Вы находитесь на странице: 1из 8

A

The Need for Q Systems Perspective in

ntr

tii
ontrol systcins havc liad a profound impact on sociely a s theories, techniques, and algorithms havc niigrated from (he laboratory to thousands ofproducts.
As llie control community continues lo improve its
solutions, ideas iirc being gcnera~edfor a new era 01
applications that arc Fundainentally diflercnt from thcir predccessors. Where applicetions once employed control systems for
grcater performance, new applications require control systcnis
for their very cxisteiicc. Examples of these types of systcins are
ultra-agilc military aircraC1, large-sp;ice-slructure observalorics,
nnd formation-flying sp;icecralt constellations. This control-enahledclass of systcms (also called high-perforinancc systems)
is changing the role of contr~Isciencc i n engineering.
Control theoreliciaos and practilioncrs havc both enthusiaslically acceptcd the challenge of controlling high-performance
sysiems. The control theory community has focused on pushing
the limits ofcontruller design, seizing (and crcating) new opporturiilies provided by iiiorc sophisticated inathematics, more cfficicnt numcrical algorilhms, and iiicrcaned coinpuler speed. Io
doing so, conlrol theoreticians liiive found ways to iucorporatc
inorc knowledge ahout what wc know and do not know ahout the
syslein into controller design. Elegant optinial conlrol theory has
been devcloped lo dcsign coiitrollers that usc minimum fuel or
gel us from here to there in miniiiiuin time. Estimation theory has
shown us how Lo extract useful signals Crom corriiptcd signals.
Robusl control theory has bccn developed Lhat allows us to describe system uncerlainty rnathcmatically. then allows us lo incorporate this description into control analysis-synthesis
methods. Adaptive control, system identificalion. and inlelligent

control have shown 11s how to obtain and then iisc new knowlcdgc of our systcnis for coiitrollcr synthesis.
Even while continuing to bc studenls of new conlrol theory,
iiiany practicing contrul cngineers arc coming Lo a sobcring realimtion. When high-per~oriiiniiccsystems demand strict performancc rcquiremeots, no control theory in ilse(f may he
satisfactory. In these cases, a new iinalysis-synthesis fniinework
m u s t be cniployed. Systcins intist he iinalyzed lroiii cnd to elid to
unclcrstanil how hoth the syslems thcinsclves, as well a s their
controllers, may bc modified to realize the ultimate objcctives of
their application.
As a result, although traditional control theory continues to
focus much rcscarch on hettcr conlroller design, practicing cont r d cngineers are being forced tu hrwiden their perspective.
They are being asked 10 analyze all liclors within control-enabled systcms that can ultirnatcly affect performance. These fittors clcarly include controller design iis acentral component hut,
depending on the applicatioii, structures, optics, signal proccssing, fiult detection, aiid computer processing may also fall inlo
the contrul engineer's consciousness.
Forltmately, control practitioners are finding thcmselves analytically well equipped for high-performance systems aiialysis
with theoretical machincry adapted from controller design melh(ids. However, thc control theory community has not yet lbcused
atleiition on systems-control a
(Forlackofashort naiiieforsy
we w i l l abbreviate it hcrcin as S-C-A-S.) Understanding this
evolving gip bclween control theory and thc iiceds OS control
practitioners can Icnd to some very pusitive results. Thcsc illdude new rcscarch directions, a broadeiiing influeiice of control
theory on other engineering disciplines, and the ~ u l i i t i i ~10nmore
and more complex multidisciplinary problems.
In this article, wc examine the iiew role of control science in
high-performance systems and its implication for control theory.
In the next section, we describc the traditionally "scrid" rnle of
control scicnce ill systems clcsign and the evolving iterative systciii-conlrol design process used for high-perSormmcc systcms.
Next, we discuss several high-pcrSormance systems that use this
iterative dcsign process. Then, we describe pas1 work that i s rclalcd lo S-C-A-S, followed by an ex;imination of fiindamciilal
cnntrol theory concepts that nlay be extended a s part of an S-CA-S theory. Finally, we present conclusions.

From Serial to Iterative Design


Control llieorists primarily study thc analysis-synth
conlrollers within a serial framework, as illustratcd iii Fig. I .
Tr.
dilltionally,
' '
application-specilic enginccrs have designcd systems. Once thcy have completed thc design, they liiivc passed
two pruducls lo the control ciigineers: (I) tlic system itself and
(2) physical requirements that the system must iicliicve aster cont r d dcsign. The contrul engineer may then model the system
within a specific malhernatical framework, ineiisurc inputs and
outputs, and possibly even pokc and prod the system to learn how
i t behaves. A t the same timc, physical control requireiiients arc
translaled into the language of control theory (gain margin, rise
time, quadratic cost, infinity-norm, ctc.). The contrnl engineers
then apply the best control theory they know fbr the particular
application. After several iterations and joggling of fundiiinciital
tradeoffs (pcrformance vs. robustncss, etc.), thcy IiopeCiilly
x h i e v e sought-after performance.

