<%BANNER%>

Coordinating Replenishment and Disposal Decisions in Closed-Loop Supply Chains

Permanent Link: http://ufdc.ufl.edu/UFE0022853/00001

Material Information

Title: Coordinating Replenishment and Disposal Decisions in Closed-Loop Supply Chains
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Karakayali, Ibrahim
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: clsc, genetic, lagrangian, lotsizing, remanufacturing, return, reusable
Industrial and Systems Engineering -- Dissertations, Academic -- UF
Genre: Industrial and Systems Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: We consider the characterization of the optimal inventory and production planning policies for multi-product closed-loop supply chain systems for direct reuse and value-added recovery by considering a class of deterministic discrete time dynamic demand and return models. In particular, we consider three problems. First, motivated by the an original equipment manufacturer (OEM) in the power generation equipment industry that provides spare part kits for power turbine maintenance services, we study a replenishment and disposal planning problem that arises in settings where customer returns are in as-good-as-new condition. These returns can be placed into inventory to satisfy future demand or can be disposed of, in case they lead to excess inventory. We propose a Lagrangian Relaxation approach that relies on the relaxation of the capacity constraints and develop a smoothing heuristic that uses the solution of the Lagrangian problem to obtain near-optimal solutions. Our computational results demonstrate that the proposed approach is very effective in obtaining high-quality solutions with a reasonable computational effort. Next, to take the uncertainty associated with directly reusable customer returns into account explicitly, we consider a procurement planning problem encountered by a catalogue retailer that also receives customer returns. To this end, we develop a discrete time dynamic deterministic demand and stochastic return model for the characterization of inventory and production planning policy. We develop solution algorithms using continuous optimization techniques, and identify the optimal solution structures. Finally, we consider more general models for value-added recovery systems. Specifically, motivated by an OEM in the automotive industry that provides vehicle maintenance and repair services for vehicles for which replacement parts are often needed, we study manufacturing, remanufacturing, and disposal planning problem with shared manufacturing and product-specific remanufacturing and disposal capacities. We develop solution approaches based on Lagrangian Decomposition and Relaxation techniques used together with a constructive heuristic and Genetic Algorithms to obtain lower and upper bound solutions of the underlying problem. We conduct extensive computational experiments to investigate efficiency and effectiveness of proposed solution algorithms. Our computational results shows that although the Lagrangian decomposition approach is computationally more efficient, the Lagrangian relaxation approach is effective in finding high-quality solutions.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Ibrahim Karakayali.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Akcali, Elif.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0022853:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022853/00001

Material Information

Title: Coordinating Replenishment and Disposal Decisions in Closed-Loop Supply Chains
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Karakayali, Ibrahim
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: clsc, genetic, lagrangian, lotsizing, remanufacturing, return, reusable
Industrial and Systems Engineering -- Dissertations, Academic -- UF
Genre: Industrial and Systems Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: We consider the characterization of the optimal inventory and production planning policies for multi-product closed-loop supply chain systems for direct reuse and value-added recovery by considering a class of deterministic discrete time dynamic demand and return models. In particular, we consider three problems. First, motivated by the an original equipment manufacturer (OEM) in the power generation equipment industry that provides spare part kits for power turbine maintenance services, we study a replenishment and disposal planning problem that arises in settings where customer returns are in as-good-as-new condition. These returns can be placed into inventory to satisfy future demand or can be disposed of, in case they lead to excess inventory. We propose a Lagrangian Relaxation approach that relies on the relaxation of the capacity constraints and develop a smoothing heuristic that uses the solution of the Lagrangian problem to obtain near-optimal solutions. Our computational results demonstrate that the proposed approach is very effective in obtaining high-quality solutions with a reasonable computational effort. Next, to take the uncertainty associated with directly reusable customer returns into account explicitly, we consider a procurement planning problem encountered by a catalogue retailer that also receives customer returns. To this end, we develop a discrete time dynamic deterministic demand and stochastic return model for the characterization of inventory and production planning policy. We develop solution algorithms using continuous optimization techniques, and identify the optimal solution structures. Finally, we consider more general models for value-added recovery systems. Specifically, motivated by an OEM in the automotive industry that provides vehicle maintenance and repair services for vehicles for which replacement parts are often needed, we study manufacturing, remanufacturing, and disposal planning problem with shared manufacturing and product-specific remanufacturing and disposal capacities. We develop solution approaches based on Lagrangian Decomposition and Relaxation techniques used together with a constructive heuristic and Genetic Algorithms to obtain lower and upper bound solutions of the underlying problem. We conduct extensive computational experiments to investigate efficiency and effectiveness of proposed solution algorithms. Our computational results shows that although the Lagrangian decomposition approach is computationally more efficient, the Lagrangian relaxation approach is effective in finding high-quality solutions.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Ibrahim Karakayali.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Akcali, Elif.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0022853:00001


This item has the following downloads:


Full Text

PAGE 1

1

PAGE 2

2

PAGE 3

3

PAGE 4

Thanksareduetomysupervisor,Dr.ElifAkcal,forherinsights,perspectives,guidanceandsupport.IthankDr.SlaCetinkaya,Dr.JosephP.GeunesandDr.SelcukErengucforservingasmydissertationcommitteemembersaswellfortheircontributionsandvaluablesuggestions.IalsothankDr.HalitUsterforhisvaluablecontributions,insights,andsuggestions.IextendmysinceregratitudetoDr._IsmailS.BakalandZekiC.Tasknfortheirwarmfriendship.Itwouldhavebeenmuchmoredicultformetogetthroughthisprocesswithouttheirendlesssupportandunderstanding.Ithankmyparents,HaticeandBakiKarakayal;mybelovedsisters,FusunandRaziye;andmylittlebrother,Burak,fortheirunconditionalloveandsupport.Finally,Ithankmywife,andbestfriendandcondantinlife,Gunseli,forhersupportandencouragementtocompletemydegree.Mywifeandmydaughter,Sla,werethebiggestmotivationforthiswork.Iknewthatwewouldbetogetherforeveruponthecompletionofmydegree.IalsowouldliketoacknowledgethenancialsupportprovidedbytheNationalScienceFoundationunderGrantNo.CMMI-0522960. 4

PAGE 5

page ACKNOWLEDGMENTS ................................. 4 LISTOFTABLES ..................................... 7 LISTOFFIGURES .................................... 9 ABSTRACT ........................................ 10 CHAPTER 1INTRODUCTION .................................. 12 1.1BackgroundandMotivation .......................... 12 1.2Overview .................................... 13 1.3Impact ...................................... 14 2LITERATUREREVIEW .............................. 15 2.1Introduction ................................... 15 2.2DeterministicDiscreteTimeModelsforInventoryandProductionPlanningforProductRecovery .............................. 15 2.3StochasticModelsforInventoryandProductionPlanningforProductRecovery 19 2.4Multi-ItemCapacitatedLotSizing ...................... 21 3REPLENISHMENTANDDISPOSALPLANNINGFORMULTIPLEPRODUCTSWITHREUSABLERETURNS .......................... 25 3.1Introduction ................................... 25 3.2SharedReplenishmentandProduct-SpecicDisposalCapacities ...... 26 3.3SolutionApproach ............................... 30 3.3.1ObtainingLowerBounds ........................ 32 3.3.2ObtainingUpperBounds ........................ 36 3.3.3SubgradientOptimization ....................... 39 3.4ComputationalExperiments .......................... 39 3.4.1RandomTestInstanceGeneration ................... 40 3.4.2DiscussionofComputationalResults ................. 42 3.5SharedReplenishmentandDisposalCapacities ................ 45 3.6ConcludingRemarks .............................. 48 4MANUFACTURING,REMANUFACTURING,DISPOSALPLANNINGFORMULTIPLEPRODUCTSWITHREMANUFACTURABLERETURNS .... 56 4.1Introduction ................................... 56 4.2ModellingAssumptionsandMathematicalFormulationofMRDPP .... 58 4.3SolutionApproaches .............................. 63 4.3.1ALagrangianDecompositionApproachforMRDPP ......... 65 5

PAGE 6

................... 67 4.3.1.2Obtainingupperbounds ................... 72 4.3.1.3Subgradientoptimization .................. 80 4.3.2ALagrangianRelaxationApproachforMRDPP ........... 81 4.3.2.1Obtaininglowerbounds ................... 82 4.3.2.2Obtainingupperbounds ................... 83 4.3.2.3Subgradientoptimization .................. 84 4.4ComputationalExperiments .......................... 85 4.4.1RandomTestInstanceGeneration ................... 85 4.4.2DiscussionofComputationalResults ................. 86 4.5ConcludingRemarks .............................. 89 5PROCUREMENTPLANNINGFORMULTIPLEPRODUCTSWITHREUSABLERANDOMRETURNS ................................ 96 5.1Introduction ................................... 96 5.2ModellingAssumptions ............................ 97 5.3PPP1:SinglePeriodandSingleProduct ................... 98 5.4PPP2:SinglePeriodandMultipleProducts ................. 102 5.5PPP3:MultiplePeriodandSingleProduct .................. 107 5.6PPP4:MultiplePeriodandMultipleProduct ................ 114 5.7ConcludingRemarks .............................. 119 6CONCLUSION .................................... 120 REFERENCES ....................................... 123 BIOGRAPHICALSKETCH ................................ 129 6

PAGE 7

Table page 3-1Parametervaluesusedincomputationalexperiments. ............... 51 3-2PerformanceoftheLRApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). ............................ 51 3-3EectofTimeBetweenReplenishment(TBR)andTimeBetweenDisposal(TBD)SetupsontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. 52 3-4EectofReplenishmentandDisposalCapacityTightnessontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. ............... 52 3-5EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. ..................... 52 3-6PerformanceofCPLEX(withaTimeLimitofthe30SecondsandaStoppingGapof0.1Percent):AverageandMaximumOptimalityGaps(%). ....... 52 3-7EectofTimeBetweenReplenishment(TBR)andTimeBetweenDisposal(TBD)SetupsontheAverageandMaximumOptimalityGaps(%)ofCPLEX. ..... 53 3-8EectofReplenishmentandDisposalCapacityTightnessontheAverageandMaximumOptimalityGaps(%)ofCPLEX. .................... 53 3-9EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)ofCPLEX. .......................... 53 3-10InuenceofIncreasingtheTimeLimitandtheStoppingGaponthePerformanceofCPLEX:NumberofUnsolvedInstances,AverageandMaximumOptimalityGaps(%). ....................................... 53 3-11PerformanceoftheLRApproachfortheRDPPwithSharedReplenishmentandDisposalCapacities:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). .................................. 54 3-12PerformanceofCPLEXfortheRDPPwithSharedReplenishmentandDisposalCapacities(withaTimeLimitofthe30secondsandaStoppingGapof0.1Percent):AverageandMaximumOptimalityGaps(%). ................... 54 3-13InuenceofIncreasingtheTimeLimitandtheStoppingGaponthePerformanceofCPLEXfortheRDPPwithSharedReplenishmentandDisposalCapacities:NumberofUnsolvedInstances,AverageandMaximumOptimalityGaps(%). 55 4-1ParameterValuesUsedinComputationalExperimentsofMRDPP ....... 93 4-2PerformanceoftheLDApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). ............................ 93 7

PAGE 8

............ 93 4-4EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. ..................... 94 4-5PerformanceoftheLRApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). ............................ 94 4-6EectofTimeBetweenActivitySetupsandCapacityTightnessontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. ............ 94 4-7EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. ..................... 94 4-8PerformanceofCPLEX(withaTimeLimitof400Seconds):AverageandMaximumOptimalityGaps(%). ................................ 95 8

PAGE 9

Figure page 3-1Problemsetting. ................................... 50 3-2Pseudo-codeoftheIdenticationofSetIpt. .................... 50 3-3Pseudo-codeoftheIdenticationofSetI .................... 50 4-1ProblemSettingofMRDPP ............................. 90 4-2Pseudo-codeoftheLDApproachfortheMRDPP. ................. 91 4-3Pseudo-codeoftheGAApproachfortheLD. ................... 92 4-4Pseudo-codeoftheLRApproachfortheMRDPP. ................. 92 9

PAGE 10

Weconsiderthecharacterizationoftheoptimalinventoryandproductionplanningpoliciesformulti-productclosed-loopsupplychainsystemsfordirectreuseandvalue-addedrecoverybyconsideringaclassofdeterministicdiscretetimedynamicdemandandreturnmodels.Inparticular,weconsiderthreeproblems.First,motivatedbytheanoriginalequipmentmanufacturer(OEM)inthepowergenerationequipmentindustrythatprovidessparepartkitsforpowerturbinemaintenanceservices,westudyareplenishmentanddisposalplanningproblemthatarisesinsettingswherecustomerreturnsareinas-good-as-newcondition.Thesereturnscanbeplacedintoinventorytosatisfyfuturedemandorcanbedisposedof,incasetheyleadtoexcessinventory.WeproposeaLagrangianRelaxationapproachthatreliesontherelaxationofthecapacityconstraintsanddevelopasmoothingheuristicthatusesthesolutionoftheLagrangianproblemtoobtainnear-optimalsolutions.Ourcomputationalresultsdemonstratethattheproposedapproachisveryeectiveinobtaininghigh-qualitysolutionswithareasonablecomputationaleort.Next,totaketheuncertaintyassociatedwithdirectlyreusablecustomerreturnsintoaccountexplicitly,weconsideraprocurementplanningproblemencounteredbyacatalogueretailerthatalsoreceivescustomerreturns.Tothisend,wedevelopadiscretetimedynamicdeterministicdemandandstochasticreturnmodelforthecharacterizationofinventoryandproductionplanningpolicy.Wedevelopsolutionalgorithmsusingcontinuousoptimizationtechniques,andidentifytheoptimalsolution 10

PAGE 11

11

PAGE 12

PohlenandM.T.Farris 1992 ; RogersandTibben-Lembke 1999 ).Toreducetheeconomicalandenvironmentalburdenofsolidwaste,productrecoverypracticesarebecomingincreasinglypopular. Productrecoverymanagement(PRM)encompassestheplanningandmanagementofabroadsetofactivitiestoreclaimtheeconomicalandenvironmentalvalueresidinginend-of-useorend-of-lifeproducts.Theeconomicalvaluethatcanbereclaimedfromtheend-of-useandend-of-lifeproductsisrelatedtothecostsavingsrealizedbyrecoveringproducts,components,andmaterialsforreuse.Similarly,theenvironmentalvalueinvolvestheenvironmentalimpactreductionrealizedbydiminishingtherelianceonlandllingandincinerationforsolidwastedisposal. Historically,boththetheoryandpracticeofsupplychainmanagementhasplacedemphasisontheforwardchannel,whichconsistsofsuppliers,manufacturers,distributors,retailers,andcustomers.Recentinterestinproductrecovery,however,hasextendedthescopeoftraditionalsupplychainmanagementbydrawingattentiontothereversechannel,whichconsistsofnal-users,retailers,collectors,andremanufacturers.Sinceproductrecoveryprocessesreturnedandusedproductsinthereversechannel,theeciencywithwhichtheseproductsareprocessedhasadirectimpactontheprotabilityofproductrecoverypractices.Moreover,asreturnsareprocessedorremanufacturedtosatisfythecustomerdemand,thisimpactstheforwardchannelowsandintroducesastrong 12

PAGE 13

Acloserexaminationofexistingproductrecoverypracticesrevealthatitispossibletodistinguishbetweentwoformsofproductrecovery: 13

PAGE 14

14

PAGE 15

Guide 2000 ),thegeneraltopicofinventoryandproductionplanningforproductrecoveryhasreceivedconsiderableattentioninrecentyears.Foracomprehensivereviewoftheliterature,wereferthereaderto( DekkerandvanderLaan 2003 ; Inderfurthetal. 2004 ; InderfurthandTeunter 2003 ; vanderLaanetal. 2004 ; RichterandDobos 2004 ).Acarefulexaminationoftheinventoryandproductionplanningforproductrecovery,however,revealsthatthislineofworkmainlyconcentratesonvalueaddedrecovery( Golanyetal. 2001 2005 ; Lietal. 2006b a ; Richter 15

PAGE 16

, 2003 ; RichterandSombrutzki 2000 ; RichterandWeber 2001 ; Yangetal. 2005 )withoneexceptionthatfocusesondirectreuse( BeltranandKrass 2002 ). Motivatedbypracticalapplicationsinhandlingreturnsincatalogueretailingandmanagingaircraftengineservicecomponentkits,BeltranandKrass( BeltranandKrass 2002 )considerasingleitemuncapacitatedlotsizingforasystemwherereturnsreceivedfromcustomersareplacedinnishedgoodsinventorywithoutanyprocessing.Theirmodelisaimedtodeterminetheoptimalquantityofitemstoprocure,quantityofitemstodispose,andinventorylevelforitemsthroughouttheplanninghorizonwiththeobjectiveofminimizingthesumofconcaveandpiecewisedierentiableprocurement,disposal,andinventoryholdingcosts.Toobtainthesolutiontothismodel,theydevelopaDPalgorithmthatrunsinpolynomialtime. RichterandSombrutzki ( 2000 )considerthesingleitemuncapacitatedlotsizingproblemforapureremanufacturingsystemwithxed-chargeremanufacturingandlinearinventoryholdingcosts.Theirmodelseekstheoptimalquantityofitemstoremanufactureineachperiodandtheinventorylevelsforremanufacturableandremanufactureditemsthroughouttheplanninghorizonwiththeobjectiveofminimizingthesumofremanufacturingsetupandinventoryholdingcosts.Theyshowthatamodicationoftheclassicallotsizingalgorithmcanbeusedtoobtainthesolution( WagnerandWhitin 1958 ).Theyalsoconsidertheprobleminthecontextofahybridremanufacturingandmanufacturingsystem,consideringxed-chargemanufacturingcostswherethemodelseeksalsotheoptimalquantityofitemsmanufacturedineachperiod.However,theydonotprovideasolutionapproachforthisgeneralproblem,butconsideraspecialcaseoftheproblemwherethequantityofremanufacturableitemsavailableatthebeginningoftheplanninghorizonissucientlylargetocovertheentiredemandthroughouttheplanninghorizon.Theydevelopadynamicprogramming(DP)approachtoobtainthesolution. RichterandWeber ( 2001 )extendthemodelformulationsgivenby RichterandSombrutzki ( 2000 )byincludinglinearremanufacturingandmanufacturingcostsand 16

PAGE 17

RichterandGobsch ( 2003 )concentrateonthesingleitemuncapacitatedlotsizingproblemforahybridmanufacturingandremanufacturingsystemconsideringxed-chargemanufacturingandremanufacturingaswellaslinearnewmaterialpurchase,remanufacturableitemacquisition,andinventoryholdingcosts.Theirmodelseekstheoptimalquantityofremanufactureditemsandmaterialtopurchase,quantityofnewitemstomanufacture,quantityofremanufacturableitemstoremanufacture,andinventorylevelsfornewmaterials,remanufacturableitems,andremanufactured/newitemsthroughouttheplanninghorizonwiththeobjectiveofminimizingthetotalcosts.TheyprovideanestedDPapproachtosolvethisgeneralproblem.Theyalsoanalyzetwospecialcasesoftheproblemwhere(i)nomaterialorremanufacturableiteminventoryisheldand(ii)noremanufacturedornewiteminventoryisheld. Golanyetal. ( 2001 2005 )considerasingleitemuncapacitatedlotsizingproblemforahybridmanufacturingandremanufacturingsystemwithmanufacturing,remanufacturing,disposalandinventoryholdingcosts.Theyshowthattheproblem(i)isNP-hardforgeneralconcavecostsand(ii)canbetransformedintoaminimumcostnetworkowproblemifallcostsarelinear. Yangetal. ( 2005 )showthattheproblemremainsNP-hardforstationaryconcavecostsanddevelopaDPapproachfortheproblem.Byexploringthecharacteristicsoftheextremepointsolutions,theydevelopaDPbasedheuristicapproachthatrunsinpolynomialtime. Lietal. ( 2006b )consideramulti-itemuncapacitatedlotsizingproblemforahybridmanufacturingandremanufacturingsystem,wheredownwardsubstitutionofitemsisallowed,i.e.,ahighergradeitemcanbesubstitutedforalowergradeitemtosatisfythedemand.Wenotethatthegradedoesnotrefertowhethertheitemisneworremanufactured,buttothecharacteristicsoftheitem,e.g.,afasterprocessorcanbesubstitutedforaprocessorwithalowerspeed,oramemorycardwithmorestoragespace 17

