Page 1
Water Research 37 (2003) 873–883
Changes in fractal dimension during aggregation
Rajat K. Chakrabortia,*, Kevin H. Gardnerb, Joseph F. Atkinsona,John E. Van Benschotena
aDepartment of Civil, Structural and Environmental Engineering, State University of New York at Buffalo, 207 Jarvis Hall, Buffalo,
New York 14260, USAbDepartment of Civil Engineering, University of New Hampshire, Durham NH 03824, USA
Abstract
Experiments were performed to evaluate temporal changes in the fractal dimension of aggregates formed during
flocculation of an initially monodisperse suspension of latex microspheres. Particle size distributions and aggregate
geometrical information at different mixing times were obtained using a non-intrusive optical sampling and digital
image analysis technique, under variable conditions of mixing speed, coagulant (alum) dose and particle concentration.
Pixel resolution required to determine aggregate size and geometric measures including the fractal dimension is
discussed and a quantitative measure of accuracy is developed. The two-dimensional fractal dimension was found to
range from 1.94 to 1.48, corresponding to aggregates that are either relatively compact or loosely structured,
respectively. Changes in fractal dimension are explained using a conceptual model, which describes changes in fractal
dimension associated with aggregate growth and changes in aggregate structure. For aggregation of an initially
monodisperse suspension, the fractal dimension was found to decrease over time in the initial stages of floc formation.
r 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Coagulation–flocculation; Fractal dimension; Particle size distribution; Shear rate; Image analysis; Pixel resolution; Alum
1. Introduction
Coagulation is a common process used in water and
wastewater treatment plants. It also affects the behavior
of particles in natural systems. Factors affecting this
process include coagulant type and dose, solution pH,
mixing intensity and particle concentration [1–3]. One of
the most important of these factors is mixing intensity
(fluid shear), which promotes interactions between
aggregates and particles, and affects overall settling
characteristics [4,5]. However, if mixing is too intense,
particles may not flocculate efficiently [6]. Spicer and
Pratsinis [7] studied the effect of various shear rates on
coagulation and found that aggregates reach an
equilibrium, steady-state aggregate structure and floc
size distribution relatively quickly at higher shear rates.
They concluded that floc breakage was the main process
responsible for maintaining a stable particle size and
limiting further growth.
During coagulation, characteristics of the particles
change as they interact with the coagulant and with each
other. The collision frequency and growth of aggregates
depend on the relative size of the colliding particles, their
surface charge and roughness, local shear forces and the
suspending electrolyte [2,3,8]. Particle concentration also
affects the rate at which particles and aggregates collide
and is therefore important in determining changes in
aggregate size.
Many models have been developed to simulate
particle aggregation. Traditional approaches, based on
the Smoluchowski equation, consider the aggregates as
impermeable spheres (e.g., [9,10]). However, recent
studies have shown that, in reality, flocs consist of
multi-branched structures that are not consistent with
the floc structure described by classical Euclidean
geometry [11,12]. Fractal concepts have provided a
new way of describing floc geometry and various
physical properties such as density, porosity and settling*Corresponding author. Fax: +1-716-645-3667.
E-mail address: [email protected] (R.K. Chakraborti).
0043-1354/03/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 3 - 1 3 5 4 ( 0 2 ) 0 0 3 7 9 - 2
Page 2
velocity [12–16]. However, if the application of fractal
concepts is to be useful and provide an improvement in
aggregation modeling, then simple and reliable methods
of measuring fractal dimensions, as well as methods to
understand the relationship between fractal dimensions
and aggregation processes, must be developed. The main
goal of the present study was to demonstrate such a
measurement procedure and to develop a descriptive
model that can illustrate expected changes in fractal
dimension during an aggregation experiment.
2. Aggregate structure and fractal dimension
Various characteristics of aggregates derived using
fractal geometry should be useful in improving our
understanding of aggregation processes and modeling.
In fact, several studies have already shown how collision
frequency functions used in particle aggregation models
are strongly dependent on the structure of the aggregate
and the corresponding fractal dimensions [8,11,13].
There are several physical interpretations of this result,
including the fact that a fractal aggregate is spread out
over a larger space than a single sphere of equal mass,
and that flow patterns around and through an aggregate
are significantly different than flow around a sphere. It is
possible that temporal changes in collision efficiency
also may be affected by floc size and structure, with a
corresponding dependence on fractal dimension, though
this relationship has not yet been investigated.
