You are on page 1of 17

1

DorfandBishop(11
th
Edition)PrenticeHall(PearsonAustralia)
FeedbackControlSystemCharacteristics
Chapter4,Section4.2
ErrorSignalAnalysis
TheclosedloopfeedbackcontrolsystemshowninthefollowingfigurehasthreeinputsR(s),T
d
(s),
andN(s)andoneoutput,Y(s).ThesignalsT
d
(s),andN(s)arethedisturbanceandmeasurement
noisesignals,respectively.

Definethetrackingerroras
E(s) = R(s) -(s)
Foreaseofdiscussion,wewillconsideraunityfeedbacksystem,thatis,H(s)=1,inthissystem.After
someblockdiagrammanipulation,wefindthattheoutputisgivenby
(s) =
0
c
(s)0(s)
1 +0
c
(s)0(s)
R(s) +
0(s)
1 +0
c
(s)0(s)
I
d
(s) -
0
c
(s)0(s)
1 +0
c
(s)0(s)
N(s)
Eqn.1
Therefore,withE(s) = R(s) - (s),wehave
E(s) =
1
1 +0
c
(s)0(s)
R(s) -
0(s)
1 +0
c
(s)0(s)
I
d
(s) +
0
c
(s)0(s)
1 +0
c
(s)0(s)
N(s)
Eqn.2
Definethefunction
I(s) = 0
c
(s)0(s)
Thefunction,L(s),isknownastheloopgainandplaysafundamentalroleincontrolsystemanalysis.
IntermsofL(s)thetrackingerrorisgivenby
2

E(s) =
1
1 +I(s)
R(s) -
0(s)
1 +I(s)
I
d
(s) +
I(s)
1 +I(s)
N(s)
WecandefinethefunctionF(s) = 1 + I(s)
Then,intermsofF(s),wedefinethesensitivityfunctionasS(s) =
1
P(s)
=
1
1+L(s)

Similarly,intermsoftheloopgain,wedefinethecomplementarysensitivityfunctionas
C(s) =
L(s)
1+L(s)

IntermsofthefunctionS(s)andC(s),wecanwritethetrackingerroras

E(s) = S(s)R(s) - S(s)0(s)I
d
(s) + C(s)N(s)
Examiningtheaboveequation,weseethat(foragivenG(s)),ifwewanttominimizethetracking
error,wewantbothS(s)andC(s)small.Obviously,designcompromisesmustbemade.

Toanalysethetrackingerrorequation,weneedtounderstandwhatitmeansforatransferfunction
tobelargeortobesmall.Thediscussionofmagnitudeofatransferfunctionisthesubject
knownasfrequencyresponse(Bodediagramsforexample).
Consideringthetrackingerror,itisevidentthat,foragiven0(s),toreducetheinfluenceofthe
disturbance,I
d
(s),onthetrackingerror,E(s),wedesireL(s)tobelargeovertherangeof
frequenciesthatcharacterizethedisturbances.Thatway,thetransferfunction
u(s)
1+L(s)
willbesmall,
therebyreducingtheinfluenceofI
d
(s).SinceI(s) = 0
c
(s)0(s),thisimpliesthatweneedto
designthecontroller0
c
(s)tohavealargemagnitude.Conversely,toattenuatethemeasurement
noise,N(s),andreducetheinfluenceonthetrackingerror,wedesireL(s)tobesmallovertherange
offrequenciesthatcharacterizethemeasurementnoise.Thetransferfunction
L(s)
1+L(s)
willbesmall,
therebyreducingtheinfluenceofN(s).Again,sinceI(s) = 0
c
(s)0(s),thatimpliesthatweneed
todesignthecontroller0
c
(s)tohaveasmallmagnitude.Fortunately,theapparentconflict
betweenwantingtomake0
c
(s)largetorejectdisturbancesandthewantingtomake0
c
(s)smallto
attenuatemeasurementnoisecanbeaddressedinthedesignphasebymakingtheloopgain,L(s),
largeatlowfrequencies(generallyassociatedwiththefrequencyrangeofdisturbances),andmaking
L(s)smallathighfrequencies(generallyassociatedwithmeasurementnoise).
Wenowdiscusshowwecanusefeedbacktoreducethesensitivityofthesystemtovariationsand
uncertaintyinparametersintheprocess,G(s).Thisisaccomplishedbyanalysingthetrackingerror
whenI
d
(s) = N(s)=0.
SensitivityofControlSystemstoparametervariations
Aprocess,representedbythetransferfunctionG(s),whateveritsnature,issubjecttoachanging
environment,aging,ignoranceoftheexactvaluesoftheprocessparameters,andothernatural
3

