Surrogate-Based Modeling and Optimization: Applications in Engineering

413 Pages·2013·16.074 MB·English

by  Slawomir Koziel, Leifur Leifsson (auth.), Slawomir Koziel, Leifur Leifsson (eds.)

Slawomir Koziel
Editors
Leifur Leifsson
Surrogate-Based
Modeling and
Optimization
Applications in Engineering
Surrogate-Based Modeling and Optimization
Slawomir Koziel (cid:2) Leifur Leifsson
Editors
Surrogate-Based
Modeling and
Optimization
Applications in Engineering
Editors
SlawomirKoziel LeifurLeifsson
EngineeringOptimization&ModelingCent SchoolofScienceandEngineering,
ReykjavikUniversity EngineeringOptimization&Modeling
Reykjavik,Iceland ReykjavikUniversity
Reykjavik,Iceland
ISBN978-1-4614-7550-7 ISBN978-1-4614-7551-4(eBook)
DOI10.1007/978-1-4614-7551-4
SpringerNewYorkHeidelbergDordrechtLondon
LibraryofCongressControlNumber:2013941933
MathematicsSubjectClassification(2010): 74P10,80M50,97M10,62P30
©SpringerScience+BusinessMediaNewYork2013
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,
broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology
nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’slocation,initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.
PermissionsforusemaybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violations
areliabletoprosecutionundertherespectiveCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
Whiletheadviceandinformationinthisbookarebelievedtobetrueandaccurateatthedateofpub-
lication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityforany
errorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,withrespect
tothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Preface
Contemporary engineering design is heavily based on computer simulations. Ac-
curate,high-fidelitysimulationsareusednotonlyfordesignverificationbut,most
importantly, to adjust parameters of the system (e.g., wing geometry, material pa-
rameters of antennas) so that it meets given performance requirements. Unfortu-
nately, accurate simulations are often computationally expensive, with evaluation
timesrangingfromhourstodaysperdesign.Consequently,designautomationusing
conventionaloptimizationmethodsisoftenimpracticalorevenprohibitive.Otheris-
suesincludethenumericalnoisethatisoftenpresentinsimulationresponses,and
theabsenceofsensitivityinformation.These,andotherproblems,canbealleviated
bythedevelopmentandemploymentofso-calledsurrogates,whichreliablyrepre-
senttheexpensive,simulation-basedmodelofthesystem/deviceofinterest,butare
muchcheaperandanalyticallytractable.
Thiseditedbookisaboutsurrogate-basedmodelingandoptimizationtechniques
andtheirapplicationsforsolvingdifficultandcomputationallyexpensiveengineer-
ing design problems. A group of international experts summarize recent develop-
mentsinthefieldanddemonstrateapplicationsinvariousdisciplinesofengineering
andscience.Themainpurposeoftheworkistoprovidethebasicconceptsandfor-
mulations of the surrogate-based modeling and optimization paradigm, as well as
to discuss relevant modeling techniques, optimization algorithms, and design pro-
cedures.
Simulation-drivendesignbasedonsurrogatemodelsplaysanincreasinglyimpor-
tantroleincontemporaryengineeringandpermitsustosolveproblemsthatcannot
besolvedotherwise,particularlybydramaticallyreducingthecomputationalcostof
thesolutionprocess.Unfortunately,recentresultsconcerningsurrogate-basedmod-
elingandoptimizationarescatteredthroughouttheliteratureinvariousengineering
fieldsand,therefore,arenoteasilyaccessibletoareaderinterestedinthistechnol-
ogy. Thus, this book should be of interest to engineers from any discipline where
computationallyheavysimulations(suchasfiniteelement,computationalfluiddy-
namics,andcomputationalelectromagneticsanalyses)areusedonadailybasisin
thedesignprocess.Theeditorsofthisvolumehopethatthepresentedmaterialwill
allow the readers to gain an understanding of the basic mechanisms of the surro-
v
vi Preface
gate modeling process and familiarize themselves with important components of
surrogate-basedoptimizationalgorithmsandtheadvantagesofemployingvariable-
fidelity simulation-driven design, as well as enable them to reduce the cost of the
designprocessaidedbycomputersimulations.
