Sunteți pe pagina 1din 20

Evaluating the Economic Impact of CMM Measurement Uncertainty

JonM.Baldwin,KimD.SummerhaysandDanielA.Campbell,MetroSageLLC,26896ShakeRidgeRoad, Volcano,CA95689

Abstract
Productmetrologyhasbeenregarded,intheconventionalviewofmanyUSmanufacturingoperations, asanecessarycostofdoingbusiness,theexpenseofwhichshouldbeminimizedintheinterestof greaterprofit.Recentyearshaveseenthebeginningsofashifttoamorecriticalapproachtothistopic. Acarefullookattherisksofincorrectaccept/rejectdecisionsduetomeasurementerrorsandthe associatedcostsofthoseerrorsrevealsthatthetraditionalapproachisoftennottheeconomicallymost advantageous.Thus,moreproductionoperationsarebeginningtocriticallyexaminetheeconomic consequencesofmeasurementand,particularly,thecostofundiscoveredmeasurementerrors. Translationofuncertaintyrelatedrisksintoeconomicoutcomesisanextremelysituationdependant exerciseandisparticularlyinvolvedwhenconsideringtheapplicationofcomplex,multiparameter measuringsystemssuchasCMMs. Inthistalk,wewillfocusonthecomplexitiesinherentintheanalyzingtheeconomicimplicationsof CMMmeasurementsforproductevaluation.Wewilldiscussthetechnicalandeconomicfactorsthat mustbeevaluatedandincludedinaconsiderationofoptimumproductacceptancestrategies,andwill illustratetheirinteractionswithasimplecostmodelthatwillbehelpfultomeasurementspecialistsand productmanagerswhomustmakedecisionsabouttheacquisitionandapplicationofmeasurement resources.

Introduction
To an ever greater degree, a major factor in manufacturing competition is the ability to produce increasinglycomplexcomponents,withbothhigheraccuracyandlowercost,whichalsodrivesproduct quality metrics such as improved function and higher reliability. One aspect of higher accuracy and lowercostinprecisionmanufacturinginvolvesoptimizingthemetrologyprocessonaneconomicbasis. However,thebusinessperspectiveonproductmetrologyoftenhasbeenthatmeasurementsdonotadd valueandthusshouldbetreatedasanexpensetobekepttoaminimum.Thisistrulythecaseonlyifa priorithereis100%certaintythatallproductionoutputiswithinspecification.Inthefarmoretypical case,somefractionofproductisoutofspecification.Identifyingandeliminatingoutofspecification productreducesahostofcostssuchasunnecessaryscrapandrework,warrantyexpense,customer dissatisfaction,damagedbrandreputationandlawsuits.Ideally,allofthesecanbecapturedincost functionsthatexpressthecostofmakingincorrectdecisions.Ifthiscanbedone,thecostofincorrect decisionsandtheireffectonprofitabilitycanbequantified.Itthenbecomestheroleofmeasurement toidentifyandminimizenegativevalueproduct.Viewedinthiscontext,metrologybecomesavalue addedactivityintheproductionprocess.

