See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226569014

Planning systems and artificial intelligence

Article in Annals of Operations Research · December 1988

DOI: 10.1007/BF02283732

CITATIONS READS

12 1,528

4 authors, including:

Klaus-Dieter Althoff Michael Richter

Deutsches Forschungszentrum für Künstliche Intelligenz Technische Universität Kaiserslautern, University of Calgary, Germany
304 PUBLICATIONS 3,409 CITATIONS 132 PUBLICATIONS 767 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Klaus-Dieter Althoff on 09 September 2016.

The user has requested enhancement of the downloaded file.

 16 (1988)I 12
Anna\ ot operalionsResearcb.

PLANNINC SYSTEMS AND ARTIFICIAL INTELLIGtrNCE

K.-D. ALTHOFF. N. KRATZ, M.M. RICHTER and P. SPIEKER

Lttute\b ol Kußeßlautern, Depa neht aI CaftPutet S.ierce, P O Bo\ 30 49,
D-t5750 Kaße^laute . W Gman),

Abstract

TradilionalO.R. syslemsare c.ftpared wilh problemsolvingin Anificial lnlclligencevia

Expert Systds. The d'scussbn@nte.son cxplicil knowledgerePreseolationTbe generd
aspedsdre illuskaledby 1woplanningsystcms:
. LESP2: A leamingsysteftld inspectionplar cene.alion
. IDA: A systemtor tinding functionsand solulio.s in @nslruclion

Xeywords:Planningsyslems,experr systems.knowledgereprese.ration,inspeclion plds,

conslruclronplans.

0, Introduction

Traditionauy,planning systemshave as a rule been generatedusing o.R.

methods.In recentyears Expert Slstems and AI methodshave also servedto
tackle these problems. Both alternativeshave their specific advantagesand
weaknesses, which in lurn rcstricttheir appropriat€fieldsof application.Roughly
speaking.the use of Al methodsappearsadequate,if lhere is a lack of uniform
mathematicalmodellingänd complexknowledgestructuresrequireflexiblepossi-
bilities of expressionwhich do not exist in the languageof conventioniilal-
gorithms.
In üe following we will discusstheseaspectsand will present two Expert
SyslemsLESP 2 and IDA as examplesof planning systemsin which AI meth-
ods are applied.

l. Asp€ctsol traditionalO.R. sistem

If o.R. melhodsare usedone tnes to describea given situation by meansof

modelling,which allowsto solveproblemsthrough mathematicaloperalionsand
algorilhms.This provides the advanlagethat due to a considerablelevel of
know-how and experiencein this domain many casescan be handled in an
optimal way. The scopeof suchmethodsis being extendedpermanently Thus it
has becomepossibleto dealwiü problemsof impressivedimensionsfar beyond
human abilities. There is no reason to replace O-R. systemsworking in a
satisfactorymannerby any other methods.

i J.C. Balt^r A.G. ScicntificPubllshingCompan

 K. D. Ahhoff et al. / PlanningslstemsandAI

There are, however,many tasksrcsulting from real applicationswhere purc

O.R. methods are not able to supply "adequate" solutions. Here the term
"adequate"is interpretedin different but alwaysinformal ways.
This is revealed
by the fact that the languagecapacitiesof knowledgerepresentationin O.R.
systemsare rather restdcted.A languagewith a limited scopeof expressionsis
not or not sufficiendy-adequate to describecomplexsituations.
First of all we will considerthe designof a mathematicalmodel lor an O.R.
procedure.Here the intuitive relationsbetweenthe objectsof the real situation
are lost,unlessthey areexplici(lypart of the model.Thesesemanlicrelationsthen
only exist in the designer'smind. They can only be communicatedby references
to agreements madeearlier; but aboveall they are nol accessibleto the system
itself. Particularlyin the caseof highly efficientprocedures,due to the simpliciry
cf the model,intentionsof the initial problemmay in addition be misrepresenled_
The proceduresfollowing the modelling stageuse agaiü a languagewith an
expressive powercomparableto the propositionirlcalculus.This doesnot exclude
at all that far more complexproblemscan be deah with. They have.however,to
be encoded,and it becomesimpossibleto recognizethem from the linguistic
point of view. The designerwho knows the code can of coursetalk about the
system,but he is no longerableto makethese"meta-refleclions"accessible to the
system.If such meta-reflections are sufficientlycomplex the user may wish to
aulomatethem.If this attempt fails due to an insufficientexpressiveness of th€
programminglanguage,then for the correspondingtasksa solution in generalor
a good solution(both in the senseof the model) may no longerbe found.
The loss of semanticinformation resultsin the fact that O.R. methodsare
more or lessusedas black box methods.It is generallyimpossible!o explainparts
of the solution to a naive user.In particular there is no possibilityof adding a
Ilexible explanation component to the system.This has two consequences:
Firstly, even a slritable solution may not be acceptedif it is not entirely
understood.The mere mathematicalconectnessof a solution would not be
consideredto be adequate.Secondly,the possibilitiesfor interactivework are
strongly restricted,too. The user can only apply his skills if he understands
sufficientlywell the actual stateof a problem solvingprocess.If this is not the
caselhe problem may not evenbe solvedat all.

