and they perform fhst in comparison with the CPU computer

TOOLS ~ METHODS FOR SPACECRAFT CON

Guillermo OrtegaEuropeanSpace Agency (ESA/ESTEC)

Keplerlaan-1, Noordwijk, AG2200, Holland

ABSTRACT

The goal of the paper is to analyze which computationalmethods and modeling tools based on Fuzzy Logic helped byGenetic Algorithms are available right now. The emphasis isput on the analyses and design of spacecraft control systemsdue to its complexity.

A SpWeCH3ft control system measures the position and attitudeof the crafl and produces guidance and rotation commands toplace it in a specific orbit and with a specific orientation.Spacecraft control units are complex systems difficult to handleduring the analysis and design phases of the engineering lifecycle.

The article describes wh~is-who in the development arena ofthe support tools and methods, which commercial products areavailable, and which of them are freely obtainable.

m-- ----- .J? .,. -I ne SUE m ux ti is ~1513ni@ d ik tXkiitkk13 Mddisadvantages of each system are carefully examined from acritical point of view.

1. FUZZY LOGIC AND GENETICALGORITHMS FOR SPACECRAFT’

CONTROL

For a relatively small servicing spacecrag the controlrequirements demand a high degree of uncertainty in criticalvehicle parameters like total mass, f&d-forward thrustimpulses, moments of inerti% center of mass, etc. Theserequirements are translated into huge complexity during thedesign phase: the navigation block is based on complicatedfiltering schemes, the guidance block is made of parametertables of considerable size, with contingency recoverysituations, and the control block must be designed usingmultiple input-output tectilques for a sixdegree-of-freedomvehicle. Much of this complexity in the design of the controlsystem cmmes from the way in which the variables of thesystem are represented and manipulated.

In the search for an easy, efficient cost-effective control design

and development technique, @ logic (FL) seems to providea method of reducing system complexity whale increasingcontrol performance. Fuzzy set theory was originallyintroduced by Prof. Zadeh in 1%5 [12][13]. Since them manyresearchers have introduced fuzzy logic techniques to solvedifferent types of control problems [3][4]. The ability to modelproblems in a simple and human-oriented way [9][26], and theability to produce smooth control actions around the set pointsmakes fiw.zy logic an especirdly suitable Candidate for use inspace applications [17][20][2 1][23].

While the fhzzy controller deals with the uncertainty of themodel of the vehicle, the genetic algorithm tries to optimize the0-7803-4778-1 /98 $ 10.00@ 1998 IEEE 3167

TROL SYSTEMS: A GENO-FUZZY APPROACH

JOS6M. Giron-SierraFacultad de Fisicas, Universidad Complutense (UCM)

Ciudad Universitaria s/n, Madrid 28040, Spain

controller for a particular constraint. The controller will try tocope with the nonlinear equations of motion of the spacecraftdynamics and kinematics, and the imperfections in sensors andactuators.

Fuzzy systems have two parameters, which can be optimized: arule database and the tizzy sets. Two approaches can he takento obtain an optimum solutiom either using the heuristicmethod, in which the control engineer obtains the best systemto satis@ the criteri~ or by findhg an analytical solution to theproblem. Gn most occasions, the design of the optimum systemrequires the knowledge of an expert operator.

Many recent publications have demonstrated the possibility ofoptimize those parameters automatically (analytically) bymeans of genetic algorithms (GA) [1][71[14]. Geneticalgorithms were introduced 30 years ago, but only recentlyhave they been recognized as a promising technique to

FUZZY CmnlCUxlc AmoRrnMs

LAla

Figure 1

optimize these types of ti.mctions. GAs are optimizationmethods based on natural evolution. They are easy to apply,

consumption of other alternatives[16][1 8][22][271.

2. METHODS TOOLS, AND STANDARDS

GcncAkzy computational methods and modeling tools to helpin the design and development of spacecratl control units are ofprimary importance to bot~ the control engineer and theSpacecraft project manager. Methods and tools will allow thespacecratl engineer to speed up the design and developmentcycle. They will also allow the project manager to cut down indevelopment time and manpower cost. Standards andconventions allow interchange of ideas, methods, and tools

among working teams.