Thustherearemultiplereasonsforconsideringtheeffectofmeasurementcostonproductprofitability, beginning with the global objective of improving competitive posture, just discussed. The effect of measurementcostbecomesyetmoreintenselyfocusedwhen,asisoftenthecase,measurementsare alsousedtocontrolthemanufacturingprocess.Processcontrolishighlydesirableasitnotonlyreduces the number of measurements (and hence costs), but it proactively adjusts the process and thus can reducethenumberofoutofspecificationcomponents.Costsarenowleveragedupasafewprocess control measurements may affect several hundred manufactured components. Additionally, an erroneousmeasurementsystemnowhastheopportunitytomisadjustthemanufacturingparameters, creatingoutofspecificationcomponents,andthentopassthemontothecustomer.Atalowerlevel, an understanding of measurement cost can be applied to optimize the contribution of measurement resources to the financial bottom line, to justify and predict the benefits of measurement resource expendituresandtofocusavailablemetrologyresourcestobestpromoteprofitability. Inrecentyearstherehavebeensignsofashiftawayfromthetraditionalviewofproductmetrology, andtowardanunderstandingofthevaluemeasurementcanbringtoamanufacturedproduct.Most noticeably,thisconcepthasgainedtractioninsemiconductormanufacturingand,tosomeextent,in chemicalmeasurement.However,ithasbeenourobservation,basedonourconsultingandsoftware activitiesintheareaofCMMmeasurementuncertaintyevaluation,thatinthefieldofdimensional measurement,and3Dmetrologyinparticular,theconventionalperspectiveremainsdominant.While therearesignsofashiftamonglargemanufacturingoperations,certainlyinthearenaofmediumto smallbusinessestherehasbeennogroundswellofchange. Itappearstheremightbeseveralreasonsthatcoordinatemeasurementhaslaggedotherareasof metrologyinthisregard.CMMsarecomplexsystems.Theiroutputissubjecttotheeffectsofmultiple, sometimesinterrelatedinfluences.ThustheuncertaintyofaparticularCMMmeasurementresultis technicallydifficultandpossiblyexpensivetoevaluate.TheaverageCMMpractitionermaybe inadequatelyequippedtoperformthisevaluationanddefenditsresults.Perhapsmostproblematicis thefactthatmeasurementcostanalysisisacombinedtechnical/businessexercise.Metrologistsmay lacksufficientlydetailedinsighttotheeconomicimplicationsoftheirmeasurements,whilebusiness managersmayfindthetechnicalaspectsdaunting. Ourpurposehereistodiscussbothtechnicalandbusinessdecisionmakingcomponentsofthisissueas itappliestocomplexmeasurementsystems,CMMsinparticular,withsomegeneralitybutinsufficient detailtohelpprepareCMMmetrologiststocommunicateconcerningtheirdataanditseconomic impactwithothersinthemanufacturingenterprise.Weareobligedtomakeclearattheoutsetthat,in ordertomakemeaningfuljudgmentsoftheeconomicsofCMMmeasurementuncertainty,itistask specificmeasurementuncertainty1thatwemusthave;moregenericexpressionsofCMMmeasurement
TaskspecificmeasurementuncertaintyistheuncertaintyapplicabletoevaluationofaspecificGD&Tparameter ofaspecificpartfeature,underparticularconditionsofmanufactureandmeasurement;forexample,The uncertaintyofthediameterofthemaximuminscribedcylinderthatwilljustfitinsidethisnominally3inch diameterhole,measuredwiththisparticularCMM,underthesespecificconditions,is0.0008inchesat95% confidence.
1

uncertaintywillnotdo.Inwhatfollows,thediscussionwillcoverinturntwodistinctphasesoftheCMM costanalysisprocess:riskanalysis,whichisapurelytechnicalexercise,andprofitabilitycalculation, whichinmanyofitsaspectsisabusinessexercise.

RecentTechnicalActivities
TherehavebeenrecentdevelopmentsintwoareasrelevanttoCMMmeasurementcostanalysis.One oftheseistheavailabilityofcommercialsoftwarepackagesfordevelopmentoftaskspecificuncertainty statementsforCMMmeasurements.Onesuchproductwasdescribedinanearlierpaperinthissession [1].Theotheristheappearanceofanumberofnationalandinternationalstandardsaddressingthe someofthetechnicalaspectsofmeasurementuncertaintyevaluation[214].Whilethesepublications dealwithmanyofthetechnicalaspectsofmeasurementcostanalysisandtwoofthem[7,8]touchon theroleofproductacceptancedecisionrulesasbusinessissues,therehasbeennodetaileddiscussion oftheeconomicimpactsofCMMmeasurementuncertainty.

