Sei sulla pagina 1di 33

Dr Lee Peng Hin : EE6203 Computer Control Systems 165 Dr Lee Peng Hin : EE6203 Computer Control Systems

tems 166

14 Observer Design Problem Formulation:

Design an observer which gives an


Motivations :
estimate of the system state from
• Some of the system states may not
the inputs and measured outputs.
be measurable and thus the state feed-
The estimate will be used to replace
back controller may not be realized.
the true states in the state feedback
• Even if some states are measurable,
controller. Two types of observers for
more sensors will be needed to im-
the state x(k).
plement the state feedback controller.
• Prediction observers : The predic-
The system that estimates the states tor estimate x̄(k) is obtained based
of the another system is called an ob- on the measurements upto y(k − 1).
server or a state estimator.
• Current observers : The current es-
timate x̄(k) is obtained based on the
measurement upto y(k), upto and
including the kth instant.
Dr Lee Peng Hin : EE6203 Computer Control Systems 167 Dr Lee Peng Hin : EE6203 Computer Control Systems 168

15 Prediction observers. The state equation of the plant is de-


scribed by,
(I). Open loop observer.
x(k + 1) = Ax(k) + Bu(k) (15.1)
The structure of an open loop observer
is given in Figure 9. The equation describing the observer
is

x̄(k + 1) = Ax̄(k) + Bu(k) (15.2)

The observer is implementable since we


know A, B and u(k).
Let

xe(k) = x(k) − x̄(k) (15.3)

be the estimation error. Then

xe(k + 1) = Axe(k) (15.4)

Figure 9: Open loop estimator


Dr Lee Peng Hin : EE6203 Computer Control Systems 169 Dr Lee Peng Hin : EE6203 Computer Control Systems 170

Remarks : (II). Closed loop observer. The struc-

• The observer is only applicable to tures of a closed loop observer is shown

stable systems. in Figure 10.

• If A is stable, then xe(k) will → 0 as


k → ∞.

• Any continuing measurement of the


system’s behaviour are not utilized.

Figure 10: Closed loop estimator


Dr Lee Peng Hin : EE6203 Computer Control Systems 171 Dr Lee Peng Hin : EE6203 Computer Control Systems 172

The model of the observer is

x̄(k + 1) = Ax̄(k) + Bu(k)


+Lo(y(k) − Cx̄(k)) (15.5)
= [A − LoC]x̄(k) + Bu(k)
+Loy(k) (15.6)

where
 T
Lo = lp1 lp2 . . . lpn

is the observer gain to be determined.


For the estimation error,

xe(k + 1)
= x(k + 1) − x̄(k + 1) (15.7)
= [A − LoC]xe(k) (15.8)

• (15.6) is a prediction estimator - mea-


surement at time k results in an esti-
Dr Lee Peng Hin : EE6203 Computer Control Systems 173 Dr Lee Peng Hin : EE6203 Computer Control Systems 174

mate of the state vector that is valid poles are at


at time k + 1. po1 , po2 , po3 , . . . , pon
• The dynamics of the estimation er- Then, the required closed loop charac.
ror is governed by the matrix [A − LoC]. poly for the observer is
• If Lo is chosen such that [A − LoC] αo(z)
is stable, xe(k) will approach zero re-
= (z − po1 )(z − po2 ) . . . (z − pon)
gardless of the value of x(0).
= z n + β1oz n−1 + · · · +
– The convergence speed is deter- o
βn−1 o
z + βn (15.10)
mined by the roots of the closed
Equate (15.9) and (15.10),
loop characteristic polynomial
det[zI − A + LoC]
det[zI − A + LoC] (15.9)
= z n + β1oz n−1 + · · · +
• The roots of (15.9) can be assigned o
βn−1 o
z + βn (15.11)
by the designer.
• The observer gain Lo can then be ob-
• Suppose that the desired observer
tained from (15.11).
Dr Lee Peng Hin : EE6203 Computer Control Systems 175 Dr Lee Peng Hin : EE6203 Computer Control Systems 176