December 1999

cd iicw classes nf systems to control


researchers, new cnntrol Lhcory hns developed. I:or cxample, the
need for undcrslandiiig locoriiotivc s t e m engines iuid ccntrifiipal governors led to linear stahility analysis I II.The use of fccdback aiiiplificrs in new electricid circuitry lcil ta cliissical control
theory 121. Thc iiccd to guide and track sp;sc vehicles helped
push the f(irmulation of slale-spiicc thcory aiid oplimal contml
iiiid esliinatioii 131. Lightly damped spacc structures Icd to dislributed paramctcr cunlrol methods 141. The cvcr increasing
complexity aiid accoinpanyiiig onccrtninly in MlMO systems
Icd tuum,p-synthesis, and othcr robust methods. Complex systans iii which iiiodcls arc hard to quantify led to fuzzy 151 and
ncuriil coiitrnl 161,
On llic other halid, thc scriiil dcsign ;ipproach has led to some
unfortunate conscqocnccs. First, Ihc "lasl step contrnl" approach
has limited tlic rule and influence uf co~ilruIscience. Must
iioncontrol engineers perccivc controls science as esoteric and
nun-intuitive. Thcrclure, ideas and techniqocs invcnled or pcrCected by the control coininunity have a hard time affecting noncontrols disciples. Sccond, arid pcrhaps iiiorc important for
conlrol science, the serial design appro;ich i s sure l o fail a s systems clcmand greater and grcatcr performance. hi thcsc cases,
ciiiitrol engineers inust work within the systcin dcsign Sramework lo help develop sysleins and contrnllcrs lhal w i l l allow rcquireinents tu bc iiicl.
The iiltcrilalivc l o the serial ;ipproacli i s iterative S-C-A-S.
Oncc ;in initial system design i s completed, control iinalysis i s
conducted using software inodels and tailorctl hardware expcriiiiciits. This antilysis leads to possihlecontrol andlor systcin inodilications nccded l o make the systcin reisible. Oncc ii feasible
system-controller s n I u t i i ~ i~si deinoiistratcd, additional changcs
lhat may uptimize [lie design arc cviiluiiled. This process conlinues iiiitil the dcsign space stabilizes. This alternative analysissynthesis approach i s illustrated in Pig. 2.
This itcnitive Crainework rcqiiircs conlrol engineers to be illvolvcd early i n the dcsign phase o f systems, to give input into
how a particular syslcin desigii may affccl conlroller design and
vicc v c r s ~It. i s only early in the desigil phnsc that major changes
can be made. Syslcm-conlroller trades rcquirc ii thorough knowledge of co~i~riil
options, frnrn Bock fccdback design to modern
tirnc-domain techniques, a s well as tlic in;ithcm;itical hackground underlying these melliods, but thcy also require the contrnl ciigineer lo know much inore lliaii coiilrollcrdcsigo. Control
engineers iiiust have in-depth knowledge n l specilic applica-

r
Control Science focus

system
Design

Controller
Design

Iterate on
Controller
Design

space structures embody dynamics that help disturbance energy


to propagate over the entire spacecraft, As a result, a11 suhsystems on such a spacecraft are dynamically connected. In addition, closed-loop conlrol changes the way these dynamics
manifest themselves. This is known as control structure interac1ion.Anearlyexampleofthisinteraction wason the NASAVoyager spacecraft (1977) shown in Fig. 3. A controlled scan
platform was mounted on the tip of a 2.3-m boom. To operate
correctly, both the platform controller and the boom had to be designed concurrcntly. Control engineers designed the platform
controller and simulated the spacecraft, boom, and platform in
closcd loop. Based on these simulations, the control engineers
modified the controllers and suggestcd boom alterations to the
boom designers.

SIM
Fig. 2. Nerarive analysis-synthesis al~prooch

Individual flexible structures oil spacecraft (such as those on


Voyager) led to spacecraft concepts that were dominated by
large flcxiblc space structures (LSS). Motivatedhy large flexible
space structure work, new spacc obscrvatories envisioned for
early in the next century will require unprcccdented control performance for instruments placed on top of these LSS. Optical elements on these large space observatories must maintain relative
stability to within a single wavelength of light!
Consider NASAs Spacc Interferometry Mission (SIM)
shown in Fig. 4. Currently under development at the Jet Propulsion Lahoratory, TRW, and Lockheed-Martin, SIM is one of a

tions. Important dynamics, hardware limitations, actuator and


sensor options, and computing power all affect the ability to
meet strict requirements.
The iterative framework also requires control engineers to he
able to develop, use, and interpret results from tools such a s
multidisciplinary simulation and specialized hardware testing.
Knowledge of general principles of mathcmatical modcling, including thc strciagths aaid wcakiacsscs of numerical methods, allows control enginecrs to ascertain thc effccts of system and
control changes on overall performance. Bventually, possible
solutions must bc tcstcdinactual hardwarc to validatesimulation
results; however, these hardware tests are rarely donc using a
full-up system. Rather, a specific test setup is designed to validate a particular solution or concept. Control engineers must
help design thcsc experiments so as to instill confidence that a
dcmonstration on a specialized testhed will translale Lo the actual
system once it is built.

