A Multi-Agent Intelligent Design System Integrating Manufacturing and Shop-Floor Control

From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved. A Multi-AgentIntelligent Design System Integrating ManufacturingAnd Shop-Floor Control SivaramBsll ubramanianand Douglas H. Norrie Division of ManufaclnringEngineering Depamnentof MechanicalEngineering University of Calgely Calgary, AB, C.=nadn T2NIN4

Abm’aet Jacksonet al. 1993], serviceability [’Makinnet ul. 1989], and reo/d~ility [Markino et al. 1994] early into the A multi-agnnt architectnre has been developedfor the design phase. These aspects of the product are integration of design, riming, and shop floor considered conjointly with the design functions. This control activities. This is based on cooperating approach can reduce the number of redesigns, thus intellismt entities in the sub-domain~which make shorteningproduct developmentlead times. In addition, decisions through negotiation, using domain-specific better products with moretml~nced overall performance knowledgeboth distributed amongthe entities and be produced, because of early consideration of all ao~ssible to them. Using this archRecmral relevant aspects of the product.life cycle. framework, an Agent Based Coty~’rent Design Achieving good manufactarab~ty and minimum Environmentsystem has been developed for featare- productioncost is often consideredto be the mostcritical based design, manufacturability evaluation, and factors when implementing concurrent engineering dypnmlc process planning. This is a multi-agent design. This brings the downstreamconcerns up front prototype system involving the following types of and helps avoid costly redesign iterations at the agent: design agent; geometric interface agent; manufacturing stage. Using Design for feature agent; part agem;machine agent; tool agent; ManufLetumbilitytools, a designer can significantly environment manager; and shop floor manager. A improvedesigns through trade-offs betweenfunctionality new technique for evaluating manufacturability is and manufacturability. Other considerations, such as introduced,based on interacting intelligent features of assembly,maintainability, and serviceability can often he the part being designed. This proof-of-concept incorporated into the design phase in a similar manner. system was developedfor three-dimensional prismatic For effective concurrent engineering, software tools are parts, with twenty-fivedifferent feature types, but can neededfor systematically collecting knowledgeand data be extended to other geometries. The system has that are related to the design, manufacturing,service and been completed and tested, and is being integrated maintenance tasks and for synthesizing optimal design into a larger multi-agent environmentincorporating with b.qlnnced overall perform*nee. The two most muting, sehecbding, and overall productioncontrol. intportant capabilities for concurrent engineeringdesign softwarerelate to the generationof all feasible designand Introduction to the evaluation of ~ility and production costs jointly with design functional performancefor each Glob~liTAtion of the economy and the resulting alternative design. competitive industrial envimment demands that the Concurrent engineering aims at quick product indusuy develop new products with shorter lead times turnover and better b,qlanced product, by integrating and better overall performance.Product design decisions various issues of product life cycle right at the design play an important role in determining the form, cost, stage. Thoughthe philosophy and the need have been reliability and marketacceptance of the final product and well studied, the implementation of the concept hence the emerging concurrent engineering (or necessitates radically new approaches to ac~hieve simultaneous engineering) design methodologyalms to concurrency in the complexmulti-objective decisions, address various life-cycle values of a productin the early based on the large volumeof interacting domain-specific stages of its design [Ishii 1990]. Instead of followingthe information. One solution is to use cooperating conventional, sequentially arranged product development intelligent entities to represent domain specific process, concurrent engmeoring design incorporates knowledge and make decisions through a negotiation considerations such as mannfacturabifity [Mabajanet al. process. Such an implementationis possible by using the 1993, Regii et ul. 1994], assembly[Jackeila et al. 1985, concept of a heterogeneous multi-agent system. This paper describes a feature-based design system for

