CIRPTemplatev4.0

OnthetrackingofindividualworkpiecesinhotforgingplantsMathiasLiewald(3),CelalettinKaradogan,BenjaminLindemann,NasserJazdi,MichaelWeyrichUniversityofStuttgart/GermanySubmittedbyTaylanAltan

Trackingeachworkpieceprovidestwomajoradvantagesinforgingtechnology.First,thematchingofphysicalworkpiecewiththemonitoredprocessinformationfacilitatesroot‐causeanalysisforproductquality.Second,thefollowingprocessstepscanbeadaptedaccordingtotheincomingworkpieceproperties to improve the robustness of hot forging process chain. The paper presents a general trackingmethodology and tagging experiments onaluminiumandsteel forgings forharshdrop‐forging technology.Furthermore,a framework forstreamingandprocessing largeamountsof real‐timedataaswellasamultidimensionalapproachtomodelandanalysetheworkpieceinformationforindividualandbatch‐trackingarepresented.Forging,DigitalManufacturingSystem,Knowledgemanagement

1.Introduction

The real‐timenetworking of technical systems over digitalisa‐tionofproductsandprocessesistherevolutionarystepinIndus‐try4.0.The fusionof virtual and realworldwith real‐timedataopensthewayforreal‐timeoptimizationofcomplexvalue‐addedsystems based on the processing ofmeasurement data and theforecast of future developments [1]. Analysis of big‐measure‐ment‐data allows the determination of anomalies, correlations,patternsandconstitutesthebaseformachinelearningandadap‐tive‐robustsystems.Hotforgingisoneofthetechnologieshavingagreatpotentialof

improvementsinthespiritofIndustry4.0[2].Theproductquali‐ty and process stability are evaluated usually after forming oreven after the heat treatment, based on properties of randomlyselectedproducts.Thestate‐of‐the‐artapplicationistheisolateddigitalisationofindividualprocessesanddisassociateddigitalisa‐tionof finishedproducts.As a result, the causeof the scatter intheproductpropertiescannotbelinkedwiththeindividualpro‐cessvariablesandparameterfluctuations.Efficientdatacommu‐nicationwithin a forging system can only be achieved by back‐trackingandlinkingonlinelabelledmeasurementswithphysicalworkpiece,otherwise,correlationsandpatternscannotbeextrac‐ted.Furthermore,thereductionofthescrapratebyadaptivepro‐cess parameters based on incomingmaterial or part propertiescannotbeachieved.Cannolly [3] has reviewed in 2005 the use of bar andmatrix

codes in assembly operations and for part tracking. Types ofcodes,markingmethodsandmachinevisionequipmentarecom‐paredandthesuperiorityofmatrixcodelabellingisemphasized.Songetal.[4]developedarobustandaccuratematerialtrackingandlocatingsolutionformaterialsstoredinlargelaydownyards.Proposed solution features barcoding and GPS‐technologies formaterialtrackingandfastretrievalinacosteffectivemanner.Daietal.[5]investigatedthetrackingcapabilityfromtheperspectiveof supply chain considering the tracking and recall costs in asupply chainwith endogenous pricing. Denkena et al. [6] intro‐duced a vibration assisted face milling technology enabling the

machiningofamatrixcodeorsimilarshapesintothecomponentsurfaceeliminatinganadditionalmarkingstep.Vedel‐Smithetal.[7]developedamatrixcodemarkingstrategyforgreensandcas‐tingswhereaflexibleinserttoolisusedtoembossthemouldit‐selfbeforethemeltispouredin.Montaninietal.[8]exploredthepossibilitiesofactiveinfraredthermographyinrestoringcoveredandabradedmarksobtainedbylaser,dotpeen,impact,pressandscribemarkingonasteelsurface.Theyreportedon tested ther‐mography techniques proved to be helpful in reducing localopticalreflectionsandobtainedenhancedreadabilitywithproperimageprocessing of the raw images. The optical tracking of ob‐jectsisanactiveresearchtopicuntilnowandcanbeappliedtoofor tracking workpieces in the forging plants. Stolkin et al. [9]addressed the visual object tracking problem in extremely poorvisibility conditions which could be transferred to forging en‐vironments. Akin et al. [10] introduced an effective DIC‐basedtracker with an improved deformable part‐based model whichtackles also scale changes in successive pictures. Kanagamalligaet al. [11] recently showed how background subtraction andfeature extraction couldbeused to improve the visual tracking.Heetal. [12] introducedanapproach for spatio‐trackingofmo‐ving objects using multiple cameras simultaneously. WhenapplicabletheseDIC‐basedtoolscanbeusedtotrackworkpieceseliminatingthenecessityofphysicaltags.Thecommonapplicationofpartmarkinginforgingcompanies

