24. Intelligent Manufacturing Applications at Ford Motor Company.pdf

NAFIPS 2005 - 2005 Annual Meeting of the North American Fuzzy Information Processing SocietyIntelligent Manufacturing Applications at Ford MotorCompanyNestor RychtyckyjManufacturing Engineering SystemsFord Motor CompanyDearborn, MJ 48121nrychtyc@ford.comAbstract. There is a common misconception that theautomobile industry is slow to adapt new technologies,such as Artificial Intelligence (Al), into the manufacturingsector. In reality, many of the earliest adaptations of AIwere in the automotive sector where such diversetechnologies as expert systems, neural networks, geneticalgorithms, and fuzzy logic were among the first to beused. In this paper we will present an overview of how Aland knowledge-based technologies are currently beingapplied at Ford Motor Company within themanufacturing arena. Some of the applications that willbe described include an Al-based approach for vehicleassembly process planning, an application of Al forergonomics analysis and a system that utilizes machinetranslation to translate assembly build instructions forFord assembly plants that do not use English as theirprimary language. We will also discuss how specifictechnologies, such as natural language processing,controlled languages, and ontologies, can be used toeffectively deal with different types of knowledge, bothstructured and unstructured, that are prevalent in themanufacturing environment. The automobile industry isamong the most competitive in the entire world, and theuse of advanced technologies, is essential in the ongoingstruggle to prosper in the global marketplace.I. INTRODUCTIONMass-market automotive production is one of the mostcomplex and dynamic processes that exist in industry today.The rapid change in both the business climate and thetechnology requires corresponding changes in the underlyingproduct development and manufacturing processes.Therefore, it is not surprising that applications using ArtificialIntelligence (Al) and Knowledge-Based (KB) technologiescan be found in many facets of the automotive industry,starting from vehicle on-board intelligent functionalities andthrough the value chain including design, manufacturing andafter-market service.In this paper, we will discuss several types of knowledgebasedapplications that have been developed and deployed formanufacturing within Ford Motor Company. The goal of thispaper is to discuss how AI can be successfully integrated intomainstream manufacturing processes and provide acompetitive advantage. There are many other applications ofAI and knowledge-based technologies within the automotiveindustry that lie outside the scope of this paper [1]. Our paperwill discuss the following types of applications that arecurrently in use at Ford:* Process Planning for Manufacturing [2]* Ergonomic Analysis of Assembly Processes [3]* Machine Translation of Vehicle Assembly BuildInstructions [4]This paper is organized in the following manner: Section IIdiscusses an existing knowledge-based application that utilizesa controlled language and a knowledge-based approach tomanage and plan the manufacturing processes at Ford'sassembly plants. This system has been in production since theearly 1990s and has demonstrated how Al can be successfullyintegrated into the core business processes for manufacturing.In Section III, we will discuss how a knowledge-basedapproach has been used to help identify potential ergonomicconcems at our assembly plants. This system has alsoprovided an immediate and quantifiable benefit to thecompany in terms of reducing injury claims. In section IV, wewill discuss how "machine translation" has been customizedand used to translate process build instructions into otherlanguages, such as German, Spanish, Portuguese and Dutch,for the assembly plants that use these languages. In section V,we will discuss how knowledge-based technologies andstrategies are useful in other enterprise application areas,notably in the development of common ontologies andterminology management. The paper concludes with adiscussion of some current work and the potential for applyingknowledge-based systems for other applications within Ford.II. Al FOR PROCESS PLANNINGThe critical need to reduce cost and improve efficiencywithin vehicle assembly at Ford has led to the development ofan Al-based system as the core for the entire manufacturingprocess planning system. The Global Study ProcessAllocations System (GSPAS) was developed to incorporate astandardized methodology and a set of common businesspractices for the design and assembly of vehicles to be used byall assembly plants throughout the world. GSPAS contains anembedded Artificial Intelligence (Al) component, known asthe Direct Labor Management System (DLMS) that was0-7803-91 87-X/05/$20.00 ©2005 IEEE. 298Authorized licensed use limited to: Dalian University of Technology. Downloaded on November 2, 2008 at 19:47 from IEEE Xplore. Restrictions apply.

