HAL Id: hal-00691272
https://hal.archives-ouvertes.fr/hal-00691272
Submitted on 25 Apr 2012
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Simulation of transient Rayleigh-Bénard-Marangoni convection induced by evaporation
O. Touazi, Eric Chénier, F. Doumenc, B. Guerrier
To cite this version:
O. Touazi, Eric Chénier, F. Doumenc, B. Guerrier. Simulation of transient Rayleigh-Bénard-
Marangoni convection induced by evaporation. International Journal of Heat and Mass Transfer,
Elsevier, 2010, 53 (4), pp.656–664. �10.1016/j.ijheatmasstransfer.2009.10.029�. �hal-00691272�
Rayleigh-Bénard-Marangoni onvetion indued
by evaporation
O. Touazi
a,b
, E. Chénier
a
, F. Doumen
b, ∗
, B. Guerrierb
a
Université Paris-Est,laboratoire Modélisationet SimulationMulti Ehelle,
MSME FRE3160 CNRS,5 bd Desartes,77454 Marne-la-Vallee, Frane
b
Univ Pierre etMarie Curie-Paris6, Univ Paris-Sud,CNRS, F-91405,
Lab FAST,Bat 502, CampusUniv, Orsay, F-91405.
Abstrat
Numerial simulation of thermal onvetion indued by solvent evaporation in an
initiallyisothermaluidisonsidered. Boththermoapillarityandbuoyanydriving
fores are taken into aount, and a riterium based on the Pelet number is used
to analyzethe stabilityof this transient problem. Critial Marangoni and Rayleigh
numbers are obtained for a large range of Biot and Prandtl numbers. Results of
the non linear simulations are ompared with a previous linear transient stability
analysisbasedon anon normalapproah andwithvisualizations performed during
PIB/Toluene solutions drying experiments. A saling analysis is developed for the
Marangoni problem and orrelations arederived to predit the order of magnitude
oftemperature and veloityasafuntion of
Bi
,M a
andP r
numbers.Key words: freeonvetion, evaporation, stability, heattransfer
1 Introdution
Evaporationof avolatiluidordrying of asolutioninduesa derease ofthe
temperatureatthe freesurfae duetothe vaporizationlatent heat.This situ-
ationan generate onvetive motionduetobothbuoyany and thermoapil-
larydrivingfores[16℄.Severalpointsmaybepointedoutwhenstudyingsuh
∗
Correspondingauthor. Tel: (33) 0169 15 3731;Fax:(33) 0169 15 80 60 Email addresses: Eri.Chenieruniv-mlv.fr (E. Chénier),doumenfast.u-psud.fr(F. Doumen), guerrierfast.u-psud.fr(B.
Guerrier).
plingbetweenthesystemandthesurroundings.Theevaporatinguxandthen
the temperaturegradientinthe uid depends onthe heat and mass transfers
withthe ambientair.Several authorshavedeveloped numerialortheoretial
studies taking into aount this oupling, for example Merkt and Bestehorn
[7℄, Colinetand o-authors [8℄, Ozen and Narayanan [9℄ and Moussy and o-
authors[10℄.Bothsurfaetension-andbuoyany-drivenonvetionmayour
as studied for example by Medale and Cerisier in several geometries but in
a nonvolatile uid [11℄. Another point that is more speially onsidered in
this paperisthe transient natureof manyexperiments.Indeed, startingfrom
an initial state where the uid is isotherm, evaporation indues a derease
of the surfae temperature. The basi temperature eld orresponding tothe
purediusiveproblem(nullveloityeld)istimedependentsothatalassial
stability analysis based onthe perturbation of a steady ondutive regime is
questionable. Then the predition of ritial onditions for the onset of on-
vetion is a omplex problem. A theoretial linear analysis based on a non
normal approah has been reently developed to take into aount the time
dependent basi state for this same problem(f. [12℄ and referenes herein).
In this paper we present a 2D non linear numerial simulation to study the
development and evolution of onvetive patterns in the transient regime,
when both buoyany and thermoapillary eets an be put forward. A large
rangeofparametersharaterizingtheproblem(Biot,Marangoni,Prandtland
Rayleigh numbers) are investigated. This numerial approah follows an ex-
perimentalstudy wherenumerousexperimentsofdryingofapolymer/solvent
solution have been performed with dierent thiknesses and visosities. Sim-
plifying assumptions follow from experimental results and are presented in
setion 2, as well as the model equations and numerial method. Setion 3
and 4 are devoted to 2D non linear numerial simulations and to the deter-
mination of onvetion thresholds for this transient problem. Comparisons
are made with experimental results and with previous results dedued from
the linear stability analysis. At last, a saling analysis is developed for the
Marangoniproblemin setion5.
