Control Systems

Part I:
Subject overview
An introduc:on to feedback

Lecture notes for ELEN90055 prepared by Michael Cantoni (c) 2011, 2012
 Welcome to ELEN90055 Control Systems ...

We will loosely follow the textbook:

G.C. Goodwin, S.F. Graebe, M.E. Salgado, “Control System Design.” Pren:ce Hall, 2001.
Our aim is to cover aspects of Chapters 2 to 10 (and 11 :me permiRng).
A number of holdings in the library.
The following books are just as good (and there are many others):
L. Qiu and K. Zhou, “Introduc:on to Feedback Control.” Pren:ce Hall, 2010.
K. Ogata, “Modern Control Engineering.” Pren:ce Hall, 5th Edi:on, 2010.
R.C. Dorf and R.H. Bishop, “Modern Control Systems.” Pren:ce Hall, 12th Edi:on, 2010.
K. Astrom and R. Murray, “Feedback Systems: An Introduc:on for Scien:sts and
Engineers.” Princeton University Press, 2008. Free electronic copy legally available at
h^p://www.cds.caltech.edu/~murray/amwiki/Main_Page .
For the mathema:cally oriented: J.C. Doyle, B.A. Francis and A.R. Tannenbaum,
“Feedback Control Theory.” Macmillan, 1992. Free electronic copy legally available at
h^p://www.control.utoronto.ca/people/profs/francis/dd.html .

M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 2

 Welcome to ELEN90055 Control Systems ...
All handouts will be on Canvas LMS - slides, workshop
manuals, problems sets - and staff and subject
information

http://www.segwaysoutherncross.com/

Three lectures per week:

-Tues and Wed 8-9am(Medical - Sunderland)
-Thurs 4.15-5.15pm (Medical - Sunderland)
- Examples class every 2nd Thurs, roughly

Workshops (2 hours/week) star:ng Fri Mar 13

3 MATLAB-based simula:on workshops; 1 workshop on
posi:on control design for a LEGO robot (over 3 weeks
with report worth 10%); 1 workshop on control design
for a LEGO robot with inverted pendulum (over 3 weeks
with report worth 10% due at end of semester)
Workshops to be completed in self-assigned groups of 3
One test on Tue 28 April 8-9am, Wilson Hall = 10%
One final wri^en exam (3hrs; hurdle) = 70%
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 3
 What is control about?
Making DYNAMICAL SYSTEMS exhibit desired behaviour:
a passenger car travel at constant speed
a collec:on of vehicles operate in a co-ordinated fashion
an internal combus:on engine achieve low emissions
a cuRng tool / robot arm follow a given trajectory
h^p://youtu.be/hyGJBV1xnJI
a jet engine run quietly h^p://www.flyingmachinearena.org/videos/

an electro-mechanical generator deliver power at constant voltage / frequency

an electronic amplifier achieve a gain characteris:c
a human body achieve specific hormone levels
a species of cell grow in culture at a specified rate
a data network achieve quality-of-service targets
an economy move towards specified infla:on figures
This must be achieved in the face of UNCERTAINTY about:
h^p://datainterfaces.com/FRM220-AC.aspx
the dynamics of components in the system
the opera:ng environment - load, interference, disturbances
FEEDBACK plays an important role ... most of this subject is about FEEDBACK SYSTEMS
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 4
 Examples of feedback ...
qi (m3 /sec)
schemaLcs
water-level (reflect physical
signal organisaLon)
flow
signals
x (m) qo (m3 /sec)

Ktesibios’ float-valve regulator - water clock circa 250 BC

Want constant flow into large vessel (where indicator floats)
Uncertain supply pressure h^p://www.ens-lyon.fr/RELIE/Cadrans/Musee/Pages/PagesGr/MuKtesibiosGr.htm

ouvlow (sub-) system block diagram (abstracLon)

_ qo signals carry informa:on
between
output

float output
input

chamber orifice system blocks

+ net flow q
that relate output signals
inflow and input signals
qi x water-level
float and 1 t
x(t) = x(0) + q( ) d
valve supply pressure A 0
(‘nominal’ + ‘disturbance’) float chamber cross-sec:on
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 5
 Examples of feedback ...
qi (m3 /sec)
schemaLcs
water-level (reflect physical
signal Points to note from the water-clock
organisaLon) example …
flow
signals of components
‘Schema(cs’ reflect the physical organiza:on
x (m) qo (m3 /sec)
‘Block diagrams’
Ktesibios’ float-valveare modelling
regulator abstrac1ons
- water expressed
clock circa 250 BC in terms of ‘signals’ and
(sub-)‘systems’
Want constant flow into larger vessel where indicator floats
Uncertain supply pressure
‘Signals’ carry quan:ta:ve informa:on (e.g. flow, pressure) as a func:on of :me
ouvlow (sub-) system
qo
‘Systems’ map ‘input’
floatsignals to ‘output’ signals,
output oden with dynamics (i.e. some
output
input

