Design and management of manufacturing systems for production quality

Design

and

management

of

manufacturing

systems

for

production

quality

Marcello

Colledani

,

Tullio

Tolio

(1)

*

,

Anath

Fischer

(1)

,

Benoit

Iung

(2)

,

Gisela

Lanza

(2)

_,

_Robert

_Schmitt

₍₂₎

_,

_Jo´zsef

_Va´ncza

₍₁₎

a

PolitecnicodiMilano,DepartmentofMechanicalEngineering,VialaMasa,1,20156Milan,Italy

b

ITIA-CNR,InstituteofIndustrialTechnologiesandAutomation,ViaBassini15,20133Milan,Italy

c_{CAD&LCELaboratory,FacultyofMechanicalEngineering,Technion-IsraelInstituteofTechnology,Haifa,Israel} d_{LorraineUniversity/CRAN,FacultedesSciences,Vandoeuvre,France}

e

InstituteofProductionScience(wbk),KarlsruheInstituteofTechnology(KIT),Kaiserstrasse12,76131Karlsruhe,Germany

f

LaboratoryforMachineToolsandProductionEngineeringWZL,RWTHAachenUniversity,Steinbachstr.19,52064Aachen,Germany

g

InstituteforComputerScienceandControl,HungarianAcademyofSciences,Budapest,Hungary

h

DepartmentofManufacturingScienceandTechnology,BudapestUniversityofTechnologyandEconomics,Hungary

1. Introduction,motivationandobjectives

Product quality and delivery reliability are key factors for successinthemanufacturingindustry.Moreover,theincreasing emphasis on sustainable production requires maintaining the resourceefficiencyand effectivenessalong theproduct,process and production system life cycle [274]. Quality, production planningandmaintenancearefundamentalfunctionsfor achiev-ingthesegoals.Theyhavebeenwidelyanalysedintheliterature overthepastseveraldecades.Theproductionplanningfieldhas developedmethods forreducing workin progress(WIP),while meeting desired production rate levels. The Statistical Quality Control(SQC)fieldhasintroducedoptimizedtoolsformonitoring thebehaviourofprocessestoachievethedesiredproductquality. The Maintenance Management field has derived policies for preserving the efficiency of degrading resources over time by offeringpro-activeandpredictivecapabilities[112].Traditionally, allthesefieldshavebeentreatedbyscientistsandindustrialists almost in isolation. Yet it is clear that equipment availability, productqualityandsystemproductivityarestronglyinterrelated. Asamatteroffact,quality,maintenanceandproductionplanning

strongly interact and jointly determine those aspects of a company’ssuccessthatarerelatedtoproductionquality,i.e.the company’s ability to timely deliver the desired quantities of productsthatareconformingtothecustomerexpectations,while keepingresourceutilizationtoaminimumlevel.

Forexample,lowWIPimprovestheabilityofidentifyingquality problemsinthesystematanearlierstagebutatthesametime makes maintenance actions more critical to the system. More inspectionsmakeitpossibletobetterassessthedegradationstate oftheresourcesyetalsoincreasethesystemlead-time.Frequent maintenance of resources tends to improve part quality, but reducestheoperationaltimeofthemachinesinthesystem,which affectstheoverallproduction.

It is clear, then, that the mutual relations among quality, productionplanningandmaintenancecontrolshouldnotbe under-estimatedwhileconfiguringandmanagingthemanufacturingsystem as a whole.Important practical questions,suchas ‘‘Whichis the expectedsystemeffectiveproductionrateifthetimetopreventive maintenanceofonemachineisreduced?’’and‘‘Whichistheeffectof increasingtheinspectionfrequencyofoneproduct featureonthe overallproductionyieldofthesystem?’’remainunsolved.Thislackof understandingresultsinsub-performingunbalancedsystemic solu-tionsthattendtoprivilegeoneofthe aspectswhilepenalizingthe overallmanufacturingsystemefficiency.

Keywords:

Manufacturingsystem Productionquality Maintenancemanagement

Thekeyquestionsthatthispaperaddressescanbeformulated asfollows:‘‘Whicharethemainindustrialproblemsrelatedtothe achievement of production quality targets?’’ ‘‘Which tools can supportthejointconsiderationofquality,productionlogisticsand resource maintenance in manufacturing system design and operation?’’ and ‘‘Which are the main technical achievements andinsightsbroughtbytheuseofthesetoolsinindustry?’’

Recently, several production paradigms havebeen proposed that are strongly related to and have an impact on production quality.Theseparadigmshaveconsiderablyreshapedthe bound-aries withinwhich thethree aspects interact. Reconfigurability

[134],flexibility[278],changeability[309]andco-evolution[280]

stresstheimportanceofaimingatastrongcoordinationbetween thedynamicsof thesystemlife cycleand thedynamics of the productandprocesslifecycles.Takataet al.[274]introducethe notion of ‘‘life cycle maintenance’’ to be in phase with such requirements. Yet, if a system evolves with faster and faster dynamics,newchallengesariseforproductionquality.Inparticular, thelong-termperformanceofthesystembecomeslessimportant, while production quality performance during system ramp-up assumesfundamentalrelevance[86].Moreover,small-lot produc-tionbecomesmorefrequentthanmassproduction.Therefore,a newproductionqualityparadigmisneededformasscustomization [60] and mass personalization [282], for open architecture products[135]andforhighproductvarietymanufacturing[79]. Availableconcepts and programmes,suchas Six-Sigma, JustIn Time, Continuous Improvement, Total Quality Management, ToyotaProduction Systemand World Class Manufacturing,are not designed for such dynamically changing contexts. A new integratedconceptofproductionqualityneedstobedevelopedto meetthisaim.

Anotherindustrialtrendthathasbeenrecentlyinvestigatedand framed[81] is the increase of the complexity of manufacturing systems,intermsofhardware,software andmanagementrules. Complexitystronglyunderminestheachievementofthe desired production quality performance. Complex systems are typically characterizedbyalternativeprocessplans[208],multipleparallel resources,parttypedependentroutings,andlatevariant differenti-ation[102].Theresultingchallengeliesintheadditionalburden placedondiagnosis,root-causeanalysis,anderrorbudgeting.

In response to these innovative aspects of manufacturing systems, multiple in-line technologies for data gathering and performancemonitoringhaveemerged.Aconsiderableamountof dataistypicallymadeavailableonmodernshopfloorsby multi-sensor technologies [304]. However, most of the time this information is treated only locally and is not spread among different company functions nor among partners within a productionnetwork.Forexample,itisnotinfrequentforaquality managementdepartmenttoignorethereliabilitystatisticsofthe machinesontheshopfloor[152].Thisbehaviourmakesithardto correlatedisruptivephenomenatakingplaceat shopfloorlevel withtheproductqualityandtogatherinsightsinthebehaviourof thesystemasawhole.Itwouldbenecessarytomovefromisolated engineeringpracticestomoreintegratedonessuchasadvocated bySystemEngineeringinitiative[105].Therefore,thesedataare not fully exploited and translated into a business competitive advantageforthecompany.

The impactof complexityon production quality iseven more significant when considering the production network level. For example,exceptfortheperiodofthedeepeconomiccrisis2009– 2010,thenumberofrecallshasbeenconstantlyincreasingalsodue thelackofinter-organizationalqualitysystems[61].Productrecalls indicatethatmanufacturingcompaniesareparticularlyvulnerable toensurequalitywhentheysourceviaaglobalsupplychainwith poorvisibility[164].Globalautomotivewarrantiesareestimatedat USD40billionperyear,i.e.a3–5%lossinsales[89].Lowpriced productionoftenleadstoqualityproblems,andoutsourcingleadsto ashiftin knowledgeconcerningtechniquesand processes.Thus, technicalfailuresaremorelikelytooccurduetocommunication failuresamongthedifferentpartiesengagedinthesupplychainand

tomissingdefinitionsfortechnicalinterfaces.Sincemostoftheflaws thateventuallycausefailuresareintroducedintheproductionphase, earlyfailureanalysiscanaverthighrecallcostsandlossofimage.

Legislationthat limitsindustrial wasteproduction, increases targetproductrecyclabilityratesandplacesthemanufacturerat the centre of the end-of-life treatment process through the ExtendedProductResponsibility(EPR)principleis anadditional driverthatstronglyinfluencestheproductionqualityparadigmby penalizingthegenerationofdefectsandwasteinmanufacturing. Moreover,sustainabilityissuesrelatedtoenergyefficient produc-tion [76] have to be taken into account while designing and operatingthesystemasawholeforadesiredoutputproduction quality-relatedperformancetarget.

