附錄1翻譯原文及譯文DocNo:P0193-GP-01-1DocName:AnalysisofManufacturingProcessDataUsingQUICKTechnologyTMIssue：1Data：20April，2006Name(Print)SignatureAuthor：D.CliftonReviewer：S.TurnerTableofContents1ExecutiveSummary.41.1Introdution.41.2TechniquesEmployed.41.3SummaryofResults.41.4Observations.52Introdution.62.1OxfordBioSignalsLimited.63ExternalReferences.74Glossary.75DataDescription.75.1Datatypes.75.2PriorExperimentKnowledge.75.3TestDescription.86Pre-processing.96.1RemovalofStart/StopTransients.96.2RemovalofPowerSupplySignal.96.3FrequencyTransformation.97AnalysisI-Visualisation.127.1VisualisationofHigh-DimensionalData.127.2Visualising5-DManufacturingProcessData.錯(cuò)誤!未定義書簽。7.3AutomaticNoveltyDetection.錯(cuò)誤!未定義書簽。7.4ConclusionofAnalysisI-Visualisation.錯(cuò)誤!未定義書簽。8AnalysisII-SignatureAnalysis.錯(cuò)誤!未定義書簽。8.1ConstructingSignatures.錯(cuò)誤!未定義書簽。8.2VisualisingSignatures.錯(cuò)誤!未定義書簽。8.3ConclusionofAnalysisII-SignatureAnalysis.錯(cuò)誤!未定義書簽。9AnalysisIII-TemplateAnalysis.錯(cuò)誤!未定義書簽。9.1ConstructingaTemplateofNormality.錯(cuò)誤!未定義書簽。9.2ResultsofNoveltyDetectionUsingTemplateAnalysis.錯(cuò)誤!未定義書簽。9.3ConclusionofAnalysisIII-TemplateAnalysis.錯(cuò)誤!未定義書簽。10AnalysisIV-None-linearPrediction.錯(cuò)誤!未定義書簽。10.1NeuralNetworksforOn-LinePrediction.錯(cuò)誤!未定義書簽。10.2ResultsofNoveltyDetectionusingNon-linearPrediction.錯(cuò)誤!未定義書簽。10.3ConclusionofAnalysisIV-Non-linearPrediction.錯(cuò)誤!未定義書簽。11OverallConclusion.錯(cuò)誤!未定義書簽。11.1Methodology.錯(cuò)誤!未定義書簽。11.2SummaryofTesults.錯(cuò)誤!未定義書簽。11.3FutureWork.錯(cuò)誤!未定義書簽。12AppendixA-NeuroScaleVisualisations.錯(cuò)誤!未定義書簽。TableofFiguresFigure1-Test90.Fromtoptobottom:Ax,Ay,Az,AE,SPagainsttimet(s)Figure2-PowerspectraforTest19afterremovalof50Hzpowersupplycontribution.Thetopplotshowsa3-D“l(fā)andspace”plotofeachspectrum.Thebottomplotshowsa“contour”plotofthesameinformation,withincreasingsignalpowershownasincreasingcolourfromblacktoredFigure3-PowerspectraforTest19afterremovalofallspectralcomponentsbeneathpowerthresholdFigure4-Azagainsttime(inseconds)forTest19,beforeremovaloflow-powerfrequencycomponentsFigure5-Azagainsttime(inseconds)forTest19,afterremovaloflow-powerfrequencycomponentsFigure6-SPforanexampletest,showingthreeautomatically-detecrminedstates:S1-drillingin(showningreen)；S2-drill-bitbreak-throughandremoval(showninred)；S3-retraction(showninblue)Figure7-Examplesignatureofvariableyplottedagainstoperating-pointFigure8-Powerspectrafortest51,frequency(Hz)onthex-axisbetween0fs/2Figure9-AveragesignificantfrequencyfuFigure10-VisualisationofAEsignaturesforalltestsFigure11-VisualisationofAxbroadbandsignaturesforalltestsFigure12-VisualisationofAxaverage-frequencysignaturesforalltestsFigure13-NoveltydetectionusingatemplatesignatureFigure14-1ExecutiveSummary1.1IntroductionThepurposeofthisinvestigationconductedbyOxfordBioSignalswastoexamineanddeterminethesuitabilityofitstechniquesinanalyzingdatafromanexamplemanufacturingprocess.ThisreporthasbeensubmittedtoRolls-RoycefortheexpressedofassessingOxfordBioSignalstechniqueswithrespecttomonitoringtheexampleprocess.TheanalysisconductedbyOxfordBioSignals(OBS)waslimitedtoafixedtimescale,afixedsetofchallengedataforasingleprocess(asprovidedbyRolls-RoyceandAachenuniversityofTechnology),withnopriordomainknowledge,norinformationofsystemfailure.1.2TechniquesEmployedOBSusedanumberofanalysistechniquesgiventhelimitedtimescales:I-Visualisation,andClusterAnalysisThispowerfulmethodallowedtheevolutionofthesystemstate(fusingallavailabledatatypes)tobevisualisedthroughouttheseriesoftests.Thisshowedseveraldistinctmodesofoperationduringtheseries,highlightingmajoreventsobservedwithinthedata,latercorrelatedwithactualchangestothesystemsoperationbydomainexperts.Clusteranalysisautomaticallydetectswhichoftheseeventsmaybeconsideredtobe“abnormal”,withrespecttopreviouslyobservedsystembehavior.II-Signaturerepresentseachtestasasinglepointonaplot,allowingchangesbetweenteststobeeasilyidentified.Abnormaltestsareshownasoutlyingpoints,withnormaltestsformingacluster.Modelingthenormalbehaviorofseveralfeaturesselectedfromtheprovideddata,thismethodshowedthatadvancewarningofsystemfailurecouldbeautomaticallydetectedusingthesefeatures,aswellashighlightingsignificanteventswithinthelifeofthesystem.III-TemplateAnalysisThismethodallowsinstantaneoussample-bysamplenoveltydetection,suitableforon-lineimplementation.UsingacomplementaryapproachtoSignatureAnalysis,thismethodalsomodelsnormalsystembehavior.Resultsconfirmedtheobservationmadeusingpreviousmethods.IV-NeuralnetworkPredictorSimilarlyusefulforon-lineanalysis,thismethodusesanautomatedpredictorofsystembehaviour(aneuralnetworkpredictor),inwhichpreviouslyidentified