Including cleaning and production phases in the eco-design of a milk evaporation process

HAL Id: hal-02887356

https://hal.inrae.fr/hal-02887356

Submitted on 2 Jul 2020

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives| 4.0

International License

eco-design of a milk evaporation process

Martial Madoumier, Catherine Azzaro-Pantel, Geneviève Gésan-Guiziou

To cite this version:

Martial Madoumier, Catherine Azzaro-Pantel, Geneviève Gésan-Guiziou. Including cleaning and

pro-duction phases in the eco-design of a milk evaporation process. Food and Bioproducts Processing,

Elsevier, 2020, 123, pp.427-436. �10.1016/j.fbp.2020.07.023�. �hal-02887356�

FoodandBioproductsProcessing 123 (2020)427–436

ContentslistsavailableatScienceDirect

Food

and

Bioproducts

Processing

jo u r n al h om ep a g e :w w w . e l s e v i e r . c o m / l o c a t e / f b p

Including

cleaning

and

production

phases

in

the

eco-design

of

a

milk

evaporation

process

Martial

Madoumier

_,

_Catherine

_{Azzaro-Pantel}

_,

_Geneviève

Gésan-Guiziou

a_{INRAE,InstitutAgro,STLO,F-35042Rennes,France}

b_{LaboratoiredeGénieChimique,UMR5503CNRS,UniversitédeToulouse,ENSIACETINPT/UPS,BP84234,F-31432}

Toulouse,France

c_{Qualisud,MontpellierSupAgro,1101AvenueAgropolis,CS24501,34093MontpellierCedex5,France}

a

r

t

i

c

l

e

i

n

f

o

Received10April2020

Receivedinrevisedform13June 2020

Accepted31July2020

Availableonline11August2020

Keywords: Eco-design Cleaning Kinetics Model Evaporation Dairysector

a

b

s

t

r

a

c

t

Thecleaningphaseisseldomincludedintheeco-designoffoodprocesses,sincefew clean-ingkineticsmodelsexist.Thegoalofthisstudywastoinvestigatebenefitsofincluding acleaningkineticsmodel,whichconsidersthreemajoroperatingparametersofcleaning (concentration,temperatureandflowrate),intheeco-designofafoodprocess.Tothisend, wedevelopedaneco-designapproachfora dairyevaporatorprocessthatincludesboth productionandcleaningphases.Acleaningkineticsmodelwasselectedtopredict clean-ingdurationasafunctionoftheoperatingparametersofcleaning.Cleaningdurationalso dependsonthefoulingsurfacedensity,whichdependsonthedurationoftheproduction phase.Foulingsurfacedensitywaspredictedusingthreehypotheticalfoulingkineticslaws. Afteroptimization,environmentalandeconomicimprovementswereobservedinprocess performance.Theevaporationprocessisoptimizedatahighcleaningtemperature(95◦C), aflow-ratesimilartothatusedduringtheproductionphaseandalowcausticsoda con-centration(<2%).Thisstudyhighlightsthattooptimizefoodprocessesinamoreprecise way,cleaningkineticsshouldbeincludedandusedtoidentifyparametersthatinfluence performanceoftheoverallprocess.

©2020InstitutionofChemicalEngineers.PublishedbyElsevierB.V.Allrightsreserved.

1.

Introduction

Cleaning-in-place(CIP)iscommonlyusedinthefoodindustry toensurehygienicsafetyoffoodsandrecoverplant perfor-mance;however,ithashighoperatingandinvestmentcosts (Tamime,2009)andenvironmentalimpacts(Eideetal.,2003). Thisisparticularlytrueinthedairyindustry,whichhaslong non-productionperiodsdedicatedtoCIP(4–6hperday)and ahugevolumeofeffluentsgeneratedbyCIP(50–95%ofthe volumeofwaste sent tothewastewatertreatment facility, regardlessofthetypeorsizeoftheplantorequipment(Marty, 2001;Sage,2005)).

∗ _{Correspondingauthor.}

E-mailaddress:genevieve.gesan-guiziou@inrae.fr(G.Gésan-Guiziou).

Eco-design and optimization of food processing have attractedmuchattention inthe pasttwodecades (Stefanis etal.,1997;Bangaetal.,2008;Erdo ˘gdu,2008;Sharmaetal., 2012).Ecodesignisanenvironmentalmanagementapproach thataimsatintegratingenvironmentalissuesintotheproduct developmentprocess,inordertoimprovetheenvironmental performanceofaproductacrossitsentirelifecycle.However, theCIPprocedurehasrarelybeenconsidered.Theliterature indicates thatthreeapproaches havebeenused toaddress CIPintheeco-designoffoodprocesses,butnoneofthemhas includedcleaningkineticsmodels,whichrepresentthemajor operatingparametersofcleaning:

• ThefirstapproachusesLifeCycleAssessmenttoidentify thealternativeCIPprocedurewiththelowest environmen-talimpactsbycomparingmultipleCIPprocedures(Eideand Ohlsson,1998;Eide,2002;Eideetal.,2003).Thisapproach https://doi.org/10.1016/j.fbp.2020.07.023

Nomenclature

Ai Reducedvariable(unitless)

Cp Heatcapacity(Jkg−1K−1)

D Tubediameter(m)

H Enthalpyrate(Jh−1)

k Kineticconstant(s−1)

Lvap _{Latententhalpyofvaporization(MJkg}−1₎

˙

m Massflowrate(kgh−1)

M0 Foulingsurfacedensity(kgm−2)

nt Numberoftubesinanevaporatoreffect(-)

rF Cleaningrate(kgm−2)

 Cleaningsolutiondensity(kg.m−3)

t Time(sorh)

T Temperature(◦CorK)

v Flowvelocity(ms−1)withv= m˙ nt(D

2) 2

X Massconcentration(kgkg−1)

analyzesdataofindustrialCIPprocedures,thusremoving theneedforpredictivemodelingofcleaningkineticsand excludingrelationshipsbetweentheproductionand clean-ingphases.

• Thesecondapproachoptimizesproductionscheduling(i.e. time management) while assuming a constant cleaning duration.Forinstance,BirewarandGrossmann(1989) min-imizedthedurationofaproductioncycleinamultiproduct batchplantwithoutconsideringthemajoroperating condi-tionsofcleaningorcleaningkinetics.Theydid,however, perform a sensitivity analysis of different scenarios of cleaningdurationinanoptimizationstrategy,asaninitial stepinpredictingcleaningkinetics.