2. Appmachesto problem solving in Al: expert systems

Sincethe mid sixties,parallel to the developmentof O-R. systems,research

groupshave been engag€din solvingproblemswhich were either not adequate
representable
in numericalforms or could not be solvedanalydcallydue to their
high complexity. Here the fundamental new idea was to solve such problems by
imitatinghumanpattemsof thoughtand behaviour.
 K..D. lltho[d al. Plannin:l rvpns rnJ 4l

Essentially,Expert Syslemsdiffer from O.R. sysremsin providing explicit

represcnaations of the availableknowledge.Here explicil knowledgerepresenta-
lion meansüat knowledgeis represented in a formalism which is similar ro the
way it is structured in the mind. These demandsrequire new programming
techniques$hich will be briefly presentedin the follo\Äing sections-T)?ical
applicalionsare problemswith varvingor dlnandc constraints.
The methods of knowledgerepresentationthat have been developedand
applied basically by Artificial Intelligenceare no longer pdmarily oiented
to$ards traditional models of computation and computer architectures.but
lowards cognilive human abilities.Hence one is farther a*'ay ftom lraditional
formalismsand availablehard' and softwarebut closerIo humanproblemsolving
abilities.Onc of the main challengesto Artificial lntelligenceis to study situa-
tions in which thesemelhodscan be appliedefficiently.

2.I. A ST]RV!]Y OF THE MOST IVPORTANT METHODS OF KNOWLEDGE REPRF]SBN.

fA f!()N

One ot the striking propertiesof methodsior (explicit) knowledgerepresenta-

tion is their strongl.vdeclarativecharacter.This ailows to representknowledge
aboul a problem wrthout already programming the acrual algorithms for a
solution-This solution is found by an infcrential componenlwhich is applied to
thc availableknowledge.In this way a large scopeof problemscan be tackled.
So far four meüods havebeenestablished:
(i) Rule-basedrepresentationis the earliestone. The knowledgeis put into
so-called"if-then' rules which are later ou processedby a rule interpreter.
This way of represenlation is basedon the assumptionthal the knowledgeof
experlscan be definedadequatelyin this "if... then' manner.Thercfore,the
earl-t E\pefl Svsle s sucb as DE-NDRAL (Buchananet al. [3]). MYCTN
(Sho lilfe [12]) or R l(Mc Dcrmott [8]), *ere purely rule-bascdsystems.
Although thesesyslemshave been successfulin principle. they neverlheless
revealedthe deficienciesof the rule-basedapproach.Declarativeknowledge
repr€sentationof contextsand nesledlacts lurns out to be very tedious;
moreoverstructuring is not supported. Proceduralknowledgeabout the
applicationof the rules which is necessäryfor their efficient use can again
onl) be expressed implicidy.
(ii) The useof predicatecalculusis a well-known,theorelicallywell-foundedand
widely favoredway of representation. Languagessuchas PROLOG (Clock
sin et al. [4]) and Krypton (Brachmanel al. [2]) afe resultsof this approach.
Yet. unlil no\l, for theselanguagestherehaveonly beenavailableinefficient
(Krypton) or incompleteand incorrect(PROLOC) deductioncomponents.
The variety of possiblemethods of representationin predicate calculus
provesfo be a handicap.Still complicalingis the fact that it is of first order
and causesproblemswilh regardlo meta-knowl€dge.
 K..D. Althollct al. Planntng'lsrem' tnd 1l

(iii) Framesplay an important paft in modellinghighly declarativeaspecrs.This

modeof representation is basedon the ideathat humanbeingsperceivetheir
environment in contexts.In addition to certain characteristicsof these
contextsthe relevanlproceduralknowledgeis also handled.lnitially, frames
arosefrom an extensionof semanticnel$orks which are nol basedon a
proceduralinlerpretation(Minsky [9]). The fundamentalnolions are inhed-
tanceand default reasoning-Although (or because)vinually all aspectscan
be expressedby framesthere is still considerablelack of experience1()use
ftamesefficiently.Thesepoints needmuch more consideration.