concept A subsumes concept B if, and only if, the set denotedby concept A includes the set denoted by concept B. The KL-ONE knowledge base can be described as a network ofconcepts with the general concepts being closer to the root ofthe tree and the more specific concepts being the leaves of thetree. A concept in the knowledge base inherits attributes fromthe nodes that subsume it. The power of the knowledgerepresentation system lies in the classification scheme. Thesystem will place a new concept into its appropriate place inthe taxonomy by utilizing the subsumption relation on theconcept's attributes.The GSPAS Al system interprets these instructions andgenerates a list of detailed actions that are required toimplement these instructions at the assembly plant level.These work instructions, known as "allocatable elements," areassociated with MODAPTS (MODular Arrangement ofPredetermined Time Standards) codes that are used tocalculate the time required to perform these actions.MODAPTS codes are widely utilized as a means of measuringthe body movements that are required to perform a physicalaction and have been accepted as a valid work measurementsystem [9]. For example, the MODAPTS code for moving asmall object with only a hand is M2; utilizing the arm gives acode of M3. The MODAPTS codes are then combined todescribe a entire sequence of actions. MODAPTS codes arethen converted into an equivalent time required to perform thataction. The output from the GSPAS Al system is sent to theappropriate assembly plants where the vehicle in question isbeing built. Engineers at the assembly plant then allocate thejob instructions among the available assembly operators. Thefollowing two figures display an example of StandardLanguage. Figure I shows a sample of a process sheet writtenin Standard Language. Figure 2 shows the results that aregenerated by the Al system for line 20 of the process sheet.The MODAPTS codes for each instruction are printed at theback of the line. Figure 3 displays the DLMS systemarchitecture and Figure 4 shows a small portion of theknowledge base.Resulting Work Instructions Generated by DLMS ForLine 20LOOSEN HEATER ASSEMBLY TURNSCREW USING POWER TOOLGRASP POWER TOOL (RT ANGLE NUTRUNNER) POSITION POWER TOOL (RT ANGLE NUTRUNNER) ACTIVATE POWER TOOL (RT ANGLE NUTRUNNER) REMOVE POWER TOOL (RT ANGLE NUTRUNNER) RELEASE POWER TOOL (RT ANGLE NUTRUNNER) Figure 2.The use of a standardized methodology and process for allvehicle programs allows us define and use a consistentapproach to the manufacturing process across multiple vehiclelines and assembly plants. Another significant advantage tousing a knowledge-based approach is easy maintainability.Since the original Al system has gone into production, wehave made thousands of changes to the knowledge base inorder to keep current with the very dynamic pace of theautomotive industry. These changes include the following:* Normal replacement and introduction of new vehicles.* Improvements and changes to the manufacturingprocess.* Introduction ofnew tooling and parts.* Introduction of new technologies, such as hybridvehicles and satellite radio.* Changes to the external business, such as the purchaseof Volvo and Land Rover.* Changes to the terminology in Standard Language.We have found that our approach to building and maintaininga knowledge base using "description logics" has proven to bevery successful.System ArchitectureProcess Sheet Written in Standard LanguageTITLE: ASSEMBLE IMMERSION HEATER TO ENGINE10 OBTAIN ENGINE BLOCK HEATER ASSEMBLY FROM STOCK20 LOOSEN HEATER ASSEMBLY TURNSCREW USING POWERTOOL30 APPLY GREASE TO RUBBER O-RING AND CORE OPENING40 INSERT HEATER ASSEMBLY INTO RIGHT REAR CORE PLUGHOSE50 ALIGN SCREW HEAD TO TOP OF HEATERTOOL 20 1 P AAPTCA TSEQ RT ANGLE NUTRUNNERTOOL 30 1 C COMM TSEQ GREASE BRUSHFigure 1.Figure 3: GSPAS Al architecture300Authorized licensed use limited to: Dalian University of Technology. Downloaded on November 2, 2008 at 19:47 from IEEE Xplore. Restrictions apply.