Example 15.1. Consider the double inte- teristic polynomial is


grator plant discussed in Example 13.1, det[zI − A +
    LoC]   
1T 1 0 1 T lp1  
x(k + 1) = x(k) = det z − + 10
0 1 0 1 0 1 lp2
 
2
T /2 = z 2 + (lp1 − 2)z + T lp2 + 1 − lp1 (15.15)
+ u(k)(15.12)
T
  Equate (15.14) and (15.15) and solving
y(k) = 1 0 x(k) (15.13)
for T = 0.1 sec,
where the position x1(k) is the mea-
lp1 = 1.2, lp2 = 5.2
surement. Design an observer so that
Theorem : Consider the discrete time
the closed loop observer poles are at
system,
z1,2 = 0.4 ± j0.4
x(k + 1) = Ax(k) + Bu(k) (15.16)
The desired characteristic polynomial
y(k) = Cx(k) (15.17)
is
Let αo(z) be a polynomial of degree n,
αo(z) = z 2 − 0.8z + 0.32 (15.14) where n is the order of the system. As-
For an observer gain of Lo, the charac- suming that the system is observable,
Dr Lee Peng Hin : EE6203 Computer Control Systems 177 Dr Lee Peng Hin : EE6203 Computer Control Systems 178

then there exists a matrix Lo such that stitutions,


the matrix [A − LoC] of the observer in K → LT Wc → WoT A → AT
o
(15.6),
From (12.2),
x̄(k + 1) = [A − LoC]x̄(k) + Bu(k)   −1
Lo = 0 . . . 0 1 [WoT] αo(AT)
T

+Loy(k) (15.18) (16.1)


has the characteristic polynomial αo(z). or
 T
Lo = αo(A)Wo−1 0 ... 0 1 (16.2)
16 Computing observer gain by Acker-
where
mann’s formula
αo(A) = An + β1oAn−1
The determination of the matrix Lo in o o I (16.3)
+ · · · + βn−1 A + βn n
the observer is the same mathematical
and αo(z) is the desired estimator char-
problem as the problem of determin-
acteristic polynomial.
ing the feedback matrix K in the pole
Transposing (15.9), we get
placement problem by using the sub-
det[zI − AT + CTLT
o] (16.4)
Dr Lee Peng Hin : EE6203 Computer Control Systems 179 Dr Lee Peng Hin : EE6203 Computer Control Systems 180

Compare with (11.4), Example 16.1. Design an observer for the


servomotor in Example 2.1, which is
det [zI − A + BK]
modeled as  
i.e. K and LT
o play the same role. In 1 0.0952
x(k + 1) = x(k)
MATLAB, 0 0.905
 
0.00484
Lo = [acker(AT, CT, p)]T, or + u(k)(16.5)
0.0952
Lo = [place(AT, CT, p)]T  
y(k) = 1 0 x(k) (16.6)

Suppose that both the desired observer


roots are all chosen to be at 0.819. Hence,

αo(z) = (z − 0.819)2
= z 2 − 1.638z + 0.617

Thus,

A2 − 1.638A + 0.671I
αo(A) =   2
0.033 0.0254
=
0 0.00763
Dr Lee Peng Hin : EE6203 Computer Control Systems 181 Dr Lee Peng Hin : EE6203 Computer Control Systems 182

Now,
 −1  
C 1 0
Wo−1 = =
CA −10.51 10.51
The observer gain is
 
 T 0.267
Lo = αo(A)Wo−1 0 1 =
0.0802
and the observer obtained is

x̄(k + 1)

 − LoC]x̄(k) + Bu(k)
= [A  + Loy(k)
0.733 0.0952
= x̄(k) +
−0.0802 0.905
   
0.00484 0.267
u(k) + y(k)
0.0952 0.0802
Dr Lee Peng Hin : EE6203 Computer Control Systems 183 Dr Lee Peng Hin : EE6203 Computer Control Systems 184
Dr Lee Peng Hin : EE6203 Computer Control Systems 185 Dr Lee Peng Hin : EE6203 Computer Control Systems 186

17 Reduced-order observers

Motivation:

• Some states can be directly ob-


tained from measurements and
thus are not necessary to be es-
timated.