Along \r.ith the idea of message

passingas a proceduralenvironmentfor
framesthe conceptof objecl-orientedand programmingwas created.Todav
främesare quite often preferredto production rules and predicatecalculus
becausethey allow to integratea natural mode of descriptionof declaralive
knowledgeinto an establishedconceptmeetingthe demandsresultingfrom
SoftwareEngineering.particularlyduring the encodingstagc.Using objecF
orienl€dconcepts,notions like modularity and infomation hiding can be
consistentlyrealiz€d.The most popularimplementatiolof this programming
styleis Smalitalk-8oby Xerox PARC (coldberg er al. [5]).

For quite a long time no1 enough importancewas attached to these

aspects.which becomesapparent when one considerslanguagessuch as
OPS-5or PROLOG. The absenceof any explicit mennsof structuringin
thesesystemshastumed out 1()be an unacceptable obstaclelbr the manage-
men! of large-scaleapplications(e.g. if severalpeople cooperalein one
protecl).Meanwhilemodular conceprs(M-PROLOG) modelledafter proce-
dural programminglanguages(MODULA-2) or world conc€pts(MULTI-
LOG) (Kaufmann et al. [7]) adaptedmore srronglvto rhe logic background
havereacheda higherlevel of popularity.
Also with respectto knowledgerepresentationframe-basedsystemshave
an advantageover productionrule-basedor logic-basedsystems.Parlicularly
in plannitg and configurationsystemsthe problem oi adequatelyrepre-
sentingthe centralobject,the plan, is a difficult task_In his articleon frames
Minsky refersto the fact that the complexityof configurationtaskscan be
reducedby meansof efficient assumptions.so-calleddefault values.These
are parametercof the plan which are givenvaluesin advanceand enableone
to definethe remainingparamelersunder adequaterestrictions.Only a later
contradictionresultsin a $,ithdrawaland changeof theseparametem.
All Expert Systemshave in common the necessity!o represenlstrategic
knowledge,generallyknown as meta-knowledge. As this meta-knowledge has
a plaoning charactereven in diagnosticsystems,frame-basedknowledge
r€presentationis a greathelp.Thereforethe structuringpossibilitiesprovided
by framesare essentialalso in diagnoslicsystems.
 K.-D. Altholl et al. / PlanningststemsandAI

(iv) In order to improve,respectivelyto increasethe expressivenessof production

rule systems"the possibilities
of structuring,hybrid systems(LOOPS,KEE,
BABYLON etc.) have been created and made available to designersof
Expert Systems.Essentially,all systemsproceedftom an object'odented
basis and add rules, logic and//or functional elements.Here the main
disadvantage is a lackingsupport for choosingan adequatemode of repre-
sentationand Jumpingto anothermode if the situationindicatesthis.

2.2. ADVANTAGES OF EXPLICIT KNOWLEDGI] REPRESENTATION

Comparing conventional programming languagesand explicit knowledge rep-

resentationwe observethe following differences(as fat as soltwareis concerned):
As shown in fig. l. the use oI AI methodsreducesthe encodedand therefore
invarianl part of the system.In additiot to the knowledgerepresentation mecha-
nismsand the inferentialcomponentthere are conventionalprograms,e.g.mask
or file managers,
In the caseof AI methodsthe explicit knowledgetakesup by far the biggest
part of the system.Here a fixed componentof the syslemis represented in a way
which is directly accessibleto the user (respectivelyto the expert). Thus the
expert'sassistance in the d€signor modificationof the systemin order to adaptit
to changingdemandscan be simplified.
Another advantageis the availability of knowledge in order to generate
explanationsautomatically.Even a simpletraceof the .ules appliedin a produc-
lion rule systemcan explainmuch belter the processof problem solvingin the
slstem than a registerdump or a procedurelrace o[ a convenlionalsystem.Th€
informationgivento a usercan vary accordingto the respecliveusert\De in the
kind of detaiisand emphasisgiven.This makesil possibleto get more sophisti
cated explanationswhich havevery little in cotnmonwith the dumps or traces

'.r'ääqgrÄllnahüil3

Fig.1. Compdson of onvenrional sirb AI planningsyslems.