Knowledge BaseFigure 4: Portion of GSPAS Al knowledge baseIII. AI FOR ERGONOMIC ANALYSISAt the 6th Annual Applied Ergonomics Conference held inDallas, Texas in March of 2003, the Institute of IndustrialEngineers (IIE) awarded the "Ergo Cup" in Training &Education to Ford Motor Company for the GSPASErgonomics Application. Ford Motor Company has beenutilizing an integrated process planning system since 1990 tostandardize the process sheet writing, create work allocationsfor the plant floor and to estimate labor time accurately. Thissystem, formerly known as the Direct Labor ManagementSystem (DLMS) is a knowledge-based system that utilizes asemantic network knowledge representation scheme. DLMSutilizes techniques from natural language processing,description logics and classification-based reasoning togenerate detailed plant floor assembly instructions from highlevelprocess descriptions. This system also provides detailedestimates of the labor content that is required from theseprocess descriptions. Techniques, such as machine translationand evolutionary computation, were integrated into DLMS tosupport knowledge base maintenance and to deploy DLMS toFord's assembly plants that do not use English as their mainlanguage.The process sheet is the primary vehicle for conveyingvehicle assembly information from the central engineeringfunctions to the assembly plants. It contains specificinformation about work instructions and describes the partsand tools required for the build process. The work that isrequired to build the vehicle according to the process sheetinstructions is then allocated among the available personnel.Work allocation requires a precise means of measuring thelabor time that is needed for any particular task. The DLMSsystem interprets these instructions and generates a list ofdetailed actions that are required to implement theseinstructions at the assembly plant level.Subsequently, the ergonomics engineers would manuallyinspect the process sheets for possible ergonomic problems.Since there may be 2000 or more process sheets for everysingle type of vehicle that Ford manufactures, this manualtype of inspection was very labor intensive and timeconsuming.An ergonomics engineer would spend upwards oftwo weeks in manually inspecting each process buildinstruction for a vehicle.In an effort to streamline this process, we developed asystem that would automate the inspection of process sheetsfor ergonomics concerns. This work resulted in thedevelopment of an AI system for ergonomic analysis withinGSPAS that checks for two types of potential ergonomicsissues: "red" and "warning." Process sheets that are flagged as"red" will not be sent to the assembly plants until those errorsare corrected. Process sheets that are flagged with "warning"messages are released to the assembly plants; however, theergonomic specialists have the opportunity to check andapprove these processes.The Ergonomics system was developed and deployed toproduction in April of 2002. This has already resulted in asavings of over $17,000,000 in injury cost avoidance as thehigh risk processes never made it to the plant floor. Thesecalculations are based on the type of injuries prevented and thedirect cost associated per injury for each type of ergonomicproblem.IV. MACHINE TRANSLATIONWe have been utilizing a Machine Translation system atFord Motor Company since 1998 within Vehicle Operationsfor the translation of our process assembly build instructionsfrom English to German, Spanish, Portuguese and Dutch.This system was developed in conjunction with SystranSoftware Inc. and is an integral part of our worldwide processplanning system for manufacturing assembly. The input toour system is a set of process build instructions that are writtenusing a controlled language known as Standard Language.The process sheets are read by an AI system that parses theinstructions and creates detailed work tasks for each step ofthe assembly process. These work tasks are then released tothe assembly plants where specific workers are allocated foreach task. In order to support the assembly of vehicles atplants where the workers do not speak English, we utilizeMachine Translation (MT) technology to translate theseinstructions into the native language of these workers.Standard Language is a restricted subset of English andcontains a limited vocabulary of about 5000 words that alsoinclude acronyms, abbreviations, proper nouns and other Fordspecificterminology. In addition, Standard Language allowsthe process sheet writers to embed comments within StandardLanguage sentences. These comments are ignored by the Alsystem during its processing, but have to be translated by theMT system. Standard Language also utilizes some structuresthat are grammatically incorrect and create problems duringthe MT process. Therefore, the development of a translationsystem for these requirements entailed considerablecustomization to the SYSTRAN translation engines as well asa lot of effort in building the technical glossaries to enable301Authorized licensed use limited to: Dalian University of Technology. Downloaded on November 2, 2008 at 19:47 from IEEE Xplore. Restrictions apply.