Figure 2 represents the cycle followed by the control engineerto design a spacecraft control system. This design is the so-cafled classical one. This type of design cycle is used with thecontrol methods like robust eigenvalue assignment, Linear

--[= “ =

8

*:TU

&8Elemklm

A@Y$u

D ‘“”

TIDS*~~ P’@&&d~-a@@Sy

‘=’ Ciiziil F==

~

c1pr0u9*- *1

Figulw 2

_tic (W method, etc. [51[61.

Figure 3 represents the cycle followed by the control engineerwhen using tizzy logic [25]. Comparing figure 2 with figure 3is possible to see the matches between both techniques.

The mission requirement part can be compared to the study ofthe ph~itx of the problem. In both cases the control engineerhas to study the problem.

NexL the engineer has to come up with a model of the plantand the correspcmdmg control archhecture. The analysis of thestability of the controller is an important step in the robustcontrol techniques scheme [8][11][15]. The simulations are awell-proven tool for the design and sofl testing of controllers:this step can be applied to both, classical robust controltechniques and FL based techniques. The testing on ground isthe ultimate integration, validation, and verification tool forcontrol design. Again, this step can be applied to boa classicalrobust control techniques and FL based techniques [10][19].

Nowadays, the computer code generation and the production ofthe associated documentation are steps, which required the helpof a computer. Most robust control design tec~lques arehelped by Computer Assisted design Tools (CSDT). Section 4of thk article will deal with the introduction on the market ofCSDT tools using fbzzy logic and optimization methods.

Methods. By methodolo~ is understood a collection ofmethods and procedures to design, cmnstruct, verify, and test aspacecraft control system.

Tools. By tools is understood a set of computer routines to aidand help in each of the tasks mentioned in the previous point.

Standards and conventions. This is defined as a group ofrules and regulations to apply when designing building andtesting the controller. The standardization helps when differentengineering groups must share a common tkunework [2].

Figure 4 shows the key cornerstones of methods, tools, andstandards in the FL-GAs control design and developmentteefuique.

A methodology to design and build FL-GAs controllers should3168

include among others the following elements:

● Guides to requirement management, analysis, design,and verification of FL-GAs based systems.

● Guides to apply software engineering standards to thecoding phase (easy if automatic code generation ispossible).

● Guides for quality assurance and test procedures.

A good tool for FL-GAs controller development wouldinclude the following elements:

● Creation and modification of universe of dkcourse,* sets and membership fimctions graphically.

● Automatic generation of rules databases based on statevariables.

● Available library for most phrtial common controlproblems; the user can pick up some building blocks andconstruct a bigger controller from them.

● Automatic optimization of membership fimctions andrule databases based on parametric probabilistic problemcharacteristics; the user can select the probabilities ofcrossover, mutation, and reproduction, and change themif necessaxy.

● Automatic generation of a fitness function to match aparticular problem.

● The automatic generation of code in C, FORTAN, Ametc.

● The generation of documentatio~ and the control of therevisions of the documents.

The goal of standards and convention is to fwilitate trade,exchange and technology transfer among engineering teamsacross the world The standardization and the establishment

~ ShluuOm

Figurs 3

of conventions would allow, among others: the uniquelabeling of designations, units, and s-~bols used, and theproduction of glossaries and thesaurus of FL and GAs terms.

FL-GAs systems will be chosen as alternative control designtechxique if they can prove that are less expensive, whilemaintaining the controller wittin the prescribed requirements.

A design control technique is less expensive than others whenit keeps the development team size smafl and thedevelopment and validation time short. By oppositio~ adesign control technique become expensive when there arefew or none existing tools available to help in its

implementation, when the technique is difficult to understandand the learning curve is pronounced, or when the output of thedesign does not meet the specifications, and multiple iterationsare needed.

3. WHo Is WHo

To be able to write this article, it was conducted a marketresearch of fiwxy logic design and development tools. Thisresearch included questions like the nature and purpose of thetool, outpu~ availability, price, suitability for spactapplications, etc.

Late in this section, the topic discussed in section 2 is explored,trying to match the tools mentioned here with suitability tomethods, standards, and conventions.

