From: AAAI Technical Report FS-96-03. Compilation copyright © 1996, AAAI (www.aaai.org). All rights reserved. The Resource-Based Configuring Technical Paradigm: Systems from Modular M. Heinrich Components. E.W. Jtingst Daimler-Benz Research Institute Abstract In the resource-basedparadigmthe interfaces through which technical systems, their componentsand their environment interact are modeledas abstractresources,and eachtechnical entity is characterizedby the types and amountsof resourcesit supplies, consumes and uses. This intuitive model,derivedin oneapplicationarea,is shownto be in concordancewith the design rationale of modular componentsystems. A simple self-organizing configuring inference procedure for the resource-basedparadigm, resource-balancing, with a description of the environment of the technical systemas the requirementspecification, is derived from the basic acceptance criterion for configurations.Fourlevels of knowledge are definedfor this paradigm and introducedin a simplerepresentationscheme which,throughits inherentlocality andmutualisolation of componentknowledge,allows efficient acquisition and maintenanceof even large component knowledgebases. 1. Introduction Configuringis the construction of a technical system accordingto the requirements of a specificationbyselecting, parametrizing andpositioninginstancesof suitable existing componenttypes from a given catalogue [24]. Thus, configuringitself does not involvethe development of new componenttypes [13]. The developmentof newcomponent types, e.g. for additionto the component catalogue,rather is a typical task in the domain of design. Configuring of technicalsystemsis an ubiquitousindustrial engineeringtask and has beena domainfor expert system application ever since the longtimeparagonR1/XCON [1]. Mostconfigurationexpert systemsagreein using framesor objects for the representation of factual knowledge about component types [2, 3]. For the knowledge about howto use the factual knowledge andhowto fred a goodconfiguration quickly, two approachesare used. Expert systemsin the tradition of XCON [1] represent the actions of a human configurationexpert throughproductionrules andmimicthe humanin a by-rote performance.Theother approachis to find and represent the principles that guide the human configurationexpert. Theprevalentconstraint-basedmodel 19 AG Berlin [4-10] treats the functional specificationof the technical systemandthe relations betweencomponents as constraints on the components andtheir attributes. Suchconstraintscan be used to check hypothetical configurations in a generate-and-testmethodology [5, 11] or as guidelinesfor selection of components [7, 9, 12]. If "key components" whichrepresentthe different functionalconstituentsof the technical system can be identified [4, 13], and the component cataloguecan be organizedinto taxonomies with root classes correspondingto the "key components",the configurationtask can be viewedas a classification problem and implemented with the taxonomy as a decision tree [12, 14], with improvement possible by makingpartial choices [15]. When"key components" themselvescan be configured, e.g. by recursively applying the process of functional specification, constraint propagation and component selection [10, 13], a hierarchical decompositionof the technical system is achieved. A mixed hierarchy of "has-parts" - relationsand"has- specialization" - relationsis an elegant and efficient representation schemefor such knowledge [7, 9, 14]. A morespecific principle, that components in a technical systemonly connectat specific interfaces or "ports", has been recognizedas an implicit understanding[16], as a primesourceof constraints on components [17, 18, 19, 20] andas a distinct "architectural"type of constraint [13]. Whenfor each component the knowledge about the componenttypes that it maybe connected to and those components that maybe containedin it is kept as an attribute of the component [17, 18], maintainabilityof the knowledge base is muchenhanced over purely rule-based expert systems, where these constraints are mixed up with sequencingknowledge,thereby increasing the very large numberof rules that haveto be maintained[21]. However, the direct referencesbetweenthe component types [13, 22, 23] makeit necessarythat every component description in the knowledgebase is re-examinedand possibly changed whenevera new componenttype is introduced to the catalogueor deletedfromit. "INsre-examination is a very demandingand time-consuming activity and could well be responsible for the huge amount of work spent on maintenanceof large knowledgebases for configuration expert systems[21]. 