Sei sulla pagina 1di 151

OPTIMALCONTROL

DesignofControlSystemsinStateSpace

2012/2013 JosSdaCosta Chapter2


DesignofControlSystemsinStateSpace
Summary

2.DesignofControlSystemsinStateSpace
2.1StateSpaceControl
2.2RegulatordesignbyPolePlacementContinuousTime
systems
2.3RegulatordesignbyPolePlacementDiscreteTimesystems
2.4SolvingPoleplacementwithMATLAB
2.5DesignofServoSystemsbyPolePlacement
2.6PolePlacementinMIMOSystems

JosSdaCosta DesignofControlSystemsinStateSpace 2
2.1StateSpaceControl
In this chapter we discuss pole placement (assignment) method in
state space.
The poleplacement method in statespace is somewhat similar to
the rootlocus method in that we place closedloop poles at desired
locations.
The basic difference is that in the rootlocus design we place only the
dominant closed loop poles at the desired locations, while in the
poleplacement design we place all closedloop poles at desired
locations.
In rootlocus we design a controller (compensator) such that the
dominant closedloop poles have a desired damping ratio and an
undamped natural n.
Inrootlocusdesignweassumetheeffectsontheresponsesofno
dominantclosedlooppolestobenegligible.

JosSdaCosta DesignofControlSystemsinStateSpace 3
2.1StateSpaceControl
Inpoleplacementdesigninstatespace:
Weassumethatallstatevariablesaremeasurableandareavailable
forfeedback.
Our initial discussions is mainly focused to SISO systems. The MIMO
case is much more complex since there is no unique solution. The
MIMO case is discussed later.
Different from specifying only dominant closedloop poles (the con
ventional design approach), the present poleplacement approach
specifies all closedloop poles.
The desired closedloop poles are located based on the transient
response and/or frequencyresponse requirements, such as speed,
damping ratio, or bandwidth, as well as steadystate requirements.

JosSdaCosta DesignofControlSystemsinStateSpace 4
2.1StateSpaceControl
Inpoleplacementdesign:
There is a cost associated with placing all closedloop poles, since
placing all closedloop poles requires successful measurements of all
state variables or else requires the inclusion of a state observer in
the system.
Thereisalsoarequirementonthepartofthesystemfortheclosed
looppolestobeplacedatarbitrarilychosenlocations thatthe
systembecompletelystatecontrollable.

JosSdaCosta DesignofControlSystemsinStateSpace 5
DesignofControlSystemsinStateSpace
Summary

2.2RegulatordesignbyPolePlacementContinuousTime(CT)
Systems
2.2.1Necessaryandsufficientconditions
2.2.2Determinationofthegainmatrix
2.2.3Locationofclosedlooppoles

JosSdaCosta DesignofControlSystemsinStateSpace 6
2.2RegulatordesignbyPolePlacementCT
ConsiderthecontinuoustimeSISOsystemgivenby
x (t ) = Ax(t ) + Bu (t ) (1)
y (t ) = Cx (t ) + Du (t ) (2)
where
x =statevector(nvector)
u =unconstrained controlsignal(scalar)
y =outputsignal(scalar)
A =n xn state matrix
B =n x1 inputmatrix
C =n x1 outputmatrix
D =constant(scalar)
Weshallchoosethecontrolsignaltobe u = Kx (3)
Thismeansthatthecontrolsignalu isdeterminedbyaninstantaneous
statefeedback.
The1 x n matrixK iscalledthestatefeedbackgainmatrix.

JosSdaCosta DesignofControlSystemsinStateSpace 7
2.2RegulatordesignbyPolePlacementCT

Thisclosedloopsystemhasnoreferenceinput(regulatorproblem).Its
objectiveistomaintainthezerooutput.
Becauseofthedisturbancesthatmaybepresent,theoutputwill
deviatefromzero.

JosSdaCosta DesignofControlSystemsinStateSpace 8
2.2RegulatordesignbyPolePlacementCT
Substitutingeq.(3)intoeq.(1)gives
(t) = (A BK)x(t)
Thesolutionofthisequationisgivenby
x(t ) = e( A-BK )t x(0) (4)
wherex(0) istheinitialstatecausedbyexternaldisturbances.
Thestabilityandtransientresponsecharacteristicsaredeterminedby
theeigenvaluesofmatrixA BK.
IfmatrixK ischosenproperly,thematrixA BK canbemadean
asymptoticallystable matrix,thusforallx(0) 0,itispossibletomake
x(t) approach0 ast approachesinfinity.
TheeigenvaluesofmatrixA BK arecalledtheregulatorpoles.
Iftheseregulatorpolesareplacedinthelefthalfsplane,thenx(t)
approaches0 ast approachesinfinity.
JosSdaCosta DesignofControlSystemsinStateSpace 9
2.2RegulatordesignbyPolePlacementCT
Theproblemofplacingtheregulatorpoles(closedlooppoles)atthe
desiredlocationiscalledapoleplacementorpoleassignment
problem.
2.2.1NecessaryandSufficientConditionforArbitraryPole
Placement
Weshallprovethatarbitrarypoleplacementforagivensystemis
possibleifandonlyifthesystemiscompletelystatecontrollable.
Thenecessarycondition:
We begin by proving that if the system is not completely state
controllable, then there are eigenvalues of matrix A BK that cannot
be controlled by state feedback.
Suppose that the system of eq. (1) is not completely state controlla
ble. Then the rank of the controllability matrix is less than n, or

JosSdaCosta DesignofControlSystemsinStateSpace 10
2.2RegulatordesignbyPolePlacementCT
Thismeansthatthereareq linearlyindependentcolumnvectorsinthe
controllabilitymatrix.
Letusdefinesuchq linearlyindependentcolumnvectorsasf1,f2,...,
fq.
Also,letuschoosen q additionalnvectorsvq+1,vq+2,...,vn such
that

isofrankn.
Then,bysimilaritytransformationofmatrixA weobtain

Nowdefine

JosSdaCosta DesignofControlSystemsinStateSpace 11
2.2RegulatordesignbyPolePlacementCT
Thenwehave

whereIq isaqdimensionalidentitymatrixandInq,isan(n q)dimen


sionalidentitymatrix.
NoticethattheeigenvaluesofA22 donotdependonK.

JosSdaCosta DesignofControlSystemsinStateSpace 12
2.2RegulatordesignbyPolePlacementCT
Thus,ifthesystemisnotcompletelystatecontrollable,thenthereare
eigenvaluesofmatrixA thatcannotbearbitrarilyplaced.
Therefore,toplacetheeigenvaluesofmatrixA BK arbitrarily,the
systemmustbecompletelystatecontrollable(necessarycondition).
Thesufficientcondition:
Ifthesystemiscompletelystatecontrollable,thenalleigenvaluesof
matrixA canbearbitrarilyplaced.
Inprovingasufficientcondition,itisconvenienttotransformthestate
equationgivenbyeq.(1)intothecontrollablecanonicalform.
DefineatransformationmatrixT by
T=MW (5)
whereM isthecontrollabilitymatrix
(6)

JosSdaCosta DesignofControlSystemsinStateSpace 13
2.2RegulatordesignbyPolePlacementCT
and

(7)

wheretheai's arecoefficientsofthecharacteristicpolynomial

Defineanewstatevectorby

JosSdaCosta DesignofControlSystemsinStateSpace 14
2.2RegulatordesignbyPolePlacementCT
If the rank of the controllability matrix M is n (meaning that the
system is completely state controllable), then the inverse of matrix T
exists, and eq. (1) can be modified to
(8)
where

(9) (10)

JosSdaCosta DesignofControlSystemsinStateSpace 15
2.2RegulatordesignbyPolePlacementCT
Thus, given a state equation eq. (1), it can be transformed into the
controllable canonical form if the system is completely state control
lable and if we transform the state vector x into state vector by use
of the transformation matrix T given by eq. (5).
Letuschooseasetofthedesiredeigenvaluesas1,2,...,n.Then
thedesiredcharacteristicequationbecomes

(11)
Letuswrite
(12)
When u (t ) = -KTx isusedtocontrolthesystemgivenbyeq.(8),the
systemequationbecomes

JosSdaCosta DesignofControlSystemsinStateSpace 16
2.2RegulatordesignbyPolePlacementCT
Thecharacteristicequationis

Thischaracteristicequationisthesameasthecharacteristicequation
u (t ) = -Kx
forthesystem,definedbyeq.(1),whenisusedasthe
controlsignal.
Nowletussimplifythecharacteristicequationofthesysteminthe
controllablecanonicalform.Referringtoeqs(9),(10),and(11),we
have

JosSdaCosta DesignofControlSystemsinStateSpace 17
2.2RegulatordesignbyPolePlacementCT

(13)
Thisisthecharacteristicequationforthesystemwithstatefeedback.
Therefore,itmustbeequaltoeq.(11),thedesiredcharacteristic
equation.

