Lecture: Anti-windup techniques

Automatic Control 2
Anti-windup techniques
Prof. Alberto Bemporad
University of Trento
Academic year 2010-2011
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 1 / 17
Lecture: Anti-windup techniques Saturation effects
Problem
G(z)
x(k)
r(k)
Kx(k)+Hr(k)
linear
controller
u(k)
saturation
Wednesday, June 2, 2010
Most control systems are designed based on linear theory
A linear controller is simple to implement and performance is good, as long
as dynamics remain close to linear
Nonlinear effects require care, such as actuator saturation (always present)
Saturation phenomena, if neglected in the design phase, can lead to
closed-loop instability, especially if the process is open-loop unstable
Main reason: the control loop gets broken if saturation is not taken into
account by the controller: u(k) = Kx(k) for some k
[1] K.J. strom, L. Rundqwist, Integrator windup and how to avoid it, Proc. American Control
Conference, Vol. 2, pp. 1693-1698, 1989
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 2 / 17
Lecture: Anti-windup techniques Saturation effects
Example: AFTI-F16 aircraft
Linearized model:
_
_
_
x =
_
0.0151 60.5651 0 32.174
0.0001 1.3411 0.9929 0
0.00018 43.2541 0.86939 0
0 0 1 0
_
x +
_
2.516 13.136
0.1689 0.2514
17.251 1.5766
0 0
_
u
y =
_
0 1 0 0
0 0 0 1
_
x
Inputs: elevator and aperon
(=ap+aileron) angles
Outputs: attack and pitch angles
Sampling time: T
s
= 0.05 sec (+ ZOH)
Constraints: 25

on both angles
Open-loop response: unstable
open-loop poles:
7.6636, 0.0075 0.0556j, 5.4530
!"##$%
&'$()*+%,
-').,
/0'$%+1
/0'$%+1
2+130*"#01)'4)50,
6$%*07)'4)50,
2)*$%)'4)50,
!+''
80*79 :);
Wednesday, June 2, 2010
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 3 / 17
Lecture: Anti-windup techniques Saturation effects
Example: AFTI-F16 aircraft
LQR control + actuator saturation 25

The system is unstable !

Actuator saturation cannot be neglected in the design of a good controller
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 4 / 17
Lecture: Anti-windup techniques Saturation effects
Example: AFTI-F16 aircraft
With a controller designed to handle saturation constraints:
There are several techniques to handle input saturation, the most popular
ones are anti-windup techniques
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 5 / 17
Lecture: Anti-windup techniques Saturation effects
Saturation function
Saturation can be dened as the static nonlinearity
sat(u) =
_
_
_
u
min
if u < u
min
u if u
min
u u
max
u
max
if u > u
max
2 1 0 1 2
1
0.5
0
0.5
1
u
min
and u
max
are the minimum and maximum allowed actuation signals
(example: 12 V for a DC motor)
If u is a vector with m components, the saturation function is dened as the
saturation of all its components (u
min
, u
max

m
)
sat(u) =
_
_
_
_
_
_
sat(u
1
)
sat(u
2
)
.
.
.
sat(u
m
)
_
_
_
_
_
_
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 6 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
The windup problem
reference output
input
To Workspace
integ
Saturation Process
1
s
Integrator
1
s
Gain
1
Derivative
du/dt
very simple process (=an integrator)
controlled by a PID controller (K
P
= K
I
= K
D
= 1)
under saturation 0.1 u 0.1
The output takes a long time to go steady-state
The reason is the windup of the integrator contained in the PID controller,
which keeps integrating the tracking error even if the input is saturating
anti-windup schemes avoid such a windup effect
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 7 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
Anti-windup #1: incremental algorithm
y(kT)
r(KT)
u(kT)
incremental
PID
1
1 z
1
G(z)
u(kT)
Sunday, May 16, 2010
It only applies to PID control laws implemented in incremental form
u((k +1)T) = u(kT) + u(kT)
where
u(kT) = u(kT) u((k 1)T)
Integration is stopped if adding a new u(kT) causes a violation of the
saturation bound
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 8 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
Anti-windup #2: Back-calculation
G(s)
!
"
e
sT
d
K
p
!
v y r u
actuator
! !
!
PID
! !
1
s
1
T
t
! "
K
p
T
i
anti-windup
e
t
Wednesday, June 2, 2010
The anti-windup scheme has no effect when the actuator is not saturating
(e
t
(t) = 0)
The time constant T
t
determines how quickly the integrator of the PID
controller is reset
If the actual output u(t) of the actuator in not measurable, we can use a
mathematical model of the actuator. Example: e
t
(t) = v(t) sat(v(t))
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 9 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
Anti-windup #2: Back-calculation
reference output
input
To Workspace
integ_aw
Saturation Process
1
s
Integrator
1
s
Gain
1
Derivative
du/dt
anti-windup scheme for a PID controller (K
P
= K
I
= K
D
= 1, T
t
= 1)
Integrator windup is avoided thanks to back-calculation
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 10 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
Benets of the anti-windup scheme
Lets look at the difference between having and not having an anti-windup
scheme
0 10 20 30 40 50 60 70 80
0
0.5
1
1.5
2
o
u
t
p
u
t