• The third approach uses a cleaning kinetics model to optimizecostsandenvironmentalimpactsofproduction scheduling of batch plants. For example, Stefanis et al. (1997)appeartohaveusedthemodelofBirdandFryer(1991) topredictalkalinecleaningkineticsasafunctionof caus-ticsoda concentrationand initialfouling oftheplant to optimizetheeco-design ofcheeseproductionlines. This approachappearsthemostadvanced,sinceitcombinesa modelofcleaningkineticswithplantoptimization. How-ever,dataonfoulingthicknessareexcluded,andimportant cleaningparameterssuchastemperatureandflowrateare notincludedasoptimizationvariables.

Theobjectiveofthisstudywasthustoinvestigate poten-tialbenefitsofacomprehensivecleaningkineticsmodelthat considerstheinfluenceofthreemajoroperatingparameters ofcleaning(concentration,temperatureandflowrate)onthe degreeoffoulingremovalintheeco-designofafoodprocess. The process under study is a falling film evaporator thatprocessesskimmilk,whichisespeciallyrelevantsince evaporation,awidelyusedunitoperationinthedairy indus-try,consumeslargeamountsofenergyand generateslarge amountsofeffluents.AccordingtoRamirezetal.(2006),energy consumedduring theevaporator cleaningphase canbe as highas70%ofthatoftheproductionphase.Moreover,water consumedduringthecleaningphasecanrepresent25–40% ofthatduringtheproductionphase(Bosworthet al.,2000). Inaddition,themodelingofevaporatorprocessingskimmilk hasalsoalreadybeeninvestigated(Madoumieretal.,2015)and

couldbenefitfromacomprehensivecleaningkineticsmodel toeco-designthewholeprocess.

Thecleaningkineticsmodelusedinthisstudywasselected fromthosealreadydescribedintheliterature.Toour knowl-edge, nospecificstudy has focusedonthe mechanismsof cleaningorhasmodeledthecleaningkineticsofevaporators fouledbymilk.JeurninkandBrinkman(1994)empirically stud-iedthecleaningofdairyevaporators,butcurrentknowledge cannotquantitativelypredicttheirfoulingandcleaning. Sev-eralotherstudiesseemsuitablehowever,sincetheyfocuson predictingcleaningkineticsofthermaltreatmentequipment fouledbydairydeposits(Gallot-Lavalleeetal.,1984;Birdand Fryer,1991;Xinetal.,2002,2004).

2.

Materials

and

methods

2.1. Descriptionoftheprocesssystem

Thesystemconsideredinthisstudyconsistsofan evapora-torsystemandtheassociatedCIPstation.Forthepurposeof thestudy,itisassumedthattheCIPstationisusedonlyto cleantheevaporatorsystem.Hypothesesaboutthe manage-mentandschedulingofproductionandcleaningarealsoused fortheeco-design.

2.1.1. Evaporator

Theequipmentunderstudyisanindustrialtubularfallingfilm evaporator.Itruns7300hayear,ofwhich6300hare produc-tionand1000haredowntime,whichcorrespondstocleaning and other operations related to productionscheduling. All data on operating conditions and associated consumption camefrom thedairy industry.Theevaporatorprocesses 20 th−1ofskimmilkwith8.5%drymattertoproduce concen-tratedmilkwith47.7%drymatter.Thestandarddurationofits productionphaseis20h.Theevaporatorisconnectedto pre-heaters,apasteurizerandacondenser(confidentialindustry source).Theevaporatorhasthefollowingmain characteris-tics:

• Four effects, inwhichmilkisevaporated successivelyat 69.3,64.2,57.3and49.0◦C.Condensatesofeacheffectare usedtopre-heatmilkbeforepasteurization.

• Asteamejectorrecycleswatervaporfromthesecondeffect, mixingitwith2300kgh−1ofsteamtofeedthefirsteffect 6049kgh−1ofsteamat0.354bar.Therefore,theevaporator consumes2300kgh−1ofsteam.

During theproductionphase,waterisconsumedbythe condenser (16 739 kgh−1),and steam is consumedby the pasteurizer (464 kg h−1) and the evaporator. According to industrialpractice(Cordsetal.,2001),theevaporatoris main-tainedundervacuumduringcleaning,whichmeansthatboth theevaporatorandtheCIPstationconsumesteamandwater duringcleaning.Consequently,theevaporatorwasassumed toconsumeasmuchwaterand steamconsumptionduring cleaningasduringproduction(16739kgh−1ofwaterand2764 kgh−1ofsteam).

2.1.2. Estimatingfouling

IncludingtheCIPprocedureintheeco-designoftheprocess requiresinformationontheamountoffoulingattheendofthe productionphase.Moreprecisely,theefficiencyandduration ofcleaningdependontheamountoffoulingtoberemoved (Xinetal.,2002).Wedidnotconsiderchangesinthenature

FoodandBioproductsProcessing 123 (2020) 427–436

429

Table1–Arbitraryestimatesoffoulingsurfacedensity (M0)after18,20and22hofevaporationtime(tprod),

startingfromthestandardvalueof1.3kg.m−2at20h. Threefoulingkineticsareconsidered:linear(M0=0.065 ×tprod),polynomial(M0=−5×10−5+0.0297×tprod+

0.0018×(tprod)2)andexponential(M0=0.0311×

exp(0.1868×tprod))withMinkg.m−2andtinh.

Productiontime,tprod(h) Foulingsurfacedensity(kgm−2)

Linear Polynomial Exponential

18 1.17 1.10 0.90

20 1.30 1.30 1.30

22 1.43 1.50 1.90

ofthefouling(i.e.compositionandstructure),althoughitcan

changedependingonoperatingconditionsoftheproduction

phase.Sincenomodelcanyetaccuratelypredicttheamount

offoulingofan evaporatorduringproductionofmilk

con-centrates,estimatesorexperimentaldataarenecessary.Caric

etal.(2009)reportedafoulingsurfacedensityof1.3kgm−2ina 4-effectevaporatorafter20hofprocessingwholemilk,while Fosteretal.(1989)demonstratedthatskimmilkproducedthe sameamountoffoulingaswholemilk.Thus,weassumedthat thesurfacedensityoffoulingafter20hofevaporationofskim milkequaled1.3kgm−2(standardvalue).

