Mathematical Modeling of Coalescence

Mathematical Modeling of Coalescence
of Oil Droplets in Water Flow
Eirik Kufs
Master of Science in Energy and Environment

Submission date: June 2008
Supervisor:
Kjell Erik Rian, EPT
Norwegian University of Science and Technology

Department of Energy and Process Engineering
Problem Description
Through this Master's thesis the goal is to study mathematical models for coalescence of oil
droplets in a water flow. The industrial basis for this project is the need for a mathematical model
that can be applied in CFD simulations for the optimization of swirl-based coalescers.
The project will be carried out in close cooperation with Aker Process Systems AS, Division of
Advanced Separation.
The following tasks should be considered in the project work:
1.
Carry out a literature review on liquid-liquid coalescence with special focus on oil droplet
coalescence in a continuous water phase. Give a thorough discussion of the problem.
2.
Formulate a mathematical model for coalescence of oil droplets in a water flow.
3.
In cooperation with the advisors, select a suitable numerical simulation tool to study
coalescence of oil droplets in a water flow. Give a description of the relevant parts of this program
system.Perform relevant numerical simulations and discuss the results thoroughly.
4.
If necessary, suggest further improvements of the mathematical model.
Assignment given: 01. February 2008

Supervisor: Kjell Erik Rian, EPT
Declaration
I hereby declare that this Masters thesis is prepared independently and
according to the rules and regulations of NTNU as well as passive directions
givenbytheDepartmentofEnergyandProcessEngineering.
Oslo27/06/08
____________________________
EirikKufs
ii
Preface
This Masters thesis is carried out at the Department of Energy and Process
Engineering, NTNU during the spring 2008. The project is carried out in
cooperationwithAkerProcessSystemsAS,DivisionofAdvancedSeparation.
I would like to thank my supervisors at NTNU during this Master thesis, Kjell
Erik Rian, Maria Fernandino and Carlos Dorao for helpful advices during the
work,aswellasMortenHanaandVishwasDindoreatAkerProcessSystemsAS
forveryinterestingproposalforthisMasterassignment.
AspecialthanksgoestomygoodfriendsthatIhavebeensharingworkingplace
withoverthelastyearforkeepingupagoodmoodthroughoutthewholeperiod
forthiswork.
Oslo27/06/08
___________________________
EirikKufs
iii
Abstract
Liquidliquid coalescers are devices used for increasing the droplet size of the
dispersedphaseincontinuousphaseflow,suchasoildropletsinwaterflow.The
efficiency of separation technologies is strongly dependent on the droplet size,
which is desirable to shift into larger droplet diameters. Theory behind
coalescence and its modeling is studied in this Masers thesis. Aker Process
Systems AS, Division of Advanced Separation Technology, provided the
assignmentproposal.
Thescopeofthisworkisaliteraturestudyonthecoalescencephenomenonand
thecloselyrelatedbreakupphenomenonandCFDmodelingingeneral.Further
a mathematical model for simulating coalescence of oil droplets in continuous
water flow is developed by the use of the commercial CFDcode FLUENT. The
basisforthemodelisaswirlbasedcoalescercalledCompactTubularCoalescer
(CTC),developedbyAkerProcessSystemsAS.
The validity of the model is evaluated before different aspects of the

performanceofthecoalescerarestudied.Severalvalidationcriteriaweretested
and were acceptable, but some weaknesses regarding lack of test cases were
detected.TheperformancetestingshowedgoodperformanceoftheCTC,itwas
abletoincreasetheSauterMeanDiameter(SMD)ofthedropletwithupto250%
for the smallest droplets (20 m) and highest volume fractions (7%).
Remarkable differences of the performance were observed as the physical
properties were changed. Higher viscosity and droplet surface tension lead to
increasedcoalescencerateanddecreasedbreakuprate.
Future work is recommended to concentrate on improving the present model

andtoinvestigatemoreaspectsofthemodel.Aneffortshouldalsobemadeto
useaEulerianapproachtomodelthedispersedphasewiththeuseofpopulation
balances, in order to be able to simulate flows with larger dispersed phase
volumefractions.
iv
TableofContents
Declaration ............................................................................................................ 1
Preface................................................................................................................... ii
Abstract ................................................................................................................ iii
ListofFigures ........................................................................................................ vi
ListofTables ........................................................................................................ vii
Nomenclature ..................................................................................................... viii
1 Introduction ..................................................................................................... 1
1.1 Motivation.................................................................................................................................................. 1
1.2 IntroductiontoDispersedTwoPhaseFlows............................................................................. 1
1.3 AimofthePresentWork ..................................................................................................................... 2
1.4 StructureoftheReport ........................................................................................................................ 2
2 PhysicalTheory................................................................................................. 3
2.1 CoalescenceasaPhenomenon ......................................................................................................... 3
2.2 CollisionbetweenParticles................................................................................................................ 3
2.3 BreakupasaPhenomenon ............................................................................................................... 7
3 CFDModelingoftheContinuousPhase .......................................................... 11
3.1 IntroductiontoCFD............................................................................................................................ 11
3.2 GoverningEquationsfortheContinuousPhase .................................................................... 11
3.2.1 MassConservationEquation(ContinuityEquation)........................................................11
3.2.2 MomentumConservationEquation .........................................................................................12
3.2.3 EnergyConservationEquation ..................................................................................................12
3.3 FiniteVolumeMethod....................................................................................................................... 12
3.3.1 SpatialDiscretization.....................................................................................................................13
3.3.2 TemporalDiscretization ...............................................................................................................15
3.3.3 AssemblyoftheDiscretizedEquations ...................................................................................16
3.4 AlgorithmforPressureBasedCalculations............................................................................. 16
3.5 TurbulenceModeling......................................................................................................................... 18
3.5.1 TransportEquationsfortheReynoldsStressEquationModel ....................................19
3.6 GeneralProcedureforUseofCFDcodes................................................................................... 19
3.6.1 Preprocessor .....................................................................................................................................19
3.6.2 Solver .....................................................................................................................................................20
3.6.3 Postprocessor ...................................................................................................................................20
4 DispersedPhaseModeling:CoalescenceandBreakup ................................... 21
4.1 EulerianandLagrangianFramework......................................................................................... 21
4.2 DiscretePhaseModel(DPM).......................................................................................................... 21
4.3 CoalescenceModeling ....................................................................................................................... 24
4.4 BreakupModeling ............................................................................................................................. 26
4.5 CouplingBetweenthePhases ........................................................................................................ 27
5 MethodologyandSimulationSetup............................................................... 29
5.1 PhysicalDescriptionoftheProblem........................................................................................... 29
5.2 GeneralAssumptionsfortheModeloftheCTC...................................................................... 29
5.3 SetupoftheCTCModel ................................................................................................................... 30
5.3.1 GeneralDescriptionoftheCTCModel ....................................................................................30
5.3.2 DiscretizationoftheModel..........................................................................................................31
5.4 HowtheSimulationsareCarriedOut......................................................................................... 31
5.4.1 InvestigatedAspects .......................................................................................................................32
6 ResultsandDiscussion.................................................................................... 35
6.1 ValidityoftheModel .......................................................................................................................... 35
6.1.1 GridIndependency ...........................................................................................................................35
6.1.2 ConvergenceofResiduals .............................................................................................................36
6.1.3 MassFlowBalance ..........................................................................................................................37
6.1.4 TimeIndependency .........................................................................................................................38
6.2 ImpactofChangingDropletsInletDiameter.......................................................................... 39
6.2.1 OnlyCoalescenceModeling..........................................................................................................39
6.2.2 BothCoalescenceandBreakupModeling............................................................................41
6.3 ImpactoftheDispersedPhaseVolumeFraction................................................................... 44
6.4 ImpactoftheTypeofDropletLiquidontheMigration ...................................................... 45
6.5 LimitationsandWeaknessesoftheModel............................................................................... 45
6.6 SuggestedImprovementsoftheModel ..................................................................................... 46
7 Conclusion ...................................................................................................... 49
8 SuggestionsforFutureWork........................................................................... 51
AppendixA,OverviewofSimulationsandLiquidProperties................................. 53
AppendixB,SettingsusedinFLUENT.................................................................... 55
AppendixC,VisualizationofDispersedPhaseConcentration................................ 59
ListofReferences ................................................................................................. 61
vi
ListofFigures
Fig.2.1:Descriptionoffactorsintheexpressionfortheimpactparameter.............4
Fig.2.2:Headoncollisionbetweentwoparticles(reflexiveseparation)..................5
Fig.2.3:Collisionwithpartialregionofinteraction(stretchingseparation). ..........5
Fig.2.4:Diagramofcollisionregimes(Ko&Ryou,2005).................................................6
Fig.2.5:Regionsforcoalescenceandseparations(Ashgriz&Poo,1990). .................7
Fig.2.6:Basictypesofdropletdeformation(Hinze,1955). .............................................8
Fig.2.7:ScrollpatternoftheKelvinHelmholtzinstability(Dhainaut,2002)..........9
Fig.2.8:ScrollpatternoftheRayleighTaylorinstability(Dhainaut,2002). ............9
Fig.3.1:AcontrolvolumearoundnodeP.............................................................................. 13
Fig.3.2:Controlvolumewithdiscretizationparameters(FluentInc.,2006)........ 17
Fig.4.1:Boundariesbetweencollisionregimes(Ko&Ryou,2005).......................... 25
Fig.4.2:Mapofflowregimes....................................................................................................... 28
Fig.5.1:Swirlelement.................................................................................................................... 29
Fig.5.2:IllustrationofthecrosssectionalareaoftheCTC............................................ 31
Fig.5.3:GridcreatedfortheCTC............................................................................................... 31
Fig.6.1:Visualizationofhowtheresultvarieswithcellsizes...................................... 36
Fig.6.2:Convergedresidualsforstationarysolutionofthewaterphase. .............. 37
Fig.6.3:Residualsforsolutionofwaterphaseafterintroductionofdroplets...... 37
Fig.6.4:SimplifiedpictureofthepartsofthecrosssectionalareaoftheCTC. .... 39
Fig.6.5:Increaseofmeandropletdiameter......................................................................... 40
Fig.6.6:Migrationofparticlesintothemiddleofthetube............................................ 40
Fig.6.7:AverageWebernumberforonlycoalescencemodeling. .............................. 41
Fig.6.8:SMDwhenbreakupmodelingisincluded. ......................................................... 42
Fig.6.9:AverageWebernumberforbothcoalescenceandbreakupmodeling.. 43
Fig.6.10:IncreaseofdiameterthroughouttheCTC. ........................................................ 43
Fig.6.11:VisualizationofdropletsandtheirdiameterinsidetheCTC. ................... 44
Fig.6.12:Increaseofmeandropletdiameterfordifferentvolumefractions........ 44
Fig.C1:ContoursofDPMconcentration,inletdiameterof20m(kg/m3)55
Fig.C2:ContoursofDPMconcentration,inletdiameterof60m(kg/m3)....55
Fig.C3:ContoursofDPMconcentration,inletdiameterof100m(kg/m3).56
vii
ListofTables
Table5.1:Overviewofthevariableparametersintheinvestigations. .................... 32
Table6.1:Comparisonbetweensixmesheswithdifferentcellsizes. ...................... 36
Table6.2:Massflowbalanceratioforallsimulations..................................................... 38
Table6.3:Resultsofrunningthetestcaseuntil6seconds. .......................................... 38
Table6.4:Performanceparametersfordifferenttypesofdropletliquid............... 45
TableA.1:Presentationofallsimulations......49
TableA.2:Materialpropertiesforthedifferenttypesofdropletliquids.49
TableB.1:SettingsforthesimulationsofthewaterphaseinFLUENT....51
TableB.2:SettingsforthesolutioncontrolsinFLUENT...52
TableB.3:SettingsusedintheDPMinFLUENT......................................52
TableB.4:SettingsfortheinjectionusedintheDPMinFLUENT....53
viii
Nomenclature
GreekSymbols
Thermaldiffusivity
p
Underrelaxationfactorforpressure
Diffusioncoefficient
Dropsizeratio
Diameterratio/arithmeticdifference
Dynamicviscosity/micro,106
Density
Surfacetension
l,s
Collisioncrosssection
Stresstensor
Generalflowvariable
LatinSymbols
Acceleration
c
Speedofsound
A
Area/amplitude
Areaoffacef
A
f
C
Generalcoefficient
cv
Heatcapacity
c0
Indicationofcell0
c1
Indicationofcell1
d
Diameter
d32
SauterMeanDiameter(SMD)
F
Force
g
Gravitationalacceleration
Jf
Massfluxthroughfacef
k
Thermalconductivity
m
Mass
n
Numberofdroplets
NVi
Viscositynumberforshearflows
NWe
Webernumberforshearflows
p
Pressure
Pl,s
Probabilityofcollisionbetweendroplets
Sourceterm
t
Time
u
Velocityinxdirection
urel
Relativevelocitybetweendroplets