Balasubramanian 3 From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved. prismatic components, implemented within the mnsiderable improvement in feature recognition frameworkof multi-ngent based concurrent engineering techniques, they do not producea sufficiently complete design. The multi-ngent architecture for integrating set of data and certain mam~re]aged information design, manufacturing and shop-floor control is also has to comefrom other sources, thus making complete described and the implementationissues associated with automationdifficult. Moreover,feature recognition is a thig i~c~l 81~ analyzed. seqt~tial process and besga~mmment evaluation of features for mamffacturability is not feasible. Feature Related Work on Feature Based Dmign based d~_igndoes not have these disadvantages, but can have the problem of having too manyfeatmm to easily Featum-bueddesign is central to the Agent Based wanage. User configurable features provide a feasible Concurrent Design Environment developed. The major solution to this problem. However, fea~ design features of several key existing feature-based design representatioas are non-uniqne, depending on the systemsare therefore briefly reviewed. selection and use of features. Concummtautomation is TheFit3t-Cut project is a feature-besed design system possible with feature based design and hence, holds more in which individual features and their corresponding promisefor concurrent engineeringdesign. procc~ plans are combined to create a part and The initial approaches to concurrent engineering manufactmingplan concurrently [Cntkoskyet al. 1988]. design used a centrafi~l knowledgebmc. A disUributed The Next-Cut feature-based design system was developed approachis advantageoussince infommtionpertaining to to work on designs from wore than one level of detail ~in.¢ Call be separately and better organized and across multiple views while maintaining consistency and nmlti-objecfive decisiomcan be achievedthrough of part representation [Cutkoskyet al. 1992]. the cooperating entities aeee~ngthese Imowledgebases. The Ar/zoua State University Features Testbed is a Such an approach is fe~ble through the concept of system for designing, documentingand evaluating parts betemgeneonsmulti-agent ~tenm. and is or~nized into a shell for product definition and In recent years, several distributed agent approaches another for mappingand applications [Shah and Rogers to concurrcmengineering have been reported, including 1989]. The Feature Modeling Shell includes a form the large-granularity Palo Alto Collaborative Testbod feature modeleras well as other modelersfor tolerances, (PACT)[Cutknsy et al. 1993] in which existing design materials, etc. Each modeler can be used in setup mode and other systems were ’wrapped’to form agents and the to create specific features and feature knowledge.These finer-granularity Designworld system [Genesereth1991] feature libraries can then be used to create part models. in whichprocesses were implementedas agents. The Qtfick Turnaround Cell System (QTC) [Anderson and Chang 1990] combines a feature-based A Multi-Agent Framework for Concurrent design system with automatic process planning. In this Engineering system, the features are related to machiningoperations, but do not correspond on a one to one basis with Materials and shop floor resources greatly affect the ~rtannfafUMing f-e~ltuge$. CADETSis a feature based man~ity analysis. A design may be design system incorporating manofact~ngconstraints manufacturshleunder one set of shop floor resources, but [Wright and I-I~nam 1989]. not another. Likewise, the stock materials from which For products designod in an external location the the part will be manufacturedcan impact the numberof sequential process of feature recognition, process stepsrequired, and hence the cost of the design. Further, planning and scheduling maybe quite appropriate and the schedule of shop floor resources wiU have a one such approachis given in [Cmetat. 19941. significant impact on the process plan to be used and A general architecture for the design with features hencothe cost of design. concgpt is given in [Dixon 1988]. A comprehensive Ideally, a productdesigner should be awareof all the review of early research in mechanical engineering technical constraints (material~processes, tools, etc.) design including features and feature based design can be and shop floor consWaints(routing, scheduling, etc.). found in Finger and Dixon [1989a, 1989b]. A more But with the yew large amountof infommtitminvolved, snmmaryof recent research in feature based design is a designer often needs the assistance of other domain given by ~lomom[1993]. experts. Thin process of information exchange, whether foUowing The conclusions can be drawn from the it be sequential or concurrent, can be very time literature review. Feature identification and definition is consuming.One solution to this ’designer’s dilemma’is neeemmyto adequately represent tolerancing and other having intelligent interacting entities whichcan supply manufacturingrelated information. There are two basic domain-specific information to validate the product approaches: recognize (identify) appropriate features design, without any input fromthe designer. after desitmiQg with geometric elements or to design The hetemgencous multi-agent architecture which from the beginning using features. Despite the has been developedat TheUniversity of Calgary involves