Aerospace Systems
Voyager
Spacecran engineers rcalizcd the importance of concurrent
S-C-A-S with the advent of flexible spacc structures. Flexible

58

Fig. 4. One possihle design for the S1iace Interjkmetry Mission.

IERE Control Sy.stems

Actively Controlled
Optics (Strategy 3)

\k

/
Vibration
Power Flow

Active

cnce of reaction whccl disturbanccs. Reaction wheels are 11111-

Disturbance Source

I
or Passive

mentum exchange devices uscd to point spacecrall and maintain


I

Vibration Isolation
(Strategy 1)

Structural Quieting
(Strategy2)

Fig. 5. Vibration control ,strategyfor SIM.

.
:

y (Meters)
1 0.5 0 -0.5

L
Fig

new generation of space observatories that will use optical interferometry to synthesize large optics using only a series of small
optics. SIM consists of a series of light collectors and other uptical elements placed on top of a flexible IO-m space structure. By
physically moving the light collectors within a two-dimensional
surface in space, the observatory will partially mimic the scicncc
return of a singlc IO-m telescope: however, to employ required
signal processing techniques, individual optics on the structure
must maintain relative positions and orientations to within
nanometers.
Thiscxtremely stringentrequirementmustbe met in the faceof
disturbances from attitude control actuators, therinal gradicnts,
solar pressure, microdynamic structural snaps, and other disturbance sources. As is typical for high-performance aerospace systems, iterative S-C-A-S is crucial for SIM. First, it is not clear
initially that there exists a control method that can meet thc systems requirements. Second, due toits highcost, thedesigncannot
bencfit from trial-and-error experience of large-scale production.
As a specific example of how iterative S-C-A-S is being uscd
for SIM, consider the problcm of optics stabilimtion in the pres-

Decsnhr 1999

attitude. As a reaction wheel spins slower or faster, it not only


produces the torque necessary for spacccraSt pointing, but it also
imparts unwanted disturbanccs into the system due to imbnlances and rriction within the reaction wheel mechanism. For
most spacecraft missions, these disturbances are negligible Lo
mission succcss; however, this is not lhe case for intcrfcrometers. Due to ultra-stringent perlormance requirements. even
small dislurbaiices ciiii be devastating to the mission. Reaction
whecl vibralion attenuation is generally bclieved to require three
complementary vibralion control strategies (Fig. 5 ) : ( I ) active or
passive reaction wheel isolation, (2) aclive or passive structural
quicting, and (3) active control of optics. Clearly, the design of
controllers Sor tach ol these elcineins Sundamentelly iinpacts the
overall systcm design.
As mentioned previously, the lirst goal of itcrative S-C-A-S
is toprove thcexistcnceolasolutiun. This is doiieforSIMusing
a combination of simulation and specialized hardware testing.
The kcy tool Sor simulation study is a inilltidisciplinary model of
the proposed interferometer (Fig. 6). This model consists of a
structural finite-element model, a linear optics modcl, and acontrol inodcl all lied together wilhin a common software framework. These models make it possible toquantitatively predict the
effect of incchanical disturbanccs on opticol performance mctrics in both open- and closed-loop configuralions. The major
testbed for vibration sludy (shown in Pig. 7) contains all neccssary systems to perform space-based astrometric measuremen1
171. Using this testbed, proposccl vibralion illtentiation solutions
can be physically impleincnted in hardware. Once a solution is
demoiistrated in bolh software and hardware, control engineers
can work collaboratively with oplics and structures engineers lo
optimizethedesig~iusingavarietyofcontrollaws,sensorsandactuators, structural configurations and materials, and/or npticiil dcsigns 181.
Other Upcoming Control-Enabled Aerospace Systems
Structurally connectcd optical interfcrometers are only one
class of systems in which iterative S-C-A-S is needed. In the future, optical interlerometer observatories will bc constructed

Fig. 7. Vibration .snrdy tcstbrd for the Space 1ntel:feromctry Mission.


Vibwtiun control .st,utcgirs and hardware are flemnnstruted 011 this
testlied 0,s proojthat nnnomef<~rstribiiiry wquirements cun be met.

59

Pig. 8. One coircepf for Terresrrial PIo~relFiwler.

without any structure connecting them. Instcad, they will usc a


coordinated fleet of spacecraft flying in precision formation.
Controllers will monitor the position and attitude of each spicecraft and keep the ensemble working as vnc unit, iis if they wcrc
rigidly connected. One such mission, Terrestrial Planet Finder, is
shown in Fig. 8. Its mission will he to direclly detcct Earth-likc
planets in other solar systems. The mulliple-agent framcwork is
also being studied for a wide variety ofothercuntrol-e~lalrlcdawospace applications. These applicatinns include fvrmatinii flying
high-altitude aircraft fix communication networks 191 and lormation coordination of robots ior Mars exploration [ 101(Fig. 9).This
article focuses on acrospacc applications; however, importatlt
challenges are sure l o cnme from addition;il nonscrospacc applications such as semiconductor mnnofacturing, chemical processing, or biomedical engineering.