factorsthataffectacontrolprocess.Intheopenloopsystem,alltheseerrorsandchangesresultin
achangingandinaccurateoutput.However,aclosedloopsystemsensesthechangeintheoutput
duetotheprocesschangesandattemptstocorrecttheoutput.Thesensitivityofacontrolsystem
toparametervariationsisofprimeimportance.Aprimaryadvantageofaclosedloopfeedback
controlsystemisitsabilitytoreducethesystemssensitivity.
Fortheclosedloopcase,if0
c
(s)0(s) > 1forallcomplexfrequenciesofinterest,wecanuseEqn.1
toobtain(lettingI
d
(s) = N(s)=0.)(s) R(s)
Theoutputisapproximatelyequaltotheinput.However,thecondition0
c
(s)0(s) > 1maycause
thesystemresponsetobehighlyoscillatoryandevenunstable.Butthefactthatincreasingthe
magnitudeoftheloopgainreducestheeffectofG(s)ontheoutputisanexceedinglyusefulresult.
Therefore,thefirstadvantageofafeedbacksystemisthattheeffectofthevariationofthe
parametersoftheprocess,G(s),isreduced.
Supposetheprocess(orplant)G(s)undergoesachangesuchthatthetrueplantmodelisG(s)+
G(s).Thechangeintheplantmaybeduetoachangingexternalenvironmentornaturalaging,orit
mayjustrepresenttheuncertaintyincertainplantparameters.Weconsidertheeffectonthe
trackingerrorE(s)duetoG(s).Relyingontheprincipleofsuperposition,wecanletI
d
(s) = N(s)
=0andconsideronlythereferenceinputR(s).FromEqn.2,itfollowsthat
E(s) + E(s) =
1
1 +0
c
(s)|0(s) + 0(s)]
R(s)
Thenthechangeinthetrackingerroris
E(s) =
-0
c
(s)0(s)
|1 +0
c
(s)0(s) +0
c
(s)0(s)](1 + 0
c
(s)0(s))
R(s)
Sinceweusuallyfindthat0
c
(s)0(s) > 0
c
(s)0(s),wehave
E(s) =
-0
c
(s)0(s)
|1 +I(s)]
2
R(s)
Weseethatthechangeinthetrackingerrorisreducedbythefactor1 +I(s),whichisgenerally
greaterthan1overtherangeoffrequenciesofinterest.
ForlargeI(s),wehave1 +I(s) = I(s),andwecanapproximatethechangeinthetrackingerror
by
E(s) =
-1
I(s)
0(s)
0(s)
R(s)
LargemagnitudeI(s)translatesintosmallerchangesinthetrackingerror(thatis,reduced
sensitivitytochangesinG(s)intheprocess).Also,largerL(s)impliessmallersensitivity,S(s).The
questionarises,howdowedefinesensitivity?Sinceourgoalistoreducesystemsensitivity,it
makessensetoformallydefinetheterm.
4

Thesystemsensitivityisdefinedastheratioofthepercentagechangeinthesystemtransfer
functiontothepercentagechangeoftheprocesstransferfunction.Thesystemtransferfunctionis
I(s) =
(s)
R(s)
,andthereforethesensitivityisdefinedasS =
11
uu
=
In1
Inu

Systemsensitivityistheratioofthechangeinthesystemtransferfunctiontothechangeofa
processtransferfunction(orparameter)forasmallincrementalchange.
ThesensitivityoftheopenloopsystemtochangesintheplantG(s)isequalto1.Thesensitivityof
theclosedloopisreadilyobtainedbyusingS =
11
uu
=
In1
Inu
.Thesystemtransferfunctionof
theclosedloopsystemisI(s) =
u
c
(s)u(s)
1+u
c
(s)u(s)