Fractal aggregation theory provides a means of
expressing the degree to which primary particles fill the
space within the nominal volume occupied by an
aggregate [16]. Fractal dimensions may be defined in
linear, planar or volumetric terms, resulting in so-called
one-, two- or three-dimensional values, respectively. For
example, considering the planar aggregate shown in
Fig. 1a, where each solid circle represents the mass of a
primary particle, it is clear that circles of different radii
drawn about the center of the aggregate enclose different
masses. The two-dimensional fractal dimension, D2; isdefined in terms of the relationship between the increase
in radius and the corresponding increase in mass
contained within the circle or, in geometrical terms,
AprD2 ð1Þ
where A is the sum of areas of all primary particles
contained within a circle of radius r: Thus, D2 may be
found as the slope of a plot of (log A) versus (log r).
Note that in Euclidean geometry, for solid circles, D2 ¼2: One- and three-dimensional fractal dimensions may
be defined in an analogous manner, but the present
discussion will focus on D2 since in this study area was
evaluated directly, using an image analysis technique. As
aggregates grow (Fig. 1b), porosity increases and the
number concentration of primary particles occupying
the nominal volume of the aggregate decreases, resulting
in a decrease in the fractal dimension.
In fractal simulation models, aggregation is usually an
irreversible process, so that once two particles stick to
each other, they do not subsequently come apart. This is
a critical limitation, when compared with naturally
occurring processes, where aggregate size is limited by
local physical-chemical conditions. In reality, there is a
restructuring of primary particles within an aggregate
due to break-up and re-formation that occurs in
response to ambient shear or other conditions [6,17].
This process is illustrated in Fig. 1b. Over time, this can
lead to stronger and more compact aggregates, with an
associated higher fractal dimension. In other words,
fractal dimension increases when the primary particles in
an aggregate are arranged more compactly. This change
in fractal dimension appears in the data of Kaye et al.
[18] and Spicer and Pratsinis [7], for example.
Another consideration for development of fractal-
based descriptions of aggregation is the variation of
fractal dimensions among aggregates in a given suspen-
sion. Based partly on this variation, as well as noting
that theoretical work has largely ignored the issue of
rearrangement of primary particles, Meakin [19] sug-
gested that traditional fractal models (e.g., diffusion-
limited aggregation or its various derivatives—see
below) can be used to develop theoretical values for
aggregate fractal dimensions, but that natural aggre-
gates should be looked at only in a composite, or
statistical sense. This is because individual aggregates
are not necessarily expected to exhibit the theoretical
Fig. 1. (a) Variation of contained mass as a function of
increasing radius. (b) Effect of breakup and reattachment,
leading eventually to stronger and more compact aggregate
structure, with associated higher fractal dimension.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883874
Page 3
scaling of an ideal fractal. The fractal dimensions for a
sample of particles can be developed following an
analogous approach as for the single particle illustrated
in Fig. 1a. In this case, however, different aggregates in a
sample are analyzed, each with its own respective length
and area. Since aggregates are not circular (or spherical),
the longest dimension, l; is usually used in place of r
[11,15,20], so
AplD2 ð2Þ
and D2 is then the slope of a plot of (log A) versus (log l).
This is the approach followed in the present study.
3. Materials and methods
A non-intrusive optical sampling technique was used
to obtain digital images of particles, which were then
analyzed to develop particle size distributions, geome-
trical properties, and calculations of the fractal dimen-
sion. The basic procedures were adapted from the recent
work of Chakraborti et al. [11], who studied aggregate
characteristics produced after mixing suspensions with
different coagulant (alum) doses. A major advantage of
their method is that it requires no sample handling, so
there is no concern for disturbing the floc characteristics
during measurement. The main difference between their
study and the present work is that in the present
experiments the images were obtained while mixing was
still taking place. A similar measurement technique was
used by Kramer and Clark [21] to observe dynamic (time
dependent) particle size behavior for particles suspended
in a turbulent flow.
The experimental apparatus consisted of a stirred 2L
jar (Phipps and Bird, Richmond, VA, USA) with
variable mixing speed, an automated stroboscopic lamp
to illuminate suspended particles in the jar and a digital
charge coupled device (CCD) camera (Kodak Megaplus
model 1.4) to capture particle images. The camera was
placed on the opposite side of the mixing jar from the
lamp, so that backlit shadows of particles were
produced. A PC served to control the camera and
provided storage for the particle images. Further details
of the experimental setup can be found in Chakraborti
et al. [12].