2
v
v
Vcol
x
X
ix
Averagedsquareoftherelativevelocitybetweendropletandfluid
Velocityvector
Collisionvolume
Impactparameter
Distancefromonedroplettotherelativevelocityvectorplacedon
thecenterofanotherdroplet
Abbreviations
CFD
ComputationalFluidDynamics
CTC
CompactTubularCoalescer
DPM
DiscretePhaseModel
FVM
FiniteVolumeMethod
Ma
Machnumber
PDE
PartialDifferentialEquation
RANS
ReynoldsAveragedNavierStokes
Re
Reynoldsnumber
RSM
ReynoldsStressEquationModel
SMD
SauterMeanDiameter
We
Webernumberforparticles
Indices
Continuousphase
crit
Critical
d
Dispersedphase
D
Drag
l
Large
p
Particle
rel
Relative
s
Small
Taylor
Taylorsanalogy
1 Introduction
1.1 Motivation
The aim of oil and gas processing facilities is to process wellhead fluids into
differentproducts.Theseproductsaretransportedtothecustomerbye.g.ships,
trailers or in pipelines. This processing of oil and/or gas for transportation or
storage is called oilfield processing. But some of the processed fluids are not
transported and sold; reinjection and disposal of fluids and gasses are usual.
Therefore, oilfield processing also includes water treatment, which means
cleaningofwaterthatistobeusedfordisposalorreinjection.Whenitcomesto
disposal, environmental issues must be considered. This puts demands on the
purenessofthewaterinordernottopollutethesea.Thiswatercontainssmall
oil or other hydrocarbon droplets from the well stream, which need to be
removed,thusseparationequipmentisneeded.
For this purpose hydrocyclones and flash drums are often used as separation
equipments. These separators have better performance with larger inlet size
distribution of the oil droplets, which are to be separated out. In order to
increase the diameter of the droplets that enter the separation device, a
coalescer could be used. Coalescers are widely used in the oil and gas. A
coalescer is a device where droplets of the dispersed phase are coalescing into
larger drops that are more amenable to removal. Aker Process Systems AS
suggested the topic of the present work; they are developing a product called
Compact Tubular Coalescer (CTCTM). The principle of the CTC is to force the
droplets into a turbulence and centrifugal flow pattern by setting the fluid in
rotationinatube.Thecentrifugalforceswillforcethewatertotheouterwalls
whilethelighterhydrocarbonwillremaininthemiddleofthetubeandcoalesce.
Mathematicalmodelsofthedeviceareimportantinordertomakethedesignas
efficient as possible. This work focuses on coalescence of oil droplets in water.
TheCTCisusedasabasisforthemodelingofthecoalescencephenomenon,and
differentaspectsoftheperformanceoftheCTCisstudied.
1.2 IntroductiontoDispersedTwoPhaseFlows
Dispersed flow occurs when one fluid is in the form of droplets within the
continuum of the other fluid. There are numerous applications to this kind of
flows, particularly in the separation part of the process industry where oil and
waterareexploitedtogetherandthenneedtobeseparatedfromeachother.In
turbulentdispersedflows,thedropletscaninteractwiththeturbulenteddiesof
the continuous phase in many different ways depending on the sizes of the
dropletsandeddies.Thisinteractioncaninfluenceandchangethestructureand
parameters of the turbulent flow. Thus, knowledge about the droplet size
distributionintwoliquidmixtureisnecessaryinordertobeabletopredictthe
final conditions in separation devices like the CTC and it will lead to better
modelinganddesignofseparationdevicescontainingdispersedflows.
1Introduction
In turbulent dispersions, both breakup and coalescence usually take place

continuously, and these processes will determine the final droplet size
distribution.
1.3 AimofthePresentWork
The main purpose of this thesis is to describe and discuss the theoretical
backgroundforliquidliquidcoalescence,withspecialfocusoncoalescenceofoil
dropletsinacontinuumwaterphase,andthenformulateamathematicalmodel
ofacasethatincludessuchcoalescence.Asmentionedabove,theperformanceof
theCTCisusedasthecaseofstudyinordertoseehowcoalescenceaffectsthe
separation performance, and which advantages the design of the CTC has. A
commercial Computational Fluid Dynamics (CFD) code, FLUENT 6.3.26, is used
tomakesimulationsofthemodel,andGAMBIT2.4isusedtomakegeometries
andgrids.Theresultsareevaluatedanddiscussedinordertofindouthowgood
themodelisandwhichfurtherimprovementsthatshouldbemade.
1.4 StructureoftheReport
Thereportstartswithaphysicaldescriptionofthephenomenonofliquidliquid
coalescence in dispersed twophase turbulent flows. It begins with a general
descriptionofwhatcoalescenceisandhowitoccurs,beforemoredetailsabout
collision between particles and the outcomes of the collisions are discussed
(chapter2).Adescriptionofthecloselyrelatedbreakupphenomenonisgivenin
thischapteraswell.ThisisfollowedbyacloserlookonCFDingeneral,andhow
regular continuous phase flows are treated in FLUENT. The numerical method
used by FLUENT, the finite volume method (FVM), is described (chapter 3).
Further,themodelingofparticlecollision,coalescenceandbreakuparetreated
and given a detailed description, and this is linked to how the CFD software,
FLUENT,usesthismodeltopredictcollision,coalescenceandbreakup(chapter
4). A chapter with computational methodology and description of the
simulations is given before results are presented and discussed. Some
recommendations about further work are also given. Appendices and list of
referencesaregivenintheend.
2 PhysicalTheory
2.1 CoalescenceasaPhenomenon
Any multiphase process involves a multitude of interactions, and these
interactionsmayinvolvecollisionsbetweendropletsandparticles.Coalescence
occurswhentwoormoresuchparticlescollideandareincontactlongenough.
The collision is caused by spatial velocity differences between particles. The
velocity differences may be caused by several factors. Prince & Blanch (1990)
mentionthreemechanismsthatcanleadtocollision:
Turbulence:Thelengthscaleoftheturbulenteddiesthatcausesrelative
velocity differences between particles must be in the order of the
diameteroftheparticle.Iftheeddiesaretoosmall,theywillnotcontain
enough energy to affect the particle motion. On the other hand, if the
eddiesaremuchlargerthanthesizeoftheparticle,ensemblesofparticles
are transported together and do not affect the relative motion between
them.Dropletcollisionscausedbyvelocityfluctuationsaresimilartothe
randommovementofgasmoleculesinthekineticgastheory.
Buoyancy: Collision is caused by difference in the rise velocities of the
particles with different size. Larger droplets can catch smaller ones on
theirway.
Laminarshear:Collisionoccursasaresultoflargecirculationpatternsin
e.g.atube.
Not every collision leads necessarily to coalescence. Several models exist for
describing when collision leads to coalescence and when it does not. For
coalescence to happen, the particles have to remain in contact with each other
forlongenoughtimefortheliquidfilmbetweenthemtodraintothecriticalsize
necessarytobreakit.Thisisacommonrequirementforallmodels.Thismeans
thatalsothefilmsthinningrateisaparameterforoccurrenceofcoalescence.
2.2 CollisionbetweenParticles
Dhainaut (2002) mentions the other possible outcomes from a collision;

bouncing, separation and shattering. When the outcome is bouncing, the
interveningfilmonthecollidingparticlessurfacespreventscoalescenceandthe
particles bounce apart. In this casethe drops maygothrough deformation, but
there is no mass exchange. Separation collision occurs when two particles
connect temporarily and separate immediately after into two or more drops.
Shatteringcollisionoccurswhenparticleswithhighrelativevelocitycollideand
breaksintoseveralsmallerparticles.
Ashgriz&Poo(1990)claimthatthemainparametersthatcontroltheoutcome
of a collision are the Weber number, the colliding particles diameter ratio, the
Reynoldsnumberandtheimpactparameter,aspresentedinEq.(2.1)(2.4).
2PhysicalTheory
Webernumber:
Diameterratio:
Impactdiameter:
Reynoldsnumber:
2
d surel
We =
(2.1)
d
= s
dl
(2.2)
2X
x=
dl + d s
(2.3)
dlurel
(2.4)
Re =
Us
ds
Ur=UsUl
Ul
dl
Fig.2.1:Descriptionoffactorsintheexpressionfortheimpactparameter.
ThefactorsintheimpactparameterfortwocollidingparticlesareshowninFig.
2.1.Xisthedistancefromthecenterofonedroptotherelativevelocityvector
placedonthecenteroftheotherdrop,whiledlanddsarethediametersofthe
largeandthesmalldrop,respectively.
Ashgriz & Poo (1990) presented two possible types of separation collisions:
reflexive separation and stretching separation. In reflexive separation a near
headon collision occurs between the two colliding particles (Fig. 2.2). In this
separation the liquid inside the temporarily coalesced drop experiences a
pressure difference between the two extremities and the center of the drop,
which pushes the liquid from the center and then the drop separates into two
newdrops.Whentwodropsofnonequalsizecollidetheinitiallargedroploses
2PhysicalTheory
somemasstothesmalldropsuchthatthesmalldropbecomesthelargestafter
thecollision.
In stretching separation only a part of each drop is in direct contact with each
other.Therestofthedropstendtocontinuetoflowintheirinitialdirection(Fig.
2.3). As a consequence, the region of interaction between the two drops is
stretched. Thus, there are two competing forces involved in the collision: the
surfacetensionintheregionofinteraction,whichisholdingthedropstogether
(draining), and the kinetic energy of the initial drops, which is stretching the
dropsandseparatingthem.Fordropsofdifferentsizesthetwocompetingforces
willdeterminethesizeofthedropsafterthecollision.Thestretchingeffectwill
cause mass transfer from the small drop to the large drop, as for the reflexive
separation, but the drainage effect will make the small drop scoop out some of
the mass of the large drop because of lower internal pressure inside the large
dropcomparedtothesmalldrop.
Fig.2.2:Headoncollisionbetweentwoparticles(reflexiveseparation).
Fig.2.3:Collisionwithpartialregionofinteraction(stretchingseparation).
In both cases of separation, the outcome may usually result in several satellite
drops. That means that the two colliding particles will produce two new
particles,butalsoseveralsmallerparticleswillarise(Fig.2.4)
Fig.2.4:Diagramofcollisionregimes(Ko&Ryou,2005).
2PhysicalTheory
By relating the impact parameter to the two types of separation collisions, it

appears that headon collisions correspond to an impact parameter equal to
zero,whileforobliquecollisionstheimpactparameterequalsone.
Coalescence occurs for impact values between the ones for stretching and
reflexive collisions. Analytical results based on experiments for when
coalescence occurs are presented in Fig. 2.5 as a function of impact parameter
andWebernumber.
2PhysicalTheory
Fig.2.5:Regionsforcoalescenceandseparations(Ashgriz&Poo,1990).
Asonecanseefromtheresults,theprobabilityforcoalescenceafteracollisionis
largestforbothlowWebernumbersandforlowvaluesfortheimpact
parameter.
AcloserstudyontheeffectsofReynoldsnumber,impactvelocity,dropsizeratio
andinternalcirculationonthecollisionandcoalescenceprocessisdoneby
Mashayek,Ashgriz,Minkowycz,&Shotorban(2003).
2.3 BreakupasaPhenomenon
As for collision between particles, breakup occurs in all kinds of multiphase
flows.Breakupofbubblesanddropsisgenerallycausedbyoneofthefollowing
mechanisms(Dhainaut,2002):
Turbulentfluctuationsandcollisions
RayleighTaylorandKelvinHelmholtzinstabilities(whichincludesrapid
acceleration)
Highshearstresses
Nonuniformityinsurfactantdistribution
All these mechanisms may lead to local shear forces of the surrounding fluid
around the droplet that are greater than the cohesive forces, which is the
criterionforbreakuptooccur.
Droplets can split up in a number of different ways that depend on the flow
pattern around them. According to Hinze (1955) there are three basic types of
deformation that can lead to breakup: lenticular, cigarshaped and bulgy.
Lenticular deformation is recognized when the droplet is flattened, forming an
oblate ellipsoid before it is further deformed into a torus, which breaks into
smaller droplets when it is being stretched. Cigarshaped deformation happens
whenthedropletismoreandmoreelongatedformingaprolateellipsoidthatis
further deformed into a long cylindrical thread, which breaks into smaller
droplets. Bulgy deformation happens when the surface of the droplet is
2PhysicalTheory
deformed locally so bulges occur and parts of the droplet become bodily
separated.ThethreetypesofdeformationareshowninFig.2.6.
Fig.2.6:Basictypesofdropletdeformation(Hinze,1955).
Also for breakup the dimensionless Weber number is used to characterize the
criterion for breakup. For shear flows this dimensionless Weber number is
givenby:
dv 2
(2.5)
NWe = c
2
v istheaverageofthesquareoftherelativevelocitybetweentheparticleand
the surrounding continuous phase across the flow field. d is the particle
diameter,cisthedensityofthecontinuousphaseandisthesurfacetensionof
thesystem.AccordingtoHinze(1955)thecriticalvalueofthisnumberdepends
on both the viscosity number of the dispersed phase and on how the relative
velocityvarieswithtime.Thedimensionlessviscositynumberthatcharacterizes
theviscosityofthedispersedphaseisgivenby:
d
NVi =
(2.6)
d
d
Here, d is the viscosity of the dispersed phase. Laboratory experiments and

theoreticalconsiderationsbyHinze(1955)showthatthecriticalWebernumber
for breakup to happen, (NWe)crit, increases with increasing NVi. Indefinitely
increaseofNVileadsto(NWe)crit,i.e.nobreakup.
For turbulent flows turbulent eddy velocities locally govern the breakup
phenomenon. These dynamic pressure forces are caused by changes in the
velocity over distances typically of the same order as the particle diameter, i.e.
much smaller than the viscous shearing action includes. Breakup occurs when
the local shear stress generated by such eddies is larger than surface tension
force. The local turbulent shear stress is defined by 1/2c u , where u 2 is the
average value of the velocity fluctuations squared. This average value can be
related to the particle diameter in order to obtain the critical value, dcrit, above
which breakup occurs. When using this method, Hinze (1955) found the
followingrelationbetweendcritandtheturbulenceenergydissipationrate,:
2
2PhysicalTheory
dcrit
5 2
5