Therefore,thesensitivityofthefeedbacksystemis S
u
1
=
1
u
u
1
=
u
c
|1+u
c
u]
2
u
uu
c
|1+u
c
u]

Or S
u
1
=
1
|1+u
c
(s)u(s)]

Wefindthatthesensitivityofthesystemmaybereducedbelowthatoftheopenloopsystemby
increasingI(s) +0
c
(s)0(s)overthefrequencyrangeofinterest.
Often,weseektodetermine S
u
1
,whereisaparameterwithinthetransferfunctionofablockG.
Usingthechainrule,wefindthat S
u
1
= S
u
1
S
u
u

Veryoften,thetransferfunctionofthesystemI(s)isafractionoftheform(s, o) =
N(s,u)
(s,u)
,
whereisaparameterthatmaybesubjecttovariationduetotheenvironment.Thenwemay
obtainthesensitivitytobyrewritingsensitivityfunctionas
S
u
1
=
olnI
olno
=
olnN
olno
_
o
0
-
oln
olno
_
o
0
= S
u
N
- S
u

where
0
isthenominalvalueoftheparameter.
Animportantadvantageoffeedbackcontrolsystemsistheabilitytoreducetheeffectofthe
variationofparametersofacontrolsystembyaddingafeedbackloop.Toobtainhighlyaccurate
openloopsystems,thecomponentsoftheopenloop,0(s),mustbeselectedcarefullyinorderto
meettheexactspecifications.However,aclosedloopsystemallows0(s)tobelessaccurately
specified,becausethesensitivitytochangesorerrorsin0(s)isreducedbytheloopgainI(s).This
benefitofclosedloopsystemsisaprofoundadvantagefortheelectronicamplifiersofthe
communicationindustry.
Anexampleisgivenbelowtoillustratethevalueoffeedbackforreducingsensitivity.
FeedbackAmplifier
AnamplifierusedinmanyapplicationshasagainK
a
,asshowninfigurebelow:
5

Theoutputvoltageis:
o
= -K
u
:
n
.
WeoftenaddfeedbackusingapotentiometerR
p
,asshownbelow.Thetransferfunctionofthe
amplifierwithoutfeedbackisT=K
a
andthesensitivitytochangesintheamplifiergainis S
K
c
1
= 1
Theblockdiagrammodeloftheamplifierwithfeedbackisshownbelow,

where[ =
R
2
R
1
andR
p
= R
1
+ R
2
.Theclosedlooptransferfunctionofthefeedbackamplifieris
I =
-K
c
1+[K
c

Thesensitivityoftheclosedloopfeedbackamplifieris
S
K
c
1
= S
u
1
S
K
c
u
=
1
1 +[K
u

IfK
a
islarge,thesensitivityislow.Forexample,ifK
a
=10
4
and[ = u.1wehave
S
K
c
1
=
1
1 +1u
3

orthemagnitudeisonethousandthofthemagnitudeoftheopenloopamplifier.

DisturbancesignalsinaFeedbackcontrolsystem
Animportanteffectoffeedbackinacontrolsystemisthecontrolandpartialeliminationofthe
effectofdisturbancesignals.Adisturbancesignalisanunwantedinputsignalthataffectsthe
outputsignal.Manycontrolsystemsaresubjecttoextraneousdisturbancesignalsthatcausethe
systemtoprovideaninaccurateoutput.Electronicamplifiershaveinherentnoisegeneratedwithin
theintegratedcircuitsortransistors;radarantennasaresubjectedtowindgusts;andmanysystems
generateunwanteddistortionsignalsduetononlinearelements.Thebenefitoffeedbacksystemsis
thattheeffectofdistortion,noise,andunwanteddisturbancescanbeeffectivelyreduced.
Disturbancerejection
WhenR(s) = N(s) = u,itfollowsthat
E(s) = -S(s)0(s)I
d
(s) = -
0(s)
1 +I(s)
I
d
(s)
Forafixed0(s)andagivenI
d
(s),astheloopgainI(s)increases,theeffectofI
d
(s)onthe
trackingerrordecreases.Inotherwords,thesensitivityfunctionS(s)issmallwhentheloopgainis
large.Wesaythatlargeloopgainleadstogooddisturbancerejection.Moreprecisely,forgood
disturbancerejection,werequirealargeloopgainoverthefrequenciesofinterestassociatedwith
theexpecteddisturbancesignals.
Inpractice,thedisturbancesignalsareoftenlowfrequency.Whenthatisthecase,wesaythatwe
wanttheloopgaintobelargeatlowfrequencies.Thisisequivalenttostatingthatwewantto
designthecontroller0
c
(s)sothatthesensitivityfunctionS(s)issmallatlowfrequencies.
Figure:Steelrollingmill.