Past Research
Thecontrol theory coinmnnity has yet to cmhracc S-C-A-S a s
a unificddisciplinc, sucli as robust control theory or systemidcntificalion theory. However, past rcscxch motiviitcd by emerging
applicetionsisleadinglowar~lsuchadiscipliiic.Thefollowing
is
not mcant to be a compreheiisive review, but rather io give examples of some work related to an S-C-A-S disciplinc. A comprehensive review would surely includc more authors and works.
Large space structures tend l o have closely packed, lightly
damped modes that start appearing at low frcqucncics. This can
Iced to instabilities in controlled struclures, as the bandwidth of
[lie cuntrollcrs oftcn ovcrlaps several structural modes. Addilionally, lhese s y s t e m are olten lionminimuin phase. Moreover,

largc spacc structures are inlinile dimcnsional systems described


by distributed-paramctcr modcls (par1ial-di lfcrential equations):
howcvcr, the controllers ulliniatcly ;ipplied to lhcse systcms are
only of finite dimension. t;initc-diincnsionol controllers inay ;iffect not only the modes considered i n the controller design process (e.g., via reduced-order modeling), bui also the (infinite)
inodcs not considered in the dcsigii process, aprohlem known iis
spillovcr. Thcsc prcviously stable, unmudclcd modes can suddenly become unstahlein the presence ofconlml. The lhcoretical
issues associated with a11 olthcsc prnbleins drew mnch allcnlion
in the 1980s and early 1990s. Prom llicsc efforts came a host o l
new theories and algorithms in the areas or distributed-paraiiicter cnntrol, spillover reduction and accoinmodalion. model rcduclion, scnsur and actuator placemenl, and frcqucncy shaping,
among olhers. Good rcvicws nf the field are givcn in Mcirnvitch
[II1and .loshi (S.M.) 1 121.
Conlrol-s&uctore interaction was one iircii that highlighted l o
researchers tlial system idcntification, reduced-order modcling,
and controllcr dcsigns are i o 1 independent problcms. For cxainple, as descrihcdin Skcltun and Hu 1 I3 1, reduced-ordcr iiiudcls for
controllcr design caiinol he conslructcd without knowledge of
which inputs the modcl must propagale. Uniorluniltely, thcsc illputs arc cxacrly tlie control ior which llic model is needed in the
h t placc. Iterative inetliods for modeling and cvntrnl are discussed i n Skelton and Hu [ I 31, Liu and Skclton [14], and Zhu and
Skelion IIS]. Similarly, system idenlirication and rvhust control
were recognized l o he coinplcnicntary problems. A good scrics nf
paperson thia subjectisgivenin Kosutelsl. 1161.0nc itcrativcapprnach to syalcin identification and robnsi contrnller design is
given hy Bayard CL al. 1171, who dcvcloped an idlemaling curveM n g and contrnl design procedure io ~xoducca scrics nf cnntrolleis that have monotonically improving robust performance.
Control-striicturc interaction also molivatcd a fcw rcscarcliC
S
I
to think about hvw hoth structures and controllers could he
ilcsigncd concorrcnily. Meirovilch [ I I I contains a good cxpanded lis1 ofresearchcrs who havc worked i n tliis lield (prior l o
1990). Hale et al. 1 181considered tlie problciii of optimal structim-control fnrcc design using a scalar cos1 functional fnr menenvering a llexiblc spacc slructure from s p e c i k d initial
condition l o specified rinal conditivn in a given amounl 01lime.
Millcr iind Shim [ 191 considered the problem of coiiibincd structural inasslconlrol-energy optimization for reducing iiiass and
suppressing vihralions using gradient-based search methods.
Lim and Junkiiis 1201 considered robustness optiinizalion of
conlrol-slriicturc design. Thcy optimized lolal mass, shhil ity robustncss, and eigenvalue sensitivity with respect to structural parameters, control partimeters, and actualor locations. Maghami
et al. 1211 sludied thc combined coiitrol-sIriiclure design using

dissipative conlroIIcrs. Smith et al. 1221 antl Grigoriailis et ?iI.