Topromoteintenseandcoordinatedresearchactivitiesaimedat developinginnovativetechnologicalandmethodologicalsolutions to the aforementioned challenges, industrial organization and fundingbodieshaverecentlylaunchedseveralactions.Forexample, atEuropeanlevel,theFactoriesoftheFuture(FoF)PublicPrivate Partnershiphasincludedthetopic‘‘ZeroDefectManufacturing’’asa priorityinitsFoF2020Roadmap.Moreover,undertheFP7callon ‘‘Zero Defect Manufacturing’’ four projects have been funded boosting cross-sectorial research and aiming at achieving the largestpossibletargetimpactforthedevelopedtechnologies.These activitiessharetheobjectiveofsupportingthedevelopmentofa knowledge-basedmanufacturingandqualitycontrolculturewithin theEU.SimilaractivitieshavealsobeenpromotedintheUSAwithin theAdvancedManufacturingPartnership(AMP).

This paper provides an overview and a framework of the industrialpractices,scientificmethodologies,andenabling tech-nologies to profitably manage the productionquality targetsin advanced manufacturing industries. It also identifies key open researchand practical issues that needto beaddressedby the researchcommunity.Thepaperisstructuredasfollows:thenext paragraph presents a set of real cases that demonstrate the industrialmotivationtotheproblem.Section2proposes anew system dynamics model for highlighting the relevant quality, maintenanceandproductionlogisticsinteractions.Sections3and 4discuss,respectively,thestate-of-theartmethodsandtoolsand the enabling technologies supporting the production quality paradigm.Finally,Section5describesthemostpromisingfuture researchtopicsinthisarea.

1.1. Industrialmotivation

Inordertohighlightthemainpracticalimplicationsrelatedto theinteractionsamongquality,productionlogisticsand mainte-nanceandtopointouthowthesechallengesarecurrentlytackled by companies, a comprehensive setof realindustrial examples have beencollected. Thesecase studies havebeengathered by analysingexistingpublications,runningindustrialprojects,both publically and privately funded, and by gathering authors expertise.Theyincludeboth traditionalproductionsectorssuch astheautomotiveandelectronicssectorandemergingsectorsof certaininterestfortheworldwidemanufacturingcontext, includ-ingtheproductionofmedicaldevicesaswellasthegreenenergy production industry. Moreover,they includea reasonably wide spectrumofmanufacturingprocesses,suchasmachining, assem-blyand forming,atboth macro and micro scales,and on both metallicandnon-metallicworkpieces.

Theindustrialcasessupportthefollowingconsiderations:  Theinteractionamongquality,productionlogisticsand

mainte-nanceaspectsisacomplexissuetobemanaged.

 This problem involves different companies and different departments within each company. The coordination and cooperationamongtheminachievingarightbalancebetween theseconflictinggoalsisseenasakeyissueforsuccess.  Dependingonthespecificproductandmarketcontext,

compa-nies tend to prioritize one of the aspects. Finding the right balancebooststhelong-termcompanyprofitability.

 Theincreasingcomplexityofproducts,processes,andsystemsis oneofthemajorchallengesforproductionquality.

 EmergingICTandinspectiontechnologiesaswellascooperation basedonrisk-sharingcontractsareseenasfundamentalenablers tomeetproductionqualitytargets.

 Emergingproductionparadigms,suchasreconfigurabilityand flexibility,posenewchallengesforproductionquality.

 Industrialcompaniesareexperiencingatrendtowardsincreased investmentsintheirabilitytoprofitablydriveproductionquality trade-offs.

1.2. Analysisofrealcases

The first example refers to the production of batteries for electricvehiclesintheelectricmobilityindustry.Thee-mobility industryisexpectedtoreachitstargetproductionby2020.

As demand is still limited, manufacturers are putting great emphasisonqualityimprovement.Dominatingtechnologiestobe adoptedfor theproduction and inspectionofbatteries are still lacking[143]. Errorpropagation is themajor causefor defects. Differentqualityteststakeplacesuchaselectrictest,stackingtest, leaking test, and temperature tests. However, not all the root causesfordefectsareknownsincethequalitymanagementisstill inalearningphase.Therefore,aspecificprocedureisadoptedto managetheramp-upduringtheintroductionofnewtechnologies in the plant. The quality planning process starts with the identification of critical product characteristics to be used to determine theproduct quality level. First, newproduction and inspectiontechnologiesaretemporarilyintegratedoff-lineinthe factorytoavoidinterferencewiththecycletimeofthemainline during the ramp-up phase (Fig. 1). In this phase, technology improvementpractices are implementedand knowledgeof the processbehaviourisgathered.Oncetheprocessismadestable,the technologiesaredevelopedason-lineintegratedoperations.Inthis context(GermanBMBFProject‘‘ProBat’’[149]),themainrelevant questionsare‘‘wheretointegratethemeasurement,withwhich technologiesandwhichstrategies?Whataretheconsequencesof these choices on the quality and production logistics perfor-mance?’’ Only by integrating quality management in factory planningcantheseimplicationsbecaptured.

Theneedsforin-lineinspectioninmulti-stagemanufacturing processesare alsoaddressed in thefollowing real-life example relatedtotheproductionofelectricalenginesfortheautomotive industryatRobertBoschGmbH(Fig.2(a)).Thisrealcaseisoneof thedemonstratorsintheMuProDFP7EU-fundedproject[200].The proposedexampleisspecificallyrelatedtotheassembly lineof electricaldrives.Thisisamulti-stagesystemtypicallyinvolving 20–30 process stages. Three main flows are found, the first dedicatedtotheassemblyofthemagneticrotor,thesecondrelated totheassemblyofthestatorandthelastrelatedtothecouplingof statorandrotortoproducethecompleteengine.

State-of-the-artinspectiontechnologiesfacilitateassessingthe quality of theengine by end-of-line testing of several product features, themost importantbeing themagnetic torqueof the rotor. However, in order to better understand the causes for deviationsandtoallowprocesscontrolandimprovementatearly process stages, innovative inspection technologies need to be developedanddistributedintheupstreamrotorassemblyprocess stages.Therotoriscomposedofasetofmagnetsassembledonthe surfaceofmultiplelaminatedstacks.Thesestacksarethenaxially assembledtoproducetherotor.Thenumberofassembledstacks determinesthespecificproducttype.

Knowing the effect of stack magnetization on the rotor magneticintensityand,ultimately,onthefinalenginetorqueis a majorissue in this manufacturing process. Thiswould allow characterizingthecorrelationbetweenproductionstages, conse-quentlycontrollingtheupstreamstackassemblystrategytoobtain thedesiredenginequalitylevels.Anotherchallengeistodetermine ad-hoc assembly strategiesthat canpreventthepropagation of defects from the early stages to the final assembly stage. In MuProD, one of the considered options exploits the quality correlationbetween thestagewhere thestacks aremagnetized andthestagewheretherotorisassembled.Adefectivestackcanbe turnedintoagoodqualityrotoriftheassemblyangleissuitably compensatedatthedownstreamstage.

The secondconsidered solution isto integrateselective and adaptiveassemblystrategiesintherotorassemblysystem[126]. Selectiveassemblyentailson-linepartinspection,clusteringparts intobinsaccordingtospecifickeyqualitycharacteristicvaluesand subsequentmatchingonlyfromcoupledclassesaccordingtosome predeterminedmatchingcriterion.Thisapproachmakespossible tochange a productquality problem intoa system design and operationproblem.Inthecaseofrotorassemblytheintroduction ofselectiveassemblycanincreaseproductionqualitysignificantly byreducingscrapandincrementingtheyieldofthesystem.

Thethirdexamplereferstothemanufacturingofsmall-lotlarge parts (i.e. planet carriers)for windmill gearboxes in the wind power sector at Gamesa (Fig. 2(b)) [200]. The continuously increasingdemandforenergyisleadingtothemanufacturingof eolictowersthatareabletoproducemorepower.Thesetowers demandlargercomponentsandrequirenewandlightermaterials for easierassembly. The machiningof componentssuchasthe planetcarrieriscritical,sinceverysmallproductfeatureshaveto bemachinedatverytighttolerancerequirements[93](normally tenthofmicronsondimensionalandgeometricalfeatures)onvery largeparts(outerdiametersupto2500mm,weightupto7000kg). Theproductionsystemadoptedinthereferencecaseiscomposed ofparallelmachiningcentresdedicatedtoroughingandfinishing operations.Thecausesofdefectsarerelatedtotheinputcasted material,partdeformationduetofixturing,toolwear,vibrations, etc. In order to achieve such highly demandingmanufacturing goals,thecompanymakesuseofahybridinspectionprocedure. ThefirstpartofthelotisextensivelymeasuredattheCMMfor compensating possible deteriorations by machining parameter adjustment.Then,thelotproductionisstarted.Foreachprocessed feature, extensive in-process part verification is carried out to

Fig.1. Procedureadoptedbythecompanyformanagingtheramp-upofnew technologiesinaplantassemblingcarbatteries.