Figure:ModelofanarmaturecontrolledDCmotorwithLoadTorqueDisturbance.
7

Asaspecificexampleofasystemwithanunwanteddisturbance,letusconsiderthespeedcontrol
systemforasteelrollingmill.Therolls,whichprocesssteel,aresubjectedtolargeloadchangesor
disturbances.Asasteelbarapproachestherolls,therollsareempty.However,whenthebar
engagesintherolls,theloadontherollsincreasesimmediatelytoalargevalue.Thisloadingeffect
canbeapproximatedbyastepchangeofdisturbancetorque.Alternatively,theresponsecanbe
seenfromthespeedtorquecurvesofatypicalmotor,asshowninthefigurebelow.

Figure:SpeedTorquecurvesofatypicalmotor
ThetransferfunctionmodelofanarmaturecontrolledDCmotorwithaloadtorquedisturbanceis
showninfigureabove.ItisassumedthatI
u
isnegligible.LetR(s)=0andexamineE(s)= (s),
foradisturbance I
d
(s).
Thechangeinspeedduetotheloaddisturbanceisthen
E(s) = -(S) =
1
[s + b +K
m
K
b
R
u
I
d
(s)
Thesteadystateerrorinspeedduetotheloadtorque, I
d
(s) = s,isfoundbyusingthefinal
valuetheorem.Therefore,fortheopenloopsystem,wehave
lim
t-
E(t) = lim
s-0
sE(s) = lim
s-0
s=
1
]s+b+K
m
K
b
R
c
[

=

b+K
m
K
b
R
c
= -
0
()
Theclosedloopspeedcontrolsystemisshowninblockdiagramformbelow.
8

Theclosedloopsystemisshownintheblockdiagramforminthefollowingfigure,where0
1
(s)=
K
m
K
b
R
u
,0
2
(s)=1([s +b),andH(s)=K
t
+K
b
K
u
.

Figure:BlockdiagramofaclosedlooparmaturecontrolledDCmotor.
Theerror,E(s) = -(S),oftheclosedloopsystemis:
E(s) = -(S) =
u
2
(s)
1+u
1
(s)u
2
(s)H(s)
I
d
(s)
Then,if0
1
(s)0
2
(s)E(s)ismuchgreaterthan1overtherangeofs,weobtaintheapproximate
resultE(s) =
1
u
1
(s)H(s)
I
d
(s)
Therefore,if0
1
(s)E(s)ismadesufficientlylarge,theeffectofthedisturbancecanbedecreasedby
closedloopfeedback.Notethat
0
1
(s)E(s) = (K
m
K
u
R
u
)(K
t
+ K
b
K
u
) = K
m
K
u
K
t
R
u
,sinceK
u
> K
b
.
Thus,westrivetoobtainalargeamplifergain,K
u
,andkeepR
u
< 2.Theerrorforthesystem
shownisE(s) = R(s) -(s),andR(s) =
d
(s),thedesiredspeed.

9

DESIGNEXAMPLES
Threeillustrativeexamplesarepresentedhere:theEnglishChannelboringmachine,theMarsrover,
andabloodpressurecontrolproblemduringanaesthesia.TheEnglishChannelboringmachine
examplefocusesontheclosedloopsystemresponsetodisturbances.TheMarsroverexample
highlightstheadvantagesofclosedloopfeedbackcontrolindecreasingsystemsensitivitytoplant
changes.Thefinalexampleonbloodpressurecontrolisamoreindepthlookatthecontroldesign
problem.Sincepatientmodelsintheformoftransferfunctionsaredifficulttoobtainfrombasic
biologicalandphysicalprinciples,adifferentapproachusingmeasureddataisdiscussed.The
positiveimpactofclosedloopfeedbackcontrolisillustratedinthecontextofdesign.
EXAMPLEEnglishChannelboringmachines
Thetunnelis38kmlongandisbored61metersbelowsealevel.Thetunnelaccommodates50train
tripsdaily.
Themachines,operatingfrombothendsofthechannel,boredtowardthemiddle.Tolinkup
accuratelyinthemiddleofthechannel,alaserguidancesystemkeptthemachinespreciselyaligned.
Amodeloftheboringmachinecontrolisshowninthefigurebelow,whereY(s)istheactualangleof
directionoftraveloftheboringmachineandR(s)isthedesiredangle.Theeffectofloadonthe
machineisrepresentedbythedisturbance,I
d
(s).