 K. D. AlrhofJ et al. / Planning sjstemt a,td AI

mertionedabove.The requirementfor this ability is not only givenby the lossof

the black box characterwhich causesan increasingdegreeof acceptance.In
addition the systemallows to check the results in dom:rinsrequiing reliablc
securitystandards(medicine.processengineeringetc.).
Knowledgerepresentation- consideredas a very highJelel programminglan-
guage.has still further advanlages:
Most representalion methodshavea specialexpressive power.which permitsto
designprogrammingsystemswith a great varietyoi functions.This is the relson
for their high level of nexibilily i.e. thesesystemscan be epplied to a *ide range
of particularproblems.The price 1C) pay is a lossof performance.This effecthas
only partially been compensatedbv more efficienl hardware. It should be
emphasizedthat in Rxpert Systemsparts of the tasks should be handled b1
conventionalprogr.rms$'heneverthis is compatiblewith the rest of the system.
Due to the declaratjvecharacterof Expert Systemsthe developmentof large
syslemscan be supportedby an incremenlaldesignprocess.Whereasthe conven-
tional procedurerequiresa systemspecificalionon thc qholc rangeof luDctions
and later changesin the designor evenencodingphaseturn oul 1()be expensive.
The conccplof protot.vpingplays an important role in Experl S.vstems. Here we
refer to a programdevelopmenlwhich providesa preliminary versirx which is
graduallyextended.For that purposeespeciall_\' the alreadyexistingconlponents
of the svstem can be used.

3, Examples

A nide rangeof applicatiol1sfor planning s.vstems can be found in construc-

tion, production planning and quality control. The t\\o Expert Syslemspresented
I n t h el o l l o s i n gc a nb r p l a c e di n l h c \ <( d t ( g u r i e \ .
IDA, an Experl S.vstem $hich supportsthe conceptionalslageof conslruolion
in Mechanical Engineering, and LESP. A learring syslem scrling to generate
inspection plans have been developedin the Laboralorium tür Wcrkzcu8-
maschinenund Betriebslehre(WZL) at the TechnischeHochschuleAachen
(Ahhoff [1]. Kralz [6]). The IDA project is now continued in a joined research
prdect (SFB 314) el thc University of Kaisefilaulem in cooperation$ith the
WZL, LESP2 is continuedtogethe.wirh the PFAFF conlpanya! Kaise$laulern.
Bolh systemsput differenl demandson kno\\,lcdgerepresenlalionmechanisms
which range.however.within the above-mentioned scope.

3,]. I-F-SP2 A LEARNING SYSTEM FOR INSPECTION PLAN GENER{T]O\

lnspection planning includesall operationsneededto plan tests which arc

economicaliynecessary to ensurethe quality of a producl. ln qualit) conlrol thc
term inspeclion planning slandsfo. developinginspectionplans.planninginspec-
tion devices.determininginspectior characleristics.lesl methods, instants of
 K. D. Ahho| et al. , PlanntIq 'vsten' and Al