todaydependsusuallyonthesizeofworkpieceandtheproduc‐tion rate. Largeworkpieceswith small production numbers aretracked by using labels, needling ormarking. Small workpiecesbeingmanufactured in high production volumes do allowbatchtracking. The ability to sample accurate process data and tocorrectly assign them to corresponding workpieces is a majorchallengeregardingtheheterogeneityofdatasourcesconcerningdataformats,protocolsandsamplingrates.Hence,softwaremo‐dulesareneededthatuniformlytransfersthecollecteddatafromthemachineleveltoafactory‐cloudlevelstorage.Modulesshouldhavean individual interface foreachdatasourceandastandar‐dizedinterfacetofurthertherecordeddata.FundamentalsofthisconceptarepavedbyFauletal.[13].

Contents lists available at SciVerse ScienceDirect

CIRP Annals Manufacturing Technology

Journal homepage: www.elsevier.com/locate/cirp

Theaimofthispaperisontheonehandtoformalisetheknow‐hownecessaryfortheworkpiecetrackinginforgingplantsusingthe state‐of‐the‐art hardware and software techniques. On theotherhand,adeepfocusisputonfurtherinvestigationofphysi‐cal tags and data driven modelling of traceability. A successfulworkpiece tracking system needs robust tags and componentswiredwith tracking software running at bothmachine and fac‐tory‐cloud level. In thiscontext,at firstageneralworkpiece tra‐cking methodology is developed and supported with multi‐dimensional modelling of data‐driven‐traceability built on PLC‐baseddataacquisitionandprocessing.Thismethodologycanbeusedtobuildatailor‐madetrackingsystem.Second,practicaltagsappropriateforforgingenvironmentareinvestigatedthoroughlyandlaser‐engravingofQR‐codesareperformedontothehotsur‐faceofforgedaluminiumandsteelworkpieces.Finally,anassis‐tancesystemforasingle/batchtrackingisintroduced.

2.Workpiecetrackingmethodology

Aproductionlineinaforgingplantmayconsistofvariouspro‐cesses depending on product itself and available infrastructure.This section introduces elements of a generalized methodologythatcanbetailoredtoavarietyofproductionsystems.Workpie‐cescanbetrackedonabatch levelor individually.Harshnessoftheprocess,tagdurability,processandtaggingspeeddeterminestracking intensity.A completeproduction linemay containpro‐cessessuchascasting,extrusion,sawing,turning,transfer,stora‐ge, heating, forging, blasting and heat treatment. Process infor‐mationcanbestoredinthefactory‐cloudandcanbeassociatedtoa specificworkpiecebymeansof amaster‐identity. It isnotne‐cessarytostorethecompletemaster‐identityonatagfixedtotheworkpiece. Local or temporary identities can be assigned andreusedatthemachinelevelprovidedthatthecorrespondingmas‐ter‐identities are distinct. In case economic and physical condi‐tions desire for the next process relabelling can be performed.Trackingofaworkpieceoveramaster‐identityusingvarious lo‐cal‐identitiesthroughouttheproductionlinerequiresnetworkingof the technical trackingsystems inreal time.Manual identifica‐tion, transfer and registration of identities can be a part of thisnetwork. With such manual networking practise the trackingproblemreduces to localized trackingofworkpieces throughoutindividualprocesses. 2.1.ProcessformalisationProcess ina forgingproduction linemaybe clustered into the

followingprocesscategories(PC): (PC1)Tagenduresthroughouttheprocess. Tagdoesnotendurethroughouttheprocessand

o (PC2)workpiecesareprocessedinsequenceo (PC3)batchprocessingisperformed.