ThedesignobjectiveistoselectthegainKsothattheresponsetoinputanglechangesisdesirable
whilewemaintainminimalerrorduetothedisturbance.Theoutputduetothetwoinputsis
(s) =
K + 11s
s
2
+ 12s + K
R(s) +
1
s
2
+ 12s + K
I
d
(s)
Thus,toreducetheeffectofthedisturbance,wewishtosetthegaingreaterthan10.Whenwe
selectK=100andletthedisturbancebezero,wehavethestepresponseforaunitstepinputr(t),
whichhas20%overshoot.IfwesetthegainKequalto20,weobtaintheresponseofy(t)dueto
stepinputr(t)whichhaslessthan4%overshootandalsothesteadystatecanbeobtainedin
relativelyshorttime.
ThesteadystateerrorofthesystemtoaunitstepinputR(s)=1/sis
10

lim
t-
c(t) = lim
s-0
s
1
1 +
K + 11s
s(s + 1)
[
1
s

= u
Thesteadystatevalueofy(t)whenthedisturbanceisaunitstep,I
d
(s)=1/s,andthedesiredvalue
isr(t)=0is
lim
t-
y(t) = lim
s-0

1
s(s + 12) + K
=
1
K

Thus,thesteadystatevalueis0.01and0.05forK=100and20,respectively.
Finally,weexaminethesensitivityofthesystemtoachangeintheprocessG(s).
S
u
1
=
s(s + 1)
s(s + 12) +K

Forlowfrequencies(|s| < 1),thesensitivitycanbeapproximatedbyS


u
1
=
s
K
whereK20.Thus,
thesensitivityisreducedbyincreasingthegain,K.Inthiscase,wechooseK=20forareasonable
designcompromise.
EXAMPLEMarsrovervehicle
ThesolarpoweredMarsrovernamedSojourner(landedJuly1997)wascontrolledbyoperatorson
Earthusingcontrolsontherover.In2004twinautonomousroversSpiritandOpportunitywere
launched.Thenewroverswere1.6mlongandweighed174kg.Spirithastravelledover7kmand
Opportunityhastravelledover10km.ThevehiclesarecontrolledfromEarthbysendingitpath
commands,r(t).
Averysimplifiedmodelofaroverisshowninfigurebelow.

Figure:Modelofaroverwithoutfeedback(openloop)

Figure:Modelofaroverwithfeedback(closedloop)
11

Thesystemmaybeoperatedwithorwithoutfeedback.Thegoalistooperatetheroverwith
modesteffectsfromdisturbancessuchasrocksandwithlowsensitivitytochangesinthegainK.
Thetransferfunctionfortheopenloopsystemis
I
o
(s) =
(s)
R(s)
=
K
s
2
+ 4s +S
,
andthetransferfunctionfortheclosedloopsystemis
I
c
(s) =
(s)
R(s)
=
K
s
2
+ 4s +S +K
,
Then,forK=2,
I(s) = I
o
(s) = I
c
(s) =
2
s
2
+ 4s +S
,
Hence,wecancomparethesensitivityoftheopenloopandclosedloopsystemsforthesame
transferfunction.
Thesensitivityfortheopenloopsystemis
S
K
1
c
=
JI
o
JK
K
I
o
= 1 ,
andthesensitivityfortheclosedloopsystemis
S
K
1
c
=
JI
c
JK
K
I
c
=
s
2
+ 4s + S
s
2
+ 4s +S +K
.
Toexaminetheeffectofthesensitivityatlowfrequencies,welets=jtoobtain
S
K
1
c
=
JI
c
JK
K
I
c
=
(S -
2
) + ]4
(S +K -
2
) +]4
.
ForK=2,thesensitivityatlowfrequencies, < u.1,is|S
K
1
c
| = u.6
Afrequencyplotofmagnitudeofthesensitivityisshowninfigurebelow.
Notethatthesensitivityforlowfrequenciesis|S
u
1
c
| < u.8,for 1.
TheeffectofthedisturbancecanbedeterminedbysettingR(s)=0andlettingI
d
(s)=1/s.Then,for
theopenloopsystem,wehavethesteadystatevalue
y() =lim
s-u
s _
1
(s + 1)(s +S)
_
1
s
=
1
S
.
Theoutputoftheclosedloopsystemwithaunitstepdisturbance,I
d
(s)=1/s,is
y() =lim
s-u
s _
1
s
2
+ 4s + S + K
_
1
s
=
1
S +K
.
12