memory
net chamber
flow qof the past like the integral map),orifice
to represent cause and effect

inflow
q‘Disturbances’ x water-level
i are
floatexogenous
and input signals which are uncertain (i.e. not known)
1 t
x(t) = x(0) + q( ) d
valve supply pressure A 0
(a ‘disturbance’)
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 6
 Examples of feedback ...
steam engine 1790s
fly-balls rise schemaLc
Wa^’s governor schemaLc with increase in
speed, which
linkage closes valve to
bu^erfly
valve reduce torque
boiler fly-balls fall
side piston with decrease in
side speed, which
opens valve to
pulley to engine
h^p://www-g.eng.cam.ac.uk/125/1950-1975/control_governor.html
increase torque h^p://www-g.eng.cam.ac.uk/125/1950-1975/control_wa^engine.html

load torque an integrator

steam _ engine iner:a again!!!
engine &
output

pressure +
input

d⇥ 1
thro^le =
(‘nominal’ + engine net dt J
input
‘disturbance’) torque torque engine speed
linkage pulley
gain gain
fly-ball
‘signals’ carry informa:on as dynamics
‘systems’ map input signals (causes)
a func1on of 1me block diagram to output signals (effects)
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 7
 Examples of feedback ...
fly-balls rise
PointsschemaLc
Wa^’s governor to note from the steam
with inc. inengine example …
speed, which
linkage
‘Equilibrium’ is reached when ALL of themakes
signals in the block-diagram remain constant
bu^erfly valve
over :me valve more closed
boiler
fly-balls fall
The uncertain loadside
torque is matched
piston bywiththe engine torque if an equilibrium is reached,
dec. in
because the input of an integrator
side is necessarily zero at equilibrium
speed, which
makes valve
pulley to engine
will an equilibrium be reached? ... thismore is a ques:on of ‘stability’ ... schemaLc
h^p://www-g.eng.cam.ac.uk/125/1950-1975/control_governor.html open h^p://www-g.eng.cam.ac.uk/125/1950-1975/control_wa^engine.html
load torque
Sta:c/constant load
block diagram or steam-pressure perturba:ons from the nominal values lead to
(a ‘disturbance’) an integrator
off-sets in engine speed at equilibrium, because the single loop integrator has memory
again!!!
of the net torque ‘transient’ _ engine iner:a
engine &
output

steam +
input

d⇥ 1
thro^le =
pressure netlarge load/pressure
enginesmall for dt J
gain in the loop can make speed off-set
input off-setsengine speed
torque torque (want this to
how much gain is required to make the off-set less than apulley
linkage specifica:on? be constant)
gain gain
fly-ball
does making the gain too big lead to problems?
dynamics

M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 8

 Examples of feedback ...
block diagram
back matched received
source termination SUBSYS1
clock distribution manager clock at 1 (sum acc.) VD voltage
lossless tx-line delay τ1
delay compensator
integrator controlled
CLK CLKOUT

τM VS ‘uncertain’ with gain delay

delay compensator
lossless tx-line delay τ2 SUBSYS2 phase error (want = 0) D
delays
system

transmission line
phase detector
clock
CLK CLKOUT

desired
ref + +
τM
clock VS

source delay compensator

lossless tx-line delay τ3 receiver phase _ +
CLK CLKOUT
(i.e.desired delay)
SUBSYS3
desired
τM VS
delay

⇥ = ⇥D +
schemaLcs phase of received clock
VOLTAGE CONTROLLED DELAY (VCD) rela:ve to clock source
CLKOUT
τD
SYS CLK
CLK
CLKM
D integrator
R C
VD RX CLK
VCD _
τM R
+

OPAMP
signals in the block diagram above
-1 BACK-MATCHED
VM U
SOURCE
V0 TERMINATION
model properLes of physical signals
V0 P

0 0

at only discrete instances of (mes

Q 0 LOSSLESS
COMPL COMPH PHASE DETECTOR TX-LINE
+
_

+ -V0
_

(rising edges of CLK at source)

WITH
VL VH DELAY τ
VS

DELAY COMPENSATOR

block diagram evolves in discrete (me

M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 9
 Examples of feedback ...
received
back matched clock at 1