(2.7)
Onlyeddiesofthesamesizeastheparticlecanleadtobreakup.Smallereddies
donotcontainenoughenergytobreaktheparticle,whilelargereddiesjustmove
the particle, but are not breaking it up. When breakup occurs, particles may
breakupintoawiderangeofsizes.
In the case of hydrodynamic instabilities at the surface of the particle, it may

breakup as a consequence of the increasing scroll pattern on the surface. This
scrollpatterncanexistasoneofthetwofamoushydrodynamicinstabilities:the
KelvinHelmholtzandtheRayleighTaylorinstability(Fig.2.7andFig.2.8).The
KelvinHelmholtz instability occurs when two parallel streams of different
velocitiesthatareadjacenttoeachotherbecomeunstableduetoperturbations.
RayleighTaylorinstabilityoccurswhentwofluidsofdifferentdensitiesinteract
suchthatadense,heavyfluidacceleratesalightfluid.
Fig.2.7:ScrollpatternoftheKelvinHelmholtzinstability(Dhainaut,2002).
Fig.2.8:ScrollpatternoftheRayleighTaylorinstability(Dhainaut,2002).
10
2PhysicalTheory
11
3 CFDModelingoftheContinuousPhase
3.1 IntroductiontoCFD
The purpose of Computational Fluid Dynamics (CFD) is to analyse systems
involving fluid dynamics, heat transfer and other associated phenomena by
numerical calculations. The areas of application are large and cover both
industrialandnonindustrialapplications.
Generally,aflowcanbedescribedbysolvingthethreeconservationequations:
Conservationofmass
Conservationofmomentum
Conservationofenergy
For incompressible flows, the equations for conservation of mass and

momentumarereferredtoastheNavierStokesequations.Theseequationsare
partial differential equations (PDEs) and thus difficult to solve analytically. A
discretization method that approximates the PDEs with a system of algebraic
equations is applied and solved numerically on a computer. The algebraic
equations are solved for small domains in space and time. The numerical
solutionoftheflowthenconsistsofthesolutioninthesediscretelocations.The
accuracy of the solution is then dependent on the quality of the discretization
methodused.
The development of CFD codes has given many advantages in engineering

analysis.Itisaveryusefultooltosimulateproblemswhereexperimentsarevery
costly.ButitshouldberemarkedthatCFDsimulationsareanapproximationof
realexperimentsandshouldnotbeconsideredasasubstitutionforexperiments,
butratheracomplementary.
ItexistsseveralCFDcodes;mostofthemarecommercial.AmongtheseareCFX,
COMSOLandFLUENT.FLUENTisusedinthepresentstudy.
3.2 GoverningEquationsfortheContinuousPhase
Thepresentworkconsidersparticledispersionandcoalescenceinafluidflow.
Thischapterwillstatetheequationsmentionedinthepreviouschapterusedfor
describingthefluidcontinuousphaseflow.Thisstudyconsiderswaterflowwith
oildropletsinit,whichmeansthatonlyincompressibleflowisconsidered.For
incompressibleflowsthedensityisconstantandnotlinkedtothepressure.The
massconservationisaconstraintonthevelocityfield;thisequation(combined
withthemomentum)canbeusedtoderiveanequationforthepressure.
3.2.1 MassConservationEquation(ContinuityEquation)
Thegeneralformofthecontinuityequationisgivenby:
+ v = Sm
t
( )
(3.1)
12
3CFDModelingoftheContinuousPhase
Sm is the mass added to the continuous phase from an eventually dispersed

second phase. In the present study this second phase is the liquid oil in water
phase.Thecontinuityequationcan,forincompressibleflows,besimplifiedto:
(3.2)
v= Sm
3.2.2 MomentumConservationEquation
Thegeneralformofthemomentumconservationisgivenby:

v + vv = p + + g+ F
t
Fisexternalbodyforceswhile isthestresstensor:
( ) ( )
()

T 2
= v+ v vI
3
( )
(3.3)
(3.4)
3.2.3 EnergyConservationEquation
The general form of the energy conservation equation for incompressible flow
canbegivenas:
T
1
+ Tv = T +
v
(3.5)
cv
t
k
Here, =
is the thermal diffusivity. The energy equation (Eq. (3.5)) for
cv
incompressibleflowsisdecoupledfromtheNavierStokesequations(Eq.(3.2)
(3.3)).ThismeansthatEq.(3.2)(3.3)aresolvedfirstfor vandpandthenEq.
(3.5)forT.
( )
3.3 FiniteVolumeMethod
FLUENT uses the FiniteVolume Method (FVM) to discretize and solve the
governing equations for the continuous phase. This method consists of three
steps:
Integrationofthegoverningequationsoverallthecontrolvolumesofthe
domain.
Discretization and conversion of the resulting integral equations into a
systemofalgebraicequations.
Solutionofthealgebraicequationsbyaniterativemethod.
In order to get a picture of how the finitevolume method works, a short

description of the three steps in the method is given with a general transport
equationasanexample,cf.(Versteeg&Malalasekera,1995).
Theconservationofageneralflowvariable,e.g.avelocitycomponent,withina
finite control volume can be expressed as a balance between the various
processestendingtoincreaseordecreaseit:
13
Rateofchange
Netrateof
Netrateof
Netrateof
ofinthe
decreaseof
increaseof creationof
controlvolume = duetoconvection + duetodiffusion + insidethe
withrespectto
intothe
intothe
control
time
controlvolume controlvolume volume
Thecorrespondingequationtothisexpressionisgivenby:
(3.6)
= v + + S
t
is a diffusion coefficient and S is a source term. For simplicity the following

presentation of the FVM for the general transport equation will be treated as
onedimensionalandwithoutsourceterms(Eq.(3.7)).
d
d d
=
u +
(3.7)
t
dx
dx
dx
Thisequationcanbefurthersimplifiedbyassumingsteadystateconditions:
d
d d
u =
(3.8)
dx
dx
dx
( )
( )
( )
( )
( )
3.3.1 SpatialDiscretization
As mentioned, the principle of the FVM is to discretize the integral form of the
governing equations. The integral form of the steady state general transport
equationwithoutsourcetermsforaninterval[w,e]inonedimensionisgivenin
Eq.(3.9).AsketchoftheactualnodeanditscontrolvolumeisgiveninFig.3.1.
x
x
2
2
W
w
P
e
E
Fig.3.1:AcontrolvolumearoundnodeP.
(uA ) (uA )
d d
= A A
w
dx e dx w
(3.9)
14
Aistheareaforthecontrolvolumefaces;itisassumedtobeequalforallfaces.
In order to obtain discretized equations for the problem, the terms inEq. (3.9)
mustbeapproximated.ByintroducingtwonewvariablesFandD,asgiveninEq.
(3.10),theapproximatedtransportequationcanbewrittenasinEq.(3.11).
F = u , D =
x
2
(3.10)
Fw = u , Dw = w
w
x
2
Fe = u , De = e
e
x
( )
( )
( )
F F = De E P Dw P W
e e w w
(3.11)
e = p and w = W
(3.12)
The next step is to approximate the face values e and w. There exist several
methods for spatial discretization. (Fluent Inc., 2006) presents a short
descriptionofsomeoftheschemesforthisdiscretization:
FirstOrderUpwindScheme:
Upwinding means that the cellface value of the current cell is derived
from the quantities of the cell upstream or upwind relative to the
direction of the normal velocity. For the firstorder upwind method the
cellcentervalueofacellisassumedtorepresentacellaveragevaluefor
the cell. Thus, the face value of the current cell is set equal to the cell
centervalueoftheupstreamcell:
SecondOrderUpwindScheme:
Secondorder accuracy is achieved at cell faces through a Taylor series
expansion of the cellcentered solution about the cell centroid. Thus the
facevalueofthecurrentcell,,iscomputedby:
e = p + rand w = W + r
(3.13)
p
W
( )
( )
andarethecellcenteredvalueanditsgradientintheupstreamcell,
whileristhedisplacementvectorfromtheupstreamcellcentroidtothe
facecentroid.
Generally the accuracy will be improved by choosing a higherorder scheme
because the firstorder scheme has a numerical diffusivity because of its
numerical dissipation term (Mller, 2007). But as the order of the scheme is
increased,convergenceismoredifficulttoachieve.Whentheflowisalignedwith
the grid, the firstorder upwind scheme may be acceptable. In other words, for
triangularandtetrahedralgridsitisgenerallymoreaccurateresultswithhigher
orderschemes,sincetheflowisneveralignedwiththegrid.Butforquadrilateral
schemestheresultsmaybeacceptable.
15
3.3.2 TemporalDiscretization
For transient problems, like the general transport equation problem, the
governingequationsmustbediscretizednotjustinspace,butalsointime.Every
termintheequationsmustbeintegratedoveratimestept,fromtimelevelnto
n+1.BywritingthetransportequationasEq.(3.14),thetimeintegrationcanbe
donelikeinordinarydifferentialequations,asshowninEq.(3.15).

d
d d
(3.14)
=
u + = f t, t
t
dx
dx
dx

t
t
(3.15)
t t dt = t f t, t dt
As for the spatial discretization there exist several schemes for temporal
discretization. (Fluent Inc., 2006) gives a short description of some of these
schemes(incompressibleflowisassumed,=const.):
FirstOrderImplicitScheme:
Intheimplicitmethodthefunctionf(t,(t))isevaluatedatthefuturetime
level.Itiscalledimplicitbecause n+1inagivencellisrelatedto n+1in
theneighboringcellthroughf(t,(t)).Thismeansthattheequationmust
be solved iteratively at each time level before moving to the next time
step.TheresultingFVMforthetransportequationcanthenbewrittenas:
( )
( ( ))
( )
n+1
n+1
n+1 n
t
3 n+1 4 n + n+1
2t
n+1
(3.16)
, n+1
SecondOrderImplicitScheme:
Second order accuracy is achieved by choosing a more accurate finite
difference stencil for the approximation of the temporal derivative. The
resultingFVMthenbecomes:
) = f (t
( ( ))
) = f (t
n+1
, n+1
(3.17)
ExplicitScheme:
The explicit method evaluates f(t,(t)) at the current time level. It is
referredtoasexplicitbecause n+1canbeexpressedexplicitlyinterms
of known values, n. The explicit method provided in FLUENT is first
orderaccurate.TheresultingFVMforthismethodisgivenby:
n+1 n
) = f (t , )
n
(3.18)
16
3.3.3 AssemblyoftheDiscretizedEquations
Whenthegoverningequationsarediscretizedwithoneoftheschemesdescribed
above, the discretized equations must be assembled to a matrix system of
algebraic equations. This matrix system can be solved by several different
techniques;however,themostpopularsolutionprocedureisbytheTDMA(tri
diagonalmatrixalgorithm)linebylinesolver(Versteeg&Malalasekera,1995).
Equation(3.19)and(3.20)showsthealgebraicequationfortheonedimensional
general transport equation without source term at each control volume, while
Eq. (3.21) shows the form of the resulting matrix system that is obtained by
combiningtheseequationsforallcontrolvolumes.
0 0
(3.19)
aW W + aP P + aE E = aP P
aP P + anbnb = a0P P0
nb
(3.20)
[ A][ ] = [B]
(3.21)
In Eq. (3.21) the term [B] contains the initial values, aP0P0, and boundary
conditions. See (Versteeg & Malalasekera, 1995) for details about how the
coefficients in the algebraic equations vary when different discretization
schemes are used, and how boundary conditions enter the discretization and
assemblyofthegoverningequations.
3.4 AlgorithmforPressureBasedCalculations
The present study does not include the energy equation in the calculations
because no heat transfer in assumed (c.f. Chapter 5.2). Thus, some special
practices related to the discretization of the continuity and momentum
equations when the solver is pressurebased (segregated) are discussed in this
chapter.
The momentum equations and the continuity equation are coupled because
everyvelocitycomponentappearsinallequations.Butthemostcomplexissue
to resolve is the pressure field since it appears in all momentum equations.
There is no equation for that resolves the pressure. If the pressure gradient is
known, the discretized equations for velocity are obtained in exactly the same
mannerasforanyotherscalar.Butformostflowcomputations,alsotheonein
this study, it is desirable to calculate the pressure field as part of the solution,
hence the pressure gradient is normally not known beforehand. If the flow is
incompressible,asitisassumedtobeinthisstudy,thedensityisconstantand
donotdependonthepressure.Thecouplingbetweenthepressureandvelocity
thus implies that if the correct pressure field is applied in the momentum
equationstheresultingvelocityfieldshouldsatisfycontinuity.
The schemes presented for discretization of the general transport equations in