Fig. 2.Electric drive producedat Robert BoschGmbH(a)andPlanetCarrier producedatGamesa(b).

avoidthegenerationofanypossibletypeofdefect,duetothehigh value of casting parts. However, these continuous machining, measuringandadjustmentloopsinterferewiththecycletimeand theproductivityoftheplant.Therefore,thisisanexamplewhere thesolution adopted by the company is strongly polarized on qualityperformance,withnegativeconsequencesforproduction logistics performance. This approach is also evident in other sectors,suchastheproductionofcriticalmechanicalcomponents, i.e.engines,intheaeronauticindustry.

A fourthexampleisrelatedtotheproductionofcustomized micro-intravascularcathetersashighvaluemedicalproductsfor theageingsocietyinthemedicaltechnologysector,asatENKIS.r.l inItaly(Fig.3).

Medicaltechnologyisoneofthemostinnovativeindustryin theworld, withan increase of 15% in turnover since 2005.As opposedtothepreviouscase,this exampleshowsacompletely differentmanufacturingcontextthatisrelatedto micro-machin-ingandmicro-formingoperationsandone-of-a-kindcustomized products.Thesetypesofcathetershaveapplicationsinoncology, angiology,angiography andangioplasty, where thedemandfor customizedsingle-useproductsisincreasingtosolvesterilization problems and to reduce the risk of contamination inherent in multiple-use products. Over the last years, a trend towards miniaturization of these devices is in place. The successful achievementofthisgoalwillfacilitatetheuseofthesecatheters insmallerarteries,thushavingagreatimpactonthenumberof curablediseasesandultimatelysavinglives.

These micro-cathetersarecomposed mainlyofa micro-tube and,dependingonthespecificapplication,aninjectionmoulded partthatmakesitpossibletocarryoutthesurgery.The micro-tubescanbeeithersinglelumenormulti-lumenfacilitatingthe transportationofmultiplesubstancestothezoneofinterest,as wellassinglelayerormulti-layerforhigh-pressureresistance.The manufacturing process is composed of four main phases: (i) materialcompoundpreparationandcontrol,(ii)micro-machining of the extrusion die (micro-milling, micro-Electrical Discharge Machining–EDM)foreachspecificparttype,(iii)micro-extrusion of themicro-tubes and (iv) finalmicro-catheter assembly. The majorcausesofdefectsarerelatedtodefectsindieproductionthat cause defects in the micro-tubes and geometrical defects generatedwithinthemicro-extrusionprocess.

Theabovedefectsleadtoanextremelyhighdefectrates(upto 70%instandardproduction).Thesedefectsarecertifiedmainlyby 100% micro-tube inspection at the end of the line, which is manually driven and expensive. This high defect rate also underminesthepossibilityofrobust productionschedulingand istranslateddirectlyintoservicelevelissues.Moreover,thishuge amountofgeneratedscrapresultsinamassivewasteflow,which is an additional cost for the company that must pay for its treatment.Thisexampleshowshowinthecontextofhighprocess variability,poorcontrollabilityandautomaticinspection,aswell asrelatively low material value,thecompany strategy maybe

highly polarized on productivity performance, thus penalizing processcontrolandfirst-time-rightqualitystrategies.

Another example is related to therecently designed engine block production line at Scania CV AB, Sweden. Scania is a worldwide manufacturer of trucks and buses. All Swedish productionwasrecentlymovedtoSodertalje.Aserialproduction linelayoutwithmultipleparallelmachiningprocessesperstage hasbeen designed withthegoal of producing differentengine blocktypesinthesamesystemataveryhighproductionrate.Asa result,some400productpathsarepossiblewhileconsideringall possibleroutingalternativesinthesystem.Theadoptionofparallel processesincreasesthereliabilityofthesystem,thusmakingit possibletoreachincreasedproductivitytargets.Nevertheless,this poses additional challenges regarding quality control and part deviationverificationwithrespecttoserialsystemlayouts.Indeed, multipleproductpathsgenerateamixingeffect,lossofprocess signatureandlossofFIFOrules,thusreducingtraceabilityinthe system,i.e.theabilitytoconnectadefectwiththeprocessthat generatedit.Moreover,inthepresenceofend-of-lineinspections, longdelaysinqualityfeedbackaregenerated.Thisclearlyreduces theabilitytocloseareactivequalitycontrolloopbutincreasesthe totalproductionrateofthesystem.Therefore,inordertoincrease thevisibilityofqualityandprocessdeviations,in-lineinspection pointsneedtobedistributedthatwillhaveapositiveimpacton qualityandanegativeimpactonproductionlogisticsperformance. This example proves that manufacturing system design affects product quality and that product inspection design affects the productionlogisticsperformanceofthesystem.

Theassessmentofcustomerperceptionofproducts in multi-stage manufacturing systems is one of the main challenges of productionandqualityengineeringandthemaintopicoftheBMW Group case study [244,248]. This real case study is also demonstration scenario of the Cluster of Excellence ‘Integrative ProductionTechnologyforHigh-WageCountries’[240].Thevehicle acousticsisa productfeaturethatisimportanttothecustomer perception of the product quality. It has very complex and multifaceted mechanisms that generate structure-borne sound, whichisthentransferredtotheinteriorofthevehicleviathecar body.Whenthenoisereachesaparticularlevelinsidethecar,it may be perceivedby thecustomer as annoying. The technical analysisshowsthattherearaxledrivehasapronouncedeffecton theacousticswithinthevehicle(seeFig. 4).

Inordertoensurethiscustomerrequirement,advancedtools forinspectionplanningandqualitycontrolmethodsinmulti-stage productionsystemsarerequired.Themanufacturingofrearaxle drives,ischaracterizedbymanyvariants,whichareproducedat

Fig.3.Multi-lumenandmulti-layercathetersformedicalapplications.

Fig.4.Listofrequirementsonthehypoidgearsetsforpassengervehicleaxledrives ofstandarddesign.

medium lot sizes (10,000 units). Different manufacturing processesareusedfordifferentvariants.Thefrequentchangein variantsleadstoahighplanningeffortforthenecessaryadaptation of the manufacturing processes. The analysed process chain is shown in Fig. 5. The main challenges in this multi-stage productionlineare:

 measurementoftheacousticperceptionofthecustomer;  parameterization of the customer specification for in-line

processinspection;

 challengingassemblyconceptfromFIFOproductiontowardsa toleranceoptimizedassemblyconcept.

In order to close the loop between customer perception, inspection planning and quality control, first of all perceived quality methods for the measurement of customer perception regardingtheacousticsoftherearaxleareneeded.Byidentifying therelevantprocessparametersinadynamicin-lineinspection theacousticbehaviourcanbethenforecasted.Evenwhen inline-inspectionsareinstalledandallprocessesarestabledeviationsin the multi-stage production system can cause critical acoustic behaviour after the differential assembly. Hence, advanced tolerance-oriented part matching tools could forecast the acousticfitofthegearwheelandringgearpair,assembledto thegearset.Thismeansbothexpandinganyunnecessarilytight tolerancestosavecostsanddefiningthecriticaltolerancesmore preciselytoensurethedesiredfunctionalityoftheendproduct. Inordertoreducescrapratestheconceptofjust-in-sequence productionhastobeextendedtoatolerance-oriented produc-tioncontrolprinciple,whereproductionandqualitycontrolare integrated.

Theproductionqualityparadigmin contextscharacterizedby deterioratingproducts,suchasfreshfoodoryoghurt,isaddressed next. Food production is pervaded by strict requirements on hygiene and delivery precision. The production plants have to quickly respond to the market demands and current order situation. A typical production sequence for yoghurt includes mixing/standardizingofmilk,pasteurization,fermentation, cool-ing, addition of fruit additives and packaging. The production planninginvolves verycomplexproblemsduetothemaximum allowed storagetime before packaging.Ifthe timethe product flows in the system exceeds this limit, it has to be scrapped. Changeovers are typically sequence-dependent (increasing fat levelispreferred givingshorterset-ups)and upto25 products variants may be produced in the same system, with different processingtimes.Intheseplantstheprimaryobjectiveistocontrol the production of the different products for reducing the changeover number (typically 100/week) and time (typically 20h/week).Secondly,theobjectiveisthereductionoftheproduct scraps(typicallyaround10%)duetoobsolescenceofinventoryby achieving a better synchronization of the process phases, an effectivejointcontrolofthetanksizes(buffers)andtheproduct quality.TheremovalofthisbottlenecksandthereductionofWIPis apriorityfortheseindustries.Therefore,inventorymanagement and line balancing play a fundamental role in achieving the productionqualitytargets.