Tostudytheinfluenceoftheproductionphaseon clean-ing operating conditions, we used three arbitrary laws to estimatefoulingsurfacedensityasafunctionofproduction timesaround20h:i)linear,ii)polynomialandiii)exponential (Table1).Thelinearlawcorrespondstoa5%changeinfouling surfacedensityperhourbeforeorafter20h.Thepolynomial andexponentiallawswereusedtosimulategreaterfouling duetohighlyconcentratedandviscousproducts.Since indus-trialdataindicatedthatproductiontimerangesfrom18to22 h,foulingwasestimatedfor18,20and22h(Table1).

2.1.3. CIPprocedure

Analysisofindustrialpracticesand the literature(Jeurnink andBrinkman,1994;Bosworthetal.,2000;Goodeetal.,2013) identifiedsimilaritiesamongcleaningprocedures.Weused this information to define a standard cleaning procedure (Table2)withfivesteps,eachwithspecificoperating condi-tions(temperatureand concentration)and duration.For all steps,theflowratewasset at130%oftheproductionflow rate,followingindustrialpractices.

Giventhisstandardcleaningprocedure(Table2),the stan-dardproductiontime(20h)andyearlyhoursofproductionand downtime,atotalof314cycles(i.e.productionandcleaning) arerunperyear.Asobservedintheindustry,whenseveral productionlines exist, especiallywhen severalevaporators areconnectedtoasinglespraydryer,schedulingconstraints resultinlagtimes.Torepresentthelagtimebetweencycles, 80minwasaddedtothetotalcycletime,thusyieldingatotal standardcycletimeof1390min.

2.1.4. TheCIPstation

TheCIPstationthatproducesthecleaningsolutionsconsists oftwosimpleunitoperations:

1) mixingofwaterandchemicaldetergentstoproducethe cleaningsolutions

2) heatingofthecleaningsolutionswithsteam(assumed effi-ciency:80%-Martínez,2017).

Bothoperationsareconsideredcontinuoussinceonlythe durationofagivencleaningstepisusedtocalculate consump-tion(i.e.water,steam,detergent).Electricityconsumptionof theCIPprocedurewasnotconsidered,sincelittleinformation onitisavailable,andCIPconsumesonlyasmallamountof electricityindairyplants(lessthan5%,accordingtoGugala etal.(2015)).

SincetheCIP stationisdesignedforsingle-usecleaning solutions,cleaningsolutionsandrinsewateraredischargedas wastewaterimmediatelyafteruse,whichisthesamepractice usedforindustrialevaporators.

Steamconsumption(correspondingtothesteamflowrate ˙

mSteamofthecleaningphaseiscalculatedas:

˙

mSteam= H˙Solution

0.8×10−6LVap_Water (1)

whereLVap_Water istheenthalpyofvaporizationofwater(atthe steamtemperature),and ˙HSolutiontheenthalpyraterequiredto heatthecleaningsolutions.

Eq. (1) assumes total condensation of steam with the exchangeofonlylatentheat. ˙HSolutioniscalculatedas:

˙

HSolution= ˙mSolution



T1

CpSolutiondT (2)

where ˙mSolution is the flow rate of the cleaning solution, CpSolutionitsheatcapacityandTitstemperature.T1andT2are respectivelytheambienttemperature(15◦C)andthecleaning temperatureofeachstepintheprocedure(Table2).

Eq.(2)usesempiricalregressionsoftheheatcapacityof causticsodaandnitricacidsolutions(Eqs.(3)and(4), respec-tively)fromAspenPlussoftware(Haydary,2019):

CpCausticsoda=3228.49–4444.73×XNaOH+16.04×T (3)

CpNitricacidsolution=3190.06–3293.52×XNitricacid+16.16×T (4)

whereXisdetergentconcentration(kgkg−1).

Table2–Standardcleaningprocedureforanindustrialdairyevaporatorfouledwithskimmilkaccordingtoindustrial practices

No. Step Detergent Temperature Duration(min)

1 Pre-rinse Water Ambient 20

2 Alkalicleaning Sodiumhydroxideat1.5% 75◦C 30

3 Intermediaterinse Water Ambient 15

4 Acidcleaning Nitricacidat1.5% 60◦C 30

2.2. Alkalicleaningkineticsmodelforanevaporator fouledbymilk

2.2.1. Choiceofthecleaningmodel

Thecleaningmodelisusedtopredictthedurationofalkali cleaning required to remove the deposit as a function of themajoroperatingparametersofthecleaning(alkali con-centration,temperature,flowrate).Manymodelshavebeen developed,mainlyforheatexchangers,toexplainand/or pre-dictthekineticsofremovingdairydeposits.Gallot-Lavallee etal.(1984)(GL)developedasimplemodelofcleaningtime asafunctionofcleaningparameters.ThemodelofBirdand Fryer (1991) and Bird (1992) was the first to consider that thecleaningrateincreasesassodiumhydroxide concentra-tionincreases,uptothepointthattheconcentrationinhibits cleaning.Otherstudiesalsoidentifiedthisbehaviorindairy processes(JeurninkandBrinkman,1994;Lötscheretal.,1994), butnostudieshaveincludedcleaningparametersinamodel topredictthekineticconstantsrequiredtocalculatethe clean-ingduration.Morerecently,longexposuretohotsurfaceswas identifiedtoinfluencefoulingcharacteristics(e.g. structure, thermalproperties,chemicalreactivity)(Ishiyamaetal.,2011). Thisphenomenon,called“aging”,alsodecreasesthe clean-ing rate, but isnot yetincluded in predictionsof cleaning duration.Xinetal.(2002,2004)developedamodeltopredict removalkineticsofwheyproteinconcentratefoulingusing avarietyofkineticparametersidentifiedfromexperimental data;however,theirmodeldidnotrelatekineticparameters tocleaningparameters(concentration,temperature,flowrate) mathematically.

Thus,duetothelackofcleaningkineticmodelsthat con-siderthe mostrecent advancesincleaningknowledge, we selectedtheGLmodelforthisstudybecauseit(i)istheonly onethatestimatesthecleaningkineticsconstantasa func-tionofconcentration,temperatureandflowrate;(ii)requires asingleparameterforfouling(i.e.initialsurfacedensity)and (iii)doesnotrequirecomplexmodelingoffouledequipment, whichisdesirablewhenoptimizingprocessesatthefactory level.

