ELEC4410
Control System Design
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes
School of Electrical Engineering and Computer Science The University of Newcastle
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 1/36
Outline
Afﬁne Parameterisation  Open Loop Unstable Model.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 2/36
Outline
Afﬁne Parameterisation  Open Loop Unstable Model.
Saturation and Slew Rate Limitations.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 2/36
Outline
Afﬁne Parameterisation  Open Loop Unstable Model.
Saturation and Slew Rate Limitations.
Antiwindup Schemes.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 2/36
Outline
Afﬁne Parameterisation  Open Loop Unstable Model.
Saturation and Slew Rate Limitations.
Antiwindup Schemes.
Reference: Control System Design, Goodwin, Graebe & Salgado.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 2/36
Afﬁne Parameterisation  Open Loop Unstable Model
Recall for afﬁne parameterisation for the stable case:
C(s) =
^{Q}^{(}^{s}^{)}
1 − Q(s)G _{o} (s)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 3/36
Afﬁne Parameterisation  Open Loop Unstable Model
If the plant contains unstable poles then its output could increase exponentially.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 4/36
Afﬁne Parameterisation  Open Loop Unstable Model
If the plant contains unstable poles then its output could increase exponentially.
Likewise the nominal model, G _{o} (s), will also contain unstable poles, hence its output can also increase exponentially.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 4/36
Afﬁne Parameterisation  Open Loop Unstable Model
If the plant contains unstable poles then its output could increase exponentially.
Likewise the nominal model, G _{o} (s), will also contain unstable poles, hence its output can also increase exponentially.
This could result in the difference between the plant and nominal model outputs being zero and hence no control action being taken.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 4/36
Afﬁne Parameterisation  Open Loop Unstable Model
If the plant contains unstable poles then its output could increase exponentially.
Likewise the nominal model, G _{o} (s), will also contain unstable poles, hence its output can also increase exponentially.
This could result in the difference between the plant and nominal model outputs being zero and hence no control action being taken.
The University of Newcastle
'0'
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 4/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now in the unstable case,
C(s) =
Q(s)
1 − Q(s)G _{o} (s) ^{,}
is still okay! But, we have to make some further points regarding stability of the sensitivity functions.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 5/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now in the unstable case,
C(s) =
Q(s)
1 − Q(s)G _{o} (s) ^{,}
is still okay! But, we have to make some further points regarding stability of the sensitivity functions.
Essentially we need to add more interpolation constraints.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 5/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now in the unstable case,
C(s) =
Q(s)
1 − Q(s)G _{o} (s) ^{,}
is still okay! But, we have to make some further points regarding stability of the sensitivity functions.
Essentially we need to add more interpolation constraints.
To ensure T _{o} (s), S _{o} (s), S _{i}_{o} (s) and S _{u}_{o} (s) are stable, we still need Q(s) stable and proper. In addition, we add further interpolation constraints:
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 5/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now in the unstable case,
C(s) =
Q(s)
1 − Q(s)G _{o} (s) ^{,}
is still okay! But, we have to make some further points regarding stability of the sensitivity functions.
Essentially we need to add more interpolation constraints.
To ensure T _{o} (s), S _{o} (s), S _{i}_{o} (s) and S _{u}_{o} (s) are stable, we still need Q(s) stable and proper. In addition, we add further interpolation constraints:
Q(s)G _{o} (s) stable. [Unstable poles of G _{o} (s) → Zeros of Q(s)]
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 5/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now in the unstable case,
C(s) =
Q(s)
1 − Q(s)G _{o} (s) ^{,}
is still okay! But, we have to make some further points regarding stability of the sensitivity functions.
Essentially we need to add more interpolation constraints.