The number of tools analyzed here are numerous (see table 1).Most of them am commercial, and few of them are heelyavailable. This relative high number of applications has beeninterpreted as an on-going maturity process in the marketplace.in comparison, few years ago, the number of available toolswas much reduced and most of them were fi-eely obtainable.This means that now commercial companies are seen tizzylogic as a potential good market. Thk can further be interpretedas a starting success of @ logic as common, easy, controlproblems solver.

Table 1, represents a snapshot of the available tools in themark& Mainly, companies tiom USA Germany, and UK arenow developing this kind of applications. The range ofplatform varies, but mainfy the target is focussed in the PCmarket.

The available columns in table 1 detail the product name, thecompany which supplies the product or the name of thedeveloper (if the product is for free). The table also shows, theoutpu$ the optimization help (Opt.) if any, the suitability forspace control applications (Spat.), the operating system used(OS), and the price.

Nearly all products are development environments. ‘Ihat is,they allow the user to desi~ develop, and test a systemcontrolled by tizzy logic sets and rule databases. The input isthe block design of the system to control. The output varies, butmost of them we able to generate C-code automatically orMATLAB@ m-files, for further prowssing [24]. TheEducational Fuzzy Control Package is a software suitable foreducational purposes which will be welcome for new comerswanting to know the principles of FL over the computer.

Few products are targeted to generate assembler codeautomatically (FLASH, FIDE). This is particularly usefid whenthe code will bc inserted into the EPROM of a micro-controller.

Table 1 shows also the suittillity of the products when

applying to spacecratl control systems.

Not mentioned in table 1 are other products like the &.zy logictoolbox tlom Boeing for their Easy-5 tool, or the fhzzy logictoolbox fiwm Mathematical (Wolfhng Research). These arealso excellent products, and available for a wide range ofplatforms.

Nearly all development environments allow the creation ofuniverses of discourse, @ sets, and membership functionsfor a particular problem (CubiCalc, A-B FLEK Fuzzy ControlManager, MATLAB ToolBox, FUZZLE, etc.). However, noneof them contain a repository of already solved problem asexample, or blocks ready to use to construct more complex

3169

problems. This can be due to the raising use of FL in themarket in thk moment.

From the point of view of the optimization, only one tool (theFL ToolBox of MATLAB@) is able to perform optimization.However, the method is not based on genetic algorithms butANFIS. None of the systems presented here uses geneticalgorithms as optimization tool. The combination FL-GAs isstill fm from Lx3mgstandard. However, a genetic algorithmstoolbox is ffeely available for MATLAB@.

Most of the products have targeted the output as C-code(CibuCalc, FUZZLE, fuzzyTECH, FuzzyCLIPS, FIDE, AB-FLEX, etc.). This means that the code generated can beintegrated in an already C-coded system with littledifficulties. However, none of the packages described hereare able to generate Ada code. Ada code is well appreciatedin a good number of spacecraft control problems (civil and

lnlE@mtaTdmb@emW (Gmmny)

Fmv Wlmc Eddmd NoCanbi lniu@DnalM. -

PaG@e (UlqLPA Fuzzy Logic WsI@l @d% No

LrYJii W?mmllPs&@8sLid I I

FUZ?LE I Woica k I CM No

1 1- 1fmvTECH f INFORM W i MATL6B No

(G9mmy) nMss,c cdl?,Ent&ledadd bs+lwti

nwletedQxi’bD39K

w Tm WhWorks cd YesF@-r#c Inc (IJaAJ

FwzjCUPS TwA l-i, code NoIlw [Usq

Iirons,

L II 1- 1

AS FLEX Gktd Tddcd Mm./@ Noser’kes(uw nws,

FLOE y#M L%% ~

miks,Cmie

FkxTml Fldie Vm!al@J& NoI* Qwp,UC.(w

FlA9i F$l$w- AsJ& h

1 1 IFIEWT Pmf ZdOh (USA) Pkts m

1 1 1Xr!my I IMSCNM(E) I PM

wes$lic,No” wi13.1 $301m

M W@s, m45woWaWT

I=FYes 2.%

UrilNOwildms,

me, g%UNIX

No m 95 EM 3100

TrM Wiodws

30 r&’i2i

H=No w% Upx

Iq.Jesi

No Wil% Upalreqmi

No Ww?. $lm3

w

Table 1

military).