PAGE 18

Lietal. ( 2006a )developatwo-itemmodelinacapacitatedsettingwhereemergencyprocurementsareallowedincaseofapotentialshortage.Theirmodelseekstheoptimalquantitytomanufacture,remanufacture,andprocureforeachitem,remanufacturableandnew/remanufacturedinventorylevelsforeachitemaswellasthequantityofalowergradeitemusedtosatisfythedemandforahighergradeitemwiththeobjectiveofminimizingthesumofxed-chargemanufacturingandremanufacturingaswellaslinearmanufacturing,remanufacturing,emergencyprocurement,andinventoryholdingcosts.Althoughdownwardsubstitutionisallowedtosatisfythedemand,thecostofsubstitutionisassumedtobezero.TheauthorsdevelopahybridDPandgeneticalgorithm(GA)approachtoobtainasolutionfortheproblem,whereGAapproachisusedtospecifythetimingofthemanufacturingandremanufacturingsetups,whereastheDPalgorithmisusedtodeterminethemanufacturing,remanufacturing,substitution,andemergencyprocurementquantities. 18

PAGE 19

Theexistingcontinuousreviewmodelsattempttocomputeinventoryandproductioncontrolpolicyparametersmainlyforvalue-addedrecoverysystems.Assumingthatallthereturnsareremanufacturedoneatatimeuponreceipt,alineofworkfocusesonsystemswithasinglestockpoint Bayndretal. ( 2003 2005 ); Fleischmannetal. ( 2002 ); Heyman ( 1977 ); vanderLaan ( 2003 ); Toktayetal. ( 2000 ); YuanandCheung ( 1998 ).Anotherlineofworkfocusesonsettingswherereturnsareremanufacturedinbatches Arasetal. ( 2004 ); MuckstadtandIsaac ( 1981 ); Teunteretal. ( 2000 ); Teunter ( 2002 ); Teunteretal. ( 2004 ); vanderLaan ( 2003 ); vanderLaanetal. ( 1996a b ); vanderLaanandSalomon ( 1997 ); vanderLaanetal. ( 1999a b ); vanderLaanandTeunter ( 2006 ).Theemphasisismainlyontheanalysisofsingle-itemsettings Arasetal. ( 2004 ); Bayndretal. ( 2003 2005 ); Fleischmannetal. ( 2002 ); Heyman ( 1977 ); MuckstadtandIsaac ( 1981 ); Teunteretal. ( 2000 ); Teunter ( 2002 ); Teunteretal. ( 2004 ); Toktayetal. ( 2000 ); vanderLaan ( 2003 ); vanderLaanetal. ( 1996a b ); vanderLaanandSalomon ( 1997 ); vanderLaanetal. ( 1999a b ); vanderLaanandTeunter ( 2006 ); YuanandCheung ( 1998 ),consideringbothnite Teunteretal. ( 2000 ); Teunter ( 2002 ); Teunteretal. ( 2004 )andinnite Arasetal. ( 2004 ); Bayndretal. ( 2003 2005 ); Fleischmannetal. ( 2002 ); Heyman ( 1977 ); MuckstadtandIsaac ( 1981 ); Toktayetal. ( 2000 ); vanderLaan ( 2003 ); vanderLaanetal. ( 1996a b ); vanderLaanandSalomon ( 1997 ); vanderLaanetal. ( 1999a b ); vanderLaanandTeunter ( 2006 ); YuanandCheung ( 1998 )planninghorizons.Onlyalimitedamountof 19

PAGE 20

Fleischmannetal. ( 2002 ); Heyman ( 1977 ),andthemainemphasisofthesemodelsisonanalyzingsystemperformanceunderaparticularinventorycontrolpolicy.TheunderlyingsystemistypicallymodeledasacontinuoustimeMarkovchain Arasetal. ( 2004 ); Bayndretal. ( 2005 ); MuckstadtandIsaac ( 1981 ); vanderLaan ( 2003 ); vanderLaanetal. ( 1996a b ); vanderLaanandSalomon ( 1997 ); vanderLaanetal. ( 1999a b ); vanderLaanandTeunter ( 2006 ); YuanandCheung ( 1998 )oraqueueingnetwork Bayndretal. ( 2003 ); Toktayetal. ( 2000 ).Tocomputetheinventorycontrolpolicyparameters,numericaloptimizationtechniques Arasetal. ( 2004 ); Bayndretal. ( 2003 2005 ); MuckstadtandIsaac ( 1981 ); vanderLaan ( 2003 ); vanderLaanetal. ( 1996a b ); vanderLaanandSalomon ( 1997 ); vanderLaanetal. ( 1999a b ); vanderLaanandTeunter ( 2006 ); YuanandCheung ( 1998 )oranalyticalapproximations Toktayetal. ( 2000 ); vanderLaan ( 2003 ); vanderLaanandTeunter ( 2006 )areused.Amorerecentlineofworkalsousessimulationmodelingandanalysistocomputeeectivevaluesforpolicyparametersortostudysystemperformance Teunteretal. ( 2000 ); Teunter ( 2002 ); Teunteretal. ( 2004 ); Toktayetal. ( 2000 ). Themajorityofexistingperiodicreviewmodelsaimtocomputeinventoryandproductioncontrolpolicyparametersforvalue-addedrecoverysystems.Whilesomemodelsaredevelopedconsideringasinglestockpoint BuchananandAbad ( 1998 ); Cohenetal. ( 1980 ); FleischmannandKuik ( 2003 ); Inderfurth ( 1997 ); KelleandSilver ( 1989 ); KiesmullerandScherer ( 2003 ); KiesmullerandvanderLaan ( 2001 ); MostardandTeunter ( 2006 ); Nakashimaetal. ( 2002 2004 ); VlachosandDekker ( 2003 ); Whisler ( 1967 ),othersfocusonsettingswithtwostockpoints Bayndretal. ( 2007 ); Inderfurth ( 1997 2004 ); Kiesmuller ( 2003 ); KiesmullerandMinner ( 2003 ); KiesmullerandScherer ( 2003 ); Mahadevanetal. ( 1999 ); Simpson ( 1978 ); TeunterandVlachos ( 2002 ),andthegeneralemphasisoftheexistingmodelshasbeenonsingle-itemsettingsconsideringsingleperiod Bayndretal. ( 2007 ); Inderfurth ( 2004 ); MostardandTeunter ( 2006 ); Vlachosand 20

PAGE 21

( 2003 ),nitehorizon BuchananandAbad ( 1998 ); Cohenetal. ( 1980 ); HeisigandFleischmann ( 2001 ); Inderfurth ( 1997 ); KelleandSilver ( 1989 ); KiesmullerandMinner ( 2003 ); KiesmullerandScherer ( 2003 ); KiesmullerandvanderLaan ( 2001 ); Simpson ( 1978 ); TeunterandVlachos ( 2002 ); Whisler ( 1967 ),andinnitehorizon FleischmannandKuik ( 2003 ); HeisigandFleischmann ( 2001 ); Kiesmuller ( 2003 ); Mahadevanetal. ( 1999 ); Nakashimaetal. ( 2002 2004 ); Whisler ( 1967 )models. Alimitedstreamofresearchonperiodicreviewmodelscharacterizestheoptimalpolicyusingdierentialcalculustechniquesinsingleperiodsettings Bayndretal. ( 2007 ); Inderfurth ( 2004 ); MostardandTeunter ( 2006 ); VlachosandDekker ( 2003 )anddynamic-programming-basedoptimizationtechniquesinnitehorizonsettings Cohenetal. ( 1980 ); FleischmannandKuik ( 2003 ); Inderfurth ( 1997 ); Simpson ( 1978 ); Whisler ( 1967 ).Also,Markovdecisionprocessmodelsaredevelopedtocomputetheoptimalinventorycontrolparameters Nakashimaetal. ( 2002 2004 ).Otherexistingworkintheareafocusesonanalyzingsystemperformanceunderaparticularinventorycontrolpolicyandusesdynamicprogramming BuchananandAbad ( 1998 ); Cohenetal. ( 1980 ); KelleandSilver ( 1989 ); KiesmullerandScherer ( 2003 )orsimulationbased Kiesmuller ( 2003 ); KiesmullerandMinner ( 2003 ); KiesmullerandvanderLaan ( 2001 ); Mahadevanetal. ( 1999 ); TeunterandVlachos ( 2002 )approachestocomputethecontrolparameters. Brahimietal. 2006 ).Itispossibletoclassifytheexistingworkonlotsizingproblemsaccordingtothe(i)numberofproducts(i.e.,single-vs.multi-item)considered,(ii)numberoflevelsinthebill-of-materials(i.e.,single-vs.multi-level)considered,and(iii)restrictiononthecapacityavailabilityofproductionresources(i.e.,uncapacitatedvs.capacitated)takenintoaccount( Brahimietal. 2006 ; Karimietal. 2003 ).Inthiscontext,theclassofproblemsweconsiderintheproposed 21

PAGE 22

Therearecomprehensivereviewsofsingle-levelsingle-item( Brahimietal. 2006 )andmulti-item( Karimietal. 2003 )capacitatedlotsizingmodelsintheliterature.Morerecently,JansandDegraeve( JansandDegraeve 2007 )reviewthemeta-heuristicapproachesdevelopedfordicultlotsizingproblems.Thegeneralsingle-itemcapacitatedlotsizingproblem( Florianetal. 1980 )aswellassomespecialcases( BitranandYanasse 1982 )areNP-hard.Thegeneralmulti-itemcapacitatedlotsizingproblemisstronglyNP-hard( ChenandThizy 1990 ).Inwhatfollows,wereviewsomeoftheexistingworkthatdevelopsexactandheuristicsolutionapproachesforthemulti-itemcapacitatedlotsizingproblem. Alineofexistingworkfocusesonthedevelopmentofexactsolutionapproachesforthesingle-levelmulti-itemcapacitatedlotsizingproblem( Baranyetal. 1984 ; EppenandMartin 1987 ; Leungetal. 1989 ).While Baranyetal. ( 1984 )proposecutgenerationtechniquewhicharethefacetsofthesingle-itemuncapacitatedlotsizingproblem, Leungetal. ( 1989 )exploretheuseofthefacetsofthesingle-itemcapacitatedlotsizingproblem. EppenandMartin ( 1987 )proposealternativeformulationsthatyieldmuchtighterlinearrelaxationmodelsforboththesingle-andmulti-itemcapacitatedlotsizingproblem. BelvauxandWolsey ( 2000 2001 )developaframeworkforthesolutionofthemulti-level,multi-resource,multi-itemcapacitatedlotsizingproblemwithbackloggingthatperformssomepreprocessing,generatescuttingplanes,andutilizesaprimalheuristicforthelotsizingproblem. Thereareanumberofpapersthatdevelopsspecializedconstructiveand/orimprovementheuristicsforthemulti-itemcapacitatedlotsizingproblem( Dogramacietal. 1981 ; DixonandSilver 1981 ; Eisenhut 1975 ; Gunther 1987 ; KarniandRoll 1982 ; KrcaandKokten 1994 ; LambrechtandVanderveken 1979 ; MaesandWassenhove 1986 ; SelenandHeuts 1987 ; Trigeiro 1989 ). Eisenhut ( 1975 )proposesasingleforward-pass 22

PAGE 23

LambrechtandVanderveken ( 1979 )and DixonandSilver ( 1981 )proposeprocedurestoremovecapacityinfeasibilities. Dogramacietal. ( 1981 )and KarniandRoll ( 1982 )proposemulti-phaseheuristicapproachesthatobtainimprovedsolutionsfrominitialfeasiblesolutions. MaesandWassenhove ( 1986 )proposeaexibleapproachthatcanusedierentrulesfortheaprioriorderingofitems,criteriausedtodecidewhetherornottoincludethedemandforaniteminthecurrentperiod,andsearchstrategy.Theirapproachincludesprocedurestoensurefeasibilityandtoimproveagivenfeasiblesolution.Alltheseapproachesbuildaplanusingarollinghorizonapproach. KrcaandKokten ( 1994 )focusonitemsratherthanperiods,andoneachiterationoftheapproach,afeasibleproductionplanforanitemisdeveloped.Thatis,oncetheplanforanitemisgenerated,theavailablecapacityineachperiodisupdated.Then,aDPbasedprocedureisusedtogeneratetheplanforthenextitem. Gunther ( 1987 )proposetostartfromalot-for-lotsolutionandbuildacapacityfeasiblesolutionusingapriorityindextocombinelots,alook-aheadproceduretoensurecumulativecapacityfeasibility,andax-upproceduretoremoveperiodcapacityinfeasibilities. SelenandHeuts ( 1987 )proposetheuseofamodiedpriorityindexforGunther'sheuristic. Trigeiro ( 1989 )proposesamulti-passapproachfortheversionoftheproblemwithsetuptimes.Intherstforwardpassaninitialplanisbuiltusingamodiedversionofthewell-knownSilver-Mealpriorityindex.Inthesecondbackwardpass,capacityinfeasibilitiesareremovedbyshiftingexcessquantities.Thenalthirdpassensuresthattheresultingsolutionsatisesthepropertiesofanextremepointsolution. Anotherlineofworkfocusesonthedevelopmentofrelaxation-basedheuristicsforthemulti-itemcapacitatedlotsizingproblem( ChenandThizy 1990 ; Diabyetal. 1992b a ; MillarandYang 1994 ; Thizy 1991 ; ThizyandWassenhove 1985 ; Trigeiro 1987 ; Trigeiro 23

PAGE 24

, 1989 ). ThizyandWassenhove ( 1985 )and Trigeiro ( 1987 )proposeLagrangianrelaxationapproachesthatfocusontherelaxationofthecapacityconstraints.While ThizyandWassenhove ( 1985 )developatransportationproblembasedupperboundingheuristictoconstructafeasibleschedulefromthedualsolutiontotheoriginalproblem, Trigeiro ( 1987 )implementsasmoothingheuristicforthesamepurpose. ChenandThizy ( 1990 )provideacomprehensiveoverviewofrelaxationapproachesandshowthattheLagrangianrelaxationofthecapacityconstraintsyieldthetightestlowerboundontheoptimalsolution. Thizy ( 1991 )investigatesthequalityofboundsgeneratedbyalternativeLagrangiandecompositionapproaches. Trigeiroetal. ( 1989 )developaLagrangianrelaxationbasedsolutionapproachfortheversionoftheproblemwithsetuptimes.Inaddition, Diabyetal. ( 1992a )and Diabyetal. ( 1992b )considersingle-andmulti-resource,respectively,versionoftheproblemanddevelopLagrangianrelaxationbasedsolutionapproaches. MillarandYang ( 1994 )alsoconcentrateontheversionoftheproblemwithsetuptimesandstudyLagrangianrelaxationanddecompositiontechniques. 24

PAGE 25

Morespecically,theproblemisconcernedwithdeterminingwhen,andhowmuch,toreplenishordisposeofeachtypeofproduct.Weconsidertheprobleminadeterministicdemandandreturnmodelingframeworkoveraniteplanninghorizonunderthepresenceofprocessingcapacitiesforreplenishmentanddisposalactivities.Wedevelopanintegerprogrammingformulationtoseekanoptimalsolutionthatcharacterizesthe 25

PAGE 26

3.2 and 3.5 ,respectively. Theremainderofthispaperisorganizedasfollows.WeproceedwithconsideringtherstvariantoftheRDPP.Section 3.2 presentsamathematicalformulationforthisproblem,andthesolutionapproachisdescribedinSection 3.3 .ResultsfromcomputationalexperimentsarepresentedandanalyzedinSection 3.4 .Section 3.5 detailshowtheproposedmathematicalformulationandthesolutionapproachcanbemodiedforaddressingthesecondvariantofRDPPwherebothreplenishmentanddisposalcapacitiesaresharedamongmultipleproducttypes.Finally,conclusionsandfutureresearchdirectionsaresummarizedinSection 3.6 26

PAGE 27

Thecustomersmayorderseveralkitsbutusefewerkitsthanordered.ThekitsthatarenotusedbythecustomercanbereturnedtotheOEM.Inourwork,weconsiderfullkitsonly.Thatis,partialkitsarenotacceptedbytheOEM.Hence,theOEMhastwosourcesforreplenishingtheinventoryofassembledkits:returnsreceivedfromthecustomersandnewkitsassembledbyusingthepartsinthepartinventories.Iftoomuchinventoryofaparticulartypeofassembledkitaccumulates,thentheOEMcan`dispose'oftheexcessinventorybydisassemblingthekitsandrestockingpartsinpartinventories.Thedisassemblyofkitscanbeperformedonworkbencheswhichdonotneedtobecongured;however,thelaborrequiredimposesadisassemblycapacityrestrictionthatlimitsthequantityofkitsthatcanbedisassembled.Therefore,thereisadisassemblycapacityrestrictionthatlimitsthequantityofaparticulartypeofkitthatcanbedisassembled,butthedisassemblycapacityisnotsharedamongdierenttypesofkits. Toensureuninterruptedsparepartkitavailability,theOEMhastocoordinatereplenishment(i.e.,assembly)anddisposal(i.e.,disassembly)decisionsforthekits,whichyieldsthemulti-itemcapacitatedlotsizingproblemofinterestinthischapter.TheproblemsettingisshowninFigure 3-1 .Specically,giventheforecasteddemandsandreturnsformultipletypesofproducts(kits)aswellasthesharedreplenishmentandproduct-specicdisposalcapacityrestrictionsineachtimeperiod,ourobjectiveis 27

PAGE 28

Wenotethatwithoutlossofgeneralitytheleadtimesassociatedwithreplenishmentanddisposalactivitiescanassumedtobezero.Moreover,therearenospeculativemotivesthatencouragereplenishmentwiththepurposeofdisposalinthefuture.Theproblemparametersareasfollows: 28

PAGE 29

Wenotethatwithoutlossofgeneralitytheleadtimesassociatedwithreplenishmentanddisposalactivitiescanassumedtobezero.Moreover,therearenospeculativemotivesthatencouragereplenishmentwiththepurposeofdisposalinthefuture.Thedecisionvariablesareasfollows: TheRDPPwithsharedreplenishmentandproduct-specicdisposalcapacitiesisformulatedasminPXp=1TXt=1(Fmptympt+ptxmpt)+PXp=1TXt=1(Fdptydpt+ptxdpt)+PXp=1TXt=1"ptipt (3{2)xdptCdptydptp=1;:::;P;t=1;:::;T; (3{3)ip;t1+xmpt=D0pt+xdpt+iptp=1;:::;P;t=1;:::;T; (3{4)xmpt (3{5)xdpt (3{6) 29

PAGE 30

(3{7)xmpt;xdpt;ipt;0andintegerp=1;:::;P;t=1;:::;T; (3{8)ympt;ydpt2f0;1gp=1;:::;P;t=1;:::;T: Objectivefunction( 3{1 )minimizesthesumofthexedcostofperformingreplenishmentanddisposalsetupsaswellasthelinearreplenishment,disposal,andinventoryholdingcoststhroughouttheplanninghorizon.Constraintset( 3{2 )ensuresthatthetotalreplenishmentcapacityrequiredforalltheproductsdoesnotexceedtheavailablereplenishmentcapacityinperiodt.Constraintset( 3{3 )ensuresthatthedisposalcapacityrequiredforeachproductpdoesnotexceedtheavailabledisposalcapacityforproductpinperiodt.Constraintset( 3{4 )istheinventorybalanceequationforeachproductpineachperiodt,whichensuresthatforeachproductp,thesumofthequantityininventorycarriedintoperiodtandthequantityproducedinperiodtisequaltothesumofthenetdemandinperiodt,thequantitydisposedinperiodt,andthequantityininventorycarriedtoperiodt+1.Constraintset( 3{5 )ensuresthatareplenishmentsetupisperformedforproductpinperiodtifthequantityofproductpproducedinperiodtispositive;thisquantityshouldnotexceedthecumulativedemanduntiltheendoftheplanninghorizon.Similarly,constraintset( 3{6 )ensuresthatadisposalsetupisperformedforproductpinperiodtifthequantityofproductpdisposedinperiodtispositive;thisquantityshouldnotexceedthecumulativereturnsreceivedsincethebeginningoftheplanninghorizon.Constraintset( 3{7 )initializesthebeginningandendinginventorylevelsforeachproductp.Constraintsets( 3{8 )and( 3{9 )ensuretheintegralityofthedecisionvariables. 3{2 )and( 3{3 ).Lett0betheLagrangianmultiplierassociatedwiththereplenishmentcapacityconstraintinperiodtandbetheTdimensionalLagrangianvectorassociated 30