y
(
t
)
0 10 20 30 40 50 60 70 80
0.1
0.05
0
0.05
0.1
i
n
p
u
t

u
(
t
)
0 10 20 30 40 50 60 70 80
5
0
5
i
n
t
e
g
r
a
l

t
e
r
m

i
(
t
)
time t
Note that in case of windup we have
strong output oscillations
a longer time to reach the
steady-state
the peaks of control signal
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 11 / 17
Lecture: Anti-windup techniques Anti-windup schemes: Ad hoc methods
Anti-windup #3: Conditional integration
The PID control law is
u(t) = K
p
(br(t) y(t)) +I(t) K
p
T
d
dy(t)
dt
= K
p
br(t) K
p
y
p
(t) +I(t)
where y
p
(t) = y(t) +T
d
dy(t)
dt
is the prediction of the output for time t +T
d
Consider the proportional band [y
l
(t), y
h
(t)] for y
p
(t) in which the
corresponding u is not saturating
y
l
(t) = br(t) +
I(t) u
max
K
p
y
h
(t) = br(t) +
I(t) u
min
K
p
where u
min
and u
max
are the saturation limits of the actuator
The idea of conditional integration is to update the integral term I(t) only
when y
p
is within the proportional band (for PI control simply set y
p
= y)
An hysteresis effect may be included to prevent chattering
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 12 / 17
Lecture: Anti-windup techniques Anti-windup schemes: General methods
Anti-windup #4: Observer approach
The anti-windup method applies to
dynamic compensators
_
_
_
x(k +1) = Ax(k) +Bu(k)
+L(y(k) Cx(k))
u(k) = Kx(k) +Hr(k)
by simply feeding the saturated
input to the observer
_

_
x(k +1) = Ax(k) +Bsat(v(k)))
+L(y(k) Cx(k))
v(k) = Kx(k) +Hr(k)
u(k) = sat(v(k))
x(k) y(k)
A,B
u(k)
C
.1$1-
-.1&%$1,+
!"#$%&'$()*+,'-..)/0).$12+$1&,#
r(k)
K
x(k) + +
H
saturation
v(k)
Wednesday, June 2, 2010
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 13 / 17
Lecture: Anti-windup techniques Anti-windup schemes: General methods
Example
The process to be controlled is obtained by exactly sampling
_
_
_
x =
_
1.364 0.4693 0.736 1.131
1.08 1.424 0.1945 0.7132
0.0499 0.8704 0.9675 0.3388
0.9333 0.8579 0.5436 0.9997
_
x +
_
0.05574
0
0.04123
1.128
_
u
y = [ 1.349 0 0.9535 0.1286 ] x
with T = 0.5s
The input u saturates between 2
Control design by pole placement: poles in e
5T
, e
5T
, e
(2j)T
Observer design by pole placement: 4 poles in e
10T
The dynamic compensator is
_
x(k +1) = Ax(k) +Bsat(v(k)) +L(y(k) Cx(k))
v(k) = Kx(k) +Hr(k)
with
K =
_
0.4718
1.5344
2.8253
2.1819
_

, L =
_
1.1821
1.1924
4.2054
3.6554
_
, H = 1/(C(I A BK)
1
B) = 8.2668
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 14 / 17
Lecture: Anti-windup techniques Anti-windup schemes: General methods
Example (contd)
Compare the results with anti-windup and without anti-windup:
0 5 10 15 20 25 30 35 40 45 50
1
0.5
0
0.5
1
1.5
o
u
t
p
u
t

y
(
t
)
0 5 10 15 20 25 30 35 40 45 50
2
1
0
1
2
i
n
p
u
t

u
(
t
)
time t
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 15 / 17
Lecture: Anti-windup techniques Anti-windup schemes: General methods
Conclusions
Conditional integration is easy to apply to many controllers, although it may
not be immediate to nd the conditions to block integration and to avoid
chattering
Back-calculation only requires tuning one parameter, the time constant T
t
.
But it only applies to PID control
The observer approach is very general and does not require tuning any
additional parameter. It also applies immediately to MIMO (multi-input
multi-output) systems
Saturation effects can be included in an optimal control formulation. In this
case the control action is decided by a constrained optimization algorithm, as
in model predictive control (MPC) techniques
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 16 / 17
Lecture: Anti-windup techniques Anti-windup schemes: General methods
English-Italian Vocabulary
anti-windup scheme schema di anti-windup
aileron alettone
pitch beccheggio
roll rollio
yaw imbardata
model predictive control controllo predittivo
Translation is obvious otherwise.
Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 2010-2011 17 / 17