CMMRiskAnalysis
Thegeneraltopicofriskanalysisinmeasurementhasbeencoveredin[10];thereaderisreferredthere forbasicinformation.Here,wewilldescribethegeneralprocessandwillgiveamoredetailed descriptionofthepropertiesofCMMmeasurementsthatmaketheirtreatmentsomewhatunique.As theseuniqueaspectsarementioned,suggestedapproachesfordealingwiththemwillbeprovided. Amanufactureditemthatissubjecttosomespecificationmaybeinoneoftwoconditions:Itmay conformtothespecification( C )oritmayfailtoconform( C ).Iftheitemissubjectedtomeasurement therearetwopossibleoutcomes:itmaypass( P )orfail( F )theinspection.Thus,therearefour possibleoutcomesofaproductinspection:Theitemmayconformtospecificationandpassinspection ( PC )oritmaynotconformtospecificationandfailinspection( FC ),bothofwhicharedesired outcomes,resultinginacceptanceofagoodpartorrejectionofabadpart.Inthepresenceof measurementerrortwootherconditionsmayberealized:Thepartmay,infact,notconformto specificationbutpassinspection( PC ).Thisisanundesirableoutcome,resultinginacceptanceofan outoftoleranceitem.Alternatively,apartthatconformstospecificationmayfailtheinspection( FC ). This,too,isundesirablesinceitleadstorejectionofagooditem.Theprobabilitiesoftheselasttwo eventsarecalled,respectively,theconsumersandproducersrisksinreferencetothepartywho usuallybearsthecostoftheerror.Therelationshipsoftheserisksareoftenrepresentedina contingencytable(Figure1).

Calculationoftheconform/nonconformprobabilitiesrequiresknowledgeoftheproductionprocess probabilitydensity.Theconformanceprobabilityisgivenby
Figure1.ContingencyTableforProductMeasurement

p (C | I ) = p( x | I )dx, p (C | I ) = 1 p(C | I )
C

where,forexample, p (C | I ) denotestheprobabilityofconformancegivenanypriorinformation, I ,

p( x | I ) istheprocessprobabilitydensityandtheintegralisovertheacceptanceregion.
Thentheconsumersriskis

RC =

xR

p ( PCx | I ) dx = p ( PC | xI )i p ( x | I ) dx
0 0 0 xR

wheretherangeofintegrationisoverallvaluesofxthatareoutsidetheacceptancezone,

p ( PC | xI 0 ) istheprobabilitythatacharacteristicknowntobenonconformingneverthelessproduces

ameasurementresultwithintheacceptancezone,seeFigure2,and p ( x | I 0 ) isthepriorprobability densityforthemeasurand.Similarly,theproducersriskisgivenby

RP =

xC

p ( FC | xI )i p ( x | I )dx
0 0

wherenowtheintegrationisovertheacceptancezone.Figure2isanillustrationoftheconsumersand producersrisksforthecaseofaGaussianerrordistribution.

Figure2.Consumer'sandProducer'sRisks

AssumptionofGaussianformfortheproductionandmeasurementprobabilitydensitiesisacommon practice,andofferstheadvantageofcomputationaltractability;thisistreatedindetailin[10].Such assumptionscanproverashinthecaseofCMMmetrologyandcanleadtosubstantialerrors.Infact, normaldistributionsoftenturnouttobeasmuchtheexceptionastherule.

ProductionProbabilityDensity
Therearetwoissuesinapplyingprobabilitydensitiesinmeasurementriskanalysis.Thefirstisto determinetheinherentprocessvariability.Thisistypicallydonebymeasuringalargesampleofthe productandplottingahistogramoftheresults.Anynonrepeatabilityofthemeasurementsystemwill besuperimposedontheproductionvariation.Ideallythecontributionofmeasurementvariabilitywill bemadenegligiblebyusingameasurementsystemwithsufficientlysmallrepeatability.Alternatively, measurementrepeatabilitycanbedeterminedbyrepeatedlymeasuringastableartifactandlookingat thedistributionoftheresults.Iftherepeatabilityofthemeasurementsystemsodeterminedis characterizedbyastandarddeviation, m ,andthetotalstandarddeviationofameasurementis T ,
2 2 theprocessstandarddeviationis p = T m .Itis,ofcourse,prudenttoperiodicallyverifythat

theprocessremainsstable. Thesecondissueischoiceofanappropriateandconvenientrepresentationoftheprobabilitydensity. Productiondensitiesfortwosidedmeasurements,e.g.size,lengthorangle,canoftenbereasonably

representedasGaussians.Onesidedmeasurements,e.g.form,positionororientation,represent instancesinwhichvaluesnearzeroarenotfrequentlyobservedandtheGaussiandistributionis thereforeinappropriate,sinceitwillassignfiniteprobabilitiestovaluesthatarephysicallyimpossible.It isconceivablethatdistributionsofinherentlyonesidedquantitiescouldbemodeledbyaGaussian distributiontruncatedatzero,butitismorepracticaltouseadistributionmodelthatnaturallygoesto zero.Thebetadistributionisonesuchthatworkswell;itistreatedinAppendixIIof[10]andillustrated inFigure3.