Chapter3.3isalsousedtodiscretizethemomentumequationsinthepressure
based solution algorithm. The steady state xmomentum equation without
sourcetermscanbeobtainedbysetting=u:
17
aPu = anbunb + p f Afx

nb
(3.22)
Equation (3.22) requires the value of the pressure at the face between cells c0
and c1, as shown in Fig. 3.2. Thus, an interpolation scheme is necessary to
calculate this value. A standard scheme is chosen in the present study. See
(Fluent Inc., 2006) for details about this scheme. FLUENT uses a colocated
schemewherepressureandvelocityarestoredatcellcenters.
Fig.3.2:Controlvolumewithdiscretizationparameters(FluentInc.,2006).
By integrating the continuity equation over the control volume in Fig. 3.2 the
followingdiscreteequationisobtained:
N faces
Af = 0
(3.23)
Jfisthemassfluxthroughfacef.
The problem with pressurevelocity linkage can be resolved by adopting an

iterative solution strategy. The strategy used in this study, and also one of the
most common strategies, is the SIMPLE (SemiImplicit Method for Pressure
Linked Equations) algorithm. The principle of the algorithm is that a guessed
pressurefield,p*,isusedtosolvethediscretizedmomentumequationsandthen
enforcethemassconservationandobtainthepressurefield.Asthenextstep,the
correctionsofthepressureandmassflux,pandJf,aredefinedasthedifference
betweenthecorrectandguessedpressureandmassflux,respectively:
p = p * +p
(3.24)
Jf = Jf * + Jf
TheSIMPLEalgorithmdoesthefollowingapproximation:
J = d f pc0 pc1
f
(3.25)
18
Here, d f =
Af
aP
Adiscreteequationforthepressurecorrection,p,inthecellisthenobtainedby
substituting the flux correction equation into the discrete continuity equation,
Eq.(3.23):
ap p = anb pnb
nb
(3.26)
p = p * + pp
(3.27)
Oncethesolutionforthecorrectionpressureiscalculated,thecellpressureand
thefacefluxarecorrectedusingthefollowingrelations:
J = J f * +d f pc0 pc1
f
(3.28)
Here, p is the underrelaxation factor for pressure. See (Versteeg &

Malalasekera,1995)fordetailsabouttheuseofunderrelaxation.Thisiterative
processisrepeateduntilconvergenceofthecorrectedvalues.
3.5 TurbulenceModeling
All flows encountered in engineering practice become unstable above a certain
Reynolds number. Turbulence is caused by the development of a chaotic and
randomstateofmotioninwhichthevelocityandpressurechangecontinuously
with time within regions of flow. Most flows of engineering significance are
turbulent, so the turbulence flow regime is of great interest in the engineering
industry.Thepresentstudydoesalsocontainturbulentflows,henceturbulence
modelingisthusneeded.
OneofthemostdifficulthurdlestotheproperuseofCFDcodesistheturbulence
modeling. In a turbulent flow, particles of fluid move unsteadily in an
unpredictable path. In most CFD codes different turbulence models are
proposed.Turbulentflowsarecharacterizedbyfluctuatingvelocityfields.These
fluctuations can be of small scale and high frequency and then become too
computationaltimeconsumingwhensimulatedinengineeringcalculations.
Turbulence modeling is based on the principle that the velocity can be

decomposed into a steady mean value and a fluctuating component of the
velocity. This is called the Reynolds decomposition. The turbulent flow is then
characterizedintermsofthemeanvaluesofflowpropertiesandsomestatistical
properties of their fluctuations. By implementing this characterization into the
momentumequationsfortheflowtheReynoldsAveragedNavierStokes(RANS)
equationscanbederived.Inordertobeabletocomputeturbulentflowswiththe
RANS equations it is necessary to develop turbulence models to predict the
Reynoldsstressesandthescalartransporttermsandclosethesystemofmean
flowequations.Thenumberofadditionaltransportequationsneededalongwith
19
the RANS equations in order to close the system classifies the most common
RANSturbulencemodels.
The different turbulence models have different pros and cons. Since the
ReynoldsStress Equation Model (RSM) accounts for the effects of streamline
curvature, swirl, rotation, and rapid changes in strain rate in a more rigorous
mannerthanoneequationandtwoequationmodels,ithasgreaterpotentialto
give accurate predictions for complex flows, such as the rotating low in the
presentstudy.Thus,thismodelischosenastheturbulencemodel,eventhoughit
isasevenequationmodelforthreedimensionalproblems(fiveequationmodel
fortwodimensionalproblems)andneedsmorecomputationalpowerandtime
tobesolved.
3.5.1 TransportEquationsfortheReynoldsStressEquationModel
Takingmomentsoftheexactmomentumequationderivestheexactformofthe
Reynolds stress transport equations. This is a process wherein the exact
momentumequationsaremultipliedbyafluctuatingproperty,theproductthen
being Reynoldsaveraged. Unfortunately, several of the terms in the exact
equationareunknownandmodelingassumptionsarerequiredinordertoclose
theequations.
The exact equation for the transport of kinematic Reynolds stress, Rij = ui u j ,
takestheformgiveninEq.(3.29)(Versteeg&Malalasekera,1995).
Rij
+ Cij = Pij + Dij ij +
ij
+ ij
(3.29)
Convection
Production
Diffusion Dissipation
Pressurestrain
Rotation
Rateofchange
The six terms in Eq. (3.29) are modeled by six different equations for the
Reynolds stress transport. Along with these, a model equation for the scalar
dissipation rate, , is solved. This makes the ReynoldsStress Equation Model
containsevenequations,whichhavetobesolvedalongwiththeexactequation
fortransportofkinematicReynoldsstress.Moredetailsaboutthedevelopment
anduseofthismodelcanbefoundin(Versteeg&Malalasekera,1995).
3.6 GeneralProcedureforUseofCFDcodes
CFDcodesareallstructuredaroundthenumericalalgorithmthatwillsolvethe
fluidflowproblems.TheCFDcodeconsistsofthreefundamentalelements:
Preprocessor
Solver
Postprocessor
3.6.1 Preprocessor
Thetaskofthepreprocessor,i.e.GAMBIT,istogathertheessentialinformation,
theinputs,neededtosolvetheflowproblem.Thisinvolvesthefollowingsteps:
Definition of the geometry, i.e. the computational domain, by creating
volumes,surfaces,edgesandpoints.
20
Generationofthegrid,i.e.subdivisionofthecomputationaldomaininto
smallersubdomains.Meshescanbebothtriangularandquadrilateral.
Specificationofboundariesandcontinuumzonetypes.
3.6.2 Solver
Once the mesh is created it is saved as a binary file, which can be read by the
solver.Whenthefileisreadinthesolver,thefirstthingtodoistomakesurethat
allthevolumescreatedarepositive.Theuserdefinesmodelsforsolver,energy,
viscosity,turbulenceandradiation.Initialandboundaryconditionsneedtobe
specified, as well as properties of the different fluids present in the system.
Different schemes for both temporal and spatial discretization are available as
described in Chapter 3.3. The user selects the desired level of convergence.
Beforetheiterationprocesscanstart,itisnecessarytoinitializetheentireflow
field, such as ambient pressure, velocity and temperature. By using different
zones in the flow field it is possible to make different initial conditions for
differentpartsoftheflowfield.Theflowcalculationscanbebasedondifferent
numerical solution techniques, but the finite volume method is the most
common technique when it comes to fluid flow calculations. The finitevolume
techniqueisusedintheCFDcodeFLUENTusedinthepresentstudy.Tocheck
the convergence of the solution FLUENT uses residuals. The residual for
continuity,velocitycomponents,andvariablesrelatedtothechosenturbulence
model is calculated in each iteration step. The residual for each variable is the
differencebetweenthevaluecalculatedatiterationN1anditerationNdivided
byitsphysicaltimestep.Theconvergencecriterioncanbedecreasedtoobtain
moreaccurateresultsbutitalsoincreasesdramaticallytheCPUtime.
3.6.3 Postprocessor
Once the solution has converged, the postprocessor provides plotting and
animationofcontoursandvectors,trackingofparticlesintheflowandseveral
othervisualizationtools.MostCFDpackages,suchasFLUENT,areequippedwith
suchtools.
21
4 DispersedPhaseModeling:CoalescenceandBreakup
4.1 EulerianandLagrangianFramework
Whenmultiphaseflowproblemsaretobesolved,therearemainlytwodifferent
classesofapproachesthatcouldbeusedformodelingofthedispersedphase:the
Eulerian and the Lagrangian modeling. In the Eulerian approach the dispersed
phaseistreatedasacontinuumthatinteractswiththecontinuousphase,while
intheLangrangianapproachthetrajectoryofeachparcelindispersedphaseis
found by solving the Lagrangian equations of mass and momentum. Both the
coalescence and breakup phenomenon can be modeled with both of the two
aboveapproaches.Thereareseveralpossibilitiesofmethodsformodelingthese
phenomena in both the Eulerian and Langrangian framework. The majority of
theinvestigationshavebeenlookingatthesephenomenaseparately.Thus,many
models that contain only one of these phenomena are made, but some models
that combine both phenomena has been made; most of these are based on the
theoryofpopulationbalances(Ramkrishna,2000).Thisapproachisbasedona
differential equation that accounts for birth and death of particles within the
flowfield.TheequationhastobecoupledwithaEulerianapproachforsolving
the dispersed phase. The present work will investigate turbulence dispersion
andthecoalescenceandbreakupphenomenausingaLangrangianapproach.In
FLUENT,themethodavailableforLangrangianmodelingofthedispersedphase
is the discrete phase model (DPM). A further description of this model and the
modelshandlingofcoalescenceandbreakuparegiveninthefollowingsections.
4.2 DiscretePhaseModel(DPM)
The discrete phase model in FLUENT provides the possibility to simulate, in

additiontothetransportequationsforthecontinuousphase,adiscretesecond
phaseinaLagrangianframeofreference.Thesecondphaseconsistsofspherical
particles that are intended to represents drops or bubbles dispersed in the
continuousphase.
This means that the discrete phase model follows an EulerLagrange approach.
The fluid (continuous) phase is solved by the timeaveraged NavierStokes
equations, while tracking of all particles, bubbles or droplets through the
calculated flow field solves the dispersed phase. This phase exchanges
momentumandeventuallyheatandmasswiththefluidphase.
(Fluent Inc., 2006) mentions some limitations of the model. Since the particles
aretrackedindividually(onebyone),thecomputationaltimemaybeverylarge.
Thus, a very important assumption made for this model is that the dispersed
phaseoccupiesalowvolumefraction,usuallylessthan1012%.Thisisbecause
the model assumes that particleparticle interactions and the effects of the
particle volume fraction on the fluid phase are negligible. Another limitation is
thatthesteadyparticleLagrangiandiscretephasemodelisonlysuitedforflows
inwhichparticlestreamsareinjectedintoafluidphaseflowwithawelldefined
entrance and exit condition. However, the unsteadyparticle discrete phase
22
4DispersedPhaseModeling:CoalescenceandBreakup
modeliscapableofmodelingcontinuoussuspensionofparticles.See(FluentInc.,
2006) for limitations on using the discrete phase model with other FLUENT
models.
The particle motion is determined by integrating the force balance of the

particle. This force balance says that the particles inertia is equal to the forces
acting on the particle. For the xdirection in Cartesian coordinates this is given
by:
gx p
dup
(4.1)
= FD u up +
+ Fx
p
dt
Fxisanadditionalaccelerationterm,FD(uup)isthedragforceperunitparticle
mass.FDisgivenfrom:
18 CD Rerel
FD =
(4.2)
pd p2 24
The drag coefficient, CD, may be taken from different relations. Morsi &
Alexander(1972)(Eq.(4.3))andHaider&Levenspiel(1989)(Eq.(4.4))provide
two different relations for the drag coefficient and details about the factors
includedintherelations.
a
a
CD = a1 + 2 + 32
(4.3)
Rerel Rerel
24
b Re
CD =
1 + b1 Rebrel + 3 rel
(4.4)
Rerel
b4 + Rerel
Intheaboveequations,uisthefluidphasevelocity,upistheparticlevelocity,is
themolecularviscosityofthefluid, isthefluiddensity, pisthedensityofthe
particle, and dp is the particle diameter. Rerel is the relative Reynolds number,
definedby:
d p up u
(4.5)
Rerel
Equation (4.1) includes additional forces, called Fx. These are important under
certain circumstances. Among these are the virtual mass force, the force
required to accelerate the fluid surrounding the particle. Another is the
additionalforcethatexistsbecauseofpressuregradientinthefluid.Bothforces
areimportantwhen>p,andaregiveninEq.(4.6)(4.7),respectively.
1 d
Fx =
u up
2
dt
p
(4.6)
Fx = upi
p xi
(4.7)
23
AthermophoreticforcemayalsobeincludedintheadditionalforceterminEq.
(4.1). This phenomenon is when small particles suspended in a gas that has a
temperaturegradientexperienceaforceinthedirectionoppositetothatofthe
gradient.See(FluentInc.,2006)fordetailsonthisforce.
Thedispersionofparticlesinthefluidphaseduetoturbulencecanbemodeled
in two ways: stochastic tracking or particle cloud tracking. The stochastic
tracking model uses stochastic methods to predict the effect of instantaneous
velocityfluctuationsontheparticletrajectories.Theparticlecloudmodeltracks
astatisticalevolutionofacloudofparticlesaboutameantrajectory.See(Fluent
Inc.,2006)fordetailsontheturbulentdispersionofparticles.
Equation (4.1) must be discretized in order to be able to solve the equation

numerically. Before doing that, it is convenient to put the equation into the
followinggeneralequationset:
dx
(4.8)
= up
dt
dup 1
=
u up + a
(4.9)
dt
The term a includes accelerations due to all other forces than the drag force.
SeveraldiscretizationschemesforsolvingEq.(4.9)areavailableinFLUENT.The
present study has used trapezoidal tracking as the highorder scheme and
implicitdiscretizationastheloworderscheme.TrapezoidaldiscetizationofEq.
(4.9)gives:
1 t t n 1
unp 1
+ u + tunp un + ta
2 p p
(4.10)
un+1
p =
1 t
1+
2 p
Fortheimplicitscheme,thediscretizationofEq.(4.9)gives:
un+1 un+1
p
t a +
n+1
n
(4.11)
up = up +
t
1+
p
For both schemes the new particle location is computed by an explicit

