O
pen
A
rchive
T
OULOUSE
A
rchive
O
uverte (
OATAO
)
OATAO is an open access repository that collects the work of Toulouse researchers and
makes it freely available over the web where possible.
This is an author-deposited version published in :
http://oatao.univ-toulouse.fr/
Eprints ID : 9586
To link to this article : DOI :
10.1016/j.compind.2013.07.004
URL :
http://dx.doi.org/10.1016/j.compind.2013.07.004
To cite this version : Jabrouni, Hicham and Kamsu Foguem, Bernard
and Geneste, Laurent and Vaysse, Christophe Analysis reuse exploiting
taxonomical information and belief assignment in industrial problem
solving. (2013) Computers in Industry, vol. 64 (n° 8). pp. 1035-1044.
ISSN 0166-3615
Any correspondance concerning this service should be sent to the repository
administrator:
[email protected]
Analysis
reuse
exploiting
taxonomical
information
and
belief
assignment
in
industrial
problem
solving
Hicham
Jabrouni
a,b,
Bernard
Kamsu-Foguem
a,*
,
Laurent
Geneste
a,
Christophe
Vaysse
baLaboratoryofProductionEngineering(LGP),EA1905,ENIT-INPTUniversityofToulouse,47Avenued’Azereix,BP1629,65016TarbesCedex,France bALSTOMTRANSPORT,RueduDocteurGuinier–BP4,65600Se´me´ac,France
1. Introduction
Forseveralyearsnow,theneedforcontinuousimprovementin
industrial products and process management has led many
companies to establish standardized procedures of Problem
Solving Methods. The objective set for these processes is to
addressproblemsthatariseatalllevelsinthecompanyincluding suppliers andthirdparties,inastreamlined, gradual,controlled andstructuredmanner.Differentprocessesofclassical problem-solvingmethodologiesareusedforthispurposeandwecancitein
particular: the PDCA process (Plan Do Check Act), the Eight
Disciplines (8D) process, theDMAICS (Define Measure Analyze
Improve Control Standardize) process or, more recently, the
ProductiveThinkingModelprocess(asurveyofproblem-solving methodologiescanbefoundin[1]).Anumberofkeythemescanbe consideredfromtheproblemsolvingliteraturefortheanalysisofa methodology[2]:generalityversusdomainspecificityofmethods;
problem structure; generic problem solving tasks; diagnostic
problemsolving;andremedialproblemsolving.Ingeneral,these
processes have in commonsteps describing the context ofthe
problemfollowedbytheanalysisstages(researchbyexpertsofthe
causesofproblem) andfinally forproposing andimplementing
correctiveandpreventiveactions[3].Iftheintroductionofprocess
problemsolvingisnowwidelyspreadinengineeringapplications, itisnotthesameforthereuseofexpertknowledgeusedinthese processes(experiencefeedback).However,theinvestmentmade, includingthosemadebyexperts,insolvingacomplexproblemis oftenconsiderableand thiscanleadtoknowledgecreationand retention[4].Knowledgemanagementmustthereforebeableto respondtosuchuncertaintiesandcontainelementsof learning-by-doingorexperiencefeedback.Forinstance,high-level
knowl-edge such as project constraint reasoning, problem resolution
methods,solution generationstrategiesor supply chain knowl-edgemustbecapturedandre-used,butasmuchasisneededfora reliableandprofitableknowledgecapitalizationandexploitation
[5]. We therefore propose in this paper to identify reuse
mechanismsofpreviouslyconductedanalysis(includingprocesses
for the reuse or recycling of analysis and the treatment of
knowledgedirectlyassociatedwiththeseactivities)toguidethe resolutionofanewproblem.
Academicresearchpublishedintheareaofexperiencefeedback canbeorganizedintotwobroadcategories.Inthefirstcategory, there are the studies that focus on the organizational aspect.
Organizationalmanagementdoesnotcovertheexplanationand
thejustification procedures,northedecision ofwhether touse competitiveselectionproceduresfortheassignmentoftechnical
analysis or not. Included in this framework is the model of
experientiallearning[6],themodeloflessonslearned[7]orthe generic modelof experience feedback systems [8]. In a second category,theemphasisismoreonknowledgerepresentationand
Keywords:
Rootcauseanalysis Semanticsimilarity Transferablebeliefmodel Experiencefeedback Railwayindustry
ABSTRACT
Totakeintoaccounttheexperiencefeedbackonsolvingcomplexproblemsinbusinessisdeemedasa wayto improve thequality ofproductsandprocesses. Only afewacademicworks, however,are concernedwith the representation and the instrumentation of experience feedback systems. We propose, in this paper, a model of experiences and mechanisms to use these experiences. More specifically,wewishtoencouragethereuseofalreadyperformedexpertanalysistoproposeapriori analysisinthesolvingofanewproblem.Theproposalisbasedonarepresentationinthecontextofthe experience of using a conceptual marker and an explicit representation of the analysis incorporating expert opinionsand thefusion of theseopinions. The experience feedback modelsand inference mechanismsareintegratedinacommercialsupporttoolforproblemsolvingmethodologies.Theresults obtainedtothispointhavealreadyledtothedefinitionoftheroleof‘‘RexManager’’withprinciplesof sustainablemanagementforcontinuousimprovementofindustrialprocessesincompanies.
* Correspondingauthor.Tel.:+33624302337;fax:+33562442708.
theassociatedinferencemechanisms,necessarytoinstrumentthe feedbackprocesses.Anexperiencefeedbackapproachissupported
where the relevant knowledge can be shared, categorized and
updated for their formal use in knowledge-based engineering
applications[9–11].
Most works done in thisobjective are based onthe use of
inference mechanisms similarto thoseproposed in Case-Based
Reasoning[12].ThisisthecaseofworksproposedbyBergmann
[13]andmorerecentlyanotherworkdescribedbyArmaghan[14]. Thispointseemsveryimportantandwewillresumeourproposal
ideas commonly used in the field of Case-Based Reasoning
(similaritysearch inparticular).Finally, thethesispresented in
[15]showstheimportanceofintegratingsubjectivedatainorder toenrichaninformationsystemfortheexperiencefeedback.Inthe
same viewpoint, Aven and Zio provide new insights on the
treatmentof uncertainties in risk analysis context tofaithfully representandexpresstheknowledgeavailabletobestsupportthe
decision making [16]. In complex domains it is usually quite
difficulttointroducecontextualinformationusingexpert’srulesas
background knowledge. However, sometimes that information
shouldbetakenintoaccountwhenmakingdecisions,asitprovides
some relevant knowledge [37]. Our propositions include both
objective and subjective views relating to analysis and the
characteristics of analysis, in problem solving processes. In
industrial practices, feedback fromtechnical investigations will
help improveprocessmanagement practicesand thequality of
productsor servicesand,thereby, strengthen feedbackand the
sharing of experience with the development of a prevention
culture,whichtargetseachemployeeaswellasmanagement.In thecontextofprocessingvariousnon-conformances,wespecified anevaluationphaseofthecriticality(Table1)oftheproblemtobe solvedtodefinethemeansputinplacetodealwithit.Forexample, theappropriateanalysisofconformancesandcorrectiveactions
may yield information that could lead to changes in resource
examinationand maintenance management, therebyimproving
qualityserviceandtheprogressofsafetyprograms.
