Robust Loop Shaping
ExamensarbeteutförtiReglerteknik
vid TekniskaHögskolaniLinköping
av
HenrikJohansson
Robust Loop Shaping
ExamensarbeteutförtiReglerteknik
vid TekniskaHögskolaniLinköping
av
HenrikJohansson
Regnr: LiTH-ISY-EX-3291-2002
Supervisor: Martin Enqvist
Henrik Jonson
Division,Department Date Språk Language
2
Svenska/Swedish2
Engelska/English2
Rapporttyp Reportcategory2
Licentiatavhandling2
Examensarbete2
C-uppsats2
D-uppsats2
Övrigrapport2
URLförelektroniskversion
ISBN
ISRN
Serietitelochserienummer
Titleofseries,numbering
ISSN Titel Title Författare Author Sammanfattning Abstract Nyckelord
Robustcontrol designhasbecomeamajor researchareaduringthe
lasttwentyyearsandtherearenowadaysseveralrobustdesignmethods
available. One exampleof such amethodis therobust loopshaping
methodthatwasdevelopedbyGloverandMacFarlaneinthelate1980s.
Theideaofthismethodistousedecentralizedcontrollerdesigntogive
thesingularvaluesoftheloopgainadesiredshape. Thisstepiscalled
LoopShapingand itis followedby aRobustStabilization procedure,
whichaimstogivetheclosedloopsystemamaximumdegreeofstability
margins. In this thesis, the robust loop shaping method is used to
designagainscheduledcontrollerforamissile. Thereportconsistsof
threeparts,where therstpart introducesthe RobustLoopShaping
controllerdesigntheoryandaGainSchedulingapproach. Thesecond
part discusses the missile and itscharacteristics. In the thirdpart a
controller is designed and ashort analysis of the closed loop system
is performed. A scheduled controller is implemented in a nonlinear
environment,inwhichperformanceandrobustnessaretested. Robust
LoopShaping is easy to use and simulations show that the resulting
controllerisabletocopewithmodelperturbationswithoutconsiderable
loss in performance. The missile should to be able to operate in a
large speedinterval. There, it is shown that a single controller does
notstabilize the missileeverywhere. Thegain scheduledcontrolleris
howeverableto doso,whichhasbeenshownbymeansofsimulations
inthenonlinearenvironment. AutomaticControl,
Dept.ofElectricalEngineering
10thJanuary2003 LITH-ISY-EX-3291-2002 http://www.control.isy.liu.se
GainScheduledMissileControlUsingRobustLoopShaping Parameterstyrdmissilstyrningmedhjälpavrobustkretsformning
HenrikJohansson
×
×
Robust control design has become amajor research area during the last twenty
yearsandtherearenowadaysseveralrobustdesignmethodsavailable. Oneexample
ofsuchamethodistherobustloopshapingmethodthatwasdevelopedbyGlover
andMacFarlanein thelate1980s. Theideaofthismethodisto usedecentralized
controllerdesignto givethesingularvaluesoftheloopgainadesiredshape. This
stepiscalledLoopShapinganditisfollowedbyaRobustStabilizationprocedure,
whichaimstogivetheclosedloopsystemamaximumdegreeofstabilitymargins.
Inthis thesis,therobustloop shapingmethod isused to designagainscheduled
controller for a missile. The report consists of three parts, where the rst part
introducestheRobustLoopShapingcontrollerdesigntheoryandaGainScheduling
approach. The second part discusses the missile and its characteristics. In the
third partacontroller is designedand ashort analysis of the closedloopsystem
is performed. A scheduledcontroller is implemented in anonlinearenvironment,
in which performance androbustnessaretested. RobustLoopShaping iseasyto
useand simulationsshowthat theresultingcontroller isableto copewithmodel
perturbationswithoutconsiderablelossinperformance. Themissile should tobe
abletooperateinalargespeedinterval. There,itisshownthatasinglecontroller
doesnotstabilizethemissileeverywhere. Thegainscheduledcontrollerishowever
able to do so, which has been shown by means of simulations in the nonlinear
environment.
ThisworkhasbeencarriedoutatSaabBoforsDynamicsABinLinköping,Sweden.
IwouldliketothankmytosupervisorsHenrikJonsonandMartinEnqvistfortheir
helpandsupportduringthisproject. IwouldalsoliketothankmyexaminerTorkel
Gladforshowinginterestinmywork. Finally,Iwouldliketothankallthepeople
atSaabBoforsDynamics ABformakingmefeellikeapartofthegroup.
Linköping,December2002
Symbols
u ≡ u(t)
inputvectorδ
a
aileron(u
1
)δ
e
elevator(u
2
)δ
r
rudder (u
3
)x ≡ x(t)
statevector˙x
thetimederivativeofx
ˆx
theestimateofx
y ≡ y(t)
outputvectorG
(s)
transferfunctionω
theangularvelocityvectorp
angularvelocityaroundthex
axisq
angularvelocityaroundthey
axisr
angularvelocityaroundthez
axisα
angleofattackβ
sideslipangleV
themissilevelocityvectoru
velocityinthedirectionofthex
axisv
velocityinthedirectionofthey
axisw
velocityinthedirectionofthez
axisV
theforwardspeedofthemissil(|V |
)I
massmomentofinertiaortheunitarymatrixI
x
thex
partofI
I
y
they
partofI
I
z
thez
partofI
Operators and functionsA
B
C
D
shorthandforthestatespace
realizationof
G
(s) = C(sI − A)
A
−1
inverseA
†
pseudoinverseA
∗−1
shorthandfor(A
−1
)
∗
A.
