You are on page 1of 4

r

Thomas Ting
ver the past scvcra1 dccades, control theorists have gcnaated an iiiiprcssive array OS new techniques [hat span a wide variety oSSields, including classical control, optimal control, adaptive control, robust control, nonlineer control, hierarchical control, and intelligent control. Concurrcntly, however, one of the dissppointincnts within sonic segments of the ctintrol community i s the perceived slow rale at which rhcse techniques have bccn adapted into industrial practicc. The purpose of this special scction i s to address this issuc and propose somc siiggcslions to ass i s t in tailoring thcory for practice from an acrospacc perspective. Control thcnrists oftcn wonder why industry i s not always anxious L o implement the latest dcvelopmcnts in control theory. The bribic ,' answcr i s that control engineers in industry will almost 81ways iinpleincnl the simplest, most cost-cfrectivc control mcthodulogy that satisfies their iiiiincdiate goals. Many applications liave huge software, production, and personnel training incentives that minimize the magnitudeof changes from a previous gencration of designs. Thus, rhc next-generation design inay oftcn he structurally identical to'thc previous one, with changes only i n some tuncd parameters (easily changeable in soltware). In more drastic cases, the changes inay include ail hoc reSinements to 8n existing control algorithm structure to trcal some dil'ficult operating conditions. Rarely, though, does ;in industrial application control design start Srom a blank sheet olpapcr.

ctiv

New control tcchnologies arc thus cmhnicedhy industry only


when there i s n clcpr aiid present nccd that c;lnnot be adcquately
addressed through currently employed methods. This necd can hc performance bascd (improved closcd-loop syslcin characteristics), c~impuretionallyhascd (less computation required), dcsign based (reduced development time), cost bascd, end there arc any iiuinbcr ol' othcr possibilities. Therefore, the ability of control theorists to inSluence industry adaptation o f new control techniques i s somewhat analogous tu h a t of ;in individual attempting to push on a ropc. It's much easier ii the individual holding tlic othcr end US the ropc i s willing to pull.

A Process for Turning Theory to Practice


The entire proccss of developing and using new control theory can he divided into three major steps: innovation, facilitalion, ;md implementation. The first step essentially refers to theurctical development o f ti new control concept. The second involves applying new control concepts L o prcvioosly untried applications and cvaluating the rcsults. The final step i s the procedure of embedding a new approach into the standard engineering design process. These three steps are oftcn performed by different groups of individuals. A standard communication tlowchart among these three groiips i s shown in Fig. I.Note that the facilitators bear primary responsibility for communicating with the other two groups. Some additional deteils ol'the three steps are discussed below.

I I

\ \ \ \

fit for a speciSic application, and on-the-fly ad hoc modifications are required. Engincers who conduct this work must possess both sufficient application-specific domain knowledge and relevant control theory capabilities. Good communication skills and tcchnical versatility areessential for this role because thesccngineers musi interface effectivcly with both the innovalors and the implemcntcrs. implementation
I
I I I I I
I I I 1