Monosized polystyrene latex microspheres with a
density of 1.05 gm/mL (Duke Scientific Corporation,
Palo Alto, CA, USA) were used as the primary particles
for the experiments. The nominal particle diameter was
9.975mm, with a standard deviation of 70.061mm.Particles were taken from a 15mL sample of aqueous
suspension with 0.2% solids content (manufacturer’s
specification) and used without any pre-treatment. The
number concentration of particles in the concentrated
suspension was 3.66� 106 particles/mL (710%). Ali-
quots of 0.06 or 0.1mL of the suspension were added to
the mixing jar along with 1L of deionized water,
resulting in initial number concentrations of 220 and
366 particles/mL, respectively.
4. Procedures
Images were taken to examine aggregate geometry
and size distributions at different times during each
experiment. The images were analyzed to track changes
in aggregate morphology for a given experiment, as well
as differences between experiments resulting from
varying coagulant (alum) dose, particle concentration
and mixing speed, or shear rate. Two levels for each of
these variables were tested, resulting in a total of eight
experiments, as summarized in Table 1. The analyses are
reported in terms of the fractal dimension and asso-
ciated particle size distribution.
A stock solution of alum was prepared by dissolving
Al2(SO4)3 � 18H2O (Fisher Scientific, Pittsburgh, PA,
USA) in deionized water to a concentration of 0.1M
(0.2M as aluminum). Alum concentrations of 3.33 and
5.33mg/L were used in the experiments. For each test,
after an initial rapid mixing period (with G ¼ 100 S�1,
where G is the velocity gradient) for 1min, the mixing
speed was reduced to either G ¼ 20 or 80 S�1 and
continued until the end of the experiment. These values
are within the normal range used for treatment
processes. All tests were conducted at room temperature
(20–231C) and a constant pH of 6.5 was maintained by
adding acid or base as required. Measurements were
taken at 10, 20 and 30min of flocculation.
The images were recorded through an imaging
window of size 2mm� 2mm. The shutter speed was
synchronized with the strobe pulses using the camera
control software of a standard particle image veloci-
meter (PIV) system (TSI Inc., St. Paul, MN, USA). A
public domain software package, NIH-Image (National
Institutes of Health, Bethesda, MD, USA), was used for
image analysis, and pixel values recorded in the image
were calibrated to the appropriate length scale using a
stage micrometer (in this study, 540 pixels=1mm).
Further details of the image acquisition and processing
techniques are described by Chakraborti et al. [12].
Before processing a particular image, image thresh-
olding was applied, where pixels were classified into two
categories, either background or particle. This step
effectively filters out random noise, as well as particles
that are not well focused. The overall image is rendered
binary, where particles with sharp edges are identified
and separated from the background. To extract the
greatest amount of information, this step requires good
quality images with the least distortion of the edge of the
particles [22,23]. A number of tests were performed with
the NIH-Image software to determine the best settings
for capturing images with good resolution, by adjusting
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883 875
Page 4
the position and frequency of the strobe light and the
camera shutter speed, depending on the particular
sample conditions. Using similar settings Cheng [22]
found that particles as small as 6mm could be resolved.
A well-focused image permits analysis of more
particles in one threshold operation, whereas an image
with relatively poor resolution requires a higher thresh-
old level to provide contrast between the image and the
background, resulting in a possible loss of information.
To avoid loss of data, several images were taken at each
measurement time by moving the camera in one plane so
that different locations of the jar could be captured. This
was done with a light-camera assembly that maintained
constant distance between the camera, light and jar. This
procedure provided guidance for choosing the best light
and camera shutter speed, for the particular residual
turbidity of a given test, and images that provided the
largest number of particles at each measurement time
were selected for further analysis.
Fig. 2 shows the aggregate parameters obtained from
the NIH-Image software following application of
thresholding. An ellipse was fitted to each aggregate in
such a way that the area of the ellipse was equal to the
area of the aggregate and the major axis of the ellipse
was used as an estimate for l (Eq. (2)).