∗ B
elementwisemultiplicationofA
andB
λ
(A)
eigenvalueofA
|β|
absolutevalueofβ
1 Introduction 1
1.1 Background . . . 1
1.2 Objectives. . . 1
1.3 Limitations . . . 2
1.4 ThesisOutline . . . 2
2 The
H
∞
Loop ShapingDesign 3 2.1 BasicRequirementsfortheClosedLoop . . . 32.2 LoopShaping . . . 4
2.2.1 TheScalingProcedure . . . 4
2.2.2 Pairingof theInputsandOutputs . . . 5
2.2.3 DecentralizedControllerDesign. . . 6
2.2.4 AlignmentofSingular Values . . . 6
2.2.5 SummaryoftheLoopShapingProcedure . . . 7
2.3 RobustStabilization . . . 7
2.3.1 NormalizedLeftCoprimeFactorizations . . . 8
2.3.2 TheStabilizingController
K
∞
(s)
. . . 92.4 GainScheduling . . . 10
2.4.1 SchedulingonEitherOneorTwoParameters . . . 11
3 The MissileModel 13 3.1 TheMissile . . . 14
3.1.1 TheGuidanceSystemoftheMissile . . . 14
3.1.2 AssumptionsandLimitations . . . 15
3.2 TheMainDynamicsoftheMissile . . . 18
3.2.1 RigidBodyDynamics . . . 18
3.2.2 TheTranslationof theMissile. . . 19
3.2.3 TheRotationoftheMissile . . . 20
3.2.4 SummaryoftheMainDynamicsoftheMissile . . . 21
3.3 ThePlant . . . 21
3.3.1 Choosing theDesignParameters . . . 23
3.3.2 TheNewton-RaphsonMethod . . . 23
4 Controller Designfor the BasicPlant 29
4.1 ThePlantUsedfortheControllerDesign . . . 29
4.2 LoopShapingandRobustStabilizationfortheBasicPlant . . . 30
4.2.1 TheScalingProcedure . . . 30
4.2.2 PairingtheInputsandOutputs. . . 31
4.2.3 DecentralizedControllerDesign. . . 33
4.2.4 AlignmentofSingular Values . . . 33
4.2.5 RobustStabilization . . . 34
5 Linear Analysis 37 5.1 TheClosed LoopSystem. . . 37
5.2 ThePerformanceoftheClosedLoopSystem . . . 38
5.3 RobustnessoftheClosedLoopSystem . . . 39
6 Nonlinear Evaluation 43 6.1 Simulations . . . 43
6.2 TheControllerDesignBasedontheBasic Model . . . 44
6.2.1 TransformationoftheReferenceSignals . . . 44
6.2.2 SimulationswithaSingleController . . . 45
6.2.3 SchedulingontheForwardSpeed. . . 47
6.3 TheControllerDesignBasedontheExtendedModel. . . 49
6.3.1 SchedulingontheAngleofAttack . . . 49
6.4 Basicvs. ExtendedModel . . . 50
7 Conclusions 51 7.1 Results. . . 51
7.2 FutureWork . . . 52
Appendices
A The SystemMatrices ofthe Nominal Plant 53
B The SystemMatrices ofthe Controller 55
C The Matlab Script Used for the ControllerDesign 57
Introduction
Thisreportisapresentationofaprojectperformedat SaabBoforsDynamicsAB
in Linköping, Sweden. For the last twenty years Saab Bofors Dynamics AB has
successfullybeendevelopingmissilesystems,wheretheSwedishdefenseisthemain
customer. Advancedautomaticcontrolisanimportantpartofthemissilesystems
of today. Hence, it is very important for Saab to stay updated in the area of
guidanceandcontrol.
1.1 Background
Robust control design has become amajor research area during the last twenty
yearsandtherearenowadaysseveralrobustdesignmethodsavailable. Oneexample
of such a method is the robust loop shaping method that was developed by K.
GloverandD. C.MacFarlaneinthelate1980s. Theideaofthismethodisto use
decentralizedcontrollerdesigntogivethesingularvaluesoftheloopgainadesired
shape. ThisstepiscalledLoopShapinganditisfollowedbyaRobustStabilization
procedure,whichaimstogivetheclosedloopsystemamaximumdegreeofstability
margins. Thismethod wasused in[8]byR. A.HydeandK.Gloverinanaircraft
application.
1.2 Objectives
The objectives of this project is to use the ideas of K. Glover and R. A. Hyde
[8] to design ascheduled
H
∞
LoopShaping controllerfor a missile application. Thecontrollerwassupposedtobeimplementedin anonlinearsimulationenviron-ment, suppliedby SaabBofors Dynamics. Furthermore, thecontroller capability
1.3 Limitations
In order to keep this projectwithin reasonablelimits there was aneed for some
restrictions. Themissileis supposedto workwellin alargeoperatingarea,which
canbedividedintospeedandaltitude. Alargeareameansthatseveralcontrollers
havetobedesigned. Inordertoreduce thenumberofdesigncases,thisprojectis
restrictedto designingandevaluatingcontrollersforaxedaltitude.
1.4 Thesis Outline
Thisreportcanbedividedintothree dierentparts. Therstpartdescribesthe
H
∞
Loop Shaping controller design and is found in Chapter 2. The controller design consists of a loop shaping procedure which species performance for theclosedloopsystem. Moreover,arobuststabilizationprocedure,whichisappliedin
orderto givetheclosed loopsystemsucientstabilitymargins, andanapproach
forgainschedulingisdescribed. Anaccuratesystemmodelisthecornerstoneofa
successfulcontroller. Moreover,sincegainschedulingis includedin thecontroller
design it is important to be able to derive several linear plant descriptions. A
methodthatmakesthiseasyisdescribedinChapter3. Thischaptercanbeviewed
asthesecondpartofthereport. Thethirdpartofthethesiscontainsanevaluation
of the designed controller. In Chapter 4a short design exampleis given and in
Chapter5theperformanceandrobustnessoftheclosedloopsystemisdiscussed.
Thebest waytoevaluateacontrollerisofcoursetoletitcontroltherealsystem.
Since it, for obvious reasons, is impractical to implement the controller in areal
missile,thenextbest thingistoevaluatethescheduledlinear
H
∞
LoopShaping controller in anonlinearsimulationenvironment. This is described in Chapter 6The
H
∞
Loop Shaping DesignThecontroldesignmethodthatisusedinthismasterthesisis
H
∞
LoopShaping. This particular method was proposed by D. C. MacFarlane and K. Glover, andwasusedbyR.A.HydeandK.Gloverin[8]. Thedesignmethodconsistsofthree
parts, alldescribed in this chapter. A Loop Shaping techniqueis used as a rst
steptospecifyperformanceoftheclosedloopsystem. Theloopshapingprocedure
is followed by a Robust Stabilization procedure with the purpose of maximizing
the stability margins of the closed loop system. Finally, an approach to Gain
Scheduling suitablefor
H
∞
controllersispresented.2.1 Basic Requirements for the Closed Loop
For a multivariable (MIMO) system there are some basic requirements for the
ClosedLoopSystem
G
c
(s)
. These canbesummarizedasfollows.•
Thetransferfunction fromthereferenceto theoutputshouldbeclosetoI
, i.e.G
c
(s) ≈ I
.•
ThetransferfunctionfromreferencetoinputG
ru
(s)
shouldnotbetoolarge.•
The sensitivity functionS
(s)
should be small so that system disturbances andmodelperturbationshavelittleornoeect ontheoutput.•
Formeasurementdisturbancesto havelittleornoeect on theoutput, the complementary sensitivity functionT
(s)
should be small. Furthermore, it shouldbesmallsothatmodelperturbationsdonotaectthesystemstability.However,becausethese designobjectivesare usuallyconicting it isnotpossible
tofulllthemall. Furthermore,themissileisapoorlydecouplednonlinearsystem
controllers that give a good compromise between the desired transfer functions
mentionedearlierandthat handlepoorlydecoupledsystems.
