• Aucun résultat trouvé

On the measurement of the fractal dimension of aggregated particles by electron microscopy : experimental method, corrections and comparison with numerical models

N/A
N/A
Protected

Academic year: 2021

Partager "On the measurement of the fractal dimension of aggregated particles by electron microscopy : experimental method, corrections and comparison with numerical models"

Copied!
11
0
0

Texte intégral

(1)

HAL Id: jpa-00210394

https://hal.archives-ouvertes.fr/jpa-00210394

Submitted on 1 Jan 1986

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.

On the measurement of the fractal dimension of aggregated particles by electron microscopy :

experimental method, corrections and comparison with numerical models

M. Tence, J.P. Chevalier, R. Jullien

To cite this version:

M. Tence, J.P. Chevalier, R. Jullien. On the measurement of the fractal dimension of aggregated particles by electron microscopy : experimental method, corrections and comparison with numerical models. Journal de Physique, 1986, 47 (11), pp.1989-1998. �10.1051/jphys:0198600470110198900�.

�jpa-00210394�

(2)

1989

On the measurement of the fractal dimension of aggregated particles by

electron microscopy : experimental method, corrections and comparison

with numerical models

M. Tence (*), J. P. Chevalier (+ ) and R. Jullien (*)

(*) Laboratoire de Physique des Solides, Bât. 510, Université Paris-Sud, 91405 Orsay Cedex, France (+) C.E.C.M.-C.N.R.S., 15 rue G. Urbain, 94407 Vitry-sur-Seine Cedex, France

Résumé. - Des images digitalisées en fond noir annulaire d’agrégats polydispersés de billes de fer sont obtenues avec un microscope électronique piloté par ordinateur. On discute des différentes méthodes capables

d’extraire la dimension fractale des agrégats et on analyse les différentes sources d’erreur. Les résultats sont

comparés avec des simulations à l’ordinateur utilisant une version polydisperse du modèle d’ agrégation par

collage d’amas. Les simulations montrent que la polydispersité n’affecte pas la valeur de la dimension fractale.

Le résultat expérimental (D =1,9 ± 0,1 ) est consistant avec un modèle d’agrégation par collage d’amas avec trajectoires linéaires.

Abstract. - Digital annular dark-field images of aggregated iron polydisperse particles are obtained using a computer-controlled STEM. Different methods are discussed on how to extract the fractal dimension of the aggregates, and the sources of error are analysed. The results are compared with computer simulations on a

polydisperse version of the cluster-cluster aggregation model. Simulations show that polydispersity does not

affect the fractal dimension. The experimental result for the fractal dimension ( D =1.9 ± 0.1) is consistent with the cluster-cluster model with linear trajectories.

J. Physique 47 (1986) 1989-1998 NOVEMBRE 1986,

Classification

Physics Abstracts

61.14 - 05.40

1. Introduction.

The work of Forrest and Witten [1] opened a new

area of study in electron microscopy, for small complicated objects, such as aggregated particles.

Through measurements of micrographs of smoke- particle aggregates, they showed that long-range

correlations exist and that these obey a power law.

The exponent in this power law was identified as the fractal (or Hausdorff) dimension [2] of the aggregate

and hence characteristic of a specific aggregation

mechanism. Since this study, considerable interest has focussed on aggregation phenomena, both through computer modelling [e.g. 3-6] and through

further experimental studies using imaging in the

electron microscope or various scattering experi-

ments [e.g. 7-11]. Owing to this combined (theoreti-

cal and experimental) approach it is now possible to

compare measured fractal dimensions in real aggre-

gates with those from computer models involving

different aggregation mechanisms and hence gain insight on the mechanism involved for specific preparation conditions. However for such a compari-

son to be valid, it is essential to examine critically

how measurements are made and what are the

sources of systematic error. The aim of this paper is

to test such a measurement on more general aggrega- tes than considered previously (particles having

different sizes and overlapping frequently) and using

the most suitable electron microscopy techniques (Digital Annular Dark-field imaging in a Scanning

Transmission Electron Microscope). We will then

discuss the possible systematic errors, their effects and how these can be corrected and finally compare the results with those obtained through computer

modelling of a polydispersed system in a cluster- cluster aggregation process [5]. We have published preliminary results of this work [12], emphasizing

the techniques of electron microscopy which have

been used.

