HAL Id: jpa-00247522
https://hal.archives-ouvertes.fr/jpa-00247522
Submitted on 1 Jan 1991
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Aggregation in an expanding cloud: experiments and numerical simulations
Françoise Dziedzinl, Robert Botet
To cite this version:
Françoise Dziedzinl, Robert Botet. Aggregation in an expanding cloud: experiments and numerical simulations. Journal de Physique II, EDP Sciences, 1991, 1 (3), pp.343-352. �10.1051/jp2:1991172�.
�jpa-00247522�
J
Phys
II1 (1991) 343,352 MARS 1991, PAGE 343Classification
Physics
Abstracts82 70R 05 40
Aggrigation in
an
expanding cloud
:experiments and numerical simulations
Frangoise
Dzledzini and Robert BotetCE B, BP n° 3, 91710, Vert-Le-Petit, France
Laboratoire de
Physique
des Sohdes, Bit 510, UnlversitdPans-Sud,
Centred'orsay,
91405Orsay,
France(Received14 November 1990, accepted17 December1990)
Rksumk. Nous avons construit un dispositlf
expdnmental
et un moddle numdnque pour dtudierl'agrbgation
d'un abrosol (oxyde detitane)
en expansion dansl'atmosphdre.
A partir del'analyse
de
photographles
pnses au microscopeblectronlque
Ibalayage,
on montre que l'abrosolagglombrb
forme des amas fractals de dimension fractale de l'ordre de 1,75. Les simulationsnumbnques
confirment quantitativement cetteparticulantb gbombtnque
Nous montronscomment nos rbsultats
numbnques
peuventcomplbter quelques
points qui ne sont pas accessiblesI
l'expbnence.
Abstract. We have set-up an
expenmental
device and a numencal model tostudy
aggregation of an aerosol (titaniumoxide) expanding
m theatmosphere
By mean of scanning microscopicanalysis,
it is shown thatagglomerated
aerosol forms fractal clusters of fractal dimension of about 75. The nurnencal simulationsquantitatively
confirm thisgeometncal
feature. We show howour numencal results can complete some points which are not available m expenments
In order to
study aggregation
of solidatmosphenc
aerosols in standard ambient cond1tlons oftemperature,
pressure andhumidity,
We have chosen astest-products
titaniumcompounds
formed
by
chemical reaction ofTiCl4
With Water Vapor[I].
Thesecompounds
are used m suchmatters as
metallurgy [2]
orphotodissociation
of Water[3].
These are also Well-known atsmoke obscurants The
elementary particles
arespherical
and therefore easy tostudy
1.
Experhnental
device.The
expenmental,device
is based on a 50sphere
made ofglass
Thls rather small Volume has been chosen because it was much easier tosample
and control the initial conditions. There are several inlets on the surface of thesphere
which makes itpossible
to add filters(for example
to check the pressure inside the
sphere)
and different apparatus which will be descnbed belowAs a first
tnal,
we havedispersed
theTiC14
inliquid phase by
mean of a nebulizer connected to one of the inlets of thesphere
But this idea was given up because the chemicalreaction between
Ticl~
and water vaporhappens
veryquickly
and takesplace
in the pipe whereconsequently
alarge
amount ofproduct
settles withouthaving enough
time to reach the inside of thesphere
itself. Todisperse TiC14 directly
inside thechamber,
we havedeveloped
a small metallic device which is able to hold and break a littlephial (glass) containing
5 ~Ll ofTicl~.
Note thatby
this way, the device allows theexpenments
to bequite reproducible,
as for thequantity
ofproduct generated
m thesphere.
Then we have first remarked that thejust-formed particles
tended to be attractedby
the metallicparts avoiding
main part of the diffusion inside the
sphere
That is the reasonwhy
we have added a thin tube whichbnngs
anitrogen
flux(nitrogen
does not react withTiC14) just
at thetop
of the brokenphlal leading
to aquasi-sphencal dispersion
of theproduct.
Moreover it is quite easy to tune the expansionvelocity
of the cloudby
mean of thisnitrogen
pressureThe
experiments
have takenplace
at the ambienttemperature (23-24 °C).
Thehumidity
is tunableby
thequality
of airfilling
thesphere
at thebeginning
of theexpenment,
and it iscontrolled in permanence
by
ahumidity probe.
A
sampling probe penetrates
at about2/3
the radius of thesphere (8.5
cm from thecenter)
and is connected to a thermal
precipitator.