WhenK=2, y() = 1S.Becauseweseektominimizetheeffectofthedisturbance,itisclearthat


alargervalueofKwouldbedesirable.AnincreasedvalueofK,suchasK=50,willfurtherreducethe
effectofthedisturbanceaswellasreducethemagnitudeofthesensitivity.However,aswe
increaseKbeyondK=50,thetransientperformanceofthesystemfortherampinput,r(t),beginsto
deteriorate.
EXAMPLEBloodpressurecontrolduringanaesthesia
Manyvitalparameters,suchasbloodpressure,heartrate,temperature,bloodoxygenation,and
exhaledcarbondioxide,arecontrolledwithinacceptableboundsbytheanaesthetist.Ofcourse,to
ensurepatientsafety,adequateanaesthesiamustbemaintainedduringtheentiresurgical
procedure.Anyassistancethattheanaesthetistcanobtainautomaticallywillincreasethesafety
marginsbyfreeingtheanaesthetisttoattendtootherfunctionsnoteasilyautomated.Thisisan
exampleofhumancomputerinteractionfortheoverallcontrolofaprocess.Clearly,patientsafety
istheultimateobjective.Ourcontrolgoalthenistodevelopanautomatedsystemtoregulatethe
depthofanaesthesia.Thisfunctionisamenabletoautomaticcontrolandinfactisinroutineusein
clinicalapplications.
Weconsiderhowtomeasurethedepthofanaesthesia.Manyanaesthetistsregardmeanarterial
pressure(MAP)asthemostreliablemeasureofthedepthofanaesthesia.TheleveloftheMAP
servesasaguideforthedeliveryofinhaledanaesthesia.Basedonclinicalexperienceandthe
proceduresfollowedbytheanaesthetist,wedeterminethatthevariabletobecontrolledisthe
meanarterialpressure.
Theelementsofthecontrolsystemdesignprocessemphasizedinthisexampleareshowninthe
figurebelow.
13

Figure:Elementsofthecontrolsystemdesignprocessforthebloodpressurecontrolexample
Fromthecontrolsystemdesignperspective,thecontrolgoalcanbestatedinmoreconcreteterms:
Controlgoal
Regulatethemeanarterialpressuretoanydesiredsetpointandmaintaintheprescribedsetpoint
inthepresenceofunwanteddisturbances.
Associatedwiththestatedcontrolgoal,weidentifythevariabletobecontrolled:
Variabletobecontrolled:meanarterialpressure(MAP).
Becauseitisourdesiretodevelopasystemthatwillbeusedinclinicalapplications,itisessentialto
establishrealisticdesignspecifications.Ingeneraltermsthecontrolsystemshouldhaveminimal
complexitywhilesatisfyingthecontrolspecifications.Minimalcomplexitytranslatesintoincreased
systemreliabilityanddecreasedcost.
TheclosedloopsystemshouldrespondrapidlyandsmoothlytochangesintheMAPsetpoint(made
bytheanaesthetist)withoutexcessiveovershoot.Theclosedloopsystemshouldminimizethe
effectsofunwanteddisturbances.Therearetwoimportantcategoriesofdisturbances:surgical
disturbances,suchasskinincisionsandmeasurementerrors,suchascalibrationerrorsandrandom
stochasticnoise.Forexample,askinincisioncanincreasetheMAPrapidlyby10mmHg.Finally,
14