2.2.2. DescriptionoftheGLmodelandpredictionof cleaningduration

The GL model predicts cleaning duration as a function of cleaningparameters(concentration,temperature,flowrate, foulingsurfacedensity).Themodelquantifiesthekineticsof milkfoulingremovalwithcausticsodaasakineticconstantk

(s−1),asfollows:

log (k)=−1.01+0.27A1+0.20A2+0.16A3−0.67M0 (5) whereM0(kg.m−2)isthefoulingsurfacedensity,andA1,A2 andA3arereducedvariablesfortheoperatingparametersof cleaning:

• Alkali solution temperature (TinK): A1= (T−273.15−75) /12

• Sodiumhydroxideconcentration(XNaOH inkgkg−1)A2= (100XNaOH−2) /1.15

• Alkalisolutionflowvelocity(vinm.s−1):A3= (v−1.1) /0.5 Oncethekineticconstantkiscalculated,thecleaningrate (rF)iscalculatedasafunctionoftime(t):

rF=M0kte−kt (6)

Table3–Costsofutilitiesandrawmaterials,and expectedrevenuefromconcentratedmilk.

Costorrevenue Value

Steam(Dkg−1)(confidentialindustrysource) 0.032 Water(Dm−3)(confidentialindustrysource) 0.85 Causticsoda–30%solution(D t−1)(industrial

data)

200 Nitricacid–56%solution(Dt−1)(industrialdata) 200 Rawmilk–fromproducer(2014mean,France)

(D kg−1)(CNIEL,2015)

0.355

50%-concentratedmilkrevenue(Dkg−1)(CNIEL, 2015)

2.636

AccordingtoBirdandFryer(BirdandFryer,1991;Bird,1992), cleaningcanbeconsideredcompletewhen,afterreachinga maximumvalue,thecleaningratedecreasesandreaches2% ofthismaximumvalue.Completecleaningtimeisthusthe timeatwhichthiscriterionismet.

Whentestedwithoperatingconditionsofalkalicleaningin thestandardprocedure(75◦C,1.5%sodiumhydroxide,0.012m s−1flowvelocity);theGLmodelpredictedacleaningduration of24min,whichissimilartothe30minofalkalicleaningused atthe industrial scale.Althoughthe model wasdeveloped forthecleaningofholdingtubes,andsincemany hypothe-sesarepotentialsourcesofvariation,thedurationpredicted isconsistentwithindustrialpractices.

2.3. Eco-designframeworkandkeyindicatorsfor optimization

AnoptimizationframeworkimplementedinanExcel spread-sheet(Microsoft,Inc.)wasusedtoeco-designtheevaporator processincludingproductionandcleaningphases. Optimiza-tionentailedsearchingforoptimalvariablesforalkalicleaning ortheproductionphasetominimizeormaximizeoneor sev-eralindicators.Thevariableswerethefollowing:

• Alkalisolutionconcentration,0.1–3.9%(boundariesofthe GLmodel)

• Alkalisolutiontemperature,55–95◦C(boundariesoftheGL model)

• Alkalisolutionflowrate,100–130%ofproductionflowrates • Productionduration,18,20or22h.

• Foulingkineticslawfortheproductionphase,linear, poly-nomialorexponential(Table1)

Indicators forboth environmental and economic objec-tivesweredefinedtoperformtheoptimization.Weconsidered three environmentalindicators(i.e.consumption ofsteam, water, and sodium hydroxide) and two economic indica-tors(i.e.yearlyproduction(concentratedmilkproducedper year) and gross profit (gross revenue minus the total cost ofutilitiesandrawmaterials(i.e.milk,steam,water, deter-gents). Althoughmilk evaporation iscommonly used as a pre-concentrationstepbeforespraydrying,grossrevenuewas related to the milk concentrate instead of the milk pow-der obtainedbyspraydrying.Includingmilkpowderinthe boundariesofthesystemwouldhaverequiredestimating con-sumptionandemissionsofspraydryingandpackaging,which laybeyondthescopeofthisstudy.Allindicatorsofcostsand revenueswerecalculatedonayearlybasis(Table3).

Twotypesofoptimizationoftheevaporationsystemwere performed:

FoodandBioproductsProcessing 123 (2020) 427–436

431

• Single-objectiveoptimization,whichcanbeconsidered eco-designwhentheobjectiveistominimizeanenvironmental indicator.WeusedExcel’sSolvertooltominimizean indi-catorbychangingdesignvariables.

• Multi-objectiveoptimization(MOO),inwhichseveral poten-tially conflicting objectives are considered. We used a genetic algorithm (Multigen code developed in a previ-ousstudy(Gomezetal.,2010))toproducePareto-efficient alternatives and a multiple-criteriadecision-making tool (M-TOPSIS,avariantofTOPSIS(TechniqueforOrder Pref-erencebySimilaritytoIdealSolution)—HwangandYoon, 1981;Lietal.,2009;Mendozaetal.,2014)torankthe com-promisesolutions.ThebasicconceptofM-TOPSISisthat the selected alternative lie closest to the “positive ideal solution” and farthest from the “negativeidealsolution” inageometricalsense.Itwasselectedforitsrationaland understandable logic, and its straightforward calculation process(Garcia-CascalesandLamata,2012).Eachobjective wasgiventhesameweight(i.e.1).

2.4. Strategyofwork

Inthisstudy,wefirstestimatedthecontributionofcleaning totheoverallperformanceoftheprocesstoconfirmthatthe cleaningphasehashighcostsandenvironmentalimpactsin theevaporationprocess.Theestimationwasperformed

i) forthestandardproductionandcleaningphases(Table2) and

ii) forthreeproductiondurations(18,20and22h)andthree foulinglevelsandkinetics(Table1).

Wethenoptimizedtheprocessconsideringfourobjectives: minimizingconsumptionofsteam,water,orsodium hydrox-ide,andmaximizinggrossprofit.

i) First,the cleaningprocedure was optimizedbased on a standardproductionduration toassess the influenceof optimizedcleaningproceduresontheoverallprocess. ii) Next,the entireprocess wasoptimized (MOO)toobtain

optimalindicatorvaluesbasedonproductionandcleaning parameters(cleaningconcentration,temperatureandflow rate,andproductionphaseduration).Bothsingle-objective andbi-objective optimizationwereperformed usingthe threefoulingkineticslaws.Thus,theproductionphasewas includedintheoptimizationprocedurethroughthefouling kineticslaws.