Thisapparatus
allows thesampling
of very small volumes(e
g 6cm~/min)
which isindispensable
in our small chamber. Particles taken in the flow come to a hot wire andprecipitate
on smallglass
lamellas situated on each side of the wire. These lamellas are then coated withgold
for observation m a scanning electronmicroscope
This coat is 4 nm thick which is much smaller than thetypical
diameter of theindividual
particles (~
40nm).
At themagnification
used tostudy
the aggregates, the surface of thespheres
appears smooth.Figure
I is asimplified
sketch of this device.Fillei
Thernial Humidity-
PhiJl
§f~
Filter'
, i<iiicr~~§Pump
Fig
[.-Sketch of theexpenmental
device2. Some
experimental
results.The
nitrogen
pressure, the relativehumidity
and the temperature caneasily
be varied. In this paper we areonly
concemed in the influence of the radialvelocity
via thenitrogen
pressure.The influence of the relative
humidity
will bepublished
elsewhere[4]
and the influence of the temperature has notyet
been doneexperimentally.
Quahtatively
we observe a radial expansion of a white cloud of titaniumcompound.
Ofcourse the
larger
is the pressure of nitrogen flux thelarger
is thevelocity
of thisexpansion
Typically,
theedge
of thesphere
is reached after about ten seconds Observation ofsamples
M 3 AGGREGATION IN AN EXPANDING CLOUD 345
by
the thermalprecipitator
showselementary particles (spherical
inshape
with a mean-size of about0.l~Lm)
as well asaggregates
of theseparticles. Figure2
istypical
of suchconfiguration.
To obtain
quantitative results,
wed1gltalize
thephotographs
taken from the electron microscope. In such a way, we havedirectly
an idea of the size-distnbution of the part of thesample
which has beenphotographed,
as well as the fractaldimension,
if any, of thelargest aggregates belonging
to the field. The method to measure this dimension is thenesting-circles
method as
explained
now for eachcluster,
a standard program transforms itsd1gltahzed
image into a binanzed image m which each
point
is I or 0 whether it isoccupied by
matter or notThen,
the center of mass of the blackpixels
is calculated and one chooses the closest blackpixel
to be theorigin
of the local coordinates. One draws circles centered on this onglnand of successive radii :
4, 6, 8, pixels,
and one counts all the blackpixels
inside each circle.This
gives
an estimation of the total mass of thecluster,
within agiven
distance from the ongln. The fractal dimension of each individual cluster could be extracted from aLog-Log
plot
of the mass versus this distance since the formula :M(R)
~
R~
holds for fractal
objects
of fractal dimension D~ 2 We shall
verify
a posteriori that all the fractal dimensions founded here are smaller than 2 But since we areworking
with manyif
Fig.
2Typical
scanning macroscopicphotographs (nitrogen
pressure =01bar,
relative humidi- ty = 73 5 9b, time = mln 30 s) The two largest aggregates of the field have a diameter of order 3 ~Lmobjects
ofrelatively
smallsizes,
we have to be very careful about the way to estimate theaveraged
value of this fractal dimension. A crude average of all values of D estimated for each cluster would give aquite
bad estimate of itsaveraged
value since the fractal dimension acts as anexponent
m the formularelating
the mass and the radius. So we have realized the averageof all masses
M(R)
contained inside a circle ofgiven
radius R for all the clusters of thecollection for which the estimated diameter was
larger
than I ~Lm. Then we have calculated the fractal dimension with theLog,Log plot
of(M(R))
versus RWe have obtained very nice
straight
lines on these doublelogarithmic plots,
and this makes believable theinterpretation
of thephotographs
in term of fractalobjects (see Fig 3).
In table
I,
wepresent
two series of results : in the first one the fractal dimension isgiven
fora nitrogen pressure of 0.I bar
(and
relativehumidity
73.5fb).
In the second one thenitrogen
pressure is 0.2 bar(and
relativehumidity
87.49b).
Sixsamplings
have been done at six different times for eachseries,
from I mm 30 s to 18 mm. The number ofanalyzed
clusters are indicated below the fractaldimensions,
so that one can estimateroughly
the confidence in theresulting
value The error bars on the values of the fractal dimension are estimated of about± 0.05. The results of Table I
appeal
some comments. We observe first that there are nosignificant
variations of this fractal dimension with time. Asaggregation
goes on, the fact thatD remains constant means that it is a
good quantity
to charactenze the internalgeometry
ofthe clusters
independently
of their sizes.Moreover,
our results for the smallest time we can do(1.5 mini
show thataggregation
process isalready
wellengaged
at this time(see Fig. 2) 1ihls
is conforted
by
theanalysis
ofexperimental
size,distribution the mean diameter increasesonly slightly
with time(an example.