Some products have targeted the output of their environmentsas MATLAB@ m-files. This makes sense with the user hasalready this software or hehhe is targeting to performsimulation under this system.

Special mention has to be made to the tlee products listedhere. Those can be a very good alternative to commercialsystems, and certainly will be very helpfi,d for a stinting pointin the development of FL control systems: FISMAT, and

Xiizzy. FISMAT is a MATLAB@ToolBox developed by Prof.Zadeh. It was developed to fmter the development of FL asalternative in control problems. Quite successfully, it can befreely used and it is suitable for an enormous kind of problems.Xfuzzy is a recent initiative from the University world. It isable to handle medium size problems with a quick response. Itdoes run in all known UNIX systems. The source code isavailable.

4. THE FUTURE OF GENO-FUZZYTECHNIQUES

The iiture of the applicability of geno-fiwzy tools andtectilques in the aerospacs industry is still unclear. It is trueevery day there are more and more tools on the market. Fuzzylogic is arising as a cheap-t%t-better alternative to commoncontrol problems. The extrapolation of this scenario to thespacecraft control problem is not immediate.

Spacecraft control engineers and project managers are ratherconservative in exploring new methods or alternatives. Inaddition, @ logic based spacecraft control systems have tobe very well proven on-ground before committing to the fight.

The market place is still quite young: diversi~ of tools withdifferent scopes, enormous range of prices, lack of highlydesirable fatures.

Spacecraft program managers will not employ FL-GAstechniques in their projects unless they are proven to be cheap,safe, and able to satisfi the agreed control specifications.Spacecratl control engineering teams will not employ FLGAstechniques unless they are proven to be efflcien~ easy to use,

S&mkds:

Took.Unitsandsymbols @

.Ghssarymdthesanms Cfl● Computer assisted design.$

~= **cQ&gaImatica Q /“

● ToolBox~ —

>/”

II WZZY GENETICLOGIC ALGORmlMS

tMeltwds

“--’ Qib3b3 R. Qualityasslualcc ~

. Pmje+t mamgmnmt&

Figure 2

and secure.

One of the duections in which the market place could mature isthe establishment of the mentioned methods, standards, andconventions. Once methods and conventions are agreed by theFL community, tools developers will have a clear direction forthe investment on the development effort. The initiative toestablish these methods and conventions must come tiom awell-recognized authority and must be endorsed by goodreputation institutions.

5. CONCLUSIONS

In the search for an easy, efficient, cost effective, control

3170

design and development technique, tizzy logic seems toprovide a method for reducing system complexity whilekeeping mntrol performance.

Since the publications of professor Zadeh [7][8], manyresearchers have introduced @ logic techniques to solvedifferent types of control problems. The atility to modelproblems in a simple and human oriented way and the abilityto produce smooth control actions around the set pointsmakes fhzzy logic an especially suitable candidate for spaceapplications.

The article aimed at presenting a snapshot of the current toolsand development environments for control problems solvingusing * logic. The introduction of genetic algorithms astool for optimizations is also considered.

The market place is still young and immature. Methods,standards, and conventions need to be established to allowthe tool developer to reach a good level of usefulness,performance, and reliability.

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

6. REFERENCES

C. Phillips, C.L. Karr, and G. Walker. Helicopterfight control with @ logic and genetic algorithms.Engineering Applications of Art@cial Intelligence,9(2):175-184, 1996.C. von Altrock. Towards fiuzzylogic standardhtion.In Proceedings of Fifth IEEE InternationalConference on FUZV $wtems, volume 3, pages 2091-2093, New Orleans, USA September 1996.C.C. Lee. Fuzzy logic in control systems: fiuzy logiccontroller-part 1 and 11. IEEE Transactiomr onSystems, Man, and Cybernetics, 20(2):404-435, 1990.D. Driankov, H. Hellendoom, and M. Reinfkmk. AnIntroduction to Fuzzy Control. Springer Verlag,1993.D. McRuer and H.R Jex. A review of quasi-linearpilot models. IEEE Transactions on Human Factorsin Electrom”cs, 8(3)23 1-249, 1967.E.H. Mamdani. Applications of fuzzy algorithms forcontrol of simple dynamic plant. Proceedings IEE,(121):1585-1588, 1974.F. Herre& M. Lozano, and J.L. Verdegay, Tuningfuzzy logic controllers by genetics algorithms,International Journal of Approximate Reasoning, Ott1995.G. Calcev, R Gore~ and 1. Dumitrache. A Popovtype approach to stability analysis of fhzzy controlsystems. In Proceedings of Fourth EuropeanCongress on Intelligent Techniques and SojlComputing, volume 1, pages 3-7, AacheL Germany,September 1996.G.J. Klir and B. Yuan. Fuzzy sets and @ logic.