2. The resource-based Ourownstudies of modularcomponent systemsand of the configuringof technical systemsfrommodularcomponents led us to a verygeneralprinciple usedby human expertsthat subsumes the connectability principle: the principle that systemsand components interact mainlythroughinterfaces whichcan be thoughtof as resources,andthat the resources demandedand the resources supplied by componentshave to be balanced. This principle, whichwasindependently recommended as a consistencycheck [11], weproposedas a basic modelfor the configuringof technical systems[24, 25]. A very similar paradigmof the demandand supply of resources waspresented as a "computationalmarketmodel for distributed configurationdesign"[26] Theconceptof resourceis an intuitive abstractionof the interactions betweencomponents and betweena technical systemandits environment. Thenotionof resourceincludes all kindsof extensive(accumulative) physical,technicaland commercial entities, both real andvirtual, that mightbe supplied by one componentof a system and consumed (exclusively)or used(temporarilyor in common) by another component (see fig. 1). Examples of resourcetypes fromthe realmof computer systemsare electrical currentat a certain voltage,coolingpower,floor space(physicalentities), card slot space, input ports, memorycapacity, software procedureinterfaces, bus connectors(technical entities), purchasecapital, constructionwork-time, softwarelicenses, supervisorattention (commercial entities). Thefact that a majorindustrial designtask (configuring) involvesonly components selected fromcatalogueshas also beenrecognizedin MechanicalEngineeringand has led to researchon "CatalogueDesign",but with little information interchangebetweenthe MechanicalEngineeringand the AI communities. Abasic idea is that cataloguedesigninvolves twophases, the selection of a system structure and the selection of real components for the genericelementsin the choosen system structure. The main research thrust, however,currently is towardstechniquesfor the designof the systemstructure. Geneticalgorithmsseemto do very wellfor this task[27]. The Constructive ProblemSolving (CPS)approach [28], with a rigorous formal treatment under an abductive reasoningscheme,followsa similar idea. Asemanticmodel of the systemstructureis incrementally built upfromgeneric components, the permissibleattribute rangesof the generic componentsare increasingly constrained through the external specifications, the emergingstructure and the emergingmutualdependenciesof the components,and the final configurationis wonby choosinga conforming set of componentsfrom the given componenttype catalogue. A logic-basedcalculuswith aboutforty constructionrules was proven soundand complete[29]. Besides the complexity issue throughthe lack of heuristic metalevelguidancerules, however,a basic problemwith logic-based incremental approaches is that logically consistent intermediate configurationstates could be isolated by arbitrarily many inconsistentstates. Were-present here the Resource-Based Paradigmbecauseit gives importantinsight into the human conceptualisationof the configuringtask, opensa novelline of reasoningfor the heuristic guidanceof configuring, discloses a type of constraints basic for configuring problemsfor whichno sophisticated algorithmsare known,and yet provesto be quite efficient in importantpracticalapplicationareas. ~ model component ~ ~’?~ [component ~ Figure1 basic relationshipsof components andresourcesin the resource-based modelof technicalsystems A resource-basedmodelcharacterizes a technical object mainlyby the types andamountsof resources it supplies, consumesand uses. With a resource-based model, the technicalsystem,its components andits environment can be describedin a single common paradigm.Thedescriptionof the environment, stating those resources and amounts demanded of the technical systemand those resources and amountsprovidedfor the technical system,obviouslycan play the role of technicalspecificationsfor the technical system(see fig. 2). Figure2 resource-based model of a technical4-" supplies -’~ consumes systemandits environment " ~ uses 2O Wedeveloped the resource-based model first for the configuring of modular computer systems, where practically all constraints are resource-based. But resource-based modelsapply as well to other modular technicalandorganizationalsystems.This can be explained fromthe fact that a technicalsystemis alwaysbuilt for a purpose:to providesomeservice, i.e. somereal or abstract resource or resources for use or consumptionby its environment.Suchresourcesdo not arise by themselvesor out of nothing, but are supplied by componentsof the technicalsystem.This, after all, is the sole reasonfor a component to becomepart of the technical system. These components mayin turn themselvesrequire other resources for their functioningwhichhaveto be suppliedby further components.At the end of that "chain of supply", some resources have to be supplied by the environmentto the technical systemfor use or consumption by its components (seefig. 