sincewewanttoapplythesamecontrolsystemtomanydifferentpatientsandwecannot(for
practicalreasons)haveaseparatemodelforeachpatient,wemusthaveaclosedloopsystemthatis
insensitivetochangesintheprocessparameters(thatis,itmeetsthespecificationsformany
differentpeople).
Basedonclinicalexperience,wecanexplicitlystatethecontrolspecificationsasfollows:
ControlDesignSpecifications
DS1 Settlingtimelessthan20minutesfora10%stepchangefromtheMAPsetpoint.
DS2 Percentovershootlessthan15%fora10%stepchangefromtheMAPsetpoint.
DS3 ZerosteadystatetrackingerrortoastepchangefromtheMAPsetpoint.
DS4 Zerosteadystateerrortoastepsurgicaldisturbanceinput(ormagnitude|J(t)| Su)with
amaximumresponselessthan_5%oftheMAPsetpoint.
DS5 Minimumsensitivitytoprocessparameterchanges.

Wecoverthenotionofpercentovershoot(DS1)andsettlingtime(DS2)morethoroughlyelsewhere
inthisunit.Theyfallinthecategoryofsystemperformance.Theremainingthreedesign
specifications,DS3DS5,coveringsteadystatetrackingerrors(DS3),disturbancerejection(DS4),and
systemsensitivitytoparameterchanges(DS5)arethemaintopicshere.Thelastspecification,DS5,
issomewhatvague;however,thisisacharacteristicofmanyrealworldspecifications.Inthesystem
configuration,weidentifythemajorsystemelementsasthecontroller,anaesthesiapump/vaporizer,
sensor,andpatient.

Figure:Bloodpressurecontrolsystemconfiguration
ThesysteminputR(s)isthedesiredmeanarterialpressurechange,andtheoutputY(s)istheactual
pressurechange.Thedifferencebetweenthedesiredandthemeasuredbloodpressurechange
formsasignalusedbythecontrollertodeterminevaluesettingstothepump/vaporizerthatdelivers
anaesthesiavapourtothepatient.
15

Themodelofthepump/vaporizerdependsdirectlyonthemechanicaldesign.Wewillassumea
simplepump/vaporizer,wheretherateofchangeoftheoutputvapourisequaltotheinputvalve
setting,or u (t) = :(t).Thetransferfunctionofthepumpisthusgivenby0
p
(s) =
0(s)
v(s)
=
1
s
.
Thisisequivalenttosayingthat,fromaninput/outputperspective,thepumphastheimpulse
responseh(t) = 1 t u.
Developinganaccuratemodelofapatientismuchmoreinvolved.Becausethephysiological
systemsinthepatient(especiallyinasickpatient)arenoteasilymodelled,amodellingprocedure
basedonknowledgeoftheunderlyingphysicalprocessesisnotpractical.Evenifsuchamodelcould
bedeveloped,itwould,ingeneral,beanonlinear,timevarying,multiinput,multioutputmodel.
Thistypeofmodelisnotdirectlyapplicablehereinourlinear,timeinvariant,singleinput,single
outputsystemsetting.
Ontheotherhand,ifweviewthepatientasasystemandtakeaninput/outputperspective,wecan
usethefamiliarconceptofanimpulseresponse.Thenifwerestrictourselvestosmallchangesin
bloodpressurefromagivensetpoint(suchas100mmHg),wemightmakethecasethatinasmall
regionaroundthesetpointthepatientbehavesinalineartimeinvariantfashion.Thisapproachfits
wellintoourrequirementtomaintainthebloodpressurearoundagivensetpoint(orbaseline).The
impulseresponseapproachtomodellingthepatientresponsetoanaesthesiahasbeenused
successfullyinthepast.
Supposethatwetakeablackboxapproachandobtaintheimpulseresponseforahypothetical
patient.Theimpulseresponsewouldinitiallyhaveatimedelay.Thisisduetothefactthatittakesa
finiteamountoftimeforthepatientMAPtorespondtotheinsertionofanaesthesiavapour.A
reasonablefitofthedataisgivenbyy(t) = tc
-pt
t u,wherep=2andtime(t)ismeasuredin
minutes.Differentpatientsareassociatedwithdifferentvaluesoftheparameterp.The
correspondingtransferfunctionis0(s) =
1
(s+p)
2
.Forthesensorweassumeaperfectnoisefree
measurementandH(s)=1.
Therefore,wehaveaunityfeedbacksystem.
Agoodcontrollerforthisapplicationisaproportionalintegralderivative(PID)controller:
0
c
(s) = K
P
+ sK