2 Problem formulation
2.1 Thermal model
Dryingexperimentsthatunderliethesimulationspresentedinthispaperhave
been performed on the system Polyisobutylene (PIB)/Toluene. Experiments
aredesribedindetailin[13℄andweonlyreallthemainpointsusedtoestab-
lishthesimpliedmodelonsideredhere.Thesolutioninitiallyattheambient
trol parameters used in the experiments are the initial thikness (
0.3mm ≤ e ≤ 14.3mm
) and the initialpolymer mass fration (0 ≤ ω P ≤ 15%
). Poly-mer solutionvisosity isstrongly sensitive topolymer onentration[14℄.The
visosity
µ
is5.5 × 10 − 4 P a.s
and2.4P a.s
forω P 0%
and15%
respetively. Several simplifying assumptions have been inferred from experimental ob-servations. They are valid only at the beginning of the drying whih is the
time under onsideration inthis paper. When evaporation begins, onvetive
patterns have been observed at the very beginning of the experiment(quasi-
instantaneous or less than 100 s after pouring the solution). They disappear
well before the end of the drying. The Lewis number (ratio of the thermal
diusivitytomassdiusivity)being very large(about
10 3
),itisassumed thatonvetive patterns observed in the rst minutes are mainly driven by ther-
maleets.Twoexperimentalobservationsdetailedin[13℄supportthisthesis.
First,a fewexperimentswere onduted withdeuterated solvent, whoseden-
sityishigherthanthepolymerdensity.Inthatase,thedensityofthesolution
dereaseswhenthe polymer onentrationinreases,leadingtoastable situa-
tionif thesolutalRayleigh-Bénardproblemisonsidered.Sinenodierenes
were found with the experiments onduted with the standard solvent, we
an exlude solutal buoyany as adominant mehanism.Seond, freesurfae
temperature elds measured by infrared amera showed that the end of free
onvetionwasrelatedtothedurationofthetransientthermalregime.Solutal
onvetion is then not taken intoaount in this paper.
Moreover, thelayer thikness isassumed to remainonstant.This hypothesis
an beadopted if
P e int ≪ 1
, where the interfae Pelet numberis dened asP e int ≡ (ev ev )/α
withe
the layer thikness,v ev
the interfae veloity due toevaporation and
α
the thermaldiusivity. Indeed whenP e ≪ 1
, the surfaedisplaement
v ev δ dif f
remains negligible ompared to the total thiknesse
during the problemharateristitime i.e. the diusion time
δ dif f ≡ e 2 /κ
. Inthe experiments,the Pelet number issmaller than
0.1
. In the same way, therateofhangeofthepolymermassfrationissmallandthephysialproperties
of the solution are assumed onstant.
The free surfae is assumed at. Surfae deformationan be negleted if the
rispationnumber
Cr ≡ (ρνα)/(γe) ≪ 1
[15℄ andif theGalileonumberGa ≡ (ge 3 )/(να) ≫ 1
, whereρ
,ν
,γ
, respetively denote uid density, visosity, surfaetension,andwheregistheaelerationduetogravity.Suhonditionsare shown to be satised for the experiments onsidered here. Then surfae
deformability an be disregarded.
Forsakeofsimpliitythenumerialanalysisisrestrited toatwo-dimensional
geometry sine the study fouses on the onset of onvetion and not on the
desription of onvetive patterns morphology. Comparison of experimental
riori. The solution layer is ontained in a retangular domain of aspet ratio
A = L/e = 20
, whereL
is the horizontal length ande
the thikness of thesolution layer. The vertial and bottom solid walls are supposed to be adia-
bati and non permeable. Evaporation ours at the free surfae only. Heat
andmass transfersbetweenthe freesurfae andthe ambient airare desribed
by global transfer oeients.