2.2 Loop Shaping
In[8]R.A.HydeandK.GloverproposeaLoopShapingprocedureusedtospecify
performanceasarststepinthe
H
∞
LoopShapingcontrollerdesign. Thepurpose ofthisstepistogivethesingularvaluesoftheloopgainadesiredshape. Thiswillaftertherobuststabilizationprocedureleadtosuitablebehavioroftheclosedloop
system. Accordingto [1] thedesiredshapeof thesingularvaluesoftheloopgain
ishigh gainforlowerfrequenciesand lowgainforhigherfrequencies. Duringthe
transitionbetweenhighandlowgain,theloopgainshouldhaveadecreaseof
−1
. InthismasterthesisprojecttheLoopShapingprocedureconsistsoffoursteps,which arecarriedoutinthefollowingorder
1. TheScalingProcedure
⇒ D
u
andD
y
. 2. PairingoftheInputsandOutputs.3. DecentralizedControllerDesign
⇒ W
p
(s)
. 4. AlignmentofSingularValues⇒ W
a
.TheresultfromtheLoopShapingprocedureisashapedplant
G
s
(s)
according to Figure 2.1. The singular values of the shaped plant should havethe desiredshape.
D
1
y
−
Wa
W
p
(
s
)
Wf
Du
G
(
s
)
Figure 2.1. Theresultof theLoopShaping procedure isaloop gainthat has
charac-teristicssuitablefortherobuststabilizationprocedureandforthebehavioroftheclosed
loopsystem. Theuseofthe
W
f
matrixwillbeexplainedinChapter4.2.2.1 The Scaling Procedure
TherststepintheLoopShapingprocedureistoscaletheinputsandtheoutputs.
Thiscanbeachieved,accordingto[1],byscalingtheinputsandtheoutputssuch
that they vary in the interval between -1and 1. A common scaling is to choose
diagonalmatrices
D
y
andD
u
sothatthephysicaltruevariablesy
p
andu
p
satisesy
p
= D
y
y
andu
p
= D
u
u
,wherey
andu
varybetween−1
and1
. Hence,therst stepoftheLoopShapingprocedure istochoosethematricesD
y
andD
u
in (2.1).D
yy = G(s)Duu ⇔
y = D
−1
y
G
(s)D
u
u
Noticethatthescalingscanbealteredduringthedesignphaseandthatpreliminary
scalingsmayberough.
2.2.2 Pairing of the Inputs and Outputs
InthesecondstepoftheLoopShaping proceduretheinputsand theoutputsare
pairedto achieveaplantthatisasdiagonalaspossible. Theinteractionsbetween
inputs and outputs reect controller capabilities in the sense that strong cross
couplingsoftenleadto poorcontrol. Hence,itis importantfortheplantto beas
diagonalaspossibletomakeiteasierforthecontrollertoachievegoodperformance
androbustness. In[1]itisshownhowtheRelative GainArray can beusedas an
indication of how the inputs and outputs should be arrangedto make the plant
as diagonal as possible. The relative gain array can be calculated according to
Denition2.1.
Denition2.1 The Relative Gain Array of an arbitrary complex-valued matrix,
A
,isdenedasRGA
(A) = A. ∗ (A
†
)
T
where
.∗
denotes elementwise multiplicationand†
denotes the pseudoinverse. Inthedenitionabove,A
isastaticmatrix. Hence,fortheplantG
(s)
theRGA
canonlybecalculatedforxedfrequencies. Thetwomostcommonfrequenciestoconsiderare
ω
= 0
and thedesiredbandwidth,ω
= ω
b
, oftheclosedloopsystem. TheRGA
ofamatrixA
hassomeusefulcharacteristics,e.g.•
Ifrows(columns)arerearrangedinA
thenthecorresponding rows(columns)ofRGA
(A)
arerearranged.• RGA
of a matrixA
is independent of scalings, e.g.RGA
(A) = RGA(D
−1
y
AD
u
)
•
Thesumof theelementsin arow(column)isalways1.For aplant evaluated in
ω
= 0
andinω
= ω
b
there are twoformal resultson howtoarrangetheinputsandoutputsoftheplant,namelyI. Theinputsandoutputsshouldbepairedsuch thatthediagonalelementsof
RGA
(G(iω
b
))
areasclosetoone,inthecomplexplane,aspossible. II. AvoidpairingsthatimpliesnegativediagonalelementsforRGA
(G(0))
.Thesecondstepistocomputetherelativegainarrayfor
ω
= 0
andforω
= ω
b
andpairtheinputsandoutputstogethersuchthattheresultsIandIIaresatised.2.2.3 Decentralized Controller Design
Asthethird stepof theLoopShapingprocedure adecentralizedcontroller
W
p
(s)
is designedsuch that the singularvaluesof theshapedplant (loop gain) receivesthedesiredshape. Theideaofthedecentralizedcontrolistodesignacontrolleras
ifthesystemwasdiagonal,i.e. to disregardtheinuence ofcrosscouplings. The
precompensator
W
p
(s)
chosenfor thecontroller designused in this thesisis, like in [8], supposed to addintegralaction to thescaled anddiagonalized plant. Thetransferfunction matrix
W
p
(s)
canbedenedaccordingto (2.2).W
p
(s) =
K
1
(1 +
1
T
1
s
)
0
0
0
K
2
(1 +
T
1
2
s
)
0
0
0
K
3
(1 +
T
1
3
s
)
(2.2)Thecoecients,
K
1
, T
1
. . . K
3
, T
3
aredesignparametersandcanbealteredduring thedesignphase. Inthenextsection, thefourth (andoptional) stepofthe LoopShapingprocedurewill bedescribed. Thisstepusesacertainmethodtoalignthe
singularvaluesofthescaled, diagonalizedandprecompensatedplantat acertain
frequency.