1231 approached tlie problem i n a diflerent way. l h c y assumed
that an active controller lias been designed a priori that meets
performance specifications. They then proceeded to change lhe
active controller, as well a s redesign the structure, to reproduce
the pcrlorinance of the original controller and striictiirc, only
now with less control power.
One of the firsl communilies tu undersland that systems and
controllers niiist he analyzed and synthcsked cuncurrcnlly was
the military aircraft industry. Many lhigli-pcrforin;incc aircraft
arc designed open-loop rmsruhk for extreme agility, As such,
these aircraft arc extremely clcpendcnt on their controllers. Furthermore, cvcry design decision, froin clioice of birdy matcritils
to aircraft shape, closcly interacts with conlrol clcsign. I n the past
several years, the aircraft design community lias Icd llie new
field of multidisciplinary design opliinization (MDO). M D O
aims to explore the interactions hetwccn striiclures, acrodyiiiimics, flight mechanics, tlicrmal dynamics, and controls by dcveloping analysis tools i n a comiiiiin software framework.
NASALangleysDivision o f Multidisciplinary Optiniizalion
1241 defines MDO theory as coinposcd of mathematical modcling, approximation thcvry, computational tradeoff theory, s n i w
reanalysis, sensitivity theory, and optiinization lhcory. M D O
rnathcniatical modeling aims to create suites of disciplinary
models that can he integvated into i i singlccnviroiimciit. Approxiinalion research 1251 a i m to acheive system performancc using
the niinimuiii amount o f needed information from ~ a c hdiscipline. Computational tradeoff aims to understand the rekitintis h i p h e t w e e n c o m p u t a t i o n c o s t antl a c c u r e c y i n
iuultidisciplinary simulation tools. Smart rcenaly.
duce the computalional load of analysis using multidisciplinary
simulation. Sensitivity theory [ZOl allows lhe rnatheinatical representationofthceffectoiachangc in a p a ~ i i n e l e r i n o n efieldon
a par;imeter in mother field (e.g., wing shapc on inaterial sclcction). Finally, optimization theory 1271 iiiins l o find efficient
ways to decompose, search, and optimise over very large design
spaces. A good series ofarticlcs in this field i s av;iilable in Livne
1281. Although often inentiancd as a vital component o f M D O ,
control theory has not been well coiiiiectcd Lo tlie M D O coinmw
nily. This disconnect, as well as the rapidly growing complexity
ofthc problems, has usually l i m i t e d M D 0 studies iiivolving control to simple P I D conlrollers. Recently, Masters and Crawley
L291 studied evolutionary design of cuntrolled slructurcs iisiiig
gcirctic a l g o r i t h m to tunc slructtiral parameters, controller parameters, and sensor and actiiiitor locations. Gutierrez 1301c o n biiicd dislurbancc, uncertainty, and sensitivity analysis, a s well
a s integrated modeling, to show how to better design controlled
high-precision struclurcs.
Finally, a number ufollier arcas have hccn studicd. To make
coirtrol systems truly robust, failures must be accul-ately imd
swiftly detected, isolated, and accoinmodated. This h a s Icd to
failure detection and isolation (FDI) theory 13 I].Estimation theory, mulliplc hypolhesis testing, and analytical redundancy liiivc
all been proposed a s methods for lault delection and isoliition
Most of these methods have been clcvcloped from conlrol-estiination theory. Optimal sensorhicluator placeincnt can have a
large impact on control performance (e.g., [32]).In addition,
computational realities such as finite word lengths and rouiid-off
error can significmtly affect controller pcrform;mce. Efforts to a
priori account for these cffccts iii the design or optiinal cuntrol-

Deceiiihw 1999

Icrs and estimators have beeu studied by Moroncy CL al. 1331 and
Liuetirl. 1341.l.o airlSkcllon (351rcporleddesiglleconolnicsby
considering the combined oplimization of contriil laws and illstrumelit selection. They assume that iiistruincnt cost i s directly
rclatcd to signal-lei-noise incasures of the instrunents.

Fundamental Control Theory Concepts


At present, iiiost pracliced S - C - A A i s dune ad hoc by ilerating hctwcen desigii and control until a suitable solution i s round.
Guiding analytical tools and design algorithms to aid designers
specializing in varied multidisciplinary applications would be
very helpfill lo converge to suitahlc designs more quickly. As
partially shown h y lhe studies ollhe last section, we may exploit
fundamentel ccintrol llieory concepts toward such design tools.
These concepts include sensitivity, uncertainty, robustness, and
ontim;ililw

Sensitivity
One or the f i r s l theoretical contributions of control theory was
the realization lhat fccdback Icads L o sensitivity reduction. This
concepl i s also extrenmly important in S-C-A-S. Scnsitivily
analysis i s already being explored in multidisciplinary design.
However, the addilion oifeedhack conlrollers changes the probIcni. Performance inusl not be redieally altered h y syslcin variatioiis clue lo structural uncertainty, cnvironment;il disturbances,
material property changes, sensor and actuation degradation,
md s o on. I t i s unclcar, Iiowever, if t h i s desensitizing i s best
achieved through change OF plant design or inodificd cniitriil authority. The extension of sensitivity theory to include both systcin design and controller design options could he f r o i t r d