FigureI l}esign Interactions intelligent interacting agents whichcollectively possess agents are knowledgeableabout their own domains and the necemuyknowledge about ,-~-ufamarability and can advise other agents about violatlom or alternative shop floor control for conconent product design and solutions in their dom~in~upon request. evaluation. These .agents dynamically access, transfer, In this multi-agent design environment, the design and evaluate muchof the lower level design information, agent carries out the design task oblivious to the transparently to the designer. Figure I shows typical constraints and restrictions of other related domains. design intem~omwithin this framework amongthe key Like any other agent, the design agent can request agents involved. information from another agent about a violation or In this architecture, the shop floor control agent alternative solution in the latter domain.He or she imposes an ’Ad~fiVe hierarchy’ over the shop floor define variations and let the other agents decide on resom~agents and.keeps track of the system state. The consequentoutcomes. In all cases, the ideal is to arrive adaptive hierarchy reflects the ability of the shop floor at multi-objective decision(s) concurrently, through the controlagent to maintain multiple virtual groupings of cooperative negotiation amongdifferent agents. resources (machine, tool, material handling equipment, The architecture can be readily extended to etc.), to accommodate the current ~ut policies incorporate other life cycle concerns such as assembly, and system status. Such an adaptive hierarchy can help service, recyclingetc. by the addition of suitable agents. imposeproduct priorities, changesin due dates and cope Thus the architecture is generic in mture. The ability to with internal distnrlmnces (machine breakdown, tool address design issues con~ntrreutly with the dynamicsof breakages,etc.), by reassiL, ning resourceagents. shop-flooris a noveltywith this apprmc,h. The. design feature agents help to define and evaluate The Agent Based Concurrent Design Environment rite local geometric, tolerance and other manufacturing (ABCDE)described below is a proof of concept prototype specifications of the product, concurrently. Each type of system implemented to demonstrate this muiti-agent feature has ’its ownfeature class, and each individual based conmrrent engineering framework. feature has its own fea~-e agent. The concept of intelfigent feauum was introduced by Mowchenkoet al. Implementation of ABCDE [1994] in eadier work on multi-agent based intelligent design. The inventory agent helps in addressing The Agent Based Concurrent Design Envimnmeut concerns about material requirements for manufacturing has been implemented using an object oriented the product. The machine and tool agents evaluate the programminglanguage (C++), whichfacilitated efficient manufactmability and produce process plans coding through object oriellted mechanismssuch as incrememally, as the design progresses. All of these parametrized types, multiple inheritance, and dynamic

N S V H I O R P O N F M L E O N O T R M M A A N N A A G O E E R R

Fig. 2 ABCDEAgeat Architecture

binding. The Advanced Modeling Extension and cbsmfcrand fillet. There are a total of twenty-five non- Application Programming Interface of Autodesk’s degematefeature types, inter--,y. The featere agents AutoCADwere used to implement the geometric user evaluate local manufeffmmbifity whenthey are first front end, and for graphic display and pmnipulation. This ~reated or whenever they get modified directly or was to provide a popular, low cost geomelxicfront end, indire~dy. Local man~ility evaluation iacludes with readily avnil#thle and proven dmedngaids. accessibility, stability, and tolmanm-sm~tcofinish Being an agent based system, the organization of compatibility,etc. ABCDEi~ m~ldnr. The overall system ai~hite~ure of The inventory agent keeps tracks of the available ABCDEis shown in Figure 2. This multi-agent system stock materials and their s lmp~ ~nd advises about is heterogeneous in nature involving the following altemativo stoc~ sizes that muldbe meal, if enonghstock different agent types: design agent, geomeuicinterface of requested size is not available. TI~ hellm the design ngent, feamm