PAGE 31

3{2 ).Similarly,letpt0betheLagrangianmultiplierassociatedwiththedisposalcapacityconstraintforproductpinperiodtandbethePTdimensionalLagrangianmatrixassociatedwithconstraintset( 3{3 ).Givenand,theLagrangianproblemobtainedbyrelaxingconstraintsets( 3{2 )and( 3{3 )canbestatedasfollows:minPXp=1TXt=1(Fmptympt+(pt+pt)xmpt)+PXp=1TXt=1((FdptCdptpt)ydpt+(pt+pt)xdpt)+PXp=1TXt=1"ptiptTXt=1tCmt subjectto( 3{4 ){( 3{9 ). Foragivenand,theLagrangianproblem( 3{10 )canbesolvedecientlytoobtainalowerbound.Moreover,usingthesolutiontotheLagrangianproblem,asmoothingheuristiccanbeutilizedtoidentifyasolutionthatdoesnotviolatereplenishmentanddisposalcapacityrestrictionswhenndinganupperbound.Usingtheselowerandupperbounds,thevaluesofthemultipliersandcanbeupdatedaccordinglyusingsubgradientoptimization. Letkbetheiterationcounter.Supposethattheterminationcriteriaaredenedby,thetolerancegapbetweenthebestlowerandupperbound,andkmax,themaximumnumberofiterations.Also,UBkandLBkdenotetheupperandlowerbounds,respectively,ontheobjectivefunctionvalueoftheLagrangianproblematiterationk.Similarly,UBandLBdenotetheincumbentupperandlowerbounds,respectively,ontheobjectivefunctionvalueoftheLagrangianproblem.Then,thegeneralLRprocedureweusecanbestatedasfollows: 1. Initialize,kmaxandtheLagrangianmultipliers.Setk=1;UB=UBk=1;andLB=LBk=. 2. While((LB(1)UB)and(kkmax)) (a) FormulateandsolvetheLagrangiandualproblemtoobtainLBk. 31

PAGE 32

If(LBk>LB)thensetLB=LBk. (c) ApplythesmoothingheuristictondafeasiblesolutiontoobtainUBk. (d) If(UBk
PAGE 33

3{11 )canbenegativesincept0.Weobservethatintheoptimalsolutiontothisproblem,if(FdptCdptpt)<0,thenydpt=1.Then,wecanmodifytheobjectivefunction( 3{11 )by(i)replacingFdptwithF0dptwhereF0dpt=maxf(FdptCdptpt);0gfort=1;:::;T;and(ii)addingaconstanttermPTt=1minfFdptCdptpt;0g.Consequently,foragivenand,themodiedsingle-productLagrangiansubproblemisgivenby:minTXt=1(Fmptympt+(pt+pt)xmpt)+TXt=1(F0dptydpt+(pt+pt)xdpt)+TXt=1"ptipt+TXt=1minfFdptCdptpt;0g subjectto( 3{12 )-( 3{17 ). Thismodiedsingle-productLagrangiansubproblemresemblesthesingle-itemuncapacitatedlot-sizingproblemwithreturns(SULSPwR)(in BeltranandKrass ( 2002 ))withonedierence:thereisanitedisposalcostassociatedwithanyexcessinventoryattheendoftheplanninghorizoninthesubproblemabove,whileanyexcessinventoryattheendoftheplanninghorizonhasnovaluein BeltranandKrass ( 2002 ).Thisdierencedoesnotinuencethestructuralpropertiesoftheoptimalsolutionpresentedin BeltranandKrass ( 2002 )butitdoesrequireamodiedDPalgorithm.Next,weproceedwithareviewofthepropertiesoftheoptimalsolutiontotheSULSPwRpresentedin BeltranandKrass ( 2002 ),andthenweextendtheDPalgorithmin BeltranandKrass ( 2002 )toobtainanoptimalsolutiontoourmodiedsingle-productLagrangiansubproblem.Asin BeltranandKrass ( 2002 ),intheremainderofthispaper,aperiodintheplanninghorizoniscalledadecisionperiod,ifareplenishmentordisposalsetuphastobeperformed.

PAGE 34

Asaresult,theplanninghorizoncanbedividedintoregenerationintervalseachofwhichhas Moreover,underamodiedzero-inventorypolicy,itcanbeshownthateachreplenishmentordisposalsetupshouldsatisfythenetdemandofanintegernumberofperiods,i.e.,amodiedzero-inventorypolicysatisestheexactrequirementsproperty. 3-2 andFigure 3-3 ,respectively.Letisaparameter. 34

PAGE 35

Letptdenotethemaximumnumberofperiodsbywhichadecisionperiod(eitherwithareplenishmentordisposalsetup)forproductpcanbepostponedbeyondperiodt.Thatis,wehavept=maxfw:twTandSt0pt0;8tt0wg.Letgp(t)denotetheoptimalvaluefunctionforproductpinperiodt,whichisdenedastheminimumcostofsatisfyingdemandandprocessingreturnfromperiodtuntiltheendoftheplanninghorizonwithzerostartinginventoryforproductp.TherecurrencerelationoftheDPalgorithmcanbestatedasgp(t)=min8>>>>>>>><>>>>>>>>:minv2ft;:::;ptgv1P`=t("pl(S`pt))+(Ipv)minv2ft;:::;ptg8><>:v1P`=t("pl(S`pt))+8><>:0pt=TandSTpt=0(I

PAGE 36

LetLBkpdenotetheoptimalsolutiontothesingle-productLagrangiansubproblemforproductpatiterationkobtainedusingtheDPalgorithmdescribedabove.Then,thelowerboundfortheLagrangianproblematiterationk,LBk,isgivenbyLBk=PXp=1LBkpTXt=1tCmt 3{10 )canbeidentiedecientlywiththeDPapproachdescribedinSection 3.3.1 .Asthisoptimalsolutionmayviolatethereplenishmentanddisposalcapacityrestrictions,wedevelopasmoothingheuristictoremovesuchinfeasibilities.Thisapproachissimilarto,yet,extends,theonepresentedin Trigeiroetal. ( 1989 ).Oursmoothingheuristichasthreephasesandisguaranteedtoremoveallreplenishmentanddisposalcapacityviolationsforafeasibleinstanceoftheproblem.Therstphaseoftheapproachisaforwardpass(i.e.,startsfromthebeginningoftheplanninghorizonandproceedstowardsthelastperiod)thatadjustsdisposalquantitiestoeliminatedisposalcapacityviolations.Thesecondphaseadjustsreplenishmentquantitiestoeliminatereplenishmentcapacityviolationsusingabackwardpass(i.e.,startsfromtheendoftheplanninghorizonandproceedstowardstherstperiod)andanadditionalforwardpassifneeded.Thethird,andlast,phaseeliminatesanyunnecessarysetupsand/orinventoryfortheproductsinthreesteps. 36

PAGE 37

1. 2. (a) Eliminatereplenishmentcapacityviolationsforperiodst>1:Webeginfromthelatestperiodtwherethereplenishmentcapacityisviolatedandconsideralltheproductsthatareproducedinthisperiod.Ifthereplenishmentquantityassociatedwithaproductissmallerthantheamountofcapacityviolation,thenweconsidertwooptionstoshifttheentirereplenishmentquantityto(i)themostrecentperiodwithareplenishmentsetupor(ii)thepreviousperiod.Ifthereplenishmentquantityassociatedwithaproductislargerthantheamountofcapacityviolation,thenweconsiderthreeoptionstoshift(i)theminimumamountpossibletoreduce/eliminatethecapacityviolationtothepreviousperiod;(ii)theentirereplenishmentquantitytothepreviousperiodor(iii)theentirereplenishmentquantitytothemostrecentperiodwithareplenishmentsetup.Foreachoption,wedeterminethemarginalcostofeliminatingthereplenishmentcapacityviolation,whichisobtainedbydividingthetotalcostassociatedwithimplementingtheoptionbythetotalamountthatcanbeshiftedbytheoption.Theseoptionsareexecuteduntilthereisnomorereplenishmentcapacityviolationinthecurrentperiodt.Then,theprocedure 37

PAGE 38

(b) Eliminatereplenishmentcapacityviolationinperiodt=1:Weconsideralltheproductsthatareproducedintherstperiod.Foralltheproducts,weconsidertheoptionofshiftingtheminimumamountpossibletoeliminatetheproductcapacityviolationortheentirereplenishmentquantitytothenextperiod.Wedeterminethemarginalcostofeliminatingthereplenishmentcapacityviolationforeachproduct.Theviolationiseliminatedbyshiftingasmanyproductsasnecessarytothenextperiod(t+1),startingwiththeproductthathasthelowestmarginalcost.Then,theprocedureisrepeatedforperiod(t+1).Attheendofthisstepwhenthecurrentperiodhasnoreplenishmentcapacityviolation,allreplenishmentcapacityviolationsareguaranteedtobeeliminated. 3. (a) Eliminateanyunnecessarysetups:Givenaproduct,westartfromperiodt=T.Ifbothreplenishmentanddisposalsetupsareperformedfortheproductinthecurrentperiod,thenthereplenishmentanddisposalquantitiesfortheproductcanbereducedbytheminimumofthesetwoquantities,i.e.,eectivelyeliminatingthethesetupthatisassociatedwiththesmallerbatchsize.Attheendofthisstepwhenthecurrentperiodist=1,unnecessarysetupshavebeeneliminatedfortheproduct.Theprocedureisrepeatedforalltheproducts. (b) Adjustreplenishmentquantitiestoeliminateanyunnecessaryinventory:Givenaproduct,westartfromperiodt=T.Ifthereisanyunusedreplenishmentcapacity,thentheproductswithpositiveincominginventoryareconsidered.Foreachsuchproduct,thecostreductionassociatedwithincreasingthereplenishmentquantityintanddecreasingthereplenishmentquantityint0(wheret0isthemostrecentreplenishmentsetupperiod,andendinginventoryispositiveforallperiodsfromt0uptoandincluding(t1)fortheproduct)isidentied.Ifthecostreductionispositive,thentheshiftisperformed,i.e.,themaximumamountofreplenishmentforproductpthatcanbemovedfromperiodt0toperiodtismoved.Theprocedurecontinuesuntileithertherearenootherproductswithpositivesavingsorallthereplenishmentcapacityisusedinperiodt.Then,theprocedureisrepeatedforperiod(t1).Attheendofthisstepwhenthecurrentperiodist=1,unnecessaryinventoryduetoreplenishmenthasbeeneliminatedfortheproduct.Thisstepisrepeatedforalltheproducts. (c) Adjustdisposalquantitiestoeliminateanyunnecessaryinventory:Foragivenproduct,westartfromperiodt=1.Ifthereisanyunuseddisposalcapacityandtheinventoryispositiveinallperiodsuptothenextdisposalsetupperiod,t0,thenthecostreductionassociatedwithdecreasingthedisposal 38

PAGE 39

(3{19) whereakisthestepsizeatiterationk.Weupdatethestepsizebyhalvingitsvaluewhenthereisnoimprovementinthelowerboundinaxednumberofiterations. 39

PAGE 40

1. (a) GenerateintegerDptforp=1;:::;Pandt=1;:::;Tuniformlyfrom[(1'D)D;(1+'D)D]. (b) Obtaintherealizedvalueofexpecteddemandforp=1;:::;Pusing^Dp=PTt=1Dpt=T. 2. (a) Settheexpectedvalueofthereturnsperperiodforp=1;:::;PusingRp=RD^Dp. (b) GenerateintegerRptforp=1;:::;Pandt=1;:::;Tuniformlyfrom[(1'R)Rp;(1+'R)Rp]. (c) LetD0+ptandD0ptdenotethepositiveandnegativenetdemandrespectively,i.e.,D0+pt=maxf0;D0ptg,andD0pt=maxf0;D0ptg,forp=1;:::;Pandt=1;:::;T.Similarly,letT+(T)denotethenumberofperiodswherenetdemandis 40

PAGE 41

3. (a) Setthereplenishmentcapacityfort=1;:::;TusingCmt= (b) Setthedisposalcapacityforp=1;:::;Pandt=1;:::;TusingCdpt=bD0minfRpt;D0ptgc. 4. (a) Settheexpectedvalueofreplenishmentsetupcostforp=1;:::;PusingFmp="pt(minf^D0+p;^Dpg)TBR2=2. (b) GenerateFmptforp=1;:::;Pandt=1;:::;Tuniformlyfrom[(1'Fm)Fmp;(1+'Fm)Fmp]. 5. (a) Settheexpectedvalueofdisposalsetupcostforp=1;:::;PusingFdp="pt(minf^D0p;^Rpg)TBD2=2. (b) GenerateFdptforp=1;:::;Pandt=1;:::;Tuniformlyfrom[(1'Fd)Fdp;(1+'Fd)Fdp]. Todevelopasetofrandominstancesofrealisticsize,wevarythenumberofproductsPandthelengthoftheplanninghorizon,T.Specically,weconsiderfourlevelsforP(20,30,40,and50)andthreelevelsforT(12,24,and36).Consequently,wehave12classesoftestinstances.Moreover,weconsidervelevelsforthecorrelationbetweentotaldemand 41

PAGE 42

3-1 .Foreach(P;T)weconsider80dierentproblemsettings,andforeachsettingwegenerate10randomtestinstances.Therefore,wegenerateandsolve9,600instancesintotal. MillarandYang ( 1994 )).Inourexperiments,theLRapproachterminateswhenthegapbetweenthelowerandupperboundislessthan0.5%oramaximumof500iterationsareexecuted.Thestepsizeissetinitiallyto2andhalvedifthelowerboundvalueisnotimprovedinveconsecutiveiterations. WesummarizetheoptimalitygapsobtainedbytheLRapproachanditsCPUrequirementsinTable 3-2 .Aswesolveatotalof9,600testinstances,eachcellinthistablecorrespondstotheaverageormaximumvaluerepresenting800instancesofthesamesize.Wenotethattheobservedaverageoptimalitygapoverallinstancesis0.25percent,whereasthemaximumisslightlyabove2percent.Wealsoobservethatforagivennumberofproducts,theaverageandmaximumvaluesoftheoptimalitygapsbecomesmallerasthelengthoftheplanninghorizonincreases.Foragivenplanninghorizonlength,theaverageandmaximumvaluesoftheoptimalitygapsarenotsignicantlyinuencedbyanincreaseinthenumberofproducts.Whenthenumberofproductsislargeand/ortheplanninghorizonisrelativelylong,(i)theinuenceofa`misplaced'setupontheoverallobjectivefunctionvaluedecreasesand(ii)thesmoothingheuristic 42

PAGE 43

Moreover,weobservethattheLRapproachiscomputationallyecientastheobservedaverageCPUrequirementoverallinstancesislessthanonesecond,whereasthemaximumislessthan22seconds. InTables 4-3 4-4 ,and 3-5 ,weprovideadetailedanalysisoftheinuenceofthedatacharacteristicsontheperformanceoftheLRapproach. Notsurprisingly,weobservethatthequalityoftheoptimalitygapsdeterioratesas(i)theTBRandTBDvaluesincreaseand/or(ii)thereplenishmentand/ordisposalcapacitiesbecomemorerestraining.ItisworthwhiletonotethattheinuenceofanincreaseinTBR(and/oradecreaseinreplenishmentcapacity)appearstobemoreinuentialthananincreaseinTBD(and/oradecreaseindisposalcapacity),whichcanbeattributedtoproblemcharacteristics.Thatis,asthereplenishmentcapacityissharedamongmultipleproducttypes,anincreaseinTBR(oradecreaseinreplenishmentcapacity)inuencesreplenishmentdecisionsformultipleproductssimultaneously,anda`mislocated'replenishmentsetupdecisioninuencestheobjectivefunctionvaluemorethana`mislocated'disposalsetupdecision.Inaddition,weobservethatthecorrelationbetweenreturnsanddemandhasaslightinuenceonthequalityofthegapsobtainedbytheLRapproach. AlthoughtheresultsinTable 3-2 provideempiricalevidencefortheeectivenessoftheproposedLRapproachtoaddresstheRDPP,aninterestingquestionthatremainsunansweredistheeectivenessoftheB&Capproachforobtainingsolutionstotheproblem.Tothisend,weconductedapreliminaryexperimentwherewesolvedaninstanceusingtheLRapproachandrecordedtheCPUtimerequired.Then,settingthetimelimittothisvalue,weattemptedtosolvethesameinstancewithCPLEX.Fornumeroustestinstances,weobservedtheoptimalitygapstobequitelarge.Therefore,weconductedan 43

PAGE 44

3-2 )andthestoppinggapto0.1percent.WesummarizetheresultsofthisexperimentinTable 3-6 Again,eachcellinthistablerepresentstheaverageormaximumvalueof800observationsfortestinstancesofthesamesize.Wenotethattheaverageoptimalitygapoverallinstancesisaround8percentwhereasthemaximumisaround34percent.InTables 3-7 3-8 ,and 3-9 ,weprovideadetailedanalysisoftheinuenceofdatacharacteristicsontheperformanceofCPLEX.Again,weobservethatthequalityofoptimalitygapsdeterioratesas(i)theTBRandTBDvaluesincreaseand/or(ii)thereplenishmentand/ordisposalcapacitiesbecomemorerestraining.WealsoobservethatthequalityoftheoptimalitygapsismoresensitivetochangesintheTBRandTBDvaluesthantothoseinthetightnessofthereplenishmentand/ordisposalcapacities. TotestwhethertheperformanceofCPLEXwouldimproveifthetimelimitwereincreasedandtheoptimalitygapwerereduced,weconductedafollow-upexperimenttoconsiderinstanceswithT=12only.Inparticular,weincreasedthetimelimitto300seconds(i.e.,aten-foldincrease)andincreasedthestoppinggapto0.5percent(i.e.,ave-foldincrease).InTable 3-10 below,wesummarizetheresultsfromthisexperiment;wereportthenumberoftestinstancesthatcannotbesolvedbyCPLEX(i.e.,theoptimalitygapcannotbereducedto0.5percentin300seconds)aswellastheaverageandmaximumoptimalitygapsuponterminationfortheunsolvedinstances.Forexample,weobservethatonly183(outof800)testinstanceswithsize(20,12)weresolvedbyCPLEXwithin30secondswithastoppinggapof0.1percent.Increasingthetimelimitto300secondsandreducingtheoptimalitygapto0.5percentallowedCPLEXtosolve9additionalinstances(i.e.,aonepercentincrease).Keepingthenumberofproductsconstantandincreasingthenumberoftimeperiods,weobservedthatincreasingthetimelimitandthestoppinggapdidnothelpCPLEXobtainsolutionsforadditionaltestinstances. 44

PAGE 45

ThemathematicalformulationinSection 3.2 canbemodiedforthisvariantoftheRDPPbyreplacingconstraintset( 3{3 )withPXp=1pxdptCdtt=1;:::;T whichensuresthatthetotaldisposalcapacityrequiredforallproductsdoesnotexceedtheavailabledisposalcapacityinperiodt. TheproposedLRapproachcaneasilybemodiedaswell.Inparticular,theapproachisnowbasedontherelaxationofconstraintsets( 3{2 )and( 3{21 ).Let=ftgdenotethevectorofLagrangemultiplierst0;t=1;:::;T,eachofwhichisassociatedwiththedisposalcapacityconstraintinperiodt=1;:::;T.TheLagrangianproblemobtainedbyrelaxingconstraintsets( 3{2 )and( 3{21 )forgivenandcanbestatedasminPXp=1TXt=1(Fmptympt+(pt+pt)xmpt)+PXp=1TXt=1(Fdptydpt+(pt+pt)xdpt)+PXp=1TXt=1"ptiptTXt=1(tCmt+tCdt) (3{22)subjectto( 3{4 ){( 3{9 ). 45

PAGE 46

3{22 )isconstant,andtheproblem( 3{22 )isseparableintoPsubproblems,eachofwhichcorrespondstoaproductp.Thesingle-productLagrangiansubproblem(forproductp)nowcanbestatedasminTXt=1(Fmptympt+(pt+pt)xmpt)+TXt=1(Fdptydpt+(pt+pt)xdpt)+TXt=1"ptipt 3{12 ){( 3{17 ). EachofthesesubproblemscanalsobesolvedecientlywiththeDPalgorithmpresentedinSection 3.3.1 ,andalowerboundcanbeobtainedeasily.Toobtaintheupperbounds,wemodifyPhase1ofthesmoothingheuristicpresentedinSection 3.3.2 asdiscussednext.Eliminatedisposalcapacityviolations: (a) Weclassifytheproductsthataredisposedofinthisperiodintotwosets.Theproductsforwhichthedisposalquantityissmallerthanorequaltotheamountofdisposalcapacityviolationareincludedintherstset.Thesecondsetcomprisesoftheproductsforwhichthedisposalquantityislargerthattheamountofdisposalcapacityviolation.Fortheproductsintherstset,weconsidertoshifttheentiredisposalquantityfortheproduct(i)tothenextfutureperiodwithanexistingdisposalsetup,or(ii)tothenextperiod(t+1).Inevaluatingthetotalcostchangeassociatedwiththeseoptions,weaccountforchangesinthetotalunitdisposalandinventoryholdingcosts.Moreover,forthesecondoption,wealsotakeintoaccountthexedcostiftheshiftrequirestheschedulingofanadditionaldisposalsetupinperiod(t+1).Similarly,fortheproductsinthesecondset,weconsidertoshift(i)theamountofcapacityviolationtothenextperiod(t+1),or(ii)theentiredisposalquantitytothenextperiod(t+1),or(iii)theentiredisposalquantitytothenextfutureperiodwithanexistingdisposalsetup.Again,inevaluatingtheseoptions,weaccountforthechangesinthetotalunitdisposalandinventoryholdingcosts.Moreover,fortherstandsecondoptions,wetakeintoaccountthexeddisposalsetupcostiftheshiftrequirestheschedulingofanadditionaldisposalsetupinperiod(t+1). Foreachproduct,wepicktheoptionthatyieldstheminimummarginalcost.Thenwepicktheproduct(and,hence,theoption)leadingtothebestmarginalcostandshifttheamountimpliedbytheoptionfortheproduct.Werepeattheprocedure 46