Inautomotivepaintshops[120],toensurehighpaintquality, multiple inspection stations are usually allocated along the cleaningandpaintingprocesses.Vehiclesfailedininspectionwill be repaired or repainted before moving to the next station. Therefore,toimprovetheperformanceofpaintshops,reducing qualityfailurerateswhilekeepingtheproductionratewithinthe target is of significant importance. An automotive paint shop typically consists of thefollowing major processes.In the pre-treatmentsection,eachvehiclebodyissubmergedinaphosphate liquidtogetalayerofcoatonthesurfaceofthesteel.Inthenext ELPOprocess(electrocoatpaintoperation)thebodyofthevehicle iscoveredwitha specialsubstancetoprotectitfromcorrosion. Then,thebodyneedstobeheatedandbakedintheELPOoven,and finallysandedtofinishtheELPOprocess.Afterwards,thereisan intermediatestagewherethepre-treatmentqualityisinspected. Thevehicleisthen movedtothesandsectionfollowedby seal inspection.Thenextisthepaintingsectionthatstartsbyspraying primeronthevehicle,whichimprovestheadherenceofthepaint to the vehicle body. Afterwards, base coat and clear coat are performed.Then,thebodyofthevehicleneedstobebaked.After thisprocess,thefinalinspection(finesse)and,incaseadefectis detected,therepairprocessesareapplied.Here,defects,suchas scratch,dirt,dentwillbeidentifiedandfixed.Afterrepair,these vehiclesaresenttothenextoperations.Inautomotivepaintshops, imperfect dirtcleaningintheupstreamsandingoperationswill result in more paint defects in downstream colour coatings. Therefore,thestagecorrelationandthemanagementofdefects throughpartre-processingarethemainissuestobeaddressedat systemlevel.

Production quality is of significant importance also in the semiconductor industryand,specifically,inwaferfabrication.A semiconductor manufacturing process has the following char-acteristics.Theproductionisperformedthroughmultiplestages. Someofthesestagesworkinbatches,includingtheslicingprocess, lapping,andpolishing.Multipleparallelprocessorsarecommonly adoptedtoachievetherequiredproductionrate.Eachproductmay undergoseveralre-entryloopsinthesystem.Theproductionyield isgenerallyverylow(around50%)andtherequirementson due-dateperformanceareverystrict.Theflowtimeisextremelyhigh thusminingthereactivenessofthequalitycontrolsystem.High priority lots typically share the production resourceswith low priority lots,thusgenerating non-FIFOproductionsequences.In this context, thecomplexity is themajorbarrier forproduction quality.

1.3. Theproductionqualityparadigm

Intheliteratureaswellasintheindustrialpracticethereare manydifferentKeyPerformanceIndicators(KPIs),orperformance measures,thatindividuallyrelatetoquality,productionlogistics andmaintenance.Inthefollowing,themostwidelyadoptedKPIs atsystemlevelareconsidered.Inmanufacturingsystemstheyare complexnon-linearfunctionsofsingleprocessorsinglestageKPIs. TypicalsystemlevelproductionlogisticsKPIsinclude:

 Theproductionrate,i.e.thenumberofpartsproducedinagiven time(alsocalledthroughput).Itisusuallymeasuredintermsof JobsPerHour(JPH).

 Thetotalinventory,orWIP,i.e.thetotalamountofpartsflowing inasystem.

 Theflowtime,i.e.thetimerequiredforpartstocrossthesystem.  The interdeparture time, i.e. the time between consecutive

deliveriesofoutputproducts.

These performance measurescan be considered in the long termorintheshortterm.Moreover,thefirstmoment(mean)or highermomentsofthesemeasurescanbetakenintoaccount.The considerationofhighermomentsintheshorttermcanbeused,for instance, to evaluate the so-called due-date performance. For example,theservicelevelofasystem,whichistheprobabilityof

Fig.5.TheBMWproductionprocess,characterizedbymanysensitivetolerances withcomplexdependencies.

deliveringalotofacertainsizeXbeforeitsfixeddeadlineT,isa due-dateperformance.Froma ‘‘quality-oriented’’pointof view, typicalKPIsofinterestare:

 The system yield, or quality buy rate, i.e. the number of conformingpartsdeliveredbythesystemoverthenumberof conformingpartsgoingintothatsystem,ina specifiedperiod oftime.Incaseof100%conforminginputflow,itissimplythe fractionofgoodpartsdeliveredbyasystem.

 Thefirst-timequality,orfirst-timerightrate,orfirst-passyield, i.e.thegoodjobratioofallthefirst-timeprocessedjobs.  Thedefectrate,i.e.thefractionofnon-conformingjobsdelivered

bythesystem.

Fromamaintenancepointofview,typicalsystemKPIsinclude:  Thesystemavailability,i.e.thetimeasystemiscapableofbeing

operationalinagiventotaltime.

Thisanalysisshowsafundamentallackofacleartaxonomyfor integratedquality,productionlogisticsandmaintenance perfor-mancemeasures[122,141].Anattempttowardstheformalization ofataxonomyhasbeenrecentlyproposedin[239].Moreover,the TotalQualityManagement(TQM)and TotalProductive Mainte-nance(TPM)paradigmshaveproposedintegratedKPIstoevaluate theeffectivenessoftheimplementationofaspecificimprovement planinindustrialcontexts.AlthoughTQMandTPMsharealotof similarities,areinfactconsideredastwodifferentapproachesin the literature. TQM attempts to increase the quality of goods, services and concomitant customer satisfaction by raising awareness of quality concerns across the organization. Total Productive Maintenance (TPM) is a systemof maintaining and improvingtheintegrityofproductionmachinesthataddbusiness valuetotheorganization.Thesemethodologiessuggestthatthe mostrelevantintegratedperformancemeasureis:

 theeffectivethroughput,orthenetthroughput,alsocalledOEE (Overall Equipment Effectiveness), that is the number of conformingpartsproducedbythesysteminagiventime.

Grounding on this background knowledge, the production qualityparadigmcanbeformulatedinthefollowingterms:

Production quality is the discipline that combines quality, production logistics, and maintenance methods and tools to maintainthethroughputandtheservicelevelofconformingparts undercontrolandtoimprovethemovertime,withminimalwaste ofresourcesandmaterials.

2. Quality,production,andmaintenance‘‘InteractionModel’’ Severalempiricalstudieshavediscussedtheinteractionamong quality,productionlogisticsand maintenance inmanufacturing systems.Forexample,in[27]asurveyapproachisusedtoidentify potentialcorrelationsbetweentheapplicationofJITandTQMlean practicesintheautomotiveand electronicindustries. Themain result of this analysis is that those companies that are more successfulinlimitingtheirinventoryandinbetterorganizingtheir production through JIT policies also achieve better quality performanceandapplymoreeffectivedefectreductionprograms. This positive correlation highlights the need for a deeper understandingoftheinteractiondynamicsbetweentheserelevant aspectsinmanufacturing.

2.1. The‘‘InteractionModel’’

The complex dynamics of the interactions among quality, productionlogisticsandmaintenancerequiresconsiderableeffort to be modelled and understood. This activity is important to identifyandexplainthemanyexistingtrade-offs.Theliterature includesmodelsdevelopedtocaptureandexplainthedynamicsof

this interaction. Among these, a very powerful set of tools is businessandsystemdynamicsCausalLoopDiagrams(CLD).These tools have been proposed for modelling complex interactions between quantitativeand qualitative variables in a number of complexbusinessmanagementproblems.Theirapplicationtothe analysis of the interactions between quality, maintenance and productivityperformanceindicatorsisreportedin[117,230,270]. The main goal of these models is to identify all possible interactions among variables and decisionsin order tosupport the definitionand implementation of continuous improvement programsthat donot failtomeet thegoalsdue tounexpected interactions.Understandingtherelevantinteractionsthenmakes itpossibletoavoidlocalimprovementsthatdeterioratetheglobal performanceduetoneglectedimpacts.CLDchartsarediagramsin whichtherelevantvariablesofaproblemarelistedandconnected bydirectedarrows.Inthisformat,ApointingatBwithapositive arrowmeansthat,giventhateverythingelseisfixed,anincreaseof variableAcausesBtoincreasemorethanitwould normally.A pointing at B with a negative arrow means that, given that everything else is fixed, an increase of variable A causes B to decrease more than it would normally. Only direct and easily explicablecause-effectconnectionshavetobereported.CLDare verypowerfultoolsforfindingexistingcontrolloopsincomplex, multidisciplinary and dynamic contexts and in making them explicit.

Althoughtheyhavebeenwidelyusedforconsultingactivities andforpolicymaking,state-of-the-artCLDmodelsdonotfocuson theproductionsystemdesignandoperationallevelsbutrathertryto seetheproblemmoregenerallyfromamanagerialpointofview.For example,the implicationsof continuousimprovement programs thattakeintoconsiderationworkermotivation,learningcyclesand company emphasis achieving performance targets, have been investigated.Oneofthehighlightedloopsisthefollowing:More defectsreducenetprocessthroughput(effectivethroughputinour notation).Thisinturnincreasestheactualversustargetthroughput gap. This negative performance increases worker effort, which positivelyaffectsgrossprocessthroughput(totalthroughput).This in turn has a positive impact on net process throughput. This reinforcementloopiscalledthe‘‘WorkHarder’’loop.