Withthe Boussinesq approximation, the Navier-Stokesand energy equations
are:
∇ ~ .~v = 0,
(1a)∂
∂t ~v + (~v.~ ∇ )~v = − 1
ρ ∇ ~ p + ν∆~v + gβ T (T − T 0 ) e ~ y ,
(1b)∂
∂t T + (~v.~ ∇ )T = α∆T,
(1)where
~v(~x, t)
is the veloity eld,p(~x, t)
is the pressure eld,T (~x, t)
is thetemperature eld,
T 0
the temperature ofthe ambient airande ~ y
the unitve-tor in the vertial diretion. In this approximation, the density
ρ
is taken tobe the density at
T = T 0
.β T
,ν = µ/ρ
,α = k/ρc
are the thermal expansionoeient, the kinemativisosity and the thermaldiusivity with
µ
the dy-nami visosity,
k
the thermalondutivity andc
the spei heat.The polymer mass fration being small, we have used the physial proper-
ties of the toluene exept for the visosity that is very sensitive to poly-
mer onentration. The following values have been used:
ρ = 865kg.m − 3
,k = 0.142W.m − 1 .K − 1
,α = 0.97 × 10 − 7 m 2 .s − 1
,β T = 1.07 × 10 − 3 K − 1
.Foreahexperiment the visosity is also assumed onstant but depends on the initial
polymer onentration, aording to an empirial law (Fig.1) dedued from
visosity measurements performed with a Low Shear 30 rheometer (oaxial
ylinders and imposed deformation)[16℄.
2.2 Initial and boundary onditions
At
t = 0
the uid is at the ambient temperatureT (~x, 0) = T 0
. To studythe stability of the system, a perturbation is imposed onthe veloity eld at
t = 0
. Sine the strutures of real experimentalperturbations are not known a priori, a random veloity perturbation with zero mean and amplituder
is implemented in the following way: numerial resolution of the problem is
ahievedusinganitevolumesheme.Foreahspatialnode,thevalueallotted
totheveloityat
t = 0
omesfromarandomdrawingofauniformdistribution between− r/2
and+r/2
. Further studies have shown that the hoie of the3.2and [12,17℄).
The solidwallsare adiabatiwithazero veloityboundaryondition.On the
free surfae loated at
y = e
, the thermalboundaryondition is :− k ∂T
∂y
!
y=e
= h th (T (x, e, t) − T 0 ) + LΦ ev
(2)The rst term of the r.h.s. is the onvetive heat transfer between the free
surfae and the ambient air where
h th
is the heat transfer oeient. Theseond term,
L Φ ev
, orresponds to the solvent evaporationwithL
the latentheat of vaporizationand
Φ ev
the evaporativeux, that an be expressed asΦ ev = h m (ρ g Sint − ρ g S ∞ )
(3)where
h m
is the mass transfer oeient,ρ g Sint
andρ g S ∞
are the solvent on-entrationinthe gas phasejustabove the interfaeand farfromthe interfae
respetively. The last one is zero, due to the important air ow rate in the
extratorhood.With the ideal gas law, we get:
Φ ev = h m
M S P V S0 (T (x, e, t))
RT (x, e, t) a(T (x, e, t), ϕ S (x, e, t))
(4)where
M S
is the solvent molar mass,a
is the solvent ativity,P V S0
is thesaturatedsolventvaporpressure,
ϕ S
isthesolventvolumefrationintheliquidphase at the interfae and
R
is the ideal gas onstant. In polymer solutionsthe ativity is lose toone for solvent volume fration greaterthan about 0.4
[18℄, so that
Φ ev
an be assumed independent of the solvent onentration at the beginning of the drying (a ≃ 1
).Moreover, giventhesmallamplitudeoftemperaturevariationsobservedinthe
experiments (a fewdegrees), it ispossibleto use a rst orderdevelopmentof
Φ ev
, i.e.Φ ev (T (x, e, t)) ≃ Φ ev (T 0 ) + ∂Φ ∂T
ev| T
0(T (x, e, t) − T 0 )
Thethermalboundaryonditionatthefreesurfaeanthen beapproximated
by the following expression
− k ∂T
∂y
!
y=e
= H th (T (x, e, t) − T 0 ) + L Φ ev (T 0 )
(5)with
H th = h th + L ∂Φ ∂T
ev| T
0Ashear stressboundary onditionisimposedatthe freesurfae,givenbythe
µ ∂v x
∂y
!