2.2.4 Alignment of Singular Values
As thefourth and last stepofthe LoopShaping procedure,thealignmentof the
singular values of the shaped plant is carried out. The goal of the alignment
procedureistochooseamatrix
W
a
suchthatthesingularvaluesofthetheloopgain aregatherednearoneatthedesiredbandwidthoftheclosedloopsystem. Theideaistousethe
W
a
matrixfrom Figure2.1to minimizethefrequencyintervalwhere boththesensitivityfunctionS
(s)
andthecomplementarysensitivityfunctionT
(s)
arelarge. Themethodisdescribedin[1]andusesthesingularvaluedecomposition(SVD).
Denition2.2 Thesingularvalue decompositionof a
n
× m
matrixA
isA
= UΣV
∗
where
∗
denotesthe complexconjugatetransposeofamatrix.
U
isan
× n
unitary (U U
∗
= I
) matrix,
Σ
isan
× m
matrixwhich has the singular values ofA
along the diagonalandzeroselsewhere andV
isaunitarym
× m
matrix.InthealignmentstepitishereonlynecessarytoconsiderSVD:sforquadratic
3 × 3
matrices. LettheSVDforthematrixA
beU
= u1
u
2
u
3
Σ =
σ
0 σ
1
0
2
0
0
0
0 σ
3
V
= v1
v
2
v
3
where
u
i
= (u
1i
, u
2i
, u
3i
)
T
and
v
i
= (v
1i
, v
2i
, v
3i
)
T
. Based on the denitions
aboveitispossibleto write
A
asA
= UΣV
∗
=
3
X
i=1
u
i
σ
i
v
i
∗
Astheequality
Av
j
v
j ∗
= u
j
σ
j
v
j ∗
holds,A
(I + αv
j
v
j ∗
)
hasthesamesingular valuesasA
apartfromthej
:thwhichhasbeenalteredtoσ
j
(1 + α)
. It ispossible toalterthesingularvaluesofthe3 × 3
matrixG
(iω)
atthefrequencyω
withthe followingmatrixW
a
= I + α
1
v
1
v
1
∗
+ α
2
v
2
v
2
∗
+ α
3
v
3
v
3
∗
where the coecients,
α
1
toα
3
, can be chosen such that the singular values ofG
(iω)W
a
willbealignednearone.2.2.5 Summary of the Loop Shaping Procedure
Themain stepsoftheLoopShapingprocedureare
1. Scale the inputs and outputs by means of matrices
D
−1
y
andD
u
so that a scaledplantD
−1
y
G
(s)D
u
isobtained.2. ComputetheRGAforthefrequencies0andthedesiredbandwidth. Pairthe
inputsandtheoutputsaccordingtothemainresultsinSection2.2.2.
3. Designadecentralizedcontroller
W
p
(s)
sothat thesingularvaluesofD
−1
y
G
(s)D
u
W
f
W
p
(s)
havethedesiredshape.4. Alignthesingularvaluesatthedesiredbandwidthbymeansofthe
W
a
matrix. Thisstepisoptional.The result of the design steps in the Loop Shaping procedure is the shaped
plant
G
s
(s)
accordingtoG
s
(s) = D
y
−1
G
(s)D
u
W
f
W
p
(s)W
a
The shapedplant
G
s
(s)
can be foundin Figure 2.1and it willbethe subject oftherobuststabilizationproceduredescribedin thenextsection.2.3 Robust Stabilization
The second part of the controller design phase is to stabilize the shaped plant
G
s
(s)
. This isdonein orderto achievemaximumstabilitymarginsforthe closed loopsystemG
c
(s)
,i.e. makingtheresultingclosedloopsystemasrobusttomodel perturbations and disturbances as possible. The shape of the singular values of2.3.1 Normalized Left Coprime Factorizations
TheRobustStabilizationprocedureproduceacontroller
K
∞
(s)
thatwillstabilize a certain class of systems. These systems are those who can be described bynormalizedleft coprime factorizations. A normalizedleftcoprimefactorization of
asystem
G
(s)
isgivenbyG
(s) = ˜
M
−1
(s) ˜
N
(s)
(2.3) whereM
˜
(s) ˜
M
∗
(s) + ˜
N
(s) ˜
N
∗
(s) = I
.
The normalizedcoprime factorizationplant descriptionof theplant
G
(s)
can bederivedaccordingtothefollowingtheorem.Theorem2.1 (Normalizedleft coprimefactorization) Let
G
(s)
begivenbyG
(s) =
A
B
C
D
anddene˜
R
= I + DD
∗
>
0
Suppose
(C, A)
is detectable and(A, B)
controllable. Then there is a normalized leftcoprimefactorizationG
(s) = ˜
M
(s)
−1
N
˜
(s)
,with˜
M
(s) =
A
+ LC
L
˜
R
1/2
C
R
˜
−1/2
and˜
N
(s) =
A
+ LC B + LD
˜
R
1/2
C
R
˜
−1/2
D
whereL
= −(BD
∗
+ ZC
∗
) ˜
R
−1
Z
isthe positive semidenite solutiontothe algebraicriccati equation(A − BD
∗
R
˜
−1
C
)Z + Z(A − BD
∗
R
˜
−1
C
)
∗
− ZC
∗
R
˜
−1
CZ
+ B ˜
R
−1
B
∗
= 0.
Proof. SeeTheorem 13.37,in[10].
2
Theclassofperturbationsthat
K
∞
(s)
stabilizesisgivenby(2.4).G
∆
(s) = ( ˜
M
(s) + ∆
M
(s))
−1
( ˜
N
(s) + ∆
N
(s))
(2.4) where∆
M
(s), ∆
N
(s)
arestable transferfunctions that representthe uncertainty in thenominalplant. Noticethat theuncertainties canintroducebothnewpolesand zerosinto the plant. This meansthat theperturbed plantmight havemore
unstablepoles andzerosthanthenominal plant. Thecontroller
K
∞
(s)
resulting fromtheH
∞
LoopShapingcontrollerdesignstabilizesperturbedsystemsG
∆
(s)
withk [∆
N
(s) ∆
M
(s)] k
∞
<
1
2.3.2 The Stabilizing Controller
K
∞
(s)
This section is a review of the almost fully automatic procedure that, given a
shaped plant
G
s
(s)
, producesa stabilizingcontrollerK
∞
(s)
. For ashaped plant givenby(2.6)G
s
(s) =
A
s
B
s
C
s
0
(2.6)thefollowingstepssummarizestheRobustStabilization procedure.