AsshowninFig.1(below),weconsider,inourwork,experience thatcorrespondstoacontainerincorporatingthecontext(problem
descriptionandanalysis),analysis(expertonfindingthecauseof
the problem) and the solution (set of actions to resolve the
problem).EiandexperiencewillberepresentedbyatripletEi=Coi,
Ai,SiwhereCoi,AiandSiarerespectivelythecontext,analysisand
solutionEioftheexperience.Abaseofexperiencecorrespondstoa
setofexperiences: BExp¼fEi; i2f1;...;ngg.InFig. 1,wealso indicatedifferentinferencemechanismsthatcanbeconstructed frominformationstoredintheexperiments.Thefirstmechanismis aninvestigationofproblemssolvedinthepastusingasimilarity
measure of context. This operation is fairly standard in the
reasoningCase-based,butwewillproposeaparticularmethod, adaptedtothesubjectmatter.Thesecondinferencemechanismis theintendedreuseofanalysisexpert.Onthispoint,theproposal
described in this article has, to our knowledge, never been
addressedintheavailablescientificliterature.Thethirdinference mechanismsuggestedinFig.1istheadaptationofthesolutions. Thispointisnotdescribedinthearticle.
Thearticleisstructuredasfollows:inSection2wepresentthe proposedcontextrepresentation(Section2.1)andthemechanisms associatedresearchpreviousexperiencerelevanttoreuse(Section
2.2). Section3 is devoted tothe description of expert analysis (Section3.1) and theprinciplesproposed toguide theanalysis whenanewproblemistobesolved(Section3.2).Finally,asample
application and the software implementation are provided to
illustratetheseprinciples(Sections4and5).Section6indicatesthe findingsofthestudy.
Table1
theCriticalityMatrix.
CRITICALITY=(5*S+2*R+Q+M)*P
S Safety 7 Riskofcatastrophic:Fatalityormultipleinjures
3 Riskofcritical:Singlefatality,severeinjury
1 Marginal:Minorinjury
0 Insignificant
R RevenueService 7 Majorimpactoncustomerrevenueservice,stoporrampup.Oneorseveraltrains
orlinesectionnotinoperation,orservicenotprovided.
3 Revenueservicebelownominal,oneormoremajorfunctionsorperformancesnot achieved.Potentialimpactonrevenueserviceorservicerampup.
1 Revenueservicebelownominal,oneormoreminorfunctionsorperformancesnot achieved.
0 NoImpact
Q QualityofService 7 Majordelayondeliveryor/anddeliverywithmajornon
3 Minordelayondeliveryor/anddeliverywithminornon
1 N/A
0 Noimpact
M Maintenance 7 Majorimpactonmaintenancecosts
3 Minorimpactonmaintenancecosts
1 N/A
0 NoImpact
P Progress 7 Noownerassignedand/orno8Dmethodologyyetinplace.
3 Ownerassigned,8Dtoolavailable,rootcauseunderinvestigationbutnotyet found,Containmentactioncompleted(D4closed)
1 Rootcauseidentified,solutionbeingdevelopedbutnotyetfinalized(D6ongoing) 0 Rootcauseconfirmed,solutionbeingimplementedandREXfeedbackprovided(D8
closed)
Table2
Differentdistributionsofbeliefmasses.
S1 S2 S3 S4 S5 1 0 0 0 0 0.4 H1 0 1 0 0.3 0.3 H2 0 0 0 0 0 H3 0 0 0 0 0 H1[H2 0 0 1 0.7 0.3 H1[H3 0 0 0 0 0 H2[H3 0 0 0 0 0 H1[H2[H3 1 0 0 0 0
2. Modelofthecontextandthesearchmechanism 2.1. Contextmodeling
In order to simplify and systematize research of past
experiences,weproposetorepresentthecontextoftheproblem bymeansofatleasttwodescriptors.Thefirstrepresentsthetypeof productorcomponentaffectedbytheproblem.Dependingonthe
available knowledge, this product may correspond to a very
general entity (e.g. a train) or a more specific entity (e.g. a
pneumatic brake). To formalize this aspect of the problem
description, theusage of a taxonomy (hierarchical relationship type‘‘is-a’’ betweenconcepts,asshowninFig.2)isparticularly suitable.Indeed,itwillbeinthedescriptioncontext,toassociatea conceptfromcorrespondingtaxonomy(herethatoftheproducts/ components).WecalltheconceptCoi‘‘component’’associatedEi
experience.
Similarly,weassociateeachexperimenttoaconceptthatbest suitsthetypeofproblemmet.Itcanalsobetakenfromataxonomy ofproblems.Thus,thecontextofanexperiencewillbedescribed bytwoconcepts:respectively,thetaxonomyofproductsandthe taxonomyofproblems.ThecontextofanexperienceEiisdenoted
bythefollowingpairCi=<Coi,Pbi>
Note that the description of the context would, where
appropriate,involveadditionaldescriptors(typeattribute–value) butthispossibilitywillnotbecoveredinthisarticle.Inthiscase,it is possibletouseadditionalsimilaritymeasuresadaptedtothe attributetypesconcerned.
2.2. Thesearchmechanism
Thesearchmechanismisproposedbasedontheassessmentof
the similarity between thecontext of a newproblem and the
contexttosolveproblemsalreadysolved.Assumethecontextof thenewproblemisdescribedasapairC0=<Co0,Pb0>,whereCo0
andPb0denotesrespectivelythe‘‘Component’’andthe‘‘Problem’’
conceptsassociatedwiththenewproblem.Thegoalis,then,to measurewhichlevelofsimilaritywitheachCi=<Coi,Pbi>ofan experiencetotheexperiencesbase.Toevaluatethissimilarity,we proposetouseameasureofsemanticsimilaritybetweenconcepts
from the same taxonomy. Taxonomies provide a classification
basedonsimilaritiesandarenaturaltohumanbeingswhooften
work by association and abstraction. In addition, taxonomies
provideastructurebasedontwobasicinferencesthatwedoevery dayandareattheheartofinformationretrieval:
Identification:abilitytorecognizetheclassofanobjectfromits characteristics.
Specialization:abilitytoaddressthemore specificcategories thanthoserequestedinasearch.
Ontologies are often equated with taxonomichierarchies of
classes, class definitions, and the subsumption relation, but
ontologies neednot be limitedto theseforms. In practice,one
can consider the taxonomy as an element contributing to the
formation of the skeleton of ontologies, without axioms
con-strainingthepossibleinterpretationsforthedefinedterms[17]. Ontologiesareinteresting,sincetherearelinkedtohumannature, existenceandpropertiesofthemindwiththeformal representa-tionofknowledge[18].