y=e
= dγ dT
∂T
∂x
!
y=e
(6)
where
γ
,the surfae tension,is alinearly dereasingfuntion of temperature.Theotherboundaryonditionattheinterfaeonernsthevertialomponent
of the veloity,
v y
. Assuming a planar surfae, and assumingthat the spatialvariationsof the evaporationux are muhsmaller thanthe ux itself,itan
beshown [12℄ that, inthe limitof
P e int ≪ 1
, the boundaryondition reduesto:
v y = 0
aty = e
. (7)Aording toexperimentalresults,thefollowingvalueshavebeen used:
H th = 28W.m − 2 .K − 1
,L = 396kJ.kg − 1
,Φ ev (T 0 ) = 3.37 × 10 − 4 kg.m − 2 .s − 1
,dγ/dT =
− 0.119 × 10 − 3 N.m − 1 .K − 1
.2.3 Non-dimensional equations
The non-dimensional form of the equations results from saling the lengths
by the thikness of the uid
e
, the veloity~v
by the diusion veloityα/e
,and time
t
and pressurep
respetively bye 2 /α
andρα 2 /e 2
.The temperature sale isθ = T ∆T − T
0 where∆T
is the dierene between the initial temperatureT 0
and the steady temperature obtained at the end of the transient regime, when the temperature is uniform in the solution. From equation 5 we have∆T = LΦ H
evth(T
0)
, that is∆T = 4.8K
for the experimental onguration. Let us note that this temperature sale is dierent from the one used in lassialstabilityanalysiswhere the basistate isharaterizedby aonstant temper-
ature gradient inthe uid.
The dimensionlessform of equations (1a-1)is:
∇ ~ .~v = 0
(8a)∂~v
∂t + (~v.~ ∇ )~v = − ∇ ~ p + RaP rθ~e y + P r∆~v
(8b)∂θ
∂t + (~v.~ ∇ )θ = ∆θ
(8)where
P r = ν/α
is thePrandtl number,Ra = β T g ∆T e 3 /(να)
isthe Rayleighnumber.
ary onditions atthe free surfae are :
− ∂θ
∂y
!
y=1
= Bi (θ(x, y = 1, t) + 1) with Bi = H th e
k
(9)∂v x
∂y
!
y=1
= − Ma ∂θ
∂x
!
y=1
with Ma = − e ∆T µα
∂γ
∂T
!
(10)
v y = 0
aty = e
. (11)with
Ma
the Marangoni number andBi
a modied Biot number that takesinto aountthe evaporativeux.
The ow and thermal behaviour in the solution are governed by four non-
dimensionalparameters(
Bi
,Ma
,P r
andRa
),thatdependonthetwoontrolparameters used inthe experiments,the initialthikness and visosity [13℄.
2.4 Numerial method
Computations are arried out with a olloated nite volume sheme with a
seond order spae and time disretization [19℄. The disrete sheme is fully
oupled.Thesetofthedisretebalaneequationsissolvedbyanunder-relaxed
Newton'smethodwith theiterativelinear solverBICGSTAB, preonditioned
by an inomplete LU fatorization. The hosen mesh is a standard regular
mesh
L ∗ e = (800 ∗ 40)
and the time step is equal to∆t = 10 − 3
. Resultsobtained with ner grids or time steps did not show any notieable hanges
inthe results.
3 Transient onvetion
3.1 Typialtemperature and veloity elds
Fig.2 gives a typial example of the time evolution of the non dimensional
temperature dierene between the bottom and the free surfae of the solu-
tion for a ongurationwhere onvetion is observed (parameters of the test
ase are given in table 1, initial veloity disturbane amplitude
r = 4
). Asa omparison, the temperature evolution obtained for a pure diusive prob-
lem is also drawn (dashed lineobtained with
Ma = 0
andRa = 0
). Severaldomains an be observed. At the beginning, heat transfer is dominated by
the diusion and the twourves are superposed (domainI). Then onvetion
atureeld (domainII).This regimeisfollowed by aquasi-steadyregime with
aslowderease ofthetemperaturedierenebetweenthe bottomand thefree
surfae (domain III). At lastfor large times the two urves orresponding to
purediusiveoronvetive regimesgotozero.Indeed the temperatureinthe
uidisuniformatthe endof thethermaltransientregime(notrepresented in
Fig.2).