5. Results and discussion
5.1. Pixel resolution
The ability of the system to accurately resolve the
smallest particles expected in a sample must be
determined in order to evaluate the scale to which
Table 1
Physical-chemical parameters used for the experiments and resulting two-dimensional fractal dimensions (D2) and particle size
distribution (PSD); n is number of aggregates
Experiment
number
Alum
dose
(mg/L)
Shear
rate
G (S�1)
Particle
conc.
(#/mL)
Observation
times
(min)
Mode in
PSD
(mm)
Fractal dimensions n
D2 R2
1 5.33 20 366 10 14 1.8370.10 0.89 86
20 18 1.6870.08 0.91 77
30 22 1.5170.09 0.89 65
2 5.33 80 366 10 14 1.8270.10 0.87 97
20 16 1.7070.08 0.91 95
30 25 1.6470.06 0.90 84
3 3.33 20 220 10 14 1.9170.21 0.72 62
20 18 1.8270.18 0.81 51
30 18 1.7370.16 0.85 54
4 3.33 80 220 10 14 1.9470.18 0.85 47
20 18 1.8870.22 0.77 45
30 18 1.8270.19 0.83 43
5 5.33 20 220 10 18 1.7270.08 0.92 80
20 24 1.6270.08 0.91 76
30 26 1.4870.08 0.91 66
6 5.33 80 220 10 18 1.7570.18 0.78 62
20 20 1.6470.12 0.84 60
30 22 1.6070.11 0.82 52
7 3.33 20 366 10 12 1.8470.09 0.81 91
20 16 1.7770.08 0.88 72
30 18 1.7070.09 0.86 58
8 3.33 80 366 10 16 1.9170.10 0.72 83
20 18 1.8670.09 0.84 76
30 20 1.7570.09 0.85 66
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883876
Page 5
geometrical properties such as the fractal dimension may
be determined. Depending on pixel size, relative to
particle size, the area of a small particle or aggregate
may be overestimated, which could result in a smaller
calculated value for D2 (i.e., if areas are overestimated
for the smallest aggregates, the resulting slope of the line
correlating area with length could decrease). In order to
illustrate this issue, consider first a plain circle,
approximated by different numbers of pixels (Fig. 3a).
When the pixel length is equivalent to the circle diameter
(first panel), then depending on the placement of the
circle relative to the pixels, from one to four pixels may
be used to ‘‘register’’ the circle. If the circle diameter is
taken to be one unit, then the ratio of the registered pixel
area (Ap) to the actual area of the circle (A0) is between
1.27 and 5.09. As the pixel resolution increases, i.e.,
more pixels are used across the circle diameter, the ratio
(Ap=A0) approaches one and its variability decreases,
since the ratio becomes less dependent on the placement
of the circle relative to the pixel locations. Fig. 4
illustrates this variability, where N is defined as the
ratio of circle diameter to pixel ‘‘length’’. In other words,
N is the number of pixels required to traverse one circle
diameter. Note that in Fig. 4 there are several area
values plotted for each resolution. These areas corre-
spond to different placement of the object, relative to the
pixel boundaries, as in the first panel of Fig. 3.
The issue of required pixel resolution was explored
further by considering a simple aggregate consisting of
three primary ‘‘particles’’, each represented by a circle as
shown in Fig. 3b. Following a similar procedure as for
the single circle, the area ratio for this case is also plotted
in Fig. 4, where N is the number of pixels required to
traverse the diameter of one of the primary circles. For
(a)
(b)
Fig. 3. (a) Variation of pixel resolution for a single circular particle; the first panel shows three different possibilities for measuring a
single primary particle, resulting in three different area values; (b) variation of pixel resolution for a simple aggregate.
Fig. 2. Parameters measured for an individual particle by image processing software.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883 877
Page 6
both cases shown in Fig. 4, the area ratio is less than
1.02 when N ¼ 20: For the resolution of the present
study, N ¼ 5:4; so that a single primary particle is
represented by approximately 29 pixels and the average
area ratio is 1.3. This implies the area measurement for a
single primary particle may be overestimated by about
30%. However, the error decreases rapidly with
increasing resolution (Fig. 4), and it also should be
noted that most aggregates analyzed were larger than
the primary particles, so the average error in area
measurement is reduced.