Intheliterature,severalsimilaritymeasureshavebeenproposed. Overall,itispossibletodistinguishbetweenmeasuresbasedonlyon thetaxonomicstructureandexploitingfurtherinformation,usually fromcorpustextsofthedomain,whichcanrefinethelevelofthe similaritybetweenconcepts.Inourproposal,sincetheexploitable corpustextsarerarelyavailableinsufficientquantities,welimit ourselvestomeasuresbasedonthetaxonomystructures.Several measureshavebeenproposedinthisbackgroundwhere consider-ationliesontheexploitationoftaxonomicalfeatures,forexample themeasureproposedbyWuandPalmer[19]:
SimWP¼ðCo0;CoiÞ¼
2N3 N1þN2þ2N3
Othersimilaritymeasuresbasedontaxonomiescanbeused
as those of Leacock and Chodorow [20],Choi and Kim [21] or
Al-MubaidandNguyen[22].Morerecently,inBatetetal.[23]an originalsimilaritymeasureisproposedtoimprovetheprevious Fig.2.Thecomponentstaxonomy.
measurements (usingonly the structure of taxonomy) without imposing the use a corpus text that is difficult to obtain and
process. According to experimental results proposed by the
authors,thismeasuregetsverygoodperformances, comparable
tosimilarmeasuresbasedontheuseofaconsistentcorpustext
[24].Thismeasureisbasedonthenotionofsuperconcept.Letus considera conceptC, a superconcept of Cis anascendant (i.e.
ancestor) of Cina giventaxonomy. For example,in taxonomy
showninFig.2,ofsuperconceptsCo32areCo3andUniversal.We write, for a concept C, the set T(C)={SC|SCis a super-concept ofC}[{C}.Themeasureisbasedonthecalculationand aggrega-tionofthetotalnumberofsuperconceptsneededtocharacterize thetwocomparedconcepts(jTðCo0Þ[TðCoiÞj)andthenumberof
commonsuperconcepts(jTðCo0Þ\TðCoiÞj).
simBSVðCo0;CoiÞ¼ÿlog2
TðCo0Þ[TðCoiÞ
j j ÿjTðCo0Þ\TðCoiÞj
TðCo0Þ[TðCoiÞ
j j
Thismeasureisnot,however,normalized.Inordertobringthis
similarity measure within a normal interval [0,1], we add a
normalizationcoefficientofthemeasureinthefollowingway:
SimJKGðCo0;CoiÞ¼ SimBSVðCo0;CoiÞ log2ðHþ2Þ ; Co 06¼Co i 1; Co0¼Co i 8 < :
whereHindicates(appoints)theheightofthetaxonomy.Forthe examplealreadyusedtoillustratethemeasureofWuandPalmer, weobtain:
SimBSVðCo0;CoiÞ¼ÿlog2
5ÿ2
5 ¼ÿlog2
3 50:75 Andforthenormalizedmeasure,weobtain:
SimJKGðCo0;CoiÞ¼
ÿlog2ð3=5Þ
ÿlog2ð5Þ ¼0:32
Wecallsim*theconceptualsimilaritymeasureselectedfrom thegroupofavailablemeasures(fortherestof thearticle, and
becausethismeasurerevealsgoodperformancein practice,we
choosesim*=simJKG).Theaggregationoftwosimilaritymeasures
calculated(onthecomponentandtheproblem)willbecarriedout usingamathematicoperator,whichwould,forexample,bebased
on a Minkowski distance of order p between two points on
Euclideanspace: SimMinkðCo0;CiÞ¼
1ÿ 1 2ð1ÿSimðCo 0;Co iÞÞPþ 1 2ð1ÿSimðCp 0;Cp iÞÞP 1P :
wherepisaparameterusedinordertomodulatetheaggregation (forp=1,weobtaintheaverageofelementarysimilaritiesandone forinfinitepamaxoperator).Thus,foreachcontextofexperience belongstotheexperiencesdatabasewecanassociateasimilarity measurewiththecontextoftheproblem.Theexperiencescanthen beselected(ornot)forreuse,accordingtoatolerablerangeofthis similaritymeasure(bythresholding).
3. Modeloftheanalysisandthemechanismoftheanalysis
reuse
3.1. Modeloftheanalysis
Theprinciple ofanalysisfrequentlyusedin problem solving
processes is the search of root causes by the deepening of
understandingofthecentralelementsorissuesassociatedtothe
investigatedproblem(RootCauseAnalysis).Inthisapproach,some potentialcausesatthefirstlevelareexpressedbyexpertsandthen, onthebasisoftheseassumptions,afurtherinvestigationiscarried outbyinterviewingexpertsontheunderlyingoriginsofthecauses describedatthefirstlevel.Dataminingdiscoveredknowledge[38]
can be used as a complementary source of knowledge for the
expertknowledge,whichmightinturnleadtoarenewedeffortin thedata miningprocess that can help tosuggestivelyincrease existingdomainknowledge[25].
Thus, step-by-step, experts are progressing toward a set of profoundcausesso-called‘‘rootcauses’’thatareconsideredtobe themostprofoundbecausetheyarefundamentallyattherootof theproblem.Studyingaprobleminsuchgreatdepthistypically implementedinthe‘‘fivewhys’’method.Herewewillnotonly
focusonthemechanismsallowingdomainexpertstogradually
express their analysis, but we shall also consider whether the analysishaspinpointedthepotentialrootcausesofproblem,and
helpeddeterminehowtomakethemsolubleandmore
controlla-blewithpreventiveactions.Weshallcallafterhypotheses,noted Hi
ktoindicatethek
eme
hypothesisassociatedwithanexperienceEi.
ForanexperienceEi,weshallnoteHialltheassociatedhypotheses.
IntheexampleofFig.3,thesehypothesesare:H111,H12,H2,H311 andH32.
Toidentifythepromisingleads,wesuggestthatbyprivileging theindividualexpressionandthegroupwareoneverycauseroot, theexpertscontributetotheemergenceofconsistentsolutions. Forthatpurpose,theuseoftheTransferableBeliefModel(TBM)
[26]is interesting. Indeed,thismodel allows simultaneously,a representationinexpertopinions(with,wherenecessary,ofthe uncertainty)andthemanipulationofasetofopinionstomerge
themandobtainsummarizedinformationaboutthehypotheses.
Asanexample,letusassumethat,inanexperience,theanalysis
phase realized by the experts have led to formulate three
hypothesesH1,H2andH3.Theframeofdiscernment(alsoknown asreasoningspace)isthen
V
={H1,H2,H3}.ByusingtheMCT,the expertcanfullyexpresshimselfonallthedisjunctives combina-tions(with‘‘or’’)ofhypothesesbydistributingaUnitarianmasson allthepossiblecombinations.Here,allthesecombinationsare: 2V={H1,H2,H3,H1[H2,H1[H3,H2[H3,H1[H2[H3} combina-tionsof2Vmustbeequalto1,i.e. PBe2VmðBÞ¼1.WecallthisdistributionofmassBBA(BasicBeliefAssignment). FromaBBA,itispossibletocalculatethemeasureofplausibilityPls andthemeasureofcredibilityBelofeverycombinationAof2V:
PlsðAÞ¼P
BjB\A6¼?mðBÞ and BelðAÞ¼
P
BjBAmðBÞ Fig.3.Rootcauseanalysis.