2. Specimens and specimen preparation.

The iron powder used in this study was prepared by

a novel cryogenic technique [13]. A solid iron ingot

is levitation melted by radio frequency induction, in

a cryogenic environment (typically liquid argon). At

the interface between the liquid argon and the molten iron a turbulent reaction occurs. Here we

believe that either vapourized iron or molten iron beads are quenched by the cryogenic liquid. The

« boiled off » argon then flows through several

decantation tanks containing hexane. From the mud which settles at the bottom of the tanks, a black powder can be obtained by drying.

Specimens for electron microscopy are prepared

either by mixing the powder with ethanol in a watch

glass, or by ultrasonically dispersing the powder in

Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:0198600470110198900

(3)

ethanol (using a standard ultrasonic cleaner). A

carbon on formvar electron microscope grid is then dipped in the ethanol suspension and allowed to dry.

Although we found no difference in aggregate appearance between these two slightly different

methods of preparation, the aggregates appear to be

more homogeneously distributed on the grid in the

latter case. We infer that this method does not

substantially break up the aggregates or allow the further combination of aggregates or particles. This

fact was confirmed by the observations. The measu- rements presented here were obtained by ultrasoni- cally dispersing the powder.

Bright-Field imaging in a conventional transmis- sion electron microscope (Fig. 1) shows that the iron

powder consists of aggregates of spherical particles.

The particles have sizes ranging from about 100 A to 2 000 A in diameter, with a mean size of about 500 A. We also note that particles overlap frequen- tly. Furthermore, some of the beads are not spheri- cal, but appear to be welded together, suggesting

that aggregation occurred at high temperature when the metal is molten or just solidifying. These points (range of size, overlap and different shapes) show

that methods of particle counting are likely to be

difficult in this case.

Diffraction patterns from such particles show that they have a bcc structure as expected. The patterns also exhibit extra rings corresponding to Fe3o4.

Dark-Field imaging demonstrates that this oxide is in the form of a thin coating ( - 15 A or so ) on the particles. This coating is not expected to affect the

measurements.

3. Experimental procedure and choice of signal for imaging.

Previous work on the measurement of the fractal dimension of aggregates from electron micrographs

first used bright-field imaging techniques and then employed either counting the particles (if these are

Fig. 1 - - Transmission electron bright-field image of a single aggregate. It is important to note that particles of

the same size need not have the same contrast and that extinctions (thickness fringes) occur in the larger particles.

all of the same size) [8-10] or digitizing the image

into a binary distribution [1]. Other authors have also suggested elaborate image treatments to pro- duce binary distributions from bright field images [14]. These treatments are all essential if one uses

bright field imaging, since the bright-field signal is

not simply proportional to the amount of matter present. This effect is well known (see e.g. [15]) and

is due to the strong interaction between the electrons and the sample (in this case the most severe limita-

tion is due to the very strong dependence of the scattering on the Bragg orientation of the crystallyne particles). Such effects can be clearly seen in the bright field image (Fig. 1) and give rise to thickness fringes and to variable intensity for particles of the

same size. The same effects occur in conventional dark-field images, where one tilts the incident beam and forms the image with part of the polycrystalline Debye ring. Here only the particles which are in the

correct Bragg orientation for this particular selected angular range (both radial and azimuthal) will be imaged. These difficulties are further compounded

when the particles overlap.

In our case, the nature of the specimens used (particles of varying size, together with frequent overlap) precluded the use of any such treatment based on conventional bright or dark field images.

