Scribd
Upload
224 views

Digital Control Kannan M. Moudgalya, Autumn 2007

This document discusses pole placement control using a state space approach. It begins by introducing the state space model and discussing how to use state feedback control to achieve a desired closed loop characteristic polynomial. It then covers controller canonical form and how this allows easily determining the feedback gain K to place the closed loop poles at desired locations. Ackermann's formula is derived, which gives K as the last row of the characteristic polynomial evaluated at the system matrix A. The document also discusses how to apply this approach when the system is not initially in controller canonical form by transforming to a new coordinate system.

Uploaded by

Ravinder Ranga
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
224 views

Digital Control Kannan M. Moudgalya, Autumn 2007

This document discusses pole placement control using a state space approach. It begins by introducing the state space model and discussing how to use state feedback control to achieve a desired closed loop characteristic polynomial. It then covers controller canonical form and how this allows easily determining the feedback gain K to place the closed loop poles at desired locations. Ackermann's formula is derived, which gives K as the last row of the characteristic polynomial evaluated at the system matrix A. The document also discusses how to apply this approach when the system is not initially in controller canonical form by transforming to a new coordinate system.

Uploaded by

Ravinder Ranga
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

1.

Pole Placement Control - State Space Approach

Start with our state space model:


x(k + 1) = Ax(k) + bu(k) + x0(k + 1).
Let u be a scalar, i.e. only one control effort.
Suppose all the states are measured.
Can we use a state feedback controller to get a desired closed
loop characteristic polynomial?
Take Z-transform:
zX(z) = AX(z) + bU (z) + x0z
The first term is:
(zI A)X(z) = bU (z)
adj(zI A)
1
X(z) = (zI A) bU (z) =
bU (z)
|zI A|
Digital Control

Kannan M. Moudgalya, Autumn 2007

Eigenvalues = Poles
Poles of transfer function = eigenvalues of A matrix.

2.

Let denote eigenvalues and v eigenvectors:


Av = v,
|I A| = 0
Roots of |zI A| gives eigenvalues of A matrix. Example:


1 2
A=
0 3


z 1 2
(zI A) =
0 z3


1
z3 2
(zI A)1 =
0 z1
(z 1)(z 3)
1 and 3 are poles of transfer function - also eigenvalues of A
Digital Control

Kannan M. Moudgalya, Autumn 2007

3.

State Space Approach to Pole Placement Controller

Recall our requirement and ss model:


x(k + 1) = Ax(k) + bu(k) + x0(k + 1).

Can we use a state feedback controller to get a desired closed


loop characteristic polynomial?

x1(k)
 x2(k)

+
u(k) = Kx(k) + v(k) = K1 K2 Kn
.
.
xn(k)
Substituting in state equation,
x(k + 1) = Ax(k) + b[Kx(k) + v(k)] + x0(k + 1)
= (A bK)x(k) + bv(k) + x0(k + 1)
A suitable K could make the system matrix behave better.
Digital Control

Kannan M. Moudgalya, Autumn 2007

4.

 Placement

 
SS - Pole
Controller
- Example
1 2
0
A=
,
b=
.
0 3
1

Eigenvalues of A are 1, 3. Unstable.



  

 



1 2
0
1 2
0 0
K1 K2 =
A bK =

0 3
1
0 3
K1 K2


1
2
=
K1 3 K2
If K1 = 0.5, K2 = 3.5,


1
2
A bK =
.
0.5 0.5
Eigenvalues are 0.25 0.6614j
Magnitude = 0.7071, stable
Digital Control

Kannan M. Moudgalya, Autumn 2007

5.

Use K to Get Good Characteristic Equation

Start with state space equation:


x(k + 1) = Ax(k) + bu(k) + x0(k + 1).
Use controller,
u(k) = Kx(k) + v(k)
Substitute in state space equation:
x(k + 1) = (A bK)x(k) + bv(k) + x0(k + 1)
Taking Z-transforms,
[zI (A bK)]X(z) = bV (z) + x0z
X(z) = [zI (A bK)]1(bV (z) + x0z)
adj[zI (A bK)]
=
(bV (z) + x0z)
|zI (A bK)|
Digital Control

Kannan M. Moudgalya, Autumn 2007

6.