Suchmeasureshavethegreatadvantageofbeingrelativelyeasy tointerpretastheprovided valuesaresimple percentages,and
thereby allowing experts to gain an understanding of the
challenging beliefs andcomplex situationsthat industrial orga-nizationsencounterintheirdailylives.
Table2illustratesvarioussituations,which canbemetwith regardtothedistributionoftheUnitarianmass.
Table 3 provides measures of plausibility and credibility associated with the assumptions in these different situations.
Situation S1 corresponds to an almost total uncertainty (the
credibilityofallthe hypothesesisequalto0,theirplausibilityis equalto1).SituationS2correspondstoacertaintyofhypothesisH1 (theplausibilityandcredibilityofthehypothesisH1areequalto1and the plausibility and credibility of other hypotheses are zero). SituationS3involvestheexclusionofhypothesisH3butaresidual uncertaintyofhypothesesH1andH2.InsituationS4weexcluded hypothesisH3;hypothesisH1isfavoredrelativelytohypothesisH2.In situationS5,apartofthemasswastransferredtotheemptyset(1), whichmeansthattheexpertdoesnotruleoutahypothesisnotyet explained(thepossibilitythatmistakesoromissionsmightarise).
Ifseveralexpertsexpressthemselves,itis possibletomerge
their opinions using a hierarchical fusion technique with the
‘‘cautious conjunction’’ operator(for experts in thesame disci-pline)andthe‘‘non-interactivedisjunction’’operator(forexperts fromdifferentdisciplines)[27].Wepickupontheworkof[28]
from a methodological viewpoint and extensively detail the
mechanisms of reuse integrating both subjective judgment of
expertsandobjectiveevaluations.Inallcases,theresultwillbea unitarydistributionofmasson2V.Sinceitisdifficulttokeepall thisinformation(combinatorialandcomputationalaspectscanbe quitelarge),weproposetosynthesizeinformationbyperforminga pignistictransformation[29](theBBAisthentransformedintoa probability distributionover the frameofdiscernment). This is calculatedasfollowsforthehypothesisHk:
subk¼ X A22V ;Hk2A mðAÞ jAjð1ÿmð?ÞÞ
The result is then a probability distribution over theset of hypothesesthat synthesizesthesubjective judgmentof experts regardingtheproblemasonethathasbeenresolved. Ata later
stage, additional tests [39] can validate (or invalidate) the
hypotheses and in this case we consider this result objective
(thehypothesisbecomesanassertionandmuchgreaterreliability isgiventotheassociatedknowledge).Thus,foreachhypothesis, wehaveasubjectivejudgmentofexpertsandanobjectiveresult (whichonehasultimatelybeenvalidatedonproblemsolving).This informationisrecorded,enrichingtheanalysisthatbecomes:Ai¼ f<Hi
k;Subik;Objik>g where Subik and Objik denote, respectively,
the summary of subjective opinions of domain experts and
objectiveoutcomesassociatedwithhypothesisHi k. 3.2. Mechanismforanalysisreuse
Themechanismwhichweproposeforthere-useoftheanalysis isasfollows.Duringthephaseofresearchforsimilarexperiences
(by using the context), we have selected a set of candidate
experiences(ofwhichthelevelofsimilarityincontextwiththe current problemis consideredsufficient (abovea given thresh-old)).
This set of experiences is noted Ea
sim, with E
a
sim¼ fEijSimMinkðC0;CiÞ
a
.Thesetofhypothesesthatweconsidertoproposeanapriori
analysis is the union of all hypotheses associated with the
experiencesofEa
sim.LetuspresumethatHsim¼[Hik;Ei2Easim.
We will assign two scores toeach hypothesis H of thisset
(whichwilldecidewhetherthishypothesiswillbeproposedornot toresolve thenewproblem). Thefirst score, called‘‘subjective score’’ and noted
s
sub(H), is calculated from the similarity incontexts and from the subjective opinions of the experts. The
second score, called ‘‘objective score’’ and noted
s
obj(H), is calculatedfromthesimilarityincontextsandfromthevalidation ofthehypotheses.Wethenhavethefollowingdefinitions:s
subðHKÞ¼C
Ei2Easimðu
ðSimMinkðC0
;CiÞ;subikÞÞ
s
objðHKÞ¼C
Ei2Easimðu
ðSimMinkðC0
;CiÞ;obji kÞÞ
where
u
indicatesafunctionoflocalaggregationandC
indicatesa functionofglobalaggregation.Itispossible,forexample,tochoose aproductoperatorforu
andindicatesamaximumoperatorforC
. 4. IllustrativeexampleTheexample wepresent in thissectionconcerns a problem
(corresponding to a simplified version of a real-life case from RailwayIndustry)offailurepoweronapneumaticbrakeofatrain. InTable4,thecontextofsixexperiencesalreadyrecordedislisted. Thecolumn‘‘Co’’istheconcernedcomponent,thecolumnPbis related toproblems.The columnssim1, sim2and simmink match
respectively to the values of similarity with the current case
(pneumaticbrake+powersupply)ofcomponent,oftheproblem
and the synthesis of both. Taxonomies used to describe the
contextsofexperiencesareshowninFig.4(forcomponents)and
Fig.5(forproblems).
Weassumethatsemantic(context-based)searchhasproduced
the outputs quantified in thefollowing paragraphs. Through a
threshold
a
=0.6,thefollowingsetisobtained:Ea0:6¼fðE1;0:68Þ;
ðE3;0:77Þ;ðE6;0:86Þg,thesecondvalueofeachcoupleindicates
the degree of similarity between the context of the current
problemwith:
A1¼f<H1;0;6;1;0>;<H2;0;3;0;0>;<H3;0;1;0;0>g
A3¼f<H1;0;1;0;0>;<H2;0;8;1;0>;<H4;0;1;0;0>g
A6¼f<H1;0;7;1;0>;<H3;0;2;0;0>;<H4;0;1;0;0>g
We deduce, using a local aggregate function ‘‘Product’’, the followingvalues: A0 1¼f<H1;0;41;0;68>; <H2;0;20;0;0>;<H3;0;07;0;0>g A0 3¼f<H1;0;8;0;0>;<H2;0;62;0;77>;<H4;0;08;0;0>g A06¼f<H1;0;60;0;86>;<H3;0;17;0;0>;<H4;0;09;0;0>g Table3
Credibilityandplausibilityofassumptions.
S1 S2 S3 S4 S5 Bel(H1) 0 1 0 0.3 0.3 Pls(H1) 1 1 1 1 0.6 Bel(H2) 0 0 0 0 0 Pls(H2) 1 0 1 0.7 0.3 Bel(H3) 0 0 0 0 0 Pls(H3) 1 0 0 0 0 Table4
Contextsandsimilaritiesofthestudiedexample.
Co Pb Sim1 Sim2 Simmink
E1 Brake Electrical 0.68 0.68 0.68
E2 Train Mechanical 0.17 0.17 0.17
E3 Pneumatic Electrical 1 0.68 0.77
E4 Bogie Pitting 0.17 0.11 0.14
E5 Motor Electrical 0.13 0.68 0.34
InthisexamplethesetHsimisdefinedinextensionbyHsim={H1,
H2,H3,H4}.