3). therefore, are usuallydesignedto supplya certain optimized amount of ofily oneresourceor of a suitablecombination of a fewresources. Common basis for the designof the modules is the system design of that modularcomponentsystem whichidentifies and specifies the physical and technical interfaces and the resourcesthat are exchangedvia these interfaces. This systemdesign stays virtually unchanged during the lifetime of the modularcomponent system and spans several generations of modules.Evenin the case where the system design is altered, the practical considerationsof upward-compatibility will only allowthe addition of somenewtypes of interfaces and resources, whichdoesnot affect anyof the older modulestypes. Thus, for a modularcomponent system, the component types will fit quite naturallyinto the resource-based model. 3. Resource-Balancing 3.1 The Principle Theidea behindthe resource-balancing principle is simple: Aconfigurationis not acceptableunless the resourceswhich the environment and the componentsdemandare each balanced by the resources whichthe componentsand the environment can maximally supply[11, 24]. This suggestsa basic configuringalgorithmwith the resourcemodel,which mosthumznexperts employconsciouslyor subconsciously: Starting fromthe resourcesdemanded by the environment as stated in the requirementsspecification for the technical system,focuson a resourcetype not yet balanced,determine the list of component types whichcan supplythat resource, select one componenttype from the list, incorporate a component of that type into the technicalsystem,andrepeat that process until for every resourcethe required amount is balanced by the amountof resources supplied by components or by the environment,with backtrackingon the decisionsas the simpleststrategyto copewith dead-ends and with the situation wheninsufficient resourcesare supplied by the environment [24]. iiiiiiiiiiiiiiiiiiiiiiiiiiiii iiiiiiiiiiiiiiiiiiiiii iiiiiiiiiiiiiiiiiiiiiii ! iiiii!i iiiiiiiiiiiii iiiiiiiiiiiiii! iiii Figure3 resource-based modelof a technical4-- supplies ~ consumes systemandits environment - ~ uses Thecomponents availablefor buildingthe technical system are not arbitrarily designed but determined by considerationsof cost-effectivenessin the trade-off between universality, whichleads to lowcost throughhigh-volume production, and adaptation, whichreduces the cost of designing-in, manufacturing and assembling the componentsinto a product. Most cost-efficient for applicationareas wherethere are only fewandstandardized resource types but large variations in the amountof resourcesfromcase to case are modularcomponent systems. A modularcomponentsystemis a collection of types of moduleseach designedwith the objectives of workingwell together and of being easily configured into technical systemsfor a widespectrumof reqmrements. Themodules, 21 3.2 The levels of knowledge This configurationprocesscorrespondsto "reasoningfrom first principles".It requiresonly ¯ systemknowledge,i.e. knowledge about the resources in the systemdesignspecificationfor the modular componentsystem, and ¯ catalogueknowledge, i.e. the technical specifications of each component typically containedin the manufacturers catalogue, to find a formallyviableconfigurationif oneexists. The heuristic knowledgethat only humanexperts can provide from their experience with the configuration processis on twofurther levels: ¯ evaluation knowledge,e.g. knowledgeabout a measure of quality of the configurationandabouthowto predict it on the componentlevel during the configuration process,that will help achievea goodconfiguration. ¯ performanceknowledge,e.g. knowledgeabout some advantageous sequencingof decisionsthat will lead to an acceptableconfigurationquickly. ~ r~erformance knowledge~ evaluation knowledge catalogue knowledge system knowledge INN I Figure 4 Levels of knowledgein a resource-based model Themeasureof quality of a correct configurationis usually its cheapnessgiven that it satisfies the qualitative and quantitative requirements. That a configuration is not acceptableunless its price is belowa givenlimit can be expressed simply by makingthe cost of a componenta (commercial)resource that has to be provided by the environment.Whenmorethan one component is applicable, a local decisionaboutthe component with the least overall cost is necessary:the list of suitable component typesmust be sortedin orderof decreasingcheapness,withthe quotient of (units-of-resourcesprovided)and(price of component) the figure-of-merit,andthe first component typeof the list must be selected. To account for the cost of further components entailed by the selection of a component when computingthe figure-of-merit, the cost of the resources consumed or used by that component mustbe estimatedand addedto its purchaseprice. In the performanceknowledge,the humanexperts neednot concern themselveswith the basic organization of the configuration process like they must in mostrule-based approaches,but only with the fine-tuningof an inherently self-organizingconfigurationprocesswhichis alreadyquite efficient dueto the supply-consumption-characteristics of the components.For simple modularcomponentsystems, no free-tuningis necessaryat all, anda static sequencing priorities assignedheuristically to the resourceshas yet provedadequatefor optimizingthe sequenceof decisions for others. 