Lecture 14 - Controllability and Observability Introduction

K. J. Åström • The concepts of controllability and observability introduced

as conditions to solve problems of state feedback and
1. Introduction observers
2. The Concepts • More insight
3. Structure of Linear Systems • Kalmans decomposition
4. Cancellation of Poles and Zeros • System structures
5. Summary • Cancellation of poles and zeros

Theme: A closer look at the controllability and observability

and the structure of linear systems.

The Concepts Algebraic Criteria

The system
dx
= Ax
dt dx
= Ax + Bu, y = Cx
y = Cx dt

Controllability: Assume that the system is at the origin initially. is controllable if the matrix
Can we find a control signal so that the state reaches a given
Wc = ( B AB A2 B ... A n− 1 B )
position at a fixed time? Notice we do not require that it stays
there! has full rank. The system observable if the matrix
Observability: Can the state x be determined from observa-  
C
tions of the output y over some time interval.  CA 
 
 
Wo = 

CA2 

 .. 
 . 
n− 1
CA


c K. J. Åström August, 2000
has full rank. 1
 Prototype of Noncontrollable System Prototype of Nonobservable System
Two identical systems driven by the same input. Intuitively: no Two identical systems whose outputs are added. Intuitively: no
way to make the systems move in opposite ways. way to find out which system generated the output.
A simple example A simple example
dx1 S dx1
S
= − x1 + u = − x1 + u
dt dt
dx2 u dx2 y
dt
= − x2 + u
dt
= − x2 + u
Σ
y = x1 + x2

The linear combination x 1 + x2 is S The linear combination x 1 − x2 S

controllable is not observable

Duality Canonical Forms

Controllability of Controllable canonical form
 −a −a . . . −a 


1 2 n
  
dx 
   1
= Ax + Bu 
 1 0 0 
 




dt 




 
 0 

dz  
0 z+  
.
= 0 1  u

 
 
 .. 

dt 
 . 
 
 

is the same as observability for 
 . 
  
 

 . 

  0
0 0 0
dx  
= AT x y =  b1 b2 . . . bn  z + Du
dt
y = BTx Observable canonical form
   
 − a1 1 0 ... 0  b1 
Controllability to observability through the transformation 
 
 
 


 − a 0 1 0 
 b2 


 2 
 
 

dz  
 .





 .



A → AT =
 . 

 z + 

 . 

 u
dt  .
 
 

. 

 
  
 

B → CT 
 − an−1 0 0 1 
 


 
  
 
− an 0 0 0 bn
Wc → Wo  
y = 1 0 0... 0  z + Du


c K. J. Åström August, 2000 2
 System Structure Kalmans Decomposition
The coordinates can be chosen so that a linear system has the A linear system can be transformed to the form
following structure    
A11 0 A13 0 B1
xc A11 A12 xc B1 A A24  B 
d dx  21 A22 A23   2
= + u = x+ u
dt xc̄ 0 A22 xc̄ 0 dt  0 0 A33 0   0 
0 0 A43 A44 0
where the states x c are controllable and x c̄ are noncontrollable.
y = ( C1 0 C2 0) x
d xo A11 0 xo
= where the state vector has been partitioned as
dt xō A21 A22 x0̄
xo  T
y = ( C1 0) xco
xō x 
 cō 
x= 
where the states x o are observable and x ō not observable  xc̄o 
(quiet) xc̄ ō

Kalmans Decomposition System with State Feedback and Observers

Partitioning of state space
u y
• Sco controllable and Soc Σ
dx
= Ax + Bu
observable dt
y = Cx
• Sc ō controllable not u = L( xm − x̂) + lr r
observable -
Soc Soc- d x̂
= A x̂ + Bu + K ( y − C x̂ )
• Sc̄o not controllable dt
observable
Replace x̂ by x̃ = x − x̂
--
Soc
• Sc̄ō not controllable not
observable dx
= ( A − B L) x + B x̃ + B ( Lxm + lr r)
dt
The transfer function is given by the subsystem S co d x̃
= ( A − K C) x̃
dt

Observer error not controllable from r. Makes a lot of sense

because we do not want reference signals to generate ob-
 server2000
c K. J. Åström August, errors! 3
 Disturbance Observer Cancellation of Poles and Zeros
The following system has been used to model constant load Formal calculations with Laplace transforms sometimes leads
disturbances to cancellation of poles and zeros.
dx
= Ax + B (u + v) Consider the system
dt
dv
=0
dt dy du
= → y( t) = u( t) + constant
y = Cx dt dt

The system can be written as Take Laplace transforms (assuming all initial values zero)
d x A B x B
= + u sY ( s) = sU ( s), → Y ( s) = U ( s) → y( t) = u( t)
dt v 0 0 v 0
x
y = (C 0) There are also design methods where it is deliberately at-
v
tempted to cancel poles and zeros. It is important to under-
Notice that the state v which models the load disturbance is not stand what happens when this is done. The decomposition of a
controllable from u. linear system gives good insight into what happens when work-
ing with transfer functions.