JosSdaCosta DesignofControlSystemsinStateSpace 18
2.2RegulatordesignbyPolePlacementCT
By equating the coefficients
of like powers of s, we get

Solving the preceding equations for the i's and substituting them into
eq. (12) we obtain

(14)
Thus, if the system is completely state controllable, all eigenvalues can
be arbitrarily placed by choosing matrix K according to eq. (14)
(sufficient condition).
JosSdaCosta DesignofControlSystemsinStateSpace 19
2.2RegulatordesignbyPolePlacementCT
We have thus proved that a necessary and sufficient condition for
arbitrary pole placement is that the system be completely state
controllable.
It is noted that if the system is not completely state controllable, but is
stabilizable, then it is possible to make the entire system stable by
placing the closedloop poles at desired locations for q controllable
modes.
Note that the remaining n - q uncontrollable modes are stable since
the system is stabilizable.
So the entire stabilizable system can be made stable.

JosSdaCosta DesignofControlSystemsinStateSpace 20
2.2RegulatordesignbyPolePlacementCT
2.2.2 Determination of the Gain Matrix K
From last results we saw that pole placement design consists to
calculate the state feedback gain matrix K based on the desired pole
location to meet the prespecified transient and steady state characte
ristics of the system response in closedloop.
Supposethatthesystemisdefinedby
= Ax + Bu
andthecontrolsignalisgivenby
u = Kx
Method1:TransformationMatrixT
ThestatefeedbackgainmatrixK thatforcestheeigenvaluesofABK
tobe1,2,...,n(desiredvalues)(ifi isacomplexeigenvalue,then
itsconjugatemustalsobeaneigenvalueofABK )canbedetermined
bythefollowingsteps:

JosSdaCosta DesignofControlSystemsinStateSpace 21
2.2RegulatordesignbyPolePlacementCT
Method 1: Transformation Matrix T (cont.)
Step1:Checkthecontrollabilityconditionforthesystem.Ifthe
systemiscompletelystatecontrollable,thenusethefollowingsteps:
Step2:FromthecharacteristicpolynomialformatrixA,thatis,

determinethevaluesofa1, a2, .. . , an.


Step3:DeterminethetransformationmatrixT thattransformsthe
systemstateequationintothecontrollablecanonicalform.(Ifthe
givensystemequationisalreadyinthecontrollablecanonicalform,
thenT = I.)Itisnotnecessarytowritethestateequationinthe
controllablecanonicalform.AllweneedhereistofindthematrixT.
ThetransformationmatrixT isgivenbyeq.(5),orT=MW
whereM isthecontrollabilitymatrixgivenbyeq.(6)andW isgiven
byeq.(7).
JosSdaCosta DesignofControlSystemsinStateSpace 22
2.2RegulatordesignbyPolePlacementCT
Method 1: Transformation Matrix T (cont.)
Step4: Usingthedesiredeigenvalues(desiredclosedlooppoles),
writethedesiredcharacteristicpolynomial:

anddeterminethevaluesof1, 2, . . . , n.
Step5:TherequiredstatefeedbackgainmatrixK canbedetermi
ned fromeq.(14),rewrittenthus
(14)

Method2:DirectSubstitutionMethod
Ifthesystemisofloworder(n 3),directsubstitutionofmatrixK
intothedesiredcharacteristicpolynomialmaybesimpler.

JosSdaCosta DesignofControlSystemsinStateSpace 23
2.2RegulatordesignbyPolePlacementCT
Method2:DirectSubstitutionMethod (cont.)
Forexample,ifn = 3,thenwritethestatefeedbackgainmatrixK as

SubstitutethisK matrixintothedesiredcharacteristicpolynomial,i.e.
|sI A + BK| = (s 1)(s 2)(s 3)
Sincebothsidesofthischaracteristicequationarepolynomialsins,by
equatingthecoefficientsofthelikepowersofs onbothsides,itis
possibletodeterminethevaluesofkl,k2,andk3.
Thisapproachisconvenientifn = 2 or3.
Forn = 4,5,6,...,thisapproachmaybecomeverytedious.
Notethatifthesystemisnotcompletelycontrollable,matrixK cannot
bedetermined.(Nosolutionexists.)

JosSdaCosta DesignofControlSystemsinStateSpace 24
2.2RegulatordesignbyPolePlacementCT
Method3:Ackermann'sFormula
Thereisawellknownformula,knownasAckermann'sformula,forthe
determinationofthestatefeedbackgainmatrixK.
Weshallpresentthisformulainwhatfollows.
Considerthesystem
= Ax + Bu
whereweusethestatefeedbackcontrolu = Kx.
Weassumethatthesystemiscompletelystatecontrollable.
Wealsoassumethatthedesiredclosedlooppolesareat1,2,...,n.
Theclosedloopstatefeedbackequationis
= (A BK)x
Letusdefine = A BK

JosSdaCosta DesignofControlSystemsinStateSpace 25
2.2RegulatordesignbyPolePlacementCT
Method 3: Ackermann's Formula (cont.)
Thedesiredcharacteristicequationis

SincetheCayleyHamiltontheoremstatesthat satisfiesitsown
characteristicequation,wehave

(15)

Weshallutilizeeq.(15)toderiveAckermann'sformula.
Tosimplifythederivation,weconsiderthecasewheren = 3.(Forany
otherpositiveintegern,thefollowingderivationcanbeeasily
extended.)
JosSdaCosta DesignofControlSystemsinStateSpace 26
2.2RegulatordesignbyPolePlacementCT
Method 3: Ackermann's Formula (cont.)
Considerthefollowingidentities:

Multiplyingtheprecedingequationsinorderby 3, 2, 1 and0,
(where0 = 1),respectively,andaddingtheresults,weobtain

(16)

JosSdaCosta DesignofControlSystemsinStateSpace 27
2.2RegulatordesignbyPolePlacementCT
Method 3: Ackermann's Formula (cont.)
Referringtoeq.(15),wehave

Also,wehave

Substitutingthelasttwoequationsintoeq.(16),wehave

Since() = 0,weobtain

(17)

JosSdaCosta DesignofControlSystemsinStateSpace 28
2.2RegulatordesignbyPolePlacementCT
Method 3: Ackermann's Formula (cont.)
Sincethesystemiscompletelystatecontrollable,theinverseofthe
controllabilitymatrix

exists.
Premultiplyingbothsidesofeq.(17)bytheinverseofthecontrolla
bility matrix,weobtain

Premultiplyingbothsidesofthislastequationby[0 0 1],weobtain

JosSdaCosta DesignofControlSystemsinStateSpace 29
2.2RegulatordesignbyPolePlacementCT
Method 3: Ackermann's Formula (cont.)
whichcanberewrittenas

ThislastequationgivestherequiredstatefeedbackgainmatrixK.