PAGE 47

(b) Weconsideralltheproductsthataredisposedofinthecurrentperiodandclassifythemintotwosetsasbefore.Foreachproductintherstset(whosedisposalquantityissmallerthanorequaltotheamountofdisposalcapacityviolation),weconsidertoshifttheentiredisposalquantityoftheproducttothepreviousperiod.Foreachproductinthesecondset(whosedisposalquantityislargerthantheamountofdisposalcapacityviolation)weconsidertoshift(i)theamountofcapacityviolationtothepreviousperiod,or(ii)theentirereplenishmentquantitytothepreviousperiod.Inevaluatingtheseoptions,weaccountforthechangesinthetotalunitdisposalandinventoryholdingcostsaswellastheadditionaldisposalsetupcost,ifneeded.Moreover,inidentifyingtheamountthatcanbeshiftedbytheseoptions,webearinmindthatreturnsthatarenotreceivedcannotbedisposedof.Themarginalcostofeachoptionisobtainedbydividingthetotalcostchangeassociatedwithimplementingtheoptionbytheamountthatcanbeshiftedbytheoption. Foreachproduct,wepicktheoptionthatyieldstheminimummarginalcostwhichisobtainedbydividingthetotalcostchangeassociatedwithimplementingtheoptionbytheamountthatcanbeshiftedbytheoption.Then,wepicktheproduct(and,hence,theassociatedoption)leadingtothebestmarginalcostandshifttheamountimpliedbytheoptionfortheproduct.Werepeattheprocedure(startingwithclassifyingtheproductsintotwosets)untilthereisnodisposalcapacityviolationinthecurrentperiodt. Ifthepreviousperiodhasnodisposalcapacityviolation,alldisposalcapacityviolationsareguaranteedtobeeliminated.Otherwise,theroutineisrepeatedforthepreviousperiod. Consideringthelowerandupperboundvaluesateachiterationk,thenewsetofmultiplierscanbeupdatedusing TotestthecomputationaleciencyofourproposedapproachforthisvariantoftheRDPP,weobtainedsetoftestinstancesusingtherandominstancegenerationschemewedescribeinSection 4.4.1 bymodifyingthedisposalcapacityforeachperiod. 47

PAGE 48

3-11 Again,wesolveatotalof9,600instances.AsinTable 3-2 ,eachcellinthistablerepresentstheaverageormaximumvalueof800observationsfortestinstancesofthesamesize.Weobservethattheaverageoptimalitygapoverallinstancesis0.22percent,whereasthemaximumislessthan2percent.Therefore,withtheproposedmodications,theLRapproachcaneasilyberevisedtoobtainhigh-qualitysolutionsforthisvariantoftheRDPPaswell.Uponadetailedinvestigationofthecomputationalresults,itispossibletomakethesameobservationsdiscussedinSection 4.4.2 inregardstotheinuenceofthedatacharacteristicsontheperformanceoftheLRapproachforthisvariantoftheRDPP. TotesttheeectivenessoftheB&Capproachinobtainingsolutionsforthisproblemvariant,wesolvedthesetestinstancesusingCPLEXwithatimelimitof30secondsandastoppinggapof0.1percentasbefore.WesummarizetheresultsofthisexperimentinTable 3-12 .Wenotethattheaverageoptimalitygapoverallinstancesisaround8percentwhereasthemaximumisaround37percent.InTable 3-13 below,wesummarizetheresultsfromanexperimentwhereweincreasedthetimelimitto300secondsandincreasedthestoppinggapto0.5percent. WeobservethatCPLEXisabletoobtaintheoptimalsolutionformoretestinstanceswithsize(20,12)forthisvariantoftheRDPP.However,thenumberofinstancesforwhichCPLEXcandetermineandverifytheoptimalsolutiondecreasesasthenumberofproductsincreases.Therefore,weconcludethatCPLEXisnoteectiveforaddressingthisvariantoftheRDPPeither. 48

PAGE 49

AnimmediateextensionofourworkthatwouldbeofpracticalrelevanceistoconsiderageneralizationoftheRDPPwhere(i)thereplenishmentand/ordisposalsetupcostsdependonthespecicproductsincludedinthesetupand/or(ii)setuptimesareexplicitlymodeled.Moreover,ourmodelscanbeextendedbyconsideringdemandbacklogging.Also,theLRapproachpresentedinthispapercanbeextendedtosettingswherecertaincapacitatedresourcesaresharedbythereplenishmentanddisposalactivities.Anotherpossibleextensionistodeveloprolling-horizonheuristics(possiblywithprovablebounds)fortheRDPP. 49

PAGE 50

Problemsetting. Pseudo-codeoftheIdenticationofSetIpt. Pseudo-codeoftheIdenticationofSetI 50

PAGE 51

Parametervaluesusedincomputationalexperiments. Level(s) Value(s) Constant 100 Variabilityofdemand('D) Constant 0.5 Variabilityofreturns('R) Constant 0.5 Variabilityofxedreplenishmentsetupcost('Fm) Constant 0.2 Variabilityofxeddisposalsetupcost('Fd) Constant 0.2 Unitinventoryholdingcost("pt) Constant 1 Unitreplenishmentcost(pt) Constant 3 Unitdisposalcost(pt) Cost 1 Unitcapacityconsumption(p) Constant 1 Timebetweenreplenishmentsetups(TBR) Low [1,3] High [3,5] Timebetweendisposalsetups(TBD) Low [1,3] High [3,5] Replenishmentcapacitytightness(D0+) Low 1.5 High 2.5 Disposalcapacitytightness(D0) Low 1.5 High 2.5 Return/Demandratio(RD) Low 0.15 High 0.75 Table3-2. PerformanceoftheLRApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). CPURequirement InstanceSize Avr.Gap Max.Gap Avr.Time Max.Time (P;T) (%) (%) (sec.) (sec.) (20,12) 0.30 1.74 0.08 0.55 (30,12) 0.28 2.34 0.07 0.78 (40,12) 0.28 1.41 0.07 1.05 (50,12) 0.28 2.02 0.06 1.25 (20,24) 0.25 1.43 0.15 2.94 (30,24) 0.24 0.91 0.11 4.30 (40,24) 0.24 0.81 0.16 5.83 (50,24) 0.24 0.83 0.17 7.36 (20,36) 0.22 0.68 0.22 8.73 (30,36) 0.22 0.91 0.28 13.19 (40,36) 0.22 0.76 0.27 17.50 (50,36) 0.21 0.62 0.19 21.80 Overall 0.25 2.34 0.15 21.80

PAGE 52

EectofTimeBetweenReplenishment(TBR)andTimeBetweenDisposal(TBD)SetupsontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. Max. (TBR,TBD) Gap(%) Gap(%) (U[1,3],U[1,3]) 0.08 0.86 (U[1,3],U[3,5]) 0.13 2.34 (U[3,5],U[1,3]) 0.37 1.34 (U[3,5],U[3,5]) 0.40 2.02 EectofReplenishmentandDisposalCapacityTightnessontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. Max. ('R;'D) Gap(%) Gap(%) (1.5,1.5) 0.33 2.34 (1.5,2.5) 0.30 1.74 (2.5,1.5) 0.19 2.02 (2.5,2.5) 0.17 1.38 EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLRApproach. Max. Gap(%) 0.15 0.28 1.64 0.30 0.28 1.74 0.45 0.26 1.51 0.60 0.20 0.94 0.75 0.22 2.34 PerformanceofCPLEX(withaTimeLimitofthe30SecondsandaStoppingGapof0.1Percent):AverageandMaximumOptimalityGaps(%). Avr.Gap Max.Gap (P;T) (%) (%) (20,12) 3.85 14.88 (30,12) 4.74 15.85 (40,12) 5.67 18.50 (50,12) 6.20 18.20 (20,24) 7.47 23.47 (30,24) 8.33 27.06 (40,24) 9.13 28.44 (50,24) 10.03 34.80 (20,36) 9.00 28.29 (30,36) 10.61 34.13 (40,36) 11.00 32.01 (50,36) 11.48 32.87 Overall 8.13 34.80

PAGE 53

EectofTimeBetweenReplenishment(TBR)andTimeBetweenDisposal(TBD)SetupsontheAverageandMaximumOptimalityGaps(%)ofCPLEX. Max. (TBR,TBD) Gap(%) Gap(%) (U[1,3],U[1,3]) 4.53 21.48 (U[1,3],U[3,5]) 4.62 22.56 (U[3,5],U[1,3]) 11.68 34.13 (U[3,5],U[3,5]) 11.69 34.80 EectofReplenishmentandDisposalCapacityTightnessontheAverageandMaximumOptimalityGaps(%)ofCPLEX. Max. (R;D) Gap(%) Gap(%) (1.5,1.5) 8.31 32.87 (1.5,2.5) 8.30 34.13 (2.5,1.5) 7.96 34.13 (2.5,2.5) 7.93 34.80 EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)ofCPLEX. Max. Gap(%) 0.15 3.04 12.16 0.30 5.62 34.13 0.45 8.45 23.74 0.60 10.95 34.80 0.75 12.57 32.87 InuenceofIncreasingtheTimeLimitandtheStoppingGaponthePerformanceofCPLEX:NumberofUnsolvedInstances,AverageandMaximumOptimalityGaps(%). 300Secondsand0.5%StoppingGap No.ofInstances GapUponTermination No.ofInstances GapUponTermination Unsolved Avr.Gap Max.Gap Unsolved Avr.Gap Max.Gap (P;T) (outof800) (%) (%) (outof800) (%) (%) (20,12) 617 4.33 14.88 608 3.85 13.16 (20,24) 774 7.57 23.47 738 7.22 21.34 (20,36) 800 9.00 28.29 800 8.40 26.06

PAGE 54

PerformanceoftheLRApproachfortheRDPPwithSharedReplenishmentandDisposalCapacities:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). CPURequirement InstanceSize Avr.Gap Max.Gap Avr.Time Max.Time (P;T) (%) (%) (sec.) (sec.) (20,12) 0.26 1.74 0.07 0.70 (30,12) 0.24 1.31 0.07 0.84 (40,12) 0.24 0.91 0.07 1.08 (50,12) 0.24 0.80 0.08 1.31 (20,24) 0.23 1.21 0.21 3.17 (30,24) 0.22 0.66 0.20 4.97 (40,24) 0.22 0.60 0.24 6.50 (50,24) 0.22 0.50 0.22 4.34 (20,36) 0.28 0.81 0.40 9.48 (30,36) 0.20 0.64 0.35 14.77 (40,36) 0.20 0.50 0.38 10.83 (50,36) 0.19 0.50 0.34 9.84 Overall 0.22 1.74 0.22 14.77 PerformanceofCPLEXfortheRDPPwithSharedReplenishmentandDisposalCapacities(withaTimeLimitofthe30secondsandaStoppingGapof0.1Percent):AverageandMaximumOptimalityGaps(%). Avr.Gap Max.Gap (P;T) (%) (%) (20,12) 3.89 14.62 (30,12) 4.78 16.37 (40,12) 5.72 18.39 (50,12) 6.25 18.84 (20,24) 7.51 23.68 (30,24) 8.38 26.92 (40,24) 9.20 28.01 (50,24) 10.13 36.97 (20,36) 8.99 28.45 (30,36) 10.55 32.54 (40,36) 11.13 32.20 (50,36) 11.50 28.91 Overall 8.17 36.97

PAGE 55

InuenceofIncreasingtheTimeLimitandtheStoppingGaponthePerformanceofCPLEXfortheRDPPwithSharedReplenishmentandDisposalCapacities:NumberofUnsolvedInstances,AverageandMaximumOptimalityGaps(%). 300Secondsand0.5%StoppingGap No.ofInstances GapUponTermination No.ofInstances GapUponTermination Unsolved Avr.Gap Max.Gap Unsolved Avr.Gap Max.Gap (P;T) (outof800) (%) (%) (outof800) (%) (%) (20,12) 616 4.38 14.62 579 4.04 13.76 (20,24) 774 7.61 23.68 738 7.26 21.56 (20,36) 800 8.99 28.45 800 8.42 26.34

PAGE 56

Sinceautomotivepartsremanufacturinghasbeenapopularpracticesincetheinceptionoftheautomotiveindustry,remanufacturableautomotivepartsarepurchasedandsoldascommodities.Therefore,insomecases,theretailersmaysellsomeofthepartstoindependentremanufacturersandreturnonlyafractionofthepartstotheOEM,whichlimitsthesupplyofremanufacturableparts.Insomeothercases,thequantityofremanufacturablepartsreceivedmaybemorethanthedemandforremanufacturedparts,inwhichcasetheOEMdisposesoftheremanufacturablepartsbysellingthemtobulkmetalrecyclersformaterialrecovery.Inordertoensureuninterruptedreplacement 56

PAGE 57

Uponacloserexaminationofautomotivepartsremanufacturingpractices,weobservethatthereareseveralfactorsthatinuencethecomplexityoftheinventoryandproductionplanningproblemsencounteredinthiscontext.Afactorthatinuencesthecomplexityoftheunderlyingproblempertainstothemanufacturingandremanufacturingcapacities.Thereareusuallyprocessingcapacityrestrictionsthatlimitthequantityofnewandremanufactureditemsthatcanbeacquiredfromnewpartandremanufacturingsuppliers.Someautomotiveparts(e.g.,enginesandtransmissions)aremanufacturedbytheOEManddierentpartssharethesamemanufacturingcapacity.Someotherparts(e.g.,alternators,distributors)arepurchasedfrompartsuppliersandhencetheydonotsharethesamemanufacturingcapacity.Remanufacturingofusedpartsareusuallyoutsourcedtoremanufacturingsuppliers.Typically,remanufacturingofsometypesofparts(e.g.,transmissionsortorqueconverters)areoutsourcedtoasingleremanufacturingsuppliertotakeadvantageoflearning-by-doingeectsandtobeabletonegotiatealowerunitcostduetoconsolidatedvolume.Therefore,somepartssharetheavailableremanufacturingcapacity.Forotherparts(e.g.,electroniccontrolunits,distributors),however,remanufacturingcanbeoutsourcedtodierentremanufacturingsuppliers,eachofwhichhashisownprocessingcapacity.Anadditionalrelatedfactorconcernstheavailabledisposalcapacity.Theremaybetransportationand/orrecyclingcapacityconsiderationsthatlimitthequantityofusedpartsthatcanbedisposedof. Inthischapter,weconcentrateontheManufacturing,RemanufacturingandDisposalPlanningProblem(MRDPP)withmultipleproducttypestoaddresstheinventoryandproductionplanningproblemforremanufacturableautomotiveparts.Morespecically,MRDPPisconcernedwithdeterminingwhenandhowmuchtomanufacture,remanufacture,anddisposeofeachtypeofproduct.Weconsidertheproblemina 57

PAGE 58

whileconsideringthemanufacturing,remanufacturing,anddisposalcapacitiesexplicitly. Theremainderofthischapterisorganizedasfollows.Section 4.2 presentsthemodellingassumptionsandamathematicalformulationfortheMRDPP.SolutionapproachesaredescribedinSection 4.3 .ResultsfromcomputationalexperimentsarepresentedandanalyzedinSection 4.4 .Finally,conclusionsandfutureresearchdirectionsaresummarizedinSection 4.5 4-1 .Therearetwostockpointsforeachproducttype,onefortheusediteminventory(UII),andonefortheserviceable(manufacturedorremanufactured)itemsinventory(MRII).UIIaccumulatesasreturnsofproductsarereceived.ExcessreturnsmaybedisposedoffromtheUII.ManufacturingandremanufacturingactivitiesreplenishtheMRII,andthedemandfortheproducttypeissatisedfromtheMRII.Aswenotedearlier,wefocusondiscretetimedynamicdemandandreturnmodelstoaddresstheinventoryandproductionplanningproblemencounteredinthissetting.Ourobjectiveistominimizethesumofthe 58

PAGE 59

Inourwork,wemakethefollowingmodelingassumptions: 1. Demandforallproducttypesthroughouttheplanninghorizonisknown.Demandforeachproducttypecanbesatisedbyusingremanufacturinguseditemsorbymanufacturingnewitems. 2. Returnsforallproducttypesthroughouttheplanninghorizonisalsoknown.ReturnsforeachproducttypecanberemanufacturedtoreplenishtheMRII(i.e.,satisfythedemand)orbedisposedof. 3. Manufacturingactivityincursaxedsetupandlinearvariablecosts.Theavailablemanufacturingcapacityissharedamongmultipleproducttypes. 4. Remanufacturingactivityincursaxedsetupandlinearvariablecosts.Thereisaprocessingcapacityassociatedwiththeremanufacturingactivityforeachproducttype. 5. Disposalactivityincursaxedsetupandlinearvariablecosts.Thereisaprocessingcapacityassociatedwiththedisposalactivityforeachproducttype. 6. ItemsinMRII-pofproducttypepincuralinearinventoryholdingcost. 7. ItemsinUII-pofproducttypepincuralinearinventoryholdingcost. 8. Backloggingofdemandisnotallowed. Theproblemparametersaredenedasfollows: 59

PAGE 60

Wehavethefollowingdecisionvariables: 60

PAGE 61

AmathematicalprogrammingformulationforMRDPPisasfollows:minPXp=1TXt=1(Fmptympt+cmptxmpt+Frptyrpt+crptxrpt+Fdptydpt+cdptxdpt+kptjpt+hptipt) (4{1) subjecttoip;t1+xmpt+xrptDpt=iptp=1;:::;P;t=1;:::;T; (4{2)jp;t1+Rptxrptxdpt=jptp=1;:::;P;t=1;:::;T; (4{3)PXp=1pxmptKmtt=1;:::;T; (4{4)xrptKrptyrptp=1;:::;P;t=1;:::;T; (4{5)xdptKdptydptp=1;:::;P;t=1;:::;T; (4{6)xmpt (4{7)xrpt (4{8)xdpt (4{9)jp0;jpT;ip0;ipT=0p=1;:::;P; (4{10)xmpt;xrpt;xdpt;jptipt0andintegerp=1;:::;P;t=1;:::;T; (4{11) 61

PAGE 62

Objectivefunction( 4{1 )minimizesthesumofxedcostofperformingmanufacturinganddisposalsetupsaswellaslinearmanufacturing,disposal,andinventoryholding(inUIIandMRII)costsforallproductsthroughouttheplanninghorizon.Constraintset( 4{2 )givestheinventorybalanceequationsforMRIIforproductpineachperiodt,whichensuresthatthesumofthequantityininventorycarriedfromperiodt1toperiodt1,thequantitymanufacturedinperiodt,andthequantityremanufacturedinperiodtisequaltothesumofthedemandinperiodtandthequantityininventorycarriedfromperiodttoperiodt+1.Constraintset( 4{3 )istheinventorybalanceequationforUIIforproductpineachperiodt,whichensuresthatthesumofthequantityininventorycarriedfromperiodt1toperiodtandthequantityofreturnsreceivedinperiodtisequaltothesumofthequantityremanufacturedinperiodt,thequantitydisposedinperiodtandthequantityininventorycarriedtofromperiodttoperiodt+1.Constraintset( 4{4 )ensuresthatthemanufacturingcapacityrequiredforallproductsinperiodtdoesnotexceedtheavailablemanufacturingcapacityinperiodt.Constraintset( 4{5 )ensuresthattheremanufacturingcapacityrequiredforproductpinperiodtdoesnotexceedtheavailableremanufacturingcapacityforproductpinperiodt.Constraintset( 4{6 )ensuresthatthedisposalcapacityrequiredforproductpinperiodtdoesnotexceedtheavailabledisposalcapacityforproductpinperiodt.Constraintset( 4{7 )ensuresthatamanufacturingsetupforproductpisperformedinperiodtifthequantitymanufacturedforproductpinperiodtispositiveandthisquantityshouldnotexceedthecumulativedemanduntiltheendoftheplanninghorizon.Constraintset( 4{8 )ensuresthataremanufacturingsetupisperformedforproductpinperiodtifthequantityremanufacturedforproductpinperiodtispositiveandthisquantityshouldnotexceedthecumulativedemanduntiltheendoftheplanninghorizon.Similarly,constraintset( 4{9 )ensuresthatadisposalsetupforproductpisperformedinperiodtifthequantityofproductpdisposedofin 62