3.

Results

and

discussion

3.1. Contributionofcleaningtoprocessperformance

3.1.1. Contributionofcleaningtothestandardproduction andcleaningphases

Forstandardproceduresoftheproductionphase(20h)and cleaningphase(Table2),theinfluenceofthecleaningphase wasnotnegligiblecomparedtothatoftheproductionphase. Althoughcleaningtooklesstimethanproduction(110min, i.e.10%ofproductiontime), itconsumedlargeamounts of steamandwaterandgeneratednon-negligible environmen-talimpacts:cleaningrepresented15%(1930MWhyear−1)of the productionphase’s steamconsumption and up to23% (24000m3_year−1_{)ofitswaterconsumption(Table4).Thus,}

Table4–Environmentalandeconomicindicatorsfor standardindustrialproductionandcleaning

parameters.Operatingconditionsforalkalicleaning: 1.5%massconcentration,75◦C,130%ofproductionflow rate,1.30kg.m−2offoulingafter20hofproduction,30 minofalkalicleaning.

Indicator Phase Value

Numberofcyclesperyear 314

Steamconsumption (MWhyear−1₎

Production 13203 Cleaning 1930 Waterconsumption

(103_m3_year−1₎ Production_Cleaning 105₂₄

Sodiumhydroxideconsumption(tyear−1) Cleaning 61 Yearlyproduction(tyear−1) Production 22671 Grossprofit(kD year−1) Production 14416

optimizingtheevaporatorprocessviathecleaningphaseis

challenging.

3.1.2. Contributionofcleaningconsideringthreefouling kinetics

Predictionsofthealkalicleaningdurationrequiredtoremove

foulingcompletelyrevealedtheinfluenceofproduction

dura-tionandfoulingkinetics(Fig.1).

Thesteamandwaterconsumptionofthecleaningphase representedasubstantialpercentageoftheconsumptionof production(13–19%and 21–27%,respectively);however, the corresponding cleaningduration rangedfrom 13 (exponen-tialkineticsafter18h)to60min(exponentialkineticsafter 22h)(Fig.1).Thelongertheproductionduration,thelonger the cleaningduration,whichisconsistentwithlonger pro-ductiondurationsincreasingthefoulingsurfacedensity.After a productionphase of22 h, cleaning duration nearly dou-bledwithexponentialfoulingkinetics(60min)comparedto linearandpolynomialkinetics(29and32min,respectively), whereas foulingsurfacedensity increasedbyonly26% and 32%,respectively. Thisindicatesthat forfluidswitha high fouling potential(i.e.exponentialfoulingkinetics),cleaning durationmustincreaseconsiderably(here,nearlydouble)if productiondurationincreasesbyafewpercentagepoints(10% of the standard duration), asobserved in the dairy indus-try.

As production duration (and thus cleaning duration) increased,steam,waterandsodiumhydroxideconsumption increased(Fig.1a–c).Moreresourceswererequiredtocleanfor longerperiods.Overall,theenvironmentalindicatorsfollowed thesametrendsasthehypothesesforcleaningdurationand fouling kinetics:consumptionofinputsincreasedas clean-ingdurationincreased,andforagivenproductionduration, it increased asfouling surface density increased.However, economic indicators followed different trends (Fig. 1d and e). For linear and polynomial fouling,yearly productionof milk concentrate and gross profit increased with produc-tion duration, sincemore product was produced per year. Thus,theincreaseinyearlyproductioncompensatedforthe decrease in productoutput per production cycle, and con-sequentlyincreasedtheyearlygrossprofit.For exponential fouling,bothyearlyproductionandgrossprofitwereminimal for 22 h of production. Despite a longer production dura-tion,thelongercleaningdurationdecreasedthenumberof productioncyclesperyear(i.e.,from314to284),thus produc-inglessproductperyear.Grossprofitfollowedthetrendfor yearlyproductioncloselybecausetheproduct(i.e.milk

con-Fig.1–Environmentalandeconomicindicatorsforstandardalkalicleaningparameters,variableproductiondurationsand threefoulingkinetics.Numbersabovehistogramsshowindicatorvaluesat18and22hofproductiondurationandthoseat 20h.

centrate)represented98% oftotalcosts,followed bysteam (1.4%).

Thus, cleaning influenced overall system performance greatly when productionduration changed. Since cleaning duration must adapt to the amount of fouling, the eco-nomicandenvironmentalindicatorsfollowdifferenttrends according to the fouling kinetics. Thus, it is necessary to understandfouling kinetics inthe evaporator toadapt the cleaningprocedureandtooptimizethesystem.Sinceyearly productionandgrossprofitfollowedsimilartrendsasa func-tion of production time and fouling kinetics (Fig. 1d and e),onlygrossprofit wasused inthe subsequent optimiza-tion.

3.2. Optimizationtoeco-designtheevaporatorprocess

3.2.1. Optimizationofcleaningprocedurewiththe standardproductionduration

Regardlessoftheindividualobjectivetobeoptimized (min-imizingconsumptionofsteam,water,orsodiumhydroxide, or maximizing grossprofit), optimal indicator values were obtainedatthelowestflowrate(100%)andhighest tempera-ture(95◦C)intheoptimizationframework(Table5).Although one might expect the highest flow rate (130%) to increase performances(higherflowrateisknowntodecrease clean-ing duration; Gallot-Lavallee et al. (1984)), the lowest flow rate compensated forthe increase in cleaningduration by

FoodandBioproductsProcessing 123 (2020) 427–436

433

Table5–Single-objectiveoptimizationoftheevaporatorprocessusingthestandardproductionphase(production duration:20h;Foulingsurfacedensity:1.30kgm−2).Optimizedcleaningoperatingparametersareshownalongwith indicatorvalues.Valueswithpositiveandnegativesignsrepresentdifferencesfromthestandardindustrialpractice. NaOH=sodiumhydroxide.

