ARTICLE IN PRESS
JID:NEUCOM [m5G;April6,2017;14:20]
Neurocomputing 0 0 0 (2017) 1–7
ContentslistsavailableatScienceDirect
Neurocomputing
journalhomepage:www.elsevier.com/locate/neucom
Application of self-organizing map to failure modes and effects analysis methodology
Wui Lee Chang, Lie Meng Pang, Kai Meng Tay
∗Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia
a rt i c l e i n f o
Article history:
Received 23 September 2015 Revised 3 December 2015 Accepted 10 April 2016 Available online xxx Keywords:
Failure modes and effects analysis Clustering
Visualization Self-organizing map Risk Priority Number Interval
a b s t r a c t
Inthispaper,aself-organizingmap(SOM)neuralnetworkisusedtovisualizecorrectiveactionsoffail- uremodesandeffectsanalysis(FMEA).SOMisapopularunsupervisedneuralnetworkmodelthataims toproducealow-dimensionalmap(typically atwo-dimensionalmap)forvisualizinghigh-dimensional data.WithregardstoFMEA,itisapopularmethodologytoidentifypotentialfailuremodesforaprod- uctoraprocess,toassesstheriskassociatedwiththosefailuremodes,also, toidentifyandcarryout correctiveactionstoaddressthemostseriousconcerns.DespitethepopularityofFMEAinawiderange ofindustries,twowell-knownshortcomingsarethecomplexityoftheFMEAworksheetanditsintricacy ofuse.Tothebestofourknowledge,theuseofcomputationtechniquesforsolvingtheaforementioned shortcomingsislimited.TheuseofSOMinFMEAisnew.Inthispaper,correctiveactionsinFMEAarede- scribedintheirseverity,occurrenceanddetectscores.SOMisthenusedasavisualizationaidforFMEA userstoseetherelationshipamongcorrectiveactionsviaamap.ColorinformationfromtheSOMmap isthenincludedtotheFMEAworksheetforbettervisualization.Inaddition,aRiskPriorityNumberIn- tervalisusedtoallowcorrectiveactionstobeevaluatedandorderedingroups.Suchapproachprovides aquickandeasilyunderstandableframeworktoelucidateimportantinformationfromacomplexFMEA worksheet;thereforefacilitatingthedecision-makingtasksbyFMEAusers.Thesignificanceofthisstudy istwo-fold, viz., the useofSOM as an effectiveneuralnetworklearning paradigm tofacilitateFMEA implementations,and theuseofacomputationalvisualizationapproachtotacklethetwowell-known shortcomingsofFMEA.
© 2017ElsevierB.V.Allrightsreserved.
1. Introduction
Failure modes and effects analysis (FMEA) is an effective problem prevention and risk analysis methodology for defining, identifying,andeliminatingfailuresofasystem,design,process,or service[1].AsearchintheliteraturerevealsthatFMEAwasexten- sivelyusedinawiderangeofapplicationdomains,e.g.,aerospace [2],automotive[1],nuclear[3],electronic[4],manufacturing[5,6], chemical[7],mechanical[8],healthcareandhospital[9],andagri- culture[10].FMEAusuallystartswithidentifyingthefailuremodes of a system or process, understanding the causes and effects of each failure mode, and determining suitable corrective actions to eliminate or reduce the risk of the respective failure modes [1]. Traditionally, the risk of a failure mode is determined by a Risk Priority Number (RPN) model[1]. The RPNmodel considers threeriskfactorsasitsinputs,i.e.severity(S),occurrence(O),and detection(D), andproduces anRPN score(i.e.multiplicationofS, O,andD)astheoutput[1].SandOareseriousnessandfrequency
∗ Corresponding author.
E-mail addresses: [email protected] , [email protected] (K.M. Tay).
ofafailuremodeanditsrootcause(s),respectively,whileDisthe effectivenessoftheexistingmeasures indetectinga failuremode beforetheeffectofthefailuremodereachesthecustomer(s)[1].
WhiletheeffectivenessofFMEAhasbeendemonstrated,three shortcomings pertaining to practical implementation of FMEA are as follows. (1) its risk evaluation and prioritization issues [2,5,11,12];(2)thecomplexityoftheFMEAworksheet[13];and(3) itsintricacyofuse[13,14].Thefirstshortcomingiswellknownand muchresearch workshavebeen conducted[2,11].The firstshort- comingsuggests that the traditionalRPN modelis susceptible to a numberof limitations,among the popular are,(1) relative im- portance among S, O andD is not taken into consideration [2], (2)differentcombinationsofS,OandDmayproduceexactlythe samevalueofRPN,buttheir hiddenrisk implicationsmaybe to- tallydifferent[5],(3)thethreerisk factorsaredifficulttobe pre- ciselyevaluated[11],(4)themathematicalformulaforcalculating RPN is questionable [11] and etc. Besides, according to a review from[11],theexistingriskevaluationmethodscanbegroupedinto fivecategories,i.e.,multi-criteriadecisionmakingmethods,math- ematicalprogrammingmethods,artificialintelligencemethods,in- tegratedmethods,andothermethods.
http://dx.doi.org/10.1016/j.neucom.2016.04.073 0925-2312/© 2017 Elsevier B.V. All rights reserved.
Pleasecitethisarticleas:W.L.Changetal.,Application ofself-organizingmaptofailuremodesandeffectsanalysismethodology,Neu- rocomputing(2017),http://dx.doi.org/10.1016/j.neucom.2016.04.073
