EXPLORATORY DATA ANALYSIS OF TRACE ELEMENTS IN CLI

EXPLORATORYDATAANALYSISOFTRACEELEMENTSINCLINKERJánosAbonyia,FerencD.Tamás*b,andJosefTritthartcUniversityofVeszprém,Hungary,H-8201VESZPRÉM,P.O.B.158aDept.ofProcessEngineeringbDept.ofSilicate-andMaterialsEngineeringcTechnicalUniversityofGraz,TVFA,A-8010GRAZ,Stremayrgasse11Tel.[phone]Fax[phone]E-mailaddresses:[e-mail]@[e-mail]:Exploratorydataanalysis,Clinker,Traceelements,Boxplot,Quantile-quantileplotAbstractTheanalysisofdatatakenfromthemeasurementsoftraceelementsinclinkersmayleadtoinsightsintothepropertiesofrawmaterialsandcanbeusedtosolvepracticalproblems:todeterminetheoriginoftheclinker(i.e.themanufacturingworks).Forthispurpose,severalhundredclinkersfromninecountrieshavebeenanalysedbyreplicatedquarterlysamplestodeterminetheirMg,Sr,Ba,Mn,Ti,Zr,ZnandVcontents.ThispaperpresentsasurveyofsimpleExploratoryDataAnalysisproceduresthathavebeenfoundtobeparticularlyusefulinthequalitativeanalysisofclinkertraceelementcontents.Fordatapresentationboxplotsandquantile-quantileplotsareproposedandusedtoanalysetherelationshipsbetweendifferentfactoriesanddifferenttraceelements.21.IntroductionForclassicaldataanalysis,datacollectionisfollowedbyimpositionofamodelandtheanalysis,estimation,andtestingthatfollowsarefocusedontheparametersofthatmodel.ForExploratoryDataAnalysis(EDA),thedatacollectionisnotfollowedbyamodelimposition;ratheritisfollowedimmediatelybyanalysiswithagoalofinferringwhatmodelwouldbeappropriate.EDAisanapproachphilosophyfordataanalysisthatemploysavarietyoftechniques(mostlygraphical)tomaximizeinsightintoadataset,uncoverunderlyingstructure,extractimportantvariables,detectoutliersandanomalies,testunderlyingassumptions,developconservativemodels,anddetermineoptimalfactorsettings.TheseminalworkinEDAiswrittenbyTukey,[1].OvertheyearsithasbenefitedfromothernoteworthypublicationssuchasDataAnalysisandRegressionbyMostellerandTukey[2],andthebookofVellemanandHoaglin[3].MostEDAtechniquesaregraphicalinnaturewithafewquantitativetechniques.ThereasonfortheheavyrelianceongraphicsisthatbyitsverynaturethemainroleofEDAistoopen-mindedlyexplore,andgraphicsgivestheanalystsunparalleledpowertodoso,enticingthedatatorevealitsstructuralsecrets,andbeingalwaysreadytogainsomenew,oftenunsuspected,insightintothedata.Incombinationwiththenaturalpattern-recognitioncapabilitiesthatweallpossess,graphicsprovides,unparalleledpowertocarrythisout.TheparticulargraphicaltechniquesemployedinEDAareoftenquitesimple,consistingofvarioustechniquesof:1.Plottingtherawdata(suchasdatatracesandhistograms).2.Plottingsimplestatisticssuchasmeanplots,standarddeviationplotsandboxplots.3.Positioningsuchplotssoastomaximizeournaturalpattern-recognitionabilities,suchasusingmultipleplotsperpage.TheaimofthispaperistopresentasurveyofsomeofExploratoryDataAnalysisproceduresthathavebeenfoundtobeparticularlyusefulinthequalitativeanalysisofthetraceelementcontentoftheseclinkers.Adetaileddescriptionofthesuggestedtoolshelpstheconstructionofbarandq-qplotsofclinkers;stilleasier,acomputerprogramhasbeenwritten(inMATLAB©)anditcanbedownloadedfrom[4]presentedamethodtoconstruct“starplots”,wherethevisualimpactofchemicaldifferencesaremorestriking.Thisstarplotmethodhasbeenfurtherdevelopedinapreviouspaper[5];tofacilitatethevisualisationofthetraceelementcontentanewstarplotispresentedwhereforeachclinkerbeingcomparedtotheproposedstandard.Thefirstpaperof“fingerprinting”i.e.todeterminetheoriginoftheclinker/cementbytraceelementanalysiswaspublishedin1993byGoguelandStJohn[6],showingtheBa,SrandMnconcentrationsofNewZealandPortlandcements.Notalltraceelementscanbeusedforfingerprinting;selectionmustfollowcertainprinciples.Themostimportantitemofselection:traceelementsof“dactylogrammaticvalue”shouldcomefromthemainrawmaterial(limestone,marl,clay)andnotfromthefuel,fromfurnaceliningorfromgrindingmediawear,andsomeotherprinciplesshouldbeobservedaswell.Morerecentlysixelementswereusedtocharacterizeclinkers:besidesthoseusedbyGoguelandStJohn[6]Mg,TiandZrcontentswerealsomeasured[7-9].ZnandVhavenodactylogrammaticvalue(theycomefromthefuel,ifwastetyresorspecialsortsofheavyfueloilareused,resp.),buttheirquantitycanbeinterestingincementperformance.Aqualitativeidentificationobviouslyrequiresadatabase,tocomparethetraceelementcontentofunknownclinkers/cementswithcharacteristicknownsamples.Forthequalitative“fingerprinting”ofclinkers,adatasetfromwelldefinedclinkersamplesisnecessary:forexample,compositeaveragesamplesofalongperiodofkilnoperation.Suchsampleswerecollectedquarterly,overalongerperiod(2-3years)inallHungariancementfactories.Anationwidesamplecollectionhasalimitedvalueonly;toobtainabroadersampling,aTechnicalCommittee„QIC”(QualitativeIdentificationofClinkersandCements)hasbeenestablishedin1996,undertheauspicesofRILEM(RéunionInternationaledesLaboratoiresd’EssaisetdeRecherchessurlesMatériauxetlesConstructions