•
Firstandforemostitisnecessarytondthesymmetricandpositivedenite solutionsX
andZ
to thecontrol algebraic riccati equation andtheltering algebraicriccati equationA
s
∗
X
+ XA
s
− XB
s
B
s
∗
X
+ C
s
∗
C
s
=0
A
s
Z
+ ZA
∗
s
− ZC
s
∗
C
s
Z
+ B
s
B
∗
s
=0
(2.7)
•
Theoptimal(smallest)γ
opt
maximizing(2.5)isgivenbyγ
opt
=
p
1 + λ
m
(XZ)
(2.8)where
λ
m
(XZ)
isthelargesteigenvalueofXZ
. However,itis,accordingto [8],provenbyexperiencethatbetterresultsareoftenachievedwithaslightlylarger
γ
. With this in mind it is usually better to setγ
to approximately1.1γ
opt
.•
DenetwomatricesF
andH
asF
=γ
2
B
s
∗
X
((1 − γ
2
)I + XZ)
∗−1
H
= − ZC
s
∗
(2.9)
Thenthestabilizingcontroller
K
∞
(s)
isgivenbyK
∞
(s) =
A
s
+ HC
s
+ B
s
F
−H
F
0
(2.10)Thecontroller
K
∞
(s)
can,forexample,beimplementedaccordingtotheblock diagraminFigure2.2. ThebenetofhavingthecontrollerK
∞
(s)
inthefeedback comparedtoaunityfeedbackisthatanabruptchangeintheinputsdoesnotexcitey
s
r
G
s
(
s
)
K
∞
(
s
)
K
∞
(
0
)
Σ
Figure2.2. Theclosedloopsystemfortheshapedplant
G
s
(s)
andthecontrollerK
∞
(s)
. NoticethestaticgainK
∞
(0)
.The implementation in Figure 2.2 is however not suitable due to the gain
schedulingapproachin this controllerdesign method. An alternative
implement-ation will bedescribed in the nextsection. There the controller is written asan
exactplantobserverplus statefeedback and this structure turns outto bemore
suitableforgainscheduling.
2.4 Gain Scheduling
Linearcontrollerdesignisoftenusedtostabilize anonlinearsystem,especiallyin
ightapplications. Sincealinearcontrollerdesignonlyisbasedonalinearmodel
one controller is usually insucient. The third step of the
H
∞
Loop Shaping controllerdesign is theGain Scheduling approach. The ideais to implement thecontroller asa plantobserverplus state feedback and to use linear interpolation
betweencontrollersofadjacentdesignpoints. In[2]itisshownhowthecontroller
K
∞
givenby(2.10)canbewrittenasanexactplantobserverplusstatefeedback. Theresultis(
˙ˆx = A
s
ˆx + B
s
u
s
+ H(C
s
ˆx − y
s
)
u
s
= F ˆx + P r
(2.11)
All matrices in (2.11) apart from
P
can be recognized from (2.10). TheP
matrixisintroducedinordertogivethetransferfunctionfromreferencesignalstooutputsignalsthestaticgain
I
. InSection5.1 itwill beshownthatP
shouldbe chosenasTheobserverimplementationisshownintheblockdiagramin Figure2.3. For
thismasterthesisprojectthecontrollerisscheduledoneitheroneortwo
paramet-ers,thespeedofthemissile
V
andtheangleofattackα
.Σ
Observer
F
P
u
s
y
s
r
G
s
(
s
)
Figure 2.3. Theobserverimplementationofthe
H
∞
LoopShapingcontroller.2.4.1 Scheduling on Either One or Two Parameters
Forthismaster thesisprojectitisonlynecessarytodesign againscheduled
con-trollerthat useseither the forward speed
V
, orV
andthe angle of attackα
, for theinterpolation.SchedulingontheForward Speed
Assume,forexample,thattheforwardspeedvariesintheinterval
[V
1
, V
2
[
andthat twocontrollerswithH
matricesH
1
andH
2
areavailable. These controllershave beendesigned using the linearizationsaroundV
= V
1
andV
= V
2
, respectively. Letη
beafunction ofV
η
=
V
− V
1
V
2
− V
1
(2.12)
Aninterpolated
H
matrixH
(η)
canthenbedened asH
(η) = (1 − η)H
1
+ ηH
2
(2.13)SchedulingontheForward Speedandthe Angleof Attack
Assume, in analogywith the previous section, that the missile forward speed
V
varies in the interval[V
1
, V
2
[
, that the angle of attackα
varies in the interval[α
1
, α
2
[
andthatfouradjacentcontrollersareavailable.Thesecontrollershavebeen designedaroundfourlinearizations.Forexample,theH
matrixforthelinearization aroundV
= V
1
andα
= α
1
iscalledH
11
. Letε
beafunction ofα
ε
=
α
− α
1
Aninterpolated
H
matrixH
(η, ε)
canbedenedasH
(η, ε) = (1 − η)[(1 − ε)H
11
+ εH
12
]+
η
[(1 − ε)H
21
+ εH
22
]
(2.15)
Itisstraightforwardto extendtheschedulingtothreeormorevariables.
Thetopicofthenextchapteristhemissileanditscharacteristics. Thechapter
willdiscussthemissile ingeneralanditwillalsoshowhowalineardescriptionof
The Missile Model
An accurate mathematical descriptionof the control objectis the cornerstone of
asuccessfulcontroller. Theplantis heretheair-to-air missile in Figure 3.1.
Al-though this missile does not exist yet, it has several features in common with
existingones. The
H
∞
LoopShapingcontrollerdesignmethod isbasedonlinear controltheory. Thus,thetopicofthischapteristodescribehowalinearplantG
(s)
, which describesthemaindynamics ofthemissilein Figure3.1,can bederived.3.1 The Missile
This sectionis asummary of themissile characteristicsand theguidance system
of themissile. Themissile studied in this master thesis projectis a bank-to-turn
missile,which meansthatwhenthemissile isgoinginto aturnitshould rstroll
inthedirectionofitsvelocityvector
V
andthenaccelerateinthedirectionofthez
axis.3.1.1 The Guidance System of the Missile
Thissectionwillgiveanintroductiontotheguidancesystemofamissile. InFigure
3.2asimplied descriptionoftheguidancesystemisgiven.