Generalizing,foranarbitrarypositiveintegern,wehavethe
AckermannsFormula tocomputetherequiredstatefeedbackgain
matrixK givenby

(18)

JosSdaCosta DesignofControlSystemsinStateSpace 30
2.2RegulatordesignbyPolePlacementCT
2.2.3 Locations of Desired ClosedLoop Poles
The first step in the poleplacement design approach is to choose the
locations of the desired closedloop poles.
The most frequently used approach is to choose such poles based on
experience in the rootlocus design, placing a dominant pair of closed
loop poles and choosing other poles so that they are far to the left of
the dominant closedloop poles.
Note that if we place the dominant closedloop poles far from the j
axis, so that the system response becomes very fast, the signals in the
system become very large, with the result that the system may
become nonlinear. This should be avoided.
Note that requiring a highspeed response implies requiring large
amounts of control energy. Also, in general, increasing the speed of
response requires a larger, heavier actuator, which will cost more.

JosSdaCosta DesignofControlSystemsinStateSpace 31
2.2RegulatordesignbyPolePlacementCT
Locations of Desired ClosedLoop Poles (cont.)
Anotherapproachisbasedonthequadraticoptimalcontrolapproach.
Thisapproachwilldeterminethedesiredclosedlooppolessuchthatit
balancesbetweentheacceptableresponseandtheamountofcontrol
energyrequired.(Wewilladdressthisapproachlatter).

Example
ConsidertheregulatorsystemshowninFigure.Theplantisgivenby
= Ax + Bu
where

and u = Kx
JosSdaCosta DesignofControlSystemsinStateSpace 32
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
Letuschoosethedesiredclosedlooppolesat
s = 2 + j4, s = 2 j4, s = 10
Wemakesuchachoicebecauseweknowfromexperiencethatsucha
setofclosedlooppoleswillresultinareasonableoracceptabletran
sient response.
DeterminethestatefeedbackgainmatrixK.
Resolution
First,weneedtocheckthecontrollabilitymatrixofthesystem.
SincethecontrollabilitymatrixM isgivenby

wefindthat|M| = 1,andtherefore,rankM = 3.

JosSdaCosta DesignofControlSystemsinStateSpace 33
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
Thus,thesystemiscompletelystatecontrollableandarbitrarypole
placementispossible.
Next,weshallsolvethisproblembyusingeachofthethreemethods
presentedbeforetocalculatethestatefeedbackgainmatrixK.
Method1: TransformationMatrixT
InthismethodthestatefeedbackgainmatrixK isgivenby

Thecharacteristicequationforthesystemis

JosSdaCosta DesignofControlSystemsinStateSpace 34
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
Hence, a1 = 6, a2 = 5, a3 = 1
Thedesiredcharacteristicequationis

Hence,1 = 14, 2 = 60, 3 = 200


Referringtoeq.(14),whereT = I forthisproblembecausethegiven
stateequationisinthecontrollablecanonicalform,thenwehave

Method2:DirectSubstitutionMethod
BydefiningthedesiredstatefeedbackgainmatrixK as

JosSdaCosta DesignofControlSystemsinStateSpace 35
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
andequating|sI - A + BK| withthedesiredcharacteristicequation,
weobtain

Thus,6 + k3 = 14,5 + k2 = 60,1 + k1 = 200 fromwhichweobtain

JosSdaCosta DesignofControlSystemsinStateSpace 36
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
Method3:Ackermann'sformula
Referringtoeq.(18),wehave

Since

JosSdaCosta DesignofControlSystemsinStateSpace 37
2.2RegulatordesignbyPolePlacementCT
Example (cont.)

and

weobtainthestatefeedbackgainmatrixK as

JosSdaCosta DesignofControlSystemsinStateSpace 38
2.2RegulatordesignbyPolePlacementCT
Example (cont.)
Asaconclusionwehave:
The feedback gain matrix K obtained by the three methods are the
same.
If the order n of the system is 4th or higher, methods 1 and 3 are
recommended, since all matrix computations can be carried out by a
computer.
If method 2 is used, hand computations become necessary because
a computer may not easily handle (only by symbolic tools) the
characteristic equation with unknown parameters k1, k2, . . , kn.
Note that matrix K is not unique for a given system, but depends on
the desired closedloop pole locations (which determine the speed
and damping of the response) selected.

JosSdaCosta DesignofControlSystemsinStateSpace 39
2.2RegulatordesignbyPolePlacement
The selection of the desired closedloop poles or the desired
characteristic equation is a compromise between the rapidity of the
response of the error vector and the sensitivity to disturbances and
measurement noises. That is:
If we increase the speed of error response, then the adverse effects of
disturbances and measurement noises generally increase.
If the system is of second order, then the system dynamics (response
characteristics) can be precisely correlated to the location of the desired
closedloop poles and the zero(s) of the plant.
For higherorder systems, the location of the closedloop poles and the
system dynamics (response characteristics) are not easily correlated.
Hence, in determining the state feedback gain matrix K for a given
system, it is desirable to examine by computer simulations the
response characteristics of the system for several different matrices
K (based on several different desired characteristic equations) and
to choose the one that gives the best overall system performance.

JosSdaCosta DesignofControlSystemsinStateSpace 40
DesignofControlSystemsinStateSpace
Summary

2.3RegulatordesignbyPolePlacementDiscretetimesystems
2.3.1Necessaryandsufficientconditions
2.3.2Determinationofthegainmatrix
2.3.3Deadbeatcontrol

JosSdaCosta DesignofControlSystemsinStateSpace 41
2.3RegulatordesignbyPolePlacementDT

Asinthecontinuouscaseinthediscretetimepoleplacementdesign:
Weassumethatallstatevariablesaremeasurableandareavailable
forfeedback.
Inthetimebeingwelimitourdiscussionstosingleinputsingle
outputsystems.
Ourproblemistoreducetheerrorvectortozerowithsufficient
speed.
Allclosedlooppolesarespecifiedinpoleplacementdesign.
Thedesiredclosedlooppolesarelocatedbasedonthetransient
responseand/orfrequencyresponserequirements,suchasspeed,
dampingratio,orbandwidth,aswellassteadystaterequirements.

JosSdaCosta DesignofControlSystemsinStateSpace 42
2.3RegulatordesignbyPolePlacementDT
ConsidertheopenloopcontrolsystemshowninFigure.The
stateequationis
x(k 1) Gx(k ) Hu (k ) (19)
y (k ) Cx(k ) Du (k )

where
x =statevector(nvector)
u =unconstrained control
y =outputsignal(scalar)
G =n xn state matrix
H =n x1 inputmatrix
C =1xn outputmatrix
D =constant(scalar)
Ifweusethestatefeedbackcontrolu = Kx.

JosSdaCosta DesignofControlSystemsinStateSpace 43
2.3RegulatordesignbyPolePlacementDT

whereK isthestatefeedbackgainmatrix(a1xnmatrix).
Thenthesystembecomesaclosedloopcontrolsystemanditsstate
equationbecomes
x(k 1) (G HK )x(k ) (20)
NotethatwechoosematrixK suchthattheeigenvaluesof
G HK arethedesiredclosedlooppoles,1,2,...,n.

2.3.1NecessaryandSufficientConditionforArbitraryPole
Placement
Likeinthecontinuoustimecase,inthediscretetimecasetheneces
sary andsufficientconditionforarbitrarypoleplacementisthatthe
systembecompletelystatecontrollable.
Theprooffollowsthesamestepsthanthecontinuoustimecase.

JosSdaCosta DesignofControlSystemsinStateSpace 44
2.3RegulatordesignbyPolePlacementDT
2.3.1NecessaryandSufficientConditionforArbitraryPole
Placement
Havinginmindthatthetransformationmatrix T = M W whereM is
thecontrollabilitymatrixgivenby
Then,thestatefeedbackgainmatrixK willbegivenby
(21)
whereais resultsfromthecharacteristicequation

andisfromthecharacteristicequationwiththedesiredeigenvalues,
i.e.