trapezoidaldiscretizationofEq.(4.8):
24
1
n
n
n+1
x n+1
p = x p + t up + up
2
(4.12)
4.3 CoalescenceModeling
Equation(4.1)(4.7)describetheparticlemotiononly.Thediscretephasemodel
does also include simulation of what happens when particles collide, both the
number of collisions and their outcomes in a computationally manner. If the
collisions and their outcomes should be calculated for each possible collision
pairthecomputationalcostwouldbeprohibitive.Thus,amodelthatreducesthe
number of calculation is needed. O'Rourke (1981) made an algorithm that
efficiently reduces the computational cost of these calculations. This algorithm
has been commonly used in many commercial applications for numerical
simulation of the droplet collision process, also in the discrete phase model in
FLUENT. The method is a stochastic estimate of collisions. Parcels, which are
statistical representations of a number of individual droplets, are used to
describe the collisions instead of particles. The equation about the collision
probability is derived using classical kinetic theory on the basis of the
assumptionthattheparticleswereuniformlydistributedinthesprayofparticles
and the droplet collision had similarity to the molecular collision process.
Anotherassumptionisthattwoparcelsmaycollideonlyiftheyarelocatedinthe
samecontinuousphasecell.Themodeldoesalsoincludepostcharacteristicsof
droplets using the balance equations of mass, momentum and energy between
beforeandaftercollision.
InordertoderivetheprobabilityofcollisionO'Rourke(1981)usestherelative
distanceandrelativevelocitybetweenthetwocollidingdroplets.Hedefinedthe
collision cross section as the circle centered on the larger droplets center that
thesmallerdropletscenterhastopasswithinsothatcollisiontakesplace:
V
Pl ,s = col
V
(4.13)
= rl + rs
l ,s
The indexes l and s represent the larger and smaller drop respectively, and r
representstheradiusofadroplet.Thecollisionvolumeisthenthecollisioncross
sectionmultipliedbythedistancetraveledbytherelativevelocity:
(4.14)
Vcol = l ,surel t
ORourkesalgorithmthendeterminesthecollisionprobabilitybycalculatingthe
probability of the smaller droplet being in the collision volume. Since the
probabilityofthedropletbeinganywhereinthecontinuousphasecellofvolume
Visuniformwithinthecell,theprobabilityofthesmallerdropletcollidingwith
thelargerisequaltotheratiobetweenthetwovolumes:
(4.15)
Thus,themeanexpectednumberofcollisionsbetweenadropletinparcelland
thedropletsinparcelsisgivenby:
nV
(4.16)
n= s col
V
25
The number of droplets in parcel s is ns. The actual number of collisions is

determinedbysamplingfromaPoissondistribution:
n
n n
(4.17)
P n =e
n!
Here,nisthenumberofcollisionsbetweentwodroplets.
Once a collision has occurred, the collision outcome must be determined.

(O'Rourke,1981)considersthreetypesofoutcomes;bouncing,coalescenceand
separation, but it ignores the formation of satellite droplets. The boundaries
betweentheregimesadoptedinthemodelforequalsizeddropletsareshownin
Fig.4.1.
()
Fig.4.1:Boundariesbetweencollisionregimes(Ko&Ryou,2005).
The criterion for coalescence is related to the coalescence collision frequency,

givenby:
24 f
Ecoal = min1,
(4.18)
We
Thefunctionf()canbefoundfrom:
()
()
f = 3 2.4 2 + 2.7
(4.19)
d
1
= l =
ds
(4.20)
Here,isthedropsizeratio:
Coalescence occurs if the impact parameter x in Eq. (2.3) is less than Ecoal.
Otherwiseseparationwilloccur.Theimpactparametersquaredischosentobea
26
randomnumberfromauniformdistributionintherange[0,1].Thiscanbedone
because in a turbulent spray all values of the impact parameter squared are
equallylikely.
Thepostcollisionpropertiesofthecoalesceddropletsarefoundfromthebasic
conservation equations of mass, linear momentum and angular momentum. In
the separation and grazing bounce regimes the new velocities are calculated
basedonconservationofmomentumandkineticenergy,nomassisassumedto
exchangebetweentwocollidingdropletparcels.Thenewvelocitiesaregivenby:
s,l d s,l3 us ,l + l ,sdl3,sul ,s + l ,sdl3,s us,l ul ,s x Ecoal

us,l =
(4.21)
s,l d s3,l + l ,sdl3,s

1 Ecoal
4.4 BreakupModeling
The discrete phase model does include two different models for describing the
breakupprocess:theTayloranalogybreakup(TAB)modelfordescribinglow
Webernumber flows, and the wave model for higherWebernumber flows
(Webernumbergreaterthan100).TheTABmodelisusedinthepresentstudy
since highspeed flows are not studied; none of the simulated cases involve
averageWebernumberhigherthanapproximately1.3.
The TAB model is based upon Taylors analogy between an oscillating and
distorting droplet and a spring mass system. The surface tension forces is
equivalent to the restoring force of the spring, the droplets drag force is
represented by the external force of the springmass system, while the droplet
viscosityforcesisanalogoustothedampingforceofthespringmasssystem.The
principle of the method is further that when the droplet oscillations grow to a
criticalvaluethemotherdropletbreaksintoseveralsmallerchilddroplets.
Theequationsgoverningadampedforcedoscillatoris:
dx
d 2x
(4.22)
FTaylor kTaylor x dTaylor
=m 2
dt
dt
Here,xisthedisplacementofthedropletsequatorfromitsundisturbedposition.
By using Taylors analogy, which describes the coefficients, and putting Eq.
x
(4.22)innondimensionalizedform( y =
),oneobtains:
C
r
b
d 2 y CF c urel ,phases Ck
C dy
(4.23)
=
y d 2d
2
2
3
C
dt
dt
r
r
b
d
d
d
Here, r is the droplet radius, CF, Ck, Cd, and Cb are constants, is the surface
tension,andcanddarethedensityofthecontinuousandthedispersedphase,
respectively.
27
To determine whether or not the droplet breaks up, the amplitude for an un
dampedoscillationforeachdropletattimestepnisused:
A=
(y
Wecrit
) (
dy dt
2
+

(4.24)
Breakupisonlypossibleifthefollowingconditionissatisfied:
(4.25)
Wecrit + A > 1
ThefollowingrelationgivesWecrit:
CF curel
,phases r
(4.26)
Wecrit =
C
C
k b
Thenexttimestepmustalsobecalculatedtofindthevalueofyanditsderivative
at the next time level. Using a discretized form of the integrated form of Eq.
(4.23) does this. The size of the child droplets is determined from the
assumptionthattheenergyoftheparentdropletequalsthecombinedenergyof
the child droplets, and the number of child droplets is determined by mass
conservation(FluentInc.,2006).
4.5 CouplingBetweenthePhases
When modeling dispersed phase problems, the most important things to
consider are the forces acting on the dispersed phase due to the effect of the
motion of the continuous phase. When only this effect is considered the
approach is called oneway coupling. When taking into account that also the
continuousphaseflowpatternisimpactedbythedispersedphasetheapproach
is called twoway coupling. In denser flows with larger volume fraction of
particles, the effects of particleparticle interaction and turbulent modulation
(effects of the dispersed phase on the continuous phase turbulent eddies)
become more prevailing and should be included in the modeling. When these
phenomena are also included, the approach is called fourway coupling. The
need for these three approaches is shown in Fig. 4.2. For very dilute flows
(volume fraction<106) the effects of the dispersed phase on the continuous
phase are negligible. Larger volume fractions (<103) imply that twoway
coupling is necessary. Even larger fractions (>103) require all effects between
thetwophasestobetakenintoaccountandfourwaycouplingshouldbeused.
28
Particlesize
107
105
Onewaycoupling
FluidParticle
Twowaycoupling
FluidParticle
Dilutesuspension
Fig.4.2:Mapofflowregimes.
103
101
Fourwaycoupling
FluidParticleParticle
Densesuspension
ThediscretephasemodelinFLUENTdoesnotincludeafullyfourwaycoupled
solutionalgorithm,itdoesnotaccountforturbulentmodulation,butitcontainsa
twoway coupled algorithm. Alternately solving the dispersed and continuous
phase equations accomplish this twoway coupling until the solutions in both
phaseshavestoppedchanging.
The present study does not consider heating/cooling, evaporation, boiling or

surface reactions, hence only momentum exchange between the two phases is
considered. Thus, only exchange of momentum between the phases is included
inthemodel.Thistermiscomputedasthechangeinmomentumofaparticleas
itpassesthrougheachcontrolvolume:
18C Re
.
D
F =
u
u
+
F
(4.27)
(
)
p
other m p t
2
dd p 24
29
5 MethodologyandSimulationSetup
5.1 PhysicalDescriptionoftheProblem
As mentioned in the introduction, the principle of the Compact Tubular
Coalescer (CTC) is to force the flow consisting of water with oil droplets into a
controlled turbulent and centrifugal flow pattern. The aim is to get the oil
droplets to migrate into the middle of the tube and coalesce so the average
diameter increases. The larger droplets can then more easily be removed from
thewaterflow.Thisisdonebyaswirlelement,whichisinsertedintoacylinder
formedpipe.ThisswirlelementispresentedinFig.5.1.TheCTCdoesnotconsist
of any moving or rotating parts, which makes it compact and easy to use and
implementamongotherprocessingequipment.Adescriptionofhowthemodel
oftheCTCismadeinthisstudyandthecomputationalconditionsispresentedin
thefollowingsections.
Fig.5.1:Swirlelement.
5.2 GeneralAssumptionsfortheModeloftheCTC
Inordertoachievethedesiredoperatingconditionsforthesimulations,several
assumptionshavetobemadeforthemodel.Thesehavetobegivenasinputsin
FLUENT.Thefollowinggeneralassumptionsconcernsallinvestigationsdonein
thepresentstudy:
Gravitation is included; thegravitational accelerationisset to9.81 m/s2

inthenegativeydirection,whichmeansstraightdownward.
Allwallsaretreatedasaluminumwallsandconsideredadiabatic,which
meansthatheattransferbetweenwallsandfluidareneglected.
The boundary condition at the inlet of the CTC is constant velocity of 2

m/s, while the pressure is held constant at the outlet, equal to the
operatingpressure(1atm=101325Pa).
Thecontinuousphaseflowisconsideredtobeturbulent.Thismeansthat
aturbulencemodelmustbechosen;theReynoldsStressEquationModel
is chosen because of its capability to account for effects of streamline
30
5MethodologyandSimulation
curvature and swirl, which is essential when modeling the swirling flow
intheCTC.SeeChapter3.5fordetailsabouttheRSMturbulencemodel.
Thefluidusedforthecontinuousphaseisliquidwater.Thepropertiesfor
water is taken from FLUENTs library of fluid properties. The density is
set to be constant since water may be considered as approximately
incompressible.Acommonlyacceptedlimitforaflowtobeconsideredas
incompressibleis(White,2003):
u
Ma = 0.3
c
(5.1)
The oil droplets are in most cases modeled as liquid diesel, which has
similarpropertiestoe.g.hydrate.Theyareinjectedfromtheinletareaof
the CTC. See Chapter 5.4 for details about the impact of the different
investigations. The properties of these liquids are also taken from
FLUENTslibrary.
The diameter distribution at the inlet of the CTC is assumed to be

uniform, i.e. all droplets entering the computational domain have the
sameinitialdiameter.
Inthebreakupmodelingitisassumedthatadropletcanonlybreakup
intotwochilddroplets.
Itisassumedthatnoheatandmasstransferbetweenthedispersedand
continuousphaseoccur.Onlymomentumexchangeisconsidered.
The problem is modeled as threedimensional, thus the 3D version of