• Therefore, a lower order observer


can be designed to estimate those
states which are not available for
measurement.

Consider the discrete time system,

x(k + 1) = Ax(k) + Bu(k) (17.1)


y(k) = Cx(k) (17.2)
Dr Lee Peng Hin : EE6203 Computer Control Systems 187 Dr Lee Peng Hin : EE6203 Computer Control Systems 188

and partitioned the state vector as • For the unmeasurable states, xb(k),
 
xa(k) we have
x(k) =
xb(k)
xb (k + 1) = Abbxb (k)+
and let 
Abaxa(k) + Bbu(k) (17.5)
xa(k) − measurable, i.e. y(k)

xb(k) − unmeasurable, i.e. remaining states the term Abaxa(k) + Bbu(k) can

Then the state space model can be de- be considered as ‘known inputs’ into

composed to the xb (k) dynamics.


    
xa(k + 1) Aaa Aab xa(k) • For the measurable states xa(k), we
=
xb(k + 1) Aba Abb xb(k) have
 
Ba
+ u(k) (17.3) xa(k + 1) − Aaaxa(k)
Bb
 
  x (k) − Bau(k) = Aabxb(k) (17.6)
a
y(k) = I 0 (17.4) 
xb (k)
the LHS term, xa(k+1)−Aaaxa(k)−

Bau(k) is ‘known measurements’ and
Dr Lee Peng Hin : EE6203 Computer Control Systems 189 Dr Lee Peng Hin : EE6203 Computer Control Systems 190

can be regarded as ‘output’ from the full-order observer equations in (15.5).


xb(k) dynamics.
x̄(k + 1) = Ax̄(k) + Bu(k)+
Compare now the full order and re- Lo(y(k) − Cx̄(k)) (17.11)
duced order observer,
x(k) ← xb (k)
x(k + 1) = Ax(k) + Bu(k) (17.7)
A ← Abb
xb(k + 1) = Abbxb(k) + (17.8)

Bu(k) ← Abaxa(k) + Bbu(k)
Abaxa(k) + Bbu(k)
y(k) ← xa(k + 1) − Aaaxa(k) − Bau(k)
C ← Aab
y(k) = Cx(k) (17.9)
The resulting reduced order observer
xa(k + 1) − Aaaxa(k)− is then
Bau(k) = Aabxb(k) (17.10) x̄b(k + 1) = Abbx̄b (k)+

The reduced-order observer to estimate Abaxa(k) + Bbu(k) + Lr[xa(k + 1)−
state xb(k) can be obtained by mak-
Aaaxa(k) − Bau(k) − Aabx̄b(k)] (17.12)
ing the following substitutions into the
Dr Lee Peng Hin : EE6203 Computer Control Systems 191 Dr Lee Peng Hin : EE6203 Computer Control Systems 192

Since y(k) = xa(k), (17.12) gives

x̄b(k + 1) = [Abb − LrAab]x̄b(k)+


Lry(k + 1) + [Aba − LrAaa]y(k)+
[Bb − LrBa]u(k) (17.13)

The error equation is

xbe (k + 1) = [Abb − LrAab]xbe (k)

where Lr is the observer gain obtained


designed according to

det [zI − Abb + LrAab]


= αo(z) (17.14)
Dr Lee Peng Hin : EE6203 Computer Control Systems 193 Dr Lee Peng Hin : EE6203 Computer Control Systems 194

or obtained using Ackermann’s formula Example 17.1. Consider the discretised sys-
from (16.2), tem in Example 13.1 for T = 0.1s,
⎤−1 ⎡ ⎤
⎡  
Aab 0 1 0.1
⎢ ⎥ ⎢ ⎥ x(k + 1) = x(k)
⎢ AabAbb ⎥ ⎢ 0 ⎥ 0 1
Lr = αo(Abb) ⎢ ⎥ ⎢ ⎥  
⎢ .. ⎥ ⎢ .. ⎥ 0.05
⎣ ⎦ ⎣ ⎦ + u(k) (17.16)
AabAbb n−1 1 0.1
 