Figure 3.2.AsimplieddescriptionoftheguidancesystemoftheMissile.
AccordingtoFigure3.2,theguidancesystemconsistsofthreeparts,whereeach
Strap Down Navigation. Whenthe missileis released thestrap down
naviga-tion system receivesinitial position, attitude, rotationand speed from the
aircraft navigation system. The strap down navigation system uses
meas-urementsfromrategyrosandaccelerometersincombinationwiththeinitial
conditionstocalculateposition,attitude, rotationandspeed. Theresultsof
thesecalculationscanbeusedbybothGuidanceandAutopilot.
Guidance. The core of the guidance system is the block Guidance, which by
predened laws decides what the missile should do. Guidance receives
in-formationconcerningthetargetpartlyfrom theaircraftandpartlyfromthe
seeker. This information, combined with the missile state variables
calcu-latedbythestrapdownnavigationsystemandthepredenedlaws,produces
demanded referencesignalstotheautopilot.
Autopilot. The autopilot receivesreference signalsas demanded output signals
in termsof desiredangular velocity aroundthe
x
axis, desiredacceleration in thedirectionofthey
axisand desiredaccelerationinthedirection ofthez
axis. These reference signals arep
d
,A
yd
andA
zd
and the main task of theautopilot istoproducecommanded ndeectionssuchthat thedesiredreferencesareachieved.
The
H
∞
LoopShapingcontrollerofthisprojectisimplementedintheshaded blockAutopilotinFigure3.2. Theautopilotreceivesdemandsinp
d
,A
yd
andA
zd
fromtheguidancesystem. DuetotheBank-to-Turnsteeringprinciplethedemandsin
p
d
,A
yd
andA
zd
aresuitableas referencesignals. 3.1.2 Assumptions and LimitationsThepurposeofthischapteristoproducelinearmodelsofthemaindynamicsofthe
missile that canbeused forthe
H
∞
Loop Shapingcontroller design. Thelinear modelof themissile iscalled aplantanditisdesirablethattheplantdescriptionhas aslow degree aspossible. Hence, it is necessary to make some assumptions
duringthemodelingofthemaindynamicsofthemissile. First,thereisaneedfor
anexplanationofthenotationinFigure 3.1.
• (x, y, z)
isthexedbodyframeofthemissile.• ω = (p, q, r)
T
is theangularvelocityof themissile
[rad/s]
.• V = (u, v, w)
T
isthemissilevelocityexpressed
in thebodyframeofthemissileand
V
= |V | [m/s]
.• α
istheangleof attack[rad]
.• β
isthesideslipangle[rad]
.The missile also has a mass moment of inertia matrix
I
, which due to the symmetryofthemissile containnomixed inertiatermsandisgivenbyI
=
I
0 I
x
0
y
0
0
0
0 I
z
[kgm
2
]
(3.1)Thenexttwosectionscontaindescriptionsofthelimitationsandtheassumptions
madeinthisproject.
TheLimitations ofthe Modeling Procedure
Inoperating conditionsthere areloads acting onthemissile, e.g. gravity, engine
thrust, engine torque aerodynamic contributions etc. Some of the loads are not
goingtobeconsideredforthemodelingof themain dynamicsofthemissile. The
gravityisneglectedduetoitssmallcontributionandtomaintainmodelsimplicity.
The contribution from the engineis neglectedbecause the rate of change of the
enginethrustismuch slowerthantheotherdynamics. Theonlyforceand torque
contributionthataregoingtobeconsideredaretheforceandtorquethatdescend
fromtheaerodynamicsofthemissile,see[5]. Theaerodynamicforceactingonthe
missileisaccordingto [6]givenby
F
a
= −q
d
S
C
C
T
C
C
N
(3.2) whereC
T
=1
C
C
=C
Cβ
β
+ C
Cδ
r
δ
r
C
N
=C
N α
α
+ C
N δ
e
δ
e
andtheaerodynamictorqueactingonthemissileisaccordingto[6]givenby
M
a
= −q
d
Sd
C
C
m
l
C
n
(3.3) whereC
l
= C
lβ
β
+ C
lp
d
2V
p
+ C
lδ
a
δ
a
C
m
= C
mα
α
+ C
m|β|
|β| + C
mq
d
2V
q
+ C
mδ
e
δ
e
C
n
= C
nβ
β
+ C
nαβ
αβ
+ C
nr
d
2V
r
+ C
nδ
a
δ
a
+ C
nδ
r
δ
r
In Equations(3.2) and (3.3)
q
d
is the dynamic pressure,d
areference length andS
= πd
2
/
4
Mach 1.5 2.0 2.5 3.0 Mach 1.5 2.0 2.5 3.0
C
Cβ
20 20 20 20C
Cδ
r
-6 -6 -6 -6C
N α
33 31 29 27C
N δ
e
8 7 6 5C
lβ
1.5 2.5 3.0 3.5C
lp
-30 -27 -24 -21C
lδ
a
-6 -6 -6 -6C
mα
-16 -14 -12 -10C
m|β|
-14 -11 -8 -5C
mq
-1900 -1600 -1300 -1000C
mδ
e
-49 -45 -41 -37C
nβ
24 22 19 16C
nαβ
-200 -175 -150 -125C
nr
-2000 -1700 -1400 -1100C
nδ
a
11 9 7 5C
nδ
r
-50 -45 -40 -35Table3.1. Theaerodynamiccoecients.
The missile ight envelopeis rather large, with a largespan in both altitude
andmissileforwardspeed. Thealtitudeandspeedintervalsthatthemissileshould
beabletooperateinare
•
Altitude:0 ≤ h ≤ 20000 m
.•
Forwardspeed:450 ≤ V ≤ 1200 m/s
.Due to thelimitations mentionedin Chapter 1 thecontrollersstudied in this
thesisarederivedfortheforwardspeedintervalmentionedearlierandforthexed
altitudeof
h
= 1000 m
.TheAssumptionsofthe Modeling Procedure
Forthemodeltobesucientlysimpleitisnecessarytomakesomeapproximations.
Besidesneglectingsomeforceandtorquecontributionsfouradditionalassumptions
aremade,namely
1. Assume that theangles
α
andβ
in Figure 3.1are small, which forexample meansthatsin β ≈ β
andtan α ≈ α
.2. Assume smallvariationsin themissile forwardspeed,i.e.