Wewilldetermineatthisstagetheinterestsofeachhypotheses of this set. We will use a global aggregate function, here the
maximum function is ‘‘max’’. We obtain the following result:
s
sub(H1)=max{0.41,0.08,0.6}=0.60Similarly, we obtain
s
sub(H2)=0.62,s
sub(H3)=0.17 ets
sub(H4)=0.09.Forobjectivescores,weobtains
obj(H1)=max{0, 9,0.0,0.8}=0.90,s
obj(H2)=0.7,s
obj(H3)=0.0ets
obj(H4)=0.0.WenoteherethattheassumptionsH1andH2canbeproposed totheuseriftheygetagoodscoreonthetwocriteria.
Taxonomy was later enriched approximately 4000concepts
through analysis of a set of working documents (examples
provided by experts). Concepts were combined using a set of
diagrams,whereprocesseshavebeenidentifiedwiththeirrelevant
inputs, the outputs, constraints and methods to support the
provisionofservices.Theseconceptswerethencompiledintoaset ofdiagramsrepresentinganontologicalmodelwithappropriate conceptualrelationsthathavebeenspecified[30].Itisimportant tosupportthefurtherdevelopmenttowarddomainontologywith axiomaticknowledge[31].
We interviewed 11 experts to assess the final taxonomy
(interviewswithexperts).Theinterviewsincludedapresentation
of 20min and werethen askedto assess the relevance of the
concepts of root causes, their relationships, and their use in
different contexts. This included a set of questions on the
Fig.4.Taxonomyofcomponentsfortheexample.
comprehensiveandconsistentinclusionofallconcepts;expertise can have an actor,the process in which thisactor is normally involved.Thisisaveryeffectiveandwidelyusedforthevalidation ofontologies[17,32].Thestrictapplicationoftheseissuesprovides ameanstoensurefullcoverageofconcepts(wheneveraconceptis consideredrelatedtoexistingconcepts,ifnecessary,itisaddedto thedomainvocabulary).
Ninemajorconceptsformtheoverallstructureofthetaxonomy component:
Vehiclebody:carcass,externaldesign, entrancefacilities,draw andbuffergear,gangway,windows.
Runninggear:supportingstructure,suspension,damping, wheel-setguidance,wheelset,runninggearandvehiclebody connec-tion,runningattachingpart,ancillaryelement.
Energysupply:currentguide,networkvoltagesystemprotection, switchanddetection,maintransformer.
Traction equipment: drive control, electrical power converter, propulsion.
Brake system: brake control,brake actuator pneumatic brake, brakeactuatorelectromagnetictrack,dynamicbraking. Auxiliarysupply:compressedairsupply,batteryequipment. Interior.
Centralcontrolandcommunication. Operationsinstrumentationandcontrol.
The integration of experience feedback methods and other
methods such as FMEA is a promising avenue for progress.
Regarding the similarity exploitation mechanism, it would be
interesting to go beyond the similarity measure based onthe
taxonomies(conceptsconnectedbyarelationshipis-a(X,Y))andto
propose a similarity measure incorporating the concept of
nomenclature (concepts connected by a relationship part-of
(X,Y)). This would enable one to obtain information related to enlargementdescriptions,which,in thiscase aretheadditional propertiesthatarenotexplicitlypartofthetaxonomystructure.
The practical implementation of the mechanism of reuse
analysisassumesthathypothesesaremadeinthesamemanner
fromoneexperiencetoanother,althoughinrealsituations,this constraint is rarelysatisfied.Indeed, experts who express their experience,ingeneral,useapersonalvocabulary;soitisdifficultto defineaformalizedexpressionofhypothesesinaconsensualand
exhaustivemannerwithintheframeworkofadynamic,iterative, and incrementalsolutionsmodel.Thisdifficultycanbe circum-ventedbycreatingataxonomyofrootcauses(liketaxonomiesof
components and problems) to assign each hypothesis with a
concept from thistaxonomy that best represented thetype of
expressedcauseswiththeunderlyingeventsandcircumstances
relevanttothatconcept.Itwouldbeeasyenoughtoassigneach concept ‘‘Cause’’two global measures(subjectiveand objective functions)usingthemechanism presentedfor theprocessingof hypotheses.
5. T-Rex:aworkflowmanagementsystem
Itisinterestingtoknowexperiencefeedbackfromthestagein theprocessingofnon-conformances,whichdependsonthelevelof informationgenerated(contexts–analysis–solutions).This dimen-sioniscriticaltomonitor,definethehistoryandtraceabilityofthe physicalflow.Thisiswhywefocusedonatechnologicalsolution orientedsoftwaretoolslike‘‘workflowmanagement’’.Workflow toolsarepresentedasatechnologicalidealtomeettheobjectives setbyreengineeringactivities[11].Workflowmanagementtightly controlstheflowofinformationaccordingtothespecificationsofa
given process. Processing tasks transfer information from one
persontoanotherinawell-definedmannerandalsoinvolvethe
developmentofcommunicationandnetworking.
ThesoftwareT-Rexisatoolsupportingtheprocessofproblem solvingandfeedbackbasedontheapproachoutlinedabove.
Wehavecompiledalistofthevariouscriteriathatmustsatisfy
T-Rexinacomprehensiveortargetedmanagementand
mainte-nanceof.Wedividedthesecriteriaintofoursections(asshownin
thefollowing table(see Table 5)): NonConformances
Manage-ment, Problem Solving, Action PlanManagement, Records, and
MeasurementsandReports.
TheT-Rexsoftwaresupportsseveralstandardmethodologies
forproblemsolving:PDCA(Plan,Do,Check,Act),8D,9S.Themain activitiesinthisprocessare:
Trainingteamproblemsolvinginastepwiseapproach. adescriptionandevaluationofcriticalityevents, characterizationoftheproblem,
analysisofeventsinordertofindtherootcausesandvalidate thisanalysis(e.g.‘‘is/isnot’’,‘‘5Whys’’Ishikawadiagram),
a proposal for a solution to the problem and its application (curativesolution),
thesuggestionofactionstopreventreoccurrenceoftheproblem (preventivesolutionandlessonslearned),
theadditionofsemanticcapacitiestothetextualsearchengines (bykeywordorsimilarity).