AllmethodscontinuoustimemethodstocalculateK havetheir
counterpartinthediscretetimecase.
JosSdaCosta DesignofControlSystemsinStateSpace 45
2.3RegulatordesignbyPolePlacementDT
2.3.2Determinationofthegainmatrix
Oncethedesiredcharacteristicequationisselected,thereareseveral
differentwaystodeterminethecorrespondingstatefeedbackgain
matrixK forthesystemdefinedbyeq.(19)whichisassumedtobe
completelystatecontrollable.
Method1:TransformationMatrixT
Asshowninthepreceding,matrixK canbegivenbyeq.(21)

(22)
wheretheai's arethecoefficientsoftheoriginalsystemcharacteristic
equation

andthei'sarethecoefficientsofthedesiredcharacteristicequation
forthestatefeedbackcontrolsystem;thatis,

JosSdaCosta DesignofControlSystemsinStateSpace 46
2.3RegulatordesignbyPolePlacementDT
DiscreteTimePolePlacementDesign
Method1:TransformationMatrixT (cont.)
ThetransformationmatrixT isgivenbyT=MW whereM andW are
given,respectively,bythediscretecounterpartofeq.(6)andeq.(7).
Ifthesystemstateequationisalreadyinthecontrollablecanonical
form,thedeterminationofthestatefeedbackgainmatrixK canbe
madesimple,becausethetransformationmatrixTbecomesthe
identitymatrix.
InthiscasethedesiredmatrixK isobtainedbysubstitutingT = MW =
I intoeq.(22).

JosSdaCosta DesignofControlSystemsinStateSpace 47
2.3RegulatordesignbyPolePlacementDT
Method2:DirectSubstitutionMethod
Iftheordern ofthesystemislow,substitute into
thecharacteristicequation

andthenmatchthecoefficientsofpowersinz ofthischaracteristic
equationwithequalpowersinz ofthedesiredcharacteristicequation

Method3:Ackermann'sFormula
ThedesiredstatefeedbackgainmatrixK canbegivenbyAckermann's
formula:
(23)
where

JosSdaCosta DesignofControlSystemsinStateSpace 48
2.3RegulatordesignbyPolePlacementDT

Method4:Diagonalization Method
Ifthedesiredeigenvalues 1,2,...,n aredistinct,thenthedesired
statefeedbackgainmatrixK canbegivenasfollows:
(24)

wherevectors1, 2,..., n satisfytheequation

Notethatthei 'sareeigenvectorsofmatrixG HK;thatis,i satis


fies theequation

JosSdaCosta DesignofControlSystemsinStateSpace 49
2.3RegulatordesignbyPolePlacementDT
2.3.3 Deadbeat Control
Indeadbeatcontrol,anynonzeroerrorvectorwillbedriventozeroin
atmostn samplingperiodsifthemagnitudeofthescalarcontrolu(k) is
unboundedwheren istheorderofthesystem.
Thedeadbeatresponseisachievedbyassigningalleigenvaluesofthe
closedlooptobezero.
Theconceptofdeadbeatresponseisuniquetodiscretetimecontrol
systems.Thereisnosuchthingasdeadbeatresponseincontinuous
timecontrolsystems.
Thesettlingtimedependsonthesamplingperiod,sincetheresponse
settlesdowninatmostn samplingperiods.
If the sampling period T is chosen very small, the settling time will also
be very small, which implies that the control signal must have an
extremely large magnitude. Otherwise, it will not be possible to bring
the error response to zero in a short time period.

JosSdaCosta DesignofControlSystemsinStateSpace 50
2.3RegulatordesignbyPolePlacementDT
Deadbeat Control (cont.)
Indeadbeatcontrol,thesamplingperiodistheonlydesignparameter.
Thus,ifthedeadbeatresponseisdesired,thedesignermustchoose
thesamplingperiodcarefullysothatanextremelylargecontrolmagni
tude isnotrequiredinnormaloperationofthesystem.
Notethatitisnotphysicallypossibletoincreasethemagnitudeofthe
controlsignalwithoutbound.
Ifthemagnitudeisincreasedsufficiently,thesaturationphenomenon
alwaystakesplace.Ifsaturationoccursinthemagnitudeofthecontrol
signal,thentheresponsecannolongerbedeadbeat.Thesettlingtime
willbemorethann samplingperiods.
Intheactualdesignofdeadbeatcontrolsystems,thedesignermustbe
awareofthetradeoffthatmustbemadebetweenthemagnitudeof
thecontrolsignalandtheresponsespeed.

JosSdaCosta DesignofControlSystemsinStateSpace 51
2.3RegulatordesignbyPolePlacementDT
Deadbeat design
Considerthecompletelystatecontrollablesystemdefinedby
(25)
Withstatefeedback ,thestateequationbecomes

Thesolutionofthislastequationisgivenby
(26)
Iftheeigenvaluesi ofmatrixG HK lieinsidetheunitcircle,then
thesystemisasymptoticallystable.
Inwhatfollows,weshallshowthat,bychoosingalleigenvaluesof
G HK tobezero,itispossibletogetthedeadbeatresponse,or

JosSdaCosta DesignofControlSystemsinStateSpace 52
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Indiscussingdeadbeatresponse,
thenilpotentmatrix

playsanimportantrole.
Consider,forexample,a4x4nilpotentmatrix:

Noticethat

JosSdaCosta DesignofControlSystemsinStateSpace 53
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Similarly,forannxnnilpotentmatrixN,wehave

Letuschoosethedesiredpolelocationstobeattheorigin,orchoose
thedesiredeigenvaluestobezero:1 = 2 = = n = 0.
Thenweshallshowthattheresponsetoanyinitialstatex(0)is
deadbeat.
Sincethecharacteristicequationwiththedesiredeigenvaluescanbe
givenby

weobtain

JosSdaCosta DesignofControlSystemsinStateSpace 54
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
andmatrixK givenbyeq.(21),canbesimplifiedtothefollowing:

(27)
ByusingthetransformationmatrixT,presentedbefore, givenby

where

JosSdaCosta DesignofControlSystemsinStateSpace 55
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Define

Definealso

Thenthestateequation(25)canbewritteninacontrollableformas

Ifweusethestatefeedback ,thenthis
lastequationbecomes

(28)

JosSdaCosta DesignofControlSystemsinStateSpace 56
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Referringtoeq.(28),wehavetheclosedloopstatematrix

JosSdaCosta DesignofControlSystemsinStateSpace 57
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Thus, -KT isanilpotentmatrix.Therefore,wehave

Intermsoftheoriginalstatex(k),wehavethesystemresponse

Thus,wehaveshownthatifthedesiredeigenvaluesareallzerosthen
anyinitialstatex(0) canbebroughttotheorigininatmostn sampling
periodsandtheresponseisdeadbeat,providedthecontrolsignalu(k)
isunbounded.

JosSdaCosta DesignofControlSystemsinStateSpace 58
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Example
Considerthesystemgivenby

(29)
DeterminethestatefeedbackgainmatrixK suchthatwhenthecontrol
signalisgivenbyu(k) = Kx(k) theclosedloopsystem(regulator
system)exhibitsthedeadbeatresponsetoaninitialstatex(0).
Assumethatthecontrolsignalu(k) isunbounded.
Resolution
Referringtoeq.(27),forthedeadbeatresponsewehave

(30)

JosSdaCosta DesignofControlSystemsinStateSpace 59
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Thesystemgivenbyeq.(29)isalreadyinthecontrollablecanonical
form.Therefore,T = I.
Thecharacteristicequationofthesystemgivenbyeq.(29)is

Thus,

Consequently,eq.(30)becomes

ThisgivesthedesiredstatefeedbackgainmatrixK tohavedeadbeat
response.

