A Case-Based Approach to the Managementand Deployment of Knowledge Assets Rick Magaldi British AirwaysInformationManagement, CorporateBusiness Solutions Waterside, POBox 10, HeathrowAirport, Hounslow,Middlesex, UK Rick.V.Magaldi@British-Airways.com.uk From: AAAI Technical Report WS-99-10. Compilation copyright © 1999, AAAI (www.aaai.org). All rights reserved. distinctions made between commonsense and specialised expertise. Knowledge in this context can be categorised as either tacit, explicit or cultural [Choo1999]. Abstract This paper outlines howcase-basedreasoningtechnologymay satisfy certain outstanding requirements in the field of knowledgemanagement(KM). The innovative handling tacit andcultural knowledge is a keyfactor that will definethe future success of commercialorganisations. The growing recognitionthat KM is principally peoplefocusedshouldgive rise to a technologicalinfrastructurecapableof supportingthis requirement.Case-basedreasoningis ideally placed to handle the storage, indexingand retrieval of both tacit and cultural knowledge.A case study is described that deals with the production and deploymentof a case-basedreasoning system for the British AirwaysConcordeOlympuspowerplant. The system makesavailable the collective past experiences of Concorde engineeringstaff to assist in problemsolving. Tacit knowledgeconsists of the learning and know-howan individual accumulates through experience. Explicit knowledge consists of what has been declared and formalised in some way. Examples include, operating manuals and safety procedures. Cultural knowledge consists of what individuals communicate through both formal and informal means. Much of what is communicatedin this way is not recorded, but is sustained verbally. It is a very important mediumof knowledge exchangeand determines howsuccessfully the organisation functions on a daily basis. Introduction Many of the challenges faced by service-based organisations operating within the current marketplace can be characterised by a need to compete on quality, products, customer service and innovation. There is a growing recognition by many such institutions, that in order to achieve these goals, there is a need to move towards a more knowledge-basedbusiness infrastructure. This in turn creates a need to understand new ways in which knowledge, people and technology can be harmonised to ensure business success [Barr and Magaldi 1996]. A great deal of effort has been expended within the business world to make knowledgeexplicit for the benefit of the organisation. Expert systems have been deployed with varying degrees of success, and mucheffort has been devoted to knowledge elicitation methodologies to facilitate the harvesting of tacit knowledge. Such approaches will continue to be useful in certain domains, but a more flexible approach is required to meet the technological infrastructure requirementsfor the successful managementand deployment of knowledgeassets. The most efficient means of supporting knowledge workers of the future will require the co-operation of several technologies rather than one. These will support a knowledge sharing environment rather than one that is strictly knowledge-based. The core infrastructure to support knowledge sharing is already available on the Internet and Intranet. Electronic Mail and Groupware provide the first layers of a knowledgemanagement system (distribution level). The actual knowledge layer (knowledge storage, indexing, and retrieval) shown Figure 1, is difficult to control by means of more traditional information systems as they are largely concerned with the logistics of data. Approachesthat will enable the knowledgelayer to becomea reality rather than remaining an abstract concept will include a mixture of data-mining, neural computing, agents, and case-based reasoning technologies. As part of this process, attention The management of knowledge as a business asset is clearly going to be high on the agenda of corporate planners as restructuring of organisations is forced by rising costs. What form these changes will take in both practical and economic terms is still being debated by industry analysts. However,what is already clear is the need for an efficient means for the identification, refinement, storage, indexing and retrieval of key business knowledge. This will be required to attain the goals of efficient knowledge sharing and of just-in-time knowledgecapture within business. Organisational knowledgestructures can be very complex, requiring detailed models of processes both within the companyand the outside world. These must combine both spatial and temporal elements of experience, with 46 mayhave been forgotten can be utilised, and relearning of what has already been experienced avoided. must be given to handling all types