Standardindustrial practice Optimizationobjectives Minimize steamcons. Minimizewater cons. Minimize NaOHcons. Maximizegross profit Optimizedcleaning parameters Concentration(%mass) 1.5 3.9 3.9 0.1 3.9 Temperature(◦C) 75 95 95 95 95

Flowrate(%productionrate) 130 100 100 100 100

Indicators and key vari-ables

Alkalicleaningtime(min) 30 3 3 15 3

Numberofcyclesperyear 314 320 +6 320 +6 318 +4 320 +6

Steamconsumption(MWhyear−1) 15133 14749 −384 14749 −384 14973 −161 14749 −384 Waterconsumption(103_m3_year−1_{)129.6 125.9 −3.7 125.9 −3.7 127.4 −2.2 125.9 −3.7}

NaOHconsumption(t.year−1_{) 61 13 −48 13 −48 2 −60 13 −48}

Yearlyproduction(tyear−1) 22671 23104 +433 23104 +433 22960 +289 23104 +433 Grossprofit(kDyear−1) 14416 14765 +349 14765 +349 14666 +249 14765 +349

Fig.2–Paretofrontforbi-objectiveoptimizationoftheevaporatorsystembasedonthestandardproductionprocedure. ArrowsidentifythetopthreealternativesaccordingtoM-TOPSISranking.

decreasingsteam,waterand NaOHconsumption.Similarly,

thehighesttemperature yieldedthe highestoverall

perfor-mance,sinceyearlycostswerelowerdespitetheneedformore

heatingpower.Theremainingcleaningparameter—

concen-tration—wasthemostrelevantparameterforoptimizingthe

process.

Regardless ofthe individual objective optimized,

single-objective optimization resulted in substantially shorter

cleaningdurations,whichimprovedtheenvironmentaland

economicindicators(Table5).Steamandwaterconsumption

decreasedby2.5%and2.8%,respectively,resultingina2.4% increase ingross profit, which corresponds to an absolute increaseof349kD per year.Thelargest improvementwas obtainedwhenminimizingsodiumhydroxideconsumption, whichdecreasedbymorethan90%(from61to2tyear−1).

Theseresultsindicatethat includingacleaningkinetics model inoptimization canimprove overall process perfor-mancegreatlyandcanalsohelpidentifyoperatingparameters towhichprocessperformanceismoresensitive.Theseresults alsodemonstratethatfortheevaporatorsystemandunder the restrictions of this study, (i) maximizing gross profit and minimizingwaterand steamconsumption are consis-tent(non-antagonistic)objectives,but(ii)minimizingsodium hydroxide consumption is antagonistic to the other three objectives.

Consequently,MOO was performedwithtwo objectives: minimize sodium hydroxide consumption and maximize gross profit. The genetic algorithm generated 10 Pareto-optimalalternatives(Fig.2).Sodiumhydroxideconsumption

variedgreatly(1.6–13.4tyear−1),but8ofthealternativeshad asimilarincreaseingrossprofitcomparedtothe standard case(ca.2%).Ifadecision-makerweretoconsiderthatthese 8alternativesprovidedessentiallythesamegrossprofit,then allalternativesexcepttheonewiththelowestsodium hydrox-ideconsumptionwouldberemovedfromthedecision-making process.Without informationfrom adecision-maker, how-ever,wekeptall8alternativesforranking.

Among the three best alternatives predicted bythe M-TOPSIS method, the best alternative corresponded to the minimumsodiumhydroxideconsumption(Tables5and6). For MOO, one could have expected a top-ranked solution with intermediate sodium hydroxide consumption, since the best alternatives representedtrade-offs between objec-tives.BecauseofhowM-TOPSISranksalternatives,however, the objectivewithgreater improvements(>80% decreasein sodiumhydroxideconsumption,ratherthanaca.2%increase ingrossprofit)hadahigherrank.Theothertwobest alterna-tivesincludedhighersodiumhydroxide consumption(0.7% and0.8%,respectively)andahighergrossprofitthanthebest M-TOPSISalternative(Table6).

The optimization framework helped identify potential increasesinprocessperformance.Amongalloptimizations, improvementsinindustrialpracticewereashighas97%for sodium hydroxide consumption, 2.5% for steam consump-tion, 2.8%forwaterconsumptionand 2.4%forgrossprofit (Tables 5 and 6). The cleaning operating parameters cor-responding totheseimprovementsdiffersignificantlyfrom those used in standard industrial practice, which

demon-Table6–TopthreePareto-optimalalternativesaccordingtotheM-TOPSISmethodforbi-objectiveoptimizationofthe evaporatorprocesswith20hofproductionduration.Bothobjectiveshavethesameweight(1).

Parameterorvariable M-TOPSISno.1 M-TOPSISno.2 M-TOPSISno.3 Optimized

parameters

Concentration(%mass) 0.1 0.7 0.8

Temperature(◦C) 95 95 95

Flowrate(%productionrate) 100 100 100

Indicatorsand miscellaneous variables

Alkalicleaningtime(min) 15 12 11

Numberofcyclesperyear 318 319 319

Steamconsumption(MWhyear−1) 14973 14934 14921

Waterconsumption(103_m3_year−1_{) 127.4 127.2 127.1}

Sodiumhydroxideconsumption(tyear−1) 1.6 8.7 9.5

Yearlyproduction(tyear−1) 22960 23032 23032

Grossprofit(kDyear−1)Grossprofitincrease(%) 14666 1.73 14711 2.05 14711 2.05

Fig.3–Optimalenvironmentalandeconomicindicatorsasafunctionoffoulingkineticslaws(linear,polynomialand exponential).The“20hproductionduration”correspondstosingle-objectiveoptimizationofthecleaningphase.

stratesthebenefitsofincludingthecleaningphaseinprocess

optimizationandeco-design.

3.2.2. Optimizationoftheentireevaporatorprocess consideringproductionandcleaningphases

Theoptimalvaluesofthefoureconomicandenvironmental

indicatorsvariedaftersingle-objectiveoptimization(Fig.3).

SeeTableS1(Supplementarymaterials)forvaluesof indica-torsandoptimizedoperatingparameters.

Including both cleaning and production phases in the optimization procedure improved indicators compared to optimizingonlythecleaningphasewithafixeddurationof20 h,regardlessofthefoulingkineticslawused(Fig.3).Steamand waterconsumptiondecreasedbylessthan1%,whilesodium hydroxideconsumptiondecreasedby41%(withexponential fouling),andgrossprofitincreasedbyca.1%(137–196kD)for alllawsused.