TheT-Rexsoftware,initscurrentversion,gavesatisfactionto thedifferentusersandthefirstpilotevaluationsoftheresultsby
theendusersare veryencouraging. Thissoftwarehasmadeit
easier for experts, especially in the process of validating the potentialrootcauses(seeFig.6).Theevaluationoftheapplication T-Rexhasledtopromisingresultsregardingboththe
responsive-ness and efficiency of the resolution. Note that beyond the
provisionoftools,implementationoffeedbackthecompanyhas beenaccompaniedbyaseriousconsiderationofthisneedwithin processesandledtothedefinitionoftheroleof‘‘RexManager’’of
which some activities is shown in the following table (see
Table6). 6. Conclusion
Theissueof experiencefeedback applied toproblemsolving processes throughsound and factual informationis in the best interestsoftheindustry.Weproposedinthiscommunicationanew enriched experience representation (resulting froma particular problemsolving)andtwomechanismstoexploittheinformation contentofcontextandanalysisbelongingtothisexperience.This proposal,whichalsosupplementstheapproachesalready estab-lishedastheFailureModesandEffectsAnalysis(FMEA),isbased ratheronapriorianalysissystemandnotonanapproachbasedon experience.Inmanyapplicationdomainstheexistingexperienced knowledgecouldbeputtomoreusesthanthoseprovidedbypriori analysissystems.Theproposedapproachalsoemulatesthepeople’s cognitivestructureandreasoningmodeltoachievebetterquality
and moreunderstandable outcomestosupport the capture and
reuseofanalysisandknowledge.
Finally, we defined performance indicators to assess the
effectivenessofexperiencefeedbackprocessthatweputinplace, thatistosayitsabilitytoneedsspecifiedbyendusers.
Measurement indicatorsareselected andapproved by
man-agementto:
analyzethetimeevolutionofthesituationinrelationtoeach objective,
makethecorrectiveorpreventivedecisionsthatareneeded, Table5
FunctionalpropertiesoftheT-Rexsoftware. Nonconformancesmanagement
Dealwithqualityissues(product,project,services,processes)includingtop 10issues
Stakeholdersidentification:
Activitymanager=personinchargeoftheplannedactivity(e.g.WP, program,project...)
QualityManageroftheactivity Technicalmanageroftheactivity Delegationbytheactivitymanager Uniqueidentifier
Linktoorigin/cause
CriticallyscoredbasedonTop10scoring(safety,revenueservice,qualityof service,maintenance,progress)
Description
Dates=targetdate,forecastdate,closuredate
Project/Program/Technology/Region/ProductLine/Platform/Sub-system/Site/ Processidentificationfromlist
Identificationofrepetitiveissues Problemsolving
8Dworkflow(AlstomTransportdefinition)withtoolssupportingcausal analysis
PDCAworkflow
Linkwithactionplanmanagement Problemclosureandarchiving Actionplanmanagement Actiontitle
Actiondescription
Linksorigin–action/causes–actions
Multi-siteandmulti-entity(ProductLine,Platform,Sub-system):accessfor monitoringandactionallocation.
Peopleidentification: Actionrequester(accountable) Actionowner(responsible)
NotificationtoactionownersviaLotusNotes Attributes:type,severity,priority,status Dates=targetdate,forecastdate,closuredate
Effort(man*month)andcostofaction(forecast,tocomplete,actual) Records,measurementsandreports
KPIextracts(pergroupofactions,versustime...): Timeopen,timetoclose,delay,closuredatetrend Costatcompletion,actualcost
Actionstobedonerate,progressindicator Listoflinked/interdependentactions
To-dolist(NonConformances/Actions/Problemscurrentornotclosedin duetime)
List/extractperRegion/ProductLine/Platform/Sub-system/Site/Project/ Process
CapabilitytogenerateExcelreport(fieldselection) Monthlyreportandreportonrequest
Commonlanguageforman-machineinterface,datarecordingand reporting=English
Table6
MechanismformonitoringREXactionsandprogress.
REXmeetings Responsibilities Attendees Frequency
ImplementationofREX capitalisation
REXManagerisinchargetofollowtheaction plan,supportedbyimpactedfunctionsthatare actionownersdependingentheREXtypology
REXManager
TheREXpilotinthecaseofissue(8D/PDCA) PQMinthecaseofRexanproject
QMSManagerorProcessOwnerinthecaseofREX onProcess
Every2weeks
REXCommitteeMeeting DefinitionofProduct/functions/projectimpart perimeter.
DefinitionofmainactionplantocapitalizeREX, withevidencesandVerificationofthe capitalisation,throughconcreteevidences chosenduringpreviousREXmeeting Deliverables:REXCardapproval+REXReview Report
REXManagerandSiteQualityDirector(optional), ProductDirector/ProjectDirector/TCEProduct Director,IndustrialDirector.SQAMgr,Chief Engineers(Includingtechnicalnetworks representatives,CCN,skillleaders,ifrelevantto thetopic),RAMSMgr
Eachmonth
Mufti-siteWorkshop amongREXManagers
Delivrables:ReadAcrossMatrix (Transversalization)
REXManagersfromdifferentsites(BECompVP orBESubsys.Directorsshouldparticipatetogive PLrecommendations)
measureandevaluatetheeffectivenessoftheassociatedactions tocontinuallyimprovetheeffectivenessofexperiencefeedback system.
Twotypesofindicatorshavebeendefinedforthispurpose: internalmonitoringindicators.
-NumberofRexcardsissued.
-Numberofcapitalizationsharesendedinmonths. KPIs(KeyPerformanceIndicator)
-Actionsoffeedbackresultedinthepastmonth(KPI). -Levelofcapitalization.
-Rexcostavoidance=costofnon-quality. -Costofobtainingquality.
Finally,thesoftwareT-Rexhasbeenfunctionallyvalidatedon severalcasesofsolvingsomeofgreatcomplexity.Currently,the stateofmaturityofthesoftwareisacommercialversion(named
ProWhy1). We wish to integrate others components of the
experiencefeedbackmodelsandinferencemechanismsforothers problemsolvingmethodologies.Thisintegrationwillallownews meansfortrackingprogressandpursuingspecifictargetsandgoals in collaboration with other relevant national and local stake-holders.
The taxonomy-based information structuring delivers the
foundationtocapitalizeknowledgealongwithassociatedbelief assignments.Taxonomiesclassifyanumberofhigh-levelconcepts andprovidewaysoftagginginformationfromcasesdescriptionin variousdomains.Taggingthecaseshencebecomesavaluableway
of assisting with the root cause analysis and possibly the
formulationofrecommendationsforimprovement.Theusercan
betterunderstandthestepsleadingtotheproblemresolutionand learn along the way. The quality (applicability, efficiency and
simplicity) of the taxonomy model and the business-oriented
evaluation are the essential elements to ensure semantic
interoperabilityandtosupportdecisionmakinginamorerigorous manner[33].
Within a domain specific target application, a dedicated
approach can aim the knowledge reuse to perform various
objectives(e.g.performance,reliabilityorsafety).So,ataxonomy
model can be semantically organized according to structuring
perspectivessuchasastructuralview,afunctionalview,aservice vieworanapplicationview.Implementationissuedealswiththe characterizationofspecializedtoolswithdedicatedindicatorsat
diverse levels as well as theprocessing (modulated by expert
beliefs to generate forthcoming opportunities and to drive
continuousimprovementonindustrialprocessissues.