PAGE 63

4{10 )initializesthebeginningandendinginventorylevelsforallproductsinUIIandMRII.Constraintsets( 4{11 )and( 4{12 )ensuretheintegralityofthedecisionvariables. Capacitatedmulti-itemlotsizingproblemisaspecialcaseofMRDPPproblem.Morespecically,whenthereturnquantityiszeroineveryperiodforeveryproduct,theproblemisequivalenttheclassicalcapacitatedmulti-itemlotsizingproblem.Sinceclassicalcapacitatedmulti-itemlotsizingproblemisNP-Hard,soisMRDPP.Therefore,wefocusonobtainingnear-optimalsolutionsforMRDPP.TothisendwedevelopaclassofsolutionalgorithmsbasedonLagrangianDecompositionandRelaxationtechniquesusedinconjunctionwithGeneticAlgorithms. IntheLagrangianDecomposition(LD)approach,weusevariableredenitionandrelaxationofsomeconstraintsetstodecomposetheproblemintopolynomiallysolvablesubproblems.WeobservethattheremanufacturingquantitydecisionvariablesconnecttheUIIandMRIIinventorybalanceconstraintsets.Hence,weredenetheremanufacturingquantityandsetupvariablesandreplacetheremanufacturingvariableintheUIIinventorybalanceconstraintset.WekeeptheoriginalremanufacturingvariablesintheMRIIinventorybalanceconstraintset,whereaswereplacetheremanufacturingvariablesintheUIIinventorybalanceconstraintsetwiththenewremanufacturingvariables.Weaddanequality(linkage)constraintsettodenetheequivalenceoftheexistingandredeneddecisionvariablestotheoriginalproblemformulation.Wethenrelaxthissetof 63

PAGE 64

IntheLagrangianRelaxationapproach,weonlyrelaxthemanufacturingcapacityconstraintset( 4{4 ).TheresultingLagrangiansubproblemsaresingleitemMRDPPproblems,whichareNP-Hardintheirownright.ForagivensetofLagrangianmultipliers,wesolvethesesubproblemsusingCPLEXtoobtainalowerboundsolutionoftheMRDPP.Weobtainupperboundsolutionsutilizingaconstructiveheuristic,whichisasmoothingheuristicthatmakesuseofthelowerboundsolutionsobtainedfromCPLEX.Again,weupdatetheLagrangianmultipliersbythesubgradientoptimizationapproachusinglowerandupperboundsolutions. Next,weanalyzethesetwoapproachesindetailbygivingthecorrespondingLagrangianproblemformulations,algorithmsdevelopedtoobtainlowerboundsandupperbounds,andthesubgradientoptimizationapproach. 64

PAGE 65

Wealsodenenewcostparametersforthenewdecisionvariables: Letcupt=Fupt=0forp=1;:::;P,andt=1;:::;T.Thenewformulationcanbestatedasfollows:minPXp=1TXt=1(Fmptympt+cmptxmpt+Frptyrpt+crptxrpt+Fdptydpt+cdptxdpt+kptjpt+hptipt+Fuptyupt+cuptxupt) (4{13) subjecttoip;t1+xmpt+xrptDpt=iptp=1;:::;P;t=1;:::;T; (4{14)jp;t1+Rptxuptxdpt=jptp=1;:::;P;t=1;:::;T; (4{15)PXp=1pxmptKmtt=1;:::;T; (4{16)xrptKrptyrptp=1;:::;P;t=1;:::;T; (4{17)xdptKdptydptp=1;:::;P;t=1;:::;T; (4{18)xmpt (4{19)xrpt (4{20)xdpt (4{21) 65

PAGE 66

(4{22)xrpt=xuptp=1;:::;P;t=1;:::;T; (4{23)yrpt=yuptp=1;:::;P;t=1;:::;T; (4{24)jp0;jpT;ip0;ipT=0p=1;:::;P; (4{25)xmpt;xrpt;xupt;xdpt;jptipt0andintegerp=1;:::;P;t=1;:::;T; (4{26)ympt;yrpt;yupt;ydpt2f0;1gp=1;:::;P;t=1;:::;T: NotethatweexchangethexrptdecisionvariableswithxuptinUIIinventorybalanceconstraintset.Constraintset( 4{23 )ensuresthatxrptandxupttakethesamevalues.Similarly,constraintset( 4{24 )ensuresthatyrptandyupttakethesamevalues.Wealsoaddauxiliaryconstraintset( 4{22 )todeneboundsforxuptandyupt.Since,theunitcostofremanufacturingforxupt,andthexedcostforyuptarezero,theformulation,i.e.,( 4{13 )-( 4{27 ),isequivalenttotheoriginalformulation,i.e,( 4{1 )-( 4{12 ). Wethenrelaxtheconstraintsetsassociatedwithmanufacturing,remanufacturinganddisposalcapacities,i.e.,( 4{16 )-( 4{18 )aswellasthelinkageconstraintsetsbetweenremanufacturingdecisionvariables,i.e.( 4{23 )-( 4{24 ).Whenconstraintsets( 4{16 ),( 4{17 )and( 4{18 )arerelaxed,theresultingrelaxedproblembecomesseparableforeachproducttypepforp=1;:::;P.Moreover,whenconstraintsets( 4{23 )and( 4{24 )arerelaxed,theresultingrelaxedproblembecomesseparableforforwardandreversechannelsubproblems.Let=ftgdenotethevectorofLagrangianmultiplierst0;t=1;:::;T,associatedwiththemanufacturingcapacityconstraintset( 4{16 ).Let=fptg,=fptgdenotethevectorofLagrangianmultiplierspt;pt0;p=1;:::;P;t=1;:::;T,associatedwiththeproduct-specicremanufacturinganddisposalcapacityconstraintsets( 4{17 )and( 4{18 ),respectively.Similarlylet=fptg;!=f!ptgdenotethevectorsofLagrangianmultiplierspt;!pt2(;1);p=1;:::;P;t=1;:::;T,associatedwiththeconstraintssets( 4{23 )and( 4{24 ).Given;;,and!,theLagrangianproblemobtainedupon 66

PAGE 67

4{16 )-( 4{18 ),and( 4{23 )-( 4{24 )canbestatedasfollows:minPXp=1TXt=1(Fmptympt+(cmpt+t)xmpt+(FrptKrptpt)yrt+(crpt+pt+pt)xrpt+(FdptKdptpt)ydpt+(cdpt+pt)xdpt(Fupt!pt)yupt+(cupt!pt)xupt+kptjpt+hptipt)TXt=1(tKmt) (4{28) subjectto( 4{14 )-( 4{15 ),( 4{19 )-( 4{22 ),( 4{25 )-( 4{27 ). ForagivensetofLagrangianmultipliers,thesolutionofLagrangianproblemyieldsalowerboundsolution. WeprovidethespecicoutlineofourLDimplementationinFigure 4-2 where`,anddenotetheiterationcounters.Theterminationcriteriaaredenedby,thetolerancegapbetweenthebestlowerandupperbounds,and`max,themaximumnumberofiterationsofLD.UponthecompletionofmaxLDiterations,newindividualsaddedtoGApopulation.UponthecompletionofmaxLDiterations,GAisexecuted.Moreover,UB`andLB`denotetheupperandlowerbounds,respectively,atiteration`.Similarly,UBandLBdenotetheincumbentupperandlowerbounds,respectively.WeproceedwithadetaileddiscussionofhowwesolvetheLagrangianproblemtoobtainalowerbound(Section 4.3.1.1 ),identifyanupperboundusingthesmoothingheuristic(Section 4.3.1.2 )andGA,andupdatetheLagrangianmultipliers(Section 4.3.1.3 ),oneachiterationoftheLDapproach. 4{28 )canbesolvedeciently.First,wenotethatthelasttermintheobjectivefunctionofthisproblemisconstant.Furthermore,theproblem( 4{28 )isseparableintoPsubproblems.Sincethelinkbetweenthereturnandthedemandisbrokenbyrelaxingtheconstraintsets( 4{23 )-( 4{24 ),eachsubproblemcorrespondingtoaproducttypepisseparableintotwosubproblemsthatcorrespondtotheforwardandreversechannelactivitiesfortheproduct.LetLR1SP1p

PAGE 68

Thesingle-productforward-channelLagrangiansubproblemforproducttypep(LR1SP1p)canbestatedasfollows:minTXt=1(Fmptympt+(cmpt+t)xmpt+(FrptKrptpt)yrpt+(crpt+pt+pt)xrpt+hptipt) (4{29) subjecttoip;t1+xmpt+xrptDpt=iptt=1;:::;T; (4{30)xmpt (4{31)xrpt (4{32)ip0;ipT=0 (4{33)xmpt;xrpt;ipt0andintegert=1;:::;T; (4{34)ympt;yrpt2f0;1gt=1;:::;T: Notethatthecoecient(FrptKrptpt)ofbinaryvariableyrptintheobjectivefunction( 4{29 )canbenegativesincept0.Weobservethatintheoptimalsolutiontothisproblem,if(FrptKrptpt)<0,thenyrpt=1.Asaresult,wecanmodifytheobjectivefunction( 4{29 )by(i)replacingFrptwithF0rptwhereF0rpt=maxf(FrptKrptpt);0gfort=1;:::;T;and(ii)addingaconstanttermPTt=1minfFrptKrptpt;0g.Consequently,foragiven,,,andthemodiedLR1SP1pforproductpisgivenby:minTXt=1(Fmptympt+(cmpt+t)xmpt+(F0rpt)yrpt+(crpt+ptpt)xrpt+hptipt) (4{36) subjectto( 4{30 )-( 4{35 ) ThemodiedLR1SP1pforproducttypepissameasthesingle-itemuncapacitatedlot-sizingwithtwosupplysources.Therefore,thezero-inventorypropertyissatisedforthisproblem.WedevelopanO(T2)DPalgorithmsimilartoWagner-Whitinalgorithmto 68

PAGE 69

Letfp(t)denotetheoptimalvaluefunctionforproductpinperiodt,whichisdenedastheminimumcostofsatisfyingdemandfromperiodtthroughtheendoftheplanninghorizonwithzerostartinginventoryforproductp.TherecurrencerelationoftheDPalgorithmcanbestatedasfollows:fp(t)=min8>>>>>><>>>>>>:minv=t;:::;TFmpt+(cmpt+pt)Svpt+v1Pl=thplSvpl+1+fp(v+1)minv=t;:::;TF0rpt+(crpt+Krptpt+!pt)Svpt+v1Pl=thplSvpl+1+fp(v+1) 69

PAGE 70

(4{37) subjecttojp;t1+Rptxuptxdpt=jptt=1;:::;T; (4{38)xdpt (4{39)xupt (4{40)jp0;jpT=0 (4{41)xdpt;xupt;jpt0andintegert=1;:::;T; (4{42)yupt;ydpt2f0;1gt=1;:::;T: Notethatthecoecient(FdptKdptpt)ofbinaryvariableydpt,and(Fupt!pt)ofbinaryvariableyuptintheobjectivefunction( 4{37 )canbenegativesincept0and!>.Weobservethatintheoptimalsolutiontothisproblem,if(FdptKdptpt)<0,thenydpt=1.Similarly,if(Fupt!pt)<0,thenyupt=1.Accordingly,wecanmodifytheobjectivefunction( 4{37 )by(i)replacingFdptwithF0dpt,andFuptwithF0uptwhereF0dpt=maxf(FdptKdptpt);0gandF0upt=maxf(Fupt!pt);0gfort=1;:::;T;and(ii)addingaconstanttermPTt=1min(fFdptKdptpt;0g+fFuptKupt!pt;0g).Consequently,forgiven,,,,!themodiedLR1SP2pisgivenby:minTXt=1(F0dptydpt+(cdpt+pt)xdpt+F0uptyupt+(cupt!pt)xupt+kptjpt) (4{44) subjectto( 4{38 )-( 4{43 ) NotethatmodiedLR1SP2pissameastheuncapacitatedlotsizingwithtwosupplysources,wherereturnsareprocessedbyremanufacturingordisposalactivities.Wedevelop 70

PAGE 71

Letfp(t)denotetheoptimalvaluefunctionforproductpinperiodt,whichisdenedastheminimumcostofprocessingreturnfromperiod1throughperiodtwithzeroendinginventoryforproductp.TherecurrencerelationoftheDPalgorithmcanbestatedasfollows:fp(t)=min8>>>>>><>>>>>>:minv=1;:::;tF0upt+(cupt!pt)Rtpv+t1Pl=vkplRlpv+fp(v1)minv=t;:::;TF0dpt+(cdpt+Kdptpt)Rtpt+t1Pl=v(kplRlpv)+fp(v1) 71

PAGE 72

4{13 )canbeidentiedecientlywiththeDPapproachesdescribedinSection 4.3.1.1 .However,thisoptimalsolutionmayviolatethemanufacturing,remanufacturing,anddisposalcapacityrestrictions,aswellaslinkageconstraintsetsforremanufacturingquantityandsetupvariables,i.e.,( 4{23 )and( 4{24 ).Moreover,thequantityofremanufactureditemsmayexceedtheavailablequantityofreturns.Weemployasmoothingheuristictoobtainupperboundoneachiteration.WeuseboththelowerboundsolutionsandsolutionsobtainedaftersmoothingheuristicateveryxnumberofiterationstogenerateaninitialpopulationfortheGA.Betweeneveryxednumberofiterations,weusethegeneticalgorithmtoimprovethecurrentbestupperbound.Moreover,attheendoftheLagrangianiterations,weuseGAagain.BelowweexplainthedetailsofthesmoothingheuristicandtheGA. 4.3.1.1 mayviolatethemanufacturing,remanufacturing,anddisposalcapacityrestrictions,aswellasthelinkageconstraintsfortheremanufacturingquantityandthesetupvariables.NotethattheremanufacturingquantitiesfromthesolutionofLR1SP1pmayviolatetheavailablereturnquantity,similarlytheremanufacturingquantitiesfromthesolutionofLR1SP2pmayviolatethedemandquantity.Hence,weapplythesmoothingheuristictothesetwosolutionsdierbyonlytheremanufacturingquantities,wheretherstone(yr;xr)isobtainedbythesolutionofLR1SP1pandthesecondone(yu;xu)isobtainedbythesolutionofLR1SP2p.Werstxtheremanufacturingquantityandthesetupvariablesforbothsolutions,thenapplythesameproceduretoobtaintwoupperboundstoMRDPPaswellastwoindividualsforthepopulationoftheGA. 72

PAGE 73

73

PAGE 74

Theaboveprocedureisrepeatedforalltheproductswithremanufacturingcapacityand/oravailablereturnorrequireddemandviolations.2.EliminatingUIIandMRIIinventorybalanceviolations: 74

PAGE 75

75

PAGE 76

(a) Weclassifytheproductsthataremanufacturedinthisperiodintotwosets.Theproductsforwhichthemanufacturingquantityissmallerthanorequaltotheamountofmanufacturingcapacityviolationareincludedintherstset.Thesecondsetcomprisesoftheproductsforwhichthemanufacturingquantityislargerthattheamountofmanufacturingcapacityviolation.Foreachproductintherstset,weconsidertoshifttheentiremanufacturingquantityoftheproduct(i)tothemostrecentperiodwithanexistingmanufacturingsetup,or(ii)tothepreviousperiod(t1).Inevaluatingthetotalcostchangeassociatedwiththeabovetwooptions,weaccountforchangesinthetotalunitmanufacturingandMRIIinventoryholdingcosts.Moreover,forthesecondoption,wealsotakeintoaccountthexedcostiftheshiftrequirestheschedulingofanadditionalmanufacturingsetupinperiod(t1).Similarly,foreachproductinthesecondset,weconsidertoshift(i)theamountofcapacityviolationtothepreviousperiod(t1),or(ii)theentiremanufacturingquantitytothepreviousperiod(t1),or(iii)theentiremanufacturingquantitytothemostrecentperiodwithanexistingmanufacturingsetup.Again,inevaluatingtheseoptions,weaccountforthechangesinthetotalunitmanufacturingandMRIIinventoryholdingcosts.Moreover,fortherstandsecondoptions,wetakeintoaccountthexedmanufacturingsetupcostiftheshiftrequirestheschedulingofanadditionalmanufacturinginperiod(t1). Foreachproduct,wepicktheoptionthatyieldstheminimummarginalcostwhichisobtainedbydividingthetotalcostchangeassociatedwithimplementingtheoptionbytheamountthatcanbeshiftedbytheoption.Then,wepicktheproduct(and,hence,theassociatedoption)leadingtothebestmarginalcostandshifttheamountimpliedbytheoption.Werepeattheprocedure(startingwithclassifyingtheproductsintotwosets)untilthereisnomanufacturingcapacityviolationinthecurrentperiodt. Then,theroutineisappliedforthepreviousperiodwhichhasacapacityviolation,andthisiscontinueduntiltherstperiodisreached.Whenthecurrentperiodist=1,allmanufacturingcapacityviolationsareguaranteedtobeeliminatedwith,possibly,theexceptionoftherstperiod. (b) Weconsideralltheproductsthataremanufacturedintheperiodandclassifythemintotwosetsasbefore.Foreachproductintherstset(whosemanufacturingquantityissmallerthanorequaltotheamountofmanufacturingcapacityviolation),weconsidertoshifttheentiremanufacturingquantityoftheproducttothenextperiod.Foreachproductinthesecondset(whosemanufacturingquantityislarger

PAGE 77

Foreachproduct,wepicktheoptionthatyieldstheminimummarginalcostwhichisobtainedbydividingthetotalcostchangeassociatedwithimplementingtheoptionbytheamountthatcanbeshiftedbytheoption.Then,wepicktheproduct(and,hence,theassociatedoption)leadingtothebestmarginalcostandshifttheamountimpliedbytheoptionfortheproduct.Werepeattheprocedure(startingwithclassifyingtheproductsintotwosets)untilthereisnomanufacturingcapacityviolationinthecurrentperiodt. Ifthenextperiodhasnomanufacturingcapacityviolation,allmanufacturingcapacityviolationsareguaranteedtobeeliminated.Otherwise,theroutineisrepeatedforthenextperiod.5.Eliminateunnecessaryinventory: (a) Westartfromthelatestperiodtwherethereisamanufacturingsetupandunusedmanufacturingcapacity.WeconsideralltheproductsthataremanufacturedinthisperiodandhavepositiveincomingMRIIinventorytotheperiod. Foreachsuchproduct,wecomputethepotentialcostreductionassociatedwithincreasingthecorrespondingmanufacturingquantityintanddecreasingitint0(wheret0istheproduct'smostrecentmanufacturingsetupperiod,andtheproduct'sendinginventoryispositiveforallperiodsfromt0uptoandincluding(t1)).Then,wepicktheproductwiththelargestcostreductionandshiftthemaximumamountpossibleforthisproductfromt0totbearinginmindthemanufacturingcapacitiesinbothperiodsaswellastheinventorybalanceconstraintsfromt0tot.Here,themaximumamountpossibleisdictatedbytheminimumof(i)theamountofincominginventoryoftheproducttoperiodt,(ii)themanufacturingquantityinperiodt0,and(iii)theavailablemanufacturingcapacityinperiodt. Theprocedureiscontinueduntileitherthereisnootherproductleadingtoacostreductionorallthemanufacturingcapacityisused.Then,theprocedureisrepeatedforthepreviousperiodwithamanufacturingsetupandunusedmanufacturingcapacity. Whenthecurrentperiodist=1,ortherearenootherperiodswithamanufacturingsetupandunusedmanufacturingcapacity,unnecessaryinventoryduetoearlymanufacturinghasbeeneliminatedforalltheproducts. 77