Furthermore we wished to try the most appropriate

of techniques recently developed. We therefore chose to use the signal from an Annular Dark-Field detector using a 100 kV scanning transmission elec-

tron microscope equipped with a field emission gun

(FEG-STEM). Furthermore since the images are to

be processed, they were acquired using a digital

scan. The instrument and the Annular Dark-Field

(ADF) detector have been fully described previously [16-18]. The digital acquisition and processing unit,

which is linked to the microscope has also been

described elsewhere [19].

The signal from the ADF detector has considera- ble advantages for this kind of work over both the

bright- and dark-field conventional images, whose

contrast is produced essentially by Bragg scattered (elastic) electrons. The ADF signal is from high angle elastically scattered electrons, inelastically

scattered electrons subsequently elastically scattered

to high angles, and high-angle elastically scattered

electrons which are then inelastically scattered. (The

inelastic scattering is largely small angle). Crystalline

contrast effects are strongly attenuated since we now

average the contrast over a large angular range (0 to

2 1T in azimuth and from one or two to about 10

Bragg angles). This means that the signal can be

considered to vary linearly with the amount of iron present up to thicknesses of about 700 A [20]. In any

case it varies monotonically to much greater thicknesses, although the linear law then becomes a

much poorer approximation.

To record images with a sufficiently large dynamic

range, due to both low intensity from the support film and high values from large particles, the output from the ADF detector is encoded into numerical values using a voltage-frequency convertor. This

(4)

1991

averages the signal over the dwell time per pixel and

hence allows high signal intensities (i.e. more than 108 counts s-1) to be recorded, when individual pulse counting is no longer feasible.

Figure 2 is an example of a digitally recorded

ADF image from a typical aggregate. We note that the overlapping particles are clearly taken into account, that particles of the same size have the

same intensity and that large particles show

thickness countours corresponding to spherical shape

with no drastic plateau effect due to failure of the linear law for large thicknesses. Thus this signal is

most appropriate to measure mass scaling in aggrega- tes.

For adequate comparison with the underlying theory and with regards to self-similarity, we have attempted to measure as large a range of aggregate size as possible, and hence we have recorded images

at 4 fixed electron optical magnifications (x 20 000 ;

x 50 000 ; x 100 000 ; x 200 000). These have been calibrated and have been found to be within 3 %,

with distortion (i.e. difference between horizontal and vertical magnification) of less than 2.5 %. To enable good resolution of the smallest particles at

the lowest magnification, 512 x 512 digital images proved necessary and this was maintained for the other magnifications, as well. To compare data from

images taken at different magnifications (i.e. because

an image pixel does not correspond to the same

value of area, and hence the intensity is not equiva-

lent to the same specimen mass) the intensity of each pixel p is normalized by y a factor M1 where M is the

magnification [21].

4. Image treatment and calculation of the fractal dimension.

The 512 x 512 images, once recorded and stored on

tape, are treated in the following manner. After

normalization for different magnifications, the computer program written for this study finds the

Fig. 2. - Scanning transmission annular dark field image.

The contrast is now approximately linearly related to the

mass.

outer limit of the aggregate (imaged one by one, i.e.

one per frame) and draws a close fitting rectangle

around it. If there are several aggregates in the frame (i.e. one large one and a small one close to it)

this routine can distinguish them as long as they do

not interpenetrate. The average intensity of the image outside this rectangle is then measured and this value is then used for the value of the back-

ground subtraction, i.e. to account for the scattering

from the carbon/formvar support film. Since the support film is the same for all aggregates analysed

on the same grid, this value is also used to normalize the intensity for each aggregate for fluctuations in the incident electron beam, and hence enable all the aggregates to be compared. This is, in general,

necessary for all electron micrographs, but especially

so here, since the intensity from the Field Emission Gun can decrease by a factor of 2 over several minutes and the tip requires frequent re-generation.