I I
I

I I I I I

I
I I I
I

I
I

I
I I
I
\

I
I I
I

\ \ \
\
\

This step consists of the nuts-and-bolts integration o f a new design approach into a mainstream engineering procedure. This is ultimately the respoiisihility of thc ciad users (typically design or developmcnt cngineers in an industrial corporation, national laboratory, or governmcnt scientific center), because they are the ones who will be using the approach on an ongoing hasis. However, thc interface hetwccn the facilitatnrs and implemcntcrs is crucial because the facilitators must take care to provide adcquare resources to ensure a completc technology lransfer. This may involve preparatinn of user-friendly software design tools, thorough yct readable technical documentation, on-site technical support, and technical training for pcrsonnel receiving thc new technology. As a last rcsort, a facilitator may lranskr to a project and begin working as an implementer, hut this is not always a prefcrred approach. Communication is the Key The absolute key to thc slructure prcscnled above is coinmunication. 111this framework, all ofthe pull has to come from the end user. The control enginecr synthesising the production designs (the implementer) has to convey his or her needs to ii facilitator (typically an advanccd dcvelopmenl engineer or engineers at an industrial research ccntcr). This facilitator then bceins to search for oossible methods for solviiie this . process, I the.txoblern. Durinr 1 or perhaps oncc candidate stnitcgies have bccn identified, the h cilitator may communicate these needs or seek assistancc Crom an appropriate control innovator. It is cssential for each party within this process to undersland and address the w e d s and requircmcnts of its "customer." They iiiust all "go the extra mile" to cnsiire that the customer can operate indepcndently with the new technology being dclivered. If not, after a few iterativc attempts lo resolve tbe problem, the engineers hecomc frustrated with the new technology and either seekalternativesolutions or try to gct by with modifications their existing solulions~ In reality, there are two m:~,jorstruclural obstacles that interfere with this ideal conlrol theory development and implemcntation process. First, control engineers in industry do not place a high priority on publicizing their technical performance roadblocks. I n an industrial setting, publishing and presenting technicalpapcrsisnotan overwhcliningpriority.'lhisisespecially true in situations whcre time dedicatcd to preparing publications would conic at the expense of existing pro,ject work. Second, in many industries, proprietary in~ormation concerns precludc the preparation of technical papcrs. Companies are highly sensitivc to the ever-increasing threat of global coinpctition and are very proteciivc of their classified information. Thus, passively identifying technical issues rclcvant to industry may he extremely difficult for control innovators to do.

:P, 1. Communication flow.hart

tbr control theorv devclmment

and implementation.

abstract in nature, is usually conducted by university professors and graduate students. The results arc communicated to others in the technical communiiy via technical presentations and conlerence or journal papers. Thc ideas emanating from this step provide a continuous flow of design options for facilitators. The communication link between the innovators and the facilitators quality of this The facilitais essential to maintaining tors must communicate their necds effectively to thc innovators, who in mnSt beresponsiveto those needs toellsurethe highest potential applicability of their idcas. It is still possible for innovators to promote new concepts through technology push, but the probability of success is greatly dependent on thc pull at the other end
~~

Facilitation This step consisis or maintaining awareness of the state of the art in control theory and analyzing the relative benefits n S new techniques in applications where performance or design improvements are required. Much of the woi-k in this step involves ensuringcompatibility between assumptions required by the theory and the reality of a specific application. Critical components of this step include model development, controller synthesis, analysis, simulation and evaluation, and model and conirollcr order reduction. Often, theoretical approaches will not he a perrect

46

IRKR Control Sy.stems

Special Section Contents


The thrcc articles in this special section havc hcen selected to provide a rangc of viewpoints about some of the current and fiiture challenges lacing the control community. The authors have pretlomiiiantly aernspace or aciidcmic hackgrounds, so their illputs may not necessarily rcflect the needs US othcr control application domains. Although I did not always agree with tlie authors vicwpoints, Ifound each d t h c articles to he enlightening. I hope you, thcreader, have asirnilarcxpericiice. I3ricf introductions to each of the tliree articles follow. Tactical Missile Control for new As described above, a successful application di~main control thcory has two requircments: a performance-based nccd aiid a iinvel control approach that can help meet this nccd. Sincc it i s unreasonable to expect a control cngiiiecr to addrcss hoth US these issues simultaneously. the initial question i s usually one such as: Arc my current control algurithiii synthesis approaches capable of producing designs Ihat meet my present (and possibly luturc) performance requirements? If the answer to this question isalfirinative,nnostlikelynotliing will changc. Iftheanswer i s no, the control engineer inay begin a search Sor p~issible alternative methods. Based on this approach, the tactical missile industry appears to bc a textbook example o f an application domain poiscd to deinand and reap the benefits of advanced control theory. Many current inissilc algorithm implcrnentations s t i l l feature variations of long-standing classical dcsigii approaches such as proportional i i i v i g a t i i i n guidance and PIU control. The cffectivencss o f such designs i s undeniable, and, until recently, they were morc than capable of achieving desired performance levels. But the fundamental performance demands of this industry are now changing. For example, the increascd agility of targets requires that missiles possess an unprcccdented level of agility over an expanded envelope of flight conditions. Concurrently, the desire for hit-to-kill capability requires devclopmcnt o l guidance strategies that can achieve pinpoint accuracy while siinultaneously meeting terminal-state constraints needed for penetration, Ridgely and McParland elucidate many ollhc control sysleni needs for present and future statc-of-the-art missile systcnis. They discuss general performancc requircments and design constraints and review the current algorithm synthesis procedurc. The erticle culminates with an identification nf several key enabling control technologies of particular intcrcsl to tlic tactical inissilc industry: lixed-structure control, paramctcr optiniization methods, linear paramctcr varying control, nonlincar dyilainic inversion, and integrated guidance and autopilot design. Past and present work in each of these areas i s reviewed, as well as outslanding challenges for the future. High-Performance Design: Flexible Space Structures Control system design has historically heen treated, in an overall system context, as a last-step optimizing technology. In most applications, thc open-loop plant structurc lias long been established and Sroren heforc the cnntrol engineer i s asked to synthcsize a controller that will optimirc tlie closed-loop syrtern performance. A I this point, tlic control engineer inust attempt to achieve prespccilied performancc and robustness goals