(17.15) y(k) = 1 0 x(k) (17.17)

Suppose that
   
x1(k) measured position of y(k)
=
x2(k) velocity to be estimated
then
   
Aaa Aab 1 0.1
=
Aba Abb 0 1
   
Ba 0.05
=
Bb 0.1
i.e. all are scalars and hence Lr is also
a scalar and there is only one estima-
Dr Lee Peng Hin : EE6203 Computer Control Systems 195 Dr Lee Peng Hin : EE6203 Computer Control Systems 196

tor pole to pick. Suppose we want the 18 Current observers.


estimator pole to be at z = 0.5, then
• The estimator discussed in the pre-
(17.14) gives
vious section is a prediction observer,
z − 1 + 0.1Lr = z − 0.5
since the estimate of x(k) is based on
⇒ Lr = 5 the measurement y(k − 1).
The estimator gain in (17.13), • We now consider a current estima-

x̄b(k + 1) = 0.5x̄b(k) + 5y(k + 1) tor that estimates states at time kT

− 5y(k) + 0.075u(k) (17.18) using the measurements at time kT .

Consider the discrete time system,

x(k + 1) = Ax(k) + Bu(k) (18.1)


y(k) = Cx(k) (18.2)

We want to estimate x(k) with x̄(k).


Dr Lee Peng Hin : EE6203 Computer Control Systems 197 Dr Lee Peng Hin : EE6203 Computer Control Systems 198

One form of current observer is given and what we expect the measure-
by the following two equations. ment to be at that time.

x̂(k + 1) = Ax̄(k) + Bu(k) (18.3) • Final estimate is x̄(k + 1).


x̄(k + 1) = x̂(k + 1) + Lc[y(k + 1) −
Now sub (18.3) into (18.4), we get
Cx̂(k + 1)] (18.4)
x̄(k + 1) = [A − LcCA]x̄(k)+
• In (18.3), x̂(k + 1) is the predicted
[B − LcCB]u(k) + Lcy(k + 1) (18.5)
estimate of the state at time (k +
1)T , based on the dynamics of the - a single equation for the estimate

system and on the signals at kT . at time (k+1)T based on the mea-


surement at (k + 1)T .
• In (18.4), estimate x̂(k + 1) is cor-
rected when the measurement y(k + Estimation error,

1) arrives at time (k + 1)T . xe(k + 1)

– Estimator gain Lc determines the = x(k + 1) − x̄(k + 1) (18.6)


weight placed on the difference be- = [A − LcCA]xe(k) (18.7)
tween measurement at (k + 1)T
Dr Lee Peng Hin : EE6203 Computer Control Systems 199 Dr Lee Peng Hin : EE6203 Computer Control Systems 200

Characteristic polynomial for the pre-


diction observer is

det [zI − A + LoC]

and characteristic polynomial for the


current observer is

det [zI − A + LcCA]

Thus, the current observer can be ob-


tained from that for the prediction ob-
server by replacing C with CA in the
Ackermann’s formula, i.e. from (16.2),
⎡ ⎤−1 ⎡ ⎤
CA 0
⎢ ⎥ ⎢ ⎥
⎢ CA2 ⎥ ⎢ 0 ⎥
Lc = αo(A) ⎢ ⎥ ⎢ ⎥
⎢ .. ⎥ ⎢ .. ⎥ (18.8)
⎣ ⎦ ⎣ ⎦
CAn 1
Dr Lee Peng Hin : EE6203 Computer Control Systems 201 Dr Lee Peng Hin : EE6203 Computer Control Systems 202

Example 18.1. Consider the discrete-time with


 
SS model of the servo motor obtained 0.033 0.0254
αo(A) =
0 0.00763
in Example 11.1.
  We have
1 0.0952  −1  
x(k + 1) = x(k) CA 2.104 −1.104
0 0.905 =
  CA2 −11.6 11.6
0.00484  −1  
+ u(k)(18.9)
0.0952 CA 0
  Lc = αo(A)
y(k) = 1 0 x(k) (18.10) CA2 1
 