Althoughthisdynamicsplaysa relevantrolein the achieve-mentofsatisfactoryproductionqualityperformance,thegoalofthis keynotepaperistoconsidermanufacturingandshopfloorrelated aspects.Therefore, anewinteractionmodel isneededwiththe specificgoalofansweringthefollowingquestion:‘‘Whatarethe cause-effectrelationsexplainingthemutualinteractionsamong quality,maintenance and productionlogisticsin manufacturing systems?’’Basedonthereal-lifeexamplesprovidedintheprevious section, in this paper an interaction model is developed and proposed.Themainobjectiveoftheproposedmodelistodefine and characterize all major sources of interactions affecting productionqualityattheshopfloorlevel.Theseinteractions are consistent with the Functional Enterprise-Control Model as proposedbytheIEC/ISO62264standard[111].

TheaggregatedrepresentationofthemodelisreportedinFig.6. Thissimplegraphshowsthatbi-directionalmutualcause-effect relationscanbefoundamongquality,maintenanceandproduction logistics. A more comprehensive definition of these links in manufacturing systems is provided in the detailed CLD model

depictedinFig.7.Thered,blueandgreenregionsrefertovariables relatedtoproductionlogistics,qualitycontrolandmaintenance, respectively. The links of greatest interest in this paper are representedbythosearrowsthatcrossregionsofdifferentcolours.

Forexample,followingthearrowsinFig.7,increasingtheWIP, inturnincreasesboththetotalthroughputandthelead-timeofthe system.Theincreaseinlead-time,however,alsocauses quality-related phenomena. Indeed, it reduces theobservability of the critical product quality feature in the system, which in turn reducestheabilitytodetectapotentiallygenerateddefectwithina shorttime.Thisdirectlytranslatesintothepropagationofmore defectsbetweentheprocessingstagesinthesystemandawasteof productioncapacityinprocessingpartsthatarealreadydefective. Alossofcapacityleadstoalossoftotalthroughput.Thisexample ofbalancingloophighlightstheimportanceofthisapproach.Ifthe effectsoftheWIPincreaseontheproductqualitywereoverlooked, qualityandproductionlogisticswouldbetreatedinisolationand erroneousdesignandmanagementdecisionscouldeasilybemade. Focusingonmaintenanceandproductionlogisticsinteractions, asecondlinkmadeexplicitbythediagraminFig.7isexplainedin the following. Equipment condition-based preventive mainte-nanceistypicallysupportedbysensorialdatacollectedfromthe fieldwhiletheequipmentisoperational.Ifthesedataareproperly analysed, they can be used to make inferences about the degradation stateof theequipment. If the samplingfrequency ofthis monitoringsystemisincreased, theabilitytodetectthe equipmentdegradationstateincreases.Thisincreasesthechances thatanundesireddegradationstatewillbecorrectlyidentifiedand preventivemaintenancepracticeswillbepromptlyactivated,thus increasing equipment reliability, decreasing the frequency of unexpectedrandomfailuresandultimatelydecreasingcorrective maintenance interventions. This maintenance-related loop also has an impact on system logistics. Indeed, less corrective maintenance generates less unplanned equipment downtime, whilemorepreventivemaintenancecausesanincreaseinplanned equipment maintenance interventions. Theses both affect the productionrate ofthesystem.If this interactionis overlooked, overallmyopicdecisionscanbetaken.

Inthefollowing,therelevantphenomenacharacterizingtwoof therealcasesinvestigatedin Section1.2 areframedwithinthe ‘‘Interactionmodel’’.Thesefewexamplesshowthatrealcasescan bemappedwithinthis‘‘InteractionModel’’.Withreferencetothe Gamesacase,thehighworkpieceinspectioneffortleadstohigh observabilityoftheproductqualitycharacteristics.Thispositively affectstheabilitytodetectandcorrectdefectsassoonastheyare generated.Thisisbeneficialforthesystemyieldbutdetrimental forthetotalthroughput,astheproductionresourcesareusedto re-processpartsandcorrectdefects.Thehighinspectioneffortalso

leads to high inspection time and, consequently, low total throughput. With referenceto theENKI case, thewide mix of personalizedpartsandthesmalllotsizesleadtoextensiveset-ups. Set-upsaredetrimentalforthetotalthroughputandforprocess deviations, thus leading to consistent generationof defects.In addition, the poor process data gathering undermines the possibility to observe the equipment degradation state, thus makingpreventivemaintenancehardtobeimplemented.This,in turn,leadstoshortplannedequipmentdowntimes,thushightotal throughput, butalsotoconsistentprocessdeviations,thushigh defectgenerationandlowyield.

Asthesefewexamplesshow,theproposed‘‘InteractionModel’’ canbeusedbyscientiststoidentifyrelevantunexploredproblems thatneedtobefurtherinvestigated,aswellasbypractitionersto motivateandgatherinsightsonunexplainedphenomenaonthe boundariesofthesethreeareas.Inthispaper,thismodelwillbe usedasareferenceframeworktostructurallyexplorethetopics already addressed in the literature and to highlight promising researchareasforthefuture.

2.2. Usingthe‘‘InteractionModel’’toclassifythescientificliterature About300papers,mostlyfromleadinginternationaljournals, havebeenclassifiedandframedwithinthe‘‘InteractionModel’’ (Fig.8).Specifically,thepapershavebeenclusteredaccordingto two-dimensionalaxes.Thefirstaxisrelatestothespecific‘‘phase’’ inthepaperwheretheinteractionisaddressed.Morespecifically, thedesignandplanningphaseandtheoperational,controlandthe managementphasehavebeentakenintoaccount.Thesecondaxis relates to the type of interaction addressed. According to the proposed interaction model, possible interactions are quality-productionlogisticsinteraction,productionlogistics-maintenance interaction, quality-maintenanceinteractions, and complete in-teractionamongquality,productionlogisticsandmaintenance.In Fig. 8, the bullets represent the cluster of papers addressing common problemsasframedwithintheinteractionmodel.The sizeofthebulletrepresentsthepopulationmagnitudeofthecluster. Mostof the contributions coverareasrelated tothe interaction among quality and production logistics, while only a few contributionsaddressproblemsunderafullyintegratedview.

3. Problemsandmethodologies 3.1. Designandplanningphase

Inthefollowingsections,theexistingliteratureaddressingthe linksbetweenquality,productivityandmaintenanceinthedesign phase are revised. The focus is mainly on systemand process related design decisions while product and tolerance design decisions,inspiteofbeingimportantfactorswithintheproduction quality target, are not explicitly considered in order to avoid deviationsfromthemainscopeonthispaper.Forarecentreview onthelinkbetweenproductdesignandqualitysee[180].

Fig.7.Detailsofthe‘‘InteractionModel’’developedinthispaper.

3.1.1. Impactofmanufacturingsystemdesignonquality

Therearemanyaspectsthatdemonstratethattheproduction systemarchitectureaffectstheproductionqualityperformance,as highlightedin[109]basedontheanalysiscarriedoutinGeneral Motors.Thisreviewhasbeenupdatedin[107],whererecentworks focusingonthisspecificlinkhavebeenframed.

Inmassproductionsystemsthesimulationworksproposedin

[25,165]contributedtotheassessmentofafundamentalprinciple:

whilechangingthesystemarchitecturethefractionofconforming productsmaydrasticallydrop.Asimilarresultwasachievedby

[136]. The authors compared six alternative configurations,

includingserialandparallel lines,andhybridconfigurations, in terms ofmultiple performancemeasures, includingthe system capability to produce parts with limited variations and the expectedavailability.Theauthorsshowthatseriallinesperform betterthan parallel lines in terms of dimensional variation of products,becausethereisonlyonepossiblepathinthepartflow and themixing effect is avoided. The mixing effectmeans that multipleprocessing stages show differentdegradationpatterns and actual capabilities and this phenomenon increases the variabilityinthekeyqualitycharacteristicsoftheoutputproducts. Themixingeffectinparallelmachinelineshasbeenfurtherstudied

in[233].Theauthorsanalysedbysimulationtheconsequenceof

themixingeffectontheabilityofperformingarootcauseanalysis attheinspectionpointsinthesystem.Otherundesired phenome-na, such as possible job order loss and sampling frequency mismatch, have also been identified in parallel processes. In

[38,94,189]itwasshownthatU-shapedlinesmayperformbetter

thanseriallinesin termsofquality ofthereleasedoutput.The reasonisthattheoperatorsassistingthelinecanvisuallydetect qualityproblems in thesystemearlierthan in seriallines and, consequently,canreactmorepromptlytothesedefects.