To ensure T _{o} (s), S _{o} (s), S _{i}_{o} (s) and S _{u}_{o} (s) are stable, we still need Q(s) stable and proper. In addition, we add further interpolation constraints:
Q (s)G _{o} (s) stable. [Unstable poles of G _{o} (s) → Zeros of Q(s)]
(1 − Q(s)G _{o} (s))G _{o} (s) stable.
[Unstable poles of G _{o} (s) → Zeros of 1 − Q(s)G _{o} (s)]
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 5/36
Afﬁne Parameterisation  Open Loop Unstable Model
Recall from last lecture, that unstable polezero cancellations should be performed analytically.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 6/36
Afﬁne Parameterisation  Open Loop Unstable Model
Recall from last lecture, that unstable polezero cancellations should be performed analytically.
To begin, we express the nominal model in its fractional form
G _{o} (s) = ^{B} ^{o} ^{(}^{s}^{)} A _{o} (s)
and assume all the poles of A _{o} (s) are unstable.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 6/36
Afﬁne Parameterisation  Open Loop Unstable Model
Recall from last lecture, that unstable polezero cancellations should be performed analytically.
To begin, we express the nominal model in its fractional form
G _{o} (s) = ^{B} ^{o} ^{(}^{s}^{)} A _{o} (s)
and assume all the poles of A _{o} (s) are unstable.
Next we choose
˜
Q(s) =
˜
P(s)
˜
E(s)
˜
where E(s) is stable. In particular the zeros of E(s) lie in a desirable region of the complex plane.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 6/36
Afﬁne Parameterisation  Open Loop Unstable Model
As we need unstable poles of G _{o} (s) to be zeros in Q(s), we can write
¯
Q(s) = ^{A} ^{o} ^{(}^{s}^{)}
˜
P(s)
E(s)
^{;}
˜
¯
P(s) = A _{o} (s) P(s)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 7/36
Afﬁne Parameterisation  Open Loop Unstable Model
As we need unstable poles of G _{o} (s) to be zeros in Q(s), we can write
¯
Q(s) = ^{A} ^{o} ^{(}^{s}^{)}
˜
P(s)
E(s)
^{;}
˜
¯
P(s) = A _{o} (s) P(s)
Then Q(s) is stable and has zeros to cancel unstable poles of G _{o} (s).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 7/36
Afﬁne Parameterisation  Open Loop Unstable Model
What about S _{i}_{o} (s) = (1 − Q(s)G _{o} (s))G _{o} (s)?
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 8/36
Afﬁne Parameterisation  Open Loop Unstable Model
What about S _{i}_{o} (s) = (1 − Q(s)G _{o} (s))G _{o} (s)?
We require the unstable poles of G _{o} (s) to be zeros in S _{o} (s).
The University of Newcastle
1 − Q(s)G _{o} (s) = 1 −
= 1 −
˜
P(s)B _{o} (s)
˜
E(s)A _{o} (s) P(s)B ¯ _{o} (s)
˜
E(s) E(s) − P(s)B _{o} (s)
˜
¯
=
˜
E(s)
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 8/36
Afﬁne Parameterisation  Open Loop Unstable Model
What about S _{i}_{o} (s) = (1 − Q(s)G _{o} (s))G _{o} (s)?
We require the unstable poles of G _{o} (s) to be zeros in S _{o} (s).
1 − Q(s)G _{o} (s) = 1 −
= 1 −
˜
P(s)B _{o} (s)
˜
E(s)A _{o} (s) P(s)B ¯ _{o} (s)
˜
E(s) E(s) − P(s)B _{o} (s)
˜
¯
=
˜
E(s)
We thus require A _{o} (s) to be a factor of
˜
¯
E(s) − P(s)B _{o} (s).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 8/36
Afﬁne Parameterisation  Open Loop Unstable Model
Which we can write as
˜
¯
¯
E(s) − P(s)B _{o} (s) = L(s)A _{o} (s)
or
¯
¯
L(s)A _{o} (s) + P(s)B _{o} (s) =
˜
E(s)
(1)
This is a standard pole assignment problem and choosing a desired
˜
E(s) and the orders of
The University of Newcastle
¯
L(s),
¯
P(s) will result in a unique solution.
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 9/36
Afﬁne Parameterisation  Open Loop Unstable Model
Which we can write as
˜
¯
¯
E(s) − P(s)B _{o} (s) = L(s)A _{o} (s)
or
¯
¯
L(s)A _{o} (s) + P(s)B _{o} (s) =
˜
E(s)
(2)
This is a standard pole assignment problem and choosing a desired
˜
E(s) and the orders of
¯
L(s),
¯
P(s) will result in a unique solution.
Note: We set
˜
E(s) = E(s)F(s).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 9/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now we have
then
The University of Newcastle
C(s) =
=
Q(s) =
˜
P(s)
˜
E(s)
^{Q}^{(}^{s}^{)}
1 − Q(s)G _{o} (s) P(s)A _{o} (s)
˜
˜
˜
E(s)A _{o} (s) − P(s)B _{o} (s) P(s) ¯
=
¯
L(s)
(3)
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 10/36
Afﬁne Parameterisation  Open Loop Unstable Model
The method to design a controller for the unstable open loop case outlined above is then:
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 11/36
Afﬁne Parameterisation  Open Loop Unstable Model
The method to design a controller for the unstable open loop case outlined above is then:
˜
Step 1 Choose _{E}_{(}_{s}_{)}_{.} (closed loop poles)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 11/36