0.258
with state feedback control gains =
0.0885
 
K = 4.52 1.12 and
 
0.742 0.0706
We use the same observer characteris- A − LcCA =
−0.0885 0.897
tic equation as used for the prediction  
0.00359
B − LcCB =
observer in Example 16.1. 0.0948

αo(z) = (z − 0.819)2 The current observer in (18.5) is im-

= z 2 − 1.638z + 0.617
Dr Lee Peng Hin : EE6203 Computer Control Systems 203 Dr Lee Peng Hin : EE6203 Computer Control Systems 204

plemented as 19 Combined control law and observer


 
0.742 0.0706
x̄(k + 1) = x̄(k) Idea :
−0.0885 0.897
   
0.00359 0.258 Using the observed states to re-
+ u(k) + y(k + 1)
0.0948 0.0885 place the true states in the state
feedback controller. Thus, the con-
troller is

u(k) = −Kx̄(k)

where x̄(k) is the estimate of the


true state x(k) obtained from an
observer.

The system structure is shown in Fig-


ure 7.
Dr Lee Peng Hin : EE6203 Computer Control Systems 205 Dr Lee Peng Hin : EE6203 Computer Control Systems 206

• The poles of the closed loop system Combining (19.2) with the estimator
are located at gain from (15.8),

p1, p2, . . . , pn po1 , po2 , . . . , pon xe(k + 1) = [A − LoC]xe(k) (19.3)
i.e. the combination of all the con-
we get the coupled equations of the
trol poles and estimator poles.
complete system as
• The characteristic polynomial of the
 
complete control system is αc(z)αo(z) . x(k + 1)
This is the Separation Principle by xe(k + 1)
  
which a controller and an observer A − BK BK x(k)
= (19.4)
0 A − Lo C xe(k)
can be designed separately yet used
together. This is shown as follows :

The closed loop system is

x(k + 1) = Ax(k) + Bu(k) (19.1)


= Ax(k) −
BK[x(k) − xe(k)](19.2)
Dr Lee Peng Hin : EE6203 Computer Control Systems 207 Dr Lee Peng Hin : EE6203 Computer Control Systems 208

Then, the characteristic polynomial is Example 19.1. The regulator for the servo
  
A − BK BK motor in Example 2.1 modeled as
det zI −  
0 A − Lo C 1 0.0952
   x(k + 1) = x(k)
zI − A + BK −BK 0 0.905
= det  
0 zI − A + LoC 0.00484
  + u(k)(19.5)
= det zI − A + BK det zI − A + LoC 0.0952
 
= αc(z)αo(z) y(k) = 1 0 x(k) (19.6)

Guidelines for choosing αc(z) and αo(z) is obtained by combining the state feeed-
back controller designed in Example 11.1
1. Choose the roots of αc(z) to satisfy
with the state observer designed in Ex-
the performance specifications and
ample 16.1. In the controller, the true
actuator limitations.
state is replaced by the estimate from
2. Choose the roots of αo(z) faster (by
the full-order observer. Thus, the reg-
a factor of 2 to 4) so that the total re-
sponse is dominated by the response
of the poles in αc(z).
Dr Lee Peng Hin : EE6203 Computer Control Systems 209 Dr Lee Peng Hin : EE6203 Computer Control Systems 210

ulator is described as To show that the closed loop poles


 
0.733 0.0952 are at 0.888 ± j0.173; 0.819; 0.819: we
x̄(k + 1) = x̄(k) +
−0.0802 0.905
    have,
0.00484 0.267  
u(k) + y(k) 1 0.0952
0.0952 0.0802 x(k + 1) = x(k) +
  0 0.905
u(k) = − 4.52 1.12 x̄(k)  
0.00484
u(k)
so that 0.0952
   
