Lecture: Integral action in state feedback control

Automatic Control 1

Integral action in state feedback control

Prof. Alberto Bemporad
University of Trento

Academic year 2010-2011

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

1 / 15

Lecture: Integral action in state feedback control

Adjustment of DC-gain for reference tracking

Reference tracking
Assume the open-loop system completely reachable and observable
We know state feedback we can bring the output y(k) to zero asymptotically
How to make the output y(k) track a generic constant set-point r(k) r ?
Solution: set u(k) = Kx(k) + v(k)
v(k) = Fr(k)
We need to choose gain F properly to ensure reference tracking
!"#$%&'$()*+,'-..

r(k)

u(k)

v(k)

A,B

x(k)

y(k)

K
',#/+,((-+

Prof. Alberto Bemporad (University of Trento)

x(k + 1)

(A + BK)x(k) + BFr(k)

y(k)

Cx(k)

Automatic Control 1

Academic year 2010-2011

2 / 15

Lecture: Integral action in state feedback control

Adjustment of DC-gain for reference tracking

Reference tracking

To have y(k) r we need a unit DC-gain from r to y

C(I (A + BK))1 BF = I
Assume we have as many inputs as outputs (example: u, y R)
Assume the DC-gain from u to y is invertible, that is C Adj(I A)B invertible
Since state feedback doesnt change the zeros in closed-loop
C Adj(I A BK)B = C Adj(I A)B
then C Adj(I A BK)B is also invertible
Set
F = (C(I (A + BK))1 B)1

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

3 / 15

Lecture: Integral action in state feedback control

Adjustment of DC-gain for reference tracking

Example
Poles placed in (0.8 0.2j, 0.3). Resulting
closed-loop:

1.4
1.2

x(k + 1)

y(k)

u(k)

1.1
0

1
0
x(k) +
u(k)
0.8
1

1 0 x(k)

0.13 0.3 x(k) + 0.08r(k)

1
0.8
0.6
0.4
0.2
0
0

The transfer function G(z) from r to y is

2
, and G(1) = 1
G(z) = 25z2 40z+17

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

10

20
sample steps

30

40

Unit step response of the closed-loop

system (=evolution of the
system from

initial condition x(0) = 00 and
reference r(k) 1, k 0)

Academic year 2010-2011

4 / 15

Lecture: Integral action in state feedback control

Adjustment of DC-gain for reference tracking

Reference tracking

Problem: we have no direct feedback on the tracking error e(k) = y(k) r(k)
Will this solution be robust with respect to model uncertainties and
exogenous disturbances ?
Consider an input disturbance d(k) (modeling for instance a non-ideal
actuator, or an unmeasurable disturbance)
&#*/0)!&.0/+1$#'!"#$%&'$()*+,'-..

d(k)
r(k)

+
+

u(k)

+
+

A,B

x(k)

y(k)

K
',#0+,((-+

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

5 / 15

Lecture: Integral action in state feedback control

Adjustment of DC-gain for reference tracking

Example (contd)
Let the input disturbance d(k) = 0.01, k = 0, 1, . . .
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0

10

20
sample steps

30

40

The reference is not tracked !

The unmeasurable disturbance d(k) has modified the nominal conditions for
which we designed our controller
Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

6 / 15

Lecture: Integral action in state feedback control

Integral action

Integral action for disturbance rejection

Consider the problem of regulating the output y(k) to r(k) 0 under the
action of the input disturbance d(k)
Lets augment the open-loop system with the integral of the output vector
q(k + 1) = q(k) + y(k)
|
{z
}
integral action

The augmented system is





  
 
x(k + 1)
A 0
x(k)
B
B
=
+
u(k) +
d(k)
q(k + 1)
C I
q(k)
0
0



x(k)
C 0
y(k) =
q(k)
Design a stabilizing feedback controller for the augmented system



x(k)
u(k) = K H
q(k)
Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

7 / 15

Lecture: Integral action in state feedback control

Integral action

Rejection of constant disturbances

&#*/0)!&.0/+1$#'!"#$%&'$()*+,'-..

d(k)
+

u(k)

+
+

.0$0-)3--!1$'4

q(k)
&#0-2+$()$'0&,#

A,B

x(k)

y(k)

Theorem
Assume a stabilizing gain [H K] can be designed for the system
augmented with integral action. Then limk+ y(k) = 0 for all constant
disturbances d(k) d