Afﬁne Parameterisation  Open Loop Unstable Model
The method to design a controller for the unstable open loop case outlined above is then:
˜
Step 1 Choose _{E}_{(}_{s}_{)}_{.} (closed loop poles)
˜
Step 2 Given A _{o} (s), B _{o} (s) & E(s) solve
¯
¯
¯
¯
˜
L(s)A _{o} (s) + P(s)B _{o} (s) = E(s)
for a unique L(s) and _{P}_{(}_{s}_{)} that we denote as L (s) and P(s).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 11/36
Afﬁne Parameterisation  Open Loop Unstable Model
The method to design a controller for the unstable open loop case outlined above is then:
˜
Step 1 Choose _{E}_{(}_{s}_{)}_{.} (closed loop poles)
˜
Step 2 Given A _{o} (s), B _{o} (s) & E(s) solve
¯
¯
¯
¯
˜
L(s)A _{o} (s) + P(s)B _{o} (s) = E(s)
for a unique L(s) and _{P}_{(}_{s}_{)} that we denote as L (s) and P(s).
Step 3
The University of Newcastle
C(s) = ^{P}^{(}^{s}^{)} L(s)
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 11/36
Afﬁne Parameterisation  Open Loop Unstable Model
Given a solution to equation (1), standard results in algebra state that any other solution can be expressed as
¯
L(s)
E(s) ^{=}
L(s)
E(s) ^{−} ^{Q} ^{u} ^{(}^{s}^{)} B o (s)
E(s)
¯
P(s)
E(s)
= P(s)
E(s)
+
Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)} E(s)
(4)
(5)
where Q _{u} (s) is any stable proper transfer function having no undesirable poles.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 12/36
Afﬁne Parameterisation  Open Loop Unstable Model
Given a solution to equation (1), standard results in algebra state that any other solution can be expressed as
¯
L(s)
E(s) ^{=}
L(s)
E(s) ^{−} ^{Q} ^{u} ^{(}^{s}^{)} B o (s)
E(s)
¯
P(s)
E(s)
= P(s)
E(s)
+
Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)} E(s)
(6)
(7)
where Q _{u} (s) is any stable proper transfer function having no undesirable poles.
Substitute (4) and (5) into (3) and we get
C(s) =
P(s)
E(s)
+
Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)} E(s)
L(s)
E(s) ^{−} ^{Q} ^{u} ^{(}^{s}^{)} B o (s)
E(s)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 12/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now say A _{o} (s) contains both desirable and undesirable poles,
A _{o} (s) = A _{d} (s)A _{u} (s)
¯
then we can write E(s) = A _{d} (s) _{E}_{(}_{s}_{)} giving the pole assignment problem of
¯
¯
¯
A _{d} (s)A _{u} (s) L(s) + B _{o} (s) P(s) = A _{d} (s) E(s)F(s)
clearly this requires
The University of Newcastle
¯ ˜
P(s) = P(s)A _{d} (s).
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 13/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now say A _{o} (s) contains both desirable and undesirable poles,
A _{o} (s) = A _{d} (s)A _{u} (s)
¯
then we can write E(s) = A _{d} (s) _{E}_{(}_{s}_{)} giving the pole assignment problem of
¯
¯
¯
A _{d} (s)A _{u} (s) L(s) + B _{o} (s) P(s) = A _{d} (s) E(s)F(s)
clearly this requires
¯ ˜
P(s) =
P(s)A _{d} (s).
Therefore we have cancellations hence,
¯
˜
¯
A _{u} (s) L(s) + B _{o} (s) P(s) = E(s)F(s)
_{(}_{9}_{)}
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 13/36
Afﬁne Parameterisation  Open Loop Unstable Model
Now say A _{o} (s) contains both desirable and undesirable poles,
A _{o} (s) = A _{d} (s)A _{u} (s)
¯
then we can write E(s) = A _{d} (s) _{E}_{(}_{s}_{)} giving the pole assignment problem of
¯
¯
¯
A _{d} (s)A _{u} (s) L(s) + B _{o} (s) P(s) = A _{d} (s) E(s)F(s)
clearly this requires
¯ ˜
P(s) = P(s)A _{d} (s).
Therefore we have cancellations hence,
¯
˜
¯
A _{u} (s) L(s) + B _{o} (s) P(s) = E(s)F(s)
_{(}_{1}_{0}_{)}
For the unstable open loop case where the plant possesses some desirable poles then the same method as stated above applies except the pole assignment problem becomes equation (8).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 13/36
Afﬁne Parameterisation  Open Loop Unstable Model
The nominal complementary sensitivity for the class of stabilising controllers is,
T _{o} (s) =
=
=
=
¯
^{B} ^{o} ^{(}^{s}^{)} P(s)
¯
¯
A _{o} (s) L(s) + B _{o} (s) P(s)
B _{o} (s) ^{P}^{(}^{s}^{)} + Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)}
E(s)
E(s)
A _{o} (s)
L(s)
E(s) ^{−} ^{Q} ^{u} ^{(}^{s}^{)} B o (s)
E(s)
+ B _{o} (s) ^{P}^{(}^{s}^{)} + Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)}
E(s)
E(s)
B _{o} (s)P(s) + Q _{u} (s)B _{o} (s)A _{o} (s)
E(s)
E(s)
A _{o} (s)L(s) + B _{o} (s)P(s)
E(s)
E(s)
B _{o} (s)P(s) + Q _{u} (s)B _{o} (s)A _{o} (s)
E(s)F(s)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 14/36
Afﬁne Parameterisation  Open Loop Unstable Model
Given the controller parameterisation for unstable plants as
C(s) =
P(s) 
Q _{u} (s) ^{A} ^{o} ^{(}^{s}^{)} E(s) 
+ 