Example of Cancellation Example of Cancellation ...

Consider the system Introduce the state representation

u s+a υ s+b y dx1

= −ax1 + ( b − a)( x2 + u)
dt
s+b s+a dx2
= −bx2 + ( a − b)u
dt
The system has the transfer function G (s) = 1. Natural y = x1 + v = x1 + x2 + u
questions: v = x2 + u

• Is the system equivalent to the system y = u? Change coordinates to

• What happens with the modes that are cancelled?
1
z1 = ( x1 + x2 ) x1 = z1 + z2
2
1
z2 = ( x1 − x2 ) x2 = z1 − z2
2


c K. J. Åström August, 2000 4
 Example of Cancellation ... Example of Cancellation ...
u y
Σ The system has two states, one state corresponding to the
Equations in new coordinates mode s = −a, is controllable and the state corresponding to
2 the mode s = −b is also controllable unless a = b. The state
dz1
= −az1 1 z1 z1 is observable but not the state z 2 . There are four interesting
dt s +a
dz2 cases:
= −bz2 + ( b − a) z1 + ( b − a)u
dt • a > 0 and b > 0:
y = 2z1 + u Σ
b−a z2 −
Σ
s+b both modes stable
v = z1 − z2 + u
Σ
υ • a < 0 and b > 0:
The state z1 is observable but not controllable mode s = −a unstable mode s = −b stable
• a > 0 and b < 0:
The state z2 is controllable but not observable
mode s = −a stable mode s = −b unstable
• a < 0 och b < 0:
both modes unstable

Lambda Tuning of PI Contollers Properties of Closed Loop System

b Response to reference signal R and load disturbance D
Process: P(s) =
s+a
b( ks + ki ) bs
ki sk + ki Y ( s) = R( s ) + 2 D ( s)
Controller: C(s) = k + = + ( a + bk) s + bki
s2 s + ( a + bk) s + bki
s s α bs
= R( s) + D ( s)
Choose controller parameters to cancel process pole at s = s +α ( s + α )( s + a)
−a, hence k i/ k = a. The loop transfer function becomes ( ks + ki )( s + a) b( ks + ki )
U ( s) = 2 R( s ) − 2 D ( s)
s + ( a + bk) s + bki s + ( a + bk) s + bki
kb α s+a α
L( s ) = = R( s) − D ( s)
s b s +α s+α

and the characteristic polynomial is s + kb. Choose controller Consider the situation when plant dynamics is slow (a small
gain k to get closed loop pole at s = −α , hence k = α /b. 0.1) and desired response fast (α large 1.0).
Are there any drawbacks with the cancellation?

The Audience is Thinking ...


c K. J. Åström August, 2000 5
 Properties of Closed Loop System ... Easy to Fix the Problem
0.8
Characteristic polynomial
0.6

0.4 s2 + ( a + bk) s + bki = s2 + 1.1s + ki

0.2

0
0 5 10 15 20 25 30 35 40
Change k i from 0.1 to 1!
1
0

0.5

−0.5 0

−0.5
0 5 10 15 20 25 30 35 40
−1 0
0 5 10 15 20 25 30 35 40
−0.5

Notice that after the initial reaction the control signal does not −1

react even if there is a large deviation. This is an effect of the 0 5 10 15 20 25 30 35 40

cancellation. The controller zero at s = -0.1 blocks transmission

Be careful with cancellation of unstable or slow modes!
of the signal e−0.1t !

Summary
• The notions of controllability and observability were
introduced by Kalman in 1960.
• The arise from questions that are natural to ask when
state models are used:
– Can the states be moved from one position to another?
– Can the state be computed from the measured signal.
• The conditions are required to solve the pole placement
and the observer problems.
• The concepts give a natural decomposition of a system
which gives insight into the cancellation problem.


c K. J. Åström August, 2000 6