Uncertainty
The need to capture oncerlainty in systems has played ii large
role in control theory. Uncertainty illso affects system design.
Ior example, in critical carly phascs of projccl design, multidisciplinary inodels are tlic only design tools av;iilable; however,
pcrformancc predictions using muhidisciplinary models are 211ways somewhat tiiicertaiii clue to several factors, including coinponcnl uncertainties, modcl reduction, and discrclization. As ii
result of these unccrkiinties, overall system designers tend to
ovcrdesign subsystems to accoiint fur uncertainty. For c x a n plc, optics arc made smoother than necessary, materials are made
stlotiger than necessary, and sensors arc made less noisy than
necessary. This conservatism also Icnds to overly expensive systerns. Control theory has a similar situiition. Control engineers
iiiiist Irade off conlrollcr perfnrmance for unccrkiinty rubustiicss. If uncertainty is overestimaled, resulting controllers cannot
meet as slrict pcrforinuncc targets as would iithcrwise hc possible. H o w to hcsl describe plant uncertainty and its effect oii coiitrol synthesis i s slill an active area of research in the control
theory community. Continued study of how to systemalically account for individual subsyslein unccrlainty i n coiiihincd systeinlcontrollcr design would hc very usefill.

Robustness
Using (he concepts of sensitivity and uiiccrtainty, rohustcontroller theory has been developed. Robusl colitrollers guarantee stability andlor performance for edmissiblc perturbations
within a predefined set. This conccpl coiild bc exlendcd to S-C-

61

coursc, wc should no1 aim to include every possible interaction


from every possible discipline. Meaningful and tractable problem formulations will he important contributions to an S-C-A-S
discipline in Lhemsclvcs.

Conclusion

Fig. 10. Modified ve,mion oJtrudifionrrl rohu.rl c o n l m dirrjimm.


~

A-S. Consider the diagram shown in Fig. I O . In the center is a


nominal system design or "plant," P.Robust control theory deals
with representing uncertainty, Axand then using chis rcprcsentalion to design arobust controller, K . Incorporating system design
options adds a new degree of freedom, reprcscntcd as 6P. The
combination ofnominalplaiit, P, and asct ofdesign changcs,6P,
resultsinanewgencralizetlplanl.Nolcthat6Pciinbecontinuous
(e.g., structural plant with continuously varying material d a m p
ing) or discrete (e.g., design option 1 or design option 2). Thc
choice ofSP has repcrcussions on both uiicortaiiity modeling and
controller dcsign. For examplc, one system design @P,) may
lead to it class of uncertainties, A , ( P , 6 P , ) ,whcrcciis another dcsign inay lead to another class of uncertainlics. A2(t',6PJ The
specific uncertainty set then affects thc design of a controller,
K , ( P , G P , , A , or
) K2(P,SP2,A2).Even without thc plant design
choices,SP, robust controller design is a difficult prublcm; however, the ability to develop systcms that guarantee spccific u ~ i certainty classes may lead to a combined system-controller that
can guarantee overall system rohustiiess.
Optimality
Both the notion of optimality and methods of oplimization
have played central roles in control theory. Oplimalily isonly defined with respect tu a pcrSormance metric. hi control theory, wc
use anumbcrofmetrics (e.g.,X2,Ww,
aodC, norms). Systems design must also be optimal with respect to a defined performance
inelric. Hctcrogeneous subsystcm inetrics inay have to be aggrcgated lo an overall system mctric. Alternatively, gross system
mctrics may need to bc evaluated a1 the subsystem level. Dcvclopment and iinderstandingolineaningfulmetrics aiid their use in
S-C-A-S will be very importanl.
Optimization aiid control theory have been lied together with
the insluelice of calculus of variations, dynamic programming,
linear atid nonlinearprograinming, and lincar matrix inequalitics
(LMI). New optimization theory will be imporlant to S-C-A-S
Simultaneous plaiiWcontroller dcsign problems inany times lead
to complex constrained optimization problems. Solutions are required Sor these problems that are bolh efficient and reliable. Of

62

As systems demand greatcr and greater perlorinancc, control


sciciicc takes on much more imporlance within system design.
Rather than being enhanced by control, new systems are enubled
by cotilrol systems. Practicing control engineers are being asked
to evaluate not only controller changes but also system changes
that can influence ullimate performance. As a result, theory and
tools for integratcd S-C-A-S would be very hclpful. Control scicncc needs to broaden its perspective and sec itself as an inlegral
part of overall systems design. It is hopcd thal this article will
help lhis reorientation proccss.
Finally, in kccping with tlic topic 0 1 this special section,
"Bridging the Cap Between the Theory and Practice ofContro1,"
this article rocused on thc implications of system-controller
analysis-synthcsis on control theory. In addition, the changing
role of control sciciicc also has hroad implications for control education, the definition of control science, ;uid the relation bctween control sciciicc and other disciplines. It is also important
Sur thc control community to continuc to address thcsc issues.

Acknowledgments
I would like to thank Tariq Samad for inviting mc to contribute lo this issue. In addition, I thank David Bayard, Mehran
Meshahi, Edward Mettler, and Edward Wong for helpful discussions, constructivc coiiniients, andor background material. This
manuscript was prepared at the Jet Propulsion Laboratory, California lnstitule of Technology, under a contract with the National Aeronautics and Space Administration.