correct translation of Ford-specific manufacturingterminology.Since Standard Language is constantly changing andevolving, it is necessary to keep the translation glossaries upto-date.It is also very critical to ensure that any changes tothe translation glossaries do not adversely impact any currenttranslations. This requirement led to development of a processto maintain and improve the accuracy of the translationglossaries and includes the following components:* The creation of a test corpus that contains a set ofsample process build instructions along with theexpected translations for each target language.* The creation and use of language modeling utilitiesthat allow us to analyze Standard Language anddetermine the usage frequency of terms and phrases.* A web-based system, known as the Systran ReviewManager, which allows users to maintain and addterminology to the technical glossaries.The use of Machine Translation technology has allowed us totranslate large amounts if data in a very cost-effective manner.The accuracy of the translation depends on the scope of thetranslation glossaries that need to be developed for the Fordspecificterminology. Our benchmarking has shown that theEnglish-German translations are 95% accurate; otherlanguages are less accurate, but this will improve as thetranslation glossaries are expanded for each language pair.We also hope to improve the accuracy of the translationprograms through the use of tagging technology, where thesource language text will contain XML tags with informationthat allows the translation engine to make better decisionsabout the source text.V. KNOWLEDGE MANAGEMENTAll of the systems described above are closely coupled withother manufacturing and product development systems. Thesesystems often use terminology that does not match the termsused within our knowledge bases. In addition, there is a lot ofunstructured knowledge that contains information about"lessons learned," "best practices" and other specificknowledge that needs to be integrated into our systems. It isextremely difficult to extract the appropriate knowledge fromthese different data sources and present it to the user in atimely and useful fashion.One example of knowledge integration is the need to beable to match part descriptions from the part database with theinformation provided in the process sheet. The process sheetis written in Standard Language and the parts database doesnot have this requirement. Our ergonomics processingrequires us to match the part from the process sheet with thepart description; if they match - we then perform someprocessing based on the part weight. In this case, many partshave multiple written representations, but represent one singlepart. We need to accommodate acronyms, synonyms,abbreviations, misspellings, misplaced punctuation and otherproblems inherent in natural language. The accuracy of ourergonomics analysis depends on how well we can find theappropriate part in the database and this requires more than asimple word match. Therefore, we need to build commonontologies that will allow us to match concepts from differentknowledge sources and leverage this information across theentire manufacturing and product development world. Ourfuture plans will focus on integrating unstructured knowledgefrom different sources, building a common ontologyframework and enabling our systems to communicate andexchange knowledge among themselves.VI. CONCLUSIONSIn this paper, we have discussed several applications withinthe manufacturing arena at Ford Motor Company that use Aland knowledge-based approached to solve critical businessproblems. The GSPAS Al system that Ford uses for processplanning has been in production for over ten years. It is usedthroughout the assembly plants to estimate the work that isneeded to build any given vehicle within Ford. We have alsobuilt a knowledge base that contains all the informationneeded to build a vehicle and have implemented StandardLanguage throughout Ford as part of the manufacturingprocess. Our success with the GSPAS Al system has allowedus to expand the usage of this technology into otherapplication areas, such as Ergonomics and MachineTranslation. We feel that these systems provide us acompetitive advantage in the manufacturing area and showhow these technologies can be utilized for other applications.References[1] Gusikhin, 0 and Rychtyckyj, N., (2004), "Al Applications in theAutomotive Industry", Proceedings ofthe Workshop on Intelligent Computingat the 3rd Mexican International Conference on Artificial Intelligence, MexicoCity, Mexico, April 26-30, 2004, pp. 7-16.[2] Rychtyckyj, N., (1999), "DLMS: Ten Years of AI for Vehicle AssemblyProcess Planning", AAAI-99/ILAI-99 Proceedings, Orlando, FL, July 18-22,1999, pp. 821-828, AAAI Press.[3] Rychtyckyj, Nestor (2004), "Ergonomics Analysis for Vehicle AssemblyUsing Artificial Intelligence", Proceedings of the 16th Conference onInnovative Applications ofArtificial Intelligence, July 25-29, 2004, San Jose,CA, pp. 793-798.[4] Rychtyckyj, N., (2002), "An Assessment of Machine Translation forVehicle Assembly Process Planning at Ford Motor Company'", MachineTranslation: From Research to Real Users, 5th Conference of the Associationfor Machine Translation in the Americas, AMTA 2002 Tiburon, CA, USA,October 2002 Proceedings, pp. 207-215, Springer-Verlag.[5] Klampfl, E., Gusikhin, O., Rossi, G., (2004), e-Workcell Application in amixed-model automotive assembly environment, Proceedings of2004JUSFA, 2001 Japan - USA Symposium of Flexible Automation, Denver,Colorado, July 19-21, UL_035.[6] Huijsen, Willem-Olaf (1998), "Controlled Language - An Introduction",in the Proceedings of the Second International Workshop on ControlledLanguage Applications (CLAW 98), May 21-22, 1998, LanguageTechnologies Institute, Carnegie Mellon University, Pittsburgh, PA, pp. 1-15.[7] McGuiness, D. and Patel-Schneider, P. 1998. Usability Issues inKnowledge Representations Systems. In Proceedings of the FifteenthNational Conference on Artificial Intelligence, 608-614. Menlo Park, CA:AAAI Press.[8] Brachman, R., Schmolze, J. 1985. An Overview of the KL-ONEKnowledge Representation System. Cognitive Science 9(2): 171-216.[9] Carey, Farrell, Hui, and Sullivan (2001), Heyde's Modapts: A Languageof Work. Heyde Dynamics PTY LTD302Authorized licensed use limited to: Dalian University of Technology. Downloaded on November 2, 2008 at 19:47 from IEEE Xplore. Restrictions apply.