Figure3.GammaProbabilityDensityforaOnesidedMeasurement

MeasurementProbabilityDensity
WhileCMMssharemanysimilaritieswithotherdimensionalmeasurementtechnologies,similar complexitiesarisewithregardtothemeasurementprobabilitydensity,dueprimarilytothemultitudeof variablesthatcaninfluencethemeasurementresultandthecomplexitiesoftheirinteractions[1]. WhileapproximatelyGaussianmeasurementdensitiesaresometimesobserved,thevarietyof distributiontypesistypicallygreatanddifficulttopredictfromintuitionoranalysis.Afewobserved measurementdistributionsarepresentedasexamplesinfigures47,whichwerechosenfortheir relativesimplicityandrepresentcaseswhereasingleerrorsourcewasknowntopredominate.


Figure4.MeasurementDistributionforaLengthmeasurement

Thisisanobservedmeasurementdistributionforalengthmeasurementmadeunderconditionswhere thegeometricerrorsoftheCMMwereknowntopredominate.Inthisinstance,aGaussiandistribution mightbeareasonableapproximation.

Figure5.MeasurementDistributionwithTemperatureVariation

Here.InFigure5,themajorinfluencewastemperaturevariation,usingacontrollerthatkeptthe temperaturebetweentwofixedlimits.Thisdistributionmightbereasonablywellmodeledasuniform.


Figure6.DiameterMeasurementwith3lobeFormError

Figure6showsthesizemeasurementdistributionproducedbyinteractionoftheprobingpatternwitha 3lobeformerroronacylindricalhole.ItexhibitstheclassicalUshapeexpectedfromthissource. Finally,Figure7showsthemeasurementdensityforaflatnessmeasurement,madeunderthesame conditionsasthelengthmeasurementofFigure4.

Figure7.FlatnessMeasurementwithpredominantlyRandomErrors

Evidently,properrepresentationofthemeasurementprobabilitydensityisarathermorecomplex problemthanmodelingtheproductiondensity.Assuggestedintheprevioussection,measurement variabilitycanbedeterminedbyrepeatedmeasurementofastableartifact.Alternatively, measurementvariabilitycanbeestimatedbysimulationtechniquessuchasthosedescribedin[1].The effortrequiredisconsiderablylessandthisis,infact,howthedistributionsinFigures47were determined.

Giventhevarietyofshapesthatcanbeobservedformeasurementdistributions,analysisrequiresa versatilerepresentation.Whitehousehassuggestedthebetadistributionprovidesasuitably comprehensivecapability[15].Thebetaprobabilitydensityisgivenby

f ( a , b, y ) =

1 ( b 1) a 1 y ( ) (1 y ) ( a, b )

where

( a, b ) = y ( a 1) (1 y )
0

( b 1)

dy

and a and b areadjustableparameters.Theabilityofthisdistributiontorepresentmanyofthe observedmeasurementdistributionsisillustratedinFigure8. Anotherworkablealternativeistousethehistogramdatadirectly,performinganumericalintegration overtheappropriateinterval.

Figure8.ExamplesoftheBetaProbabilityDistribution

Insummary,theconsumersandproducersrisksdependontheproductionandmeasurement probabilitydensities,andtheacceptancelimits.Knowingandunderstandingtheserisksprovide

solutionstothetechnicalquestionsabouttheconsequencesofbaddecisions.Theyareofvalueinand ofthemselvesifthetechnicalrequirementsofanacceptanceprocessareknownanditisnecessaryto determineifaspecificCMMisuptothemeasurementchoreathand.Theyare,aswell,aprerequisite todealingwithbusinessconcernsaboutproductprofitability.