References
I. I.I J.C. Miirwell. "On C O Y C ~ O ~ S Phil~~,~onhicd
."
Mapacirir. vol. 35, m.
385-398, 1868.
121 G.P. Franklin, J.D. Powell, and A, Emani-Naeini. Fc'ecdhackCo,rlml of
I)ynandc Syrler,ir, Addison-Wcsley, 1994.

131 A.E. Hryson, Applied Oprirnnl Coritml: Optirnimtio,i. Ertimalion, urrd


Cmrrml, Taylor and Francis, 1981,

141 V.B. Marks andC.R. Keckler, NASA !Vork.?hopon Distrii,ufedPr,ra,ne~


Ier'Modvling ami Cmtml of l+.tihle Aerospace Syslenw, Pasadena, NASA
Cnnfcrence Publication 3242, 1994
151L.X. Weng,A Cormein A,LLyLogicasdConIroi, PrmIiccIIdI, 1997.
[ h ]B. Koska, Neurd N e t w ~ kmid
, ~ Fuzzy Sysrcrizr, Preotice-Hall, 1992.

171G.W. Neat, 1.W. Mclady, and B. Lurie, "Vibration attenuation approach


for spacobomc qxiciil inlcrfemmeters."
Trirns. Contr. Syst. Techtrol.,
v d 6, ria. 6 , pp. 889-700, 1998.

[ R I S.S. Joshi, "Optimal doniper plisement for spaceborne intecferamctcrs


using 11-infinity imrm optimization," submittcd to Anierimn Conlmh CO$,
Cbicaga, 2000.

191J.D. Wolfe, D.S. Chichkii, iind J.L. Spcycr, "Decentralized C O ~ I ~ O I ~ Clor


IS
vchiclc hrmatirm flight," Pmc. AIAA Gaidmce, Control,
'md Naui,qaIio,i Ch$,San Diego, CA, ATAA Papcr 96-3833, 1996.
~iniiiiiiiiietliieiial

I 101S . S . loshi, "Systcms theory iippmachfornulon


ination control in unccmin mviionmciits,'' submilt
i r s arid Auronmion. San Francisco, 2OOll.

IEEE Control S y s t e m

[II]L. Mciroviteh, Dynamics am1 Control (If S f r r ~ r ~ n ' eWilcy


r.
snd Sans,
1990.
[I21 S.M. Joshi, Contml < $ L u g e I~'1exihlcSpwecrafr ( L ~ c t a r eN o m
Conrrol and Information ,Sciences), vol. 131, Springcr~Vcrliig.1989.

1261 R.T. Bieilran, J.S. Samareh, and L.L. Green, "Parallel computation of
sensitivity derivatives with application to acrorlynimic optimimtion of il
wing," NASA Arms Coiipulurionol Aerosciences W w h h o p , Augosl 1998.

iti

1271N. Aleniindmv, "Mullilevel mcthoils for optimal design," E,icyclopdia


C.A. Ploud;ia and P.M. Piirdalos, (eds.).Kliiwcr Auidernic
r,~Oi~timiizntion.

[ 131R.E. Skeltonand A. Hu."Modeliogstructures ~ ~ r ~ ~ l i l r ~ l t l eConisig~~,).


PLttLI)T
a r r ~ i s ~ n ~ c t uyo^.
r e s20,
.

1.3. pp. 303.309, 198s.

[ 141 K. Liu and R,E. Skelton, "Inlegrated inodeling and ~ ~ n t i o design


ll~i
withapplication to Flcxiblestructurccuntrol,"Auf~~,n,u~ic~~,
vol. 29, no. 5, pp.

1291- I 3 14, 1993.


[ I S G.M. Zhu and R.E. Skelton, "Integrated inodeling atid Contiol (07 the
large spacecraft control laboratory experiment facility,"
frol, ond Uyrrarnics, vol. 17, no. 3. pp. 442-450, 1994.

Guidmrce, C m -

1161R.L. Kosut, G.C. Goodwin, nod M.P. Polis (Guest Editors), "Special issite m system idcntificatian fbr robust ~ ~ i m
dcsign-lntmduction,"
il
IELP
liuns. Aufoiiiat. Conn:, vol. 37, no. 7, 1992.

Publishcrs, i n press.

12x1li. Livne (Guest Editor), "Spcchl ismc on multidisciplinary deiigii optimication,"AlAA .I.Aircraji, vol. 36, no. I. 1999,
[29] R.1'. Masters aiitl H.?. Cmwley, 'Bvolntionary design of conrrolled
slmcloras,"AlAA J, Airwafi, vol. 36, no. I, pp. 209-217, 1999.

1301 H. Gutierm, Performance Asscssment and Enlranccnienlof Precision


Contiullcd Struclurcs During Conccptual Dcsign, P1r.L). Uisscrt;ition. Departincnt of Aeronautics and Astronuilics, MIT, 1999.
[3 I I II. Pallan, 1'. Prank, onil R. Clark, Fctuit Uiqnosi.?in Dyauwic Sysfenir:
Theory and Applicofiun, Pientice Hall, 1989.