The stream lines and the temperature on the free surfae are given in Fig.3
for three dimensionless times t=0.5, t=0.6 and t=0.65. We an observe the
vanishing of the entral ell between t= 0.5 and t=0.65 that orresponds to
the seond peak that an be observed on the temperatureevolutionin Fig.2.
It is followed by the quasi-steady regime,with a onstantnumberof ells.
The objetive of this paper is to analyze the three regimes, as a funtion
of the four non dimensional parameters that haraterize the problem,
Bi
,Ma
,P r
andRa
numbers. The transition from domain I to domain II, i.e.the onset of onvetion is analyzed in setion 4. Setion 5 is dediated to
the haraterization of the quasi-steady regime, using saling laws to get an
estimation of the temperatureand veloity in this regime.
Toharaterizethepreseneofobservableonvetion,ariteriumbasedonthe
thermal Pelet number was hosen. The thermal Pelet number
P e = e v/α
ompares the relative importane of advetion and diusion. The veloity
v
used in the estimation of the Pelet number is the maximal value of theveloity norm. Then onvetion will be onsidered signiant if, when the
systemissubmittedtoaninitialveloity perturbation,there isatime
t
wherethe perturbation is signiantly amplied, i.e. suh as
dP e(t)/dt > 0
andP e(t) > 1
.3.2 Eet of the initial disturbane amplitude
As the problem is sensitive to initial onditions, we have rst performed a
preliminarystudy to analyze the inuene of the amplitude of the initialdis-
turbaneonthe timeevolutionofthePeletnumber.The parametersusedfor
this test are given inTab.1.
Fig.4a shows the temporal evolution of the Pelet number in log sale. At
the beginninga linearregime an be observed witha dereasing of the initial
perturbation followed by its ampliation. But at large
t
, in the non linearregime,
P e
beomes quasi independent of the initialperturbation.If we on- sider the hosenriterium(P e > 1
),all thetests lead tothe same onlusion,i.e.onvetionisobserved;onlythetimedelaydependsontheamplitudeofthe
disturbane. The orresponding time is about
t = 0.1
forr = 4
andt = 0.03
for
r = 400
. Fig.4b shows the Pelet number normalized by its initialvalue,sothat the transitionbetween the linearand nonlinear regimean belearly
observed.
Sine the general trends observed in the three regimes and the ourrene of
onvetiondonotdependstronglyontheperturbationamplitude(atleastina
largedomain),weuse
r = 4
inthe following.A moredetailedstohasti anal-ysis of the inuene of the initialperturbation struture has been performed
[17℄ andgivesthe same onlusions:hangingthe initialperturbationindues
only small hanges in the pattern wavelength, in the temperature or in the
Peletvariations.Thethresholdvalueorrespondingtotheonsetofonvetion
isthereforelittlesensitivetothe initialperturbation.This assumptionison-
rmedin setion4.3 by the goodagreementbetween the thresholds obtained
with the non-linear simulationsand the linear stability analysis. Howeverthe
thresholdmustnotbeunderstoodasapreisedelimitationbetweenstableand
unstable domains but rather as a transition region (a fator of about two is
obtained for dierent initialperturbations). This "blurring" eet is inherent
tothe transientharater of the problemunder study [12℄.
4 Stability
4.1 Inuene of the dynami visosity
The visosity of solutions understudy being very sensitive tothe initialpoly-
mer onentration, we rst study the inuene of the dynami visosity
µ
on the Pelet transient behavior. Both buoyany and surfae tension driving
foresare taken intoaount(Rayleigh-Bénard-Marangonionguration)and
resultsare given inFig.5for
e = 1
mmand fordierentvaluesof the dynamivisosity
µ
.For the largest visosities, the Pelet number is a monotone dereasing fun-
tions and the initialperturbation dies down. For smaller visosities, there is
an ampliation of the initial perturbation after some times. However this
ampliation may be very weak and for visosity larger than
5mP a.s
thePelet number is always smaller than one, so that this onguration is "sta-
ble" aording to the riterium dened in setion 3.1. The ritial visosity
orrespondingtothe onsetofonvetionliesbetween
4mP a.s
and5mP a.s
forthis thikness. In the next setion, the same study was performed for other
thiknesses inorder to ompare simulationsand experimentaldata.