Optimalproductiondurationscanbeidentified,whilehow muchthefoulingkineticslawinfluencesprocessperformance

FoodandBioproductsProcessing 123 (2020) 427–436

435

Table7–Bi-objectiveoptimizationresultsasafunctionoffoulingkineticslawwiththestandardindustrialcaseandwith bi-objectiveoptimizationfor20hofproductionduration.

Standard industrialcase M-TOPSISno.1 with20h production duration Foulingkinetics lawusedwith M-TOPSISno.1

Linear Polynomial Exponential

Productionduration(h) 20 20 22.0 21.9 20.3

Concentration(%mass) 1.5 0.1 0.1 0.1 0.1

Temperature(◦C) 75 95 95 95 95

Flowrate(%productionrate) 130 100 100 100 101

Alkalicleaningtime(min) 30 15 18 21 17

Numberofcyclesperyear 314 318 292 292 313

Steamconsumption(MWhyear−1) 15133 14973 15054 15091 15025

Waterconsumption(103_m3_year−1_{) 129.6 127.4 127.3 127.5 127.6}

Sodiumhydroxideconsumption(tyear−1) 61 1.6 1.7 2.0 1.8

Yearlyproduction(tyear−1) 22671 22960 23185 23135 22987

Grossprofit(kDyear−1) 14416 14666 14815 14780 14682

isafunctionofproductionduration.Tominimizesteam

con-sumptionand sodiumhydroxideconsumption,the optimal

productionduration wasca.18 h,withthehypothesisthat

exponential fouling kinetics minimize consumption ofthe

threeinputs.Tominimizewaterconsumption(except with

exponentialfouling)andmaximizegrossprofit,theoptimal

productiondurationwasca.22h.

Resultsofbi-objectiveoptimizationand M-TOPSIS

rank-ing of the fouling kinetics laws showed that the trade-off

alwayscombinedthelowestdetergentconcentration(much

lowerthan themaximum2%sodiumhydroxide

concentra-tiondeterminedbyBirdandFryer(1991)andBird(1992),the

highesttemperatureandthelowest(ornearlyso)flowrate,as observedwitha20hproductionduration(Table7).Although anoptimizedproductiondurationtendedtohavehigher con-sumption,grossprofit increasedby1%withlinear fouling, whichrepresentsa2.8%increasecomparedtothestandard industrialcase.Thus,tooptimizegrossprofit,itseems appro-priate to includeproduction duration among optimization variables.Doingsoalsowidenstherangeofpossibleoperating parametersforcleaningandproduction,andthuswidensthe possibletrade-offswhenoptimizingtheprocess.

Optimizationofcleaningandproductionphaseparameters revealstheexistenceofoptimaloperatingconditions,which canresultinbetteroverallprocessperformancethan optimiz-ingonlycleaning.Eco-designoftheevaporatorprocessthus benefitsfromincludingcleaningandfoulingkineticsinthe MOOframework.

4.

Conclusion

Thisstudydevelopedaneco-designapproachforadairy evap-orationprocessthatcombinesacleaningkineticsmodeland fouling kineticshypotheses. Resultsindicatethat consider-ing bothcleaning and productionphaseswhen optimizing theevaporatorprocesscanresultinimprovementscompared tobothindustrialpractices,evenwithuncertaintyinfouling kinetics.Doingsocanalsohelpestimatepotentialeconomic savingsandreductioninenvironmentalimpacts.The evapo-rationprocesswasoptimizedatahighcleaningtemperature (95◦C), aflowratesimilar tothatusedduringthe produc-tionphase andalow detergentconcentration(<2%).These improvementscanalso beappliedwithlittle to no invest-ment,sincecleaningparameters(concentration,temperature, flowrate)and productionduration influencemainly(if not

only)theoperationoftheprocess.Beforetheseimprovements areimplemented,however,theeco-designapproachcouldbe developedfurtherbyconsideringrecentadvancesin model-ingthecleaningprocessusingcomputerfluiddynamics(e.g. Joppaet al. (2017)),along withdeeper knowledgeof clean-ingkineticstopredictprocessperformanceaccurately.Special attentionshouldbepaidtotheestablishmentandintegration ofcleaningmodelassumingnonlinearinfluencewithNaOH concentration.

Declaration

of

interests

The authors declare that they have no known competing financialinterestsorpersonalrelationshipsthatcouldhave appearedtoinfluencetheworkreportedinthispaper.

Declaration

of

Competing

Interest

Theauthorsreportnodeclarationsofinterest.

Acknowledgements

TheauthorsthanktheFrenchNationalInstitutefor Agricul-ture,FoodandEnvironmentandtheBrittanyRegionGeneral Councilforfundingthisresearch.TheyalsothankBenoitColin (S.A.Thermique-GénieChimique-Evaporation),MehdiDif (ElodysInternational),GilbertSteunou(Euroserum)andGaëlle Tanguy(FrenchNational InstituteforAgriculture,Food and Environment)forfruitfuldiscussions.

Appendix

A.

Supplementary

data

Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10 .1016/j.fbp.2020.07.023.

References

Banga,J.R.,Balsa-Canto,E.,Alonso,A.A.,2008.Qualityandsafety modelsandoptimizationaspartofcomputer-integrated manufacturing.Compr.Rev.FoodSci.FoodSaf.7(1),168–174. Bird,M.R.,1992.CleaningofFoodProcessPlant.Ph.D.Thesis.

Bird,M.R.,Fryer,P.J.,1991.Experimentalstudyofthecleaningof surfacesfouledbywheyproteins.FoodBioprod.Process.69 (1),13–21.

Birewar,D.B.,Grossmann,I.E.,1989.Incorporatingschedulingin theoptimaldesignofmultiproductbatchplants.Comput. Chem.Eng.13(1),141–161.

Bosworth,M.E.D.,Hummelmose,B.,Christiansen,K.,Technical report,COWIConsultingEngineersandPlannersAS,Denmark 2000.CleanerProductionAssessmentinDairyProcessing. Caric,M.,Milanovic,S.,Akkerman,C.,Kentish,S.,Tamime,A.,

2009.Technologyofevaporators,membraneprocessingand dryers.In:DairyPowders&ConcentratedProducts.Wiley VCH,Chichester,UK.