22 Additionally, exception knowledgemust be represented. Instead of expressiongincompatibilityof components (our first approach),it has provenmosteffective to describe hopelesssituations in termsof resourcebalancesituations. For morecomplicatedsituations, simplerules whichchange the priority of resourcesatisfaction or component placement haveprovenadequate. 3.3. Resource Quality Thesimpleresourcemodeldescribedaboveis not sufficient to adequatelyrepresent reality becauseoften resources whichsuperficially seemthe same will in someaspects really be slightly or significantlydifferent, e.g. elecrical current is only equivalent whendelivered at the same voltage andpolarity or frequency,or, moreaccuratelyyet, withina specific voltageinterval. This fact weexpressby associatingquality attributes with the resourcetypes, e.g. voltage intervals or voltage and tolerance, polarity or frequencywith electrical current, rotational speed with torque, access time with memory capacity, baudrates with RS232-connections. Resource-Balancing becomesmuch moreintricate then, as the qualities haveto be compared in order to determinewhichprovidedresourcesare compatible with whichrequired resources. 3.4. Knowledgerepresentation Withthe resource-basedmodeleach of these four distinct levels of knowledge (see fig. 4) can be acquired incrementally and quite independently and can be representedin well-structuredknowledge bases. This leads to a decisiveimprovement in the maintainabilityespecially of the knowledge base whichwiUscale-upwell for the large knowledge basesencounteredin practical applications: Thesystem knowledge,i.e. knowledgeabout the types of resourcesthat arise fromthe systemdesignof the modular component system,can be organizedin a resource taxonomy basedon resource similarity, whichcan be exploited for resourcesubstitution decisions.Thequality attributes are representedin value-slotsof the resourcetype classes. Thecatalogueknowledgeis the largest and mostvolatile knowledge base. It contains the knowledge about the types of component that are availablefor configurations.Theideal representation paradigmfor the componenttypes are the classes of an object-oriented language, where the similarities between componenttypes can be used to construct a taxonomywith the catalogue componentsas leaves of the class tree. The types of resources that a componentmaytypically supply, consumeor use are introducedas value-slotsat suitablesuperclassesof the class tree togetherwith default valuesfor the amountof resource. Thusthe catalogue components can easily be entered into the knowledgebase by specializing an appropriate superclassandfilling in the specific valuesof the resource amountsfor that type of component. Theheuristic knowledge of humanexperts, i.e. exception knowledge, evaluation knowledge and performance knowledge, has natural places of attachmentin the classes of the component taxonomyor of the resource taxonomy: The evaluation knowledgeis easily expressible by the purchase-pricevalue for the component, andby specifying the per-unit-valueas a resourceattribute, whichcan be used in a standardprocedure as defaultfor the cost entailedbythe component in the computationof the figure-of-merit. Theperformanceknowledgeabout quick waysto reach an acceptableconfigurationis representedby a static priority attributedto resourcetypes or superclassesin a value-slot. Theexception knowledge,captured as rules on resource ¯ typesandamounts,is representedin value-slotsof resource types and their superclasses as part of the performance knowledge. This knowledgerepresentation scheme,by describingeach component type only in termsof the resourceseachinstance of that component type supplies, consumes anduses, avoids all direct references to other components and effectively isolates the knowledge about components from each other. All the knowledge, includinganyheuristic knowledge, even in formof rules, is organizedlocally withinthe compact and efficient structure of the componentand resource taxonomies,and can be found and accessed quickly and predictably by a humanexpert. Anewcomponent type thus can be addedto the component cataloguewithout reference or changeto older component type descriptions, and any