FLUENT6.3.26isrun.
All simulations of the present work are done on a computer with an Intel Core
DuoCPUwithacapacityof2.2GHz.Allsettingsusedforsimulationofthewater
phase in FLUENT are summarized and presented in Table B.1. Settings for
solutioncontrolsarepresentedinTableB.2.
5.3 SetupoftheCTCModel
5.3.1 GeneralDescriptionoftheCTCModel
Becauseofthecomplexityoftheflowwithbothturbulenceandswirlingflow,the
CTC is modeled as threedimensional. The geometry for the model is made to
simulate in the best possible way what will happen in a real physical CTC. The
geometry is quite simple. It consists of a cylinder in which a swirl element is
inserted(Fig.5.1).ThecrosssectionalareaoftheCTCisshowninFig.5.2.Both
theouterwallandtheswirlelementareconsideredasaluminumwalls,whilethe
inletvelocityandtheoutletpressureareheldconstant.
31
Fig.5.2:IllustrationofthecrosssectionalareaoftheCTC.
5.3.2 DiscretizationoftheModel
The grid is created in GAMBIT. The volume is meshed by means of a TGrid
meshing scheme. This means that the mesh primarily consists of tetrahedral
meshelements,buthexahedral,pyramidal,andwedgeelementsmaybeincluded
whereappropriate.Thegridismadewithanequaldistanceof0.0035mbetween
eachnode.Thegridconsistedof176205cellsintotal.Thelengthis1mandthe
diameter is 4 cm. The swirl element consists of 12 loops. See Chapter 6.1.1 for
detailsofhowthegridsizeischosen.ThegridispresentedinFig.5.3.
Fig.5.3:GridcreatedfortheCTC.
5.4 HowtheSimulationsareCarriedOut
Thesimulationsarecarriedoutduringthreesteps.Thefirststepwastoinitialize
the flow field. This was done with the settings defined for the boundary
conditionsattheinlet.Noparticleinjectionisdoneinthisstep.Thesecondstep
was to run this simulation stationary until convergence was reached. This was
done by first reaching convergence with a firstorder upwind spatial
discretization method for the momentum and turbulence equations, and then
switching to a secondorder upwind method to achieve a fully converged
solutionoftheflowfield.Thedefaultvaluesfortheunderrelaxationfactorsare
chosen.Thethirdstepwastodefinetheparticleinjectioninthediscretephase
model and run the simulation of the entire flow field when the particles were
32
included.Thecontinuousphase(water)wasstillmodeledasstationary,stillwith
asecondorderupwindmethodinordertoachievegoodaccuracy.Thediscrete
phasewasmodeledunsteady.Thesimulationswererunwithtimestepsof0.005
secondsforthediscretephaseuntilatotaltimeof2seconds.Theendtimefor
the simulations was chosen on the basis of the fact that the solution did not
change any more when it was carried out for longer time, i.e. the solution was
steady.SeeChapter6.1.4fordetailsaboutthecontroloftimeindependency.All
settings for the DPM and the injection of droplets that are constant for all
investigationsarepresentedinTableB.3andTableB.4,respectively.
5.4.1 InvestigatedAspects
SeveralaspectsoftheperformanceoftheCTCareinvestigated.Beforetheresults
regarding performance of the CTC are presented, some aspects regarding
validity of the model are presented and discussed. Regarding the results
achieved, the first thing of interest was to investigate the coalescence
performancefordifferentsizesofthedropletsattheinlet.Thisinvestigationis
done when both coalescence and breakup is included, and also when just
coalescence is included. These results are compared and discussed. All further
simulationsweredonewithbothcoalescenceandbreakupmodeling.Thenext
investigation was to study the impact of the volume fraction of the dispersed
phase. The volume fraction was varied from 1% to 7%. Another investigation
was done with some other liquids with different density, viscosity, and droplet
surfacetensionasthedispersedphase.Anoverviewoftheinvestigationsandthe
parametersthatarevariedinthedifferentcasesispresentedinTable5.1.Please
consultTableA.1forafullpresentationofallsimulations.
Coalescence/ Dropletliquid
Volume
Inletdiameter(m)
Breakup?
fraction
Case1 Only
Diesel
5%forall
20,40,60,80,100,120,140,
coalescence
simulations
160,180,200
Case2 Coalescence
Diesel
5%forall
20,40,60,80,100,120,140,
andbreakup
simulations
160,180,200
Case3 Coalescence
Diesel
1%,3%,5%,
100forallsimulations
andbreakup
7%
Case4 Coalescence
nPentane,Diesel, 5%forall
100forallsimulations
andbreakup Gasoil,Fueloil
simulations
Table5.1:Overviewofthevariableparametersintheinvestigations.
The performance of the CTC is mainly recognized by its ability to increase the
meandiameterofthedropletsduringtheirtransportthroughthedevice.Thisis,
ofcourse,aresultofthecoalescenceprocess.Acoupledparameteristheability
tomigratethedropletsintothemiddleofthedevice,whichwillmakeiteasierto
for the droplets to coalesce. In all investigations the mean diameter used for
comparisonistheSauterMeanDiameter,d32.Thisisdefinedasthediameterofa
spherethathasthesamevolume/surfacearearatioasaparticleofinterest.Itis
a common measure in fluid dynamics for estimating the average particle size.
Therelationforthegeneralmeandiameterdjkisdefinedas:
33
(d )
jk
jk
()
f (d )dd
d j f d dd
dk
(5.2)
Thedistributionfunction,f(d),isequaltothetotalnumberofdropletswhenitis
integratedfrom0to.Hence,theSauterMeanDiameter,d32,canbeexpressed
as:
d32
N d
=
N d
i
i
2
Niisthenumberofdropletsinthedomaini.
(5.3)
34
35
6 ResultsandDiscussion
The purpose of this Masters thesis was, as mentioned, to make a model for
simulating coalescence of dispersed oil droplets in continuous water flow. The
makingofamodeloftheCompactTubularCoalescer(CTC)didthis.Hence,there
were several aspects of interest. The validity and accuracy of the model is of
greatinterest,andregardingtheperformanceofthedevicetheincreaseofmean
diameter from the inlet to the outlet is crucial. The following chapters present
and discuss the results achieved. All diagrams are plotted with black dots for
eachsimulatedresult,andadarkredtrendlineisaddedinordertoseethetrend
oftheresults.
6.1 ValidityoftheModel
WhenperformingCFDsimulationsitisimportanttoevaluatethevalidityofthe
simulations. There are several aspects of the simulations that can tell us
somethingabouthowaccurateandvalidtheachievedresultsare.Thisworkhas
focused on four things: grid independency, converged residuals, mass flow
balance, and time independency. Grid independency means that the results are
relatively constant even if the simulations are done on a mesh with smaller
computational cells. Converged residuals mean that the residuals reach a value
thatissmallerthantheselectedconvergencecriterion.Thismeansthatsolutions
varywithasmallenoughvaluebetweenthelastiterations.Massflowbalance
is a control of that the same amount of water that enters the device, must also
.
comeoutontheotherside,i.e. m water ,in = m water ,out .Timeindependencyischecked

in order to control convergence of the discrete phase since it is simulated as
unsteady. The control consists of checking that the solution does not vary
anymoreasthesimulationisiteratedformoretimesteps.
Allvalidationsimulationsaredonewithatestcasewithdropletsofsize100m
with only coalescence included (not breakup), a volume fraction of 5 %, and
liquid diesel as dispersed phase, except from the mass flow balance evaluation
where all simulations are considered. All other properties and parameters are
equaltothosethatareconstantforallsimulationsduringthisstudy.
6.1.1 GridIndependency
Sixsimulationsoftheexactsamecase,butwithsixdifferentcellsizesweredone
inordertocheckthedependencyofthegrid.TheoverallSauterMeanDiameter
(SMD),d32,forthewholedomainwasselectedascomparingparameterbetween
the different simulations, while the case with the finest grid was chosen as
reference.TheresultsofthistestarepresentedinTable6.1andinFig.6.1.
36
6ResultsandDiscussion
Numberofcells
Timeforsimulation Meandiameter,d32
221512
Approx.5hours
203.50m
176205
Approx.3hours
203.99m
141568
Approx.2.5hours
210.59m
112787
Approx.2hours
214.24m
63655
Approx.1hours
228.50m
32375
Approx.0.5hours
247.56m
Table6.1:Comparisonbetweensixmesheswithdifferentcellsizes.
Deviation
0%
0.24%
3.49%
5.28%
12.29%
21.65%
Griddependency
260
OverallSMD
250
240
230
y=2E09x20.000x+263.8
220
210
200
190
0
50000
100000
150000
200000
250000
Numberofcells
Fig.6.1:Visualizationofhowtheresultvarieswithcellsizes.
One can see that the deviation decreases as the number of cells in the grid is
increased.Thetrendlinethatflattensoutwhenthenumberofcellsisincreased
visualizesthis.Forthetwocaseswithlargestamountofcellsthedeviationlies
beneath 1 %, which is quite acceptable. The solution of the four grids with the
largestcellsizearevaryingtoomuchcomparedwiththeonewiththesmallest
cellsize,andcannotbeconsideredasgridindependent.Becausethemeshwith
the smallest cell size is the most timeconsuming this mesh is not preferable.
Thus, the grid with 176205 cells is chosen since the result is within the
acceptabledeviationlimitanditisnotverytimeconsuming(around3hoursfor
onesimulation).
6.1.2 ConvergenceofResiduals
Asmentioned,thewaterphasewasiterateduntilconvergenceoftheresiduals,
first for firstorder upwind method and secondly for secondorder upwind
method. This was done without introduction of the oil droplets. Thus, a
converged, stationary solution was achieved before the droplets were
introduced. The convergence criteria for all residuals were 103. They are
presentedinFig.6.2.
37
Fig.6.2:Convergedresidualsforstationarysolutionofthewaterphase.
Afterintroducingtheoildropletstheresidualsincreasesabitbeforetheyflatout
andbecomestotallystable(thelargestataround102)(Fig.6.3).Thestabilizing
happens when the time stepping has reached the time for the droplets to pass
through the whole device. Even though the residuals do not decrease, the
solution can still be valid since the residuals are very stable and at a quite low
value. Thus, other convergence controls like the mass flow balance and time
independency should be evaluated in order to make a conclusion about the
accuracyandvalidityofthesolution.
Fig.6.3:Residualsforsolutionofwaterphaseafterintroductionofdroplets.
6.1.3 MassFlowBalance
.
The mass flow balance ratio, defined as
m water ,out
.
for all simulations are
m water ,in
presentedinTable6.2.SeeTableA.1fordetailsabouteachcasenumber.
38
Casenumber
Massflowbalanceratio
1
0.999853
2
1.000132
3
0.999869
4
1.000458
5
1.000116
6
1.000456
7
0.999991
8
0.999456
9
1.000463
10
1.000048
11
1.000194
12
1.000056
13
0.999615
14
0.999626
15
0.999978
16
0.999668
17
1.000011
18
0.999540
19
1.000204
20
0.999789
21
1.000014
22
0.999737
23
0.999453
24
0.999448
25
0.999974
26
1.000169
Table6.2:Massflowbalanceratioforallsimulations.
All values of the mass flow balance ratio are very close to 1, which means that
thisconvergencecriterionisfulfilledforallsimulations.Thisisagoodindicator
forvalidsolutions.
6.1.4 TimeIndependency
In order to check if the solution remained approximately unchanged when the
simulationwasrunforlongertime,thetestcasewasrununtilthetotaltimeof3
seconds for the dispersed phase, the water phase is all the time simulated as
steady.Therestofthesimulationsarestoppedatatotaltimeof2seconds.The
simulation that is run until 3 seconds is chosen as reference. The result is
presentedinTable6.3.
Totaltime
Meandiameter,d32
Deviation
3seconds
0%
203.98m
2seconds
0.074%
204.13m
Table6.3:Resultsofrunningthetestcaseuntil6seconds.
Thisresultshowsthatthechangefrom2to3secondsisverysmall,under0.1%.
Thismeansthatthesolutionisapproximatelytimeindependent.
Thepreviousvalidationtestshaveshownthatthemodelusedinthesimulations
of this thesis is approximately both grid and time independent, and the mass
flow balance shows convergence with very small margin of error. The only
39
criterion that is not fulfilled, as it preferably should have been, is the