within a heavily constrained Sramcwnrk. Ifthesc goals are nnt attain;ihle via iteration on the controller design, they are typically deemed unachievnble and are relaxed. The implications nf such an approach arc dramatic. Control theory comes to be viewed, not as an enabling technology, but as a means to do the best you ciiii with an cxisling plant striicture. System hiirdware designers do not routincly consult with control engineers a priori ahout the drawbacks of a potcntially slow or saturating actuator, or the hcnefits of a slightly innre expcnsivc hut morc accuratc sensor. These types of decisions, made independently of the control enginecr, often havc a significant impact oii overall closed-loop system pcrSormance. Joshi addresses this issuc by outlining a ncw role for control scieiicc in high-performancesystems and its implication for cnntrol theory. Spccilically, he challcnges the last-step control approach that has histnrically limited the influenceof control scicncc and identifies the need for ii new systcm-coiitr~il-analysis-synthesis (S-C-A-S) apprmich that links system and controller design. This apprnach advocates cxpansion o f the standard design process to include siinultaneous iterations on hoth the system and contrnler design. Such an approach rcquires control engineers to get involvedearlicr in the design phase when maior systcinchangcs arc still possible. The examples in this arlicle focus 1111 tlexible spacc structure applications and how the cost and performance requirements of such applic:itions ncccssitate an S-C-A-S design. Past controls research rclaled to this design approach i s reviewed, and future research needs concentratingon sensitivity, onccrtainty, robustncss, and optimality are identilicd. An Innovators Viewpoint An nld saying in the control lield is Ihal most ol the work in a control dcsign i s in developing the model. With the current emphasis on virtual design through computer-aided design packages, and tlie desire to implement inore sophisticatcd, model-based control techniques, this saying appcars to be more true today than ever. Modern control techniques can achievc rnagnificcnt results with state-space models. Not cven the inost robust controller, however, can fully compensatc for a poor modcl. Complicating this issue is the fact that developing a quality inodel typically requires a significant amount of domain cxpertise. For instancc, i m aernspace enginccr who i s highly proficient at dcveloping aircraft models may not hc the best choice for developing models Sor a chcniical process control system. Converscly, a chemical engineer inay not bc tlie hest choicc for developing aircraft models. A good control design modcl i s one that slrikcs a proper balance between fidelily and simplicity. To achieve this halancc, tlje engineer must have an adequatc feel for which charactcristics nf the system must he included in the model iiiid which may ,he discarded. This i s usually hest gnined through firsl-hand cxpericnce. Bernstcin provides an overall perspectiveof some ufthe challenges wc a11 face in tailoring control theory to practicc. These thoughts areclassified as general only in the sense that they are iiutdrawnfroin aspcciOcapplicalionspcrspectivc.The tlicmeof his article i s that each olus in thecontrol held must pay attention to tlctails because, in most cascs, cven tlic sinilllest unaddressed detail can thwart innplementaliono f new control ideas. This article i s suhdividcd into two m j o r sections: one that focuses on modeling issues and another that focuses on control issues.

