Reliability Growth

ObservingandPredictingTrendsinReliability

Today’scomplexproductsrequirethatcompaniesfocusonestablishingdetailedplansandproceduresfordevelopingandmanufacturingproductsthatmeetspecifiedreliability,maintainability,andotherperformancerequirements.Reliabilitygrowthreferstoawell-definedprocessforidentifyingandcorrectingreliabilityproblemsearlyinthedesignprocesssothatthereliabilityofaproductincreasesor”grows”astheproductgoesthroughsuccessivedevelopmentstages.Reliabilitygrowthprogramsshouldbeestablishedforallnewproductsandforallexistingproductsundergoingmajorredesignssothattheimprovementduetochangesindesignandmanufacturingprocessescanbeeasilytracked.
Duringtheearlydesignstage,areliabilitygoalissetforaproduct.Becausefailuredatafromprototypetestingisnotyetavailable,theinitialreliabilitygoalisoftenbasedoneitherthefailuredataforsimilarproductsorthefailuredataforthesubcomponentsoftheproduct.Duringthedevelopmentstage,productprototypesgenerallyundergoextensivetestingsothatdeficienciesinrespecttodesign,engineering,andmanufacturingcanbeidentifiedandcorrected.
Atypicalreliabilityimprovementtestconsistsofoperatingproductprototypesforseveralweeksinthesametypesofenvironmentsinwhichcustomerswilleventuallyoperatetheproduct.Ateamofprojectengineersandtechniciansanalyzeeveryfailurethatoccurs,determiningrootcausesforfailuresanddevelopingdesignandmanufacturingimprovementsthataretoeithereliminateorreducetherecurrencesofthesefailures.Asthetestingcontinues,theimprovementsdevelopedbytheteamareincorporatedintotheprototypesothatproductreliabilitycontinuestoimprovethroughoutthetestingperiod.

AchievingReliabilityGrowth
Thethreemostimportantstepsintheiterativeprocessforachievingreliabilitygrowtharedepictedinthefeedbackloopthatfollows:

Mostoftheproblemsencounteredduringtestingarelikelytobecomponentfailuresandmanufacturingdeficienciesthatcouldnotbeforeseenintheearlydesignphase.Becausevariousperformancerequirementscanconflict,optimizingadesigntomeetonerequirementcancausethedesigntofailtomeetanotherrequirement.Thus,iterationsofdesignsareoftenneededtocorrectallofthecomponentselectionandmanufacturingdeficienciesthatarefoundduringprototypetesting.
Inadditiontoimprovingreliability,earlyimplementationofareliabilitygrowthprogramminimizestheimpactonproductionschedulingandtotalproductcost,especiallysincethecostsassociatedwitheitherredesigningaproductlateinthedevelopmentcycleorretrofittingproductsalreadyinthefieldareextremelyhigh.Developingaproductthatmeetsreliabilityrequirementsalsoensuresthattheproductultimatelymeetsuserneedsandhasanacceptabletotallifecyclecost.Settinginterimreliabilitygoalsthataretobeachievedduringtestingensuresthatresourcesareallocatedefficiently.

ReliabilityGrowthData
Thebasicprincipleofanyreliabilitygrowthmodelistoapplythetestingresultsanddatapointstodetermineifthereliabilityoftheproductisgrowingsufficientlytomeetthereliabilityrequirementsfortheproduct.Thetypesofdatausedforpredictingreliabilitygrowthare:
•ReliabilityData.Thereliabilityoftheproductisrecordedatdifferentpointsintime.Thereliabilityisaratioofthenumberofunitsstillfunctioningandthenumberofunitsthatenteredthestage.
•Success/FailureData.Theitemistestedandcaneithersucceedorfail.Successdatacanconsistofacodeindicatingtheoutcome,suchasSforSuccessorFforFailure;or,itcanconsistofacodeindicatingthefailuremodeforeachfailure.Success/failuredatamightalsoindicatethenumberoffailuresthatoccurredwhenseveralunitsaretestedtogether.
•FailureTimeData.Thetimetofailureforanitemistrackedusingeithercumulativeornon-cumulativeoperatingtimes.Failuretimedataisthemostcommonlyusedtypeofdatainreliabilitygrowth.Thisdatacanbeforonefailureorsystem,oritcanbeformultiplefailuresormultiplesystems,whichiscommonlycalledintervalorgroupeddata.
Thefailuredatacollectedduringprototypetestingisusedtodeterminewhetherthereliabilitygoalfortheproductislikelytobemetorexceededbythetimetheproductisscheduledtobeputintofull-scaleproduction.Althoughmanymethodsexistformodelingthereliabilitygrowthprocess,theDuane/Crow-AMSAAmodelisconsideredthebestpractice.In1964,J.T.Duane,anengineerattheAerospaceElectronicsDepartmentofGeneralElectronicCompany,publishedapaperdemonstratinghowalearningcurveapproachcouldbeusedtomonitorthecontinuingreliabilityimprovementsintheearlystagesofdevelopingcomplexelectromechanicalandmechanicalsystems.Duaneexplainedhowthiswasbecausethelessonslearnedfromfailureswereusedtorefinedesigns.
AccordingtoDuane,agraphofthecumulativeMTBFversusthecumulativeoperatingtimeplottedonlog-logpaperfellclosetoastraightline.Graphingthislearningcurveprovidesameansofmeasuringandpredictingreliabilityduringaperiodofproductchange.Dr.LarryCrowlateraddedpowerfulstatisticalcapabilitiestoDuane’spostulateforlearningcurvemodelingtocreatethemodelsowidelyrespectedandimplementedtoday.AnexampleofareliabilitygrowthchartgeneratedbytheDuane/Crow-AMSAAmodelfollows:

Reliabilitygrowthisgenerallyquantifiedbygraphinganyofthefollowingthreemeasuresovertime:
•Theincreaseinthemissionsuccessprobability(reliability).
•TheincreaseinMTBF.
•Thedecreaseinfailurerateasafunctionoftime.
Reliabilitygrowthchartsdepicttrendsthatareusedtoforecastfailuresasafunctionofadditionaltesttimeorcalendartime,therebymakingplanningforredesignandtestresourceseasier.Inadditiontoallowingyoutodeterminewhetherreliabilityrequirementswillbeachieved,reliabilitygrowthchartscanhelpyoutodeterminethetimeneededtomeettheserequirementsandtheassociatedcosts.Extrapolatingagrowthcurvebeyondthecurrentlyavailabledatashowswhatreliabilityaprogramcanbeexpectedtoachieveprovidingthattheconditionsofthetestandtheengineeringefforttoimprovereliabilityaremaintainedattheirpresentlevels.
Ifthereliabilitygrowthgraphindicatesthatthereliabilitygoalisnotgoingtobemetorexceeded,thentheproductdesignmustbeimproved.Thismightrequiretheuseofmorereliablecomponentsorredundancy.Additionalresourcesmightalsohavetobedevotedtodesigningandmanufacturingamorereliableproducttomeettherequireddeliverydate.

OtherUsesforReliabilityGrowthCharts

Inadditiontoanalyzingtheresultsfromtestingnewlydevelopedorredesignedproducts,theDuane/Crow-AMSAAmodelcanbeusedto:
•Predictfuturefailures.
•Trackfleetsofrepairablesystems.
•Provideaccuratetrendingofsignificanteventsformanagement.
•Forecastsafetyincidentsthatcanbecontrolled.