SS Approach to Pole Placement Controller

|zI (A bK)| is the closed loop characteristic polynomial.


We may want the characteristic polynomial to be equal to
c(z) = z n + 1z n1 + + n1z + n
= (z 1)(z 2) (z n)
How does one find K to satisfy the above requirement?
Can one always find such a K?
Can equate the coefficients and solve for K. But not an easy
problem.
We will show that if the pair (A, b) is in controller canonical
form, to be defined next, K can be easily determined.

Digital Control

Kannan M. Moudgalya, Autumn 2007

7.

Controller Canonical form

(A, b) - controller canonical form if A and b are as in


a1 a2 an2 an1 an
1
1
0
0
0
0
0

.
0

.
1
0
0
0

A= .
, b=

0
.

0
0
1
0
0
0
0
0
0
1
0
Can show that (Problem 14.1)
|zI A| = z n + a1z n1 + + an
With u = Kx + v, we obtained
x(k + 1) = (A bK)x(k) + bv(k) + x0(k + 1)
Digital Control

Kannan M. Moudgalya, Autumn 2007

8.

SS Approach to Pole Placement Controller

A BK =

a1 K1 a2 K2

1
0

0
1

...

0
0
0
0

an2 Kn2 an1 Kn1 an Kn

0
0
0

0
0
0

1
0
0

0
1
0

(z) = |zI (A bK)| =


z n + (a1 + K1)z n1 + + (an1 + Kn1)z + (an + Kn)
If the desired characteristic polynomial is
c(z) = z n + 1z n1 + + n1z + n,
Equating coefficients of powers of z in (z) and c(z), can
determine [K1 Kn] = [1 a1 n an]

Digital Control

Kannan M. Moudgalya, Autumn 2007

9.

Ackermanns formula

Suppose that A and b are in controller canonical form.


Then the characteristic equation of A is given by
z n + a1z n1 + + an1z + an = 0
As per Cayley Hamiltons theorem, a matrix satisfies its own
characteristic equation:
An + a1An1 + + an1A + anI = 0
Evaluate the desired characteristic polynomial
c(z) = z n + 1z n1 + + n1z + n
at A to arrive at
c(A) = An + 1An1 + + n1A + nI
= (1 a1)An1 + + (n1 an1)A + (n an)I
Digital Control

Kannan M. Moudgalya, Autumn 2007

10.

Ackermanns Formula

Recall controller canonical form:


a1 a2 an2 an1 an
1
1
0
0
0
0
0

.
0

.
1
0
0
0

A= .
, b=

0
0
0
0
1
0
0
0
0
0
1
0
0
Pre-multiply powers of A by en:
eTn A = eTn1
eTn A2 = (eTn A)A = eTn1A = eTn2
..
eTn An1 = eT1
Digital Control

10

Kannan M. Moudgalya, Autumn 2007

11.

Ackermanns Formula - Continued

c(A) = (1 a1)An1 + + (n1 an1)A


+ (n an)I
Pre-multiplying eTn
eTn c(A) = (1 a1)eTn An1 + + (n1 an1)eTn A
+ (n an)eTn
= (1 a1)eT1 + + (n1 an1)eTn1
+ (n an)eTn


= 1 a1 n1 an1 n an = K
Thus the pole placing controller is given by
K = eTn c(A) = last row of c(A)
Digital Control

11

Kannan M. Moudgalya, Autumn 2007

What if A, b are not in Cont. Canonical Form


(A, b) not in controller canonical form

12.

M such that M 1AM = A, M 1b = b, (A, b) in


controller canonical form, iff C = [b Ab An1b] is
nonsingular - assume this for now
Define new state x: x = M x. M is nonsingular, constant
State space equation x(k + 1) = Ax(k) + bu(k) becomes,
M x(k + 1) = AM x(k) + bu(k)
x(k + 1) = M 1AM x(k) + M 1bu(k)
Define new variables,
4

A = M 1AM,

b = M 1b

State space equation in new coordinates (u not changed):


x(k + 1) = Ax(k) + bu(k)
Digital Control

12

Kannan M. Moudgalya, Autumn 2007

13.