Points to note from the clock robot example …

source termination SUBSYS1
clock distribution manager
lossless tx-line delay τ1
delay compensator

Abstrac:on CLK yields

CLKOUT an addi1ve disturbance model of the uncertain component

τ M V S
‘uncertain’
system
Zero ‘error’ between a ‘measurement’
delay compensator
delays
lossless tx-line of a signal and a constant ‘reference’ that enters
delay τ SUBSYS2 2

like a disturbance is achieved at equilibrium if there is an integrator in the loop

clock
CLK CLKOUT

τM VS
For the clock management
delay compensator
feedback
lossless tx-line delay τ system, increasing the loop gain can lead to
3

divergenceCLK of the error and in this case

CLKOUT the system does not reach equilibrium
SUBSYS3
desired
τM VS
delay
ek vk +
ref sk+1 = sk + ek +
Avcd
+ schemaLcsvk = Apd · sk
_
k
SYS CLK
e = ⌧ A · sk+1
. sk+1 = integrator
(1 A)sk RX k+1
CLK ref
A = Avcd · Apd = sk+1 (1 A)sk A · sk+1
<latexit sha1_base64="pPH9jgCo/7A8c96txzTd0BEc1TM=">AAACFnicbZDLSgMxFIYz9VbrrerSTbAIbiwzIthlixuXFewFOmXIpKdtaOZicqZYhj6FG1/FjQtF3Io738b0AmrrD4Gf75yT5Px+LIVG2/6yMiura+sb2c3c1vbO7l5+/6Cuo0RxqPFIRqrpMw1ShFBDgRKasQIW+BIa/uBqUm8MQWkRhbc4iqEdsF4ouoIzNMjLn1Wo24kQ7mjFS12Ee0yHvDMeU5cbTH9obGDOyxfsoj0VXTbO3BTIXFUv/2lu50kAIXLJtG45doztlCkUXMI45yYaYsYHrActY0MWgG6n07XG9MSQDu1GypwQ6ZT+nkhZoPUo8E1nwLCvF2sT+F+tlWC31E5FGCcIIZ891E0kxYhOMqIdoYCjHBnDuBLmr5T3mWIcTZKTEJzFlZdN/bzoGH9zUSiX5nFkyRE5JqfEIZekTK5JldQIJw/kibyQV+vRerberPdZa8aazxySP7I+vgEEzJ9A</latexit>

+ (⇥ref ) signals in the BD above model

= (1 A)(s
proper1es sksignals
of physical
k+1 ) = (1 at A)ek
discrete :mes (specifically, the
en = (1 A)n e0 and so we need |1 A| < 1 for the error to converge
rising edges of clock at source)
if |1 A| is close to 1 then error transients last for BD evolves
a long :me discretely
... in :me
this is an addi1onal ABSTRACTION
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 10
 Examples of feedback ...
op-amp
series two-port
comparator model
+ schemaLcs
+ +
_ + +
+
+ V_s Vi A·V
_ i Vo
V_s _ Vf + _ _ Rl
Vo Rl load
R1
R2
_ R1 feedback
R2 network

ignores all non-ideali:es except the large

block diagram but uncertain finite-gain
Vs + Vi op-amp Vo Vi = V s Vf ⇤
gain
_ Vo = A · Vi Vo = A · (Vs · Vo )
A ⇥
V f = · Vo

Vf feedback ⇥
gain A for large A 1
Vo = · Vs · Vs
1+ ·A
overall gain is only a func(on of the feedback network if the
= R2 /(R1 + R2 ) op-amp gain is large ... overall gain is ROBUST to variability in A!
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 11
 Examples of feedback ...
posi:on sensi:ve laser
photodiode (PD) Atomic force microscopy (AFM) uses
can:lever feedback for instrumenta1on
Goal: make the ‘z’ control force
sample ‘balance’ the force between the sample
‘x,y’ sweep piezo http://www.nanolab.polimi.it/
and the sharp can:lever :p in order to
command drive achieve a reference deflec:on
‘z’
command
Recording this force as the sample is
signal computer power progressively scanned in the ‘x,y’ plane,
amplifier control amplifier under the can:lever, yields an image
schemaLc The transient behaviour of the ‘z’
computer actuator control loop limits the ‘x,y’ scan rate
http://www.jpk.com/dna-molecule-on-mica.315.en.html
+ power
control