The impact of buffers on production quality has also been analysedintheliterature.TheLeanProductionareahasshownthat the reduction of inventory has a positive impact on product quality, since quality defects are identified earlier and are not propagatedthroughoutthesystemstages[310].Asamatteroffact, ToyotaProductionSystem(TPS)advocatesseein-processstocksas waste(muda),whichoftenhidesproductionproblems.However, fromtheManufacturingSystemEngineeringareaitisknownthat theproduction rateof the systemis positively affected by the presence of buffers, since they decouple the behaviour of the unreliable machines [65]. This trade-off has been studied analyticallyin[52]and[131]fromanintegratedquality-logistics pointof view. The authorsfound casesin which theeffective throughput is maximized for a given buffer capacity. This behaviourisduetothe couplingof twocontrastingeffects, in thepresenceofremoteorubiquitousinspections,whereaproduct featuremanufacturedatacertainprocessingstageisinspectedat amonitoringstationlocatedfurtherdownstreamintheline.One effectisthepositive impactofthebuffercapacityonthetotal throughputofthesystem.Theothereffectisduetothedelayofthe quality information feedback when remote inspection is per-formed. Processed parts do not instantaneously reach the inspectionpoint,butarestoredin theinventoryqueuebefore beingmeasured.Largebuffersbetweenthemonitoredstationand theinspectionpointincreasethetimepartsspendinthisportion ofsystem. Thiscauses longreactiontimein identifyingout of controlconditionsanddecreasesthesystemyield.Thisbehaviour generatesinterestingconsiderationsonthejointdesignofbuffers andqualitycontrolparametersinmanufacturinglines[51].Nada

etal.[201]developedacomprehensiveframeworktoaddressthe

aforementionedissueduringthedesignphaseofmanufacturing systems.AConfiguratorCapabilityIndicator(CCI)isdeveloped,by using hierarchical fuzzy inference, to select the most proper architecturalparametersofthesystem,underproductionquality considerations.

Thedesignofin-processbuffershasarelevantimpactonthe productqualityalso inthose industriesproducing perishableor deterioratingcomponentsandproducts.Thequalitycharacteristics

of perishable products deteriorate over time. For example, as commentedinSection1.2,infoodindustrythereisamaximum storagetimebeforepackaging.Theproducthastobescrappedif thetimespentinthesystemoverpassesacertainfixedlimit.This problemhasbeenaddressedin[169].Aprojecttodetermine cost-efficient waysof speeding upthe croissant processinglines of ChipitaInternationalInc.isreported.Theinstallationofaproperly sized in-process bufferled toa reduction in failure impact on productqualityandanincreaseofthesystemefficiency.In[168] the authors focused on the production rate of asynchronous productionlinesinwhichmachinesaresubjecttofailures.Ifthe failureofamachineislongenough,thematerialunderprocessing in theupstreammachinesmust bescrappedby thesystem.In

[295]a transientanalysis isproposed todesign thesizeof the

buffersneededindairyfillingandpackaginglines.Thedistribution oftheflowtimeinunreliablemulti-stagemanufacturingsystems wasevaluatedin [260]. Thismethodcansupportthedesign of buffersforachievingacertainacceptedscraprateinperishable good production. In [272] an inventory model for perishable products with randomperishability and alternatingproduction rate is proposed. As shown in these works, buffers should be designed by using an integrated production quality oriented approach.

Inmachiningandassemblyoperationsithasbeenshownthat thedesign decisions concerningthe systemoperatingspeed are strongly correlatedtothe productquality [214]. Improvingthe machines’processing ratehasa positive impact on thesystem productionrate,butmaynegativelyaffectsthesystemyield.For example,inroboticassemblythequalityoftheproductionprocess isrelatedtotherobotrepeatabilityandtheoutputrateisrelatedto therobotspeed. Robotrepeatabilitydeteriorateswiththerobot speed[129].Thisbehaviourhasbeeninvestigatedin[186].The authors modelled multi-stage systems with quality-quantity couplingmachines.In these machines,thecorrelationbetween efficiencyandyieldismadeexplicitthroughananalyticrelation. Themethodsupportsthedesignoftheoptimalprocessingspeedof themachinesinthesystemandhasbeenappliedtoanautomotive casestudy[12].

The link between mix flexibility and quality in flexible machiningsystemshasalsoreceivedattentionintheliterature. A taxonomy for flexible manufacturing systems is proposed in

[278].Aflexibilityevaluationtoolbox inmodernmanufacturing

systemsisaddressedin[91].Moreover,amethodforassessingthe flexibility of a manufacturing system, in an uncertain market environment,underlifecycle considerationsis developedin[7]. Partmixflexibilityprovidestoasystemtheabilityofprocessing different part types with relatively limited set-up times and changeovercosts.Thelevelofflexibilityofa systemaffectsthe product quality [161]. There are few examples showing that system flexibility is positively correlated with product quality

[232].In[303]theauthorarguesthatflexiblemodularassembly

systems support the achievement of higher product quality. However, increased flexibility can also deteriorate quality. For example,consideraflexibleautomotivepaintshop[298].When shiftingbetweendifferentpartbatchescharacterizedbydifferent colours a certain amount of defective parts that need to be reworkedisproducedinthetransientperiodsincethecolouris contaminated by the one used for the previous batch. This phenomenonisclearlystronglyaffectedbydecisionsconcerning theset-uptimesandthejobsequencing(seeSection3.2.5).

The effect of the design of reconfigurable systems on the production logistics and quality performance has alsoreceived attentioninthepast.[23,101]presentedanapproachfordesigning system reconfiguration options according to a multi-criteria decision making framework (Fig. 9). Starting fromthe analysis of the product feature and demand requirements, and from a database of available equipment modules a system-level tool generates different potentialreconfiguration alternatives. Their KPIsareevaluatedwithinasimulationenvironmentanddominant solutions areselected.ElMaraghy andMeselhy [80] proposeda

framework to study the relation between maintainability and qualityinchangeablemanufacturingsystems.

The impact of complexity in manufacturing and assembly systemsonseveralperformancemeasuresincludingqualityand production logistics metrics has been revised in [81,144]. Moreover, in [103] the impact of different plant complexity sourcesonproductqualitywasinvestigatedbasedontheanalysis ofrealdatafromanautomotivecompany.Theresultsprovethat thereexistsanegativecorrelationbetweenthenumberofchassis producedintheplantandquality.Althoughsystemcomplexityhas manydimensions,productvarietyseemstobethemostimportant factoraffectingproductionqualityperformance.Infact,thenumber ofproductvariantsdecreasestheabilityoflearningfromrepetitive operationsandincreasestheprobabilityofhumanerrors.Thelink between assembly system design for product variety and performance was explored in [102]. Many papers address the issue of quality and human induced errors in mixed-model assembly systems [325]. Mixed-model assembly systems were recognized as enablers for mass customization manufacturing. However, highly proactive and knowledgeable workforce is neededtoeffectivelyimplementmixed-modelsystemsin indus-try.In [2] thequality and productivityperformance of mixed-modelassemblysystemsunderhumanerrorswasevaluated.In

[271]itwasreportedthatabout20%ofthedefectsintheFujiXerox

Chinaphotocopy machine assembly systems was connected to operatorserrors.Itwasthesecondcausefordefectsintheanalysed plant. Product and process related complexity metrics were proposed to tackle this problem. The link between complexity and performance measuresin mixed-model assembly was also algebraicallyanalysedin[1].

3.1.2. Impactofprocessplanningonquality

Manufacturingprocessplanningisamongthemostknowledge intensive decision-making activitiesundertaken in factories. In thisactivitytheproductinformationismappedontotheavailable informationforthevariousexistingmanufacturingresourcesto determineaplanofactiontoconverttherawmaterialintothefinal product.Process planningis normally carried outby a specific humanresourceanddependsonindividualexperience. Method-ologiessupportingprocess plannershavebeendeeply analysed sincethe80s[42]and,overthelast10years,asignificantnumber ofsoftwaretools focusingon Computer-AidedProcessPlanning (CAPP)approacheshavebeendeveloped[293].CAPPsystemsuse, among others, artificialintelligence methods to enable human operators to select the most appropriate operations for manufacturing.Currentlytheknowledgeusedwithinthisactivity hasbeenbasedon nominalmodelsof manufacturingresources

[40,313]. While the nominal information pertaining to

manufacturingresourceisstaticanddoesnotchangeovertime, the capabilities of physical resources do, due to wearing of mechanicalcomponentsandtools.Capabilityprofiling[203]isa methodforrecordingthesechangesinthevariouscapabilitiesof manufacturingresources.Withcapabilityprofilingtechniques,itis

possibletooptimizethegenerationoftheprocessplantodevelop solutions thatareappropriatefortheactualavailablehardware andsoftwareratherthanthenominalvalues.Capabilityprofiles are generatedby combiningthenominalresourcemodelswith actual values obtained from sensors on the shop floor and predictivemodels.

