Natural Gas Technology

June 2011
Kjell Kolsaker, EPT
Submission date:
Supervisor:
Norwegian University of Science and Technology
Department of Energy and Process Engineering
Modelling of Multistream LNG Heat
Exchangers
Joan Soler Fossas
I

Abstract
The main goal of this thesis is to find out if a liquefied natural gas multistream heat
exchanger numerical model is achievable. This should include several features usually
neglected in nowadays available heat exchanger models, such as flow maldistribution,
changes in fluid properties and heat exchanger dynamic behaviour. In order to accomplish
thatobjectiveasimplercaseismodelled.Effortsareputinachievingnumericalstability.
A counter flow natural gas and mixed refrigerant heat exchanger is modelled. Some
importantcharacteristicsoftheobtainedmodelare:(1)itallowsadynamicstudyoftheheat
exchanger, (2) mass flow rate is a consequence of inlet and outlet pressure difference, (3)
fluidpropertieschangeistakenintoaccount,(4)itpresentsatimestepcontrolfunctionand
(5)fluidmovementisnotneglected.
Someinterestingnumericalbehavioursincludedinheatexchangersmodelsdesignthathave
beenobservedduringthecourseofthisthesisarediscussed.Forinstance,thecomparisonof
theeffectsofchoosingoneheattransferscorrelationoranother.
Dynamic response of the modelled heat exchanger during start up and during an abrupt
changeinmixedrefrigerantinlettemperatureareshownanddiscussed.

II

Preface
This Master thesis was completed at the Department of Energy and Process Engineering of
the Norwegian University of Science and Technology (NTNU) from February 2011 to June
2011.
First, I would like to thank my supervisor, NTNU professor Kjell Kolsaker, for his help and
guidelines during the course of this Master thesis. I specially appreciate that he has always
beenopenedtomyquestionsanddoubts,nomatterthetimeittooktosolvethem.
I would also like to thank to ivind Wilhelmsen, who belongs to SINTEF staff. His advices
during all the semester helped mea lot. Specially, at the end, when his help in structuring
thethesisbecamereallyvaluable.
Iwouldliketothankthiscountryanditspeople,NorwayandtheNorwegians,forgivingan
alldayexampleofthewelldonework.
I would also like to thank my friends, in Norway and abroad, because no thesis could give
whatafriendshipdoes.
I would like to thank my parents for the excellent examplethat they have always been, for
the unconditional love that they have always shown and for how fortunate they make me
feel.
Finally,IwouldliketothankNriabecauseshealwaysmakesthethingslookbrighter.
III

TableofContents
Abstract.......................................................................................................................................I
Preface........................................................................................................................................II
Listoffigures..............................................................................................................................V
Listoftables..............................................................................................................................VI
Nomenclaturetable.................................................................................................................VII
1 Backgroundandobjective..................................................................................................1
1.1 Introduction.................................................................................................................1
1.2 Aimofthethesis..........................................................................................................3
1.3 Descriptionofthethesis..............................................................................................4
2 Preliminaryconcepts..........................................................................................................5
2.1 Heatexchangers..........................................................................................................5
2.1.1 Heattransferandpressuredropcharacteristics.................................................5
2.2 Finitedifferencemethods.........................................................................................11
3 Calculations.......................................................................................................................14
3.1 Caseofstudy..............................................................................................................14
3.1.1 Counterflowheatexchanger:Naturalgasliquefaction....................................15
3.2 Themodel..................................................................................................................20
3.2.1 Procedure...........................................................................................................20
3.2.2 Structureandoperation.....................................................................................21
4 Resultsanddiscussion......................................................................................................32
4.1 Evaluationofthemodelrobustnessandtimeconsumption....................................32
4.1.1 Phasechangeathighpressure...........................................................................32
4.1.2 GnielinskiandDittusandBoelterheattransfercorrelationsnumericaleffects35
4.1.3 Advantagesofthevaryingrelaxingcoefficient..................................................37
4.1.4 Effectoftimesteplengthandgridmeshingonthenumericalstability............37
4.2 Caseresults................................................................................................................39
4.2.1 Thermaldynamicsduringthestartup...............................................................41
4.2.2 Thermaldynamicsduringanabruptchangeintheshellinlettemperature.....44
5 Conclusionsandrecommendationsforfurtherwork......................................................48
5.1 Conclusions................................................................................................................48
5.2 Recommendationsforfurtherwork..........................................................................49
IV

6 Bibliography......................................................................................................................50
7 Appendix...........................................................................................................................51

Listoffigures
Figure1TypicalBreakdownofLiquefactionPlantCapitalCost[1].........................................1
Figure2SeveraltypicalassumptionsimpactontheHEdesigndependingontheHE
effectiveness[2].........................................................................................................................2
Figure3Twophaseflowtypes.................................................................................................6
Figure4Geometricinterpretationofthedifferentapproximations.....................................12
Figure5Shellandtwotubesmultistreamheatexchangerscheme......................................14
Figure6Shellandtubeheatexchangerscheme...................................................................15
Figure7Naturalgasinputpressuredependenceontime....................................................17
Figure8Mixedrefrigerantinlettemperaturedependenceontime.....................................17
Figure9Mixedrefrigerantinletpressuredependenceontime...........................................18
Figure10Discretizedschemeoftheheatexchangermodelled............................................21
Figure11Modeloperationscheme.......................................................................................23
Figure12Tubesidecellenergybalancescheme.....................................................................25
Figure13Numericallystablesituation..................................................................................26
Figure14Numericallyunstablesituation..............................................................................26
Figure15Fixedcoefficientrelaxingsolutionscheme............................................................27
Figure16Behaviouroftherelaxingthesolutionfunctiondevelopedandusedinthemodel
..................................................................................................................................................28
Figure17Modeltimestepfunctionscheme.........................................................................30
Figure18Vapourfractionvs.Enthalpyat50bar...................................................................32
Figure19Heattransfercoefficientdependencevs.Enthalpyat50bar................................33
Figure20Vapourfractionvs.Enthalpyat54bar...................................................................33
Figure21Heattransfercoefficientvs.Enthalpyat54bar.....................................................34
Figure22Vapourfractionvs.Enthalpyat55bar...................................................................34
Figure23Heattransfercoefficientvs.Enthalpyat50bar.....................................................35
Figure24NumberofiterationsnecessarytoachieveconvergencewhenGnielinski
correlationisused....................................................................................................................36
Figure25NumberofiterationsnecessarytoachieveconvergencewhenDittusandBoelter
correlationisused....................................................................................................................36
Figure26Numberofiterationsnecessarytoachieveconvergencewhenafixedrelaxing
coefficientvalueisused...........................................................................................................37
Figure27Mixedrefrigerantinlettemperaturedependenceontime...................................39
Figure28Codeofcoloursusedwhenplottingdifferentcellsofasamestream..................39
Figure30Mixedrefrigeranttemperatureinseveralshellcells.............................................40
Figure31Naturalgasmassflowrateduringtime.................................................................41
Figure31Naturalgascellstemperaturesduringt<150s.......................................................42
Figure32Mixedrefrigerantcellstemperaturesduringt<150s............................................42
Figure43Naturalgascellstemperaturesduring500s<t<600s.........................................45
Figure44Mixedrefrigerantcellstemperaturesduring500s<t<600s..............................45

VI

Listoftables
Table1Naturalgasandmixedrefrigerantscomposition.....................................................15
Table2Tubeandshellgeometricaldata..............................................................................16
Table3Aluminiumproperties...............................................................................................16

VII

Nomenclaturetable
A
1
, A
2
, A
3
Friedelscorrelationconstants
J
h
hydraulicdiameter
frictionfactor
() polynomiallinkingfunction
Fr Froudenumber
g gravitationalacceleration
0 massflux
b heattransfercoefficient
E intensiveenthalpy
I length
m massflowrate
N totalcellsnumber
P pressure
q
ii
heatflux
Rc Reynoldsnumber
S surface
t time
: specificvolume
I volume
wc Webbernumber
x
c
thermodynamicequilibriumquality
z streamwisecoordinate
Greeksymbols
o voidfraction
[ channelaspectratio
z thermalconductivity
p viscosity
p density
o surfacetension
twophasepressuredropmultiplier
Superscripts
twophase
k timestep
Subscripts
o accelerational
c contraction
c expansion
fluid
g saturatedvapour
i cellposition
in channelinlet
l saturatedliquid
lom laminarflow
lo liquidonly
out channeloutlet
S Shell
sp singlephase
t tube
tot total
tp twophase
tronstransitionalflow
tur turbulentflow
w wall
1

1 Backgroundandobjective
1.1 Introduction
Therearealotofsituationswherenaturalgastransportationbypipelineisoverlyexpensive.
Oneofthemostcommonwaystotransportthegasinthesesituationsistoliquefyit.
Since the combinedcycle turbine power plants are getting more and more common, the
LiquefiedNaturalGas(LNG)demandwillkeepgrowing.Thereforeitisjustifiedtostudyand
develop all the necessary tools to reduce the cost, both operational and capital, of a LNG
plant.
LNGprojectsareinherentlycapitalintensive,withtheliquefactionprocessrepresentingthe
50% of total project cost [1] as shown in figure 1. The liquefier is the key area where a
processdesignercanmakelargestcostsavingsandinfluencetheprojectviability.