of knowledge, including textual, graphical, and pictorial forms. Within British Airways case-based reasoning is being investigated as one potential means of managing its knowledgeassets [Burchell 1997]. These can be described as the insights and experiences gained by the organisation with regard to the products, technologies and markets that British Airways owns and uses to generate revenue. One area at an advancedstage is the application of case-based reasoning techniques to aircraft maintenance. Knowledge Infrastructure Aircraft Maintenance Figure 1. Knowledge sharing via Intranet and Internet Because case-based reasoning is grounded in a model of humancognition based on concrete experience rather than on dry abstractions it is natural and easy to grasp. It has an intuitive appeal to manyof those involved in knowledge management and is therefore gaining acceptance as a foundation KMtechnology (e.g. [Watson 1997]) from many organisations. A recent application example that combines case-based reasoning, relational database, and web technologies has been produced by Western Air Ltd. of Fremantle, WesternAustralia. They have deployed a case-based reasoning system that operates on the world wide web. It supports a sales force deployed over a wide geographical area whoare required to produce quotations for complexair conditioning system configurations. The benefits derived to date include increased profits ($1M), increased sales staff capability, release of experts to focus on more complex tasks, and increased customer satisfaction [Gardingen and Watson 1998] Case-Base Reasoning The basic assumptionbehind case-based reasoning is that if something has been experienced in the past, there maybe features of that experience that can be utilised to solve somecurrent problem or aid understanding of sometask or other. The general approach is to avoid fundamental problemsolving by taking more direct means. The strategy is dependenton the availability of someefficient meansof storing useful experiences, and indexing these for later retrieval by meansof an efficient retrieval engine. A case-based reasoning system therefore enables remembering from experience to take place. This is achieved by retrieval of those experiences most similar to somecurrent situation. By this process past insights that 47 Airline operating schedules require that swirl and efficient problemrectification takes place after aircraft defects are reported, if expensive delays are to be avoided. In such situations, the maintenance engineer is very likely in addition to the use of maintenancemanuals, to apply rich insights gained from experience, and in doing so, better gaugethe utility of certain diagnostic approaches,or assess the merits of particular maintenance actions. An experienced person is likely to ask such questions as "have I seen this problem before?", and "what produced an effective fix to this type of problem?"~ However, the uniformity of this type of approach can be very variable, due to a variety of factors including staff inexperience. It would be extremely helpful under these circumstances to be able to have a repository of diagnostic experiences that can be madeavailable to all engineering staff, including inexperienced engineers. For this reason case-based reasoning is being applied to support these requirements, and by doing so, to ensure that aircraft can be consistently and expertly maintained. The overall effect of this will be to minimise departure delays due to engineering problems and lower operating costs [Magaldi 1994). Current case-based reasoning work within British Airways Engineering is focused on support to diagnostics on the Concorde Olympus powerplant. This is a very complex assembly, comprising of a variable geometryintake system, main engine assembly, reheat system and reverse thrust assembly. Accurate diagnostics and repair procedures are essential to mitigate the effects of departure delays (these cost £1000 per minute) due to engineering problems, and maintain a cost effective operation, with a requirement for a very high standard of passenger service. To meet these needs, British Airways has produced a prototype case-based diagnostics system based on commercially available software from CaseBank Technologies IncTM. of Canada. The software is called SpotLightTM, and is optimised for aircraft maintenance diagnostic tasks. It runs under Microsoft Windows 95TM or taken place operationally since 1984. The following types of problemsare currently represented: NT 4.0 TM and is capable of links to any ODBC compliant database. Additional features include communicationswith programs using COMor OLEstandards, works on LANs (TCP/IP support) WANs, or as a stand-alone system. It can also support dial-in or internet access. ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ The current system configuration being deployed by British Airwaysis a stand-alone system for the duration of current operational trials. However,it is hopedto extend this into a networkedmulti-user facility in the future. SpotLightTM OlympusDiagnostic Tool The SpotLight TM case-based Olympus engine diagnostic tool assists maintenance technicians in solving problems with Olympus engine systems or equipment more efficiently, by telling themwhethera problemsimilar to the one that they are facing has already been solved at another place or time. Fire Warning& Extinguishing Intake Failure Low Power Oil System Rapid Acceleration Reheat Failure Reverse Failure Surge Fuel Vibration The system allows a user to input engine fault symptoms and associated observations in response to system directed questions. This is achieved by using a mouseto point and click on screen icons and hot-spots. Very little keyboard use is required. The purpose of this dialogue is to provide the user with a list of cases containing the most likely causes of the observed symptoms- the most likely being listed first. The system is considered as ideal for use in time-constrained situations, and should help reduce SpotLightTM facility has three major components:the casebase, the software that searches the case-base, and the user interface. Previously solved problems and their solutions are stored as cases. p ............. ~~,"i ~ ~ I’ t t t t ..~.,,~2.,=,,mm~ ~d,,. ~, ..,~ ~s .............. We, s theFTRag Sho~ng ? DidN~accelera~ nndmeJntedn n venue consi~te~ Ambient Temp. TOR &MinIdle? DidReheet Reselec~on h~ve 8We~le~ ? Was reheat problem evident onBolhECA~ MAIN ) &ECA,( ALT DidIgni~on eppeor Io bewoddng? .... liii ~ ;- -- i? i? ? ? i? Figure 2. SpotLightTM ConcordeOlympusDiagnostic Tool Figure 3. MainDiagnostic Session Window (initial state) A User can simply point and click on the symptomsthat best describe the current problem. SpotLightTM will ask pertinent questions to differentiate amongpossible cases and bring forwardthe cases in the case-base that best fit the problem.The user can then decide if the solution in any of the cases presented by SpotLightTM is the right one for the current problem. unproductive time spent during fault diagnostics, particularly whentimely decisions have to be made. The system, like a human,should also improve in accuracy as experience increases. This will be achieved in SpotLight TM by constantly adding new cases that have successful outcomes,as they occur. A seed case-base has been constructed around 270 cases dealing with a variety of operational defects relating to the Concorde Olympus powerplant and installation. These cases represent reported and recorded events that have A Typical Consultation A diagnostic session begins by selecting the appropriate diagnostic area, e.g., Intake Failure, LowPower, Oil System. On the basis of this selection, the system will 48 respond by introducing a list of questions to be answered. These questions are those that the system has judged most suitable for initiating a useful diagnostic dialogue for a particular diagnostic area. On answeringa given question, a search can be initiated by clicking on the ’search’ button, or several questions answeredat once before this is done. Whena search is initiated, the system search engine extracts those cases that most closely match the evidence so-far presented, and excludes those cases that have no ,/ W’n~.m Clelss is ~lrent? .¢" ~h, i~i¢~’i~q~ ---~ ~ ----:--~_ Figure 4. increasing) Consultation i P,eheettFoJture: ~__~ - -- ~, ............. t What w~$FuelRowb~heMour? t Was theFTRe,g Shoving ? t W~S theCON Ught( Amber ) illumin~ed? W~s ECA ( ALT ) Le~e selected DidReheet Resele~on hen,e enyerred ? WetsE~(M,~N) Lene selected? DidN2eccelerele andmeJnt~n a value c~nsislent with~bientTemp, TOR &MinIdle? Wo.s reheelproblem evidBra on BothEC~ ~N) ~ EC~ ALT )? DidIgnalon eppeer to be~rking? i answering further questions. Cases displaying high similarity scores can then be selected and their histories examined. Selection is achieved by ’highlighting’ the particular case to be examined, and then activating the ’view" button. The process is repeated for each case of interest and can be carried out in any order. Also, questions previously answered can have their values changed and newsearches carried out. ? 