Ackermanns Formula

New state x: x = M x
A=M

AM , b = M 1b

A, b are in controller canonical form


Can design K by equating coefficients:
u(k) = Kx(k) + v(k) = KM 1x(k) + v(k)
In original coordinates, controller: K = KM 1. Substitute in
x(k + 1) = Ax(k) + bu(k)
x(k + 1) = Ax(k) bKx(k) + bv(k)
= (A bK)x(k) + bv(k)
How useful is this characteristic polynomial?
Digital Control

13

Kannan M. Moudgalya, Autumn 2007

14.

Ackermanns Formula

Characteristic polynomial in new coordinates:


(z) = |zI (A bK)|
1

Substitute I = M 1M , A = M AM , b = M 1b


1
1
1

= zM M M AM M bKM
Pull out M 1 from left and M from right:
= |M 1 [zI (A bK)] M |
Determinant of products is product of determinants:
= |M 1M ||zI (A bK)| = (z)
Ch. polynomial of original and transformed system are the same.
Controller: K = eTn c(A).
Digital Control

14

Kannan M. Moudgalya, Autumn 2007

15.

Direct Controller Expression

Recall
x = M x, A = M 1AM,
K = KM 1
K = eTn c(A)

b = M 1b

Evaluate powers of A:
2

A = M 1AM M 1AM = M 1A2M


3

A = A A = M 1A2M M 1AM = M 1A3M


2

n1

Ab = M 1Ab, A b = M 1A2b A
b = M 1An1b
h
i


n1
1
n1
b Ab A
b
C = b Ab A
b =M
= M 1C
Digital Control

15

Kannan M. Moudgalya, Autumn 2007

16.

Evaluating Controller Expression Directly

Recall
C = M 1C
Therefore, C has full rank, if and only if C is of full rank.
K = eTn c(A) = eTn M 1c(A)M
This is because,
n

A = M 1AnM
Let us Calculate K:
K = KM 1 = eTn M 1c(A) = eTn CC 1c(A)
= eTn C 1c(A)
Ackermanns formula
Digital Control

16

Kannan M. Moudgalya, Autumn 2007

17.

Existence of Matrix M

There exists a nonsingular matrix M such that

a1 a2 an2 an1 an
1
0
0
0
0

1
0
0
0
0
1
M AM = ..
,
.

0
0
1
0
0
0
0
0
1
0


1
0

..
.
1
M b=
0

0
0

if and only if Rank C = n.


We show this fact as follows. Let the jth column of M be
denoted as mj , i.e.


M = m1 m2 mn
Write the top equation as AM = M the big matrix. Also,
b = M the unit vector = m1
Digital Control

17

Kannan M. Moudgalya, Autumn 2007

18.

Existence of Matrix M

a1 a2
1
0


 0
1
m1 mn
..

0
0
0
0



A m1 mn =

an2 an1 an
0
0
0

0
0
0

1
0
0
0
1
0

Equating the columns of the first equation, we obtain


Am1 = a1m1 + m2
Am2 = a2m1 + m3
..
Amn1 = an1m1 + mn
Digital Control

18

Kannan M. Moudgalya, Autumn 2007

19.

Existence of Matrix M

Recall the relations from the previous slide:


Am1 = a1m1 + m2
Am2 = a2m1 + m3
..
Amn1 = an1m1 + mn
With m1 = b, these are equivalent to
m1 = b
m2 = Am1 + a1m1 = Ab + a1b
m3 = Am2 + a2m1 = A2b + a1Ab + a2b
..
mn = An1b + a1An2b + + an1b
Digital Control

19

Kannan M. Moudgalya, Autumn 2007

Existence of Matrix M
m1 = b
m2 = Am1 + a1m1 = Ab + a1b
m3 = Am2 + a2m1 = A2b + a1Ab + a2b
..
mn = An1b + a1An2b + + an1b


is equivalent to m1 m2 mn =

1 a1 a2 an1
1 a1 an2



n1

1 an3
b Ab A
b

..
1

20.

This completes the proof


Digital Control

20

Kannan M. Moudgalya, Autumn 2007

You might also like