CMMCostAnalysis
Inordertounderstandtheeconomicimplicationsofmeasurementdecisionsandtooptimizethisfacet ofthemanufacturingoperationitisrequiredtoreducethetechnicalrisksofmeasurementdecisionsto theirimpactonproductprofitability.TypeIandTypeIImeasurementerrorswillbothincurcosts. Rejectionofapartthatis,infact,good(TypeIerror)resultsinlossoftheentireamountinvestedin productionofthatitem.Thesecostsshouldbereadilyaccessibleandwellknown.Theywillinclude,for example,thecostsofmanufactureandinspectionaswellasofmaterials,scrapdisposaland,perhaps, unnecessaryrework.Someofthecostsarisingfromacceptanceofapartthatis,infact,bad(TypeII error)areequallyaccessible.Examplesmightbewarrantyandfailureanalysiscosts.Othercost consequencesofaTypeIIerrormaybemoredifficulttoquantifybutcaneasilyexceedthereadily documentedcosts.Examplesincludedamagedcustomerperceptionofproductquality,lossoffuture salesandcostoflawsuits.

LossFunctions
Inordertoquantifythecostofincorrectdecisionsaboutproductconformance,itisnecessaryto associatedeviationofthequalitycharacteristicfromtheidealwithaneconomicloss.Thisassociationis generallyexpressedasalossfunction.Realisticassessmentofproductioneconomicsanddevelopment ofoptimummeasurementstrategieswilldependcriticallyonchoiceofanappropriatelossfunction. SomepossiblechoicesareshowninFigure9forthecaseofabilateraltolerancewhere,generally,the nominalvalueofthequalitycharacteristicisoptimum.


Figure9.QualityLossFunctions;BilateralToleranceor"NominaltheBest"

Theclassicallossfunctionhasoftenbeenastepfunction,whichstatesthatcostsdonotdependonthe actualvalueofthequalitycharacteristicaslongasitiswithinspecification.Althoughthestepfunction hasbeenwidelyusedinthepastandismathematicallyconvenient,itisnowgenerallyacknowledged thatthequadraticlossfunction,

LT ( x ) =

k II 2 x T ) 2 (

originallyproposedbyTaguchi[16],isabetterexpressionofthecostconsequencesoflesstangible factorssuchasthecustomersqualityperception.Here, k II isthecostofaTypeIIerror, isthe tolerancewidthand T isthenominalortargetvalue.Inthismodel,anydeviationfromthenominal valueincursacostpenalty.TheTaguchilossfunctionisalsorecognizedtoembodysomedegreeof unrealism,inthatincreaseswithoutlimit.Morerealistically,therewillgenerallybeadegreeof nonconformancebeyondwhichthefulllosswillhavebeenrealized.Thisrealitycouldbe accommodatedbychoosingadeviationatwhichthelossfunctionundergoesatransitionfroma quadraticexpressiontoaconstantvalueofmaximumloss.This,too,presentssomedifficultyinthatthe suddentransitiontoconstantlossseemsunrealisticandhasledtoproposalofavarietyoflossfunctions basedoninvertedprobabilitydistributions.Theearliestoftheseistheinvertednormalloss,originally suggestedbySpiring[17]andhavingtheform

( x T )2 LS ( x ) = K 1 exp 2 2

where K isthemaximumlossduetoaTypeIIerror, T isthetargetvalueand isashapeparameter suchthatif = / 4 thelossat T willbeveryclose(within0.9997of) K . OneoftheinterestingandproblematicfeaturesofCMMevaluationsofGD&Tparametersisthat, contrarytothemoregeneralqualityenvironment,wherethemajorityofqualitycharacteristicsareof thebilateraltoleranceornominalthebesttype,manyGD&Tparametersarelessthebetteror singlesidedvariety.Analogouslossfunctionscanbegeneratedforthesinglesidedtolerancecase, Figure10,ascanmorecomplicatedfunctionstodealwithinstanceswheredeviationsinonedirection shouldbepenalizeddifferentlythanthoseintheother,seeforexampleFigure11.Unfortunatelyfor theCMMmetrologists,theliteratureonthiscategoryoflossfunctionsisrelativelysparse.

Figure10.LossFunctionsforaUnilateralTolerance

Figure11.NonsymmetricLossFunctionExample

Choiceoftheappropriatelossfunctionforaparticularproductisgenerallybeyondtheexpertiseand knowledgebaseoftheCMMmetrologist;heretheneedforcollaborationafinancialanalystfamiliar withthecustomerbaseandsalesenvironmentisindicated.Themeasurementspecialistwillalmost certainlyfindfinancialandsalespersonnellackinginunderstandingofthebusinessofmeasurement.A mutualeducationalexchangewillalmostcertainlyberequired.