E(s) 

L(s) 
^{Q} ^{u} ^{(}^{s}^{)} B o (s) E(s) 
E(s) ^{−} 
and that u = Ce where e is the error signal we can write (note the argument s has been dropped)
The University of Newcastle
u = Ce
P
A
o
E
=
+ Q _{u}
E
L
E − Q _{u}
B
o
E
e
= ^{E}
L
^{} P
E
e + Q _{u}
A
o
E
e + Q _{u}
B
o
E
u
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 15/36
Afﬁne Parameterisation  Open Loop Unstable Model
Afﬁne Parameterisation: Unstable Open Loop Case
Prestabilising Controller
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 16/36
Afﬁne Parameterisation  Open Loop Unstable Model
The parametrisation as discussed above leads to the following parameterised version of the nominal sensitivities:
S o (s) = A _{o} (s)L(s) − Q u (s) B _{o} (s)A _{o} (s)
E(s)F(s)
E(s)F(s)
T o (s) = B _{o} (s)P(s) + Q u (s) B _{o} (s)A _{o} (s)
E(s)F(s)
E(s)F(s)
S io (s) = B _{o} (s)L(s)
E(s)F(s)
S uo (s) = A _{o} (s)P(s)
E(s)F(s)
− Q _{u} (s)
+ Q _{u} (s)
(B _{o} (s)) ^{2}
E(s)F(s)
(A _{o} (s)) ^{2}
E(s)F(s)
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 17/36
Saturation, Slew Rate Limitations and Anti windup Schemes
What is Saturation?
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 18/36
Saturation, Slew Rate Limitations and Anti windup Schemes
What is Saturation?
FACT: All actuators saturate!
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 18/36
Saturation, Slew Rate Limitations and Anti windup Schemes
What is Saturation?
FACT: All actuators saturate!
Objective: How do we deal with the nonlinear affects of Saturation and Slew Rate Limitation? {u(t) and u˙ _{(}_{t}_{)} are limited}
Control System with Saturation on the Plant Input
^{d} o