1171D.S. Bayard, Y . Yam,ilndT.. Mettler."Ac~itciionhrjointoptin~izntion [321P.G. Msghimi ant1 S.M. Jonhi,"Scnsor;md ilctuaturplnccinenl foortlexi~
of idenlificetiun and robust ~ ~ i i t d ,1E6h
' '
Tvmis. A~ttoiiior.Co,zfr., vol. 37, ble spitcc strwtwcs,'' IEEE 'lrn,is. Aemvpacc and Eluctronic , S y f . , vol. 29,
no. 7, pp. 986~991,1992.
no. 2. pp. 345.351. 1993.
[ 181A.I.. M81c,R.J. Lismvski, m d W.8, Dahl, "Opliinal Siniiiltiiiiews stcuctwal ;md c m t i o l design of m;ocuvering flexible spivxcraft," .I Guidmce.
Conool, a,tdDynmrici, vol. 8, no. 1. pp. 86-93, 1985.

[33] P. Moroney, A. Willskey, and 1'. Hunpt, "Round-off iiuisc acid scaling in
thcdigital iiiiplcmcntatiiiii ufcoiitid compen\alors," I
Speecli, nnrldipioi Processing, YUI.
31, pp. 1464-1477, 1983.

[ 191 D.P. Miller iiiiil 1. Shim, "Gnaliolt-bose~lcombincdi l i i i ~ l ~ ciind


a l control optimization." J. Guidrrme, Co,ioul, m d I>y,zamics, vol. IO, no. 3, p p

1341K,X Liu, R . t . Skclton, t w d I<. Cirigoriadis, "Oplimnl ~onlrollersIbrlinite word-length implcmentotion." I E E E T m m Auforrroi. C m i r . ,vol. 37, no.
9,pp. 12~4~1304,
1992.

291-298, 1987.
1201K.H. I.im ancl J.L. Juakins, "llobustncss oplimiriltion ofslnlctul-Jl and
control parsmctcrs." .I Guidance. Cormol, a,idl)y,ra,irios. vol. 12, inn. I,pp.
89-96, 1989.
1211P.G. Maghami, S.M. Joshi, and D E . Price, "lntegnilcd coiitrols-stiucturns design metl~odolugyf o r ilcxible spacecraft," .I. ,Spuc~crufr orid
Rockas, vol. 3 2 , no. 5, pp. 839-844, 1995,
1221M.J. Smith, K.M. Grigoriadis, and R.E. Skclton, "Optimal ,nix of passive and naive control in striicmes," J . Gwidmxe, Conrr,,l, r m l I)yimnics,
vol. 15,m. 4,pp. 912-919, 1992.
1231K.M. Grigaiiadis, G. Zhu, andR.E. Skelton,"Oplirnal ndcsignollineia
systcms," .I Dynamic ,Sysre,nr Meii.sur-erne,ir irrid Contml-l)uns. ASME,
vol. I 18, 110. 3, pp. 598-605, 1996.

[24] hltp:iif~aad~www.larc.nils.?.gov/mdob/MDOHl,1999.
1251J.J. Korlc, J. Dum, A. Salns. N. Alcnmdrov, W. I'nllett, and G . Orient,
"Multidisciplinary approach tu lincar amspike nol%le optimization." 3 r d
Joint Pwlxdsion CO?$,AIAA Paper 97-3374, Seattle, 1997.

December 1999

1351J.R. Lu iiiiil K.E. Skdton, "lntcgrated instlamcntation ;md coiitr~ldesip," Irif. J. Co,ur., vol. 72, no. 9, pp. 799~814.1999.
Snitjay S. Jushi rcccived tlic H.S.E.E.
dcprcc with distinction from Cornell University i n 1990. H e rcccived
M.S. (1992) and Ph.D. (1996) degrecs i n Elcctiictil lingineering from UCI ,A, During this graduate slridies, hc
joined the Jct Pmpulsiun Labanitory, Califcmiia Ins&
tatcofl'ecl~nalugy,in the Guid.loceaadCaiitlol Anelyhis Group. Correnlly, Dr. hshi i s II Technical I.cad i n
the NASA Origins Program h,r spme hiised ioterlcrometry, rocusing on mul1idisciplinal.y modcling and con~incclianicillinstrunicnts. 111ailrlitiun, hc is ti researchcrin
the Mars robotics pcogram, focusing 011 coordirutlcdautoiiomous rnvcrs. Dr.
Joshi's research interests incliitle inultidisciplintiry modeling-designcontrol, spacecraft and robot autonomy. iailurc detectionand isokninn, arli~plive syslems, and rlistiihutedcontrol and wising. Dr. Joshi serves 11s tin Associate Editor on tlic IEEE Control Systems Society Confcreocc Edilorisl
L3"Zld

63

Вам также может понравиться