JosSdaCosta DesignofControlSystemsinStateSpace 60
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Letusverifythattheresponseofthissystemtoanarbitraryinitialstate
x(0) isindeedthedeadbeatresponse.
Sincetheclosedloopstateequationbecomes

Iftheinitialstateisgivenby

wherea andb arearbitraryconstants


JosSdaCosta DesignofControlSystemsinStateSpace 61
2.3RegulatordesignbyPolePlacementDT
Deadbeat design (cont.)
Thenwehave

Thus,thestatex(k) fork = 2, 3, 4, becomeszeroandtheresponseis


indeeddeadbeat.

JosSdaCosta DesignofControlSystemsinStateSpace 62
DesignofControlSystemsinStateSpace
Summary

2.4SolvingpoleplacingwithMATLAB
2.4.1Continuoustimesystems
2.4.2Discretetimesystems
2.4.3Deadbeatdesign

JosSdaCosta DesignofControlSystemsinStateSpace 63
2.4SolvingpoleplacingwithMATLAB
Poleplacement problems can be solved easily with MATLAB for the
continuous and discrete time cases for both the SISO and MIMO cases.
The commands for the SISO and MIMO cases will be explain as well
but the design for MIMO systems itself will be dealt latter since for a
specified set of closedloop poles the statefeedback gain matrix K is
not unique and we have an additional freedom (or freedoms) to
choose K.
There are many approaches to constructively utilize this additional
freedom to determine K but most of them will require knowledge of
MIMO analysis in the frequency domain that we will address later.
For example, one common choice is to maximize the stability margin.
The pole placement based on this approach is called the robust pole
placement.
This technique can be applied to both SISO and MIMO systems but in
the MIMO case drives the pole placement design to have a unique
solution.
JosSdaCosta DesignofControlSystemsinStateSpace 64
2.4SolvingpoleplacingwithMATLAB
2.4.1 Continuous time case
MATLABhastwocommands acker andplace forthecomputation
offeedbackgainmatrixK.
Thecommandacker isbasedonAckermann'sformula:
Thiscommandappliestosingleinputsystemsonly.
Thedesiredclosedlooppolescanincludemultiplepoles(poleslocatedat
thesameplace).
Thecommandplace maximizethestabilitymargin(robustpoleplace
ment):
Thiscommandappliestosingleandmultipleinputsystems.
Requiresthatthemultiplicityofpolesinthedesiredclosedlooppolesbeno
greaterthantherankofB.
Requiresthattherebenomultiplepolesinthesetofdesiredclosedloop
poles.
Forsingleinputsystems,thecommandsacker andplace yieldthesameK.
JosSdaCosta DesignofControlSystemsinStateSpace 65
2.4SolvingpoleplacingwithMATLAB
Continuous time case (cont.)
Notedthatwhenthesingleinputsystemisbarelycontrollable,some
computationalproblemmayoccurifthecommandackerisused.
Insuchacasetheuseoftheplacecommandispreferred,provided
thatnomultiplepolesareinvolvedinthedesiredsetofclosedloop
poles.

Useofcommandacker orplace
Tousethesecommandsweneedfirsttointroducethematricesinthe
program:
A matrix, B matrix, J matrix
whereJ matrixisthematrixconsistingofthedesiredclosedloop
polessuchthat

JosSdaCosta DesignofControlSystemsinStateSpace 66
2.4SolvingpoleplacingwithMATLAB
Continuous time case (cont.)
Thecommandsyntaxare
or
Notethatwecancrosschecktheresultwiththecommand
eig(AB*K)
toverifythatK thusobtainedgivesthedesiredeigenvalues.

Example
Considertheregulatorsystemtreatedinthepreviousexample.
Thesystemequationsis
= Ax + Bu
where

JosSdaCosta DesignofControlSystemsinStateSpace 67
2.4SolvingpoleplacingwithMATLAB
Example(cont.)
Byusingstatefeedbackcontrol u = Kx ,itisdesiredtohavethe
closedlooppolesats = i, (i = 1, 2, 3) ,where
1 = 2 + j4, 2 = 2 j4, 3 = 10
DeterminethestatefeedbackgainmatrixK withMATLABandobtain
theresponsetotheinitialcondition: xT (0) 1 0 0 T

Resolution
Commandacker Commandplace

JosSdaCosta DesignofControlSystemsinStateSpace 68
2.4SolvingpoleplacingwithMATLAB
Example(cont.)
Toobtaintheclosedloopresponsetothegiveninitialcondition
x(0),wesubstituteu = Kx intotheplantequationtoget

T
= (A BK)x , x T
(0) 1 0 0
Toplottheresponsecurves(x1 versust,x2 versust,andx3 versust),we
mayusethecommandinitial.
Wefirstdefinethestatespaceequationsforthesystemasfollows:
= (A BK)x + Iu
y = Ix + Iu
whereweincludeu (athreedimensionalinputvector).
Thisu vectorisconsidered0 inthecomputationoftheresponsetothe
initialcondition.
Thenwedefine

JosSdaCosta DesignofControlSystemsinStateSpace 69
2.4SolvingpoleplacingwithMATLAB
Example(cont.)
andusetheinitialcommandasfollows

wheret isthetimedurationwewanttouse,suchas

Thenobtainx1,x2,andx3 asfollows:

andusetheplotcommandtoobtaintheresultingresponsecurves.

JosSdaCosta DesignofControlSystemsinStateSpace 70
2.4SolvingpoleplacingwithMATLAB
Example(cont.)
Theprogram
inMATLABis

JosSdaCosta DesignofControlSystemsinStateSpace 71
2.4SolvingpoleplacingwithMATLAB
Example(cont.)
Theresultingresponsecurvesare

JosSdaCosta DesignofControlSystemsinStateSpace 72
2.4SolvingpoleplacingwithMATLAB
2.4.2 Discrete time case
Considerthesystem

where

Notethat

Hence

DetermineasuitablestatefeedbackgainmatrixK suchthatthesystem
hastheclosedlooppolesat

JosSdaCosta DesignofControlSystemsinStateSpace 73
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

Letusfirstexaminetherankofthecontrollabilitymatrix.Therankof

is2.Thusthesystemiscompletelystatecontrollable,andtherefore
arbitrarypoleplacementispossible.
Thecharacteristicequationforthedesiredsystemis

Hence

Method1:TransformationmatrixT

JosSdaCosta DesignofControlSystemsinStateSpace 74
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

Noticethestateequationisalreadyinthecontrollableform,and
thereforethetransformationmatrixT becomesI:

Hence

Method2:Ackermannsformula

JosSdaCosta DesignofControlSystemsinStateSpace 75
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

where

Thus

JosSdaCosta DesignofControlSystemsinStateSpace 76
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 77
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 78
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 79
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 80
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 81
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 82
2.4SolvingpoleplacingwithMATLAB
Discrete time case (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 83
2.4SolvingpoleplacingwithMATLAB
2.4.3 Example of deadbeat control
Considerthesystemgivenby

DeterminethestatefeedbackgainmatrixK suchthatwhenthecontrol
signalisgivenby u = Kx theclosedloopsystem(regulatorsystem)
exhibitsthedeadbeatresponsetoaninitialstatex(0).
Assumethatthecontrolsignalu(k)isunbounded.
Resolution
Forthedeadbeatresponse,thedesiredcharacteristicequationis

where1 = 0 and2 = 0.
JosSdaCosta DesignofControlSystemsinStateSpace 84
2.4SolvingpoleplacingwithMATLAB
Example of deadbeat control (cont.)
Forthedeadbeatresponse,thestatefeedbackgainmatrixK canbe
givenby

where

AMATLABprogramforthedeterminationofstatefeedbackgainmatrix
K isgivenby

JosSdaCosta DesignofControlSystemsinStateSpace 85
2.4SolvingpoleplacingwithMATLAB
Example of deadbeat control (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 86
2.4SolvingpoleplacingwithMATLAB
Example of deadbeat control (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 87
2.4PoleplacementProblems
Problem 1
Considerthesystem d3y d2y dy
3
+ 5 2
+ 3 + 2y = u
dt dt dt

DeterminethestatefeedbackgainmatrixK tohavearapidresponsewitha
lowovershootfortheclosedloopsystem.