inspection,test costs,test data processing,and inspectionplace as well as the

demandsfor staff.
LESP 2 is conceived as an Expert System shell in the area of inspection
planning which allows (within a certain range) to producequite easily specific
Expert Systemssatisfying the needs of different applications.Therefore its
particular task is, in addition to modelling generalmechanismsand knowledge, to
offer the possibility of integratingspecificknowledgeand data of a particular
companyinto the system.Accessto thesedata is providedbestby a suitabledata
bank connection.For the integrationof domain specificknowledgeit is essential
that LESP 2 has explicitly representedthe adequatestructures and mechanisms.
Only in this way one can get acc€ssto the semanticcontelts of the problems.
The representationin LESP 2 includesgeneralknowledgeabout insp€ction
planning, knowledge about the pdncipal way developing a plan (e.9. a simple
modelling of time and sequencing conditions), and knowledge about taking
advantage of analogous situations. Thus LESP 2 orientates towards the
VDIIVDEIDGQ slandard2619(VDI [14]) about inspectionplanning.Its target
is to structurethe various problems,to standardie conceptsand to obtain a
widÜ accept€dinspection procedure.
Mormver. LESP 2 offers the possibility of integrating specific knowledge
structuresand domain dependentmethods.Theseare for instancegeneralguide-
lines for inspectionplanning, specific instructions for the particular field of
productionuüder consideration,the classificationof the spectrumof products,
the empiricalknowledgeof the personinvolvedin inspectionplanningor produc-
tion schedulingas well as particular company interval methodsof taking ad-
vantageof similadtiesand analogies.
The procedurein LESP 2 js subdividedby the cyclesof the operationsheet
corresponding to lhe inspection plan. For every edited cycle the system suggest
the correspondinginspectionsteps.If this cycleis accomplishedthrough modifi-
cation of anotherone carriedou1earlierLESP 2 directly generatesan inspection
step by analogy.Thus the similarity of the cyclesis used to reducethe search
spac€of the possible inspection steps. This, however, requires knowledge about
the relations between all the cycles. The demands made on LESP 2 with respect
to knowledgerepresentationresult in the possibilityof adequatelyrepresenting
knowledgeabout inspeclionplanningand the relationsof the different cyclesby
productionrules,whereasthe structudngof this knowledgeis achievedby the use
of frames. Thus, e.g. a cycle is interpreted as a ftame with knowledge about the
relevant qualily requirements and characteristics.The knowledge about similari-
ties and the company dependent spectrum of products is modelled in object
hierarchies.
It should have becomeclear that realizing LESP 2 by conventional represen-
tation mechanisms would be very time-consuming. Nevertheless LESP 2 is
dependent on the integration of the efficiency provided by a (conventional) data
base svstem.
l0 K.-D. Altholl ?t al. Plannnqrt,t?mtandAl

:].2. IDA A SYSTEMFOR FINDING FUNCTIONS AND SOLUTIONS IN ']ONSTRUCTION

The purposeof the IDA proJectis to support in mechanicalengineeringthe

stageof designnamed "conception",in which functions and technicalrealiza-
tions are determined.First a detailedfunctional descriptionof the object to be
designedis generated.This descriptiotrhas to meel c€rtainrequirements-Subse-
quently, on this basis, the system searchesfor technical realizationsfor the
correspondingsubfunctions.In doing so differentlevelsof abslractionhaveto be
considered.
Here the fundamentalproblem is to restrict the cornplexityconcerningthe
choiceof possiblesubfunctionsand their technicalreälizalionson the different
levels of specificationby means of appropdate heudstics (use of particular
boundary conditions).Moreover,il appearsdifficuh to describeand to handle
dependenciesamong objects ol the same or different level of detail.
This problem descriptionalreadymakesthe difficuities of an computeraided
approachusingconventionalprogrammingapparent-Due to the possiblealtema-
tives of seleclionand combinatiol and the resultinginterdependencics such an
approachhas presently a very iimited chanceof success.A further problem
originateslrom the ne€d for continuousupdatesof rhe fast growing technical
knowledge.Today conventionalsystemsdo not supply satisfactorysuppofi in this
direction.Thesereasonseemto justify an Al approachto problemsdealingwilh
planningand configurationon differentlevelsof abstraclion.
As in the LESP 2 system the realizationis based on a hybrid mode of
.epresentationof the rel€vantdesignknowiedge.This is due to the fact that
neither an entirely ob.ject-orientednor an exclusivelyrule-basedrepresentation
adequatelymirrors the givenknowledgestructures:
The object-orienledrepresentationservesto describefunctions and technical
realizationsin termsof their desiredpropertiesand restriotions-
The rule-basedcontrol sefles to guide the inferential processas $ell as to
representdependenciesamong objects on the same or diflerenl levels of
abstraction.Here the choiceof a particular obje,cttriggersoff the applicalion
of a rule. An applicationof a rule changeslhe conditionsand requiremenlson
certainobjectsof higherlevelsol abstraction.i.e. rules modify the context of
objects.

The aclual state of specificationis stored in a dl,namic data basis.If a new

object is generaled,then by meansof adequaterules the actual context ol ä11
involved objectsin the d),namicdata basisis modified before another step of
specificalionis started.
It would be fatal to believethat a systemas descdbedabovecould be realized
without considerationof its environmentin construclion.So one has to support
in!€rfacesto databasesystems(containingtechnicalrealizations),to calculation
programs,and to CAD-systems.Only in combination uith these traditional
 K.-D. Althoff et al. / Planningsystems
and Al l1

systemsan expert systemlike IDA can become a powerful tool \r.ithin the
conslructionprocess.