Decemher 1999

47

The modeling scction emphasizes llic nced for lurlhcr rcsearch in 1he modcling area. In certain cascs, it calls lor oil-line idcntification techniques for developing murc :scurale models, while in other cases it secks to reduce the clfort associaled with modcling. The main idea is to develop a modcl of appropriate complexily 10 gcl tlic job done. Thc control ~ectioiidiscusses complic;itions that lypically arise in control algorithm iiiiplcmcntations-coinputing power requirements. transitions hotwccn continui~usand discrete-time systems, impact of saturations, limitations imposed by sciisor noise, and distinclions bctwceii siiiooLli and nonsinooth noiilinearities. Although these issues inay iiot seem critical i n control theory dcvelopment, if not properly addressed they can easily discouragc or iiltogether prcclmle implementation ofthese new ideas.

some OS thc rcasoiis Sor thc gap belwccn the lhcory arid practice ol'conlml and to suggest some ways to bridge this gap. Thc specific appli~ations discussctl are aerospace oncs, iilthough inany ofthegcncralpointsmadcarerelevant tootherdumninsss wcll.

Acknowledgments
I want to thank Tnriq Samad C o r his assislance in dcvcloping Ilicco~iceptforll~is spccia1iasuc;ind inidentifyingaiidrecruiting some of the contribuling iiutliors. This prujcct w ~ i u l dnot have been pussihlc without his help. Thanks also to Jim Cloulier for his a ~ s i ~ l a n c in c identilying potenliel contrihulors, Finally, 1 waul lo thank each olthc contribuling authors lor taking the tiinc to thoughtfully compose and share their llioughts and pcrspcclivcs with our rcadcrship.
'l'homss I,. Ting mccivcd B.S. dcgrccs i n Elecriical Engiiacring iind Matlmaalics in 1982 aiitl M.S. iliiil l'11,Il. dcglocs i n ElcctiiCal Enginecriog io 19x4 and 1987, rcspcctivcly, $111 from the Univcisity 0 1 Illiaois-Urlxwi. Siiicc 1994 lie liiis bccn with Gcncral Moiors Rcswrcli iiiitl Devclopmcnt and Plcming i n Warrcn, Michigan, whc11. he is currcnlly a staff re^ reaich aigincer. His CIKI~ work focuscs 01, &xcliq>-

Conclusions

Thc control engineering discipline, likc almost any technicel field, can he dividcd into sevel-al broad subclassilicatioiis: innovation, facilitation, and implcmentation. For tlic inosl parl, cach of these functions is performed by ii differenl set of cngineers. Effective internclion iiinong these lhrcc groups is an csscntial factor in lailoring control theory for pnicticc. Good co~niiiuiiicai o x and developing u i f i e d Inodel-h;iaerl iiiitoinolive tion between iiinovaturs hiid facilitators and betwccii fiicilitators ix~weltiiiiii coiiliiil and diagnaslic aigorithais. klc lias lpiior work experience and iinplementcrs is a prerequisitc for rapid, succcssful technoli n Ihe aernsp;sc indusll-y, desigrriiig ani1 tmilyiiog Right ~ o n t r osyslcins l lor ogy transfer. Hoiicywcll Tcchiiirlogy Centcr (Ibnnerly Himcywcli Systems Sr Rcscarch The contributing authors to this special section rcpreseiil in- Ccnler) fruin 1988 10 1494, imd i n robotics, clcvcloping p;ith planning mtl dustry, government, and academia, thus offering a widc range of iiiolioii coiilrol algorithms for Stinilia Naliuniil Ltihnwlories l i i m 19x7 IO perspectives. Thc oh,jective of lhis spccial seclion is to articulate 1988. DI-.'I'hg is a mcinbcr of I

4R

You might also like