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

8 / 15

Lecture: Integral action in state feedback control

Integral action

Rejection of constant disturbances

&#*/0)!&.0/+1$#'!"#$%&'$()*+,'-..

d(k)
+

u(k)

+
+

.0$0-)3--!1$'4

q(k)
&#0-2+$()$'0&,#

A,B

x(k)

y(k)

Proof:
The state-update matrix of the closed-loop system is

  


A 0
B
K H
+
C I
0
The matrix has asymptotically stable eigenvalues by construction
h
i
x(k)
For a constant excitation d(k) the extended state q(k) converges to a

steady-state value, in particular limk q(k) = q

q
=0
Hence, limk y(k) = limk q(k + 1) q(k) = q
Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

9 / 15

Lecture: Integral action in state feedback control

Integral action

Example (contd) Now with integral action

Poles placed in (0.8 0.2j, 0.3) for the augmented system. Resulting closed-loop:


 
1.1 1
0
x(k + 1) =
x(k) +
(u(k) + d(k))
0 0.8
1
q(k + 1)

y(k)

u(k)

q(k) + y(k)

1 0 x(k)

0.48 1 x(k) 0.056q(k)

Closed-loop simulation for x(0) = [0 0]0 , d(k) 1:

3
2.5
2
1.5
1
0.5
0
0.5
0
Prof. Alberto Bemporad (University of Trento)

10

20
sample steps
Automatic Control 1

30

40
Academic year 2010-2011

10 / 15

Lecture: Integral action in state feedback control

Integral action

Integral action for set-point tracking

&#*/0)!&.0/+1$#'!"#$%&'$()*+,'-..

d(k)
r(k)

+
-

q(k)
&#0-2+$()
$'0&,#

0+$'4&#2)-++,+

u(k)

+
+

A,B

.0$0-)3--!1$'4

x(k)

y(k)

Idea: Use the same feedback gains (K, H) designed earlier, but instead of feeding
back the integral of the output, feed back the integral of the tracking error
q(k + 1) = q(k) + (y(k) r(k))
|
{z
}
integral action

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

11 / 15

Lecture: Integral action in state feedback control

Integral action

Example (contd)
3.5

x(k + 1)

q(k + 1)

1.1 1
x(k)
0 0.8
 
0
+
(u(k) + d(k))
1
q(k) + (y(k) r(k))
|
{z
}

3
2.5
2
1.5
1
0.5

tracking error

y(k)
u(k)

0
0

0 x(k)

0.48

1 x(k) 0.056q(k)

10

20
sample steps

30

40

Response for x(0) = [0 0]0 ,

d(k) 1, r(k) 1

Looks like its working . . . but why ?

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

12 / 15

Lecture: Integral action in state feedback control

Integral action

Tracking & rejection of constant disturbances/set-points

Theorem
Assume a stabilizing gain [H K] can be designed for the system
augmented with integral action. Then limk+ y(k) = r for all constant
disturbances d(k) d and set-points r(k) r
Proof:
The closed-loop system




 


x(k + 1)
A + BK BH
x(k)
B 0
d(k)
=
+
q(k + 1)
C
I
q(k)
0 I
r(k)



x(k)
C 0
y(k) =
q(k)
d(k)
has input r(k) and is asymptotically stable by construction
h
i

x(k)
For a constant excitation d(k)
the
extended
state
converges to a
q(k)
r(k)

steady-state value, in particular limk q(k) = q

q
=0
Hence, limk y(k) r(k) = limk q(k + 1) q(k) = q
Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

13 / 15

Lecture: Integral action in state feedback control

Integral action

Integral action for continuous-time systems

The same reasoning can be applied to continuous-time systems
x(t)

Ax(t) + Bu(t)

y(t)

Cx(t)

Augment the system with the integral of the output q(t) =

Rt
0

y()d, i.e.,

(t) = y(t) = Cx(t)

q
{z
}
|
integral action

The augmented system is



d x(t)
dt q(t)
y(t)

=
=

A
C

0
0



x(t)
q(t)

x(t)
q(t)

B
0


u(t)

Design a stabilizing controller [K H] for the augmented system

Implement
Zt
u(t) = Kx(t) + H

(y() r())d

0
Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

14 / 15

Lecture: Integral action in state feedback control

Integral action

English-Italian Vocabulary

reference tracking
steady state
set point

inseguimento del riferimento

regime stazionario
livello di riferimento

Translation is obvious otherwise.

Prof. Alberto Bemporad (University of Trento)

Automatic Control 1

Academic year 2010-2011

15 / 15