4. Prospects

As the secondexamplehas illustrated it is not sufficient to considerExpert

Systemsin completeisolalion ftom other aspectsonl) on an abstractand high
level as it has been done in the past. It is often rather necessaryto make use of
certain fast algorithms.data base operationsor conventionalC).R. programs.
Sucha systemthen requiresinterfecesbetweenits compo[ents.
The speedof the conventionalalgodthmsis due to the fact that a compara-
tively greatnumberof operationsis carriedout in a uniform and possiblyparallel
way. The problemsarisingfrom applicationsusuallydo not inmediately allow to
separatesucha uniform part. Thereare two approaches which offer themselves
to
solve such problems:
1) The algorithmis calledby the systemto deal with a non-uniformquestion;the
resultis usuallya severereductionof lhe efficiency.
2) The systemcalls the algorithm 10 solve a modilied and a more uniform
problemwhich can be efficientlysolvedby this algorithmand is still usefulfor
the main task.

The main problem is to subdivide the original lask into parts, such that
classicalalgorithmscan be appliedas indicaledin 2).
Along wilh an increasingapplicationof Expert Syslemsin real situalionsthis
type of interfacequeslionsand the adequatedistribution of tasks will become
more and more essential.The actual staleof the art in this areais, however,still
very dissatisfying.Only the connectionto data basesshowssomeinitial progress
(Spieker[13]). Nevertheless. the future possibilitiesof applying Expert Systems
will decisivelybe in{luencedby suchdevelopments.

References

Ill K. D. Allhoff, What are experl syslems?.i!: txpet Sleenß tu l'tuduLtion Enqüeetikg
(Springefvcrlag. 1987)20 14.
[2] R.J. Brlchnan. R.E. Fikesand H.J. Levesquc.KRYPION: a iunclional apPtuachto knowl-
edgerepreserhtio!, IEEE Computer16 (oclober 19E3)67 7.1.
t3l B.G. Buchanan.G. Southerländand E. Geirgenbauh.HeuristicDENDIL{L: A programlor
generatingexpldalory h)pothesesin organicchemistry.in: Machin( Ihtelllg.hce5, .ds. B.
Mellzer and D. Michie (New York: Elsevie..1969).
[4] w.F. Clocksina.d C.S.Mellish.Prosrammirsin PROLOC (Spnngd-verla8,1981)
[5] A. coldberg and D. Robson..tn/lrllt-80: The hhguage and ns lnvleaerr"rio, (Addison
Wesley.1983).
 12 K.-D. Ahhoff et al. / PlanningsystemsandAI

16l N'. ka1z. Enrwicklung ejnes Solsüe noduls zur re.hn€runlerslützlenFunktong und
Linungsfindungln der Konstruklionsphase Konzipieren.Diplonarbeit Rw.rH Aachen,1986.
l7l H. Kauifnan lnd A. crumbach, Representing and mdipularing Lno*,]edgewirhin -$.orldJ,.
\D: Ptur. ol the fi6t lht. Canl on ExpetrDdtah6e Srrrenr, Cbarleslon,Soulh Carolina:ADril
lo'tb. J | \erschberC.
[8] J. McDermolt, R1: A rule-based.onfigürer of computers]slems.ComputerScienceDepan,
hent, Carnegic-Mellon University.1980.
l9l M. Minskr. A rramework for representing knowle d9e. iD: pslchotos\ al Canputet rlision, ed.
P.H.Winnon (Mccraw-Hiu. New York. 19?5).
[10] M.M. Richter. Expeflensysleme und konvenrionell€programme UnFrschiedeund KoDp,
lüng.proDlem<. in tp.ß,hhtt lah"ologp. Wa,h,,un. Atbcr.l.'ste . ed. R. Henn. ,zum
50. GebDrtstag von Lothar Spärh(SpringerVedao, ro appeü.
[11] M.M. Richtcr,Sone abstracrproblemsin knosledgerepresenradon, in: E\pen Ju.lgenentond
nrler "!rr?ra. ed. J Mumpower,NATO ANSI Series(SpringerVerlag)ro appear_
Ir2l EH. Shorrliffe. M'ICIN: Computerhased ned!.dt un\uhanans (American Etsevier/No(h_
H.'llard \eu Yorl. l9 /o,.
[1]l P. Spieker, Möglichleiren der Kopplung von PROLOG und einem Da(enbanks]ren,
äusgeführtam Beispielde. tDM. DiplomarbeilRWTH Aachen,1985.
[14] VDI/VDE/DGQ Ri.hdinie 2619:Prülpla.uns.Juti 1982.

View publication stats