PAGE 78

Givenaproduct,westartfromthelatestperiodtwherethereisaremanufacturingsetupfortheproduct.IfthereisanyunusedcapacityinperiodtandtheincomingMRIIinventoryispositiveinallperiodsfromthepreviousremanufacturingsetupperiodt0fortheproduct,thenwecomputethepotentialcostreductionassociatedwithdecreasingtheremanufacturingquantityint0andincreasingtheremanufacturingquantityint.Ifthecostreductionispositive,thenweshiftthemaximumamountpossiblethatcanbemovedfortheproductfromt0totbearinginmindtheremanufacturingcapacitiesinbothperiodsaswellastheUIIandMRIIinventorybalanceconstraintsfromt0tot. Here,themaximumamountpossibleisdictatedbytheminimumof(i)theminimumamountofMRIIinventoryoftheproductfromperiodt0toperiodt,(ii)theremanufacturingquantityinperiodt0,and(iii)theavailableremanufacturingcapacityinperiodt.Theprocedureisrepeatedforthepreviousperiodwitharemanufacturingsetupandunusedremanufacturingcapacity.Whenthecurrentperiodist=1ortherearenootherperiodswitharemanufacturingsetupandunusedremanufacturingcapacityfortheproduct,unnecessaryMRIIinventoryduetoearlierremanufacturinghasbeeneliminatedfortheproduct. Aboveroutineisrepeatedforalltheproducts. (c) Givenaproduct,westartfromtheearliestperiodtwherethereisadisposalsetupfortheproduct.IfthereisanyunuseddisposalcapacityinperiodtandtheUIIinventoryispositiveinallperiodsuptothenextdisposalsetupperiodt0fortheproduct,thenwecomputethepotentialcostreductionassociatedwithdecreasingthedisposalquantityint0andincreasingthedisposalquantityint.Ifthecostreductionispositive,thenweshiftthemaximumamountpossiblethatcanbemovedfortheproductfromt0totbearinginmindthedisposalcapacitiesinbothperiodsaswellastheUIIinventorybalanceconstraintsfromttot0.Here,themaximumamountpossibleisdictatedbytheminimumof(i)theminimumamountofUIIinventoryoftheproductfromperiodttoperiodt0,(ii)thedisposalquantityinperiodt0,and(iii)theavailabledisposalcapacityinperiodt.Theprocedureisrepeatedforthenextperiodwithadisposalsetupandunuseddisposalcapacity.Whenthecurrentperiodist=Tortherearenootherperiodswithadisposalsetupandunuseddisposalcapacityfortheproduct,unnecessaryUIIinventoryduetodelayeddisposalhasbeeneliminatedfortheproduct. Aboveroutineisrepeatedforalltheproducts. Recallthatweemploytheaboveheuristicusingremanufacturingquantitiesxrorxu)separately.Therefore,weobtaintwoupperboundsolutions.Ifanyoftheseupper 78

PAGE 79

4{2 )and( 4{3 ),withsomepenaltyaddedtotheobjectivefunction. TogeneratethepopulationforaniterationofGA,weusethelowerboundsolutionandsmoothingheuristic.Weobtainfourchromosomesateveryxednumberofiterations.Therstchromosomeisobtainedusingthesetupvariables'valuescorrespondingtothemanufacturingandtheremanufacturingfromthesolutiontotherstsubproblem,andthesetupvariables'valuescorrespondingtothedisposalfromthesolutionofthesecondsubproblem,i.e.,[ym;yr;yd].Thesecondchromosomeisobtainedbyusingthesetupvariables'valuesofmanufacturingfromthesolutionoftherstsubproblem,andthesetupvariables'valuesofremanufacturinganddisposalfromthesolutionofthesecondsubproblem,i.e.,[ym;yu;yd].Thethirdandfourthchromosomesareobtainedfromthemodicationofthesolutionsobtainedfromlowerboudingusingthesmoothingheuristics. AfteraxednumberofLagrangianiterationsareexecuted,GAisemployedusingthegeneratedpopulationuptothatiteration.Weusetheobjectivefunctionvalueasatnessvalueforeachindividualinthepopulation.Fromthecurrentpopulationwerstchoose 79

PAGE 80

WeprovidethespecicoutlineofourGAimplementationinFigure 4-3 whereMdenotesthegivenpopulationsize,denotestheiterationcounter,andmaxdenotesthenumberofgenerations.LetP1andP2denotetwoparentchromosomes.Similarly,letC1andC2denotetwochildrenchromosomes.WeletL(j)denotetheobjectivefunction(tness)valueofindividualj. AfterLagrangianiterationsstopseitherwithpredenedgaphasbeenreachedorpredenednumberofiterationsexecuted.WeusetheGAbyusingthemostrecentpopulationtondabetterupperboundsolution.Wereportthesolutionwithbestobjectivefunctionvalue. 80

PAGE 81

`);`+1pt=max(0;`pt+a`(UBLB`)(xrptKrptyrpt) `);`+1pt=max(0;`pt+a`(UBLB`)(xdptKdptydpt) `);`+1pt=`pt+a`(UBLB`)(xrptxupt) `;!`+1pt=!`pt+a`(UBLB`)(yrptyupt) `: Weupdatethestepsizebyhalvingitsvaluewhenthereisnoimprovementinthelowerboundinaxednumberofiterations. 4{4 ).Whenconstraintset( 4{4 )isrelaxed,theresultingLagrangianproblembecomesseparableforeachproductp;p=1;:::;P.EachsubproblemisasingleproductMRDPP,anditisNP-Hard.Hence,weuseCPLEXtosolvetheLagrangiansubproblems.Insteadofsolvingthesubproblemstooptimality,a1%optimalitygapisusedtoobtainsolutionsfaster. Let=ftgdenotethevectorofLagrangianmultiplierst0;t=1;:::;T,associatedwiththemanufacturingcapacityconstraintsinset( 4{16 ).ThentheLagrangianproblemcanbestatedasfollows:minPXp=1TXt=1(Fmptympt+(cmpt+t)xmpt+Frptyrt+crptxrpt+Fdptydpt+cdptxdpt+kptjpt+hptipt)TXt=1(tKmt) (4{45) subjectto( 4{2 )-( 4{3 ),( 4{5 )-( 4{12 ). 81

PAGE 82

WeprovidethespecicoutlineofourLRimplementationinFigure 4-4 where`anddenotetheiterationcounters.Theterminationcriteriaaredenedby,thetolerancegapbetweenthebestlowerandupperbounds,and`max,themaximumnumberofiterationsofLR.WhentherearemaxLRiterations,thenGAisapplied.Moreover,UB`andLB`denotetheupperandlowerbounds,respectively,atiteration`.Similarly,UBandLBdenotetheincumbentupperandlowerbounds,respectively.WeproceedwithhowwesolvetheLagrangianproblemtoobtainalowerbound(Section 4.3.2.1 ),ndanupperboundusingthesmoothingheuristic(Section 4.3.2.2 )andGA,andupdatetheLagrangianmultipliers(Section 4.3.2.3 ),oneachiterationoftheLDapproach. 4{45 )isseparableintoPsubproblems.LetLR2SPpdenotesthesubproblemforproductpforp=1:::;P.WenotethatthelasttermintheobjectivefunctionofthisproblemisconstantanddoesnoteectthesolutionoftheLagrangianproblem.ThesingleproductLagrangiansubproblem(forproductp),i.e.,LR2SPpcanbestatedasfollows:minTXt=1(Fmptympt+(cmpt+t)xmpt+Frptyrpt+crptxrpt+Fdptydpt+cdptxdpt+kptjpt+hptipt) (4{46) subjecttoip;t1+xmpt+xrptDpt=iptt=1;:::;T; (4{47)jp;t1+Rptxrptxdpt=jptt=1;:::;T; (4{48)xrptKrptyrptt=1;:::;T; (4{49)xdptKdptydptt=1;:::;T; (4{50)xmpt (4{51) 82

PAGE 83

(4{52)xdpt (4{53)jp0;jpT;ip0;ipT=0 (4{54)xmpt;xrpt;xdpt;jptipt0andintegert=1;:::;T; (4{55)ympt;yrpt;ydpt2f0;1gt=1;:::;T: WesolveLR2SPpforproductpbyCPLEXwith1%optimalitygapforp=1;:::;P.LetLBpdenotebestobjectivefunctionvalueobtainedfromCPLEX,thenthelowerboundfortheMRDPP,LB,isgivenby:LB=PXp=1(LBp)TXt=1(tKmt) 4{45 )isobtainedbysolvingLR2SPpforp=1;:::;PbyCPLEX.Thisoptimalsolutionmayonlyviolatethesharedmanufacturingcapacityrestrictions.Weemploysmoothingheuristicstoobtainanupperboundsolution.Moreover,weuseGAtoimprovetheupperboundsolution.WeusethelowerboundsolutionandsmoothingheuristictogeneratethepopulationforGAforxednumberofLagrangianiterations.Weobtainfourchromosomesateachiteration.WethenuseGAtondabettersolutionfromthecurrentpopulation. Therstchromosomeisobtainedusingthesetupvariables'valuesofthemanufacturing,remanufacturinganddisposalactivitiesobtainedfromthelowerboundsolution.Thesecondchromosomeisobtainedbyapplyingmutationtotherstchromosome.Werandomlychooseaxedsetofproducts,andforeachchosenproduct,werandomlychoosexedsetofperiods.Foreachoftheseproductsandperiods,werandomlychoosemanufacturing,remanufacturing,anddisposalactivityandchangethecurrentassociatedsetupvariablevalue,from1to0or0to1.Thethirdchromosomeisobtainedbyapplying 83

PAGE 84

RecallthatthelowerboundsolutionisobtainedinSection 4.3.2.1 mayonlyviolatethemanufacturingcapacityrestrictions.WexthemanufacturingcapacityviolationsbytheroutineexplainedinSection 4.3.1.2 .Iftheobtainedupperboundisbetterthanthecurrentbestupperbound,thenthecurrentbestupperboundisupdated.ThisupperboundsolutionisusedasathirdchromosomeforthecurrentLagrangianiteration.Thefourthchromosomeisgeneratedbyapplyingmutationtothisupperupperbound. AfteraxednumberofLagrangianiterations,GAisemployedusingthegeneratedpopulationuptothatiteration.WeusetheGAgiveninSection 4.3.1.2 .Theonlydierencecomesfromthesizeandqualityofthepopulation. Lagrangianiterationsstopseitherwithpredenedgaphasbeenreachedorpredenednumberofiterationshit.Weconductthegeneticalgorithmoncemoreusingthemostrecentpopulationtondabetterupperboundsolution.Wereportthesolutionwithbestobjectivefunctionvalue. Weupdatethestepsizebyhalvingitsvaluewhenthereisnoimprovementinthelowerboundinaxednumberofiterations. 84

PAGE 85

Inourexperiments,webeginbygeneratingthedemands,returns,andmanufacturing,remanufacturinganddisposalcapacitiesbyinducingcorrelationsamongthemforeach(P;T)pair.Werstgeneratethedemands,thenbyusingthecorrectionbetweendemandandreturn,andrealizeddemandvalue,wegeneratereturns.Wethengeneratethereturnparametersinducingcorrelationbetweendemandsandreturns.Withthegivencapacitytightness,wegeneratethemanufacturingcapacitiesbyusingaveragerealizeddemandineachperiod.Similarly,wegeneratetheremanufacturinganddisposalcapacitiesforeachproductandeachperiodbyusingthecapacitytightnesslevelandtherealizedreturnanddemandvalues. Inourexperiments,weconsiderinstanceswithconstantunitmanufacturing,unitremanufacturing,andunitdisposal,andinventoryholdingcostsforUIIandMRII.Wegeneratethexedmanufacturing,remanufacturing,anddisposalsetupcostsbasedontheinventoryholdingcosts.Inparticular,webeginbygeneratingthetimebetween 85

PAGE 86

Similarly,wegeneratethexeddisposalsetupcostforeachproductineachperiodconsideringtheaveragereturn,TBDvalue,inventoryholdingcost,andvariabilityofxeddisposalsetupcosts. Todevelopasetofrandominstances,wevarythenumberofproductsPandthelengthoftheplanninghorizon,T.Specically,weconsidertwolevelsforP(10,20)andtwolevelsforT(12,24).Consequently,wehave4classesoftestinstances.Moreover,weconsidervelevelsforthecorrelationbetweentotaldemandtototalreturnsthroughouttheplanninghorizon(from0.15forlowto0.75forhighinincrementsof0.15),twolevelsforeachofTBM,TBR,andTBD(U[1,3]forloworU[3,5]forhigh),andtwolevelsforeachmanufacturing,remanufacturinganddisposalcapacitytightness(1.5forlowor2.5forhigh).WeconsidertheparametervaluessummarizedinTable 4-1 Foreach(P;T)weconsider20dierentproblemsettings,andforeachsettingwegenerate10randomtestinstances.Therefore,wegenerateandsolve800instancesintotal. 86

PAGE 87

WesummarizetheoptimalitygapsobtainedbytheLDapproachanditsCPUrequirementsinTable 4-2 .Aswesolveatotalof800testinstances,eachcellinthistablecorrespondstotheaverageormaximumvaluerepresenting200instancesofthesamesize.Wenotethattheobservedaverageoptimalitygapoverallinstancesis6.86percent,whereasthemaximumis15.34percent.Wealsoobservethatforagivennumberofproducts,theaverageandmaximumvaluesoftheoptimalitygapsbecomesmallerasthelengthoftheplanninghorizonincreases.Foragivenplanninghorizonlength,theaverageandmaximumvaluesoftheoptimalitygapsbecomesmallerbyanincreaseinthenumberofproducts. Moreover,weobservethattheLDapproachiscomputationallyecientastheobservedaverageCPUrequirementoverallinstancesislessthan17seconds,whereasthemaximumislessthan37seconds. InTables 4-3 and 4-4 ,weprovideadetailedanalysisoftheinuenceofthedatacharacteristicsontheperformanceoftheLDapproach. Weobservethatthequalityoftheoptimalitygapsdeterioratesasthetimebetweenactivitysetupsincreaseand/or(ii)theactivitycapacitiesbecomemorerestraining.Itisworthwhiletonotethattheinuenceofanincreaseincapacitytightnessistobemoreinuentialthananincreaseintimebetweenactivitysetups.Inaddition,weobservethattheoptimalitygapsobtainedbytheLDapproachdeterioratesasthecorrelationbetweenreturnsanddemandincreasesinboththeaverageandmaximumoptimalitygap. 87

PAGE 88

WesummarizetheoptimalitygapsobtainedbytheLRapproachanditsCPUrequirementsinTable 4-5 .Wenotethattheobservedaverageoptimalitygapoverallinstancesis0.95percent,whereasthemaximumis3.69percent.WeobservethesameinuenceofnumberofproductsandlengthofplanninghorizonontheoptimalitygapsoftheLRapproachasintheLDapproach.Consequently,thequalityofthegapidentiedbytheLRapproachisbetterforinstanceswithalargernumberofproductsand/oralongerplanninghorizon.OurcomputationalresultsshowedthattheproposedLDapproachismoreecientthanLRapproach,butLRapproachismuchmoreeectivethanLDapproach. InTables 4-6 and 4-7 ,weprovideadetailedanalysisoftheinuenceofthedatacharacteristicsontheperformanceoftheLRapproach. Surprisingly,weobservethatthequalityoftheoptimalitygapsofLRapproachdoesnotdeteriorateasmuchasLDapproachwhenthetimebetweenactivitysetupsincreaseand/or(ii)theactivitycapacitiesbecomemorerestraining.Moreinterestingly,averageoptimalitygapsalmostremainsameindependentfromthecapacitytightnessand/ortimebetweenactivitysetups.Inaddition,thecorrelationbetweenreturnsanddemanddoesnothavemucheectonthequalityofLRapproach. AlthoughtheresultsinTable 4-2 provideempiricalevidencefortheeectivenessoftheproposedLRapproachtoaddresstheMRDPP,aninterestingquestionthatremainsunansweredistheeectivenessoftheB&Capproachforobtainingsolutionstothe 88

PAGE 89

4-8 Again,eachcellinthistablerepresentstheaverageormaximumvalueof800observationsfortestinstancesofthesamesize.Wenotethattheaverageoptimalitygapoverallinstancesisaround6.22percentwhereasthemaximumisaround32percent.ThisclearlyshowsthatourLRapproachoutperformstheCPLEX. 89

PAGE 90

WeconductedextensivecomputationalexperimentstoinvestigatethecomputationaleciencyoftheLDandLRapproaches.OurcomputationalresultsshowedthattheproposedLDapproachiscomputationallymoreecientthantheLRapproach,buttheLRapproachismoreeectivethantheLDapproach. AnimmediateextensionofourworkthatwouldbeofpracticalrelevanceistoconsiderageneralizationoftheMRDPPwhere(i)themanufacturingand/orremanufacturingand/ordisposalsetupcostsdependonthespecicproductsincludedinthesetupand/or(ii)setuptimesareexplicitlymodeled.Moreover,MRDPPwithsharedremanufacturingcapacityand/orlineardisposalcostscanbepossibleextensions.Oursolutionalgorithmscanbeappliedtotheseproblemswithslightmodications.Also,theLRapproachpresentedinthispapercanbeextendedtosettingswherecertaincapacitatedresourcesaresharedbythemanufacturingandremanufacturingactivities. Figure4-1. ProblemSettingofMRDPP 90

PAGE 91

Figure4-2. Pseudo-codeoftheLDApproachfortheMRDPP. 91

PAGE 92

Pseudo-codeoftheGAApproachfortheLD. Figure4-4. Pseudo-codeoftheLRApproachfortheMRDPP. 92

PAGE 93

ParameterValuesUsedinComputationalExperimentsofMRDPP Level(s) Value(s) Constant 100 Variabilityofdemand Constant 0.5 Variabilityofreturns Constant 0.5 Variabilityofxedmanufacturingsetupcost Constant 0.2 Variabilityofxedremanufacturingsetupcost Constant 0.2 Variabilityofxeddisposalsetupcost Constant 0.2 UnitMRIIinventoryholdingcost Constant 1 UnitUIIinventoryholdingcost Constant 0.5 Unitmanufacturingcost Constant 3 Unitremanufacturingcost Constant 2 Unitdisposalcost Constant 1 Unitcapacityconsumption Constant 1 Timebetweenmanufacturing,remanufacturing,anddisposalsetups Low U[1,3] (TBM;TBR;TBD) High U[3,5] Manufacturing,remanufacturinganddisposalcapacitytightness Low 1.5 ('M;'R;'D) High 2.5 Return/Demandratio Low 0.15 (RD) High 0.75 PerformanceoftheLDApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). CPURequirement InstanceSize Avr.Gap Max.Gap Avr.Time Max.Time (P;T) (%) (%) (sec.) (sec.) (10,12) 7.68 15.34 11.65 19.25 (20,12) 6.43 12.63 16.72 28.56 (10,24) 7.21 14.68 15.21 27.14 (20,24) 6.12 11.24 22.43 36.79 Overall 6.86 15.34 16.50 36.79 EectofTimeBetweenActivitySetupsandCapacityTightnessontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. Max. (TBM,TBR,TBD) ('M;'R;'D) Gap(%) Gap(%) (U[1,3],U[1,3],U[1,3]) (1.5,1.5,1.5) 5.95 13.23 (U[3,5],U[3,5],U[3,5]) (1.5,1.5,1.5) 8.95 17.34 (U[1,3],U[1,3],U[1,3]) (2.5,2.5,2.5) 4.92 10.72 (U[3,5],U[3,5],U[3,5]) (2.5,2.5,2.5) 7.62 16.34

PAGE 94

EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. Max. Gap(%) 0.15 4.79 9.45 0.30 5.57 10.79 0.45 6.67 14.84 0.60 7.86 16.92 0.75 9.42 17.34 PerformanceoftheLRApproach:AverageandMaximumOptimalityGaps(%)andCPURequirement(sec.). CPURequirement InstanceSize Avr.Gap Max.Gap Avr.Time Max.Time (P;T) (%) (%) (sec.) (sec.) (10,12) 0.98 3.69 14.85 52.25 (20,12) 0.94 2.43 27.25 98.68 (10,24) 0.97 3.21 44.52 138.53 (20,24) 0.92 2.98 72.91 356.12 Overall 0.95 3.69 39.88 356.12 EectofTimeBetweenActivitySetupsandCapacityTightnessontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. Max. (TBM,TBR,TBD) ('M;'R;'D) Gap(%) Gap(%) (U[1,3],U[1,3],U[1,3]) (1.5,1.5,1.5) 0.94 1.59 (U[3,5],U[3,5],U[3,5]) (1.5,1.5,1.5) 1.03 3.69 (U[1,3],U[1,3],U[1,3]) (2.5,2.5,2.5) 0.91 1.31 (U[3,5],U[3,5],U[3,5]) (2.5,2.5,2.5) 0.92 1.44 EectofCorrelationBetweenReturnsandDemandontheAverageandMaximumOptimalityGaps(%)oftheLDApproach. Max. Gap(%) 0.15 0.89 3.69 0.30 0.94 2.72 0.45 0.93 2.43 0.60 0.98 1.91 0.75 1.01 2.34