Problema 2
1
Considerthesystem G (s) =
s2

Determinethestatefeedbackgain,usingtheAckermannsformula,toplace
theclosedlooppolesats =1 j.

JosSdaCosta DesignofControlSystemsinStateSpace 88
2.4PoleplacementProblems
Problem 3
Considerthediscretetimemodelofasystem

where

a)DeterminethestatefeedbackgainmatrixK suchthatthesystemwillexhibit
adeadbeatresponsetoanyinitialstate.
b)Assumingthatthestateiscompletelymeasurablesothattheactualstate
x(k)canbefedbackforcontrol,determinetheresponseofthesystemtothe
initialstate wherea,b,andc arearbitraryconstants.

JosSdaCosta DesignofControlSystemsinStateSpace 89
DesignofControlSystemsinStateSpace
Summary

2.5DesignofServoSystemsbyPolePlacement
2.5.1Continuoustimecase
2.5.1.1Plantwithintegrator
2.5.1.2Plantwithoutintegrator
2.5.2Discretetimecase

JosSdaCosta DesignofControlSystemsinStateSpace 90
2.5DesignofServoSystemsbyPolePlacementCT
2.5.1 ContinuousTime
Uptoknowweaddressthepoleplacementdesignforregulator
systems,i.e.systemwherethereferenceinputisalwayszeroora
nonzeroconstant.Thatis:
Openloop = Ax+Bu, y = Cx + Du, r=0
Controlaction: u = Kx
Closedloop: = (A BK)x, y = Cx + Du, r = 0

Designproblem:
FindthegainmatrixK
whichguaranteesthat
theclosedloopspoles
areinaprespecified
place.
JosSdaCosta DesignofControlSystemsinStateSpace 91
2.5.1DesignofServoSystemsbyPolePlacementCT
2.5.1.1DesignofServoSystemwhenthePlanthasanintegrator
Assumethattheplanthasanintegrator(type1plant)andisdefinedby
= Ax+Bu (u scalar) (31)
y = Cx (scalar) (32)
r = r(t) referenceinput(scalar)
Figureshowsageneralfeedbackconfigurationoftheservosystem
whentheplanthasan
integrator.
Hereweassumedthat
y = x1.
Inthepresentanalysis
weassumethatthere
ference inputr isastep
function.
JosSdaCosta DesignofControlSystemsinStateSpace 92
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
where

(33)

and

Assumethatthereferenceinput(stepfunction)isappliedatt = 0.
Then,fort > 0,thesystemdynamicscanbedescribedbyequations
(31)and(33),or
(34)
JosSdaCosta DesignofControlSystemsinStateSpace 93
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
Weshalldesigntheservosystemsuchthattheclosedlooppolesare
locatedatdesiredpositions.
Thedesignedsystemwillbeanasymptoticallystablesystem,y() will
approachtheconstantvaluer,andu() willapproachzero.
Noticethatatsteadystatewehave
(35)
Notingthatr(t) isastepinput,wehaver() = r(t) = (constant)fort > 0.
Bysubtractingeq.(35)fromeq.(34),weobtain
(36)
Define

JosSdaCosta DesignofControlSystemsinStateSpace 94
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
Theneq.(36)becomes
(37)
whichdescribestheerrordynamics.
Thedesignoftheservosystemhereisconvertedtothedesignofan
asymptoticallystableregulatorsystemsuchthate(t) approacheszero,
givenanyinitialconditione(0).
Ifthesystemdefinedbyeq.(37)iscompletelystatecontrollable,then,
byspecifyingthedesiredeigenvalues1 ,2,...,n forthematrix
A - BK,matrixK canbedeterminedbythepoleplacementtechnique
presentedbeforefortheregulatorsystem.
Thesteadystatevaluesofx(t) andu(t) canbefoundasfollows:
Atsteadystate(t = ),wehave,fromeq.(34),

JosSdaCosta DesignofControlSystemsinStateSpace 95
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
SincethedesiredeigenvaluesofA - BK areallinthelefthalfs plane,
theinverseofmatrixA - BK exists.
Consequently,x() canbedeterminedas

Also,u() canbeobtainedas

Example
Designaservosystemwhentheplanttransferfunctionhasan
integrator.Assumethattheplanttransferfunctionisgivenby

JosSdaCosta DesignofControlSystemsinStateSpace 96
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
Thedesiredclosedlooppolesare 2 2 3 and 10.
Assumethatthesystemconfigurationisthesameasthatshowninthe
previousFigureandthereferenceinputr isastepfunction.
Obtaintheunitstepresponseofthedesignedsystem.
Resolution
Definestatevariablesx1,x2,andx3 asfollows:
x1 =y, x2 =1 and x3 = 2
Thenthestatespacerepresentationofthesystembecomes
= Ax+Bu, y = Cx
where

JosSdaCosta DesignofControlSystemsinStateSpace 97
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
ReferringtoServosystemFigure(slide92)andnotingthatn = 3,the
controlsignalu isgivenby

where

ThestatefeedbackgainmatrixK canbeobtainedeasilywithMATLAB
withthefollowingprogram

JosSdaCosta DesignofControlSystemsinStateSpace 98
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
ThestatefeedbackgainmatrixK isthus

UnitStepResponseoftheDesignedSystem:
Theunitstepresponseofthedesignedsystemcanbeobtainedas
follows:
Since

fromeq.(34)thestateequationforthedesignedsystemis

(38)

JosSdaCosta DesignofControlSystemsinStateSpace 99
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
andtheoutputequationis

(39)

Solving Equations (38) and (39) for y(t) when r is a unitstep function
gives the unitstep response curve y(t) versus t.
Other way to obtain numerically the unitstep response curve is
through a MATLAB Program like:

JosSdaCosta DesignofControlSystemsinStateSpace 100


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)

Theresultingunitstepresponsecurveisshownbelow

JosSdaCosta DesignofControlSystemsinStateSpace 101


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)

JosSdaCosta DesignofControlSystemsinStateSpace 102


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has an integrator (cont.)
Notethatsince

wehave

Atsteadystatethecontrolsignalu becomeszero.

JosSdaCosta DesignofControlSystemsinStateSpace 103


2.5.1DesignofServoSystemsbyPolePlacementCT
2.5.1.2 Design of Servo System when the Plant has no integrator
Iftheplanthasnointegrator(type0plant),thebasicprincipleofthe
designofatype1servosystemistoinsertanintegratorinthe
feedforward pathbetweentheerrorcomparatorandtheplant,as
showninFigurebelow.

TheblockdiagramshownintheFigureaboveisabasicformofthe
type1servosystemwheretheplanthasnointegrator.

JosSdaCosta DesignofControlSystemsinStateSpace 104


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
From the diagram, we obtain
(40)
(41)
(42)
(43)
where isoutputoftheintegrator(statevariableofthesystem,scalar)
Weassumethattheplantgivenbyeq.(40)iscompletelystatecontrol
lable.
Thetransferfunctionoftheplantcanbegivenby
(44)

JosSdaCosta DesignofControlSystemsinStateSpace 105


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
Toavoidthepossibilityoftheinsertedintegratorbeingcanceledbythe
zeroattheoriginoftheplant,weassumethatGp(s) hasnozeroatthe
origin.
Assumethatthereferenceinput(stepfunction)isappliedatt = 0.
Then,fort > 0,thesystemdynamicscanbedescribedbyanequation
thatisacombinationofEquations(40)and(43):

(45)

Weshalldesignanasymptoticallystablesystemsuchthatx(), (),
andu() approachconstantvalues,respectively.
Then,atsteadystate, (t) = 0,andwegety() = r.