0.306 ~Lm, 0.308 ~Lm, 0.388 ~Lm, 0.378 ~Lm forrespective
times. I mm 30 s, 6 mm, 13 mm, 18 mm, for
nitrogen
pressure 0.I bar and relativehumidity
73 5
fb,
the errors on the diameter are of order 0.I ~Lm). In the two series of experiment, wehave found D
=
75 ± 0.08 for the value of the fractal dimension
Log(M) 4
D 1.68
O,5 1,0 1,5 2,0
Log(R)
Fig
3.-Onetypical
experimental result(nitrogen
pressure= 02bar, M is the mass
averaged
over13 clusters of a same sample) The saturation for Log (R) ~ l 5 is due to the finite sizes ofthe clusters The scales are
arbitrary
M 3 AGGREGATION IN AN EXPANDING CLOUD 347
Table I.
Experimental fractal
dimensionsfor d#jferent
times and two pressuresof
nitrogenflux.
The relativehumidity
ts indicated below the pressurefor
the two experiments. The smallnumbers below the values
of
thefractal
dimensions are the numberof
clusters used to extract the valueof
the dtmensions.PN~)
0.I bar 0.2 bartime
(RH)
(73.5 fb) (87.4 fb)
mln 30 1.62 70
(23j (6j
6 mm 83 79
('21 (61
13 min 1.70 75
(1 14) (8)
18 mm 1.67 1.79
(26) (18)
A
striking
result is that the fractal dimension does not seem todepend
on the radialvelocity
of
expansion
of the cloud(see
TableI).
In fact arapid sight
on thisproblem
could lead to theerroneous conclusion that the clusters should be less compact than clusters grown in
homogeneous
conditions since, because of the radial drift which tends tokeep
the clusters far the ones from theothers,
the number ofpossible
collisions between two of them isconsiderably lower,
so thatjust sticking by
the tips ispredominant.
As we shall see, thissimple
reasoning is wrong for fractalobjects,
butthe
reason of this failure is not immediate.Apart
anexplanation
of the value of the fractal dimension the reasonswhy
we have tried toexplain
these featuresby
a numencal simulation of this sort ofphysical aggregation
within this radialgeometry
are that this numericalapproach
gives access to the first times ofcoagulation,
which cannot be reached
experimentally
and which iscertainly important,
asquoted
before3. Numerical model.
Contrary
to colloidsdispersed
in a viscousliquid
like water, and for which it is easy to avoidcoagulation by
agitation, in mostapplications
aggregation of aerosols takesplace
in anhomogeneous
medium(steady
state gas of lowviscosity)
but withinhomogeneous starting
cond1tlons. Theexpenment
described in Part I is atypical example
of such conditions where a competition exists between expansion of the aerosol cloud andaggregation
of the aerosol.To
study
the influence of a radial drift on thegeometrical
structure ofagglomerated
aerosols we have
imagined
a numencal model which wepresent
now a set of Nsphencal
particles
of same diameter d is put in a continuous infinite space This set simulates the imtalphlal
of theexperiment
and is defined as follows a cluster of Nparticles
is constructedaccording
to a variant of the Bennett-scheme[5] (see Appendix)
which realizes very compactclose-packed
randomobjects (see figure
of theAppendix)
At this stage the diameter of theparticles
is setequal
to I and it will be the definition of the unit oflength
m theproblem.
Then the diameter of all the Nparticles
is decreased to the desired value d. This initialization allowsto obtain a disordered set of ind1vldual
disjointed particles
and toadjust
the value of the initial concentration to any value between 0 and the most compactpackage
of iso-diameterspheres (about
0 61by
thisway).
At time t one gJves to each cluster
(or
individualparticles)
its actual radialvelocity
v~ whose modulus is
r/t [6] (this
is a solution of the Namer-Stokes equation of a ball of viscous gasexpanding radially
in anempty space).
Here r is the distance of its center of mass to thecenter of
explosion (center
of mass of the initalconfiguration).