Figure1TypicalBreakdownofLiquefactionPlantCapitalCost[1]
Intheliquefactionprocessthemostimportantelementsaretheheatexchangers.Theserule
thebehaviouroftheprocessandrestricttheproductionandefficiencyoftheplantinevery
situation.
Due to their safe and costeffective designs and the need for higher effectiveness and
efficiency, MultiStream Heat Exchangers (MSHE) are preferred rather than the normal two
stream heat exchangers in many energyintensive industrial and cryogenic processes.
Naturalgasliquefactionisoneofthem.
A multistream heat exchanger is a single unit in which multiple hot and cold streams
exchangeheatsimultaneously.Plate&fin,spiralwoundandmultipassshellandtubearethe
mostcommontypesofMSHEs.ThisthesisreferstoamultipassshellandtubeMSHE,which
2

is the result of hundreds of thin shell and two tubes working in parallel. Usually, a low
pressurerefrigerantsuchasapureormulticomponentrefrigerantflowsdowntheshellside.
Thetubesideisoccupiedbythehotstreams;usuallyeachstreamfillsonetube.
Heat exchangers are the main components in cryogenic processes. In LNG plants they
represent 2030% of the investment cost [2, 3]. In addition, their performance affects the
sizinganddesignofotherequipment,namelycompressorsandtheirpowerdrivers.
It has been studied by several authors how thermodynamic considerations make cryogenic
processesverysensitivetotheHEperformance[2,4,5].
AcryogenicHEinaLNGplantisexpectedtohaveanefficiencyhigherthan90%.Thismeans
thatthedesignhastobereallyaccurateandithasbeennoticedthatseveraleffectsthatare
neglectedwhendesigningahightemperatureHEshouldnotbeneglectedwhendesigninga
HEforcryogenicapplications.
These nonnegligible effects are: changes in fluid properties, flow maldistribution,
longitudinal thermal conduction and heatinleakage. The consideration of this effects
dependontheeffectivenessthataHEisexpectedtoperform:

Figure2SeveraltypicalassumptionsimpactontheHEdesigndependingontheHEeffectiveness[2]
Thus,asithasbeenexplainedMSHEsplayanimportantroleintheoperationandthedesign
of a LNG plant. It is the engineers priority to find the way to improve their design and to
predicttheirperformance.Thereforelotsofheatexchangerssimulationsmodelshavebeen
developed in the last decades and are used extensively for both designing plants and
evaluatingtheactualperformanceofaplant.
However,overthelastyearsithasbeenseenthatthereisaneedtoimproveHEmodelsfor
cryogenic applications. This is specially the case when it comes to reducing the number of
3

assumptions of ideal conditions that are used. One of the reasons is the pursuing of more
energy, cost and space efficient plants. Features used to achieve this include among others
mixed refrigerant and more compact heat exchangers. These features will in many cases
lead to conditions where ideal assumptions are no longer valid. If this is not taken into
accountwhendesigningtheplant,itcouldgiveoperationalchallengesandaswellleadtoan
energyconsumptionthatishigherthananticipated.
Because of its own nature, MSHEs with phase change involve several physical phenomena
thatshouldbeconsideredwhenmodellingthem.
Characteristicsitisimportanttoaccountforinthedesignphaseinclude:
Distributionoftwophasefluidsintoseveralparallelchannels.
Instabilitiesinchannelswithevaporation/condensation.
Maldistributionoftwophasefluidswithinheatexchangerschannels.
Two phase non equilibrium conditions (both thermodynamic and fluid dynamic
effectslikeslip).
Heatexchangerdynamicbehaviour,forinstanceatstartupandshutdown.
Both steady state and dynamic heat exchanger models are widely used in the design
processes of heat exchangers and overall LNG production plants. Detailed plant operability
analysisrequiresgooddynamicheatexchangermodels.Astheplantcomplexityincreases,it
isseenthatthesekindofanalysisareimportantpartsoftheplantprocessaswell.
The heat exchanger models which are used are usually based on enthalpy balance
calculations or on heat transfer and pressure drop correlations. The latter method is
normallymoredetailedandgeometrydependentthanthefirstone.
The correlation based models will in addition to correlations, based on empirical results,
contain some kind of assumptions in order to simplify the heat exchanger modelling.
Assumptions of equal two phase distribution, stable flow conditions and thermodynamic
equilibriumarecommon.
SINTEF Energy Research is currently running a competence building project (KMB: Low
emissions LNG systems) with contribution from NTNU and industry partners, focusing on
bothfundamentalphysics,heatexchangerdesignandsystemoptimisation.
This thesis aims to produce a model capable of including part of the problematic here
explained.ThegoalistoproduceatoolthatallowsimprovedLNGprocessdesigns.
1.2 Aimofthethesis
TheobjectiveofthisthesisistofindoutifitispossibletodevelopaMSHEnumericalmodel
that includes most of the problematic already explained. This model should simulate a
multistreamheatexchangerworkingwithnaturalgas.Themodelshouldbeabletosimulate
4

satisfactorily phase changes in transient conditions. Minimum simplifications as possible

shouldbeassumedinfluidpropertiesfield.
Main model characteristics that will be appreciated are numerical stability, accuracy and
CPU time consumption. Specially, first one will be the most important since the other two
canbeimprovedlaterifastablemodelisdeveloped.
To achieve this conclusion, a simpler model will be developed. A counter flow two streams
shell and tube heat exchanger will be modelled. This model should be a tool that allows a
heatexchangerdesignerknowmoreabouttheLNGprocessdynamics.
Effortswillbeputinachievingaworkingnumericallystablemodel.Experimentalvalidityof
theresultsisnotthegoalofthisthesis,hence,therecanbesomeoftheuseddatathatmay
not be as accurate as desired in a final heat exchanger design. Besides, the optimisation of
themodelcanbedoneafterwardsinfurtherworkresearches.
1.3 Descriptionofthethesis
Inchapter2isanexplanationofthepreliminaryconceptsthatarenecessarytounderstand
thisthesis.Chapter2.1focusesontheheatexchanger,whatitis,whereitisusedandhowit
is designed. Since a heat exchanger model is developed in this thesis, chapter 2.1.1 goes
deeperinthedesignissueandtheheattransferandpressuredropcorrelationsusedinthe
model are explained there. Chapter 2.2 is a basic theory explanation of finite different
methodsanditsbasicconcepts.
Calculations belong to chapter 3. First, in chapter 3.1, the case modelled is explained in
detail. Chapter 3.2 is an explanation of the model solution methodology, the procedure to
achievethesatisfactorycodeanditsstructureandoperationaredescribedinit.
Chapter4consistsoftheresultsandobservationsthatareproductofthisthesis.Inchapter
4.1 some relevant numeric characteristics that have been observed during the model
development are presented and discussed. Chapter 4.2 is an illustration and discussion of
theresultsobtainedfromthecaseofstudy.
Chapter5summarizesthemostimportantandrelevantissuesthatareproductofthisthesis.
Afutureworkproposalisdoneinchapter5.2.
5

2 Preliminaryconcepts
2.1 Heatexchangers
Shah and Sekulic give the following heat exchanger definition [6]: A heat exchanger is a
device that is used for transfer of thermal energy (enthalpy) between two or more fluids,
between a solid surface and a fluid, or between solid particulates and a fluid, at differing
temperaturesandinthermalcontact,usuallywithoutexternalheatandworkinteractions.
Inmostheatexchangers,thefluidsareseparatedbyaheattransfersurface,andideallydo
not mix. A heat exchanger consists of a heat exchanging elements such as core or a matrix
containing the heat transfer surface, and fluid distribution elements such as headers,
manifolds,tanks,inletandoutletnozzles...Usually,therearenomovingparts.
Theheattransfersurfaceisasurfaceoftheheatexchangercorethatisindirectcontactwith
fluidsandthroughwhichheatistransferredbyconduction.
Thebalancedifferentialequationthatgovernsthefluidenthalpyvariationwhenthefluidis
incontactwiththeheattransfersurfaceisthefollowing:
poI
oE
ot
= m
oE
oz
+oq Eq. (1)
Wherethelasttermintherighthandoftheequationisthedifferentialheatflow,whichis
definedbythenextequation:
oq = b oS
],w
(I
w
-I
]
) Eq. (2)
Notice that the heat transfer coefficient is then the parameter which will define the heat
exchangervolumeinadesign,ortheonethatwillgovernitsefficiencywhenevaluatingthe
HEperformance.
AnotherimportantpartofaHEdesignisthepressuredropthatiscausedbythefluidgoing
throughit.Pressuredropwilldirectlydefinethecapitalandoperationalcostsoftheprocess
wheretheheatexchangerisused.Afluidthatsuffersahighpressuredropgoingthrougha
HEwillprovokehighcostsincompressorsorpumpsinvestmentandpowerconsumption.
Thus, heat transfer and pressure drop in heat exchanger are two important features that
which need a special attention of the designer. Therefore, chapter below gives a further
explanationofthemandofthewaytheyhavebeenmodelledinthisthesis.
2.1.1 Heattransferandpressuredropcharacteristics
Heat transfer and pressure drop inside a heat exchanger are two complex phenomena that
havebeenstudiedindecades.Researchershavebeenconcernedwithdevelopingpredictive
models for several industries, such as traditional ones like steam and nuclear power
6

generation, chemical and petroleum, etc. or newer ones like electronics and all micro
channelrefrigerationsystemsthatitincludes.
2.1.1.1 Twophaseflowtypes
It is known that pressure drop and heat transfer characteristics depend strongly on two
phase flow types. Depending on mass and heat flows and on vapour fraction, apart from
othervariables,twophaseflowtypesare:
Bubbleflow:Liquidfillsallthepipewhilesomesmallvapourbubblesareinamixture
withit.
Slugflow:Bigvapourbubblestravelthroughthepipeseparatedbyaliquidfilm.
Annularflow:Liquidcoversonlyathinfilmthroughthepipewallwhilevapourflows
throughaninsidechannelwithsomedropletsinit.