ProcessesbelongingtoPC1donotdamagethetagappliedontotheworkpiece,sothetagcanbeidentifiedflawlesslyafterprocessing.Examples are storage, transfer or heat treatment. Workpiecesqualified as scrap can be identified and registered easily. Pro‐cesses belonging to PC2 do damage the tag, but the sequentialprocessing allows the tracking of the identity within and afterprocessingaccordingtothe identityscannedbeforetheprocess.Partscanbequalifiedasscrapduringorrightafter theprocess.Especiallyscrapdeterminedandseparatedduringprocesshastoberegisteredinrealtime,whichcanbeautomatedorperformedbymanual triggering. State‐of‐the‐art applications such as forcemeasurementsandpartproximitysensorslocatedatgripperscanbe used for recognition ofmissing parts.Processes belonging toPC3 damage the tag, the individual identity of produced part islost sincemanyworkpieces areprocessed together. In this case

individualrelabellingisnotpracticalduetohighproductionrateor constraints given by logistics. Such processes (e.g., heat‐treatment,blastcleaning)consequentlypermitonlybatchestobetracked.Individualinformationofworkpiecesprocessedtogethercanbemergedina“statisticalrepresentation”storedinthefacto‐ry‐cloudandonlycanbeassociatedtobatches.TaggingtimebeinglongerthantheprocesstimeincaseofPC2

hinders the tracking of all the workpieces individually unlessmultipletaggingstationsareemployed.Valuablelinkedinforma‐tion on processes and workpieces however can be created forsomeoftheworkpieceswithasingletaggingstation.Batch trackingcanbeenrichedby inlinemeasurements to im‐

proveon‐sitetheinformationcontentofthetracking.Henceforthtrackingwithadditionalindividualinformationcanbeperformedduringtherestoftheprocesschain.Forexample,theworkpiecesize,usuallyhavingahighscatterfrequency,isanimportantissueforprocesscontrolandforposteriorbig‐dataanalysisonparame‐terdependency.On‐sitemeasurementscanbeextendedwithfastgeometrymeasurements (such as shear face of a billet) orwitheddy‐currentmeasurements,ifapplicable.The operative tracking actions are tagging with the (part or

batch) identity and tag scanning to recall the (part or batch)identityand thecorresponding information in the factory‐cloud.The computational tracking actions are locating the identity inthefactorynetwork,providingtherequiredinformationfromthefactory‐cloud to the processmachine and storing the generatedproductioninformationtothefactory‐cloud.Locatingtheidentityinthefactorynetworkiseitherachievedbyonsitetagscanningor“virtually”withouttheaidofaphysicaltagontheworkpieceandcan be named as a virtual tag. If a strict transfer protocolwithsequentialprocessingis implementedforarangeofprocesses,avirtuallocationtrackingcanbeperformedusingonsitebusy/freelocationswitches.These locationswitchescanberealisedeitherby part sensors or digitalmonitoringwith image processing. Intheabsenceofastricttransferprotocol,locationswitcheshavetobesupportedbymotiontrackingtoperformthevirtualtracking.Motion tracking could be implementedwith a networkof videocamerasandimageprocessingtechniques.Cameralocationsandmotionpathshavetobeplannedtoyieldarobustconfiguration.Thecategoryofaprocesscanbealteredbyaprocessredesign

to improve the traceability.Workpiece conveyors or workpiececarrierswithastrictpreservationofprocessingorderinsteadofadisorderlytransferaretrivialexamplesforsuchanimprovement.At least, small workpiece containers providing reduced batchsizes can be used to increase the specificity of the statisticalrepresentationoftheassociatedinformation.2.2.PLC‐baseddataacquisitionandprocessingThe individual interfaces of the connector softwaredo runon

controldevicesandthestandardized interfaceruns inthe facto‐ry‐cloud. The individual interface is basically implemented asstatemachinethatcanbeintegratedintothePLCcode.Twoma‐jor cycles canbedistinguished. Firstly, a “read”‐cycle that scansthe protocol structure and extracts data depending on the bussystem. Secondly, a “write”‐cycle that uses aRESTmodel to de‐scribetheextracteddata.Themeasuredprocessparametersareconvertedandmappedontothemodelandpostedtothestandar‐dized interface that runs on the cloud server. The standardizedinterface forwards the data for further processing. It is eitherstored in a relational databasewhere data analytics techniquescanbe applied or forwarded to a real‐timedatabase to conductad‐hoccalculationsandonlineanalyticalprocessing(OLAP).Thereal‐time database is used for the association of workpieceidentityandprocessparametersalongtheentireprocesschain.Itallowsanonline trackingofsinglepartsorbatches.ThegeneralconceptisdepictedinFigure1.