One gives also to each cluster its dilTusion coefficient D= D
o/k
[7~ whereDo
is the coefficient for an individualparticle
and k is the number ofparticles (that
is : itssue)
of the considered cluster. At thispoint
one tnesto increase the time of an amount At, and all the clusters move at the same time. To achieve
these movements one calculates the individual
displacement
of each cluster as the sum of the radialdisplacement (which
is just theproduct
of thevelocity
v~ andAt)
and the Browniandisplacement
which isequal
tofi
ina random direction of the space. The trial value of At is chosen such that no
displacement
of a cluster can exceed the valued/2
in modulus.Overlaps
ofparticles
canhappen.
If it is the case, we search thelargest
value of At which avoidsoverlap
but allows contacts, andby
this way :aggregation.
Th1s value is givensolving
as many 4thdegree
equations as there areoverlaps
The direction of the Browman motion of each cluster iskept
constantthroughout
this search(only
the modulus of thisdisplacement
is affectedby
thechange
ofAil.
Thissticking
on contact is irreversible m thesense that, once
stuck,
twoparticles
can never unstick. Clusters arengld
and so we do notstudy
here the influence of thetemperature
in this process.The fractal geometry of these clusters are
investigated
as a function of time and of the adimensionalparameter
p =Do/vod,
where vo is the initial dnftvelocity
Thlsparameter
characterizes the relativeimportance
of the radial dnft and the Brownian motion, and it may be estimatedlying
in between I and 10 in ourordinary physical
conditions. Note that thelarger
is p, the smaller is the radialexpansion
of the cloud.4. Numerical results.
At the
beginning,
the fractal dimension of the clusters have been calculated in two different ways. The first one consists inplotting
image of all thelargest
clustersprojected
on threeorthogonal planes
and todigitallze
then binanze thesepictures
We treat them withthe
same program as described in theexpenmental part.
This insuresexactly
the same treatment for theexperimental
clusters and for the simulated ones The second one consists inplotting
theloganthm
of the number ofparticles
of each cluster versus thelogarithm
of its radius of gyration for all clusterspresent
at agiven
time. Theslope
of thestraight
line(if any)
gives avalue of the fractal dimension of the collection of
clusters,
since we know the formulaN~Rf'
(this
isequivalent
to the formula given in section II but for a collection of clusters of same fractal dimensionD).
In tableII,
the fractal dimensions obtainedby
the two methods areshown and can be
compared
for two numencal simulations. One sees thatagreement
issatisfying.
Th1s result gives confidence in the method(the
firstone)
used inexpenmental
set- up, andexplains why,
for all thefollowing,
numencal results have been obtained with thesecond method
(easier
to use on acomputer).
We have
checked, by
this method that the fractal dimensions do notdepend
on the time.But the poor statistics due to the small number of clusters at small
times,
lead us to take in fact all the clusters any timethey
appear in the simulation to have better statistics to extract thecommon fractal dimension. Th1s
independence
of the fractal dimension with respect to time isM 3 AGGREGATION IN AN EXPANDING CLOUD 349
Table II. -Numerical
fractal
dimensionsfor
twodiameters,
with two methodY the nesting- circles with threedifferent projections
onorthogonal planes,
and the methodsof
the radiusof
gyration
(N
=
2
000,
p=
5).
Diameter o.90 0.91
j~ projection
1.74 1.67
along
X E~
~
projection
(
1.70 1.67£
along
Yc- o
i projection
~ l.60 1.67
( along
ZMethod of the radius
of
gyration
71 1.72well-known for
closely
related models as standard Cluster-Cluster aggregation[8].
This is due to the scale-invanance of the system.We have
systematically
studied the influence of the parameter p on the value of the fractaldimension,
for pbelonging
to the range 1-10(see Fig. 4)
We have found the value D = 1.72 ± 0.10. One remarks onfigure
4 that the first points are wellsupenmposed
overabout one decade
(in radius).
This confirms that the fractal dimension is not even the same,but_
we are in presence of the same statisticalobjects.
The value of the fractal dimension must be
compared
with the value measured in the expenments D= 175 ± 0.08. So either
experimental
or numencal results agree to say that there is no influence of the radial dnft on the geometry of theclusters,
even if the size-distnbut1on is indeed very sensitive to this parameter.
The reason of this absence of influence is that two clusters close one to the other have a
small relative dnft
velocity (in regard
to their relativedistance)
so that Brownian motionremains
predominant
as for the standardagglomeration
process. Thisexp1alns
alsowhy
theexperimental
or numencal values of the fractal dimensions are identical(within
errorbars)
to the standard cluster-cluster one.We have
changed
the value of another parameter d(and
so the 1nltial concentration, asexp1alned before).