Inaproductionenvironmentofresourceswithmixedcapability profiles and varying reliability, both process planners and schedulerstendtogiveprioritytomachineswithmoreadvanced andunfailingservices.Allinall,thisresultsinanuneven,distorted loadoftheseresources:whiletheyarebusyallthetime,othersare idling. Thethroughput of suchso-calledflexible job shopscan, however,substantiallybeimprovedifproductsaremanufactured via alternative routings. Nonaka et al. [208] presents a CAPP methodthat,departingfromthegeometricproductmodelandthe description of machining resources, generates a portfolio of process plans withthe objective to maximizethe throughput. Themodelisopentoincludequalityrelatedconstraints,too.Next, efficient loadbalancing and operation sequencingmethods are applied toschedule flexible job shopsby using the alternative routings(i.e.processplans)forproducingthesameproduct.The methodthat maximizeda workshop’sthroughput provedtobe robust and applicable even in large-scale industrial scenarios

[209].Thegenerationofalternativeprocessplansisalsothemain

objectiveoftheNetworkPartProgram(NPP)approach[87]andof non-linear process planning, in general. Non-linear process planninggoesbeyondthestaticandstrictlysequentialnatureof traditional process plans that are often carried out without considering the manufacturing system information [139]. The idea of network part program is to delete non-technological constraintsfromamongtheoperations,transformingthe sequen-tial part program into a network of operations, each one characterizedbyasetofG-Minstructions.Forinstance,thepart programforthemachiningofapalletcanbeeasilybuiltand re-builtaccordingtotheworkpiecesthatarereallymountedonthe fixtureasaconsequenceofchangesindemandmixandquantity

[218].Incaseofunavailableresourcespartprogramscanbeeasily

adjustedandeventuallysplitondifferentmachinetools. AfirstattempttodevelopNPPonindustrialscalewasledinthe ItaliannationalprojectNetPP[21]wheretheapproachwaslimited totheproductionofpalletsmountingonesingleparttypeonone work area. Non-linear process planning able to support the configuration of multi-fixtures (pallet) with differentparts has beenlaterdeveloped[217]formanagingsmallbatchesandahigh numberofproductvariants.Currently,12installationsoftheNPP are available inEurope. A process planningapproach basedon network part program has been developed in the DEMAT EU project[64]foramanufacturingsystemcomposedofultra-light, eco-compatibleandenergyefficientmachinetools.Other Europe-anprojects,suchasENEPLAN[83],analyseandproposenon-linear processplanningtechniquesforhybridprocesses,suchasmilling, turningandlasercutting.Oneofthekeychallengeswhileapplying theNPPinindustryistheneedforspecificprocedurestoprovidea qualitycertificationoftheentireNetworkPartProgram, consider-ingalltheprocesspathalternatives,insteadofonlycertifyingone specificpartprogram,astypicallydonewiththetraditionalG-code partprograms.

3.1.3. Inspectionplanninginmulti-stagesystems

Inspectionplanningdealswiththedefinitionofthepartquality inspectionsintheproductionsystemandwiththedefinitionofthe multi-sensor system for process monitoring. Both technologies serve as data gathering systems to feed SQC, Statistical Process Control(SPC)andConditionBasedMaintenance(CBM)procedures withusefulinformationtoperformamachineandprocessstate diagnosis and the implementation of corrective or preventive actionstorestorein-controlmanufacturingsystembehaviour.This diagnosis-orientedstrategyfocusesonthenear-zerolevelofdefect generation. Here, thepartqualityinspectionplansfeed product quality assurance and the consequent activation of defect

managementstrategies,includingscrap,reworkandrepair.These strategies allow smoothing the defect propagation throughout processstages andtothefinalcustomer.A reviewofthemost advanced automatic inspection and process data gathering technologiesisprovidedinSection4.Theuseofthesetechnologies forcomplexproductvalidationisrevisedin[180].Whileproduct inspectionallocation techniques have been revised since 1980

[225],lessattentionhasbeengiventoprocesssensordistribution

strategies.Concerningpartinspectionplanning,twomajortasks havetobesolved:

 Inspectioncharacteristicsidentificationandanalysis.Thenecessary inspectioncharacteristicshavetobeidentifiedandanalysedat eachprocessstage.

 Inspection process conception and allocation. According to the identified and analysed inspection characteristics, inspection strategies have to be developed, which define the test procedures,casesandresourcesandaligntheinspectionsteps tothetestsequenceinmulti-stageproductionsystems.

Thefirsttaskisessentialfortheoverallsuccessofinspection planning,sinceallcharacteristicswhichareneglectedmightcause fataldamagestotools,personnel,productsorcustomersand,on thecontrary,unreasonableinspectionscauseinefficientteststeps and increased process complexity. A common consequence of wrongcharacteristicsidentificationintheplanningphaseis the occurrenceofNo-Fault-Foundfailuresmadevisibleduringtheuse phaseoftheproduct[179,220,221].Thesearein-tolerancefailures due to unexplored interactions during the process/inspection planning phases, performed without taking into consideration processcapabilityprofiles.Hence,[242]introducedtheconceptof perceivedquality,which providesmethodologiestoidentifyand measure customer demands, and add the requirements from different product stakeholders in order to develop a holistic product and system specification [11,243]. The risk assessed specificationsaretheinputforthesecondmajortaskofinspection planning,i.e.theinspectionstrategyplanningandexecution.This phaseentails:

 Determinationofthepointintimeofinspection(when?).  Determinationofthepropertechnologiesforinspection(how?).  Determinationoftheinspectionextend(howmuch?).

 Determinationoftheinspectionlocation(where?).  Determinationoftheinspectionpersonnel(who?).  Selectionoftheinspectionequipment(whereby?).

Although heavy interdependencies between the inspection planningstepsdoexist[246],scientificapproachesmainlyfocus on the optimization of single inspection planning tasks. These works mainly address the inspection extend problem, against economicalKPIsusingstatisticalmethods[82,125].VanVolsemet

al. [287] derives an algorithm for the cost-optimal inspection

extend,place,typeandamountofinspectionstations.Thelackofa holistic consideration of all inspection planning tasks was addressedbySchmittet al.[250]wherethemodelwasextended to calculate business cases based on the risk attitude of the inspectionmanagement.Withthisidea,theoptimalsolutionisthe onethatmaximizesthedecisionmaker’svalueofbenefit.

Concerningtheareaofsensorsallocationforprocess monitor-ingonlyfewrecentcontributionsareavailable.Inamanufacturing process,sensordistributioninvolvesthedeterminationof:(i)the workstationsatwhichtoplacethesensingdevices;(ii)thenumber of sensors required at individual stations; (iii) the location of sensorswithinindividualstations.Threemajortypesofproblems have been considered in the literature [178]: (i) for a given numberof sensors,find theoptimal sensor locations;(ii) find the minimal number of sensorsas well as the corresponding locations;and(iii)giventhedistributionofqsensors,whereto distribute additional s sensors. These formulations lead to a constrainednon-linearoptimizationproblem[71,108].Wangand

Nagarkar [302] proposed a two-level hierarchical approach to solveproblems(i)and(ii)simultaneously.

Part quality inspection and process sensor planning have a strong impact on the production logistics performance of the manufacturing system. Ithas beenshown by [238] that, for a productionlinewith15machines,theeffectiveproductionrateof thesystemifinspectionstationsarepoorlyallocatedcanbe15% lowerthantheonecorrespondingtoagoodallocationofthesame number of inspection stations. As investigated in [55] three fundamentalphenomenadeterminethiseffect.Firstly,ifacritical product feature is remotely monitored, a quality information feedback delay is generated. If dedicated inspectionstages are designed, i.e. each critical product quality characteristic is measured by a dedicated inspection device, local monitoring shouldbeadopted.However,inordertosaveequipmentcostsand toincrease the inspectionsystemlife-cycle, reconfigurableand flexible inspection technologies have been recently proposed

[15,137]whichareabletoadapttoandtomeasureasetofproduct

features.Inthiscase,remotemonitoringisinevitable.

Secondly,thepartinspectioninterfereswiththecycletimeof thesystem,whileprocessmonitoringactivitiestypically donot. Therefore, a more extensive product inspection provides more accurateinformationabouttheproductqualitybutdecreasesthe totalproductionrateofthesystem.Thirdly,asitwillbediscussed in Section 3.2.3, the implementation of defect management strategies affectsthesystemdynamics andits performance.An algorithmtoallocateinspectionstationsinordertomaximizethe throughput ofconforming parts,consideringtheeffectof these three phenomenaunderpredeterminedinspectiontechnologies andtasks,hasbeenproposedin[188].Moreovertheconceptof quality bottleneckin a system in addition to the traditionally investigatedproductivitybottleneckconcepthasbeenformulated

in[296].Aqualitybottleneckisastageinamulti-stagesystemthat

moreseverelyaffectsthesystemyield[185].Identifyingquality andproductivitybottlenecksisanimportantactivityfor prioritiz-ingsensorandpartqualityinspectiondistribution.However,more extensiveresearchoninspectionandsensorplanningfor produc-tionqualitytargetsshouldbedeveloped,jointlytakingintoaccount alltheaforementionedaspects.