0.711 0.0898 u(k) = − 4.52 1.12 x̄(k)
x̄(k + 1) = x̄(k) +  
−0.510 0.798 0.711 0.0898
  x̄(k + 1) = x̄(k) +
0.267 −0.510 0.798
y(k)  
0.0802 0.267
  y(k)
u(k) = − 4.52 1.12 x̄(k) 0.0802
 
y(k) = 1 0 x(k)
It can be checked that the poles of the
closed loop system are located at

0.888 ± j0.173, 0.819, 0.819


Dr Lee Peng Hin : EE6203 Computer Control Systems 211 Dr Lee Peng Hin : EE6203 Computer Control Systems 212

Hence,
 
Now we look at an example on the com-
1 0.0952
x(k + 1) = x(k) + bination of a reduced order observer
0 0.905
 
−0.02188 −0.00542 and state feedback controller.
x̄(k)
−0.4303 −0.1066 Example 19.2. The servo motor in Exam-
 
0.711 0.0898
x̄(k + 1) = x̄(k) + ple 2.1 is modeled as
−0.510 0.798  
 
0.267 0 1 0.0952
x(k) x(k + 1) = x(k) +
0.0802 0 0 0.905
 
The closed-loop state equation is, 0.00484
  u(k) (19.7)
x(k + 1) 0.0952
x̄(k + 1)  
⎡ ⎤ y(k) = 1 0 x(k) (19.8)
1 0.0952 −0.02188 −0.00542  
⎢ ⎥
⎢ 0 0.905 −0.4303 −0.1066 ⎥ x(k) We are measuring position x1(k) and
= ⎢
⎢ 0.267

⎣ 0 0.711 0.0898 ⎥
⎦ x̄(k)
0.0802 0 −0.510 0.798
will estimate velocity x2(k). As in Ex-
 
x(k) ample 11.1, the closed loop controller
= Acl
x̄(k)
characteristic equation (11.15) is cho-
The closed-loop poles are
sen to be
λi(Acl ) = 0.888 ± j0.173; 0.819; 0.819
αc(z) = z 2 − 1.776z + 0.819 = 0 (19.9)
Dr Lee Peng Hin : EE6203 Computer Control Systems 213 Dr Lee Peng Hin : EE6203 Computer Control Systems 214

In Example 16.1, the estimator roots (17.13),


are chosen to be at z = 0.819. The
x̄b(k+1) = [0.905−(0.903)(0.0952)]x̄b (k)
same choice is made here but the re-
+ 0.903y(k + 1) + [0 − (0.903)(1)]y(k)
duced order observer is first order. Hence,
+ [0.0952 − (0.903)(0.00484)]u(k)
αo(z) = z − 0.819 (19.11)
From the plant equations above and or

(17.3) and (17.4), the partitioned ma- x̄b(k + 1) = 0.819x̄b(k) + 0.903y(k + 1)


trices are, − 0.903y(k) + 0.0908u(k)
Aaa = 1, Aab = 0.0952, Ba = 0.00484 Here, x̄b(k) is the estimate of velocity
Aba = 0, Abb = 0.905, Bb = 0.0952 x2(k). Since we have considered the

From the Ackermann’s formula in (17.15), measurement y(k) to be the measure-


ment at the present time, the imple-
Lr = (0.905 − 0.819)(0.0952)−1 (1)
mentation of the observer is more ob-
= 0.903 (19.10)
vious if we replace k with (k − 1)in the
The reduced order observer is given in
Dr Lee Peng Hin : EE6203 Computer Control Systems 215 Dr Lee Peng Hin : EE6203 Computer Control Systems 216

observer equation, or

x̄b(k) = 0.819x̄b (k − 1) + 0.903y(k) x̄b(k + 1) = 0.717x̄b(k)

− 0.903y(k − 1) + 0.0908u(k − 1) + 0.903y(k + 1) − 1.313y(k)