Figure1.Conceptandsystemarchitecture.2.3.MultidimensionalmodellinganddatadriventraceabilityToachieveanefficientmonitoringandtracking,anOLAPmodel

has to be created under consideration of the data spaces of thesingleprocess steps.Themodel is separated intodimensionsoftime, space and product information, Figure 2. The dimensionscontainaninternal,hierarchicalstructurewhereeachbranchingis described by an element. Basic elements are associated withdata vectors and mark the innermost layer of the data model(producttypePwithidx).Higherlayers(batchAofproducttypeP) cumulate the information of lower layers by applying aggre‐gation procedures. Thus, heterogeneous data can be integratedintoonehomogeneousmodelduetothefactthatitispossibletoassociatedifferentlayerswithdifferentgranularities.Thisfactorenables the tracking of a flexible amount of parts by navigatingthrough layers. Themodelling approach is visualized as cube inFigure2.AccordingtoShinetal.[14],threemajorpropertiesarerequired forn‐dimensionalmodelstructures: ‘commonality’, ‘re‐usability’and‘composability’.Subsequently,authorsintendtowi‐den the ‘composability’‐aspectby ameta‐modelling approach todefine thresholdsandrules forprocessparameters toenableaneffectivelimitmonitoring.Regardingthedescribedprocesschain,varyingenvironmentalconditions indeeddoaffectthequalityofprocessandproduct.Forexample,decreasingqualityofcoolantsdependsonnumerousfactorsincludingusefullife(timedim.)andmachine (space dim.). Consequently, these side effects can beformalizedandmappedonthedimensionstructureandprojectedontothemeta‐cube.Hence,anautomatedmonitoringonthebasisofexpertknowledgecanberealized.

Figure2.Multidimensionalmodelofprocessandworkpieceinformation.

3.Practicaltagsandtaggingexperiments

The design of a complete tracking system considers appliedprocesses, part geometry and material. The discussion here islimited to the comparison of the known tag characteristics andthe investigations on the not yet tested characteristics. In thiscontext, laser‐engraving tests on hot steel and aluminium areconducted and the results are comparedwith needling. Part orbatch tracking solutions for operations such as machining,

sawing, transfer or storage are already in industrial praxis. Theharsh forging environment, possibly covering heating, forging,blasting and heat treatment allow RFID, labelling, needling,marking,laser‐engravingandvirtualtaggingtobeusefulthrough‐out theseprocesses.Tagsshouldnotcauseanyharm, theymustendure and could be tagged around 500°C for aluminium and1200°C for steel. Oxide‐layer has to be removed before taggingandhinderedafterwardsbutthetagshouldendurethedevelop‐mentoftheoxide‐layer.AheatproofRFIDisdurableupto350°Catthetransponder,[15],withasizeof10×7×3cmbeingpracticaloncontainersforbatchtracking.Therearealsolabelsdurableupto 1250°C, providing flexible size and content for individual orbatchtracking[16].Thedifficultyariseswhileattaching,keepingandremovingthelabels.Althoughtheyarejustsuperficial,need‐ling,markingandlaser‐engravingbecomeapartoftheworkpie‐ce.Theyaffectthegeometryandeventhemateriallocally.Table1compares functionality of needling, marking, laser‐engravingbasedonmarketresearchconductedduringexperiments.QR‐Codecanprovidesuccessful recognitionevenapartof the

label is damaged. Figure 3 demonstrates on the left a QR‐Codegeneratedby a fibre‐laser onaluminiumat 550°C anddamagedafterwards by collisions with other parts during shaky batchtransferinacontainer,yettherecognitionisfunctional.Figure3demonstratesontherightcoldmarkedsteelwhichiscoveredbyoxide‐layerforthreehoursat900°C.Visiblemarkingsontheoxi‐de‐layerandonthecleansurfaceafterremovaloftheoxide‐layerare both identified by QR‐Code readers. If the QR‐Code readershouldfail,theattachedcleartextisthoroughlylegible.

Figure3.QR‐Codesgeneratedbyfibre‐laser:Aluminiumat550°Cdama‐gedbycollisions(left);SteelatRTcoveredbyanoxide‐layerdevelopmentat900°Cforthreehours(right).Allcodesarerecognisedsuccessfully.