9 i? ? ? ? :? i? ? example (similarity Figure 5. Consultation example(similarity scores at or near 100%) scores matchingcharacteristics. So, from having a large number of cases initially, the potential numberof cases of interest is reduced through a series of steps. Finally, a few cases remain, these being judged by the system as having the highest matching characteristics with the current problem. Ranking of the matched cases is done numerically (1100%), and by colour coding, red, yellow and green. Red meansa low score, yellow is mediumand green is high. In practice, Yellow and green scores signify cases that are worthlooking at, rather than just green alone. It is important to note that when lists of questions are generated by the system, these can be treated in any order. Also, there is no obligation to answer them all, only as many as cover observed symptoms. Figs. 3. & 4. show examples of SpotLight TM on-screen information. A consultation session can be interrupted at any time and saved for later progress. A saved session remains the secure property of the user until it is shared for general viewing,or it is cancelled. Figure 6. Case Information Window Whenexamining a particular case, a new windowbecomes available containing a complete range of information about that case, Fig. 6, and includes: ¯ ¯ ¯ ¯ SymptomObservations Causes Repairs Explanations Each case follows a commonformat making it simple to comparedifferent cases and their outcomes. On completion of the initial question and answer session, the significant cases can be openedor search continued by 49 Conclusions to Date Initial assessment of case-based reasoning technology within British Airways has been very positive. The SpotLightTM diagnostic facility is now being extended to incorporate more of the events (potentially 2500+) that have occurred over a period of 20 years. These are being introduced to the system over a six month evaluation period in consultation with the user. In parallel with this, the Olympussystems model used by the reasoning engine is being refined with the help of Concorde engineering experts. Steve Mott, Powerplant Performance Engineer, for their valuable technical contribution, enthusiasmand support to the aimsof the project. This hands-on participation by experts in system build and refinement, gives a great deal of validity to the knowledge output from the system. It is essential that user community have confidence and trust in knowledgesharing systems of the type being described. Choo, C. W. The Intelligent Organization: Mobilizing Organizational Knowledgethrough Information Partnerships. 1999.Facultyof InformationStudies, Universityof Toronto. Also being given attention at this stage is the development of a return-on-investment plan that will help British Airwaysassess the key financial benefits to be achieved from the current and future application of case-based reasoning technology. The experience gained on the SpotLight TM project will be used to drive other programmes. The strategy for the broader deployment of case-based reasoning technology within British Airways is being developed, and is one that will initially focus on the logistics and economics of building, validating and maintaininglarge case-bases. It is realised that for existing knowledge sources to be exploited, methodologies, standards and procedures need to be developed. The ultimate success of first-generation case-based reasoning systems supporting knowledge managementwill depend on the facility to automatically, generate cases from diverse data sources, thus eliminating the need to enter large numbers cases by hand. References Barr, J. M., Magaldi, R. V. Corporate Knowledge Management for the Millennium. In Advancesin Case-BasedReasoning, Smith, I., Faltings, B. eds. 1996. Lecture Notesin Artificial Intelligence1168.Berlin.: Springer-Verlag. Burchell, B.1997, Computerson the case. WorldEngineering July/Aug:42-43. Gardingen, D., Watson,I. A WebBased CBRSystemfor HVAC Sales Support. In Applications and Innovations in Expert SystemsVI, Milne, R., Macintosh,A., &Bramer,M.eds. 1998. Proceedings of Expert Systems 98. Cambridge. : SGES Publications. Magaldi, R.V. Maintaining Aeroplanes in Time Constrained OperationalSituations using Case-BasedReasoning.In Advances in Case-BasedReasoning,Hatton, J. M., Keane,M., &Manago, M. eds. 1994. Lecture Notes in Artificial Intelligence 984. Berlin.: Springer-Verlag. Watson, I. Applying Case-BasedReasoning: Techniques for Enterprise Systems.1997. MorganKaufmann. Finally, British Airways Concordeengineering staff have recognised the case-based reasoning approach as one complementary to their own way of thinking and problem solving. Users currently involved in trials and evaluation of the SpotLightTM system, are very enthusiastic about the potential for such systems in the future, and have now requested that other Concordesystems be considered for inclusion within a future facility. Acknowledgements author thanks Phil D’Eon CEO,and the SpotLight development team of CaseBank Technologies Inc TM, for their continuous support to the Concorde Olympus programme. Also, the following British Airways Staff. Steve Lincoln, Concorde Fleet Technical Engineer and TM The 50