From Example 11.1, the controller is The control system is implemented as


  follows. A measurement y(k) is made
u(k) = − 4.52 1.12 x(k)
at t = kT . The observer state is cal
which can be implemented as
from
u(k) = −4.52y(k) − 1.12x̄b (k)
x̄b(k) = 0.717x̄b (k − 1)
Hence we can write the observer equa- + 0.903y(k) − 1.313y(k − 1)
tion as
Then, the control input is cal, using
x̄b(k + 1) = 0.819x̄b(k) + 0.903y(k + 1)
 u(k) = −4.52y(k) − 1.12x̄b (k)
−0.903y(k)+0.0908 −4.52y(k)−1.12x̄b (k)
Dr Lee Peng Hin : EE6203 Computer Control Systems 217 Dr Lee Peng Hin : EE6203 Computer Control Systems 218
Dr Lee Peng Hin : EE6203 Computer Control Systems 219 Dr Lee Peng Hin : EE6203 Computer Control Systems 220
Dr Lee Peng Hin : EE6203 Computer Control Systems 221 Dr Lee Peng Hin : EE6203 Computer Control Systems 222

Show that a pole-placement design im-


plemented with a current observer has
a controller-estimator transfer function
U (z)
Y (z)
given by

U (z)
= −zK zI − A + LcCA+
Y (z)
−1
BK − LcCBK Lc

We have the current estimator as,

x̄(k + 1) = (A − LcCA)x̄(k)+
(B − LcCB)u(k) + Lcy(k + 1)

Taking z-transforms, we have

z X̄(z) = (A − LcCA)X̄(z)+
(B − LcCB)U (z) + LczY (z)
Dr Lee Peng Hin : EE6203 Computer Control Systems 223 Dr Lee Peng Hin : EE6203 Computer Control Systems 224

But, since u(k) = −K x̄(k), 20 System with inputs

U (z) = −K X̄(z) Consider the discrete time system,


Hence, x(k + 1) = Ax(k) + Bu(k) (20.1)

X̄(z) = z zI − A + LcCA+ y(k) = Cx(k) (20.2)
−1 and
BK − LcCBK LcY (z)
u(k) = −Kx(k) + Krr(k)
Therefore,
 where
U (z)
= −zK zI − A + LcCA+ r(k) = system input
Y (z)
−1
BK − LcCBK Lc Kr = constant to be determined
to satisfy specifications

The closed loop system is

x(k + 1) = [A − BK]x(k)
+BKrr(k) (20.3)
y(k) = Cx(k) (20.4)
Dr Lee Peng Hin : EE6203 Computer Control Systems 225 Dr Lee Peng Hin : EE6203 Computer Control Systems 226

which gives
Y (z)
= C[zI − A + BK]−1BKr
R(z)
• Kr do not affect location of closed
loop poles.

• Design of state feedback K can pro-


ceed as before.

• Choose Kr e.g. to satisfy DC gain


specifications.
Dr Lee Peng Hin : EE6203 Computer Control Systems 227 Dr Lee Peng Hin : EE6203 Computer Control Systems 228

21 Integral control

• The idea of integral control by state


augmentation is to add an integra-
tor so as to obtain an integral of the
error signal.

• Integral control is to eliminate steady


state error due to constant reference
input commands.

• One way to intro an integrator in the


mathematical model of a closed loop
system is to intro a new state vec-
tor that integrates the difference be-
tween the command vector r(k) and Figure 11: Block diagram for integral control with full state feedback

the output vector y(k). See Fig 11.


Dr Lee Peng Hin : EE6203 Computer Control Systems 229 Dr Lee Peng Hin : EE6203 Computer Control Systems 230

The plant is Apply pole placement technique for sys-

x(k + 1) = Ax(k) + Bu(k) (21.1) tems with inputs.

y(k) = Cx(k) (21.2)

Augment the state with xI , the inte-


gral of the error, e = y − r.

xI (k + 1) = xI (k) + e(k)
= xI (k) + Cx(k) − r(k)
 
x(k)
Define the new state vector as .
xI (k)
Then,
    
x(k + 1) A 0 x(k)
=
xI (k + 1) C 1 xI (k)
   
B 0
+ u(k) − r(k)
0 1
and  
  x(k)
u(k) = − K KI + KNxr(k)
xI (k)

Potrebbero piacerti anche