Figure3Twophaseflowtypes
Intheupperfigurethedifferentflowtypescanbeobserved.Churnandwispyannularflow
canbeconsideredmiddletermsbetweenslugandannularflow.
Twophasecorrelationsdescribedbelowbasetheircalculationsontheinfluenceofeachflow
typewhenaphasechangeoccurs.
2.1.1.2 Heattransfer
2.1.1.2.1 Single phase
Two popular single phase heat transfer coefficient correlations for turbulent flow (Rc
Suuu)inapipearetheonedevelopedbyDittusandBoelter[7]andGnielinski[8].
DittusandBoeltercorrelationis:
Nu = Nu
tu
= u.u2SRc
0.8
Pr
0.4
Eq. (S)
7

AndGnielinskicorrelationis:
Nu = Nu
tu
=

8
(Rc -1uuu) Pr
1 +12.7
_

8
(Pr
2
3
-1)
Eq. (4)
= (1.82log(Rc) -1.64)
-2
Eq. (S)
It is decided to use Gnielinski heat transfer correlation since it was noticed after some
simulations that Dittus and Boelter correlation introduced higher numerical oscillations in
thesystem.
Hausen[9]correlationwillbethesinglephaselaminarflowheattransfercorrelationusedin
themodel.So,for2Suu Rc:
Nu = Nu
Ium
= S.66 +
u.u668
J
h
I
Rc Pr
1 +u.u4 [
J
h
I
Rc Pr
2 3
Eq. (6)
IftheReynoldsnumberiswithintheupperlimitofHausencorrelationandthelowerlimitof
the Gnielinski correlation, Nusselt number will be calculated as an interpolation of both to
guaranteeNusseltcontinuity.For2Suu < Rc < Suuu:
Nu = intcrpolotion(Nu
Ium
(Rc), Nu
tu
(Rc), Rc) Eq. (7)
It is reminded that once the Nusselt number has been defined, heat transfer coefficient is
foundbythefollowingequation:
b
sp
=
Nu z
sp
J
h
Eq. (8)
2.1.1.2.2 Two phase
2.1.1.2.2.1 Evaporation
Liu and Winterton [10] correlation is used when calculating evaporation heat
transfer.
b
tp
= [(Eb
sp
)
2
+(Sb
nb
)
2

0.5
Eq. (9)
Where b
sp
is DittusBoelter single phase heat transfer coefficient for the liquid
(alreadyexplained),b
nb
isthenucleateboilingcoefficient:
b
nb
= SSP

0.12
(-log
10
(P

))
-0.55
H
w
-0.5
(q
ii
)
0.67
Eq. (1u)
AndEandSare:
8

E = _1 +x
c
Pr
I
_
:
g
:
I
-1]_
0.35
Eq. (11)
S = (1 +u.SSE
0.1
Rc
Io
0.16
)
-1
Eq. (12)
2.1.1.2.2.2 Condensation
There are less available condensation heat transfer correlations than evaporation
ones.Shahfilmcondensationcorrelation[11]hasbeenchosenhereforbeinganoldbroadly
testedcorrelation.
b
tp
= b
I
_(1 -x
c
)
0.8
+
S.8x
0.76
(1 -x
c
)
0.04
P

0.38
_ Eq. (1S)
Where x is the vapour quality, p

is reduced pressure P

= PP
ctcuI
and b
I
is
DittusBoeltersinglephaseheattransfercoefficientfortheliquid.
2.1.1.3 Pressuredrop
ConsideringthestudiesthatLeeandMudawardevelopedinevaporatorsfield[12],total
pressuredropinaheatexchangercanbeexpressedas:
P
tot
= P
c
+P
sp,pc
+(P
]
+P
u
)
tp
+P
sp,subsupc
-P
c
Eq. (14)
Pressure drop inside a heat exchanger where phase change is happening includes the
sudden contraction loss at the channels inlet and a sudden expansion recovery at the
channelsoutlet.However,thesetwotermsarenotincludedwhencalculatingpressuredrop
in this thesis because they are considered to be out of the study. A singlephase pressure
drop term is included corresponding to the flow before the phase change conditions are
achieved,meaningavapourprecoolingincondensersoraliquidpreheatinginevaporators.
The two phase pressure drop term consists of frictional and accelerational components.
Finally, another single phase flow term is included corresponding to subcooling or super
heating that can occur in condensers or evaporators, respectively, when phase change is
completed.
2.1.1.S.1 Twophase piessuie uiop mouels
2.1.1.3.1.1 HomogeneousEquilibriumModel(HEM)
Thehomogenousequilibriummodelisbasedontheassumptionthatthetwophasemixture
behaves as a pseudosingle phase fluid with mean properties weighted relative to vapour
and liquid content, and that only latent heat may be exchanged between the phases.
Propertyvariationsresultingfrompressurechangesalongthechannelresultincomplicated
terms that account kinetic energy, flashing and compressibility. The resulting pressure
gradientmaybeexpressedas:
9

-_
JP
Jz
]
tp
=
2
tp
J
h
0
2
(:
I
+ x
c
:
Ig
) +
4q
ii
0:
Ig
J
h
[b
Ig
+0
2
:
Ig
(x
c
:
g
+(1 -x
c
):
I
)
1 + 0
2
_x
c
J:
g
JP
+(1 -x
c
)
J:
I
JP
] _1 -
0
2
:
Ig
(x
c
:
g
+(1 -x
c
):
I
)
b
Ig
+0
2
:
Ig
(x
c
:
g
+(1 -x
c
):
I
)
_ - _
0
2
:
Ig
_x
c
JE
g
JP
+ (1 - x
c
)
JE
I
JP
]
b
Ig
+0
2
:
Ig
(x
c
:
g
+(1 - x
c
):
I
)
_
Eq. (1S)
The first term in the numerator of Eq. (13) is the frictional gradient and the second the
accelerational.Thedenominatorincludeskineticenergy,flashingandcompressibilityterms.
The twophase pressure drop can be determined by integrating Eq. (13) numerically along
thestreamwisedirection.
P
tp
= _ -_
JP
Jz
]
tp
L
tp
0
Eq. (16)
Thetwophasefrictionfactor
tp
isafunctionofthetwophaseReynoldsnumberRc
tp
:
Rc
tp
=
0J
h
p
tp
Eq. (17)
Thereareseveraltwophaseviscositymodels,McAdams[13]proposedthefollowing:
1
p
tp
=
x
c
p
g
+
(1 -x
c
)
p
I
Eq. (18)
2.1.1.3.1.2 SeparatedFlowModel(SFM)
In separated flow models gas and liquid are considered to flow apart from each other. In
these types of models accelerational and frictional terms are treated separately. The first
oneisexpressedintermsofpipeinletandoutlet:
P
u
= 0
2
(_
:
g
x
c,out
2
o
out
+
:
I
(1 -x
c,out
)
2
(1 -o
out
)
_ -_
:
g
x
c,n
2
o
n
+
:
I
(1 -x
c,n
)
2
(1 -o
n
)
_ Eq. (19)
WherethevoidfractionisdeterminedfromZivis[14]relation:
o = _1 +_
1 -x
c
x
c
] _
:
I
:
g
_
2
3
_
-1
Eq. (2u)
OneofthemostpopularfrictionalpressuredropcorrelationistheoneproposedbyFriedel
[15]:
P
]
=
2
Io
0
2
I
tp
:
I
J
h
x
c,out
_
Io
2
Jx
x
cut
x
in
Eq. (21)
10

Io
2
= A
1
+
S.24A
2
A
3
Fr
tp
0.045
wc
tp
0.035
Eq. (22)
A
1
= (1 -x
c
)
2
+x
c
2
_
p
I

go
p
g

Io
_ A
2
= x
c
0.78
(1 -x
c
)
0.224
A
3
= _
p
I
p
g
_
0.91
_
p
g
p
I
]
0.19
_1 -
p
g
p
I
]
0.7
Eqs. (2S,24,2S)
Fr
tp
=
0
2
gJ
h
p
2
wc
tp
=
0
2
J
h
p o
p =
1
x
c
:
g
+(1 -x
c
):
I
Eqs. (26,27,28)
It is decided to use HEM pressure drop correlation as the default one. SFM with Friedel
correlationmaybeusedtocompareresultsifnecessary.
2.1.1.S.2 Singlephase piessuie uiop mouel
Followingrelationisusedtodeterminepressuredropinsinglephaseflow[16,17]:
P
sp
=
2I
sp
J
h

sp
0
2
: Eq. (29)
The single phase friction factor
sp
has different values depending on the flow regime.Two
linkinginterpolationfunctionshavebeendefinedtoguaranteefunctioncontinuity.
For Rc 19Su:

sp
=
Ium
=
24
Rc
(1 -1.SSSS[ +1.9467[
2
-1.7u12[
3
+u.9S64[
4
-u.2SS7[
5
) Eq. (Su)
Where[isthechannelaspectratio,whichisequaltooneforacircularcrosssectionpipe
andthen:

Ium
=
14.2296
Rc
Eq. (S1)
For19Su < Rc < 2uSu

sp
= intcrpolotion(
Ium
(Rc),
tuns
(Rc), Rc) Eq. (S2)
For 2uSu Rc 19uSu:

sp
=
tuns
= u.u79Rc
-0.25
Eq. (SS)
For19uSu < Rc < 2uuSu

sp
= intcrpolotion(
tuns
(Rc),
tu
(Rc), Rc) Eq. (S4)
For 2uuSu < Rc

sp
=
tu
= u.u46Rc
-0.2
Eq. (SS)
11

2.2 Finitedifferencemethods
Inthischapterdifferentbasicfinitedifferencemethodsareexplained.
Theprincipleofafinitedifferencemethodisthatthederivativesinthepartialdifferential
equationareapproximatedtolinearcombinationsoffunctionvaluesatthegridpoints.So,
havingadomainandagridlikethefollowing:
o < x < b t > u
x

= o +ib, i = u, 1, , N, b =
b -o
N

t
k
= kt, k = u, 1, ,
Thefunctionvaluesforeachgridpointwillbe:
u

= u(x

)
Firstorderderivativecanbedefinedwithanyofthesethreeequalities:
ou
ox
(x ) = lim
h-0
_
u(x +b) -u(x )
b
_ = lim
h-0
_
u(x ) -u(x -b)
b
_

= lim
h-0
_
u(x +b) -2u(x ) -u(x -b)
2b
_ Eqs. (S6,S7,S8)

Resultingfromthatdefinition,threefirstderivativeapproximationscanbedone:
Theforwardfinitedifference

+
u(x ) =
u(x +b) -u(x )
b
= u
i
(x ) +
b
2
u
ii
() Eq. (S9)
Thereforetheabsoluteerroroftheforwardfinitedifferenceisproportionaltob
andtheapproximationisreferredasafirstorderapproximation.
Thebackwardfinitedifference

-
u(x ) =
u(x ) -u(x -b)
b
= u
i
(x ) -
b
2
u
ii
() Eq. (4u)
Whichisagainafirstorderapproximation.
Thecentralfinitedifference
12

0
u(x ) =
u(x +b) -2u(x ) -u(x -b)
2b
= u
i
(x ) +
b
2
6
u'''() Eq. (41)
Theerrorisproportionaltob
2
sotheapproximationisreferredasasecondorder
approximation.Notethat:

0
u(x ) =
1
2
(
+
u(x ) +
-
u(x )) Eq. (42)

Figure4Geometricinterpretationofthedifferentapproximations
Approximationtohighordersderivativescanbeobtainedfromtheformulasforlowerorder
derivatives.Thecentralfinitedifferenceschemeforthesecondorderderivativeu
xx
is:

0
2
u(x ) =
+

-
u(x ) =
+
u(x ) -u(x -b)
b
=
u(x -b) -2u(x ) +u(x -b)
b
2

= u
xx
(x ) +
b
2
12
u
(4)
() Eq. (4S)
Soitisasecondorderapproximation.
Theonedimensionheatequationwithasourceisanequationofinterestinthisthesis.Itisa
parabolicpartialdifferentialequation:
ou
ot
=
o
ox
_[
ou
ox
] +(x, t) Eq. (44)
If [ is a constant value and the partial derivatives are presented in their shorter way, the
equationcanbewrittenlikethis:
u
t
= [u
xx
+(x, t) Eq. (4S)
TheforwardEulermethod(forwardintime,centralinspace)is:
u

k+1
-u

k
t
= [
u
-1
k
-2u

k
+u
+1
k
b
2
+

k
Eq. (46)
13

k = u, 1, , i = 1, 2, , N -1,
wheretheboundaryconditionsu(o, t)andu(b, t)aregiven,
andwhereu

k
isanapproximationtou(x

, t
k
)and

k
= (x

, t
k
).
Themaincharacteristicofthisschemeisthatthereisonlyoneunknownvaluewhichisthe
function value in the next time step for this grid point u

k+1
. That means that each new
functionvalueineachgridpointcanbecalculatedapartfromtheothers.
However, to guarantee stability in a forward scheme the t should satisfy the time step
restriction:
u < t <
b
2
2[

This time step restriction can imply a big disadvantage when talking about computational
timeconsuming.
The truncation error of this method is first order in time and second order in space
0(t +b
2
).
On the other hand, the same equation written in backward difference scheme (also called
implicitscheme)is:
u

k+1
-u

k
t
= [
u
-1
k+1
-2u

k+1
+u
+1
k+1
b
2
+

k+1
Eq. (47)
In an implicit scheme the only known value is u

k
so a linear equation system needs to be
solvedforeverynewtimestep.
The truncation error is also 0(t +b
2
). The good point of this backward Euler method
schemeisunconditionallystable,meaningthattcouldtakeanyvalue.
Comparisonbetweenthesemethods
The explicit or forward finite difference scheme it is usually easier to implement because it
goesoneequationbyoneequationandthatmakesitsimplytodevelop.However,thetime
step restriction is a hard disadvantage because it can cause high computational resources
consumption.
Theimplicitorbackwardfinitedifferenceschemehasthereallyattractivepropertyofbeing
always stable, it does not matter which is the value of time step. That property gives the
model a really robust behaviour which is a valuable characteristic when programming.
However, this scheme is more difficult to implement because of the matrix system that
needstobecreated.
14

3 Calculations
3.1 Caseofstudy
ThisthesisfocusesinashellandtubemultistreamLNGheatexchanger.Specifically,onlyits
basic element is going to be studied. A multipass shell and tube MSHE is composed of
hundredsofshelltubeswhileeachofthemhasastubesinsideasstreamsareleft.

Figure5Shellandtwotubesmultistreamheatexchangerscheme

Since the dynamics of a multistream shell and tubes heat exchanger is a complex
phenomenon to model, a simpler case will be studied in this thesis. The dynamics of a
counterflowshellandtubeheatexchangerhavebeenmodelled.Inthetubesidenaturalgas
isliquefiedwhileintheshellsidemixedrefrigerantvaporizes.Althoughitisasimplercase,it
hasprovedtobeaconsiderablechallenge.
15

Figure6Shellandtubeheatexchangerscheme
Sinceflowmaldistributionispartoftheheatexchangersproblematicwherebetterdesigning
toolsareneeded,massflowinthismodelisnotafixedvaluebutdependsonthedifference
betweentheinletandoutletpressures.
Anotherimportantissuethathastobetakenintoaccounttoachievebetterheatexchanger
designs is fluid properties dependence on state. In the model, fluid properties are given by
TP_library, which is a thermodynamic package created by SINTEF Refrigeration Engineering
whichreturnsitsuserseveralfluidpropertiesforacertainthermodynamicstate.
3.1.1 Counterflowheatexchanger:Naturalgasliquefaction
Inthiscasethenaturalgasphasechangefromliquidtogasismodelled.
3.1.1.1 Generaldata
Thecompositionofthenaturalgasusedinthemodelisthefollowing:
Table1Naturalgasandmixedrefrigerantscomposition
Naturalgas Mixedrefrigerant
MolefractionCH
4
0.94480 0.29130
MolefractionC
2
H
6
0.00580 0.38870
MolefractionC
3
H
8
0.01890 0
MolefractionnC
4
H
10
0.00110 0.22710
MolefractionN
2
0.02950 0.09290
Criticalpressure[bar] 56 44.16

16

SincethroughTP_librarycriticalpressuresarenotdirectlyobtainable,naturalgasandmixed
refrigerant critical pressures have been set at these approximate values. When accurate
resultsaredesired,thesevaluesshouldbecheckedandreplacedifnecessary.
Thegeometricaldataoftheheatexchangeris:
Table2Tubeandshellgeometricaldata
Tube Shell
Innerdiameter[mm] 4 10
Roughness[m] 1 1
Length[m] 2

Thetubematerialisaluminium.Itspropertiesareconsideredtobeconstantandare:
Table3Aluminiumproperties
Density[kg/m
3
] 2700
Intensiveheatcapacitance[J/kgK] 930.77

Aluminiumheatconductivityisnotdisplayedbecauseitisassumedtobeinfinite.
3.1.1.2 Startingconditions
Whenthesimulationbegins,thenaturalgasinthetubesideisconsideredtobeinauniform
superheated vapour state at 298K and 49.2 bar. Tube wall temperature at the t = us is
297K through all its length. Finally, mixed refrigerant starting conditions are superheated
vapourat296Kand4.7barthroughalltheheatexchangerlength.
3.1.1.3 Boundaryconditions
S.1.1.S.1 Tube siue
Natural gas inlet enthalpy is the one corresponding to superheated vapour at 298K and it
keeps constant during all the simulation time. Outlet pressure is kept constant at 49.2 bar
whileinletpressureraisesfrom50att = ustill55baratt = Sus.
17

Figure7Naturalgasinputpressuredependenceontime
S.1.1.S.2 Shell siue
Mixed refrigerant inlet enthalpy at t = us is the one corresponding to superheated vapour
at296Kbutitdecreasestillbeingsubcooledliquidat110Katt = 6us.
Once steady state conditions have been reached, mixed refrigerant inlet temperature is
raised again to observe the thermal dynamic behaviour of the heat exchanger. Inlet
temperature raises from 110K at t = Suus to 250K at t = S2us where is maintained till
t = S4usanddecreasedagainto110Katt = S6us.

Figure8Mixedrefrigerantinlettemperaturedependenceontime

18

Inletandoutletpressuresfollowananalogouslawtonaturalgasonebutwithlowervalues.