JosSdaCosta DesignofControlSystemsinStateSpace 106


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
Noticethatatsteadystatewehave

(46)

Notingthatr(t) isastepinput,wehaver() = r(t) = r (constant)fort > 0.


Bysubtractingeq.(46)fromeq.(45),weobtain

(47)
Define

JosSdaCosta DesignofControlSystemsinStateSpace 107


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
ThenEquation(47)canbewrittenas

(48)

where
(49)
Defineanew(n+1)thordererrorvectore(t) by

ThenEquation(48)becomes
(50)

JosSdaCosta DesignofControlSystemsinStateSpace 108


2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
where

andeq.(49)becomes
(51)
where

Thestateerrorequationcanbeobtainedbysubstitutingeq.(51)into
eq.(50):

Ifthedesiredeigenvaluesofmatrix - (thatis,thedesiredclosed
looppoles)arespecifiedas1 ,2,...,n+1 ,thenthestatefeedback
gainmatrixK andtheintegralgainconstantkI,canbedeterminedby
JosSdaCosta DesignofControlSystemsinStateSpace 109
2.5.1DesignofServoSystemsbyPolePlacementCT
Design of Servo System when the Plant has no integrator (cont.)
thepoleplacementtechniquepresentedfortheregulatorsystem,
providedthatthesystemdefinedbyEquation(50)iscompletelystate
controllable.

Notethatifthematrix

hasrankn+1,thenthesystemdefinedbyeq.(50)iscompletelystate
controllable.

JosSdaCosta DesignofControlSystemsinStateSpace 110


2.5.1DesignofServoSystemsbyPolePlacementCT
Example
Considertheinvertedpendulum
controlsystemshowninFigure.
Weareconcernedonlywiththemotion
ofthependulumandmotionofthecart
intheplaneofthepage.
Itisdesiredtokeeptheinvertedpendu
lumuprightasmuchaspossibleandyet
controlthepositionofthecart,forinstan
ce,movethecartinastepfashion.

Tocontrolthepositionofthecart,weneedtobuildatype1servo
system.
Theinvertedpendulumsystemmountedonacartdoesnothavean
integrator.

JosSdaCosta DesignofControlSystemsinStateSpace 111


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Therefore,wefeedthepositionsignaly (whichindicatesthepositionof
thecart)backtotheinputandinsertanintegratorinthefeedforward
path,asshowninFigurebelow.

JosSdaCosta DesignofControlSystemsinStateSpace 112


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Thelineartimeinvariant equationsfortheinvertedpendulumsystem
are

If

Thenequationsoftheinvertedpendulumsystembecome
(52)
(53)
Letusdefinethestatevariables

JosSdaCosta DesignofControlSystemsinStateSpace 113


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Then,referringtoeq.s (52)and(53)andthepreviousFigureand
consideringthecartpositionx astheoutputofthesystem,weobtain
theequationsforthesystemasfollows:

where

JosSdaCosta DesignofControlSystemsinStateSpace 114


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Forthetype1servosystem,wehavethestateerrorequationasgiven
byeq.(50):
(54)
where

andthecontrolsignalisgivenbyeq.(51)

where

JosSdaCosta DesignofControlSystemsinStateSpace 115


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Toobtainareasonablespeedanddampingintheresponseofthe
designedsystem(forexample,thesettlingtimeofapproximately4 ~ 5
secandthemaximumovershootof15% ~ 16% inthestepresponseof
thecart),letuschoosethedesiredclosedlooppolesats = i

Beforeweproceedfurther,wemustexaminetherankofmatrixP,
where

Therankofthismatrixcanbefoundtobe5.

JosSdaCosta DesignofControlSystemsinStateSpace 116


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Therefore,thesystemdefinedbyEquation(54)iscompletelystate
controllable,andarbitrarypoleplacementispossible.
UsingMATLABwecomputethestatefeedbackgainmatrixK,i.e.

JosSdaCosta DesignofControlSystemsinStateSpace 117


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Thatis

and

UnitStepResponseCharacteristicsoftheDesignedSystem:
Thestepresponseinthecartpositioncanbeobtainedbysolvingthe
followingequation:

Theoutputofthesystemisx3(t),or

JosSdaCosta DesignofControlSystemsinStateSpace 118


2.5DesignofServoSystemsbyPolePlacementCT
Example (cont.)
ByusingMATLABweobtainthestepresponsecurvesofthedesigned
system

JosSdaCosta DesignofControlSystemsinStateSpace 119


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)

Noticethat,toobtaintheunitstepresponse,weenteredthe
command

JosSdaCosta DesignofControlSystemsinStateSpace 120


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
where

JosSdaCosta DesignofControlSystemsinStateSpace 121


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
Noticethat(t) [= x3(t)] hasapproximately15%overshootandthe
settlingtimeisapproximately4.5sec.(t) [= x5(t)] approaches1.1.
Thisresultcanbederivedasfollows:
Since
or

weget
Sinceu() = 0,wehave,

JosSdaCosta DesignofControlSystemsinStateSpace 122


2.5.1DesignofServoSystemsbyPolePlacementCT
Example (cont.)
andso

Hence,forr = 1,wehave

Itisnotedthat,asinanydesignproblem,ifthespeedanddampingare
notquitesatisfactory,thenwemustmodifythedesiredcharacteristic
equationanddetermineanewmatrix .
Computersimulationsmustberepeateduntilasatisfactoryresultis
obtained.
JosSdaCosta DesignofControlSystemsinStateSpace 123
2.5.2DesignofServoSystemsbyPolePlacementDT
2.5.2 DiscreteTime case
Like the case of design of discrete regulator systems by pole placement
the design of discrete servo systems follows the steps of the
continuoustime case design.
In the servo system it is generally required that the system have
one or more integrators within the closed loop. Unless the plant to be
controlled has an integrating property, it is necessary to add one or
more integrators within the loop to eliminate steadystate error to
step inputs.
Like in the continuoustime case one way to introduce an integrator in
the mathematical model of a closedloop system is to introduce a new
state vector that integrates the difference between the command
vector r and the output vector y.
The integrator can be included as part of the pole placement
formulation that was presented before.

JosSdaCosta DesignofControlSystemsinStateSpace 124


2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
ConsidertheservosystemshownintheFigurebelow,wherethe plantis
assumedtobecompletelystatecontrollableandcompletelyobservable
andthattheplantdoesnothaveanintegrator.

Theplantstateequationandoutputequationare
x(k+1) = Gx(k) +Hu(k) (55)
y(k) = Cx(k) (56)
JosSdaCosta DesignofControlSystemsinStateSpace 125
2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
Theintegratorstateequationis
(57)
where

Eq.(57)canberewrittenasfollows:

(58)
Thecontrolu(k) isgivenby
(59)
OurdesignparametersarematricesK1 andK2.
JosSdaCosta DesignofControlSystemsinStateSpace 126
2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
ProcedurefordeterminingmatricesK1 andK2
FromEquations(55),(58),and(59),weobtain

(60)
Notingthatu(k) isalinearcombinationofstatevectorsx(k) andv(k),
defineanewstatevectorconsistingofx(k) andu(k),ratherthanx(k)
andv(k).Thenweobtainfromeqs.(55)and(60)thefollowingstate
eq:

(61)

JosSdaCosta DesignofControlSystemsinStateSpace 127


2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
ProcedurefordeterminingmatricesK1 andK2 (cont.)
Theoutputequation,eq.(56),canbewrittenasfollows:
(62)

Toapplythepoleplacementtechniquedirectlytothedesignofthe
presentservosystem,considerthecasewherethecommandvector
r(k) isaconstantvector(stepinput)atk = 0,sothat

Theneq.(61)becomes

(63)