Table1Comparisonoffunctionalityofcommercialtags Marking Needling FiberLaser‐Engraving(50W)

Content Cleartext QR‐Code,Cleartext

Markingdepth

Limited Strikeenergy

Beamenergyandrepetitiveshots

TaggingTime

~1sec. ~4sec. ~8sec.for50WFiberLaser(increaseswithsizeandsurfacepreparation,reduceswithpower)

Clarity Missingcontrast

Missingcontrast

Superiorduetosurfacepreparation

Size Markersize Needlesize Flexible

Resolution ‐ Fixed Flexiblewithinalabel

SurfaceCurvature

Slightcurvatures Flexiblewithadaptivefocusing

ClearText Highquality Dotted Dottedwithflexibleresolution

Temperat. Testedat550°CforAluminium,1200°CforSteel

Investment ~10.000€ ~30.000€

Scanningautomation

Opticalrecognition

QR‐Codereader,OCR

QR‐CodeReader,OCR(Superiorifsurfacepreparationisperformed)

Blastingresistance

Poor.Increasewiththedotsizeanddepth.

Needling and laser‐engraving can both be used to mark QR‐Codesandclear textson the surface. Since the resolutionof theneedling is limitedwith theneedlediameter themarkingsize isnot flexible. The depth of dots can be influenced by the impactenergy.Overlappingdots byneedling is not practical but this istotally harmless for a laser‐engraver and can be performed toincrease the depth of each dot and even the resolution of theengraving. A QR‐Code togetherwith the content in clear text isdifficult to generate by needling compared to laser‐engraving.Figure 4 compares a high resolution laser‐engraving (where asingledotisgeneratedwithmanytinydots)withtheresultoftheneedlingtodemonstrateclearlythelimitsoftheneedlingandthecapabilitiesoflaser‐engraving.Alaser‐engravingwithsingle‐dot‐resolution is also provided to demonstrate the differencebetween the dot characteristics. The larger and deeper dotcharacteristicsprovidesabetterdurabilityagainstblast‐cleaning.

Figure 4. QR‐Codes generated by needling (left), low resolution laser‐engraving (middle) and high resolution laser engraving (right), all withsuccessfulopticalrecognition.

4.Experimentalassistancesystemforsingle/batchtracking

In order to test the system architecture, a simulative processchainisused.ThePLCsoftheenvironmentrunwithacycletimeof1mswhichprevents the lossof relevantprocess informationdue to low sampling rates. Thedata space includes18differentsensorvalues(realandsimulated)of5consecutiveprocesssteps.The standardized interface runs in the cloud and forwards thedata in JSON format to theOLAP database. The transmission ofthe data can be controlled in variousways. Firstly, a constantlyfrequented query can be executed by the REST server in thecloud.Alternatively,anevent‐drivendatatransfercanberealizedthroughthecustomizedinterfaces.In case of different amounts of parts being tracked, the ETL

stack maps the new information on the hierarchical dimensionstructures.Thisprojectioncausesthecreationofhigherlayersforbatchtracking.Asaresult,newelementsaregeneratedwithinthehierarchy.Therefore,differentprocessdatastructuresofvaryingproductsandamountscanbeadapted.Thesame flexibilityapp‐liestoanyotherelementofanyotherlayerinthehierarchicaldi‐mensionstructure.Hence,itispossibletotrackdifferentamountsof parts by navigating in the hierarchical dimension structure.Batches can be associatedwith the corresponding process dataandcumulatedvaluesofthemeasureddata.Incontrasttosingleworkpieces,thetrackingofmultiplepartsdemandsthesafeesti‐mationwhetherallpartsfulfilthequalityrequirements.Thenavi‐gationtothebatchlayerautomaticallymonitorsallrelevantpara‐meters characterizing thebatch and gives theuser an overviewonaveragevalues.Thesevaluesarecalculatedbasedonallpartsofthebatch.Thewholedataintegrationprocessiscyclicallyexe‐cuted with parallel computing enabling online monitoring andtracking. It is implementedasweb interface fora flexibleaccessfromavarietyofdevices.Thefirstviewallowsthelimitmonito‐ring of process parameters so the exceedance of limit values islighted.Theusernavigatesinallthreemajordimensionsoftime,spaceandproduct.Thus,hecanselectsingleworkpiecesorbat‐chesatacertainprocessstepatapointintimeofchoiceandthe

accordingprocessdatawillbedisplayed.Inthesecondviewtra‐jectories and deviations are observable providing an efficientcontrolofprocessparameters.