[10]

[11]

[12]

[13]

Prentice Hall, 1995.H. Yousse~ C.Y. Chkm~ and G.R Yu. On-lineLQG-fiwzy approach to failure detectio~ isolatio~and reconguration of control surfiwes. In AMAGuidance Navigation and Control Conference, AIAA96-3799, 1996.K. Tanaka and M. Sugeno. Stability analysis anddesign of fimzy control systems Fuzzy Sets andSystems, 45(2):135-156, 1992.L. A. Zadeh, K. Fu, and M. Shlmurq Fuzzy Sets andTheir Applications to Cognitive and Decisionpmcesse~ Academic Press, Inc., 1975.L. A. Zadeh., Fuzzy sets, In Fuzzy Sets forIntelligent Systems, pages 27-64. Morgan Kaz@mann

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

Publishers, Inc., 1993.L. M. Freem~ K. K. Kumar, C. L. Karr, and D. L.Meredith, Tuning @ logic controllers using geneticsalgorithms: Aerospwx applications, In Conference onAerospace Applications of Art@ial Intelligence, pages351-358, Dayton, OH, USA Oet 1990.L.X. Wang. Adaptive Fuzzy Systems and Control.Prentice-Hall, Englewood Clfs, 1994.M. Brown and C. Harris. Neuro-fkzzy adaptivemodeling and control. Prentice Hall, 1994.M. Steinberg. Development and simulation of an F-18fhzzy logic automatic carrier landing system. In 2ndIEEE conference on Fuzzy Systems, pages 797-802, SanFransisco, CA 1993.M. Steinberg. Potential role of neural networks andfuzzy logic in fight control design and development. fnAIAA Aerospace Design Conference, AIAA-92-0999,frvine, CA 1992.M. Sugeno, M.F. Gti~ and A. Bastian A iizzyhierarchical control of an unmanned helicopter. fnIEEE 3rd Conference of Fuz~ Systems, Koreq 1993.S. Daley and K.F. GI1l,Attitude control of a spacecraftusing an extended self organizing fhzzy logiccontroller, In Proceedings of the Institute ojMechanical Engineering, volume 201, No. C2, 1987.S. E. Woodard, D. P. Gar& C. Y. Tyan, and P. P.Wang Application of Fuzzy Logic Control to aGlmballed Payload on a Space Platform, Journal ojInformation Sciences: Applications, Vol. 4, No. 3, pp.143-166, 1996.S. Isaka and A.V. sebal~ An optimization approach forfiuzy controller design, IEEE Transactions on $vstems,Man, and Cybernetics, 22(6) :1469-1 472, nov 1992.T. SUZUly,K, k’asu~ S. Yoshikaw~ K. Yam@ and N.Yoshi~ An Application of Fuzzy Afgorithm toThruster Control System of a Spaceera& 19thInternational Symposium on Space Technology andScience, Yokohama (Japan), May 15-24, 1994, pp 303-310.R. Babueska Designing fiwzy control systems withMATLAB and SIMULINK. fn Proceedings EuropeanConference on Fuzzy Systems and Intelligent

[25]

[26]

[27]

Techniques, Aache~ Germany, 1994.T. Takagi and M. Sugeno. Fuzzy identification ofsystems and its application to modeling and control.IEEE transactions on Systems, man, and Cybernetics,15(1):1 16{132, 1985.T. Terano, K. Asai, and M. Sugeno. Fuzzy SystemsTheory and Its Applications. Academic Press, Boston,1992.W.A. Kwon& K.M. Passino, and E.G. Laukonen.Expert supervision of fuzzy learning systems for faulttolerant aircrafl control. Proceedings of IEEE, 83(3),1995.

3171