convergence of the residuals after the dispersed phase is introduced. But the
residuals do not in any way diverge; in fact they stabilize at a quite low value.
Conclusively,withthreeapproximatelyfulfilledconvergencecriteriaandthelast
onenotdirectlyunfulfilled,itisbasisforsayingthatthemodelisvalidandquite
accurate.Hence,itisausefulmodelforthesimulationsofwhichresultsfollowin
thenextchapters.
6.2 ImpactofChangingDropletsInletDiameter
Ten different inlet diameters for the droplet are investigated. The Sauter Mean
Diameter(SMD)forthedropletsinthelastquarteroftheCTCisusedtoestimate
thediameterincreaseofthedropletsthroughthedevice.Thisisdonebecauseit
will give a better picture of the real outlet conditions than if all droplets in the
whole domain where considered. Many of these droplets are just in the
beginningoftheirwaythroughthedeviceandhavenotbeeninfluencedbythe
effectoftheCTCconfiguration.ThecapabilityoftheCTCtogatherthedroplets
in the middle of the tube is of great interest. This effect, called the migration
effect,willleadtohighercollisionfrequencyandhencehighercoalescencerate.
The migration effect is visualized by comparing the total mass of droplets that
lies inside the inner half of the cross sectional area of the CTC to the mass of
dropletsthatareintheouterhalf.AsfortheevaluationofincreaseofSMD,the
migrationeffectisalsoevaluatedatthelastquarterofthedevice.
Innerhalf
Outerhalf
Fig.6.4:SimplifiedpictureofthepartsofthecrosssectionalareaoftheCTC.
6.2.1 OnlyCoalescenceModeling
Thefirstsimulationsaredonewithonlycoalescencemodeling,i.e.nobreakup
modeling. The results regarding increase of SMD during the transport through
theCTCarepresentedinFig.6.5aspercentageincreaseplottedagainstdroplet
inletdiameter.
40
y=0.001x+1.687
IncreaseofSMD
200%
150%
100%
50%
0%
0
50
100
150
200
Inletdiameter[m]
Fig.6.5:Increaseofmeandropletdiameter.
Partofdropletsinthemiddle
By considering Fig. 6.5 one can see that the increase is quite high for all inlet
diameters, from approximately 160 % up to 220 %. This observation clarifies
thatthecoalescenceprocessdoestakeplaceandtheextentofitislargeenough
tocoalesceenoughdropletswitheachothersothattheincreaseofthedroplets
diameterisremarkable.Thediameterincreasethroughthedeviceisquiteinthe
same order of magnitude for all inlet sizes, but the trend line shows a small
growth in percentage diameter increase for the larger inlet diameters. This
agreeswiththefactshowninFig.6.6:thatthelargerdropletsmigratemoreinto
themiddleofthetubeandincreasethecollisionfrequency.Thisisbecausethe
centrifugal effects have larger impact on larger particles than the smaller ones.
Butinaregularstraighttubeonewouldexpectthecoalescenceratetodecrease
with larger droplet sizes. This is because the total number of particles is much
largerforsmallinletdiameters(themassflowrateisthesame);hencethereare
moreparticlesthatcancollide/coalesceandincreasetheSMD.
120%
100%
80%
y=0.002x+0.436
60%
40%
20%
0%
0
50
100
150
200
Inletdiameter[m]
Fig.6.6:Migrationofparticlesintothemiddleofthetube.
41
Another argument for larger coalescence rate for smaller particles is that
coalescenceasaresultofdropletcollisionis,asdiscussedinchapter2.2and4.3,
most probable for small values of both the Weber number and the impact
parameter.ItcanclearlybeseenfromFig.6.7thattheparticlesaverageWeber
numberincreases(approximatelylinearly)whenthedropletdiameterincreases.
AverageWebernumber
1.40
1.20
y=0.005x+0.016
1.00
0.80
0.60
0.40
0.20
0.00
0
50
100
150
200
Inletdiameter[m]
Fig.6.7:AverageWebernumberforonlycoalescencemodeling.
Although,asdiscussed,thetrendisthatthecoalescencerateandincreaseofSMD
increases slowly with the increase of inlet diameter. This shows that the
increasingeffectoftheCTCwithincreasinginletdiametersisabletocounteract
the decreasing effect the physics of the coalescence phenomenon has on the
coalescencerateforlargerparticles.
6.2.2 BothCoalescenceandBreakupModeling
Inordertobeabletomakeamorerealisticsimulationofwhatwillhappeninthe
CTC,boththecoalescencephenomenonandtheoppositebreakupphenomenon
need to be modeled. The next simulations are carried out with the use of
FLUENTs breakup model for low Weber number flows, the TAB model. The
resultsregardingSMDofthedropletsattheendofthetubearepresentedinFig.
6.8.
PercentagechangeofSMD
42
200%
150%
100%
y=2.435e0.00x
50%
0%
0
50
100
150
200
Inletdiameter[m]
Fig.6.8:SMDwhenbreakupmodelingisincluded.
The results from this simulations show, as for the modeling with no breakup,
thatthereisaremarkableincreaseoftheSMDforalltestedinletdiameters,from
approximately70%upto220%.But,unliketheresultsforthesimulationswith
no breakup modeling, the percentage increase of SMD diminishes as the inlet
diameters are increased. This observation is supported by the theory about
breakupdiscussedinChapter2.3.Forbreakuptooccur,acriticalvalueofthe
Webernumber,(NWe)crit,exists.IfthevalueoftheactualWebernumberdoesnot
passthiscriticalvalue,breakupwillnotoccur.Byrelatingthedropletdiameter
to this critical Weber number, a critical diameter, dcrit, was found. Hence the
droplet diameter is directly linked to the probability of breakup, and larger
droplet diameters increase the probability of breakup. But even though the
breakuprateislargerforthelargerdroplets,thecoalescencerateisevenlarger.
ThisisbecauseofthecentrifugalflowpatterntheCTCmakes.TheresultingSMD
at the outlet has increased with a minimum of 70 % (for the largest droplets)
fromtheinlet.Fig.C.1C.5showcontoursoftheconcentrationofthedispersed
phase for five different inlet diameters in a longitudinal cross sectional area of
theCTC.
TheaverageWebernumberforthedropletsintheCTCispresentedinFig.6.9.
Compared to the results with no breakup modeling the increase of Weber
number with the inlet diameter flattens out for larger inlet sizes. The droplet
Weber number is, as discussed, strongly coupled with the diameter of the
droplets.Thus,theincreasingbreakuprateforlargerdropletscausesstagnation
oftheincreaseoftheSMDandthedropletWebernumber.
43
AverageWebernumber
0.70
0.60
0.50
y=0.211ln(x)0.483
0.40
0.30
0.20
0.10
0.00
0
50
100
150
200
Inletdiameter[m]
Fig.6.9:AverageWebernumberforbothcoalescenceandbreakupmodeling.
As observed from the previous results, the droplets diameters increase more
and more due to coalescence (more than breakup) as the droplets travel
through the device. This is visualized in Fig. 6.10 where the diameters of all
droplets (or parcels that FLUENT works with, c.f. chapter 4.2) are plotted as a
function of the distance traveled through the tube. The result can be
approximated with a trend line that shows linear increase of the droplets
diameters as they travel through the tube. This is also visualized in Fig. 6.11
where the droplets are presented and they are colored by the size of the
dropletsdiameters.
500
450
Dropletdiameter
400
y=73.68x+106.7
350
300
250
200
150
100
50
0
0.00
0.20
0.40
0.60
0.80
1.00
PositionintheCTC
Fig.6.10:IncreaseofdiameterthroughouttheCTC.
44
Fig.6.11:VisualizationofdropletsandtheirdiameterinsidetheCTC.
6.3 ImpactoftheDispersedPhaseVolumeFraction
Four different inlet volume fractions were investigated in order to see if this
parameterhadsomeimpactontheperformanceoftheCTC.Theresultsobtained
arepresentedinFig.6.12.
PercentageincreaseofSMD
155%
150%
145%
140%
y=0.127ln(x)+1.858
135%
130%
125%
120%
115%
110%
0%
1%
2%
3%
4%
5%
6%
7%
8%
Inletvolumefraction
Fig.6.12:Increaseofmeandropletdiameterfordifferentvolumefractions.
The coalescence process takes to a larger extent place for higher volume
fractionsthanforthelowerones.Thiscoincideswiththeexpectedresults;when
therearemoredropletsinthedomainthechangeofcollisionincreases.Thiscan
be explained by looking at the theory of collision volumes (defined bye Eq.
(4.14)). When the concentration of droplets are large, the amount of collision
volumeislarge,thustheprobabilityofonedroplettobeinthecollisionvolume
ofanotherdropletislarger.Hence,theprobabilityofcollisionandcoalescenceis
larger.
45
6.4 ImpactoftheTypeofDropletLiquidontheMigration
The last investigation was to evaluate the effect on the CTC performance by
changing the type of liquid for the droplets. Four different liquids are
investigated: npentane, diesel, gas oil, and fuel oil. These oils have different
materialproperties;npentaneisthelightest,whilefueloilistheheaviest,most
viscous oil. This testing is done with inlet diameter of 100 m and a volume
fractionof5%forallfoursimulations.TableA.2showsthematerialproperties
of the four tested types of liquid. Table 6.4 shows the results achieved for the
fourdifferentdropletliquids.
Dropletliquid
IncreaseofSMD
Part of droplets in the

middle
nPentane
100.91%
86.26%
Diesel
147.29%
85.21%
Gasoil
155.26%
70.80%
Fueloil
265.72%
59.66%
Table6.4:Performanceparametersfordifferenttypesofdropletliquid.
The results show a clear tendency regarding the migration of droplets into the
middleofthetube:thelighteroilswithlowerdensityandviscositymigratemore
into the middle of the tube. Lower density ratio between the oil and the water
makes the centrifugal forces more effective on the water. The heavier water is
pulledawayfromthecenterofthetubeduetothecentrifugalflowpattern.But
on the other hand, the results show that the increase of SMD is larger for the
heaviestoils.Thiscontradictswiththefactthatthelightestoilsgathermorein
the middle of the tube increase the collision frequency. But there are some
considerable differences in the other physical properties: viscosity and droplet
surface tension. Increased viscosity leads to increased Weber number between
the droplet und the surrounding fluid, which further leads to reduction of the
breakuprate.AsonecanseefromTableA.2,theviscosityismuchlargerforthe
heavieroils.Theotherpropertythatdiffersbetweenthedifferentliquidsisthe
droplet surface tension. This parameter is included in the expression for the
Weber number between the droplets. Larger droplet surface tension leads to
smaller values of this Weber number, which further leads to increased
coalescence rate. Thus, decreased breakup rate combined with increased
coalescenceratemakestheincreaseofSMDlargestforfueloil,eventhoughthe
migrationeffectislowestforthisliquid.
6.5 LimitationsandWeaknessesoftheModel
ItshouldberemarkedthataCFDmodeldoesveryoftenhavethepossibilityto
bebetter,moreaccurateandmoreapplicable.Thisrequiresalotoftestingand
modification,andthisprocessisverytimeconsuming.Thetimeavailableforthis
Mastersthesisdidnotallowformoreworkonthemodelthanwhatispresented
inthepreviouschapters.Neverthelessitisveryimportanttobeawareofwhat
weaknesses and limitations the present model does have. Hence a short
discussionofthissubjectfollows.
Chapter 6.1.2 showed that convergence for the residuals after the dispersed
phaseinjectionwasintroducedwasnotreached.Still,astablevaluewasreached,
46
so no divergence occurred. Only 5 iterations for the continuous phase per

iteration for the discrete phase were allowed. A closer look at the residuals
showsthattheresidualsvaluesweredecreasingduringthe5iterationsbetween
the iterations of the discrete phase, but down to values below 103 for all
residuals. Allowing even more iterations for the continuous phase between the
discretephaseiterationswouldprobablyresultedinmoreacceptablevaluesfor
theresiduals.
The test of grid independency in chapter 6.1.1 was done for six grids with
different cell sizes. The tests showed that the difference of the test parameter
between the chosen grid and the grid with smallest cells was very small;
increasingthenumberofcellswitharound25%didnotchangetheresultwith
morethan0.24%.Thiswasagoodindicatorofgridindependency.Butitshould
be remarked that this test was only done with coalescence modeling, and only
one parameter was tested for grid independency. Although this was the main
resultofinterest,someotherresultsshouldhavebeencheckedforthedifferent
gridsinordertobemorecertainofgridindependency.Inaddition,onlyonetype
ofgridwastested,soideallysomeothergridtypesshouldalsobetestedtocheck
forgridindependency.
Thetimestepusedfortheunsteadysimulationofthedispersedphase(t=0.005
s)wasrelativelylarge,andshouldhavebeentestedwithlowervaluesinorderto
checkthatthesolutionisnottimestepdependent.
Regarding applicability of the model, it should be mentioned that the model is

based on FLUENTs built in model for dispersed phase flow, the discrete phase
model(DPM).Thismodelhasaseriouslimitationwhenitcomestothevolume
fractionofthedispersedphase.Thedispersedphasehastobesufficientlydilute;
itsvolumefractionshouldnotexceed10%(FluentInc.,2006).
6.6 SuggestedImprovementsoftheModel
Experiencesdonebythesimulationsinthisworkshowthatthemodelusedmay,
without fundamental changes, be tested more and eventually improved. As
discussed above, some more testing and modification of computational
parametersshouldbedoneinordertogetmorecertainvalidationofthemodel.
Moretimeshouldalsobespentonoptimizingthemodelregardingbothaccuracy
andefficiency.Themodelshouldbeascomputationallyinexpensiveaspossible
withoutloosingaccuracy.Thegridwasequidistantandnotrefinedinanyareas.
Thus,themeshcanbemodifiedandrefinedinimportantareasoftheflowfield,
i.e.inareaswithhighconcentrationofdropletsandclosetowalls,andcoarsened
in less important areas. FLUENT provides the opportunity to adapt the grid
easilyinareaslikethis(wheretheconcentrationofthedispersedphaseislarge).
The time step for the unsteady simulation of the dispersed phase was also
discussedintheprevioussection,andthisshouldbeloweredinordertocheck
for and avoid timestep dependent solution. The computational time of the
present model is not very long and expensive compared to many other CFD
simulations.Thus,arefiningofthemeshandaloweringofthetimestep,which
47
will increase the computational time dramatically, should be tested in order to

