r , x (0)1
x  Ax
1 
(Augmented)  
y  Cx   Du

where
 A 0    B   W  

A  , B    ,W    , C  C 0 , D  D
C 0  0 0
13
 14.5 PI-Type Tracking Controller

Step 2: Obtain K= [K1 k2] from the given A , B

and Q and R .

Step 3: Implement the PI-type tracking controller

u = -K1x - k2z
= -K1x + k2(r - y)dt

14
 14.5 PI-Type Tracking Controller

Ex: Consider the previous scalar system x   x  2u  d , x(0)= 0

with disturbance d = 1 and x(0)= 0. yx

 x   a 0   x  b   w   0 
From the system,  z    c       u+   d -   r
0   z   0   0  1 
a = -1, b = 2, w = 1, c =1.   
The augmented system is  1 0   x   2   w 0 
       u+   d -   r
1 0   z   0   0  1 
A B

Find the optimal tracking controller to minimize J

1 
J
2 
0
x 2  4 z 2  u 2 dt
15