CNIEL-CentreNationalInterprofessionneldel’Economie Laitière,URL:2015.L’économielaitièreenchiffres-Edition 2015.http://fr.calameo.com/books/002230051347964ef558f. Cords,B.,Dychdala,G.,Richter,F.,2001.Cleaningandsanitizing

inmilkproductionandprocessing.In:Dekker,Marcel(Ed.), AppliedDairyMicrobiology.,Seconded,Inc.

Eide,M.,2002.Lifecycleassessment(LCA)ofindustrialmilk production.Int.J.LifeCycleAssess.7(2),115–126. Eide,M.H.,Ohlsson,T.,1998.Acomparisonoftwodifferent

approachestoinventoryanalysisofdairies.Int.J.LifeCycle Assess.3(4),209–215.

Eide,M.H.,Homleid,J.P.,Mattsson,B.,2003.Lifecycleassessment (LCA)ofcleaning-in-placeprocessesindairies.Lwt-FoodSci. Technol.36(3),303–314.

Erdo ˘gdu,F.,2008.OptimizationinFoodEngineering.CRCPress. Foster,C.L.,Britten,M.,Green,M.L.,1989.Amodelheat-exchange

apparatusfortheinvestigationoffoulingofstainlesssteel surfacesbymilkI.Depositformationat100C.J.DairyRes.56 (2),201–209.

Gallot-Lavallee,T.,Lalande,M.,Corrieu,G.,1984.Cleaning kineticsmodelingofholdingtubesfouledduringmilk Pasteurization1.J.FoodProcessEng.7(2),123–142.

Garcia-Cascales,M.S.,Lamata,M.T.,2012.Onrankreversaland TOPSISmethod.Math.Comput.Model.56(5),123–132. Gomez,A.,Pibouleau,L.,Azzaro-Pantel,C.,Domenech,S.,Latgé,

C.,Haubensack,D.,2010.Multiobjectivegeneticalgorithm strategiesforelectricityproductionfromgenerationIV nucleartechnology.EnergyConvers.Manage.51(4),859–871. Goode,K.R.,Asteriadou,K.,Robbins,P.T.,Fryer,P.J.,2013.Fouling

andcleaningstudiesinthefoodandbeverageindustry classifiedbycleaningtype.Compr.Rev.FoodSci.FoodSaf.12 (2),121–143.

Gugala,M.,Zarzecka,K.,Kapela,K.,Sikorska,A.,2015.Analysisof EnergyConsumptionintheFacilitiesoftheCiechanówDairy Cooperative.J.Ecol.Eng.16(3).

Hwang,C.-L.,Yoon,K.,1981.Methodsformultipleattribute decisionmaking.In:MultipleAttributeDecisionMaking. Springer.

Ishiyama,E.M.,Paterson,W.R.,IanWilson,D.,2011.Optimum cleaningcyclesforheattransferequipmentundergoing foulingandageing.Chem.Eng.Sci.66(4),604–612.

Jeurnink,T.J.M.,Brinkman,D.W.,1994.Thecleaningofheat exchangersandevaporatorsafterprocessingmilkorwhey. Int.DairyJ.4(4),347–368.

Joppa,M.,Köhler,H.,Rüdiger,F.,Majschak,J.-P.,Fröhlich,J.,2017. ExperimentsandSimulationsontheCleaningofaSwellable SoilinPlaneChannelFlow.HeatTransf.Eng.38(7–8),786–795. Haydary,Juma,2019.ChemicalProcessDesignandSimulation:

AspenPlusandAspenHysysApplicationsEdition.Hardcover Wiley&SonsLtd.,JohnWiley&Sons,ISBN:

978-1-119-08911-7;448Pages.

Li,C.,Zhang,X.,Zhang,S.,Suzuki,K.,2009.Environmentally consciousdesignofchemicalprocessesandproducts: multi-optimizationmethod.Chem.Eng.Res.Des.87(2), 233–243.

Lötscher,D.,Henck,M.,Gallmann,P.,1994.HöhereDosisreinigt wederbessernochschneller.DeutscheMilchwirtschaft45 (17),783–784.

Madoumier,M.,Azzaro-Pantel,C.,Tanguy,G.,Gésan-Guiziou,G., 2015.Modellingthepropertiesofliquidfoodsforuseof processflowsheetingsimulators:applicationtomilk concentration.J.FoodEng.164,70–89.

Martínez,I.,2017.Heatexchangers,lecturesonheatandmass transfer.Ind.Alim.Agr.

Marty,P.,2001.Maîtrisedelaconsommationd’eaudansles industriesagro-alimentaires.Industriesalimentaireset agricoles118(5),35–39.

Mendoza,L.-F.M.,Boix,M.,Azzaro-Pantel,C.,Montastruc,L., Domenech,S.,2014.Biodieselproductionfromwaste vegetableoils:combiningprocessmodelling,multiobjective optimizationandlifecycleassessment(LCA).Comput.Aided Chem.Eng.,235–240.

Ramirez,C.A.,Patel,M.,Blok,K.,2006.Fromfluidmilktomilk powder:EnergyuseandenergyefficiencyintheEuropean dairyindustry.Energy31(12),1984–2004.

Sage,M.,2005.Valorisationdeseauxdeprocédésdel’industrie laitièreenméthanisationetdénitrification.PhDthesis.École NationaleSupérieured’AgronomiedeRennes,France. Sharma,S.,Rangaiah,G.P.,Cheah,K.S.,2012.Multi-objective

optimizationusingMSExcelwithanapplicationtodesignof afalling-filmevaporatorsystem.FoodBioprod.Process.90(2), 123–134.

Stefanis,S.,Livingston,A.,Pistikopoulos,E.,1997.Environmental impactconsiderationsintheoptimaldesignandscheduling ofbatchprocesses.Comput.Chem.Eng.21(10),1073–1094. Tamime,A.Y.,2009.Cleaning-in-Place:Dairy,FoodandBeverage

Operations.JohnWiley&Sons,Chichester,UK.

Xin,H.,Chen,X.,Özkan,N.,2002.Cleaningrateintheuniform cleaningstageforwheyproteingeldeposits.FoodBioprod. Process.80(4),240–246.

Xin,H.,Chen,X.D.,Özkan,N.,2004.Removalofamodelprotein foulantfrommetalsurfaces.AIChEJ.50(8),1961–1973.