component type description can be removed- together with all the knowledge pertainingto it - withoutconsequence for the rest of the component knowledgebase. Combined with the effect of representing not "by-rote" performancebut "principle-based" knowledge,even very large knowledge bases can be expectednot to showany of the maintenance. problemsrule-based configuration expert systems are plaguedwith. Theuniversality of this knowledge representation scheme for all kinds of modular technical systems makesit economicallyfeasible to provide powerfulinteractive standardtools for the acquisition and maintenance of the knowledge bases. The taxonomic structure of the component and resource type catalogues is exploited best with the familiar browsers displaying the class tree graphically and allowing access to the knowledge interactivelyvia the class nodeicons. 23 4. A Configuration Shell for Modular Systems The resource-based modeland the resource-balancing principle wasdevelopedfromstudies for our configuration expert system shell COSMOS (COnfiguration Shell for MOdularSystems). 4.1 The COSMOSArchitecture Figure 5 showsthe systemstructure of COSMOS with the resource catalogue and the componentcatalogue. Both resource types and componenttypes are represented as classes in an object-orientedlanguage.Mostresearcheffort wasspent towardsknowledgeacquisition and maintenance tools and towardsdebugging andexplanationfacilities that supportthe domainexpert in the development and tuningof the application-dependent knowledge base for the configuration shell. The resource types and component types andtheir superclassesare visualizedin browsers,the interactive tools are based on those browsersaugmented with dynzmicallyconstructedfill-in-the-blanks masksand menusfor value selection. [ list of parts, [ System structure of the expert system shell COSMOS The inference engine of the expert system shell COSMOS uses a blackboardarchitecture (see fig. 6) with a decision record, resourcemodelandbalancesheets, and the agenda. Thedecisionrecordcontains,for eachdecisionstep, the list of all viablecomponent typesin orderof decreasingutility, wherethe f~rst and highest-ranking componenttype is considered selected.Thebalancesheetstally for eachtypeof resource the amountrequired by the selected components against the amountsuppliedby the selected components. Theenvironment,carrying the requrrementsspecification, is enteredas the selectedcomponent in the first decision.In each inference step, from the columnof the selected components in the decision record, the balance sheet is updated, fromthe balancesheets the unbalancedresources are determinedandtheir treatmentput on the agenda,the actionwiththe highestpriority is chosen,either the selection of a component or the positioning of a component. For componentselection, all componenttypes from the component cataloguethat can supplythat resourcetype or a compatibleresourcetype are evaluatedin their prospective environment andlisted in orderof decreasingutility andthis list is takenas nextentry to the decisionrecord. system specification ii!i!iii!iiiii ii!!’ ’ iiiiiiiiii i’,’,iiii’,iiii’ !ii !!’,i’, Iiil i:ii?i?:i:ii!:!:!:!!!!!iiiiii:?:i:i:i:i:iiii!:iii!!!~iii?iii~ii:i i::::::::::::::::::::::::::::: component catalogue I [ componentY The COSMOSexperience Thoughaware of manypossible improvementsto the inferenceprocedure,in our ftrst implementation of the basic configuration expert system shell COSMOS wewantedto explorehowwell the simplebasic algorithmabstracted from humanexpert behavior would serve in practical applications. Wespeedily included an improvement that enables components to balancecertain resources locally, whichis a necessitye.g. for spatially distributed systems with local powersupplies. COSMOS has since been successfully used for a numberof prototype expert systems, e.g. for the configuring of Programmable Logic Controllers, ConveyorBelt Systems, and Switchgear System Controllers. A commercial re-implementationhas been successfully applied to the configuring of PLCsystemsincluding software and also including the configuring of the communication network hardware (Conquest, AEG/Schneider/MODICON). I u, is-~-,vo~] i ............... 