eventuallyobtainamoreaccurateandreliablesolutionoftheproblem.
48
49
7 Conclusion
The results of the simulations showed that the model is able to predict the
performance of the CTC and the coalescence process of oil droplets of some
different oils. The change of SMD of the droplets is the important parameter
regardingtheperformanceofthedeviceandwasthusinvestigated.
AllcaseswithdifferentinletdropletdiametersresultedinincreaseoftheSMD,
rangingfromaround70%to220%increase.Theincreasewaslargestforthe
smallestinletdiametersandsmallestforthelargestinletdiameters.Theresults
showed that the configuration of the CTC was very effective. The droplets are
gatheredinthemiddleofthetubeandthecollisionfrequencyisincreased.When
no breakup modeling was included the increase of SMD was larger for large
inlet droplets. The discussion in Chapter 6.2 concluded that one would expect
largerdropletstohavealowercoalescenceratethanthesmallerones.
Inclusionofbreakupmodeling,whichisnecessarytogetarealisticsimulation,
resultedinlowerincreaseofSMDforlargedroplets(>6080m).Accordingto
thediscussioninChapter6.2thiswasexpected,becausetheprobabilityofbreak
up increases with the diameters of the droplets. Even though the inclusion of
breakup resulted in lower performance of the CTC, the performance was still
verygoodforthesmallestparticles(SMDincreasearound200%)andacceptable
forthelargestparticles(SMDincreasearound70%).
Othersimulationsresultedintheconclusionthatlargervolumefractionsofthe
dispersedphasegeneratehigherincreaseoftheSMD.Thephysicalpropertiesof
the droplet liquid were found to be a very important factor on the CTC
performance. Oils with high viscosity and large droplet surface tension had a
highcoalescencerate,eventhoughthemigrationeffectwaslow.
The study of the validation of the simulated results shows that the model
delivers results that are, according to several controls, both quite accurate and
valid. Several validation controls are done. Almost all tests gave satisfying
results,butsomemoretestsandsomemodificationshouldbedoneinorderto
increasevalidityandaccuracy.Somelimitationswerepointedout;inadditionto
the problem with the breakup modeling the model missed the applicability of
handlingdispersedphasevolumefractionslargerthan10%.
50
7Conclusion
51
8 SuggestionsforFutureWork
Additional work within the problems discussed in this work should first of all
concentrateonimprovementofthemathematicalmodeldeveloped.Asdiscussed
in chapter 6.5 and 6.6 the model should undergo some more testing to ensure
that it is grid independent and timestep independent, i.e. adaption of the grid
and reduction of the time step for particle tracking. More time should also be
usedonoptimizingthemodelregardingbothaccuracyandefficiencyinorderto
get a model that is as economic as possible within the frames of acceptable
accuracy.
Secondly,someotheraspectsofthemodelshouldbetested.Thismaybetesting
of the effect of changing the diameter and length of the CTC. What must the
length of the device be if the diameter is increased in order to have the same
performance? Also, it would still be interesting to check the performance of he
CTConevenmorekindsofoils,andalsoothermediaascontinuousphase.The
model should also be tested for more realistic inlet distributions than the
uniformdistributionusedinthisstudy.
It would also be interesting to see the result of another approach for the same
case of study, e.g. a Eulerian approach. This is a multiphase model provided in
FLUENT. In order to model coalescence and breakup with this approach, the
theoryofpopulationbalancesmustbeincluded.Thereexistapurchasableadd
onmoduleforthelatestversionofFLUENT,whichincludespopulationbalance
modeling.Thisapproachwillnothavethesamelimitationregardingthevolume
fractionofthedispersedphase.
52
8SuggestionsforFutureWork
53
AppendixA,OverviewofSimulationsandLiquidProperties
Table A.1 and A.2 contain an overview of all performed simulations and the
propertiesofallusedliquids,respectively.
Case
Coalescence Breakup
number
modeling
modeling
1
Yes
No
2
Yes
No
3
Yes
No
4
Yes
No
5
Yes
No
6
Yes
No
7
Yes
No
8
Yes
Yes
9
Yes
Yes
10
Yes
Yes
11
Yes
Yes
12
Yes
Yes
13
Yes
Yes
14
Yes
Yes
15
Yes
Yes
16
Yes
Yes
17
Yes
Yes
18
Yes
Yes
19
Yes
Yes
20
Yes
Yes
21
Yes
Yes
22
Yes
Yes
23
Yes
Yes
24
Yes
Yes
25
Yes
Yes
26
Yes
Yes
TableA.1:Presentationofallsimulations.
Droplet
liquid
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
nPentane
Gasoil
Fueloil
Volume
Fraction
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
5%
1%
3%
7%
5%
5%
5%
Inlet
diameter
20m
40m
60m
80m
100m
120m
140m
160m
180m
200m
20m
40m
60m
80m
100m
120m
140m
160m
180m
200m
100m
100m
100m
100m
100m
100m
Dropletliquid
Chemical
Density
Viscosity
Notation
(kg/m3)
(kg/ms)
npentane
C5H12
626.0
0.000229
Diesel
C10H22
730.0
0.002400
Gasoil
C16H29
830.0
0.003320
Fueloil
C19H30
960.0
0.048000
Water
H2O
998.2
0.001030
TableA.2:Materialpropertiesforthedifferentliquidsusedinthestudy.
Droplet Surface
Tension(N/m)
0.015508
0.026326
0.019036
0.030000
54
AppendixA,OverviewofSimulationsandLiquidProperties
55
AppendixB,SettingsusedinFLUENT
Table B.1 B.4 present a complete overview of all settings used for the
simulationsinFLUENT.
Category
MODEL
Sub
Category1
Solver
Energy
Viscous
Radiation
OPERATING Pressure
CONDITIONS
Gravity
Sub
Category2
Solver
Formulation
Time
Space
Gradient
Option
Model
Reynolds
StressModel
Reynolds
Stress
Options
NearWall
Treatment
Model
Constants
Operating
Pressure
Reference
Pressure
Location
Choice/Value
Choice/Value
PressureBased
Implicit
Steady
3D
GreenGaussCellBased
OFF
ReynoldsStress
LinearPressureStrain
WallBCfromk
Equation
WallReflectionEffects
StandardWall
Functions
Defaultvaluesforall
constants
OFF
101325Pa
x=0,y=0,z=0
Gravitational
Acceleration
BOUNDARY Default
Interior
CONDITIONS Interior
Fluid
Fluid
Water(liquid)
Inducer
Wall
Defaultvalues
Inlet
Velocity
VelocityMagnitude
inlet
Turb.KineticEnergy
Turb.Diss.Rate
KandTurb.Intensity
DiscretePhaseBCType
OuterWall
Wall
Defaultvalues
Outlet
Pressure
GaugePressure
Outlet
Turb.KineticEnergy
Turb.Diss.Rate
KandTurb.Intensity
DiscretePhaseBCType
TableB.1:SettingsforthesimulationsofthewaterphaseinFLUENT.
x=0m/s2,y=9.81m/s2,
z=0m/s2
Defaultvalues
2m/s
0.3m2/s2
1m2/s3
Escape
101325Pa
0.3m2/s2
1m2/s3
Escape
56
Category
Equations
PressureVelocity
Coupling
UnderRelaxation
Factors
Choice/Value
Flow,Turbulence,
ReynoldsSresses
SIMPLE
Choice/Value
Pressure
0.3
Density
1
BodyForces
1
Momentum
0.7
Turb.Kin.Energy
0.8
Turb.Diss.Rate
0.8
Turb.Viscosity
1
ReynoldsStresses 0.5
Disc.Pha.Sources 0.5
Discretization
Pressure
Standard
Momentum
2.orderupwind
Turb.Kin.Energy
2.orderupwind
Turb.Diss.Rate
2.orderupwind
ReynoldsStresses 2.orderupwind
TableB.2:SettingsforthesolutioncontrolsinFLUENT.
Category
DISCRETE
PHASE
MODEL
SubCategory1
Interaction
SubCategory2
Interactionwith
continuousphase
Choice/Value
ON
Choice/Value
UpdateDPM
SourcesEvery
FlowIteration
OFF
Numberof
ContinuousPhase
Iterationsper
DPMIteration
UnsteadyParticle
Tracking
10
ON
ParticleTimeStep
Size
NumberofTime
Steps
Tracking
Parameters
DragParameters
0.005s
Defaultvalues
Defaultvalues
PhysicalModels
Options
SprayModel
Numerics
Options
TrackingScheme
Selection
Defaultvalues
DropletCollision
DropletBreakup
Defaultvalues
Defaultvalues
SeeTable5.1
SeeTable5.1
Particle
Treatment
Tracking
TableB.3:SettingsusedintheDPMinFLUENT.
57
Category
InjectionType
ReleaseFrom
Surfaces
ParticleType
Material
Diameter
Distribution
PointProperties
SubCategory
Surface
Inlet
Choice/Value
Choice/Value
Inert
SeeTable5.1
Uniform
0m/s
0m/s
2m/s
SeeTable5.1
0s
10s
SeeTable5.1
Turbulent
DiscreteRandom ON
Dispersion
WalkModel
RandomEddy
OFF
Lifetime
TimeScale
0.15
Constant
TableB.4:SettingsfortheinjectionusedintheDPMinFLUENT.
Xvelocity
Yvelocity
Zvelocity
Diameter
StartTime
StopTime
TotalFlowRate
Stochastic
Tracking
58
59
AppendixC,VisualizationofDispersedPhaseConcentration
FigureC.1C.5presentscontoursofthedropletconcentrationinfivecaseswith
differentinletdiameterofdieseldroplets.
Fig.C1:ContoursofDPMconcentration,inletdiameterof20m(kg/m3).
60
AppendixC,VisualizationofDispersedPhaseConcentration
61
ListofReferences
1.Ashgriz,N.,&Poo,J.Y.(1990).Coalescenceandseparationinbinarycollisions
ofliquiddrops.JournalofFluidMechanics,221,183204.
2.Dhainaut,M.(2002).LiteratureStudyonObservationandExperimentson
CoalescenceandBreakupofBubblesandDrops.SINTEF,MaterialsTechnology.
SINTEF.
3.Elgobashi,S.(1991).Particleladenturbulentflows:directsimulationand
closuremodels.Appliedscientificresearch,48(3),301.
4.FluentInc.(2006).FLUENT6.3User'sGuide.FluentInc.
5.Haider,A.,&Levenspiel,O.(1989).DragCoefficientandTerminalVelocityof
SphericalandNonsphericalParticles.Powdertechnology,58(1),6370.
6.Hinze,J.(1955).FundamentalsoftheHydrodynamicMechanismofSplittingin
DispersionProcesses.AICheJournal,3(1),289295.
7.Ko,G.H.,&Ryou,H.S.(2005).Modelingofdropletcollisioninducedbreakup
process.InternationalJournalofMultiphaseFlow,31(6),723738.
8.Mller,B.(2007).IntroductiontoComputationalFluidDynamics.Introduction
toComputationalFluidDynamics.
9.Mashayek,F.,Ashgriz,N.,Minkowycz,W.,&Shotorban,B.(2003).Coalescence
collisionofliquiddrops.InternationalJournalofHeatandMassTransfer,46(1),
7789.
10.Morsi,S.,&Alexander,A.(1972).AnInvestigationofParticleTrajectoriesin
TwoPhaseFlowSystems.JournalofFluidMechanics,55(SEP26),193208.
11.O'Rourke,P.(1981).Collectivedropeffectsonvaporizingliquidsprays.Thesis,
LosAlamosNationalLab.,NM(USA).
12.Prince,M.J.,&Blanch,H.W.(1990).Bubblecoalescenceandbreakupinair
spargedbubblecolumns.AIChEJournal,36,14851499.
13.Ramkrishna,D.(2000).Populationbalances:theoryandapplicationsto
particulatesystemsinengineering.SanDiego:AcademicPress.
14.vanLeer,B.(1977).TowardstheUltimateConservativeDifferenceScheme.
IV.ANewApproachtoNumericalConvection.J.Comput.Phys.
62
ListofReferences
15.Versteeg,H.,&Malalasekra,W.(1995).AnIntroductiontoComputational
FluidDynamics:TheFiniteVolumeMethod.Harlow,England:PearsonEducation
Limited.
16.White,F.M.(2003).FluidMechanics.

Mathematical Modeling of Coalescence

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Mathematical Modeling of Coalescence

Hochgeladen von

Copyright:

Verfügbare Formate

Mathematical Modeling of Coalescence

of Oil Droplets in Water Flow

Master of Science in Energy and Environment

Norwegian University of Science and Technology

Formulate a mathematical model for coalescence of oil droplets in a water flow.

If necessary, suggest further improvements of the mathematical model.

Assignment given: 01. February 2008

The validity of the model is evaluated before different aspects of the

Future work is recommended to concentrate on improving the present model

In turbulent dispersions, both breakup and coalescence usually take place

Dhainaut (2002) mentions the other possible outcomes from a collision;

By relating the impact parameter to the two types of separation collisions, it

Here, d is the viscosity of the dispersed phase. Laboratory experiments and

In the case of hydrodynamic instabilities at the surface of the particle, it may

For incompressible flows, the equations for conservation of mass and

The development of CFD codes has given many advantages in engineering

Sm is the mass added to the continuous phase from an eventually dispersed

In order to get a picture of how the finitevolume method works, a short

controlvolume controlvolume volume

is a diffusion coefficient and S is a source term. For simplicity the following

The schemes presented for discretization of the general transport equations in

aPu = anbunb + p f Afx

The problem with pressurevelocity linkage can be resolved by adopting an

Here, p is the underrelaxation factor for pressure. See (Versteeg &

Turbulence modeling is based on the principle that the velocity can be

The discrete phase model in FLUENT provides the possibility to simulate, in

The particle motion is determined by integrating the force balance of the

Equation (4.1) must be discretized in order to be able to solve the equation

For both schemes the new particle location is computed by an explicit

The number of droplets in parcel s is ns. The actual number of collisions is

Once a collision has occurred, the collision outcome must be determined.

The criterion for coalescence is related to the coalescence collision frequency,

s,l d s,l3 us ,l + l ,sdl3,sul ,s + l ,sdl3,s us,l ul ,s x Ecoal

s,l d s3,l + l ,sdl3,s

The present study does not consider heating/cooling, evaporation, boiling or

Gravitation is included; thegravitational accelerationisset to9.81 m/s2

The boundary condition at the inlet of the CTC is constant velocity of 2

The diameter distribution at the inlet of the CTC is assumed to be

The problem is modeled as threedimensional, thus the 3D version of

comeoutontheotherside,i.e. m water ,in = m water ,out .Timeindependencyischecked

The mass flow balance ratio, defined as

for all simulations are

criterion that is not fulfilled, as it preferably should have been, is the

Part of droplets in the

so no divergence occurred. Only 5 iterations for the continuous phase per

Regarding applicability of the model, it should be mentioned that the model is

will increase the computational time dramatically, should be tested in order to

Das könnte Ihnen auch gefallen