Figure9Mixedrefrigerantinletpressuredependenceontime
3.1.1.4 Heattransfercoefficientandpressuredropgradientdefinition
Thecorrelationsselectedtobeusedinthemodelweredescribedinchapter6butthefinal
definition through that makes them continuous through all fluid states has yet to be done.
Single phase and two phase correlations need to be linked by some smooth functions,
hence, there have to be defined some vapour fractions limits that these functions will link.
Here, the lower limit is vapour fraction equal to 0.3 and the upper limit is vapour fraction
equalto0.7.
Equationsthatdefinetheheattransfercoefficientarethefollowing:

Analogously,equationsthatdefinepressuredropgradientare:

19

3.1.1.5 Simulationaccuracy
Thegridconsistsof40cellsandtheconvergencecriterionisequalto10
3
.
20

3.2 Themodel
3.2.1 Procedure
Duringtheperiodthisthesishasbeendeveloped,thesameprocedurehasbeenfollowedto
create and evaluate different models. When developing a model, it has been applied the
from simple to complex rule, that means that first the whole structure of the model was
built without taking into account characteristics that could introduce more instabilities. So
first a pipe with fixed and defined wall temperature was modelled where fluid properties,
heattransfercoefficientandpressuregradientwereconsideredtobeconstant.Andstepby
step,morecomplexitywasintroduced.
Once the pipe model proved to work correctly with constant fluid properties, real
thermodynamics were included by the use of TP_library. So, first the model was checked
withpurewater,becauseofthebetterknowledgeofthethermodynamicalbehaviourofthis
fluid.Thenthemodelwascheckedwithpuremethaneastheworkingfluidwhichhasmuch
more similarities with natural gas but it is still a pure component. Finally, the model was
tested with natural gas which as a multicomponent fluid with the effects on its properties
thatcarriesrepresentedahigherchallenge.
When real fluid properties had been tested satisfactorily in the model, heat transfer and
pressure drop correlations were included. First, heat transfer coefficient correlation was
included while pressure drop was set constant or following a linear rule with mass flow.
Thentheoppositesituationwasfollowed,pressuredropcorrelationwasincludedwhileheat
transfercoefficientjustfollowedasimplerule.Finally,bothcorrelationswereincluded.
At the end, once the pipe model had proved to give satisfactory results, the counter flow
heatexchangerwasmodelledfollowingthesamecalculationstructureasinthepipemodel.

21

3.2.2 Structureandoperation
The model developed in this thesis allows simulating the dynamics of a counter flow heat
exchanger where the two fluids suffer a phase change. In this model the mass flow of a
streamisnotdefinedbutdependsonthedifferenceoftheinletandoutletpressures.

Figure10Discretizedschemeoftheheatexchangermodelled
TheprogramhasbeenwritteninMATLAB.
Inthismodelthestateofthefluidissetthroughitsenthalpyandpressure.Othervariables
liketemperatureandvapourfractionbecomeaconsequenceofthem.Thestateofthewall
isdefinedbyitstemperature.
Someassumptionsaretakeninthismodel:
Stableflow.
Homogenousflowregime.
Thermodynamicequilibrium.
Nonaxialheatconduction.
Infinitewallthermalconductivity(walltemperatureisconsideredtobethesameon
bothsidesofonewallcell).
Therearesomeimportantcharacteristicsofthefinalandsatisfactorymodeldevelopedthat
shouldbementioned.Firstoneisthatenthalpiesandwalltemperaturescalculationsfollow
22

an implicit scheme. The second one is that it presents a varying time step (between
established limits) depending on time derivatives. Also, a function that governs the
weighting coefficient value in solution relaxation has been implemented. Further
explanationofalltheseisgiveninthischapter.
Theoperationofthemodelfollowstheschemebelow:
1. Massflowandpressurescalculationofbothshellandtubestreams.
2. Enthalpiescalculationofbothshellandtubestreams.
3. Convergencecheckingonthestreamsobtainedsolutions.
4. Relaxingstreamssolutions.
5. Other streams variables calculation (temperatures, vapour fractions, heat flows,
etc.). A flash call gives the new temperatures and new vapour fractions, then, new
heattransfercoefficientsandnewheatflowscanbefound.
6. Newwalltemperaturescalculation.
7. Convergencecheckingonthewallobtainedsolution.
8. Relaxingwallsolution.
9. If both streams and wall convergences were achieved, results are stored and new
time step calculations will start. If convergence was not achieved, a new iteration
startsagainwiththerecentcalculatedvalues.
23

Figure11Modeloperationscheme
24

It is asked the reader to note that it is an iterating scheme and that, during calculations
between time steps t
k
and t
k+1
, as many iterations as necessary will be done till
convergenceisreachedforallcells.
3.2.2.1 Pressuresandmassflowcalculation
The total stream pressure drop is the result of the integration of each differential pressure
dropalongthefluiddirection:
P
1otuI
= _
oP
oz
Jz
L
0
Eq. (Su)
Whendiscretized,thetotalpressuredropbecomesasumofeachcellpressuredrop.Taking
intoaccountthattheothervariablesarekeptconstantduringthiscalculation,pressuredrop
dependsonlyonthemassflow.
P
1otuI
= P

(m )
N
=1
Eq. (S1)
A combination of different numerical methods is used to find the mass flow rate which
corresponds to the boundary pressure drop. The secant method, the false position method
and the halving intervals method are used here. First, a couple of iterations are done using
the secant method to find the upper and the lower limits for the false position method.
Then,acertainnumberofiterationsaredonewiththefalsepositionmethodusingitsrobust
but also fast characteristics. If convergence is not achieved after this certain number of
iterations, the secant method is used to find the mass flow rate. The halving intervals
methodisusedonlywhenconvergencehasnotbeenreachedbytheothertwomethods.
3.2.2.2 Enthalpiescalculation
Thefollowingbalanceequationneedstobesolvedinthemodelforeachstream:
poI
oE
ot
= m
oE
oz
+oq Eq. (S2)
Aschemeoftheenergybalanceforeachcellinthetubeisthefollowing:
25

Figure12Tubesidecellenergybalancescheme
Whendiscretized,theimplicitschemeformoftheEq.50is:
p

k+1
I
E

k+1
-E

k
t
= m
k+1
(E
-1
k+1
-E

k+1
) +q

k+1
Eq. (SS)
Wheresuperscriptsdenotetimeandsubscriptsdenoteposition.
However,heatflowdependsonseveralvariables(temperature,vapourfraction,...)thatwill
befoundwiththeflashcall,whichdependagainontheenthalpy.So,heatflowstakevalues
fromlastiterationbecauseisnotpossibletodirectlyisolatetheenthalpy.
Asithasbeenexplainedinthetheorypart,animplicitschemerequiressolvingasystemof
equations.Thelinearsystemthathereneedstobesolvedis:

MATLAB presents a really powerful characteristics when operating with matrices, so apart
from the robust characteristics of the implicit scheme itself, the program gives an extra
advantageinthispartofthemodel.
26

3.2.2.3 Solutionrelaxation
Oncepressuresandenthalpieshavebeencalculated,newstateshavebeenfound.However,
sometimes these results would lead to numerical instabilities. So new states cannot always
be the results obtained in the calculation. In some cases it is needed to do a weighting
relativebetweennewnewstateresultsandtheoldnewstateresults,thisiscalledrelaxing
thesolution.
y = X y
ncw
+(1 -X) y
oId
u < X 1 Eq. (SS)
Hereydenotesstateandisanenthalpyandpressurecouple,istheresultofaniteration.X
istheparameterusedtodotheweighting.InthismodeltheparameterXishasnotafixed
valuebutdependsonthecalculation.
When iterating, there are two situations types of situations that could occur. First type is
thateachiterationgivesanewstatethatitisclosertotherealsolutionthantheonebefore,
then it is a stable situation. Second type is that each iteration gives a new state that is
further from real solution than the one before. Both types are shown in the following
figures:

Figure13Numericallystablesituation

Figure14Numericallyunstablesituation
27

Theusualstrategywhenrelaxingthesolutionistofixthevalueoftheweightingcoefficient
for all the situations. If the coefficient is low enough and/or the convergence criterion is
wide enough, that brings to a stable situation. However, this strategy is not optimal when
thesituationisstablefromthebeginning.Inastablesituation,aweightingcoefficientvalue
lower than one will lead to a higher number of the necessary iterations to achieve
convergencethanwhenthevalueofitisone.