Reykjavik,Iceland SlawomirKoziel
March2013 LeifurLeifsson
Contents
Space Mapping for Electromagnetic-Simulation-Driven Design
Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
SlawomirKoziel,LeifurLeifsson,andStanislavOgurtsov
Surrogate-BasedCircuitDesignCentering . . . . . . . . . . . . . . . . . 27
Abdel-KarimS.O.HassanandAhmedS.A.Mohamed
Simulation-DrivenAntennaDesignUsingSurrogate-Based
Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
SlawomirKoziel,StanislavOgurtsov,andLeifurLeifsson
PracticalApplicationofSpaceMappingTechniquestotheSynthesisof
CSRR-BasedArtificialTransmissionLines. . . . . . . . . . . . . . . 81
AnaRodríguez,JordiSelga,FerranMartín,andVicenteE.Boria
The Efficiency of Difference Mapping in Space Mapping-Based
Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
MuratSimsekandNeslihanSerapSengor
BayesianSupportVectorRegressionModelingofMicrowaveStructures
forDesignApplications . . . . . . . . . . . . . . . . . . . . . . . . . 121
J.PieterJacobs,SlawomirKoziel,andLeifurLeifsson
ArtificialNeuralNetworksandSpaceMappingforEM-BasedModeling
andDesignofMicrowaveCircuits. . . . . . . . . . . . . . . . . . . . 147
JoséErnestoRayas-Sánchez
Model-BasedVariation-AwareIntegratedCircuitDesign . . . . . . . . . 171
TingZhu,MustafaBerkeYelten,MichaelB.Steer,andPaulD.Franzon
ComputingSurrogatesforGasNetworkSimulationUsingModelOrder
Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
SaraGrundel,NilsHornung,BernhardKlaassen,PeterBenner,and
TanjaClees
vii
viii Contents
AerodynamicShapeOptimizationbySpaceMapping . . . . . . . . . . . 213
LeifurLeifsson, SlawomirKoziel, EirikurJonsson, and
StanislavOgurtsov
EfficientRobustDesignwithStochasticExpansions . . . . . . . . . . . . 247
YiZhangandSerhatHosder
SurrogateModelsforAerodynamicShapeOptimisation . . . . . . . . . . 285
SelvakumarUlaganathanandNikolaosAsproulis
Knowledge-Based Surrogate Modeling in Engineering Design
Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
QianXu,ErichWehrle,andHorstBaier
SwitchingResponseSurfaceModelsforStructuralHealthMonitoring
ofBridges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
KeithWorden,ElizabethJ.Cross,andJamesM.W.Brownjohn
Surrogate Modeling of Stability Constraints for Optimization of
CompositeStructures . . . . . . . . . . . . . . . . . . . . . . . . . . 359
S.Grihon,E.Burnaev,M.Belyaev,andP.Prikhodko
EngineeringOptimizationandIndustrialApplications . . . . . . . . . . . 393
Xin-SheYang
Space Mapping for Electromagnetic-
Simulation-Driven Design Optimization
SlawomirKoziel,LeifurLeifsson,andStanislavOgurtsov
Abstract Spacemapping(SM) has beenoneof the mostpopular surrogate-based
optimizationtechniquesinmicrowaveengineeringtodate.Byexploitingtheknowl-
edge embedded in the underlying coarse model (e.g., an equivalent circuit), SM
allowsdramaticreductionofthecomputationalcostwhileoptimizingelectromag-
netic (EM)-simulated structures such as filters or antennas. While potentially very
efficient,SMisnotalwaysstraightforwardtoimplementandsetup,andmaysuffer
fromconvergenceproblems.Inthischapter,wediscussseveralvariationsofanSM
optimization algorithm aimed at improving SM performance for design problems
involvingEMsimulations.TheseincludeSMwithconstrainedparameterextraction
and surrogate model optimization designed to overcome the problem of selecting
preassigned parameters for implicit SM, SM with response surface approximation
coarse models that maintain SM efficiency when a fast coarse model is not avail-
able, and SM with sensitivity which takes advantage of adjoint sensitivity (which
hasrecentlybecomecommerciallyavailableinEMsimulators)toimprovethecon-
vergence properties and further reduce the computational cost of SM algorithms.
EachvariationoftheSMalgorithmpresentedhereisillustratedusingareal-world
microwavedesignexample.