4.2. [ Figure6 basic architectureof the COSMOS inferenceengine After the resource throughwhicha component is contained in another componenthas been summarilyprovided, the separatestep of positioningthe component gives an identity to that resourceandassociatesit with the component. Backtracking occurswhenthe list is empty.It is performed by returningto the previousdecisionin the decisionrecord, deletingthe first component typeof that list, whichis added to a list of all rejectedcomponents for that decisionstep, and proceedingfrom there. Backtrackingalso occurs whena hopelessresourcebalancesituation (accordingto exception knowledge) is detected. Animpossibleconfigurationtask is determinedif backtrackingreaches the first line which contains the environment as the selected component. Theexplanation engine is yet simple. For each step, a reconstructionof the balancesheet to the state prior to the selection of a component, together with knowledge of the resourceconsideredandthe evaluationresults for the list of viablecomponents, allowsto explainthe "whyselect ... ?" questions. Thelist of rejected componenttypes, which includesthe attribute relevantfor the rejection,takes careof the muchmoreinteresting "whynot select ... ?" - questions. 24 The acquisition and maintenanceof knowledgefor the component catalogues provedto be as easy as expectedof our resource-basedknowledgerepresentation. Throughthe COSMOS tool set, knowledgecan be entered by an expert without intercession of a knowledgeengineer. The knowledge(including communicationsand structured components)was acquired from paper catalogues and entered with less than four person weekseffort. Some training andexperience,of course, is neededfor the design of the basic representation structure and the ftrst representation of exceptions. The bulk of maintenance work,i.e. the introductionof newcomponents, the phasing out of discontinuedcomponents andthe updatingof prices, can nowbe performedby marketingpersonnel alone. The transfer of price informationfromEXCEL databasesto the COSMOS knowledgebase is automated. Mostnotable from the point of viewof marke~g wasthat the informationin the componentcatalogue, while organized locally like a paper-based catalogue,is free of its side-effectsand, withthe tools provided,mucheasier to maintain.Someideas about printing future paper catalogues directly from the knowledge base are entertained. 5. HierarchicalConfiguring TheResource-Based Paradigmis not limited to configuring of classical modulartechnical systemslike PLCwherethe components are concreteandthe structure of the productis simple. A very importantindustrial application area are industrial plants and installations whichare constructed froma large variety of different modularcomponent family systems. Suchapplications are well within the scope of automatic configuring by a suitable expansion of the resource-based approach to both the real and abstract components of such plants. In our approach,the top modelinglevel consists of the process elements as viewed by the project planning engineers.In manyapplicationareas, these abstract process elements exhibit a modular behaviour and functional capabilities whichcan be easily capturedand adequately expressedin a resource model.Theoverall task is then solvedby selecting and connectingenoughof suitable such process elements. However,each of the abstract process elementsitself will usually needto be configuredfrom smallerabstract or real components. Theresourcesprovided by the abstract component determinethe resourcesrequired of its parts. In manycases, the internal parts of the process elementscan be treated like additional components of the wholesystem,as long as the implicationsfor the scopeof the resource-balancing are takeninto account.Alternatively,the abstract component can be thoughtof as the environment for its internal component, andconfiguredin isolation. In both cases, the configuringprocessshouldbe structured into hierarchical layers, and the configuring done sucessively with knowledgebases for resources and compoennts appropriatefor just onelevel. An application domainwell suited to this hierarchical configurationapproachis the constructionof conveyor-belt systems.Figure7 sketchesthe multiplelayers andillustrates the component relationships typically encountered: TransportTask Belt Segment Belt Segment Mech.Power 2 kW@ 50rprn&MK2 Belt Segment 1 B$Contro[@ 200sps BertS eg rnentControlTas k.3 Gear1/24 Motor3.0kW Electr. Power BSC3 Subroutine LightBarrier CPU Bin arYB~o~lnlnput 3,0 kW@400V60Hz&C’ld 3,2 kW@ 400V60Hz&Ctd MCCUnit 7kW400V utputBoard Bin aB~grd :::::::::::::::::::::::::::::::: :::::::::::::::::::::::::::::::::::::::::::::::::::::: -::55:::::::-::: ::5~:-----:~ ............................... . ..........<..-~.-.....-.-.-.-..-_~-.-.-.-...-.~-......:’:-~-~::~--!: :::::::::::::::::::::::::: ..-.-.-.-................ .................... :Z~_-" ............. X’~,~-~ ........................................................... i,,,~,,o, [ E nv~ron rn~t. ,nfrastruclu re Figure7. Resource-Based Multi-Level Hierarchical Configuring of a Conveyor BeltSystem startingwiththeProcess Task 25 I Therequirementspecification is given on the Process Levelin formof a "TransportTask"Processandits partial transportresourcerequirements,e.g. transportdistanceat a certainvelocity, or changesin transportdirection. Onthe ProcessLevel, the required resourcesare then satisfied by selecting suitable cataloguecomponents, e.g. Belt Segments,until the required transport distance is reached.Thesecomponents maybe thoughtof as abstract or real. EachBelt Segmentthen posts its required resources, e.g. mechanicalpowerof certain amountand quality and controlof certain type, whichis satisfied on the Technology Level by selecting suitable motorsand gears or control programs andentailed sensors/actors,respectively. Aquality attribute is often usedas a meansto express mechanicalor geometricconstraints at the interface over whichthe resource is exchanged.Atypical examplein mechanicaldesign is the exchangeof mechanicalpower between a motorandload. Thequality attributes thenare the rotationalspeedof the motoraxle, the standardfor the shaft couplingand the standard for the flange mountingof the motor.In the exampleof Fig. 7, the mechanical interface is represented in the qualityattribute "Stub"at onelevel andas the quality attribute "ShaftForm" at another. The motor needs a certain amount of switchable electrical powerat a certain voltage,whichis satisfied by selecting a MotorControl Center (MCC) unit of suitable powerhandlingcapability, andsubsequentlysatisfying the requirements of the MCCby selecting a MCCFrameto house the unit. The voltage and the frequency of the electrical powerare affixed as quality attributes to the exchanged resource "electrical power". Thecontrol program,on the other hand, requires the resourcesof suitable subroutinesoftwareandsensors, which in turn require memory, sensor inputs andCPUpowerfroma control system, e.g. a Programmable LogicController. At that level, the quality and powerof the resource-based configurationtechniquehas alreadybeenestablished. Withsuch a complexsystem, even a resource-based configuration algorithm will get into trouble whenthe numberof componenttypes to choose from gets large enough, A muchmoreprofoundproblem,however,is that this conveyorbelt systemis not located at onepoint, whereit doesn’t matterwhois suppliedby whom,but is distributed over space and has a specific and necessary internal structure, whereit is importantthat the right amountof resourceis suppliedat the right place. 26 In systemswitha spatialextentor internalstructure,the balancingof resourcesis usually muchmoreintricate than the basic idea conveys.In suchsystems,it is necessary,but not sufficient that resourcesare balancedwithinthe whole system. Instead, most resources mustbe balancedwithin somescope. For resources which are involved in establishing "part-of-" or "contained-in-"relationships, whichwecall "containing" resources due to the role they play in the placementof components, the scopeis simplythe container (the component whichprovidesthe containingresources). For someother resources,it is obviousthat theyare scoped similarlyto containingresources,e.g. mass. In manydomains, like the conveyor-belt transport systems, the service of a technical systemdoes not only dependon the numberof components that constitute it but also on the structure in whichthe components are connected, e.g. the questionof whichother belt segments are controlled by the control systeminstance depicted in Figure 7, and whichother control systeminstancescontrol the other belt segmentsand transport systemsubsystems. For cataloguedesignof suchtechnicalsystems,the total configurationtask has to be partitionedinto configurationon multiplelevels. Foreachlevel, the refinement can be donein oneof three ways: ¯ the system/ subsystem is configuredout of smaller components froma catalogue,with a fixed structure, as for the "controlsystemlevel" in Figure7. ¯ the resourcesthat the system/ subsystemhas to provide are decomposed andcbon~redinto separate requirements manually or automatically,as in the "transport task", ¯ the system/ subsystemis replacedby or hierarchically refined into a decomposition of generic components and links, wherethe components determinehowthe higher level resourcesare transformedinto the lowerlevel resourcesand the links determinewhichcomponent has to provide whichresources to whichcomponents.This techniquewill be prevalenton the processlevel andthe technologicallevel. 