Figure15Fixedcoefficientrelaxingsolutionscheme
In the model developed in this thesis the coefficient has not a constant value but changes
depending on the results of iteration. Actually, every new time step the coefficient starts
havingavalueequaltoX
0
.ThevalueofX
0
issetbytheuserandcannotbehigherthan1.In
thesimulationwhichresultsareshowedinthisthesis,thisstartingvalueis:
X
0
= u.S
After each iteration, results are checked and the coefficient is decreased or kept constant
dependingonthem.Thelogicalrulethatgovernsthecoefficientdecreasingisbasedonthe
followingtwocharacteristicsofanunstablesituation(seefigure8):
1. Absolutedifferencebetweenstatesraises:
|y
2
-y
1
| < |y
3
-y
2
|
2. Differencebetweenstateschangesitsigneveryiteration:
y
2
-y
1
> u y
3
-y
2
< u
Ifanunstablesituationisdetected,theweightingcoefficientisreduced:
X
ncw
= u.9S X
Following that methodology the solution is relaxed only when need, so a fastest and stable
solutionisachieved:
28

Figure16Behaviouroftherelaxingthesolutionfunctiondevelopedandusedinthemodel
Theweightingcoefficienthasaminimumvaluethatissetbytheuser.
As it has been explained, fluids states are set by a pressureenthalpy couple, and there are
twostreamsandseveralcellsforeachstream.Therelaxingcoefficientusedineachiteration
tocalculatenewstateswillhavethesamevalueforallcellsandinbothstreams,thismeans
that if one cell is found to be in an unstable situation the relaxing coefficient value will
decreaseforallofthem.
3.2.2.4 Flashcall
Once the solution relaxation ends, pressures and enthalpies values have been definitely
established,whichmeansthatnewfluidstateshavebeenfound.Laststepistofindtherest
of the variables that are basic for the calculus or may be interesting for the analysis. These
variables are: temperature, vapour fraction, heat flow, heat transfer coefficient, etc.
Temperature,vapourfractionandliquidandvapourmolarcompositionsarefoundthrough
theTP_librarypressureenthalpyflashcall.
Itcannumericallyhappenthatenthalpyvalueislowenoughthatitsrelatedtemperatureis
0K, if that happens all the fluid properties give false values and the calculus of the model
crash.Therefore,aminimumoutputtemperaturewasset,iftheflashoutputtemperatureis
lowerthan60Kthecellisconsideredtobeinliquidstateat60K.
Similar situation happens when temperature rises above 1000K, so when that happens the
cellisconsideredtobeinvapourstateat1000K.
Apartfromflashoutputvariables,heatflowneedstobecalculated.Thedifferentialformof
theheatflowis:
oq = b oS
],w
(I
w
-I
]
) Eq. (S6)
Here b = b(E, P, ) is the convection coefficient, oS is the differential fluidwall contact
surfaceandI
w
andI
]
arethewallandthefluidtemperature.
29

Whendiscretized:
q

k
= b

k
S
ccII
(I
w
k
-I

k
) Eq. (S7)
The convection coefficient is calculated by an internal function which includes all the
necessary correlations which are linked by the pertinent linking functions. An important
characteristic that has to be mentioned is that whether a condensing or an evaporating
correlation is used in the two phase area depends only on the inlet conditions. If the fluid
enterstheheatexchangerasliquid,thenanevaporatingcorrelationwillbeused.Ifthefluid
enters the heat exchanger as vapour, then a condensing correlation will be used. That is
mentioned because, in a cross flow heat exchanger, heat flow direction could swap locally,
whichcouldmeanavapourfractionincreasingofatwophasemixturethatwasexpectedto
condense or, a vapour fraction decreasing of a two phase mixture that was expected to
evaporate.
3.2.2.5 Timestepfunction
Timestepinthemodelisnotconstant,itischangeddependingontimederivativesofsome
of the values of the variables in it. These variables are pressure, enthalpy, mass flow and
temperatures. Although the first two are enough to define a state, the other ones have
proved to give an accuracy increasing when taking into account. A new time step is set
considering that a variable cannot change its value more than a fraction that is set by the
user. Then, considering this fraction and the variable time derivative, the maximum time
stepforeachvariableisfoundandtheminimumofallischosen:
t
P,mux
=
y P

k
JP
Jt
, t
H,mux
=
y E

k
JE
Jt
, t
1,mux
=
y I

k
JI
Jt
, t
m ,mux
=
y m
k
Jm
Jt

t = min(t
P,mux
, t
H,mux
, t
1,mux
, t
m ,mux
)
Minimum and maximum time step values are set by the user in order to prevent
unreasonablyshortorlongtimesteps.
Since pressure, enthalpy and temperature values depend on the cell, a previous cell
maximumtimederivativeselectionhastobedone.
Timestepiscalculatedwhennewtimederivativesarenotstillknown,solastvaluesareused
to calculate them. However, real new time derivatives may be different than the ones
before,sovariablechangeswillbedifferentthanpredicted.
30

Figure17Modeltimestepfunctionscheme
Although time derivatives can be different than predicted no instabilities will be provoked
becauseofthat.Infact,theobjectiveofthisfunctiontodetailedresultsduringdynamicsbut
allowingfastresultswhensteadystateisreached.
3.2.2.6 Linkingfunctions
When heat transfer coefficient and pressure drop correlations are used, a link has to be
created between the single phase and the two phase areas in order to guarantee function
continuity and, then, numerical stability. It was also observed that just continuity was not
alwaysenoughtoguaranteestabilityandthatitsderivativeneededtobealsocontinuous.In
order to achieve that objective, linking polynomial functions are used in that case. The
degree of these polynomials is three and they achieve the continuity of both the function
anditsderivative.
Thefirstofthesetwosmoothfunctionslinktheliquidsinglephaseareawiththe0.3vapour
qualitytwophasearea.Thesecondsmoothfunctionlinksthe0.7vapourqualitytwophase
area with all vapour single phase area. These relatively wide ranges of vapour qualities are
neededtoachievethedesiredcharacteristicsofthefunctions.
3.2.2.7 Wallenergybalance
Consideringthatitisassumedthatwalltemperatureisequalonbothfacesofthepipeinthe
samecell,theenergybalanceequationthatgovernsthewalltemperatureisthefollowing:
p
w
oIc
p,w
JI
w
Jt
= b
]1
oS
]1,w
(I
]1
-I
w
) +b
]2
oS
]2,w
(I
]2
-I
w
) Eq. (S8)
Wherewdenoteswalland1and2denoteshellandtubefluidsrespectively.
WhenEq.(50)isdiscretizedfollowingtheimplicitscheme:
31

p
w
I
w,ccII
c
p,w
I
w,
k+1
-I
w,
k
t
= b
]1
k+1
S
]1,w
(I
]1
k+1
-I
w,
k+1
) +b
]2
k+1
S
]2,w
(I
]2
k+1
-I
w,
k+1
) Eq. (S7)
Walltemperatureoftimestept
k+1
canbeisolatedfromEq.(51):
I
w,
k+1
=
b
]1
k+1
S
]1,w
I
]1
k+1
+b
]2
k+1
S
]2,w
I
]2
k+1
+
p
w
I
w,ccII
c
p,w
t
I
w,
k
b
]1
k+1
S
]1,w
+b
]2
k+1
S
]2,w
+
p
w
I
w,ccII
c
p,w
t
Eq. (S9)

32

4 Resultsanddiscussion
4.1 Evaluationofthemodelrobustnessandtimeconsumption
Themainobjectiveofthisthesiswastocreateanumericallystablemodel.Therefore,allthe
functions used in it have been analysed. Linking functions have been created when it has
beennecessarytoavoiddiscontinuitiesthatcouldcarrynumericalinstabilities.Thisstudyof
thenumericalbehaviourofthemodelhasrevealedsomefactsthataredescribedbelow.
4.1.1 Phasechangeathighpressure
SinceLNGisproducedathighpressuresjustbelowitscriticalpoint,whichinthisthesishas
beensetat56bar,itwasfirstdecidedthatfinalinletandoutletpressuresduringsimulations
wouldbe55and54.2barrespectively.However,thesesimulationsbroughtalwaysa
numericalinstabilitywhoseoriginwasfinallyfoundandisexplainedhere.
It has been observed that the natural gas phase change at high pressure is a phenomenon
where vapour fraction varies abruptly with enthalpy. Besides, it has been noticed that it
could be considered a function discontinuity when pressure is high enough. That
discontinuity has severe consequences on other functions of the model, like heat transfer
coefficientone,andthatprovokesanirresolvablenumericalinstability.
Thefollowingfiguresshowthevapourfractionandheattransfercoefficientdependenceon
enthalpy at different pressures. Distance between dots in the phase change is 100 kJ/kg
whichisashortintervalconsideringtheenthalpymagnitude.
Resultswhenpressureisequalto50bar:

Figure18Vapourfractionvs.Enthalpyat50bar
33

Figure19Heattransfercoefficientdependencevs.Enthalpyat50bar
Results when pressure is equal to 54bar (here, dots enthalpy interval when u < x
c
< u.1
wassetequalto1kJ/kgtoassurethattherewasadiscontinuity):

Figure20Vapourfractionvs.Enthalpyat54bar
34

Figure21Heattransfercoefficientvs.Enthalpyat54bar
Results when pressure is equal to 55bar (here, dots enthalpy interval when u < x
c
< u.6
wassetequalto1kJ/kgtoassurethattherewasadiscontinuity):

Figure22Vapourfractionvs.Enthalpyat55bar
35

Figure23Heattransfercoefficientvs.Enthalpyat55bar
Noticethatat54baritalreadyappearsadiscontinuityatlowvapourfractions.However,at
55barthisdiscontinuitygoesfromx
c
= utox
c
= u.S8,andthatprovokesthatheattransfer
suffers also a discontinuity that goes almost from its minimum value to its highest value,
whichwouldbeafocusofenormousnumericalinstabilities.
Sincevapourfractioncorrespondencewithenthalpydependsontheflashcall,andthisone
depends itself on the TP_library, there is no possibility of avoiding this vapour fraction
discontinuity without losing all the advantages and flexibility that TP_library allows.
Therefore, it is decided to reduce the final inlet pressure to 50 bar, which is also a high
pressurewhereinphasechangevariablespresentabruptlyvariationsbutiscontinuous.
4.1.2 GnielinskiandDittusandBoelterheattransfercorrelationsnumerical
effects
Another relevant observation that has been noticed during simulations is that correlation
election can have an important impact on the numerical oscillations created during
calculations,hence,onCPUtimeconsumption.
Two simulations were run to compare Gnielinski and Dittus and Boelter heat transfer
correlations numerical effects. All the parameters are the same as described in chapter 8
except from some accuracy parameters that have reduced to increase simulation speed.
Thesechangesare:
Grid:Thecellsnumberhasbeenreducedto20.
Convergencecriterionissetat10
3
.
Timestepfunction:t
mn
= 1s.
36