JosSdaCosta DesignofControlSystemsinStateSpace 128


2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
Fromeq.(57),forthestepinput,weobtain:

or
Thereisnosteadystateerrorintheoutputwhenthecommandinputis
astepvector.
Also,atsteadystateeq.(63)becomes

(64)

Letusdefinetheerrorvectorsby

JosSdaCosta DesignofControlSystemsinStateSpace 129


2.5.2DesignofServoSystemsbyPolePlacementDT
ProcedurefordeterminingmatricesK1 andK2 (cont.)
Then,subtractingeq.(64)fromeq.(63),weobtain

(65)
Thedynamicsofthesystemaredeterminedbytheeigenvaluesofthe
statematrixappearingineq.(65).
Eq.(65)canbemodifiedtoread

(66)
where

(67)

JosSdaCosta DesignofControlSystemsinStateSpace 130


2.5.2DesignofServoSystemsbyPolePlacementDT
ProcedurefordeterminingmatricesK1 andK2 (cont.)
Ifwedefine

(68)
theneqs.(66)and(67)become,respectively,
(69)
and
(70)
JosSdaCosta DesignofControlSystemsinStateSpace 131
2.5.2DesignofServoSystemsbyPolePlacementDT
ProcedurefordeterminingmatricesK1 andK2 (cont.)
Thecontrollabilitymatrixcanbewrittenasfollows:

(71)
whichisrankn + m, sincetheplant(55)iscompletelystatecontrol
lable.Thus,thesystem(69)iscompletelystatecontrollable.Therefore,
thepoleplacementtechniqueappliestothiscase.
Oncethedesiredclosedlooppolesarespecified,matrixK canbe
determinedbythepoleplacementtechnique.
UsingmatrixK thusdetermined,wecanobtainmatricesK1 andK2 as
follows.First,notethat

(72)

JosSdaCosta DesignofControlSystemsinStateSpace 132


2.5.2DesignofServoSystemsbyPolePlacementDT
Discrete Time case (cont.)
Then,fromEquations(68)and(72),wehave

Hence,weobtain

(73)

ThedesiredgainmatricesK1 andK2 maybedeterminedfromeq.


(73).
NotethatintheMIMOcasethematrixK isnotunique.

JosSdaCosta DesignofControlSystemsinStateSpace 133


2.5.2DesignofServoSystemsbyPolePlacementDT
Example
Considerthedigitalcontrolofaplantbyuseofstatefeedbackand
integralcontrol.Assumethatthesystemconfigurationisthesameas
thatshowninFigure.

Assumealsothatthepulse
transferfunctionofthe
plantis

DetermineanintegralgainconstantK1,andastatefeedbackgain
matrixK2,suchthattheresponsetoaunitstepcommandinputis
deadbeat.

JosSdaCosta DesignofControlSystemsinStateSpace 134


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Resolution
Weshallfirstobtainastatespacerepresentationfortheplantpulsetransfer
function.Bycomparingthegivenpulsetransferfunctionwiththestandard
form

we find

Then,wecanobtainthefollowingstatespaceequations forthe plant:

x(k+1) = Gx(k) +Hu(k) (74)


y(k) = Cx(k) (75)

JosSdaCosta DesignofControlSystemsinStateSpace 135


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
where

Notethatthisplantiscompletelystatecontrollableandcompletely
observable.Thereforepoleplacementapplies.
WeshallnowdeterminetheintegralgainconstantK1 andthestatefeedback
gainmatrixK2.
Inthepresentsystemwerequiretheresponsetothestepcommandinputto
bedeadbeat.Thus,wemustplacetheclosedlooppolesofthesystem at the
origin.
Referringtoeqs.(69)and(70),wehave

JosSdaCosta DesignofControlSystemsinStateSpace 136


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
where

and

OurproblemhereistodeterminematrixK sothattheclosedlooppolesof
thesystemareattheorigin,orthedesiredcharacteristicequationis

Byusingthepoleplacementtechnique,K matrix canbedeterminedeasily.


ReferringtoAckermann'sformula,weobtain

where

JosSdaCosta DesignofControlSystemsinStateSpace 137


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Thus,thegainmatrixK isgivenby

(76)

ThedesiredintegralgainconstantK1,andthestatefeedbackgainmatrixK2,
areobtainedfromeq.(73).

JosSdaCosta DesignofControlSystemsinStateSpace 138


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Notingthat

isnonsingularweobtain

JosSdaCosta DesignofControlSystemsinStateSpace 139


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)

(77)

Fromeq.(77)weobtaintheintegralgainconstantK1

(78)

ThestatefeedbackgainmatrixK2,isgivenby

(79)

JosSdaCosta DesignofControlSystemsinStateSpace 140


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Next,letusdeterminetheoutputy(k).Fromeq.(75)wehave

Toobtainoutputy(k),weshallfirstdeterminethestatevectorx(k) and
signalv(k).FromthepreviousFigure,wehave

(80)
(81)
(82)
(83)

JosSdaCosta DesignofControlSystemsinStateSpace 141


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Hence,fromeqs.(80)and(83)weobtain

(84)
Also,fromeqs.(81),(82),and(84)weget

(85)

JosSdaCosta DesignofControlSystemsinStateSpace 142


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Combiningeqs.(84)and(85),weget

(86)
whichcanberewrittenas

(87)
Sincethecommandinputr(k) isaunitstepinput,wehave

JosSdaCosta DesignofControlSystemsinStateSpace 143


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Letusassumethattheinitialstateis

JosSdaCosta DesignofControlSystemsinStateSpace 144


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)

Similarly,

and

Theoutputy(k) isobtainedasfollows:

JosSdaCosta DesignofControlSystemsinStateSpace 145


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Similarly,

Deadbeatbehaviour
Noticethat

wherek = 4, 5, 6... Sinceu(t) fort 4T (whereT isthesampling


period)isconstant,thereisnointersampling oscillationintheoutput.
Thus,theresponseofthesystemisdeadbeat.

JosSdaCosta DesignofControlSystemsinStateSpace 146


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Note that the output y(k) reaches unity in at most four sampling
periods and will stay there in the absence of disturbances or new
command inputs.
See for example the sample unitstep response sequence shown in
next Figure.

JosSdaCosta DesignofControlSystemsinStateSpace 147


2.5.2DesignofServoSystemsbyPolePlacementDT
Example (cont.)
Underspecialinitialconditions,forexample,a = b = c = 0 andd = 1,
theoutputreachesunityinthreesamplingperiodsandstaysthere,or
y(k) = 1 fork = 3, 4, 5, ...

JosSdaCosta DesignofControlSystemsinStateSpace 148


2.6PolePlacementinMIMOSystems
Summmary
Tobeaddlatter

JosSdaCosta DesignofControlSystemsinStateSpace 149


2.6PolePlacementinMIMOSystems
Pole Placement When the Control Signal Is a Vector
Thusfar,wehaveconsideredthepoleplacementdesignproblemwhen
thecontrolsignalisascalar.Ifthecontrolsignalisavectorquantity(r
vector),theresponsecanbespeededup,becausewehavemore
freedomtochoosecontrolsignalsu1(k),u2(k),..,,u,(k)tospeedupthe
response.Forexample,inthecaseofthenthordersystemwitha
scalarcontrol,thedeadbeatresponsecanbeachievedinatmost11
samplingperiods.Inthecaseofthevectorcontrolu(k),thedeadbeat
responsecanbeachievedinlessthannsamplingperiods.Inthecase
ofthevectorcontrol,however,thedeterminationofthestatefeedback
gainmatrixKbecomesmorecomplex.Weshallpresentsuchacasein
AppendixC.

JosSdaCosta DesignofControlSystemsinStateSpace 150


2.6PolePlacementinMIMOSystems
Summary
Tobeaddlatter

JosSdaCosta DesignofControlSystemsinStateSpace 151

Potrebbero piacerti anche