In tableIII,
the fractal dimensions for parameter p =5,
N= 1000 and d
TableIII.-Numerical
fractal
dimensions versus the diameterof
individualparticles (i.e.
versus the initial
concentration) for
the parameters N= 2
000,
p= 5
Diameter 0 90 0 91 0 93 0.95 0 97
Fractal
Dimension 1.71 1.72 1.74 71 74
Log(N)
a
a
$f°
~a
~
~
~f
a~
~
# all
a.
m
»
«
« w
»
I
M 3 AGGREGATION IN AN EXPANDING CLOUD 351
Acknowledgements.
It is a
pleasure
to thank M.Cabane,
C.Tiret,
R Julhen and Ph. Adam forpermanent help
and discussion. Th1s work was
supported by
contract n°25073/89/ETCA/CEB/DED.
Appendix.
The Bennett-scheme
[5]
is analgorithm
used to construct very compact structures. The idea is to start with a set of threeadjacent particles forming
anequilateral tnangle.
Then one addssuccessively
individualparticles
of same diameter in such a way that the addedparticle
touches three other ones
(in
a d-dimensional space, this maximum condition writes.tangent
to d other
particles).
An economic way to achievethis,
is to make a list of all thetriplets
ofparticles
which have the two properties : eachparticle
of thetriplet
touches the twoothers,
and one can put at least a fourthparticle touching
the three others. The number of suchtriplets
in this list growsjust
as the surface of the cluster(so
it is aN~'~-process,
where N is the total number ofparticles
of theaggregate).
To insuregood
randomness of thecluster,
one adds a newparticle
at a site chosenrandomly
among all thepositions
of the availabletnplets.
This is the version used here One reactuahzes then the list of
triplets
and so on. In his workBennett choose
systematically
the site the closest to the center of mass of theaggregate
Atypical
cluster of 1024particles
grownby
this way is shown infigure
5.Fig 5 Typical 024
particles
cluster obtainedby
the Bennett-like scheme as descnbed m the nextJOURNAL DE PHYSIQUE II T I,M3, MARS [WI 16
References
[1] PASCAL P., Nouveau Traitk de Chimie Mmdrale, Tome IX
(Masson,
Pans) 1963POLYACHENOK L. and NOVIKOV G., Obshch. Prikl. Khim 5
(1972)
31[2] BAUER O, US Patent Offce 3, 425, 796
(1969)
TORU I., MASAYOSHI K, MASAn M, Bull Chem Sac Japan 45
(1972)
2343[3] Zuco V V, ANTIPIN L M, NEMODRUK A A, and SHOSTAKOVSKII M F, Zhurnal Anahticheskoi Khimu 38
(1983)
831KOROLEV V. V. and SHOKIMA N T, Zavod Lab. 37 (1971) 1332
KUTTY T. R N., AVRIDAITKAI M, Mat Res Bull. 23
(1988)
725.[4] LESAFFRE-DzIEDzINL F, Formatlon
d'Agrdgats
d'Adrosol.Expdnences, Analyse
Fractale etSimulations, Thdse de Doctorat, Unlversitb Pans 7
(1990)
DzIEDzINL F and BOTET R m preparation
[5] BENNETT C H, J
Appl. Phys.
43(1972)
2727[6] The geometry of the aggregates do not seem to
depend
on the precise law givingv~ as
long
as v~ remains centro-symmetnc[7] Here we suppose a priori
tiat
the clusters haveso tenuous structure that the fnction force is
proportional
to the total number ofparticles
of the cluster Thishappens
forexample
with fractal clusters of fractal dimension< 2. For more compact
objects,
one has toreplace
this formulaby
D=
Do/R
where R is atypical
radius of the cluster(m
the Stokesregime)
Once again thegeometncal
features do notdepend
on the precise vanation of D with k or R, aslong
as D is a
decreasing
function of k, cf BOTET R,
JULLIEN R and KOLB M
,
J
Phys
A Letters 17 (1984) L75[8] For a recent review BOTET R and JULLIEN R. Phase transitions 24
(1990)
691.[9] The elastic
light
scattering of iso-fractalpolydisperse
aggregates is proportional to the secondmoment of the size-distribution. See for
example
DJORDJEVIC Z B, Ph. D. Thesis MI T(1984) for a detailed presentation of this result