˙V ≈ 0
.3. Assume that themassofthemissile isconstant, i.e. neglectdecrease ofthe
massofthemissilecausedbythecombustionoffuel.
3.2 The Main Dynamics of the Missile
Themaindynamicsofthemissileisdescribedwithbasicrigidbodydynamicsand
suitableapproximations.
3.2.1 Rigid Body Dynamics
When a rigid body is inuenced by forces it will go into translation and when
inuencedbytorqueitwillstartrotating. Hence,itisnecessarytounderstandthe
dynamicsof therigid body. Thetoolsused to create amathematicaldescription
areold but powerful. IsaacNewton rst postulated therelationshipbetweenthe
force actingon arigid body and itsaccelerationand this relationis hencecalled
Newton'slaw of motion. Thefactthattheappliedtorqueequalstherateofchange
of angularmomentum was rstpostulated byLeonardEuler and this equalityis
usuallyknownas Euler'sequation. Theyaretwowellknownlawsofnature buta
remindermightbeappropriate.
Denition3.1 (Newton'slaw of motion) Theforceonarigidbodyequalsthe
massof the rigidbodytimesitsinertial acceleration:
F = ma.
Denition3.2 (Euler'sequation) Theappliedtorqueequalstheinertialrateof
changeof the angularmomentumofarigid body:
M = ˙
H
Theangularmomentum canbeexpressedin termsofmassmomentofinertiaand
therotationoftherigidbody,i.e.
H = Iω
. Noticetheuseofinertial in thetwo denitionsabove. Inordercalculate theinertialtimederivativeofarigid body itis necessaryto pay attention to its rotation. The theorem below shows how the
inertialtimederivativeofarigidbodycanbecalculated.
Theorem3.1 (Inertial derivative) The time derivative of a vector
v
with re-spect to the inertial (i
) frameis related tothe time derivative with respectto the bodyframe(b
) byd
dt
i
v =
d
dt
b
v + ω
bi
× v
whereω
bi
isthe angularvelocity ofthe bodyframeand
v
isthe vectorinthe body frame.Proof. Theproofisgivenin[9].
2
The main purposeof derivinga mathematical descriptionof the missile is to
use it for the controller design. Theearlier mentioned simplications havebeen
madeinordertomaketheresultingmodeleasiertowork with,hopefullywithout
changingthemaindynamicsofthemissiletoomuch. Themodelingapproachisto
3.2.2 The Translation of the Missile
Forthemissile translationitissensibletoexpresstheacceleration
a
inDenition 3.1asthetimederivativeofthemissilevelocityV
. Thistimederivativeisobtained byapplyingTheorem3.1onthemissile velocityaccordingto(3.4).a = ˙V + ω × V
(3.4)Sinceonlytheaerodynamicforce
F
a
isconsidered,thetranslationofthemissile isobtainedbycombining(3.4)andDenition3.1. Theresultis
F
F
ax
ay
F
az
= m
˙u
˙v
˙w
+ m
p
q
r
×
u
v
w
(3.5)Equation(3.5)containsthetimederivativesof thesuitablestatevariables
u, v
andw
. The mathematicaldescriptionwhich isthepurposeofthis sectionshould expressthetimederivativesofthestatesasfunctionsof thestatesandtheinputsanditisthereforesensibleto rewrite(3.5)accordingto(3.6).
˙u =
F
ax
m
+ rv − qw
˙v =
F
ay
m
+ pw − ru
˙w =
F
az
m
+ qu − pv
(3.6)Itispossibletoinclude thetwoangles
α
andβ
asstatesinsteadofu, v
andw
. Todothatthereisaneedforsomefurtherassumptions. AsseeninFigure3.1thefollowing relationbetweenthe angles,
α
andβ
, and the velocities,u, v
andw
, is obvious.sin β =
v
V
tan α =
w
u
(3.7)The assumption mentioned earlier considering small angles is useful and the
approximations
sin β ≈ β
andtan α ≈ α
are in this application sucient up to|α| ≤ 30
◦
and
|β| ≤ 30
◦
. Hence,therelationshipbetween
α
andβ
andu
andw
isβ
=
v
V
α
=
w
u
(3.8)
Forsmallanglesitispossibletoassumethat
V ≈ V x
,whichmeansthatu
≈ V
, andifthevariationsin thespeedV
aresmallthen˙V ≈ 0
and sois˙u
. Thisleads,combinedwith Equations(3.6)and (3.8),to tworst orderdierential equations in
α
andβ
˙α =
F
az
mV
+ q − pβ
˙β =
F
ay
mV
+ pα − r
(3.9)The forces
F
az
andF
ay
areobtained fromF
a
in (3.2)and this results in the descriptionoftheα
andβ
dynamicsaccordingto (3.10).˙α = − pβ + q −
q
d
S
mV
(C
N α
α
+ C
N δ
e
δ
e
)
˙β =pα − r −
q
d
S
mV
(C
Cβ
β
+ C
Cδ
r
δ
r
)
(3.10)
3.2.3 The Rotation of the Missile
The total torque acting on the missile equals the rate of change of the angular
momentumofthemissile. Thisis,duetothefactthat
H = Iω
,thedescriptionof therotationof themissile. TheinertialtimederivativeofH
isthen givenbythe followingequation.˙
H = I ˙ω + ω × Iω
(3.11)The Euler equation from Denition 3.2 combined with (3.11), leads to the
followingequation.
M
M
ax
ay
M
az
=
I
I
x
y
˙p
˙q
I
z
˙r
+
p
q
r
×
I
I
x
y
p
q
I
z
r
(3.12)Suitable statevariablesin (3.12)are
p, q
andr
. Byextractingthetime deriv-ativesofthesevariables from(3.12)thefollowingexpressionisobtained.˙p =
M
x
I
x
+
I
y
− I
z
I
x
qr
˙q =
M
y
I
y
+
I
z
− I
x
I
y
pr
˙r =
M
z
I
z
+
I
x
− I
y
I
z
pq
(3.13)Equation(3.13)incombinationwith(3.3)leadstothreerstorderdierential
equationsin
p
,q
andr
accordingto (3.14).˙p =qr
I
y
− I
z
I
x
+
q
d
Sd
I
x
(C
lβ
β
+ C
lp
d
2V
p
+ C
lδ
a
δ
a
)
˙q =pr
I
z
− I
x
I
y
+
q
d
Sd
I
y
(C
mα
α
+ C
m|β|
|β| + C
mq
d
2V
q
+
+ C
mδ
e
δ
e
)
˙r =pq
I
x
− I
y
I
z
+
q
d
Sd
I
z
(C
nβ
β
+ C
nαβ
αβ
+ C
nr
d
2V
r
+
+ C
nδ
a
δ
a
+ C
nδ
r
δ
r
)
(3.14)3.2.4 Summary of the Main Dynamics of the Missile
The main dynamics of the missile, described by equations (3.10) and (3.14), are
summarizedhere.