Thismeansofextensibilitytootherdomainscanenablethe
manipulation of similarity measurement and exploitation of
reuseanalysisthataredefinedinourapproach.Forexample,this
is possible in a maintenance application where the major
indicatorsarerelatedtoconceptsofreliabilityengineeringand systemsafety[34].Inthiscontext,itwouldseemconvenientfor
industrial systems facing high failures to opt, among the
reliabilityengineeringinstrumentsavailable,forsystem regen-erationtorestoretheiravailability[35].Availabilityofthesystem is defined on the basis of the possible states of the required functionsforamissionandnotablyinthemodelingmethod;the startingpointremainstheconceptualtaxonomiesmodelforthe descriptionofsystems.Whensystemfailuresarebeingaddressed,
the maintenance recognizes the justification for additional
knowledge that may be gained by studying the problems
associatedwithspecificrootcauses[36].Asaresult,theproposed
methodologyfromtheproblem-solvingcontexthasapragmatic
potentialofextensibilitytootherdomainsinwhichthesuggested
techniqueswouldbedeployedthroughadjustmentsto
appropri-atemechanisms.
Acknowledgments
Theauthorswish tothankElisabethKuntzandEricReubrez, employeesofthe‘‘CenterofInformation,Decisionand Communi-cationforEnterprises’’,forthedevelopmentoftheT-Rexsoftware/ ProWhy1.
References
[1]T.DeJong,Problem-solvingmethodologies,in:KimberlyKempf-Leonard(Ed.), EncyclopediaofSocialMeasurement,Vol.3, ElsevierInc.,2005,,pp.171–177, ISBN:978-0-12-369398-3.
[2]J.DeMast,J.Lokkerbol,AnAnalysisoftheSixSigmaDMAICmethodfromthe perspectiveofproblemsolving,InternationalJournalofProductionEconomics 139(2)(2012)604–614.
[3] B.Kamsu-Foguem,T.Coudert,L.Geneste,C.Beler,Knowledgeformalizationin experiencefeedbackprocesses:anontology-basedapproach,Computersin In-dustry59(7)(2008)694–710.
[4] A.S.Choo,K.W.Linderman,R.G.Schroeder,Methodandcontextperspectiveson learningandknowledgecreationinqualitymanagement,JournalofOperations Management25(4)(2007)918–931.
[5]C.A.Costa,M.A.Luciano,C.P.Lima,R.I.M.Young,AssessmentofaProductRange Modelconcepttosupportdesignreuseusingrulebasedsystemsandcasebased reasoning,AdvancedEngineeringInformatics26(April(2))(2012)292–305.
[6]D.A.Kolb, Experiential Learning:Experience as the SourceofLearningand Development, PrenticeHall,EnglewoodCliffs,NJ,1984.
[7]R.Weber,D.W.Aha,B.I.Fernandez,Intelligentlessonslearnedsystems,Expert SystemswithApplications17(2001)17–34.
[8]H.Rakoto,P.Clermont,L.Geneste,Elaborationandexploitationoflessonslearned, IntelligentInformationProcessing(2002)297–300.
[9]P.Bermell-Garcia,W.J.C.Verhagen,S.Astwood,K.Krishnamurthy,J.L.Johnson, D. Ruiz,G.Scott, R.Curran, Aframeworkformanagementof knowledge-basedengineeringapplicationsassoftwareservices:enablingpersonalization andcodification,AdvancedEngineeringInformatics26(April(2))(2012)219– 230.
[10]W.J.C.Verhagen,P.Bermell-Garcia,R.E.C.vanDijk,R.Curran,Acriticalreviewof knowledge-basedengineering:anidentificationofresearchchallenges,Advanced EngineeringInformatics26(2012)5–15.
[11]M.ElKhadiri,A.ElFazziki,Howworkflowsystemsfacilitatebusinessprocess reengineering and improvement, InternationalJournalof ComputerScience Issues9(2)(2012)447–461.
[12]J.Kolodner,Case-BasedReasoning, MorganKaufmannPublishers,1993.
[13] R.Bergmann,ExperienceManagement:Foundations,DevelopmentMethodology, andInternet-BasedApplications.LectureNotesinComputerScience,vol.2432, Springer,2002.
[14] N.Armaghan,Contributiona` unsyste`mederetourd’expe´riencebase´ surle raisonnementa` partirdecasconversationnel:applicationa` lagestiondespannes demachinesindustrielles,InstitutNationalPolytechniquedeLorraine,2009(PhD Disertation)(inFrench).
[15]C.Tea,Retourd’expe´rienceetdonne´essubjectives:quelsyste`med’information pourlagestiondesrisques,Paristech,ENSAM,2009(The`sededoctorat).
[16]T.Aven,E.Zio,Someconsiderationsonthetreatmentofuncertaintiesinrisk assessmentforpracticaldecisionmaking,ReliabilityEngineeringandSystem Safety96(1)(2011)64–74.
[17]T.R.Gruber,Towardprinciplesforthedesignofontologiesusedforknowledge sharing,InternationalJournalofHuman–ComputerStudies43((November)5–6) (1995)907–928.
[18]O.Cairo´,S.Guardati,TheKAMETIImethodology:knowledgeacquisition, knowl-edgemodelingandknowledgegeneration,ExpertSystemswithApplications39 (9)(2012)8108–8114.
[19]Z.Wu,M.Palmer,Verbsemanticsandlexicalselection,in:32ndAnnualMeeting oftheAssociationforComputationalLinguistics,NewMexicoStateUniversity, LasCruces,NewMexico,(1994),pp.133–138.
[20]C.Leacock,M.Chodorow,in:C.Fellfaum(Ed.), CombiningLocalContextand WordNetSimilarityforWordSenseIdentification, MITPress,Cambridge, Mas-sachusetts,1998,pp.265–283.
[21]I.Choi,M.Kim,Topicdistillationusinghierarchyconcepttree,SpecialInterest GrouponInformationRetrieval(2003)371–372.
[22]H.Al-Mubaid,H.A.Nguyen,Measuringsemanticsimilaritybetweenbiomedical conceptswithinmultipleontologies,TransactionsonSystems,Man,and Cyber-netics–PartC39(2009)389–398.
[23] M.Batet,D.Sanchez,A.Valls,Anontology-basedmeasuretocomputesemantic similarityinbiomedicine,JournalofBiomedicalInformatics(2010).
[24] D.Sa´nchez,M.Batet,IsernD,Ontology-basedinformationcontentcomputation, Knowledge-BasedSystems24(2)(2011)297–303.
[25] F.Alonso,L.Martı´nez,A.Pe´rez,J.P.Valente,Cooperationbetweenexpert knowl-edgeanddataminingdiscoveredknowledge:lessonslearned,ExpertSystems withApplications39(8)(2012)7524–7535.
[26]P.Smets,R.Kennes,Thetransferablebeliefmodel,Artificialintelligence66(2) (1994)191–234.
[27]M.Ha-Duong,Hierarchicalfusionofexpertopinionsinthetransferablebelief model,applicationtoclimatesensitivity,InternationalJournalofApproximate Reasoning49(3)(2008)555–574.
[28]H.Jabrouni,B.Kamsu-Foguem,L.Geneste,C.Vaysse,Continuousimprovement throughknowledge-guidedanalysisinexperiencefeedback,Engineering Appli-cationsofArtificialIntelligence24(8)(2011)1419–1431.