3.1.4. Qualitycontrolplanninginmulti-stagesystems

In multi-stage systems the design of an effective and cost-efficient quality control strategy is of critical importance. For recentreviewsofqualitycontrolplanningmethodsinmulti-stage systems see [262,283]. The major challenges that have been tackledbyresearchersinthisareaincludemulti-stagevariational propagationmodellingforqualitycontrol,processmonitoring,and rootcauseidentificationformulti-stagesystems.Thefirstareawill be revised in Section 3.2.1. Concerning process monitoring for multi-stagesystems,SPCisthemaintechniqueusedinpracticefor quality and process monitoring. Control charts are the most commonly adopted tools. However, most conventional SPC techniquestreatthemulti-stagesystemasawholeandlackthe capabilitytodiscriminateamongchangesatdifferentstages[192]. Toovercomethisproblem, multivariatecontrolchartsbasedon principal components and partial least squares analyses seem attractivefor multi-stagesystems. Morerecently,somespecific SPC techniques have been developed to exploit the detailed structureofmulti-stagesystemstoachievehighdetectionpower anddiagnosticcapability.Forexample,anexponentialweighted moving average scheme has been proposed as a monitoring methodformulti-stagesystems[312,326].IntheSPCarea,aftera processchangeisdetected,thediagnosisofrootcausesisleftto human operators.Significant progress has been made towards intelligent root cause diagnostics. These methodologies can be roughlyclassifiedas(i)statistical-estimation-basedmethods[72] and(ii)pattern-matching-basedmethods[175].Bothmethodsare basedonmathematicalmodelsthatlinkthesystemerrorandthe systemqualitymeasurements.Asamatteroffact,themajorityof available SPC approaches tackle the quality control planning

problem by selecting the optimal control chart parameters (samplingfrequency,samplesizeandcontrollimits)withrespect to an economic objective function. Multiple criteria including production logistics and maintenance performance are usually neglected. According to [262] the complexity of multi-stage systems requires a holistic system-level approach for effective qualitycontrol.Byintermeshingandlinkingclosed-loopquality controlsystems atvarious levelsofthecompanyunambiguous rulesfordecisionsatengineeringandorganizationallevelsemerge (Fig. 10).

The continuous alignment between actual and target state enables continuous improvement to beinstitutionalized in the company.InthisdirectionWiendahlintroducedthecharacteristic curves in order to describe production flows based on the bottleneck theory and theory of constraints (TOC) [307]. The underlyingideaistojointlyperformqualitycontrolplanningatthe facility and process control levels. In order to improve the transparencyofqualitycontrolplanninginmulti-stagesystems, themethod‘‘QualityValueStreamMapping’’[96],canbeusedto develop an optimal configuration of quality control along the processchain[148].Bymeansof‘‘QualityValueStreamMapping’’ theoccurrenceofdefects,theeffectiveintegrationofinspection stations as well as the design of quality control loops can be systematicallyvisualized,analysedandimproved.

3.1.5. Personnelallocationinmulti-stagesystems

The human factor has a fundamental role in achieving the required production quality performance of a manufacturing system[308].Thehumanelementis consideredasa keyfactor inallthediscussedcompanyfunctions,i.e.production,qualityand maintenance. Rootcause analysisand finalproductverification stillmainlygroundonhumanlydrivenoperations,alsoinhighly automatedcontextssuchastheautomotiveindustry.Asamatter offact,alltraditionalqualityimprovementprograms,suchasthe World ClassManufacturing, ground on the attitudeof workers towards problem solving and waste elimination. Moreover, correctiveandpreventivemaintenanceproceduresrequirehighly skilledpersonneltobeperformedincompliancywiththetarget timesandcostrequirements.Furthermore,complex manufactur-ingandassemblytasksstillentailmanualoperationsinalmostall industrial sectors. Eventhe human-robotinteraction paradigm, thatiscurrentlyunderinvestigationandtesting[138],stressesthe importance of the role of humans in advanced manufacturing systems for performing non-repetitive assembly tasks. Due to theseimplicationsbetween theworkforceorganizationand the operationalperformance ofa plant,theallocation and manage-ment of personnel in manufacturing systems have motivated significantamountofworkinthepast.Sterman[270]showedthat theproductionqualitystrategyfixedbythecompanystrategicgoals mayactivatevirtuous(workwiser)orvicious(workharder)loops inthebehaviouroftheworkforcetowardsthesetargets,depending onthevisionimposedbythemanagement.Fromanoperational pointofview,theproblemofallocatingmaintenancepersonnelin complexmulti-stagesystems hasbeeninvestigated. Automated flow lines where human operators are allocated to cope with

machinebreakdownsandothertaskssuchasinspection,support and control have been considered. In these systems, machine failuresandconsequentrepairactionsplayadominanteffecton theperformance.Tocopewithmachinefailures,arepaircrewis usuallydedicatedtotheline.However,inordertosaveoperating costs,therepaircapacityisgenerallylimitedandtherepaircrew availabilitycanbeaperformancebottleneckforthewholeline.In otherwords,whenafailureoccursandalloperatorsarebusyin repairing otherstations,themachineis forcedtowaitbeforea repair intervention is started, queued with other contingent maintenancerequests.Thiskindofmachineidlenessisknownin literatureasinterference[300]andtherelatedproblemisknownas the‘‘MachineInterferenceProblem’’(MIP)[269].Aliteraturesurvey on methods to solve MIPs can be found in [98]. An advanced approachtoiterativelysolvethisproblemhasbeenproposedin

[140].Theauthormodelledtheoriginalsystemastwointeracting

systemsthefirstbeingtheautomatedflow lineandthesecond beingtherepaircrewsystem.Thisapproachhasbeenappliedtoa realengineblockproductionlineatScaniain[45],showinggreat operationalbenefitsfortheplantobtainedbyoptimallyallocating repairmentostations. Moreover, theeffectofmixed workforce skill levels has been analysed in [48]. In [75] the problem of distributinganavailablerepaireffortin thesystem,considering the impact on the system dynamics have been solved by an analytical approach. Currently, approaches that include quality considerations in these frameworks are not available. From a qualitypointofview,anapproachtooptimallydesignSPCcontrol chart parameters, also considering the limited manpower, is proposedin[311].Anotherareathathasreceivedattentioninthe lastyearsistheanalysisoftheimpactoftheworkforcebehaviour ontheoperationalperformanceofthesystem.In[106]theimpact ofworkersabsenteeismontheperformanceofassemblylinesis investigated. The authors found that specific cross-training strategiesmayreducethelossofperformanceduetoabsenteeism. However, cross trainingmay reducequality [108,182]. Further-more,theimpactofworkforcelearningonthesystemperformance improvementduringtheramp-upphasehasbeenanalysedin[67]. Althoughpreliminaryapproachesexist,theanalysisoftheimpact of workforce on the production quality performance of a manufacturing system is a relatively new researcharea where themainchallengesisthedifficultyindevelopingreliablemodels ofthehumanbehaviour.

3.1.6. Advancedintegratedbusinessmodelsforproductionquality Business modelinnovation isa relativelynewconcept inthe manufacturing industry. Traditionally, innovation in this sector wasprimarilybasedontechnologyinnovation.Inthelastdecade, industry competitiveness has been stained by the increasing turbulenceofthemarket.Tofacethissituation,companieswere motivated to innovate their business models towards the establishment of long-termrelationships withtheir customers, and theprovisionofvalue-addedservicesbeyond thetechnical

ones [306]. The topic of proposing new business models for

machinetool buildershasattractedtheattentionofresearchers andindustrialistonlyinrecentyears.Theconceptof Reconfigur-able ManufacturingSystems [18]pavedthe waytotheidea of establishing a collaborative relation between producer and customers todesign andmanage thesystemoverits life-cycle. Later in [258] the idea of delivering services for the system adaptationandenablemodulere-usefordifferentcustomerswas proposed.Astrategicapproach fordevelopingsuchserviceshas been suggested [256]. Moreover, a CIRP Collaborative Working Group–IndustrialProductServiceSystems(IPS2_{)[187]–hasbeen}

launched, with the objective of investigating benefits and operating modes for implementing theproduct-service idea in B2Brelations.Theimplementationofsuchconceptsisthecoreidea of the EU funded project RobustPlaNet [229]. In this project, guidelines to select the bestbusiness model and collaboration mechanismdependingonthestakeholders’situationareproposed. In recent years full-service contracts and reliabilitywarranties