+
K
I
s
=
K

s
2
+ K
P
s +K
I
s

whereK
P
,K

, onJ K
I
arethecontrollergainstobedeterminedtosatisfyalldesign
specifications.Theselectedkeyparametersareasfollows:
SelectKeyTuningParameters
ControllergainsK
P
,K

, onJ K
I

Webegintheanalysisbyconsideringthesteadystateerrors.Thetrackingerror(witb I
d
(s).=0and
N(s).=0)is
16

E(s) = R(s) - (s) =


1
1 +0
c
(s)0
p
(s)0(s)
R(s)
or
E(s) =
s
4
+2ps
3
+ p
2
s
2
s
4
+ 2ps
3
+ (p
2
+ K

)s
2
+ K
P
s + K
I
R(s)
Usingfinalvaluetheorem,wedeterminethatthesteadystatetrackingerroris
lim
s-0
sE(s) = lim
s-0
R
o
(s
4
+ 2ps
S
+ p
2
s
2
)
s
4
+ 2ps
S
+ (p
2
+ K

)s
2
+ K
P
s + K
I
= u
whereR(s) = R
o
sisastepinputofmagnitudeR
o
.Therefore, lim
t-
c(t)=0.
WithaPIDcontroller,weexpectazerosteadystatetrackingerror(toastepinput)foranynonzero
valuesofK
P
,K

, onJ K
I
. Theintegralterm,K
I
/s,inthePIDcontrolleristhereasonthatthe
steadystateerrortoaunitstepiszero.ThusdesignspecificationDS3issatisfied.
Whenconsideringtheeffectofastepdisturbanceinput,weletR(s) = uandN(s) = u.Wewant
thesteadystateoutput(s)tobezeroforastepdisturbance.Thetransferfunctionfromthe
disturbanceI
d
(s)totheoutput(s) is
(s) =
-0(s)
1 +0
c
(s)0
p
(s)0(s)
I
d
(s)
=
-s
2
s
4
+ 2ps
3
+(p
2
+ K

)s
2
+ K
P
s +K
I
I
d
(s)
WhenI
d
(s) =

c
s
,wefindthat
lim
s-0
s(s) = lim
s-0
-
o
s
2
s
4
+ 2ps
S
+ (p
2
+ K

)s
2
+ K
P
s + K
I
= u
Therefore, lim
t-
y(t)=0.Thusastepdisturbanceofmagnitude
o
willproducenooutputinthe
steadystate,asdesired.Thesensitivityoftheclosedlooptransferfunctiontochangesinpisgiven
byS
P
1
= S
u
1
S
P
u

WecomputeS
P
1
asfollows: S
P
u
=
u(s)
p
p
u(s)
=
-2p
s+p

and
S
u
1
=
1
1 +0
c
(s)0
p
(s)0(s)
=
s
2
(s + p)
2
s
4
+2ps
3
+ (p
2
+ K

)s
2
+K
P
s + K
I

therefore,
S
P
1
= S
u
1
S
P
u
=
2p(s +p)s
2
s
4
+ 2ps
3
+(p
2
+ K

)s
2
+K
P
s +K
I

17

WeneedtoevaluatethesensitivityfunctionS
P
1
,atvariousfrequencies.Forlowfrequencieswe
canapproximatethesystemsensitivityS
P
1
byS
P
1
=
2p
2
s
2
K
I

Soatlowfrequenciesandforagivenpwecanreducethesystemsensitivitytovariationsinpby
increasingthePIDgain,K
I
.SupposethatthreePIDgainsetsareavailableandweplotmagnitudeof
thesensitivityS
P
1
asafunctionoffrequencyforeachPIDcontroller.Inthisway,wecanselectaPID
thatgivesthesmallestsystemsensitivitytochangesintheprocessparameter,p.

You might also like