PAGE 95

PerformanceofCPLEX(withaTimeLimitof400Seconds):AverageandMaximumOptimalityGaps(%). InstanceSize Avr.Gap Max.Gap (P;T) (%) (%) (10,12) 3.98 8.98 (20,12) 6.23 17.42 (10,24) 5.42 14.61 (20,24) 9.25 32.05 Overall 6.22 32.05

PAGE 96

MostardandTeunter 2006 ).Insomecases,thereturnratecanbeashighas75percent( MostardandTeunter 2006 ).Sincealargeproportionofthesereturnedproductsarereceivedinverygoodcondition,thesereturnscanbeplacedintoinventorytosatisfyfuturedemandwithlittleprocessing,suchasinspection,cleaning,testing,repackagingetc.Atypicalcatalogueretaileroersmultipleproductsthroughitscataloguesandinternetsitesandthe(nishedgoods)inventoryforaproductcanbereplenishedbyeitherprocuringproductsfromthesuppliersorprocessingandrestockingthereturnsreceivedfromconsumers.Inthissetting,foreectiveinventoryplanningandcontrol,thereturnsreceivedfromconsumersmustbetakenintoaccountwhilemakingprocurementdecisions.Animportantissuethatneedstobetakenintoaccountistheuncertaintyassociatedwiththereturnstream,whichcanbesignicantinthecontextofcatalogueretailing. Inthecontextofsparepartkittingforindustrialequipmentmaintenance,atypicalsupplierhasservicecontractswithastablecustomerbase.Thisallowsthesuppliertousehistoricaldatatoestimatethefuturedemandandreturnquantitiesaccuratelyfordierenttypesofproducts.Inthecontextofcatalogueretailing,however,retailer'sabilitytoforecastreturnsaccuratelyislimited,asthequantityand/ortimingofconsumerreturnsarehighlyvariable;inuencedbyseveralfactorssuchasbuyer'sremorseormismatchbetweencustomer'sexpectationsabouttheproduct.Typically,aretailerprocurestheproductsfromasupplieratthebeginningofaniteplanninghorizon.Reusablereturnsreceivedfromcustomersthroughouttheplanninghorizoncanbeplacedintoinventoryastheyarrivetosatisfythedemandinfutureperiods.Clearly,foroptimalinventoryplanningandcontrolforthissetting,returnsmustbetakenintoaccountexplicitlywhendecidingontheprocurementquantityatthebeginningoftheplanninghorizon.Ourfocus 96

PAGE 97

Morespecically,theproblemisconcernedwithdeterminingtheprocurementquantitiesformultipleproducttypesatthebeginningoftheplanninghorizon.Weconsidertheproblemindeterministicdemandandstochasticreturnframeworkoveraniteplanninghorizonunderniteprocurementcapacity.Inourwork,weconsiderfourvariantsofthePPP.Webeginbyconsideringtheprobleminasingle-periodsettingforasingleproducttype(PPP1)andextendthismodeltoincludemultipleproducttypes(PPP2).Then,westudytheprobleminmultiple-periodsettingforasingleproducttype(PPP3)aswellasformultipleproducttypes(PPP4). Theremainderofthischapterisorganizedasfollows.Weproceedwiththemodellingassumptionsin 5.2 WedevelopmathematicalmodelsandsolutionapproachesforasingleproducttypeinSections 5.3 and 5.4 forPPP1andPP2,respectively.Weanalyzethesemodelswhentheprocurementcostsarelinearandwhenthereisaxedcomponentinadditiontothelinearcomponent.Similarly,weinvestigatemodelsandsolutionapproachesformultipleproducttypesinSections 5.5 and 5.6 forPPP3andPPP4,respectively.Finally,conclusionsandfutureresearchdirectionsaresummarizedinSection 5.7 97

PAGE 98

Thesequenceofeventsareasfollows: 1. Manufactureditemsforproductpareprocuredpriortothebeginningoftheplanninghorizon,denotedbyxp,forp=1;:::;P. 2. Returnsforproducttypepinperiodtarereceivedforp=1;:::;Pandt=1;:::;T. 3. Demandforproducttypepinperiodtissatisedforp=1;:::;Pandt=1;:::;T. 4. Excessstockforproducttypepattheendofperiodtiscarriedtothenextperiodandinventoryholdingcostincurredforp=1;:::;Pandt=1;:::;T. 5. Excessdemandisbackorderedorshortagecostisincurredforproducttypepinperiodt,forp=1;:::;Pandt=1;:::;T. Theobjectiveistominimizethesumofexpectedprocurement,returnprocessing,backordering,andinventoryholdingcosts. 98

PAGE 99

Rearrangingtheterms,thetotalexpectedcostcanbewrittenasfollows: Then,wecanformulatethePPP1asfollows: minE[C(x)] (5{3) subjecttoxd: Firstorderconditionfor( 5{2 )isgivenby: Secondorderconditionfor( 5{2 )isgivenby:d2E[C(x)] (5{6) WehaveE00[C(x)]0.Hence,( 5{2 )isconvex,andtherstorderconditioncanbeusedtocharacterizetheoptimalprocurementquantity,x.Notethatwhenxdrstderivativeispositive,sotherstorderconditionscanonlybesatisedwhenx
PAGE 100

+hx=dF1cn+h +h Notethatwehavecn<(+h)F(d)h.Hence,xalwayssatises( 5{4 ). Intheclassicalnewsboy(newsvendor)problem,thecriticalfractileplaysanimportantroletocharacterizetheoptimalorderquantity.Recallthatthecriticalfractileistheratioofthecostofunderagetothesumofthecostsofunderageandoverage.InPPP1,thedemandisdeterministic,butreturnsarerandom.Wecaninterpret( 5{7 )asfollows:Thecostofunderageinourproblemis(cn+h).Also,thecostofoverageis(cn).Thatis,ifweunderestimatethequantityofreturnsbyoneunit,and,asaresult,weprocureonemoremanufactureditem,thenthecostofthisextraunitisitsprocurementcostplustheinventoryholdingcost,i.e.,(cn+h).Similarly,ifweoverestimatethereturnbyoneunit,and,asaresultprocureoneunitmanufactureditemless,thenthecostofnotprocuringthisunitisthepenaltycostofunsatiseddemandminustheprocurementcost,i.e.,(cn).Ifweusethecriticalfractileformulausingtheseunderageandoveragecosts,weobtain( 5{7 ).Therefore,ifweknowthecostofunderage,cu,andcostofoverage,co,thenwecandeterminetheoptimalprocurementquantityusing:x=dFcu 5{2 ),canberewrittenasfollows:E[C(x;z)]=Cnz+ 100

PAGE 101

LetMdenoteasucientlylargenumber.Then,theproblemformulation( 5{3 )-( 5{4 )becomes:minE[C(x;z)] (5{9) subjecttoxMz (5{12) Theoptimalsolutionto( 5{9 )-( 5{12 )canbecharacterizedasfollows.Letx0betheprocurementquantityidentiedbysolving( 5{3 )-( 5{4 ).Ifx0>0,thenthetotalexpectedcostisE[C(x0;1)].Moreover,E[C(0;0)]denotesthetotalexpectedcostwhentheprocurementquantityiszero.Wecancomparethesetwovalues,andiftheformerissmaller,thentheoptimalprocurementquantityisx0(i.e.,x=x0);otherwisetheoptimalprocurementquantityis0(i.e.,x=0). Recallthatwhencn<(+h)F(d)h,x0=dF1(cn+h +h),thenE[C(x0;1)]=Cn+(cn+h)x0+(cr+h)E[r]hd+(+h)Zdx00(dx0r)f(r)dr=Cn+cnd(cn+h)F1(cn+h +h))+(cr+h)E[r]+(+h)ZF1(cn+h +h)0(F1(cn+h +h)r)f(r)dr WealsohaveE[C(0;0)]=(cr+h)E[r]hd+(+h)Zd0(dr)f(r)dr 101

PAGE 102

+h);z=1.Morespecically,wecanndaconditiononthexedprocurementcostsuchthattheoptimalprocurementquantityiseitherx=dF1(cn+h +h)orzero.Letdenotethebreak-evenvalueofthexedcost,whichcanbeidentiedasfollows:=(cr+h)E[r]hd+(+h)Zd0(dr)f(r)dr(cnd(cn+h)F1(cn+h +h))+(cr+h)E[r]+(+h)ZF1(cn+h +h)0(F1(cn+h +h)r)f(r)dr)=(+h)Zd0(dr)f(r)drhdcnd(cn+h)F1(cn+h +h))+(+h)ZF1(cn+h +h)0(F1(cn+h +h)r)f(r)dr! +h).Otherwise,ifCn,thenitisoptimalnottoprocure,i.e.,x=0. 102

PAGE 103

(5{16)PXp=1pxpK Objectivefunction( 5{15 )isseparableforeachproductp,butconstraint( 5{17 ),sharedprocurementcapacityconstraint,linkstheproducts.Weconsiderrelaxingconstraint( 5{17 ),usingaLagrangianmultiplier,>0.When( 5{17 )isrelaxed,( 5{15 )-( 5{17 )becomesseparableforeachproducttypepforp=1;:::;P.ForagivenLagrangianmultiplierwehavethefollowingsubproblemforeachproducttypep:minE[C(xpj)]=Zdpxp0(cnp+p)xp+crprp+p((dpxp)rp)fp(rp)drp+Z1dpxp(cnp+p)xp+crprp+hp(rp(dpxp))fp(rp)drp) subjecttoxp0 (5{19) Aftersomealgebraicmanipulation,( 5{18 )canberewrittenasfollows:E[C(xpj)]=(cnp+p+hp)xp+(crp+hp)E[rp]hpdp+(p+hp)Zdpxp0(dpxprp)fp(rp)drp Firstorderconditionfor( 5{20 )is:E0[C(xpj)]=(cnp+p+hp)(p+hp)Zdpxp0fp(rp)drp Secondorderconditionfor( 5{20 )is:E00[C(xpj)]=8>><>>:(p+hp)fp(dpxp)xp
PAGE 104

5{22 )isnon-negative,( 5{20 )isconvex.Hence,theoptimalprocurementquantityforagivenLagrangianmultipliercanbeidentiedusingtherstorderconditionsfortheproducttypes.ForagivenLagrangianmultiplier,,theoptimalprocurementquantityxpisgivenby:xp=8>>><>>>:dpFpcnp+p+hp 5{17 ),thenitisoptimalto( 5{15 )-( 5{17 ).However,ifthissolutiondoesnotsatisfy( 5{17 ),thenwecansearchfortheoptimalvalueofthatyieldsthebestsolutionthatsatises( 5{17 ).First,weidentifythelargestpossiblevaluethatcantake,whichisgivenbythemaximum((p+hp)Fp(dp)hpcnp)=pvalueforp=1;:::;P. Wedevelopthefollowingalgorithmwhichseeksfortheoptimalsolutionto( 5{15 )-( 5{17 )byperformingasearchovertheLagrangianmultiplier.Let>0beanerrorboundonthecapacityconstraintsuchthatifasolutionthathasacapacityusagein[K;K]isidentied,thealgorithmstops. Initialization:Initialize.Setk0;k0;min0;maxmaxp=1;:::;Pn(p+hp)Fp(dp)hpcnp 104

PAGE 105

FindanInitialSolution:Forp=1;:::;Pndxkp:xkp=(dpFp(cnp+hp FindaSolution:Setk=min+max ReportSolution:Reportxp=xkpforp=1;:::;P,and=k. InAlgorithmAPPP2,werstinitializetheiterationcounter,,andminimumandmaximumvaluesthatcantake.Wethenndaninitialsolutionconsideringnocapacityconstraint,i.e.,when=0,inStep1.Ifthissolutionisfeasible,thenitisalsooptimalto( 5{15 )-( 5{17 ).Otherwise,wesearchfortheoptimalvalueiterativelybyhalvingtheintervalofateachiteration.Whenwendasolutionthatsatises( 5{17 )withanerrorboundof,westopthesearchandreportthecurrentsolutionasanoptimalsolution. Wealsoanalyzetheproblemiftheprocurementcosthasaxedcostcomponentinadditiontothelinearcomponent.Assumethatthereisaxedcost,Cnpforproductpforp=1;:::;P.Letzpbeadecisionvariableforproductpthattakesthevalueofonewhentheprocurementquantityispositive,i.e.,xp>0,andzerootherwise.( 5{15 )-( 5{17 )canberewrittenwhenthereisaxedcostasfollows:minE[C(!x;!z)]=PXp=1CnpzpZdpxp0cnpxp+crprp+p((dpxp)rp)fp(rp)drp+Z1dpxpcnpxp+crprp+hp(rp(dpxp))fp(rp)drp) 105

PAGE 106

(5{26) WecanstillusetheAlgorithmAPPP2.However,thesolutionofthesubproblemforproductpmaychangedependingonthexedcost.Morespecically,itmaybelesscostlynottoprocureproducttypepwhenthereisaxedcost.RecallthatinSection( 5.3 ),weidentifytwosolutions,i.e.,whenz=1andz=0.Then,weevaluatetheobjectivefunctionvaluesofthesesolutions.Theoptimalsolutionisthesolutionthatyieldsthelowerobjectivefunctionvalue.Subproblemforproducttypepcanbesolvedutilizingasimilarapproach.Wenextpresentthesubproblemformulationforproductpwhenthecapacityconstraint( 5{24 )isrelaxed:minE[C(xp;zpj)]=Cnpzp+(cnp+p+hp)xp+(crp+hp)E[rp]hpdp+(p+hp)Zdpxp0(dpxprp)fp(rp)drp subjecttoxpMzp (5{29) Foranygiven,zpcanbeeitherzeroorone.Hence,thesolutionsforbothcasescanbeevaluatedtoidentifytheoptimalsolution.Foragiven,whenzp=1theprocurementquantity,x0pisgivenby:x0p=8>>><>>>:dpFpcnp+p+hp

PAGE 107

ThesecondandthirdstepsofAlgorithmAPPP2canbemodiedtosolvethesubproblems.Thatis,foragivenLagrangianmultiplier,twosolutionsforeachproducttypephavetobecompared,i.e.,(x0p;1)and(0,0),andtheonethatyieldsthelowerobjectivefunctionvalueistheoptimalsolutionfortheproduct.Hence,wecanuseAlgorithmAPPP2,whenthereisaxedcostofprocurement. (5{30) subjecttox+r1(w)d1=I1(w) (5{31)It1(w)+rt(w)dt=It(w)t=2;:::;T (5{33) 107

PAGE 108

5{30 )minimizesthetotalexpectedcost.Thersttermisthetotalprocurementcost,secondtermistheexpectedprocessingcostofreturns,andthethirdandfourthtermsaretotalexpectedinventoryholdingandbackorderingcosts,respectively.Constraint( 5{31 )ensurestheinventorybalanceconstraintfortherstperiod.Constraintset( 5{32 )ensurestheinventorybalanceforperiods2throughendoftheplanninghorizon.Constraint( 5{33 )ensuresthenonnegativityofprocurementquantity. Byusing( 5{31 )and( 5{32 ),wecanndthenetendinginventoryinanyperiodtintermsofprocurementquantity,x,cumulativedemand,Dt,andcumulativereturn,Rt(w)asfollows: Notethatforanyperiodtifx
PAGE 109

Since( 5{38 )isnon-negative,Lt(x)isconvex. Wenowreformulatetheproblemwithoutinventoryvariablesbyusing( 5{34 )and( 5{36 )asfollows:minE[C(x)]=cnx+TXt=1(crtE[rt]+Lt(x)) (5{39) subjecttox0 (5{40) Recallthattheinventoryholdingandbackorderingcostsateachperioddependsontheprocurementquantity,x.Byusingdierentintervalsofx,theobjectivefunction,( 5{39 ),canberewrittenasfollows:

PAGE 110

TherstorderconditionofE[C(x)]isasfollows: Notethatthederivative,E0[C(x)],isnondecreasinginxfortworeasons,(i)Fj(Djx)isnonincreasinginx,and(ii)tincreasesinthesummationPTj=t(hj+j)Fj(Djx).WeformallypresentthispropertyinthefollowingProposition. Proof. 5{41 ),xcanbeinanyoftheT+1intervals.LetDT+1bedenedasDT+1=1.Intervalforperiodtcanbedenedas[Dt1;Dt)fort=1;:::;T+1.ItisenoughtoshowthatthederivativeofE[C(x)]is(i)nondecreasinginanyintervalt,i.e.,[Dt1;Dt),and(ii)nondecreasingthroughincreasingintervals,i.e.,t0andt00with1t0
PAGE 111

Thedierencebetweentherstderivativevalueswhenprocurementquantitiesarex2andx1is: SinceFj(Djx1)Fj(Djx2)forj=t;:::;T. (ii)Letx1andx2betwoprocurementquantitieswithx12[Dt01;Dt0)andx22[Dt001;Dt00),where1t0
PAGE 112

Proof. (i) (ii) (iii) Proof. (i) SinceE0[C(Dt)]<0,E0[C(DT)]>0andE0[C(x)]continuousandnondecreasing,thereexistssomeDt0,thenx=0,otherwise,0
PAGE 113

Initialization:Initializek1;xmin0;xmaxD1 Calculaterstderivativevaluesatxminandxmax:E0[C(x)]jxmin=cn+TXt=1htTXj=k(hj+j)Fj(Djxmin)E0[C(x)]jxmax=cn+TXt=1htTXj=k(hj+j)Fj(Djxmax) Checkconditions: { ifE0[C(x)]jxmax<0,thenkk+1;xminDk1;xmaxDk,andgotoStep1. { ifE0[C(x)]jxmax=0,thenx=xmax,andgotoStep4. { ifE0[C(x)]jxmax>0,thenkk,andgotoStep3. Findtheoptimalsolutionvialinesearch:Notethatx2(Dk;Dk+1).Solvethefollowingequation,andidentifyx. ReportSolution:Reportx. Wenotethatwhenk=T,therstderivativeatx=DTisalwayspositive,hence,thesolutioncannotbeunbounded.AlgorithmAPPP3identiesasolutionafteranitenumberofiterations,becausesameintervalwillnotberevisited.TheperformanceofthealgorithmalsodependsonhowecientlythelinesearchidentiestheoptimalsolutionatStep3. 113

PAGE 114

114

PAGE 115

subjecttoxp+rp1(w)dp1=Ip1(w)p=1;:::;P Objectivefunction( 5{42 )minimizesthesumofexpectedprocurement,processing,inventoryholdingandbackorderingcoststhroughouttheplanninghorizon.Constraintset( 5{43 )capturestheinventorybalanceequationsforproductpinperiod1.Constraintset( 5{32 )representstheinventorybalanceequationsforproductpineachperiodtfort=2;:::;T,whichensuresthatforeachproductpthesumofquantityininventorycarriedintoperiodtandthequantityofreturnsprocessedisequaltothesumofthedemandinperiodtandthequantityofinventorycarriedtoperiodt+1.Constraint( 5{45 )ensuresthatthetotalprocurementcapacityrequiredforalltheproductsdoesnotexceedtheavailableprocurementcapacity.Constraintset( 5{46 )ensuresthenonnegativityofthedecisionvariables. Byusing( 5{43 )and( 5{44 )foreachproducttypep,wecanndthenetendinginventoryinperiodtintermsofprocurementquantityxpfortheproducttypep

PAGE 116

Notethatforanyproductpforanyperiodtifxp0denotetheassociatedLagrangianmultiplier.When( 5{50 )isrelaxed,( 5{49 )-( 5{51 )becomesseparableforeachproducttype.ThecorrespondingLagrangian 116

PAGE 117

ForagivenLagrangianmultiplierwehave:minE[C(!xj)]=PXp=1(cnp+p)xp+TXt=1crptE[rpt]+Lpt(xp)!K subjecttoxp0p=1;:::;P Notethatthelasttermof( 5{52 )isconstantandtheproblem( 5{52 )-( 5{53 )isseparableforeachproducttypep.Hence,foragivenLagrangianmultiplierwehavePsubproblems.Subproblemforproducttypepisasfollows:minE[C(xpj)]=(cnp+p)xp+TXt=1crptE[rpt]+Lpt(xp) subjecttoxp0 (5{55) SubproblemforproducttypepissameasPPP3problem.Therefore,AlgorithmAPPP3canbeusedtosolveeachsubproblem.WedevelopanalgorithmsimilartoAlgorithmAPPP2thatseeksfortheoptimal.WerstsolveeachsubproblemwhentheLagrangianmultiplierisequaltozero,i.e.,=0.Ifthesolutionobtainedsatisesthecapacityconstraint,thenthissolutionisalsooptimaltooriginalproblem,( 5{49 )-( 5{51 ).Otherwise, 117