ProfitCalculation
Methodsofmeasurementcostcomputationcanvaryconsiderablyincomplexityandsophistication.At oneextreme,elaborateandhighlydetailedmodelshavebeendevelopedforanalyzingthecostto benefittradeoffsofsemiconductormeasurement;forexample,see[18].Ontheotherhand, significantlysimplermodelsareusefulinmostinstances,especiallyforapplicationtoestablished metrologyfacilitieswherethereisalonghistoryofmeasurementcostdata.Suchasimplemodelis usedintheexampleprofitcalculationsthatfollow.Specifically,weusehereamodelderivedfromone originallyproposedbyWilliamsandHawkins[19,20]todealwithGaussianerrordistributionsandthe Taguchilossfunction,andexpandedbyustocovertherangeoferrordistributionsandlossfunctions describedabove. Intheiroriginaltreatment,WilliamsandHawkinsdevelopedanexpressionfortheperunitprofitunder conditionsofmeasurementerror:

k 2 UnitProfit = k s II 2 + (T ) P ( F ) k I 2

where k S isthesellingpriceperunit, k I isthecostofmanufacturingandtestingoneproductunit(also equaltothecostofaTypeIerror), k II istheestimatedcostofaTypeIIerror, T isthenominalortarget value, isthetolerancebandwidth, and arethemeanandstandarddeviation,respectively,of thequalitycharacteristicasseenbythecustomer(thatis,afteracceptance)and P ( F ) istheprobability ofaunitpassingitsinspectiontest.Thisexpressionincludestheeffectsofsaleprice,TypeIandTypeII errors,andcustomerperceptionofquality.Analogousexpressionshavebeendevelopedforotherloss functionsanddistributions.

Multipleparameters
Acomplicationuniquetocoordinatemeasurement,ascomparedtomostotherdimensional measurementtechnologies,arisesfromthefactthatitisusualformultiplecharacteristicstobe evaluatedonthesamemeasurementapparatusinasingleinspectionoperation.Itismostusefulto endupwithasinglelossparameterthatexpressestheneteconomiceffectofallproductnonidealities, whichleadstotheconsiderationofhowtomostappropriatelycombinethelossesarisingfromthe variousdimensionalrequirements.Ithasbeensuggestedthatthelossfunctionsfortheindividual parameterscanbeaddedtoformanetmultivariatelossfunction[21],butthisseemsintuitively extremeforacasewherethecasewhereacommonmeasurementapparatus,acommonenvironment andinsomecasesagainstcommondatumsandcommonproductionmachinerymakelikelysome degreeofcorrelationamongthevariousGD&Tparameterevaluations.Forthisreason,wepreferthe

recommendationofWilliamsandHawkins[20]thatthelossfunctionsbecombinedastheirrootsumof squares,

LT =

( E ( L ))
i i

whichtendstolessentheseverityofthepenaltyformeasurementsysteminducedcorrelation.Finally, itshouldberecognizedthatthedevelopmentofmultivariatelossfunctionsforcorrelatedcharacteristics remainsanactivefieldofresearchandthatnewdevelopmentsaretobeexpected.

Examples
Inthissectionwepresentafewspecificexamplesoftheeffectsofvariousmeasurementvariableson productprofitability,asillustrationsofthevalueofperforminglosscalculations.