6. Conclusions Theresource-based paradigm reflects the designrationale of modularcomponent systemsand captures a basic principle that human expert configurators are guided by. Resource-balancing is reasoningfrom"ftrst principles"and "deepknowledge", but with a simpleself-organizing basic inferenceprocess. In the resource-based model,knowledge aboutconfiguring technicalsystemsis layeredinto fourdistinct levels with well-definedresponsibilities andintuitive appeal.These knowledgelevels allow a well-structured knowledge representation with compactand mutually isolated knowledge for each component type. This isolation of knowledge aboutcomponents fromeach otheris the key to efficient maintenance of evenlarge componentknowledgebases. As first experiences corroborate,the resource-basedparadigm overcomesthe maintenance bottleneck for the caseof configuring technical systemsfrommodularcomponents. Theuniversality of the resource-basedparadigmmakes elaboratetools economically feasible. Through the available tools andthe inherentpowerof the resource-based model, the humanexperts can readily shun the services of knowledge engineersin knowledge acquisitionandneednot becomeinvolvedwith routine knowledge maintenance. Forresearch,the resource-balancing principleholdsmuch potential for moreadvanced inferencetechniques,andthe resource-based modelcouldbe an interestingstartingpoint for model-based design. References: [10] Steinberg, L.I: Designas Ref-mementPlus Constraint Propagation. The VEXED Experience. Proc. AAAI’87, 830-835. [11] Neumann,B.,, Weiner, L: AnlagenkonzoptmadBilanzverarbeittmg. 3. Workshop Planen und Konfigurieren (1989), Arbeitpapiere der GMD 388(1989), 209-224. [12]Cunis, R. et al.: PLAKON,EIN I.)BERGREIFENDESKONZEPT ZUR WISSENSREPR)i~EN’I’ATION UND PROBLEMLOSUNGBEI PLANUNGS- UND KON~-’IGURIERUNGSAUFGABEN. Prec. Expertensysteme ’87: Konzepteund Werkzeuge,Niirnberg, 1987, 406-419. [13] Mittal, S., Frayman,E : Towardsa generic modelof configuration tasks. Proe. IJCAI ’89, 1395-1401. [14]Baginsky, W. et al.: BASIC ARCHITECTURAL FEATURESOF CONFIGURATION EXPERTSYSTEMS FOR AUTOMATION ENGINEERING. Prec. IEEE/SICE AI’88, 603-607. [15]Mittal, S., Frayman, F.: Making Partial Choioes in Constraint Reasoning Problems. Prec. AAAI’87, 631-636. [16]Hakkarainen, K. et ah A KNOWLEDGE-BASED CONFIGURER FOR EaMBEDDEDCOMPUTERSYSTEMS SOFTWARE: REAL-LIFE EXPERIENCE. Prec. n~.h-~JSICE AI’88, 608-613. [17]Eschefle, A., Sachs, S.: Ein interaktives Werkzeugmar Konfiguration modularer Rechnersysteme. WIMPEL ’88, 322-328. [18]Eich, E. et al: CONSTRUCT - ein wissensbasierter An.sam zur Konfigurierung. 2. AnwendefforumExpertensysteme 1988, Duisburg. [19]Vitins, M.: KEN- eine Expertensystemumgebuag f/ir dis Konfiguration teclmischer Systeme. Bulletin S.E.V. 80(1989)3, 118-122. [20] Birmingham, W. et al.: MICON:A Single-Board ComputerSynthesis Tools. ~ CIRCUITS ANDDEVICESMAGAZIN,4(1988)1, 37-46. [1] McDermott,J.: RI: A Rule-Based Configurer of ComputerSystems. Artificial Intelligence 19 (1982) 39-88. [21] Soloway,F_. et al.: Assessing the Maintainability of XCON-In-RhME: Coping with the Problems of a VERYLarge Rule-Base. Proe. AAAI’87, 824-829. [2] Haugeneder,H. et al.: Knowledge-Based Configuration of Operating Systems - Problems in Modeling the Domain Knowledge. Prec. GI-KongreB"W’tsseasbasierte Systeme" 1985, 121-134. [22]Lehmann, E. et al.: SICONFEXein Expertensystem ftir die Konfiguration eines Betriehssystems. 15. GI-Jahrestagung, 1985, 792-805. [3] Borgida, A. et al.: KNOWLEDGE REPRESENTATION AS THE BASIS FOR REQUIREMENTS SPECIFICATION. Computer 18(1985)4, 82-91. [23]Dias, P.d.C., B/istschli, M.: Der lq.lrt¢atz der frame-basierten ExpertensystemShell KENfttr die Konfiguration nachriehtentechnischer Produkte. 2. AnwenderforumExpertensysteme 1989, Duisburg. [4] Mittal, S.,, Araya, A.: A Knowledge-BasedFrameworkfor Design. Prec. AAAI’86, 856-864. [24] Heinrich, M.: EIN GENERISCI-IES MODPJJ~ ZUR KONFIGURATION TECHNISCHER SYSTEME AUS MODULAREN KOMPONENTEN. 3. Workshop Planen und Konfigurieren (1989), Arbeitpapiere der GMD 388(1989), 49-57. [5] Lan, M.S. et al.: A KNOWLEDGE-BASED APPROACHTO PRINTINGPRESS CONFIGURATION. Prec. CAIA ’87, 38-43. [25]Heinrich, M.; Jtingst, W.E.: A Resource-Based Paradigm for the Configuring of Technical Systems from Modular Components. Proo. Seventh ~ Conf. on AI Applications (CAIA’91), Miami Beach, 1991, 257-264 [6] Frayrnan, E,, Mittal, S.: COSSACK: A Constraint-Based Expert System for Configuration Tasks. In: Sriram, D. and tLA. Adey (eds.), Knowledge-BasedExpert Systems in Engineering: "Planning and Design, 1987. [7] Wu, H. et al.: ISCS - A TOOLKIT FOR CONSTRUCThNG SYSTEM CONFIGURATORS. Prec. AAAI’86, 1015-1021 .[8] Pierick, J.: A KNOWLEDGE REPRESENTATION TECHNIQUE FOR SYSTEMS DEALING WITH HARDWARECONFIGURATION. Prec. AAAI’86, 991-995. [9] Streeker, H.: Configuration Using PLAKON - An Applications Perspective. Prec. 3. Int. GI-KongreB"W’tssensbasierte Systeme"1989, 352-362. [26] Wellma,, M.P., "A Computational Market Model for Distributed Configuration Design", .Proc. AAAl"94, pp 401-407. [27] Carlson, S.E.; Pegg, T.A.: GeneticOperators for Catalog Design. Prec. Advances in Design Automation 1995, Boston, MA. [28] Klein, 1L: Model Representation and Taxonomic Reasoning .in Configuration Problem Solving, Prec. 15th GWAI,1991, 182-194. [29] Klein, tL; Buchheit, M.; Nutt, W.: Configuration as Model Construction: The Constructive Problem Solving Approach. in: Gero and Sudweeks(eds):Artificial Intelligence in Design ’94, 201-218. 27