PAGE 118

Initialization:Initializek0;k0;min0; FindanInitialSolution:Findxkpbysolvingsubproblem( 5{54 )-( 5{55 )forproductpbyusingAlgorithmASPMS,forp=1;:::;P FindaSolution:Setk=min+max 5{54 )-( 5{55 )forproductpbyusingAlgorithmASPMS,forp=1;:::;P ReportSolution:Reportxp=xkpforp=1;:::;P,and=k. InAlgorithmAPPP4werstinitializetheiterationcounter,andboundsof.Wethenndaninitialsolutionconsideringnocapacityconstraint,i.e.,when=0inStep1byusingtheAlgorithmAPPP3.Ifthissolutionisfeasiblethenitisalsooptimalto( 5{49 )-( 5{51 ).Otherwise,wesearchforoptimalvalueiteratively,byhalvingtheintervalofateachiteration.Whenwendasolutionthatsatises( 5{51 )withanerrorboundof,westopthesearchandreportthecurrentsolutionasanoptimalsolution. Assumethatthereisaxedcostcomponentforprocurementforeachproductp,Cnp,inadditiontolinearcostcomponent.WecanuseAlgorithmPPP4tosolvethisproblem. 118

PAGE 119

AnimmediateextensionofourworkthatwouldbeofpracticalrelevanceistoconsiderageneralizationofthePPPwhere(i)demandcanbemodeledasarandomvariabletotakedemanduncertaintyintoaccountand/or(ii)multipleprocurementopportunitiesareconsideredthroughouttheplanninghorizon. 119

PAGE 120

Inourstudy,wefocusedonthecharacterizationoftheoptimalinventoryandproductionplanningpoliciesformulti-productCLSCsystemsfordirectreuseandvalue-addedrecoverybyconsideringaclassofdeterministicdiscretetimedynamicdemandandreturnmodels.Wealsoconsideraprocurementplanningproblemfordirectreusesystemswithdeterministicdiscretetimedynamicdemandandstochasticreturns.Mostofthecurrentliteratureoninventoryandproductionplanningforproductrecoverymodelsforuncapacitatedsingle-productenvironments.Hence,ourcontributiontoliteratureisrelatedtoconsideringmulti-productenvironmentswithcapacityrestrictions.Moreover,wedevelopecientsolutionapproachesthatarebasedondynamicprogramming,relaxation,decomposition,andmeta-heuristicapproachestoaddressthisclassofmodels. Intherstpartofourstudy,wefocusedonreplenishmentanddisposalplanningproblemformulti-productdirectreusesystems.TheproblemofinterestwasmotivatedbyOEMinthepowergenerationequipmentindustrythatprovidessparepartkitsforpowerturbinemaintenanceservices.Toensureuninterruptedsparepartkitavailability,theOEMhastocoordinatereplenishmentanddisposaldecisionsforthekits.Wedeterminethetimingofthereplenishmentanddisposalsetups,alongwiththeassociatedreplenishmentanddisposalquantitiesformultipleproducttypesthroughoutaniteplanninghorizonsoastominimizevariablereplenishment,disposal,andinventoryholdingcostsaswellasxedreplenishmentanddisposalcosts.Inparticular,weconsideredtwovariantsoftheproblem.Intherstvariant,thereplenishmentcapacityissharedamongmultipleproducttypeswhilethedisposalcapacityisproductspecic.Inthesecondvariant,boththereplenishmentanddisposalcapacitiesaresharedamongmultipleproducttypes.Wedevelopedecientsolutionapproachforbothvariantsoftheproblems.Ourcomputationalresultsshowthattheproposedapproachesiseectiveinobtaininghigh-qualitysolutionsforrealistically-sizedinstancesoftheRDPPwithareasonable 120

PAGE 121

Inthesecondpartofourstudy,westudiedamanufacturing,remanufacturinganddisposalplanningproblemwithmultipleproducttypesforvalue-addedrecoverysystems.Problemofinterestismotivatedbytheoriginalequipmentmanufacturerintheautomotiveindustrythatprovidesvehiclemaintenanceandrepairservicesforvehiclesforwhichreplacementparts(products)areoftenneeded.Wedevelopedtwosolutionapproaches:LagrangianDecomposition(LD)andLagrangianRelaxation(LR).InLDapproach,wedecomposetheproblemintosubproblemsthataresolvedbypolynomialtimeDPapproaches.InLRapproach,however,weuseCPLEXtosolvetheresultingsingleproductproblems.OurcomputationalresultsshowedthatalthoughtheLDapproachiscomputationallymoreecient,theLRapproachiseectiveinndinghigh-qualitysolutions.ConsideringageneralizationoftheMRDPPwhere(i)themanufacturingand/orremanufacturingand/ordisposalsetupcostsdependonthespecicproductsincludedinthesetupand/or(ii)setuptimesareexplicitlymodeledwouldbepracticallyrelevantextensionofourwork.Moreover,MRDPPwithsharedremanufacturingcapacityand/orlineardisposalcostscanbepossibleextensions.Oursolutionalgorithmscanstillbeappliedtotheseproblemswithslightmodications.Also,theLRapproachpresentedinthispapercanbeextendedtosettingswherecertaincapacitatedresourcesaresharedbythemanufacturingandremanufacturingactivities. Inthelastpartofourstudy,wefocusedonprocurementplanningproblemformulti-productdirectreusesystems.Theproblemofinterestwasmotivatedbycatalogueretailingapplicationswherereturnsareinas-good-asnewconditionandcanbeusedtosatisfydemandwithlittleprocessing.Tothisend,weconsideredadiscretetimedynamicdeterministicdemandandstochasticreturnmodelforthecharacterizationofoptimal.Westudytheprocurementplanningproblemsincatalogueretailingapplications 121

PAGE 122

122

PAGE 123

Aras,N.,T.Boyac,V.Verter.2004.Theeectofcategorizingreturnedproductsinremanufacturing.IIETransactions36319{331. Barany,I.,T.J.VanRoy,L.A.Wolsey.1984.Strongformulationsformulti-itemcapacitatedlotsizing.ManagementScience30(10)1255{1261. Bayndr,Z.P.,N.Erkip,R.Gullu.2003.Amodeltoevaluateinventorycostsinaremanufacturingenvironment.InternationalJournalofProductionEconomics81{82597{607. Bayndr,Z.P.,N.Erkip,R.Gullu.2005.Assessingthebenetsofremanufacturingoptionunderone-waysubstitution.JournaloftheOperationalResearchSociety56286{296. Bayndr,Z.P.,N.Erkip,R.Gullu.2007.Assessingthebenetsofremanufacturingoptionunderone-waysubstitutionandcapacityconstraint.Computers&OperationsResearch34(2)487{514. Beltran,J.L.,D.Krass.2002.Dynamiclotsizingwithreturningitemsanddisposals.IIETransactions34437{448. Belvaux,G.,L.A.Wolsey.2000.Bc-prod:Aspecialisedbranch-and-cutsystemforlotsizingproblems.ManagementScience46(5)993{1007. Belvaux,G.,L.A.Wolsey.2001.Modellingpracticallot-sizingproblemsasmixedintegerprograms.ManagementScience47(7)724{738. Bitran,G.R.,H.H.Yanasse.1982.Computationalcomplexityofthecapacitatedlotsizeproblem.ManagementScience301121{1140. Brahimi,N.,S.Dauzere-Peres,N.M.Najid,A.Nordli.2006.Singleitemlotsizingproblems.EuropeanJournalofOperationalResearch1681{16. Buchanan,D.J.,P.L.Abad.1998.Optimalpolicyforaperiodicreviewreturnableinventorysystem.IIETransactions301049{1055. Chen,W.H.,J.M.Thizy.1990.Analysisofrelaxationsforthemulti-itemcapacitatedlot-sizingproblem.AnnalsofOperationsResearch26(1)29{72. Cohen,M.A.,W.P.Pierskalla,S.Nahmias.1980.Adynamicinventorysystemwithrecycling.NavalResearchLogisticsQuarterly27289{296. Dekker,R.,R.vanderLaan.2003.Inventorycontrolinreverselogistics.V.D.R.Guide,L.N.VanWassenhove,eds.,BusinessAspectsofClosed-LoopSupplyChains,InternationalManagementSeries,vol.2.CarnegieBoschInstitute,149{174. Diaby,M.,H.C.Bahl,M.H.Karwan,S.Zionts.1992a.Capacitatedlot-sizingandschedulingbyLagrangeanrelaxation.EuropeanJournalofOperationalResearch59444{458. 123

PAGE 124

Dixon,P.S.,E.A.Silver.1981.Aheuristicsolutionprocedureforthemulti-item,single-level,limitedcapacity,lot-sizingproblem.JournalofOperationsManagement223{39. Dogramaci,A.,J.C.Panayiotopoulos,N.R.Adam.1981.Thedynamiclot-sizingproblemformultipleitemsunderlimitedcapacity.AIIETransactions13(4)294{303. Eisenhut,P.S.1975.Adynamiclotsizingalgorithmwithcapacityconstraints.AIIETransactions7(2)170{176. Eppen,G.D.,R.K.Martin.1987.Solvingmulti-itemcapacitatedlot-sizingproblemsusingvariableredenition.OperationsResearch35(6)832{848. Fleischmann,M.,R.Kuik.2003.Onoptimalinventorycontrolwithindependentstochasticitemreturns.EuropeanJournalofOperationalResearch15125{37. Fleischmann,M.,R.Kuik,R.Dekker.2002.Controllinginventorieswithstochasticitemreturns:abasicmodel.EuropeanJournalofOperationalResearch13863{75. Florian,M.,J.K.Lenstra,A.H.G.RinnooyKan.1980.Deterministicproductionplanning:algorithmsandcomplexity.ManagementScience26669{679. Golany,B.,J.Yang,G.Yu.2001.Economiclot-sizingwithremanufacturingoptions.IIETransactions33995{1003. Golany,B.,J.Yang,G.Yu.2005.Erratum:Correctingaproofin\Economiclot-sizingwithremanufacturingoptions".IIETransactions37787. Guide,V.D.R.2000.Productionplanningandcontrolforremanufacturing:industrypracticeandresearchneeds.JournalofOperationsManagement18467{483. Gunther,H.O.1987.Planninglotsizesandcapacityrequirementsinasinglestageproductionsystem.EuropeanJournalofOperationalResearch31223{231. Heisig,G.,M.Fleischmann.2001.Planningstabilityinaproductrecoverysystem.ORSpektrum2325{50. Heyman,D.P.1977.Optimaldisposalpoliciesforasingle-iteminventorysystemwithreturns.NavalResearchLogisticsQuarterly24385{405. Inderfurth,K.1997.Simpleoptimalreplenishmentanddisposalpoliciesforaproductrecoverysystemwithleadtimes.ORSpektrum19111{122. Inderfurth,K.2004.Optimalpoliciesinhybridmanufacturing/remanufacturingsystemswithproductsubstitution.InternationalJournalofProductionEconomics90325{343. 124

PAGE 125

Inderfurth,K.,R.H.Teunter.2003.Productionplanningandcontrolofclosed-loopsupplychains.V.D.R.Guide,L.N.VanWassenhove,eds.,BusinessAspectsofClosed-LoopSupplyChains,InternationalManagementSeries,vol.2.CarnegieBoschInstitute,149{174. Jans,R.,Z.Degraeve.2007.Meta-heuristicsfordynamiclotsizing:Areviewandcomparisonofsolutionapproaches.EuropeanJournalofOperationalResearch1771855{1875. Karimi,B.,S.M.T.FatemiGhomi,J.M.Wilson.2003.Thecapacitatedlotsizingproblem:areviewofmodelsandalgorithms.Omega31365{378. Karni,R.,Y.Roll.1982.Aheuristicalgorithmforthemulti-itemlot-sizingproblemwithcapacityconstraints.IIETransactions14(4)249{256. Kelle,P.,E.A.Silver.1989.Purchasingpolicyofnewcontainersconsideringtherandomreturnsofpreviouslyissuedcontainers.IIETransactions21(4)349{354. Kiesmuller,G.2003.Anewapproachforcontrollingahybridstochasticmanufacturing/remanufacturingsystemwithinventoriesanddierentleadtimes.EuropeanJournalofOperationalResearch14762{71. Kiesmuller,G.,S.Minner.2003.Simpleexpressionsforndingrecoverysysteminventorycontrolparametervalues.JournaloftheOperationalResearchSociety5483{88. Kiesmuller,G.,C.W.Scherer.2003.Computationalissuesinastochasticnitehorizononeproductreccoveryinventorymodel.EuropeanJournalofOperationalResearch146553{579. Kiesmuller,G.,E.vanderLaan.2001.Aninventorymodelwithdependentproductdemandsandreturns.InternationalJournalofProductionEconomics7273{87. Krca,O.,M.Kokten.1994.Anewheuristicappraochforthemulti-itemdynamiclotsizingproblem.EuropeanJournalofOperationalResearch75332{341. Lambrecht,M.R.,H.Vanderveken.1979.Heuristicproceduresforthesingleoperation,multi-itemloadingproblem.AIIETransactions11(4)319{326. Leung,J.M.Y.,T.L.Magnanti,R.Vachani.1989.Facetsandalgorithmsforcapacitatedlotsizing.MathematicalProgramming45331{359. Li,Y.,J.Chen,X.Cai.2006a.Heuristicgeneticalgorithmforcapacitatedproductionplanningproblemswithbatchprocessingandremanufacturing.InternationalJournalofProductionEconomics. 125

PAGE 126

Maes,J.,L.N.VanWassenhove.1986.Asimpleheuristicforthemulti-itemsinglelevelcapacitatedlotsizingproblem.OperationsResearchLetters4(6)265{273. Mahadevan,B.,D.F.Pyke,M.Fleischmann.1999.Periodicreview,pushinventorypoliciesforremanufacturing.EuropeanJournalofOperationalResearch151536{551. Millar,H.H.,M.Yang.1994.Lagrangianheuristicsforthecapacitatedmulti-itemlot-sizingproblemwithbackordering.InternationalJournalofProductionEconomics341{15. Mostard,J.,R.Teunter.2006.Thenewsboyproblemwithresalablereturns:Asingleperiodmodelandcasestudy.EuropenJournalofOperationsResearch16981{96. Muckstadt,J.A.,M.H.Isaac.1981.Ananalysisofsingleiteminventorysystemswithreturns.NavalResearchLogisticsQuarterly28237{254. Nakashima,K.,J.Arimitsu,T.Nose,S.Kuriyama.2002.Analysisofaproductrecoverysystem.InternationalJournalofProductionResearch403849{3856. Nakashima,K.,J.Arimitsu,T.Nose,S.Kuriyama.2004.Optimalcontrolofaremanufacturingsystem.InternationalJournalofProductionResearch42(17)3619{3625. Pohlen,T.L.,IIM.T.Farris.1992.Reverselogisticsinplasticsrecycling.InternationalJournalofPhysicalDistributionandLogisticsManagement22(7)35{47. Richter,K.,I.Dobos.2004.Production-inventorycontrolinaneoq-typereverselogisticssystem.H.Dyckho,R.Lackes,J.Reese,eds.,SupplyChainManagementandReverseLogistics.Springer,163{180. Richter,K.,B.Gobsch.2003.Themarket-orienteddynamicproductrecoverymodelinthejust-in-timeframework.InternationalJournalofProductionEconomics81{82369{374. Richter,K.,M.Sombrutzki.2000.RemanufacturingplanningforthereverseWagner/Whitinmodels.EuropeanJournalofOperationalResearch121304{315. Richter,K.,J.Weber.2001.ThereverseWagner/Whitinmodelwithvariablemanufacturingandremanufacturingcost.InternationalJournalofProductionEco-nomics71447{456. Rogers,D.S.,R.S.Tibben-Lembke.1999.GoingBackwards:ReverseLogisticsTrendsandPractices.ReverseLogisticsExecutiveCouncil. 126

PAGE 127

Simpson,V.P.1978.Optimumsolutionstructureforarepairableinventoryproblem.OperationsResearch26(2)270{281. Teunter,R.,E.vanderLaan,K.Inderfurth.2000.Onthenon-optimalityoftheaveragecostapproachforinventorymodelswithremanufacturing.OMEGATheInternationalJournalofManagementScience28409{415. Teunter,R.H.2002.Economicorderquantitiesforstochasticdiscountedcostinventorysystemswithremanufacturing.InternationalJournalofLogistics:ResearchandApplications5(2)161{175. Teunter,R.H.,E.vanderLaan,D.Vlachos.2004.Inventorystrategiesforsystemswithfastremanufacturing.JournaloftheOperationalResearchSociety55475{484. Teunter,R.H.,D.Vlachos.2002.Onthenecessityofadisposaloptionforreturneditemsthatcanberemanufactured.InternationalJournalofProductionEconomics72257{266. Thizy,J.M.1991.AnalysisofLagrangiandecompositionforthemulti-itemcapacitatedlot-sizingproblem.INFOR29(4)271{283. Thizy,J.M.,L.N.VanWassenhove.1985.Lagrangeanrelaxationforthemulti-itemcapacitatedlotsizingproblem:Aheuristicimplementation.IIETransactions17(4)308{313. Toktay,L.B.,L.M.Wein,S.A.Zenios.2000.Inventorymanagementofremanufacturableproducts.ManagementScience46(11)1412{1426. Trigeiro,W.W.1987.Adual-costheuristicforthecapacitatedlotsizingproblem.IIETransactions7567{72. Trigeiro,W.W.1989.Asimpleheuristicforlotsizingwithsetuptimes.DecisionSciences20(2)294{303. Trigeiro,W.W.,L.J.Thomas,J.O.McClain.1989.Capacitatedlotsizingwithsetuptimes.ManagementScience35(3)353{366. vanderLaan,E.2003.AnNPVandACanalysisofastochasticinventorysystemwithjointmanufacturingandremanufacturing.InternationalJournalofProductionEconomics81{82317{331. vanderLaan,E.,R.Dekker,M.Salomon.1996a.Productremanufacturinganddisposal:Anumericalcomparisonofalternativecontrolstrategies.InternationalJournalofProductionEconomics45489{498. 127

PAGE 128

vanderLaan,E.,G.Kiesmuller,R.Kuik,D.Vlachos,R.Dekker.2004.Stochasticinventorycontrolforproductrecovery.R.Dekker,M.Fleischmann,K.Inderfurth,L.N.VanWassenhove,eds.,ReverseLogistics:QuantitativeModelsforClosed-LoopSupplyChains.Springer,181{220. vanderLaan,E.,M.Salomon.1997.Productplanningandinventorycontrolwithremanufacturinganddisposal.EuropeanJournalofOperationalResearch102264{278. vanderLaan,E.,M.Salomon,R.Dekker.1999a.Aninvestigationoflead-timeeectsinmanufacturing/remanufacturingsystemsundersimplePUSHandPULLcontrolstrategies.EuropeanJournalofOperationalResearch115195{214. vanderLaan,E.,M.Salomon,R.Dekker,L.VanWassenhove.1999b.Inventorycontrolinhybridsystemswithremanufacturing.ManagementScience45(5)733{747. vanderLaan,E.,R.Teunter.2006.Simpleheuristicsforpushandpullremanufacturingpolicies.EuropeanJournalofOperationalResearch1751084{1102. Vlachos,D.,R.Dekker.2003.Returnhandlingoptionsandorderquantitiesforsingleperiodproducts.EuropeanJournalofOperationalResearch15138{52. Wagner,H.M.,T.M.Whitin.1958.Dynamicversionoftheeconomiclotsizemodel.ManagementScience589{96. Whisler,W.1967.Astochasticinventorymodelforrentedequipment.ManagementScience13(9)640{647. Yang,J.,B.Golany,G.Yu.2005.Concavecostproductionplanningproblemwithremanufacturingoptions.NavalResearchLogistics52443{458. Yuan,X.-M.,K.L.Cheung.1998.Modelingreturnsofmerchandiseinaninventorysystem.ORSpektrum20147{154. 128

PAGE 129

_IbrahimKarakayalearnedB.S.andM.S.degreesinindustrialengineeringfromtheMiddleEastTechnicalUniversityinAnkara,Turkey,in2001and2003,respectively.HebeganhisgraduatestudiesattheDepartmentofIndustrialandSystemsEngineeringattheUniversityofFloridainAugust2003.HewasaninternintheEnterpriseOptimizationGroupatUnitedAirlinesinChicago,Illinois.Hismainareasofresearchareproductionplanningandcontrol,mathematicalprogramming,andclosed-loopsupplychains. 129