MeasurementUncertaintyandGuardBands
Itiscommonpracticetoguardbandmeasurementstoreducetheincidenceandcostofmeasurement errors.Mostcommonly,itisthecostofTypeIIerrorsthatisofmajorconcern;hencetheacceptance zoneisreducedbysomeamount(knownasstringentacceptance)andanincreasedprobabilityof rejectingagoodpartisacceptedintheinterestofnotexposingthecustomertobadparts.Itisusefulto know,foragivenmeasurementuncertainty,theguardbandchoicethatwillmaximizeprofit.Inthis example,weareconsideringthemeasurementofa100mmdiametershaftwithtolerancelimitsof1 mm.Theproductionprocessiscentered;thatisitproducespartswithameansizeof100mm.The productionstandarddeviationis0.33mm.Themeasurementprocessisunbiased.Thesellingpriceof onepartis$30,thecostofproducingapartis$7.50,andthecostofshippingabadpartistakentobe $300.Thisistypicalofwhatmightberegardedasacriticalorhighconsequencepart,wheretheratio ofTypeIIerrorcosttosellingpriceisoftenfoundtobe 10. Guardbandingisrequiredinordertoachieveprofitability,withtheoptimumguardbandmultiplierbeing ontheorderof0.65.Thereisasignificanteffectofmeasurementuncertainty,withtheprofit approachingzeroatthehighendoftherangestudied.


Figure12.EffectofMeasurementUncertaintyandGuardbandingonProfit

MeasurementBias
Measurementbiasisproperlyconsideredtobeafactortobediscoveredandeliminated,howeverits completeremovalisnotalwayspossibleandtheeffectofbiasonprofitabilitycanbesignificant.Hereis anexample,usingthesamegeneralconditionsasthepreviousonethatillustratesthepotential magnitudeofbiasdrivenlossofprofit.Themeasurementstandarddeviationwastakenheretobe0.1 mmandaguardbandclosetotheoptimumindicatedintheearlierexamplewasused.Anundiscovered oruncorrectedmeasurementbiasofthesamemagnitudeasthemeasurementuncertaintycancausea profitdecrementofalmost7%.


Figure13.EffectofMeasurementBias

MeasurementStrategy
AsregardsmeasurementpointplacementstrategywithCMMs,thegeneralwisdomismoreisbetter. Thisexampleillustratesthatthisisnotguaranteedtobethecase.Thefeatureofinterestis,again, cylindrical.Figure14showstheeffectonperunitprofitassamplingdensityisincreased,assumingthe featureformerrorisrandomoverthefeaturesurface.Thesamplingpatternusedconsistedoftaking halfthetotalnumberofpointsoneachoftwocircularsectionsneareachendofthefeature,anot unreasonablestrategyiftheformerrorisknowntoberandom.TheupperplotofFigure14showsthat theuncertaintyofthediametermeasurementdecreasessteadilywithincreasingpointdensityas conventionalwisdomsuggests. Figure15illustratestheresultsofthesamesetofmeasurementstrategieswhentheformerrorhasa3 lobeshapewiththesameamplitudeasthemaximumrandomexcursion.Theuncertaintythenadopts anoscillatorybehaviorthatisreflected,althoughnottoanextremedegreeoncethesamplingdensity risesto7pointsperlevel,intheperunitprofit.


Figure14.EffectofSamplingStrategyonProfit,RandomFormError

Figure15.EffectofSamplingStrategyonProfit,3lobeFormError

AvailableTools
TwocommerciallyavailablesoftwarepackagesforCMMmeasurementuncertaintyevaluationare knowntotheauthorsofthisreport;onewasdescribedinanearlierpaperinthissession[1].Itwas usedtosupplythemeasurementuncertaintyevaluationsusedinthispaper.Theotherisavailableasan addontotheCMMcontrolanddataanalysissoftwareprovidedbysomespecificCMMvendors.The nextreleaseofthefirstmentionedofthesepackageswillmakeavailableatoolkittocomputeproduct profitabilityandoptimumguardbandselections. Someoftheprofitabilitycalculationspresentedinthispapercanbeperformedwithspreadsheet programs.Mostofthem,however,requirenumericalintegrationofprobabilitydensitieswhichare difficulttoperforminspreadsheetapplications.Commercialsoftwareintendedfornumerical computation(forexampleMATLAB2orMathematica3)ismuchbettersuitedforthecomputationof measurementcostsandprofitability.ApackageofMATLABfunctionsdevelopedbyuswasusedto preparethedataforthispaper.