4.Conclusion

Authorsproposeageneralworkpiecetrackingmethodologyforhotforgingplantscoveringprocessformalisation,multidimensio‐nalmodellingofdata‐driven‐traceabilityandPLC‐baseddataac‐quisitionandprocessing.Practicaltagsarereviewedandtaggingexperiments are performed to extend the possibilities of indivi‐dualworkpiecetracking.Therealisationof individualworkpiecetrackingforhighproductionratesrequiresre‐designofprocesseslike transfer, storage heating and heat treatment based on theintroducedprocessformalisation.Real‐timeregistrationofscrapsisarequirementinthetrackingproblem.QR‐Codewithcleartextextension is suggested formanually aided automation.QR‐Codestandards are based on the unit square dot fully covering thegrid‐cell. Needling and laser‐engraving in the single‐dot‐resolu‐tionhowevergeneraterounddotswithout fillingthecorrespon‐ding grid‐cell. Furthermore, the colour and contrast in a dottedgrid‐cellisnotoptimal.ReadabilityofaQR‐Codedependsontherecognitionofadot,havingadistinguishedcontrast froma freegrid‐cell. Since DIC based identification use this standard tech‐niquewhichisbiasedtowardstheprintedcodeswithsharpcon‐trast difference, identificationof the engraved codesmay fail incaseofbadlighting.Inthisstudy,standardsystemsareusedandmostlysuccessfulresultsareobtainedwithproperlighting.How‐ever,theneedofimprovedsoftwareforDICbasedidentificationisalsofelt.

5.Acknowledgement

This work was supported by the German Federal Ministry forEconomic Affairs and Energy within the framework of theProgrammeforCompetitionDigitalTechnologiesforBusiness.References[1] Bauernhansl, T., Krüger, J., Reinhart, G., Schuh, G., 2016, WGP‐StandpunktIndustrie4.0.,WissenschaftlicheGesellschaftfürProduktionstechnikWGPe.V.[2] Liewald, M, Karadogan, C., Felde, A., Lodwig, R., 2017, Development andIntegration of Digital Technologies in the Forging Process Sequence, NewDevelopmentsinForgingTechnology,245‐256.[3] Connolly, C., 2005, Part‐tracking labelling and machine vision, AssemblyAutomation,25/3:182‐187.[4]Song,L.,Mohammed,T.,Stayshich,D.,Eldin,N.,2015,ACostEffectiveMaterialTracking and Locating Solution forMaterial Laydown Yard, Procedia Engineering,123:538‐545.[5]Dai,J.B.,FanL.,Lee,N.K.S,Li,J.,2017,Jointoptimisationoftrackingcapabilityandpriceinasupplychainwithendogenouspricing,IJPR,55/18:5465–5484.[6]Denkena,B.,Grove,T.,Seibel,A.,2016,DirectPartMarkingbyVibrationAssistedFaceMilling,ProcediaTechnology,26:185‐191.[7]Vedel‐Smith,N.K.,LenauT.A.,2012,Castingtraceabilitywithdirectpartmarkingusingreconfigurablepin‐typetoolingbasedonparaffin–graphiteactuators, JournalofManufacturingSystems,31:113‐120.[8]Montanini,R.,Quattrocchi,A.,Piccolo,S.A.,2016,Activethermographyandpost‐processing image enhancement for recovering of abraded and paint‐coveredalphanumericidentificationmarks,InfraredPhysTechnol.,78:24‐30.[9] Stolkin, R., Greig, A., Hodgetts,M., Gilby, J., 2008, An EM/E‐MRF algorithm foradaptivemodel based tracking in extremely poor visibility, ImageVisionComput.,26/4:480‐495.[10] Akin, O., Erdem, E., Erdem, A., Mikolajczyk, K., 2016, Deformable part‐basedtracking by coupled global and local correlation filters, J. Vis. Commun. Image R.,38:763‐774.[11] Kanagamalliga, S., Vasuki, S., 2018, Contour‐based object tracking in videoscenesthroughopticalflowandgaborfeatures,Optic,157:787‐797.[12]He,T.,Mao,H.,Yi,Z.,2017,Movingobjectrecognitionusingmulti‐viewthree‐dimensionalconvolutionalneuralnetworks,NeuralComput&Applic,28:3827‐3835.[13] Faul, A., Jazdi, N. and Weyrich M., 2016, Approach to interconnect existingindustrialautomationsystemswith the industrial internet.21st IEEEInternationalConferenceonEmergingTechnologiesandFactoryAutomation(ETFA)2016.[14] Shin, S.J., Woo, J. and Rachuri, S., 2014, Predictive analysis model for powerconsumptioninmanufacturing,ProcediaCIRP15:153‐158.[15]http://sawcomponents.de/produkte/saw‐sensorik‐und‐rfid/[16]http://heatproof.eu/products/