The « stripped and normalized » images of aggre- gates are analysed to yield fractal dimensions by two

different methods, both based on a scaling relation

between mass and size. In the first method used,

which is similar to the method proposed by Forrest

and Witten [1], the scaling relation is probed inter- nally for each aggregate. That is, for a given origin

chosen to be on a particle in the central portion of

the aggregate, concentric squares are drawn, and

within each square, the intensity (assumed now to be linearly proportional to the mass) is summed. To minimize the effects of the arbitrary choice of origin,

three origins are in fact used and for each size of square the arithmetic mean of the mass, for the three

origins, is taken. This tends to reduce the effects

owing to the breakdown of scaling and to the finite

size of the particles for small squares. From In/In

plots of mass versus size of square, a curve is

obtained, the slop of which gives a value of D, the fractal dimension. We will call this the « nesting

squares » method.

The second method used simply seeks to verify the scaling relation between mass and size for all the aggregates. Thus we simply integrate the intensity

over the whole aggregate and relate this to a measure of the size of the aggregate. We have chosen to use the value of J a2 + b2 for the size,

where a and b are the sides of the close fitting rectangle used previously. From a log-log plot of

mass versus size for all the aggregates measured, we

obtain a set of points, which can be fitted to a

straight line, again with slope D. We have also tried

using other measures of aggregate size (e. g.

but we have found that this makes only a very slight difference on the value of the slope D (about 10 % of the standard deviation on

the value).

5. Results of the measurements.

Altogether 36 aggregates were measured with an approximately normal distribution in size (or rather operational magnification). From this set, 3 aggrega-

(5)

tes were excluded. One because they were too many other small aggregates, another because the image

was saturated (streaking in the background) and a

third because the aggregate was very small (about

50 particles) and essentially linear and so judged non-representative. Of the 33 remaining, 3 consisted of a large and a small aggregate on the same frame.

The small aggregates in these 3 frames were also eliminated since we think that the magnification

used did not lead to appropriate sampling. This then

leaves 33 aggregates in the analysis, with two orders

of magnitude in « mass » between the smallest and

the largest.

Concerning what we call the « nesting squares » method, a number of problems were encountered.

The log-log plots produced curves which could be considered to have a linear central point. This was

not always clear, since fluctuations occurred even in the central portion. Some examples are given in our preliminary paper [12]. In that work [12] we had

estimated the slope of these linear portions using a

ruler. Here we have carried out a least-squares fit,

but without any weighting (cf. Forrest and Witten [1]

who weighted according to the intensity in the square). However, if we decide to discard part of the

curve, either the top, bottom or both since it is

expected that these will not be linear and that this

departure from linearity can be accounted for by

discrete finite size effects and by truncation effects,

we find that we obtain a continuous variation in the values of D depending on how much of the curve is discarded. For example if we discard only the top end (eliminating truncation effects) and assuming

that the averaging of different origins has reduced

particle size effects at the lower end, then we obtain

the following values from a least squares fit :

The percentages given in the bracket correspond to

the percentage of the total size of the curve that has

Fig. 3. - Log-log plot of mass/size for 33 aggregates measured. A least-squares fit on the slope yields

D =1.89 ± 0.09.

been fitted. It is thus clear that it is difficult to decide which value can be. considered as « correct » and

why. This will be discussed fully in the next section,

but it should be noted that the values obtained here

by this method are all systematically smaller than that estimated ( D = 1.8 ± 0.2) in our preliminary

report [12]. This is due to two factors, one which was

a systematic overestimate of the slope estimated

with a ruler (!) and the other is that we now use a correct background subtraction technique. We now

subtract a measured average background, which is necessarily larger than the minimum pixel value used

before. This will necessarily lead to a smaller value

of the fractal dimension.

The result for the total mass/size scaling method

for all the aggregates is more satisfactory. Figure 3 is

a log-log plot for the 33 aggregates studied. The scatter of point is quite considerable, but a least squares fit is physically reasonable and this gives :

The physical meaning of this value will be discussed later in terms of the various models for aggregation.

However several remarks about the measures can now be made. Firstly the last value is significantly

different from the value obtained by the « nesting

squares » method, and secondly it is slightly less than

the value we had previously reported [12] of

D = 2.0 ± 0.1. This decrease can be explained by

the improved method of background subtraction discussed above. The problem of background sub-

traction has not been of relevance to other authors up to now since they have either counted particles or

binarised their images.