So, once these modifications were done, both simulations were run until t = 4us. The
following figures show the number of iterations necessary to solve each time step for each
simulation:

Figure24NumberofiterationsnecessarytoachieveconvergencewhenGnielinskicorrelationisused

Figure25NumberofiterationsnecessarytoachieveconvergencewhenDittusandBoeltercorrelationisused
ThemeannumberofiterationsusedinthesimulationwhereGnielinskicorrelationhasbeen
used is 15,175 iterations per simulated second, while in the one where Dittus and Boelter
37

hasbeenusedis43,125.BothcorrelationswouldbevalidtouseinthemodelbutGnielinski
correlationprovedtogiveafastersolution,sothisisthecorrelationusedinthemodel.
4.1.3 Advantagesofthevaryingrelaxingcoefficient
Sinceafunctionwasdevelopedjusttocontroltherelaxingcoefficientvalue,itwasnecessary
toproveitsusefulness.
Same simulation parameters as the ones used to compare correlations were used in a
simulationwithafixedrelaxingcoefficientvalue(andusingGnielinskicorrelation).
The relaxing coefficient was fixed at 0.2 and the number of iterations necessary to reach
convergenceeachtimeisshowedinthefollowingplot:

Figure26Numberofiterationsnecessarytoachieveconvergencewhenafixedrelaxingcoefficientvalueisused
The mean number of iterations when the coefficient is constant is 27,6 iterations per
simulated second, while when it varies is only 15,125 (as showed in figure 24). So, the
functiondevelopedtocontrolthevalueofthiscoefficientprovestogiveafastersolution.
4.1.4 Effectoftimesteplengthandgridmeshingonthenumericalstability
In linear numerical implicit scheme where parameters in it have a constant value like ones
explainedinchapter7,timesteplengthdoesnotjeopardizenumericalstability.Asexplained
in chapter 7, only the explicit scheme is under a time step length restriction to avoid
instabilities.
However,themodeldevelopedinthisthesisisafollowsanonlinearimplicitscheme.Lotsof
parameters vary with the state (like fluid properties). As a consequence, it has been
observedthattimesteplengthdoescompromisethenumericalstabilityofthemodel.Asan
38

example,samesimulationastheoneheresimulatedwithaminimumtimestepof5sinstead
of2sanditconductedtoanumericalinstabilitythatmeantthefailureofthesimulation.
Similar consequence has been observed when talking about grid meshing. Same simulation
runwith20cellsasgridnumber,insteadof30,producedaseverenumericalinstability.
Itisreasonabletothinkthatbigcellsunderbigchangeswillproducemoreoscillations,and
sometimesinstabilities,thansmallercellsundersmallerchanges.

39

4.2 Caseresults
Thischaptershowtheobtainedresultswhensimulatingthecasepresentedinchapter3.1.
Case of study boundary conditions were set in order to study the behaviour of the heat
exchanger during start up conditions and during an abrupt change in the inlet mixed
refrigeranttemperature.
Itisremindedthatduringthestartup,bothmixedrefrigerantandnaturalgasinletpressures
increase till t = 6us, and from then on are kept at a constant value. Natural gas inlet
temperatureiskeptconstantat298Kduringallsimulationtime.However,mixedrefrigerant
inlettemperaturefollowslawdependingontimethatisshowedinthefigurebelow:

Figure27Mixedrefrigerantinlettemperaturedependenceontime
Whentheresultsofastreamareshownforseveralcells,acodeofcoloursisusedtoknow
thepositionofthesecells.Thiscodeistheonerepresentedinthenextfigure:

Figure28Codeofcoloursusedwhenplottingdifferentcellsofasamestream

40

Figure29Mixedrefrigeranttemperatureinseveralshellcells

Figureaboveshowsthetemperaturesduringthesimulatedtimeofseveralcellsintheshell
side.Someinterestingissuescanbeexplainedfromit.
First of all, it shows how the start up conditions last till after t = 1uus although mixed
refrigerant inlet temperature was kept constant from t = 6us. So, the system needs
approximately40stoachievesteadystateconditions.
Figure 29 also shows the effects of the abrupt change in the mixed refrigerant inlet
temperature(seefigure27).Note thatMRinlettemperatureiskeptconstantduring20sat
220Kbuttheheatexchangerstillrespondstoitasifitwasapeakinput.
Focusingonthenumericbehaviouronthemodel,animportantcharacteristicthatfigure30
illustrates is the good response of the time step function. Note that time step length
decreases when the system is working in transient conditions and as a result detailed
information is obtained from the model. Besides, it can also be observed that time step
length increases when steady state conditions are reached and that means less CPU time
consumption.
Since the system reaches steady state conditions at t = 1uus, from then on the time step
length is 30s. That could have provoked the lost of much of the information about the
abruptchangeconsequencesthatoccursatt = Suus.Toavoidthis,themodelisforcedto,
first, reach exactly t = Suus and then, in the next calculation use the time step with
minimum length. This way of solving this problematic is called event handling and some
programslikeordinaryequationssolversinMATLABusethesameprocedure.
41

Naturalgasmassflowrateduringthesimulatedtimeisshowninthefollowingfigure:

Figure30Naturalgasmassflowrateduringtime
In figure 30 it can be observed again the dynamic behaviour of the system, here, NG mass
flow rate during the simulated time. It is interesting to focus in the decrease that the mass
flow rate suffers when the MR inlet temperature changes increases abruptly. The cause of
thatphenomenonisthatnaturalgasoutletphasechangesfromliquidtovapour.Whenthe
cold stream increases its temperature more and more natural gas in vapour state emerges
fromthetubeoutlet,consideringthatpressuredropisconstantfromt = 6us,thisincrease
inthevapourfractionprovokesareductioninthemassflowrate.
MRmassflowratedecreasesasaconsequenceofthesamephenomenon.MRinletvapour
fractionincreaseswiththeinlettemperature.Consequently,massflowrateisreducedwith
inlettemperatureincreasing.
4.2.1 Thermaldynamicsduringthestartup
Oneofthemostappreciatedqualitiesofthismodelwillbeitscapabilityofsimulatingstarts
upandshutdowns.Therefore,detailedthermalresultsofthestartupinthissimulationare
givenbelow.
42

Figure31Naturalgascellstemperaturesduringt<150s

Figure32Mixedrefrigerantcellstemperaturesduringt<150s
Upper figures show some cells temperatures of each stream during the start up. It can be
seen how MR inlet cell reaches 110K just after t = 6us and is kept constant at that value
from then on. However, the rest of the system be considered to be in steady state till
t = 1uus.Someinterestingframesofwhathappensduringthisperiodaredisplayedbelow:
43

Figure33Temperaturesandvapoursfractionsatt=20sandatt=40s
Figure 33 shows NG, wall and MR temperatures along the length of the heat exchanger in
two different time steps, at t = 2us and at t = 4us. It also shows vapour fraction of each
streamatthespecifiedtimesteps.
Figure 33 clearly illustrates the influence of the phase change in the wall temperature. It is
reminded that NG, wall and MR starting temperature were 298K, 297K and 296K,
respectively,so,bothstreamsstartingconditionsweresuperheatedvapour.However,both
frames show that wall temperature reduces considerably its difference with MR
temperature when two phase MR flows through the shell side. The cause of that is the
increasingoftheheattransfercoefficientvalue;atwoheattransfercoefficientvalueismuch
higherthanthesinglephaseone.
44

Figure34Temperaturesandvapoursfractionsatt=44sandatt=100s
Figure34showsagaintwoframesofthestartup.Theseonesbelongtothesystemstateat
t = 44sandatt = 1uus(steadystate).
First feature that has to be highlighted is the rapid NG phase change. As it has been
explainedinchapter7.1,NGphasechangeathighpressuresisaabruptphenomenon.Here,
itcanbenoticedhowitbecameallliquidinjust4s(comparingwithfigure33att = 4us).
Second feature that should be commented about the first frame is that this liquid NG
achieves higher heat transfer values and that tights the wall temperature to higher values.
Thiseffectisillustratedintherisingofthelastcellstemperature.
Third and last important displayed characteristic that should be discussed refers to the
steadystateconditionsatt = 1uus.Itcanbeobservedthatwalltemperatureisonlycloser
toNGtemperaturewherethephasechangeoccurs.Inthepreviousmeterswalltemperature
is almost the same as MR temperature. After the phase change, wall temperature drops
again to a middle value within both streams temperatures. This feature is, again,
consequenceoftheheattransfercoefficientvaluedependingphasearea.
Moredetailedframesaboutthestartuparedisplayedintheappendix.
4.2.2 Thermaldynamicsduringanabruptchangeintheshellinlettemperature
Thesesecondresultsshowthethermaldynamicbehaviouroftheheatexchangermodel
whenjustoneinletparameterismodified.
45

Asexplainedinchapter6.1anddisplayedinfigure27,shellinlettemperaturesuffersarapid
change between 500s and 560s. This change in the inlet conditions has an effect in the
wholeheatexchangerthatisinterestingtostudy.