The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 18/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Saturation
u = sat{uˆ } =
u min
uˆ
u max
if uˆ
if u _{m}_{i}_{n} ≤ uˆ
if uˆ
< u min ,
≤ u _{m}_{a}_{x} ,
> u max .
^{d} o

The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 19/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Ignoring the presence of saturations can cause long undesirable transients in the closed loop.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 20/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Ignoring the presence of saturations can cause long undesirable transients in the closed loop.
The transients are due to the controller states having ‘wound up’ to large values.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 20/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Ignoring the presence of saturations can cause long undesirable transients in the closed loop.
The transients are due to the controller states having ‘wound up’ to large values.
In a PID controller, there is only one state that is subject to windup  the integrator state!
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 20/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Ignoring the presence of saturations can cause long undesirable transients in the closed loop.
The transients are due to the controller states having ‘wound up’ to large values.
In a PID controller, there is only one state that is subject to windup  the integrator state!
Therefore in a PID controller, an antiwindup scheme involves limiting the integrator state in the some way.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 20/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Antiwindup scheme for a PI controller
e
^{K}
p
+
When the controller is in the linear region of the saturation
uˆ
_{e}
= K _{p} (1 + ^{K} ^{i} ).
s
Thus the state of the controller is updated only with the actual plant input uˆ .
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 21/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The key properties of an antiwindup scheme are:
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 22/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The key properties of an antiwindup scheme are:
The state of the controller should be driven by the actual (i.e. constrained) plant input.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 22/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The key properties of an antiwindup scheme are:
The state of the controller should be driven by the actual (i.e. constrained) plant input.
The states of the controller should have a stable realisation when driven by the actual plant input.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 22/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The Problem of Windup in IMC
We consider here open loop stable plants.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 23/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The Problem of Windup in IMC
We consider here open loop stable plants.
The block diagram shows an IMC structure where saturation is included on the input to the plant. The initial states of the system are depicted as x(t _{0} ) and xˆ (t _{0} ). We also assume G _{o} (s) = G(s).
Internal Model Control with Saturation on Plant Input
^{d} o
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 23/36
Saturation, Slew Rate Limitations and Anti windup Schemes
When uˆ is in the linear region of the saturation,
y = Q(s)G _{o} (s)r + (1 − Q(s)G _{o} (s))d _{o} + IC(x − xˆ )
where IC is a transfer function relating the initial conditions to the output.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 24/36
Saturation, Slew Rate Limitations and Anti windup Schemes
When uˆ is in the linear region of the saturation,
y = Q(s)G _{o} (s)r + (1 − Q(s)G _{o} (s))d _{o} + IC(x − xˆ )
where IC is a transfer function relating the initial conditions to the output.
Once the initial transient has decayed we see that the output is what we would expect, i.e. only a function of the reference and disturbance.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 24/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} ) the loop is nonlinear
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )
the loop is nonlinear
G _{o} is driven by uˆ
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} ) the loop is nonlinear
G _{o} is driven by uˆ G is driven by u
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} ) the loop is nonlinear
G _{o} is driven by uˆ G is driven by u
Now, if saturation ends at time t _{0} ,
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} ) the loop is nonlinear
G _{o} is driven by uˆ G is driven by u
Now, if saturation ends at time t _{0} , the loop is once again linear
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )

the loop is nonlinear 


G _{o} is driven by uˆ 


G 
is driven by u 
Now, if saturation ends at time t _{0} ,

the loop is once again linear 


G 
and G _{o} are both driven by uˆ 
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )

the loop is nonlinear 


G _{o} is driven by uˆ 


G 
is driven by u 
Now, if saturation ends at time t _{0} ,

the loop is once again linear 


G 
and G _{o} are both driven by uˆ 

But x(t _{0} _{)} _{} _{x}_{ˆ} (t _{0} _{)} since we had _{u}_{ˆ} u 
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )

the loop is nonlinear 


G _{o} is driven by uˆ 


G 
is driven by u 
Now, if saturation ends at time t _{0} ,

the loop is once again linear 


G 
and G _{o} are both driven by uˆ 

But x(t _{0} _{)} _{} _{x}_{ˆ} (t _{0} _{)} since we had _{u}_{ˆ} u 
Hence we will again see a transient in the output.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
During saturation (i.e. uˆ < u _{m}_{i}_{n} or uˆ > u _{m}_{a}_{x} )

the loop is nonlinear 


G _{o} is driven by uˆ 


G 
is driven by u 
Now, if saturation ends at time t _{0} ,

the loop is once again linear 


G 
and G _{o} are both driven by uˆ 

But x(t _{0} _{)} _{} _{x}_{ˆ} (t _{0} _{)} since we had _{u}_{ˆ} u 
Hence we will again see a transient in the output.
This transient is due to the mismatched states in the plant and the controller, i.e. x(t _{0} ) xˆ (t _{0} ), and is called windup.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 25/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Antiwindup for IMC
Strategies aimed at reducing the effects of windup are, of course, called antiwindup.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 26/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Antiwindup for IMC
Strategies aimed at reducing the effects of windup are, of course, called antiwindup.
An initial idea for antiwindup  Saturate the model input as well!
^{d} o
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 26/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Antiwindup for IMC
Strategies aimed at reducing the effects of windup are, of course, called antiwindup.
An initial idea for antiwindup  Saturate the model input as well!
^{d} o
Now x(t _{0} ) = xˆ (t _{0} ) during and after saturation.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 26/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The problems with this are:
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 27/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The problems with this are:
1. The total system from uˆ to y is nonlinear, so a linear compensator Q is no longer an inverse of the process.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 27/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The problems with this are:
1. The total system from uˆ to y is nonlinear, so a linear compensator Q is no longer an inverse of the process.
2. e and uˆ are completely independent of the saturation.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 27/36
Saturation, Slew Rate Limitations and Anti windup Schemes
The problems with this are:
1. The total system from uˆ to y is nonlinear, so a linear compensator Q is no longer an inverse of the process.
2. e and uˆ are completely independent of the saturation.
For further insight into this problem, we digress slightly, and examine a feedback realisation of _{Q}_{(}_{s}_{)}_{.}
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 27/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Assume, for simplicity, that G _{o} (s) is minimum phase. Then
G
_{o} = G
o
i
−1
i
and Q(s) = F _{q} (s)G _{o} (s).
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 28/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Assume, for simplicity, that G _{o} (s) is minimum phase. Then
G
_{o} = G
o
i
−1
i
and Q(s) = F _{q} (s)G _{o} (s).
Lets also assume here that F _{q} (s) is chosen to make Q(s) biproper
Q(s) = n _{n} s ^{n} +·+ n _{1} s + n _{0}
m _{n} s ^{n} +·+ m _{1} s + m _{0}
where n _{n} 0 and m _{n} 0.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 28/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Assume, for simplicity, that G _{o} (s) is minimum phase. Then
G
_{o} = G
o
i
−1
i
and Q(s) = F _{q} (s)G _{o} (s).
Lets also assume here that F _{q} (s) is chosen to make Q(s) biproper
Q(s) = n _{n} s ^{n} +·+ n _{1} s + n _{0}
m _{n} s ^{n} +·+ m _{1} s + m _{0}
where n _{n} 0 and m _{n} 0.
The high frequency gain of Q(s) is
q _{∞} =
_{s}_{→}_{∞} Q(s) = ^{n} ^{n}
lim
m
n
0.
The University of Newcastle
Lecture 6: Afﬁne Parameterisation  Open Loop Unstable Model. Antiwindup Schemes – p. 28/36
Saturation, Slew Rate Limitations and Anti windup Schemes
Assume, for simplicity, that G _{o} (
Lebih dari sekadar dokumen.
Temukan segala yang ditawarkan Scribd, termasuk buku dan buku audio dari penerbitpenerbit terkemuka.
Batalkan kapan saja.