Keywords Computer-aideddesign(CAD)·Microwaveengineering·
Simulation-drivenoptimization·Electromagnetic(EM)simulation·
Surrogate-basedoptimization·Spacemapping·Surrogatemodel·High-fidelity
model·Coarsemodel
1 Introduction
Space mapping (SM) [1–3] was originally developed in the 1990s to deal with
computationally expensive design problems in microwave engineering [3,4]. Au-
tomated design closure of structures evaluated using electromagnetic (EM) simu-
S.Koziel(B)·L.Leifsson·S.Ogurtsov
EngineeringOptimization&ModelingCenter,SchoolofScienceandEngineering,Reykjavik
University,Menntavegur1,101Reykjavik,Iceland
e-mail:[email protected]
S.Koziel,L.Leifsson(eds.),Surrogate-BasedModelingandOptimization, 1
DOI10.1007/978-1-4614-7551-4_1,
©SpringerScience+BusinessMediaNewYork2013
2 S.Kozieletal.
lations is still a challenging task today, mostly due to the high computational cost
of accurate, high-fidelity EM simulation. The presence of massive computing re-
sourcesdoesnotalwaystranslateintocomputationalspeedup.Thisisduetoagrow-
ingdemandforsimulationaccuracy(whichrequires,amongotherthings,finerdis-
cretizationofthestructure),aswellasthenecessityofincludingimportantinterac-
tionsbetweenthestructureunderdesignanditsenvironment(e.g.,antennahousing
andconnectors,etc.),whichincreasesthecomputationaldomainand,consequently,
slows down the simulation. At the same time, multiphysics simulations (e.g., in-
cluding thermal effects) become more and more important, further contributing to
thecomputationalcostofthesimulation.Asconventionaloptimizationalgorithms
(e.g.,gradient-basedschemeswithnumericalderivatives)requiretens,hundreds,or
eventhousandsofobjectivefunctioncallsperrun(dependingonthenumberofde-
sign variables), the computationalcost of the whole optimizationprocess may not
beacceptable.
SMbelongstoabroaderfamilyofmethodscalledsurrogate-basedoptimization
(SBO)techniques[5–7].SMandmostotherSBOmethodsshareamainstructure,
in which the direct optimization of the expensive (here, EM-simulated) structure,
referredtoasthehigh-fidelityorfinemodel,isreplacedbyiterativerefinementand
reoptimization of a low-fidelity (or coarse) model. The coarse model is a physics-
basedrepresentationofthehigh-fidelityone.Itislessaccuratebutsupposedlymuch
fasterthanthelatter.Anexampleofacoarsemodelinmicrowaveengineeringisan
equivalentcircuitwhichdescribesthesamestructureasthefinemodelbutusingcir-
cuit theory rather than full-wave EM simulation. The SM surrogate is constructed
by enhancing the coarse model through auxiliary transformations, usually linear,
with parameters of these transformations obtained in what is called the parameter
extraction(PE)process[8],whichisatrademarkofSM.PEisexecutedtoreduce
themisalignmentbetweentheresponsesofthespace-mappedcoarsemodelandthe
finemodelatalimitednumberofdesigns,usuallythosethathavealreadyemerged
duringtheSMoptimizationrun.ThebenefitofSMliesinthefactthateachSMiter-
ationusuallyrequiresevaluationthehigh-fidelitymodelatasingledesign(theone
obtained by optimizing the current surrogate model), and—for a well-performing
algorithm—onlyafewiterationsarenecessarytoyieldasatisfactorydesign.
SM has been successfully applied to optimize a number of microwave compo-
nents, the majority of which are filters [1–4] and impedance transformers [8], but
alsoantennas[9–11],etc.Asmentionedbefore,oneofthefundamentalprerequisites
ofSM isthattheunderlyingcoarse modelbefast, sothatthecomputationalover-
head related to parameter extraction and surrogate model optimization performed
ateachiterationofthealgorithmcanbeneglected.Forthisreason,equivalentcir-
cuitmodelsarepreferred.Unfortunately,reliableequivalentcircuitmodelsarenot
availableformanystructures,includingbroadbandantennas[12]andsubstratein-
tegratedstructures[13].Also,ifEMcouplingbetweenthedeviceofinterestandits
environmenthastobetakenintoaccount(e.g.,anantennamountedinacellphoneor
onavehicle),full-wavesimulationisprobablytheonlywaytoevaluatethesystem.
ApossiblelackoffastmodelsposesadifficultyforSM,becausethecomputational
overhead related to multiple evaluation of the low-fidelity model, particularly due
toPE,maydeterminethetotalCPUcostoftheSMprocess.