[29] P.Smets,DecisionmakingintheTBM:thenecessityofthepignistic transforma-tion,InternationalJournalofApproximateReasoning38(2005)133–147.
[30] B. Kamsu-Foguem, Knowledge-basedsupport innon-destructive testing for healthmonitoringofaircraftstructures,AdvancedEngineeringInformatics26 (October(4))(2012)859–869.
[31] B.Kamsu-Foguem,G.Diallo,C.Foguem,Conceptualgraph-basedknowledge representation for supporting reasoning in African traditional medicine, Engineering Applications of Artificial Intelligence 26 (4) (2013) 1348– 1365.
[32]C.Welty,N.Guarino,Supportingontologicalanalysisoftaxonomicrelationships, DataandKnowledgeEngineering39(1)(2001)51–74.
[33]M.H.Karray,B.Chebel-Morello,N.Zerhouni,Aformalontologyforsemanticsin maintenanceplatforms,AppliedOntology7(3)(2012)269–310.
[34]B.Kamsu-Foguem,D.Noyes,Graph-basedreasoningincollaborativeknowledge managementforindustrialmaintenance,ComputersinIndustry(2013),http:// dx.doi.org/10.1016/j.compind.2013.06.013.
[35]M. Monnin,B. Iung,S. Olivier,Dynamic behaviouralmodel for assessing impactofregenerationactionsonsystemavailability:applicationtoweapon systems,ReliabilityEngineeringandSystemSafety96(March(3)) (2011) 410–424.
[36]P.PotesRuiz,B.KamsuFoguem,D.Noyes,Knowledgereuseintegratingthe collaborationfromexpertsinindustrialmaintenancemanagement, Knowledge-BasedSystems(2013),http://dx.doi.org/10.1016/j.knosys.2013.06.005. [37]B. Kamsu-Foguem, G.Tchuente´-Foguem, L. Allart,Y. Zennir,C. Vilhelm,H.
Mehdaoui,D.Zitouni,H.Hubert,M.Lemdani,P.Ravaux,User-centeredvisual analysisusingahybridreasoningarchitectureforintensivecareunits,Decision SupportSystems54(1)(2012)496–509.
[38]B.KamsuFoguem,F.Rigal,F.Mauget,Miningassociationrulesforthequality improvementoftheproductionprocess,ExpertSystemswithApplications40(4) (2013)1034–1045.
[39] B.KamsuFoguem,E.Simeu,Optimizingconstructionofscheduleddataflow graphforon-linetestability,TheMediterraneanJournalofComputersand Net-works8(4)(2012)125–133,ISSN1744-239.
Dr. Hicham Jabrouni received his BSc degree in Industrial and ComputerEngineering in 2007 from NationalSchoolofAppliedSciencesinAgadir,Morocco. Hereceived his MS and PhDdegrees in industrial EngineeringfromNationalPolytechnicInstitute, Tou-louse,Francein2008and2012respectively.Hewas previouslyPhdQualityEngineeratAlstomTransport, Tarbes,France. He was leading the process of the implementationof8D methodologyand Experience feedbackapproachinseveralproductline&Business UnitofTransport sector.Heis presently Quality& continuousimprovementManageratALSTOMThermal PowerforGeneratorsMaintenanceBusinessUnit(main activity:servicesfornuclearpowerplantsgenerators),Belfort,France.Hismain areaofresearch(inpartnershipwithProductionEngineeringResearchLaboratory ofENIT)isExperiencefeedbackandknowledgemanagementprocessesincluding data fusion and knowledge discovery from imperfect data domains. His developmentsincludesimilaritiesmeasurestechniquesthatexploitExperiences modelingandfusiontechniques.Hisworkhasappearedinprestigiousjournalsand hasbeenpresentedatmanyinternationalconferences.
Dr. BernardKamsu-Foguemis currentlya tenured AssociateProfessorattheNationalEngineeringSchool ofTarbes(ENIT)ofNationalPolytechnicInstituteof Toulouse(INPT)andleadshisresearchactivitiesinthe ProductionEngineeringLaboratory(LGP)ofENIT-INPT, a research entity (EA1905) of the University of Toulouse.HehasaMaster’sinOperationalResearch, CombinatoricsandOptimization(2000)fromNational Polytechnic Institute of Grenoble, and a PhD in Computer Science and Automatic (2004) from the UniversityofMontpellier2.Hegotthe‘‘accreditationto superviseresearch’’(abbreviatedHDR)fromINPTin 2013, reflecting a consistent researchand typically around15–20publicationsinpeerreviewedjournals.InENIT,hismaincoursesare orientedonartificial intelligence methods, Ontology engineering,information systems, knowledge management and Visual analysis in Human–computer Interaction. His current research work concerns the focuses on Knowledge RepresentationandReasoning,Datamining(theanalysisstepofthe‘‘Knowledge DiscoveryinDatabases’’process)andKnowledgeManagementforCollaboration andDecisionSupport.Hisarticlesproposemethodologiesandrepresentationsthat arerelatedtoSemantic-basedInformationandEngineeringSystemswithparticular emphasesbothonknowledgeandengineeringapplications(e.g.quality,industrial maintenance,constructionandtelemedicine).
Prof. Laurent Genesteis Professor at the National EngineeringSchoolofTarbesoftheNational Polytech-nicInstituteofToulouse(ENIT-INPT).Hereceivedhis PhDdegreeoftheUniversityPaulSabatier(Toulouse)in 1995 and an accreditationto superviseresearch in 2002. He is currently head of the ‘‘Cognitive and Decisional Systems’’ ofthe ProductionManagement Laboratory in Tarbes. His current research interest relatetoknowledgeengineeringandmorespecifically toexperiencefeedbackandlessonslearnedforproblem solvinginindustrialorganizations.Heisco-authorof morethanonehundredpapersforinternationaland nationaljournalsandconferences.
Christophe Vaysse is currently Train Sub-System Director(TractionandPowerPack)atAlstomTransport, basedinTarbes.HehasbeengraduatedEngineerin 1992 inAutomationandComputerScienceat Poly-tech’Montpellier/ex-ISIM/University of Science and Technology of Montpellier. He performed an MBA ‘‘Dualcompetenceinmanagement’’attheUniversityof Nancy-Luxembourgin2002.Hespent13yearsinthe automotive domain, startingwith SAGEM-AUTOLIV, from1994,asdesignengineerintheairbagsdesign centerofOsny.From1996,5yearsforPRESSACGroup, he was Project Manager for electronic embedded automotive systems basedin Germany.From 2000 weworkedabroadinMexicoforARVINMERITORLVSasDoorsSystemsProgram ManagerfortheNewBeetledevelopment.BackinEuropefrom2003,hebecame QualityDirectorbasedinSaint-Die´ (88),Designcenter&Manufacturingoflatches (VW/PSA/FORD).Enteringintherailwaysdomainfrom2007asQualityDirectorin Tarbesduring3years,hedevelopedanR&DprogramwiththeENITofProblem SolvingandReturnofExperience.From2010,hebecameTrainSubsystemDirector forTractionandPower-packSystem.