6. Projection and finite size effects in the K nesting

squares w method.

Many authors [e.g. 2] state that the projection of a 3-

dimensional aggregate to 2-dimensions will not affect the value of the fractal dimension D calculated by

the « nesting squares » method for values of D less than 2. For D > 2, this method cannot give more

than an effective fractal dimension equal to 2, whilst the mass/size method would yield the correct value

in principle. In fact this reasoning is correct only for

infinite aggregates, and for D less than but close to 2, finite size effects will affect the projection, and

hence the results obtained using the « nesting squa-

res » method.

This idea can be demonstrated by considering

some kind of « ideal » continuous representation of

a finite-sized 3-d fractal of maximum radius R and with spherical symmetry, defined by the following

mass density :

The coefficient in front of the term rD -3 is calculated

so as to normalize the density to the total mass M of

the aggregate. The consideration of an abrupt cut off

(6)

1993

at r = R is not necessarily very physical, but we

believe that this will only be a secondary effect. The

essential point is that this cut-off exists, thus taking

the finite size into account explicitly.

This can be easily projected by calculating the

mass m contained in a cylinder of radius ), whose

axis goes through the origin. Integrating over the variable r, one has:

with

After some straightfordward changes of variables,

one finds

This expression can be used to calculate m ( c ) numerically. To see how this quantity behave, when

C -+ 0, one can rewrite it as :

and then, when ( --+ 0

When ( --+ 0 one can expand the diverging integral :

Thus when § - 0, m ( ( ) contains two competing

terms in §2 and ( D. One can write in general:

and

This can be seen on the log-log curves shown in figure 4, when, for D > 2, the slope of the curves, as

In i tends to - oo, becomes independent of D. A

R

more quantitative picture can be obtained by defi- ning a size dependent dimension Deff ( f ) ; this is given by the slope of the log-log curve m ( § ) for a

given f. R Thus:

Fig. 4. - Results for the mass m ( § ) contained in a

cylinder of radius in the case of a fractal « sphere » with finite cut-off R. The figure shows a plot of In m ( § ) (where M is the total mass of the sphere), against In

for various values of D.

R

Figure 5 is a plot of D eff ( § ) versus D for different

This curves gives an idea of the error made in the

« nesting square » method, since from the curves obtained by this method, values of De ff can be obtained ; i.e. when the effective fractal dimension is evaluated at a point corresponding to a square

going to half the edge distance, a quarter the edge distance, etc. For example one sees that for a

reasonable value of eIR, of order 1/8 or 1/16, the

difference between the true fractal dimension and the effective (or apparent) one is of the order of 0.2

Références

Documents relatifs

The operations of burnishing ball were performed according to plans ofexperiments of ‘Box–Behnken’, an optimal regime was obtained and a mathematical model wascleared for predicting

Typically, biomarkers obtained from the Zero Set and the Adapted Box algorithms have shown good discriminating power in the early detection and differential diagnosis of

Dans les entreprises de 10 salariés ou plus des secteurs concurrentiels(1), le nombre moyen d'heures supplé- mentaires par salarié à temps complet(2) déclarées par les entreprises

L’approximation des électrons sans masse qui nous intéresse dans cette thèse mène également à des modèles fluides pour la dynamique des électrons.. Cependant, la

STRUCTURES BY MEANS OF ELECTRON MICROSCOPY AND ELECTRON DIFFRACTION.. - L'utilite de la microscopie Blectronique et de la diffraction electronique dans l'ktude des transitions

The difference in the last network with D = I-S is easily understood when one thinks that its function is to transport people from or to the center of urban ensemble, the town of

Dénoncer, le racisme de MEETIC, rester entre les gens de notre race, à première vu ou c'est ce qu'il préconisent, parce que lors des rendez vous, je les ai tous contaminer

Cette stratégie a par exemple été utilisée lors de la synthèse de la (-)-halosaline : une métathèse cyclisante entre une des doubles liaisons terminales et le cycle central à