˙p =qr
I
y
− I
z
I
x
+
q
d
Sd
I
x
(C
lβ
β
+ C
lp
d
2V
p
+ C
lδ
a
δ
a
)
˙q =pr
I
z
− I
x
I
y
+
q
d
Sd
I
y
(C
mα
α
+ C
m|β|
|β| + C
mq
d
2V
q
+
+ C
mδ
e
δ
e
)
˙r =pq
I
x
− I
y
I
z
+
q
d
Sd
I
z
(C
nβ
β
+ C
nαβ
αβ
+ C
nr
d
2V
r
+
+ C
nδ
a
δ
a
+ C
nδ
r
δ
r
)
˙α = − pβ + q −
q
d
S
mV
(C
N α
α
+ C
N δ
e
δ
e
)
˙β =pα − r −
q
d
S
mV
(C
Cβ
β
+ C
Cδ
r
δ
r
)
(3.15)There is need to be able to use several linearized models for the
H
∞
Loop Shapingcontrollerdesign. Theselinearizationscanbederivedfrom (3.15).3.3 The Plant
Theplantusedforthe
H
∞
LoopShapingcontrollerdesignisalineartransfer func-tionG
(s)
mappingtheinputu
totheoutputy
. Thisplantisalineardescription ofthemaindynamicsofthemissile inFigure3.1anditisdesirablethattheplantisasimplebut yet accurate descriptionof themissile. The resultof theprevious
sectionwasEquation(3.15),whichisanonlineardescriptionofthemaindynamics
ofthemissile, andfrom thisequation aplantisderived. Sincegainschedulingis
includedinthecontrollerdesignitisnecessarytobeabletoproduceseverallinear
Theresultfromthissectionwillbetheplant
G
(s)
mappingtheinputu
tothe outputy
anditcan eitherbewritten asastatespacerealizationaccordingto(
˙x = Ax + Bu
y = Cx + Du
(3.16)orasatransferfunction matrixaccordingto
G
(s) = C(sI − A)
−1
B
+ D
(3.17)In(3.16)
x
isthestatesdened asx =
p
q
r
α
β
(3.18)and the input
u
is the n deections of theaileron, the elevator and therudder accordingtou =
δ
δ
a
e
δ
r
(3.19)Inboth(3.16)and (3.17) thematrices
A
,B
,C
andD
are includedandthese matricesarecalledsystemmatrices.Themaindynamicsofthemissiledescribedbyequation(3.15)willinthefuture
bereferredto as
˙x = f(x, u, h, V )
(3.20)andthelineardescriptionof (3.20)includesthetwosystemmatrices
A
andB
and hasthefollowingappearance˙x = Ax + Bu
(3.21)The followingtheorem, which can be found in [1], is used to calculate the
A
andB
matrices.Theorem3.2 (Linearization) If
f(x, u, h, V )
isdierentiableinaneighborhood of the stationary pointx
0
,u
0
,h
0
andV
0
it is possible to approximate equation (3.20)with˙z = Az + Bv + g(z, v)
(3.22)where
z = x − x
0
,v = u − u
0
andg(z,v)
|z|+|v|
→ 0
when|z| + |v| → 0
. The two matricesA
andB
aregiven byA
=
∂f
(x, u)
∂x
x=x
0
,u=u
0
,h=h
0
,V =V
0
(3.23) andB
=
∂f
(x, u)
∂u
u=u
0
,x=x
0
,h=h
0
,V =V
0
(3.24)Proof. SeeTheorem
11.1
in [1].2
A plant description of the missile can be obtainedby choosing theC
andD
matricesof (3.16)suchthat thedesiredoutputs areobtained. AccordingtoThe-orem3.2 thelinearizationis onlyvalid in aneighborhood of thestationary point
x
0
andu
0
of (3.20)andthispointsatisesf(x
0
, u
0
, h
0
, V
0
) = 0
(3.25) The Newton-Raphson method is a techniquethat makes it possible to nd anumericalsolution
x
0
,u
0
,h
0
andV
0
to(3.25). Thismethodisrestrictedtosystems withn
equationsandn
unknowns. Thisis notthecasefor(3.25), which hasve equations and tenunknowns. The unknowns are theve states, thethree inputsignals,the speed of themissile
V
and thedynamic pressurewhich is afunction ofV
and thealtitudeh
. Beforeitispossibletocalculate thestationarypointsof (3.25)usingtheNewton-Raphsonmethoditisthusnecessarytochooseveofthetenunknownsasdesignparameters.
3.3.1 Choosing the Design Parameters
Theightenvelopeisdividedintothespeedofthemissile
V
andthealtitudeh
ac-cordingtoSection3.1.2. ThisimpliesthatV
andh
aresuitabledesignparameters. Themain focus forthis masterthesisis to evaluate controllersfor xedaltitudesandvariationsin speed. Theangleofattack
α
isagainschedulingparameterand shouldalsobechosenasadesignparameter. Therearetworemainingdesignpara-metersto choose. Theangular velocity
p
aroundthex
axisisasensiblechoiceas thefourthdesignparameterduetothefact thatp
d
isareferencesignal. Another referencesignalistheaccelerationin thedirectionof they
axis,which makesthe sideslipangleβ
agood choiceasthefthdesignparameter.3.3.2 The Newton-Raphson Method
Asseenin theprevioussectionthevedesignparametersarechosenas
•
The angular velocity aroundthex
axis,p
.•
Theangleofattack,α
.•
Thesideslipangle,β
.•
The forward speed of the missile,V
.•
Thealtitude,h
.Thisleavesasystemofveequationsandveunknownswhichaccordingto[7]
canbesolved by using theNewton-Raphson method. With the designpointsas
xedvalues(3.20) canberewrittenas