Figure35Naturalgascellstemperaturesduring500s<t<600s

Figure36Mixedrefrigerantcellstemperaturesduring500s<t<600s
FiguresaboveshowsomeofNGandMRcellstemperaturesduringthisperiod.Noticethat
althoughMRinlettemperaturekeepsconstantduring20sat220K,onlyMRinletcell
46

achievesastablestateduringthistimewhiletherestofthesystemisinconstantlychange
duringit.
Figure35showshowtheeffectsofthisabruptlychangearerapidlydampedinthetubeside.
Noticethat,sincethemodeltakesintoaccountthefluidmovement,thereisadelay
betweentheincreasingoftheMRinletandoutlettemperatures.
Aninterestingfigurethatshowstheconsequencesoftakingintoaccountthisfluid
movementisdisplayedbelow:

Figure37NG,wallandMRtemperaturesandNGandMRvapourfractionsatt=516sandatt=520s.Crosstemperatures
phenomenoncanbeobservedatt=520sframe
Figure37showsstreamandwallstatesatt = S16sandatt = S2us.Thealreadyexplained
effectsofthephasechangeonthetemperaturescanbeobservedagain.However,sincethe
model takes into account fluid movement, some interesting features can be extracted of
them.First,itcanbeseeninbothframesthatthereisaminimumatz = 9minthevapour
fractionthatcanonlybenoticedbecausefluidmovementeffectisnotneglected.
Notice that at t = S2us MR temperature at shell inlet rises above NG and wall
temperatures. That means that there cold and warm streams swapped locally. That is a
consequence of taking into account the fluid movement. Since it is a counter flow heat
exchanger,NGisfacingMRoutlettemperatureatitsinletandthatprovokesthatNGoutlet
temperature can be lower than MR temperature when MR inlet temperature is increased
rapidlyenough.
However,themodeldoesnottakeintoaccountthiscoldandwarmstreamsswappingwhen
calculatingheattransfercoefficientvalue.ThismeansthatMRtwophaseheattransfervalue
47

whenitislocallywarmerisstillcalculatedthroughanevaporationcorrelation,whichisnot
correct.However,ithastobeconsideredthatchangingheattransfercoefficientcorrelation
followingheatflowdirectionwouldprovokeadiscontinuityintheheattransferfunctionthat
wouldtriggerasevereinstability.
More frames about walls and streams states during this period are displayed in the
appendix.

48

5 Conclusionsandrecommendationsforfurtherwork
5.1 Conclusions
Duringthedevelopmentofthisthesissomeinterestingnumericalbehavioursrelatedtoheat
exchanger where phase change occurs have been observed. It has been proved that if the
pressureatwhichthephasechangetakesplaceishighenough,itcancreateadiscontinuity
in the vapour fraction dependence on enthalpy. This discontinuity affects severely the
numeric behaviour of the model, it provokes discontinuities on other functions like heat
transfercoefficientone,andthatobviouslyleadstoanumericalinstability.
A second observation noticed when simulating in this thesis is the importance of the
correlations selection. Whether one or other correlation is chosen has an impact on CPU
timeconsumption.Forthesameinputsimulationparameters,ithasbeenprovedthatDittus
and Boelter single phase heat transfer correlation leads to a higher number of necessary
iterationstoachieveconvergencethanGnielinskis.
Anotherrelevantoutputofthisthesisisthedevelopmentofasimplefunctionthatvariesthe
weightingcoefficientvaluewhenrelaxingthesolution.Thisfunctionhasbeentestedagainst
acoefficientwithafixedvalueandithasprovedtoproduceafastersolution.
Ithasalsobeennoticedthattimesteplengthandcellsizehaveanimpactonthenumerical
instabilityofthemodel.
Ithasbeenshownthatthemodelresultsprovidedetailedinformationoftheheatexchanger
dynamics. Time step control can be an appreciated feature of the model. Also, accounting
fluid movement has shown some interesting results that would have been skipped if it had
beenneglected.
The objective of this thesis was to find out if it is possible to develop a numerically stable
multistreamheatexchangermodelwherethermaldynamicscouldbesimulatedandstudied.
Pursuing an answer to this objective a stable counterflow heat exchanger model has been
satisfactorily developed. This model aims to be a suitable base to continue in the design of
theMSHEmodel.

49

5.2 Recommendationsforfurtherwork
Recommendationsforfurtherworkcanbeclassifiedintothethreefollowinggoals:
1. Optimisationofthenumericalpartofthemodel.
2. Researchofthesuitableheattransferandpressurecorrelations.
3. Introductionofthedevelopedtoolintopracticaluse.
The model developed in this thesis has still to overcome some numerical challenges till
presentingthesuitablecharacteristicstobeusedasaheatexchangerdesigntool.Stabilityof
the simulations must be guaranteed, hence, further study on the time step length and grid
meshing should be considered. It is the writer opinion that improved relaxing solution and
time step functions could give much more robustness to the model. Besides, the
improvementofthesefunctionscouldalsomeanareductionintheCPUtimeconsumption,
whichisthenumericalchallengethatshouldbefacedoncestabilityisguaranteed.
Furtherresearchshouldbedoneinthecorrelationsfield.Findingthecorrelationsthatbetter
correspond to the experimental results has not been the goal of this thesis. Therefore, an
evaluation of the different available correlations and its accuracy should be carried out.
However,italsohastobetakenintoaccountthenumericconsequencesthatthiscorrelation
maybring.Itisremindedthatithasbeenshowninthisthesisthatcorrelationselectionhas
aneffectonthemodelnumericalresponse.
Finally, once these and future challenges have been conquered, the model should be used
for the real design of a heat exchanger. Model flexibility during the design process and the
accuracyoftheresultsmaybethemainappreciatedfeatures.
50

6 Bibliography
1. Auiian }. Finn, u.L.}., Teiiy R. Tomlinson, LNGtechnologyforOffshore
andmidscaleplants. 2uuu.
2. }ulio Cesai Pacio, C.A.B., Areviewonheatexchangerthermalhydraulic
modelsforcryogenicapplications. 2u11.
S. Fieuheim A0, B.R., Possibilitiesofcostreductionsinbaseload.
E0R0uAS96. Pioceeuings fiom the Euiopean applieu ieseaich
confeience on natuial gas, 1996: p. 1u114.
4. Kanoglu N, B.I., Rosen NA, Performanceanalysisofgasliquefaction
cycles. Int. }. Eneigy Res, 2uu8.
S. RF, B., Cryogenictechnology. 0llman's encyclopeuia of inuustiial
chemistiy. WileyvCB veilag umbB anu Co, KuaA, 2uuu.
6. Waiien N. Rohsenow, }.P.B., Young I. Cho, eu. Handbookofheat
transfer. 1998.
7. F. W. Bittus, B., Heattransferinturbulentpipeandchannelflow. 19Su.
8. unielinski, v., Newequationsforheattransferinturbulentpipeand
channelflow. Int. Chem. Eng, 1976: p. SS9S68.
9. Stefan S. Beitsch, E.A.u., Suiesh v. uaiimella, Acompositeheattransfer
correlationforsaturatedflowboilinginsmallchannels. Int. }. Beat
Nass Tiansfei, 2uu9. 52: p. 211u2118.
1u. Z. Liu, W., Ageneralcorrelationforsaturatedandsubcooledflow
boilingintubesandannuli,basedonanucleatepoolboilingequation.
1991.
11. Shah, N.N., Ageneralcorrelationforheattransferduringfilm
condensationinsidepipes. 1978.
12. Issam Nuuawai, }.L., Twophaseflowinhighheatfluxmicrochannel
heatsinkforrefrigerationcoolingapplications:PartIpressuredrop
characteristics. 2uu4.
1S. W. Qu, I.N., Measurementandpredictionofpressuredropintwophase
microchannelsheatsinks. Int. }. Beat Nass Tiansfei, 2uuS. 46: p.
27S727SS.
14. Zivi, S.N., Estimationofsteadystatevoidfractionbymeansofthe
principleofminimumentropyproduction. ASNE }. Beat Tiansfei 1964.
86: p. 2472S2.
1S. }.u. Colliei, }.R.T., Convectiveboilingandcondensation. thiiu eu.,
0xfoiu 0niveisity Piess, 1994.
16. F. P. Inciopeia, B.P.B., FundamentalsofHeatandMassTransfer,fifth
ed. 2uu2.
17. R. K. Shah, A.L.L., LaminarFlowForcedConvectioninDucts:ASource
BookofCompactHeatExchangerAnalyticalData. 1978.

51

7 Appendix
Detailedresultframesduringsimulationsstartupperiod:

Temperaturesandvapourfractionsatt=8sandatt=16s

Temperaturesandvapourfractionsatt=8sandatt=16s
52

Temperaturesandvapourfractionsatt=40sandatt=48s

Temperaturesandvapourfractionsatt=56sandatt=64s
53

Temperaturesandvapourfractionsatt=72sandatt=80s

Temperaturesandvapourfractionsatt=90sandatt=100s

54

SomeframesduringtheabruptchangeoftheMRinlettemperatureare:

Temperaturesandvapourfractionsatt=500sandatt=504s

Temperaturesandvapourfractionsatt=508sandatt=512s
55

Temperaturesandvapourfractionsatt=516sandatt=520s

Temperaturesandvapourfractionsatt=524sandatt=528s
56

Temperaturesandvapourfractionsatt=532sandatt=536s

Temperaturesandvapourfractionsatt=540sandatt=544s
57

Temperaturesandvapourfractionsatt=548sandatt=552s

Temperaturesandvapourfractionsatt=556sandatt=560s
58

Temperaturesandvapourfractionsatt=570sandatt=580s

Temperaturesandvapourfractionsatt=590sandatt=600s