6 ICMAS-9$ From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved. Every tool in the shop-floor is represented by an be manufactomble. The manufacturing knowledge autonomoustool ngenL The tool agent Ires knowledge implemented in the system has been collected and about its shape, schedule and toleran~ capabilities in synthesized from a widevariety of somces. combination with a particular machine and work Being a proof of concept prototype application, the mmmial, under certain operating parameterx This is numberof features and the degree of intelligence he.t~ in aummat~toot and operming pammmr by the system is limited for a real world problem. selection for a particular process request from a given Howeverthe architecture is modular and generic, and machine. As was noted cariier, the selectiun preoess is henceadditional features can he incorporatedover time. a~._,~d ~ bidding by the int©~l machine a~L A Design Session withABCDE The part agent dynanfically updates and w~intain¢ itself as the t~xmitotyfor bothproduct data and The designer can start the process by using a knowledge, as the design prngreme~ The product data preproeessedblank geometry.It shouldhe notedt lint this inclmim the geometric and ~ specifications blank geometry may only be tentative and can be while the product knowledge includes the conditions changedin the midst of the design process, if need arises. under which a particular specification was accepted as The inventory agent will automatically check the mamffnctumble.This helps in consistency validation available stock sizes and shapes for the required quantity both when the conditions change and also when the and advise the design agent accordingly. The design productis to be modifiedlater on in its life. agent mayevaluate available stock-oost options or may There are two agent coordinators, namely the request the purchase department for stock procurement, envigoBlnl~tt mann~er alld the shop-floor ma~q~l’, and or maypostpone tiffs discussion until later, in whichcase all of the other agents interact with each other through the processrepeats at a later stage. them. The presence of agent coordinators helps in The design agent (designer) selects a feature for the eliminating ~ communication among agents part, which requests the environment mnnRger to whichis irrelevantto the current m~ andin thedynamic instantiate this partio~tAr feature type on the blank The restmctu~g of agent groups (e.g. upon machine/tool managerredirects the request to the appropriate feature breakdown or addition). The environment manager agent. Next, the feature agent will request the design merely acts as a message redirector and possesses no agent, through the manager,for dimensional, lucationai, extra knowledgeof its own. tolerance and other m~mufa~uringrelated information. The shop floor w~nngerplays a vital part in shop After collecting enoughinforw~don, it starts evaluating floor control, in addition to its function and message local manufactorabilityin termsof accessibility, stability, redirection. It helps maintain virtual groupings of shop and so on. Violation of any constraint is promptly floor remurc~s to accommodate current wnnngement notified. If this agent has altered any existing feature, it policies and system status. It also helps in imposing informs the appropriate agent(s) to cant out consistency hierarchy within a group as required. For example, it checking. Anyunresolved issue is notified to the design broadcasts about a feature requirement only to those agent by the concerned agent. All of the relationship machine agents that are in the highest level of the information gathered in the process is stored for use hierarchy (primmyproce~ machines, such as milling or duringfuture alterations. drilling). These machine agents may then request the Subject to successful completion of local shop floor ~nagerto sub-contract to machinesat a inanity evaluations, the featnre agent requests lower level in the hierarchy (secondaryor tertiary process the environment managerto evaluate global concerns. machines, such as grinding, honing, or lapping). Apart This request is redirected to shop floor managerwho in from avoiding unnecessmy communicationamong agents tom auctions these reqmentsto machineagents irrelevant for the present task, this also helps in the belonging to the highest level of hierarchy in the machine agents remaining independent of system status different virtual groupings. As described earlier, these (i.e. they need not keep track of machineagents chooseappropriate processes and tooting, additiun/deletion/regrouping of resources). The shop evaluate global concernsand tO" to sub-contract for their floor nmn~ralso resolves the bidding process, in the shortcomings. This pruccss repents downto the lower case of morethan one bidder. It also notifies the design level of hierarchy until either the specifications are met agent if either noneor only incompletespecifications can or the lowest level in the hierarchy has been reached. The be satisfied. later case meansthat the shop floor has insufficient A primmyconcept in implementation was to make to meetthe requirements, reflected by nil bids the system a design advisory aid and not to place any for the feature. The success or failure of the biddin$ restrictions on conflict resolution. This meansthat the process is reported back by the shop floor managerto the designer is not foreed to clear up any or all unresolved design agent, back through environment mnn~ger and issues before proceedingfurther, but if he does resolve all feature agent, for further action. issues before proceedingso the product is guaranteed to

t_ i I_ F 4 -I F 4 ,’

Lever Bracket

Figure $

A successful feature thus far is accepted by the pan ~s. It should be noted that the process plans agent as manufafturable. A failure triggers suggestions, generated by the machineagents would l~nutin the same if any are available, by the feature agent to the design but the routing and schedule would change acamiing to agent. The final resolution is left to the design agent. the lowest bid. Workis in progress to integrate these Theproduct is thus designed by repeated instantiation of aspects of manufacturingand shop floor control into the features with concurrent manufacturability evaluations. Agent Based Concurrent Design Environmentsystem. Figure 3 shows a simple example of a product designed using the ABCDEsystem, involving only a few of the Conclusion different types of features available. The system is undergoingfurther testing and evaluation. The volumeof domain specific information involved and the complexity of addressing, integrating and Further Work automatingvarious aspects of product life cycle requires new approaches to co~t engineering d,_-_~"gn. A The alternative processes, operating parameters and heterogeneous multi-agent framework for concurrent tools chosen by the machine agents are components design provides a feasible solution for or~mmi~ins design which wouldnormally be included in the general process and ~m~fnctufing knowledge and integrating design plans of the parts. These plans can be madespecific by with manufacturingand shop floor control, through the using these componentswith setop planning, operation use of intelligent interacting entities capable of making sequencing and free tuning of the operations. These multi-objective decisions, concurrently through specific plans can be combined with routing and cooperation and negotiation. The Agent Based scheduling through a cost based model and bidding Concurrent Design Environment implemented as a process [13]. Production control for the entire demand proof-of-concept prototype of a feature-based design quantity can then be achieved through a repeated bidding system for prismatic components,amply illustrates the