References
[1]D.A.Campbell,J.M.Baldwin,K.D.Summerhays,ApplicationsofComputerSimulationtoCMM UncertaintyEvaluation,MeasurementScienceConference,Pasadena,CA,March2010. [2]DraftTechnicalReport155301,GeometricalProductSpecification(GPS)CoordinateMeasuring Machines(CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart1:Overviewand GeneralIssues,InternationalOrganizationforStandardization,Geneva. [3]DraftTechnicalReport155302,GeometricalProductSpecification(GPS)CoordinateMeasuring Machines(CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart2:Expert Judgment,InternationalOrganizationforStandardization,Geneva. [3]TechnicalReport155303,GeometricalProductSpecification(GPS)CoordinateMeasuringMachines (CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart3:ExperimentalUncertainty AssessmentfoetheSubstitutionMethodUsingCalibratedObjects,InternationalOrganizationfor Standardization,Geneva,2004. [4]TechnicalReport155304,EstimatingGeometricalProductSpecification(GPS)CoordinateMeasuring Machines(CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart4:Uncertainty AssessmentusingComputerSimulation,InternationalOrganizationforStandardization,Geneva,2008.

MATLABisaregisteredtrademarkofTheMathWorks. MathematicaisatrademarkofWolframResearch,Inc.

[5]TechnicalReport155305,EstimatingGeometricalProductSpecification(GPS)CoordinateMeasuring Machines(CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart5:Statistical EstimationfromMeasurementHistory,InternationalOrganizationforStandardization,Geneva. [6]DraftTechnicalReport155306,EstimatingGeometricalProductSpecification(GPS)Coordinate MeasuringMachines(CMM):TechniquesforDeterminingtheUncertaintyofMeasurementPart6: ExperimentalUncertaintyAssessmentusingUncalibratedObjectsinCombinationwithAnalysisofBias, InternationalOrganizationforStandardization,Geneva.

[7]142531,GeometricalProductSpecifications(GPS)InspectionbyMeasurementof WorkpiecesandMeasuringEquipmentPart1:DecisionRulesforProvingConformanceor NonconformancewithSpecifications,InternationalOrganizationforStandardization,Geneva,1998.


[8]B89.7.3.12001,GuidelinesforDecisionRules:ConsideringMeasurementUncertaintyDetermining ConformancetoSpecifications,ASME,NewYork,2001. [9]B89.7.3.32002GuidelinesforAssessingtheReliabilityofDimensionalMeasurementUncertainty Statements,ASME,NewYork,2002. [10]TechnicalReportB89.7.4.12005MeasurementUncertaintyandConformanceTesting:Risk Analysis,ASME,NewYork,2005. [11]TechnicalReportB89.7.52006MetrologicalTraceabilityofDimensionalMeasurementstotheSI UnitofLength,ASME,NewYork,2006. [12]TechnicalReportB89.7.3.22007GuidelinesfortheEvaluationofDimensionalMeasurement Uncertainty,ASME,NewYork,2007. [13]JCGM100:2008,EvaluationofmeasurementdataGuidetotheexpressionofuncertaintyin measurement,BureauInternationaledesPoidsetMeasures,Svres,France,2008. [14]JCGM101:2008,EvaluationofMeasurementDataSupplement1totheGuidetotheExpression ofUncertaintyinMeasurementPropagationofDistributionsusingaMonteCarloMethod,Bureau InternationaledesPoidsetMeasures,Svres,France,2008. [15]D.J.Whitehouse,BetaFunctionsforSurfaceTypology,AnnalsoftheCIRP,42(1),4917(1978). [16]G.Taguchi,E.A.Elsayed,T.C.Hsiang,QualityEngineeringinProductionSystems,McGrawHill,New York,1989. [17]F.A.Spiring,TheReflectedNormalLossFunction,CanadianJournalofStatistics,21(3),32130 (1993). [18]S.YSohn,H.U.Moon,CostofOwnershipModelforInspectionofMultipleQualityAttributes,IEEE TransactionsonSemiconductorManufacturing,16(3),45671(2003).

[19]R.H.Williams,C.F.Hawkins,TheEconomicsofGuardbandPlacement,1993InternationalTest Conference,21825. [20]R.H.Williams,C.F.Hawkins,EconomicConsiderationsinToleranceDesign,1994InternationalTest Conference,739. [21]Y.Ma,F.Zhao,AnImprovedMultivariateLossFunctionApproachtoOptimization,Journalof SystemsScienceandSystemsEngineering,13(3),31825(2004).

S-ar putea să vă placă și