‘z’ control force

_ D/A amp. +
algorithm
deflec:on
reference

piezo drive
AFM
image
PD + signal can:lever of DNA
A/D
amplifier dynamics
atomic forces
block diagram sensor (a ‘disturbance’)
= can:lever deflec:on
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 12
 General structure of feedback control systems ...
noise load disturbances

desired system plant

outputs controller actuators (dynamics
to control) actual system
outputs

sensors
measured system
outputs
noise
The following simpler block diagram abstrac:on is frequently used to model feedback
control systems for the purpose of performance analysis and design
di do
e + +
r + u + + y
feedback plant
_ compensator model
+
+ n

Sensor noise n enters the loop in same way as the reference r!!!
how can the system output follow the reference AND be insensi1ve to sensor noise?
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 13
 What is the alternaLve to feedback control?
d

_
feedforward +
r + u plant + y
controller
f 1⇥ ⇤ f⇥ ⇤

f h i denotes a causal mapping between signal spaces and f 1

h i its inverse
WE WANT r = y = f hui + d (= u=f 1
hr di
The problem of control can be thought of in terms of inversion as shown above ... but
this scheme requires
the plant model to be exact and the inverse to be realized exactly (not robust)
suppose there is a delay in the plant ... how do you invert a delay?
this requires a view into the future!??! ... this is not causal/physical!!!
the model and its inverse must be such that bounded-inputs yield bounded-outputs
the disturbance must be measurable to compute the control signal
The feedforward scheme is asking for a lot ... exact tracking of reference
These generally unachievable requirements can be relaxed by looking for approximate
inverses and achieving desired behaviour asympto1cally (i.e. in the limit of :me passing)
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 14
 Approximate inversion by high-gain feedback ....

r + high gain u y
plant
_ h⇥ ⇤

ŷ plant model
f⇥ ⇤

u = h⌅r f ⌅u⇧⇧ =⇤ r f ⌅u⇧ = h 1

⌅u⇧
for high-gain h
=⇤ u=f 1
⌅r h 1
⌅u⇧⇧ ⇥ u=f 1
⌅r⇧ =⇤ y⇥r
When the plant model is exact and no disturbances act on the system, this approximate
inversion scheme ( open-loop control ) is equivalent to the feedback loop below

r + e u y
high-gain
_ plant
feedback

The two approaches differ in terms of the effects of disturbances and modelling errors
feedback can provide some robustness to such uncertainty
but beware ... high loop-gain can lead to instability and sensor noise problems
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 15
 Important observaLons so far ...
The idea of feedback finds very broad applica:on
Feedback involves sensing and actua:on for control
Feedback can de-sensi1ze a system to uncertainty in the
components and the opera:ng environment
Integral ac:on and/or large loop-gain can result in good
equilibrium behaviour:
rejec:on of unknown disturbances
reference regula:on
http://www.rubicon.com.au/

Feedback can also lead to bad behaviour if poorly designed:

long transients
instability with divergence of signals
sensi:vity to sensor noise
To properly account for the merits and limita:ons of
feedback control in the design of a system we need an
appropriate set of mathema1cal tools
for modelling & formula:ng specifica:ons; analyzing
performance; and controller synthesis Total Channel Control (TCC)
this subject focuses on some ‘classical’ tools Rubicon Water + UniMelb
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 16
 By the end you should be able to ...
Apply fundamental tools for studying of linear :me-invariant dynamical systems

Topics include: differen:al equa:on models; step response; convolu:on models; transfer
func:ons; :me-domain interpreta:on of poles and zeros; frequency response; Bode plots

Chapters 3 and 4 of Goodwin, Graebe and Salgado (GGS) and Part II of the notes

Most of this should be ‘review’ of material already seen in previous subjects

Assess the stability and performance of linear :me-invariant feedback control systems in terms of
the open-loop characteris1cs

Topics include: closed-loop sensi:vi:es; closed-loop stability; root-locus plots; Nyquist plots;
rela:ve stability; robustness to model uncertainty; fundamental limita:ons; the internal model
principle; feedforward compensa:on

Chapters 5, 8 and 10 of GGS and Part III of the notes

Design feedback controllers to meet stability and performance specifica:ons

Topics include: propor:onal (P), integral (I), lag and lead (D) compensa:on; PID control and
empirical tuning; classical loop-shaping; polynomial approaches to pole-placement

Chapters 6 and 7 of GGS and Part IV of the notes

Apply techniques for dealing with actuator constraints (:me permiRng - unlikely) - Chapter 11 GGS
M.Cantoni (c) 2011, 2012 ELEN90055 Control Systems: Part I 17