8 ICMAS-95 From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved. ~ of this approach. This framework ean readily Kwok, A.D., and Norrle, D.H. 1992. IAO: A Multi- inmrpomteother aspects of the product life cycle such as Paradigm ProgrammingEnvironment. In Proceedings of assemblyor recyclingas .m~be required.The the International Conference on Object Oriented modnMrityand flexibility offered by the multi-agent Manufacturing Systems, 219-226. Calgmy,Canada. approach makes R an invaluable tool to automate and Mahajan P.V.; Corrado Poli; Rosin, D.W.; and Wozny, imp~nentmmmeat engineering practice. M.J. 1993. Design for Stamping: A Feature Based Approach. In Proceedings of the ASMENational Design Engineering Conference, 29-50. ]~akino, A.; Barkan. P.; and Pfaff, P,. 1989. Designfor Anderson, D.C.; and Chang, T.C. 1990. Geometric Serviceability. In Proceedings of the ASMEWinter Reamnin8 in Feature Based Design and Process AnnualMeeting, 117-120. planning: Computersand Graphics, 14(2):10-19. Marco, D.P.; Eubanks, C.F.; and Ishii, K. 1994. Cutlmsky, M.R., Tenen~nmJ.M., and Muiler, D. 1988. Compatibility ,~’~is of Product Design for Features in Procem Based Design. In ~ of the Recyclabll/ty and Reuse. In Proceedings of the ASME ASMEInternational Computers in Engineering International Computersin Engineering Conference and Conference and Exhibition, 55%562. San Francisco, Exhibition, 105-112. California Mowc~ako,M.; Norrie, D.FL; and ]3alaqubrnmsmiaq, S. Cmkosky, M.R.; Tenanbaum, J.M.; and Brown, D.R. 1994. Intelligent IndependentFeatures: 1992. Working with Multiple Representations in a Features Which Ensure Their OwnManufacturability. Concurrent Demga System. Journalof Mechanical Forthcoming. Des/gn, 114(3):515-524 Regli W.C.; Gupla, S.K.; and Nan, D.S. 1994. Feature Cutkosky, M_R.et aL 1993. PACT:An Experiment in Recognition for Manufacturability Analysis. In Integrating Concurrent Engineering Systems. Computer, Proceedings of the ASMEInternational Computers in 25(1):28-37. Engineering Conferenceand Exhibition, 93-104. Dixon, J.R. 1988. Desit, nlng with Features: Building Salomons,O.W.; Van Houten, F.J.A.M.; and Kals, l-I.J.J. Man_t~acUningKnowledge into More Intelligent CAD 1993. Review of Research in Feature Based Design, Systems. In Proceedings of the Manufactming Journalof ManufacturingSystems, 12(2 ): 113 - 132. International Conference,1:51-57. Atlanta, Georgia. Shah, J.J.; and Rogers, M.T. 1989. Feature Based Finger, S.; and Dixon, J.1L 1989a. A Reviewof Research Modelling Shell: Design and Implementation. In in MechanicalEngineering Design. Part I: Descriptive, Proceedings of the ASMEComputer and Engineering Priscriptive and Computer Based Models of Design Conference, 255-269. Processe~Research in Engineering Design, 1(1):51-68. Shoham, Y. 1993. Agent-Oriented Programming. Finger, S.; and Dixon, J.IL 1989b. A Reviewof Research Arti~cial Intelligence 60:51-92. in M[efhanifal Engineering Design. Part II: Wright, T.L.; and Hannam,ILG. 1989. A Feature Based Represelllation&,Analy~s and Designfor the Life Cycle. Design for Manufacture: CADCAMPackage, Computer Reseaevh in Engineering Design, 1(2):121-137. Aided Engineering Journal, 215-220. Genesere~ tCLIL 1991. Designworld. In Proceedings of the IEEEConf. on Robotics and Automation. 2785-2788. Los Alanms,California. Gu, P.; Balasubra~anlan, S.; and Norrle, D.H. 1994. Bidding-Based Profess Plant~mg and Scheduling in a Multi-Agent System. FoRhcoming. Isbii, K. 1990. The Role of Computersin Simultanmus Engineering. In Proceedings of the ASMEInternational Computersin Engineering Conference and Exhibition, h217-224. Jackeila, M.; Pal~Jam-bres,P.; and Ulsoy, A.G. 1985. Progralnming Op~ Sugg(~tiOllS in the Design Concept Phase: Applications to the Boothroyd Assembly Charts. ASMEJournal of Mechanisms, Transmissions and Automafionin Design, 107(2):285-291. Jackson, S.D.; Sutton, J. C.; and Zorowski, C. F. 1993. Design for AssemblyUsing Fuzzy Sets. In Proceedings of the ASMENational Design Engineering Conference, 117-122.

Balasubramanian 9