To further investigate the potential effect of pixel
resolution on calculations of D2; the data for aggregate
length and area for a sample were divided into sub-
ranges, based on length, and D2 was calculated for each
sub-range and compared with the values obtained for
the sample taken as a whole. These calculations are
shown in Table 2, for experiments 1 and 2, where it can
be seen that there is negligible difference in D2 values
calculated over the different size ranges, although there
is a significant change in D2 for the different times. This
comparison indicates that potential problems with area
calculations for the smaller aggregates did not influence
D2 values for the sample as a whole, although it is
possible that D2 values are slightly underestimated for
the smaller size sub-range (note that an overestimation
in area for the smaller aggregates would tend to result
in smaller slopes, and therefore smaller D2; for
the regression of Eq. (2)). Even smaller divisions in size
ranges were not considered, due to limited data available
within each sub-range (see Table 2).
5.2. Images
Images captured at the three different times show the
evolution of aggregate structure during flocculation.
After 10min of mixing, the initially monodispersed latex
particles formed flocs of variable shapes and sizes
(Fig. 5). With additional time, larger flocs were formed
and examples of some of these flocs are shown in
Figs. 5b and c. Compared with images taken after
10min, the images at later stages were composed of
fewer but larger particles. The larger aggregates in the
image taken after 30min (Fig. 5c) appear to be more
porous and spread out than the smaller aggregates
observed at earlier times and should be associated with
lower fractal dimensions, as previously noted.
5.3. Particle size distribution
Fig. 6 shows the evolution of the particle size
distribution for experiments 1 and 2, using the major
axis of the ellipse fitted to the aggregates as a
characteristic length (l). Similar characteristics of the
distributions were observed for the other experiments
and are not shown here. The same alum dose and
particle concentration were used in these two experi-
ments, with only the mixing speed varied (see Table 1).
For both experiments, there is a clear movement of the
peak of the size distribution (the mode) towards larger
size classes with time. This movement is rather regular
for experiment 1 (Fig. 6a), whereas the peak moved
more at the 30min measurement for experiment 2,
relative to earlier times (Fig. 6b). At 30min, the mode
for experiment 2 is slightly larger than for experiment 1.
However, a greater portion of the distribution in
experiment 1 is in larger sizes, suggesting that the lower
G condition was more favorable for aggregation.
Plots of particle size distribution were generated for
all the experiments and they all showed the occurrence
of a broad range of particle sizes, particularly at later
times when both large and small aggregates were
present. A heterogeneous suspension is believed to lead
to higher aggregation rates [4,8,16]. This appears to be
consistent with the changes in size distribution and
gradual movement of the peak of the size distribution
plots. In other words, on average there is a slightly
greater change in the position of the peak size between
0
0.5
1
1.5
2
2.5
3
3.5
4
0 10 0 30 0 50
N
Ap/
Ao
single circle
cluster
Fig. 4. Dependence of area ratio on pixel resolution.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883878
Page 7
20 and 30min, relative to the change between 10 and
20min (see Table 1).
5.4. Fractal dimension
Calculated values of D2 for the eight experiments are
shown in Table 1. The mean values and 95% confidence
levels (7values in Table 1) were calculated from
regression analyses based on Eq. (2). The correlation
coefficients (R2) for all regressions range between 0.72
and 0.92, indicating a good correlation between logA
and log l: In addition, values for D2 compare well with
previous studies examining fractal dimensions for flocs
produced using alum. For example, Gorczyca and
Ganczarczyk [20] reported D2 values for alum flocs
formed in inorganic clay and mineral suspensions to be
between 1.71 and 1.97 (70.05) and Li and Ganczarczyk
[24] calculated fractal dimensions of alum flocs to be
between 1.59 and 1.97 from the measurements carried
out by Tambo and Watanabe [25]. Except for the 30-min
measurements of experiments 1 and 5, the present values
fall within the ranges of these earlier studies. It also is
interesting to note that these values are consistent with
cluster–cluster modeling results obtained by Meakin
[26]. This similarity may be worth further investigation
in future studies.
When the alum dose is low, as in experiments 3 and 4,
the initial (10min) D2 values are close to the Euclidean
dimension (D2 ¼ 2), since there was apparently insuffi-
cient coagulant to cause significant growth of aggregates
at that time. A sharp decrease in the measured D2 values
is observed for experiments 5 and 6, particularly for
experiment 5. Although the early fractal dimensions (at
10min) for these two experiments were almost identical,
the different D2 values at later stages apparently resulted
from the lower shear rate applied for experiment 5.
When comparing measurements at any corresponding
time, experiment 5 produced the lowest D2 and largest l
values. This effect is also seen in comparing experiments
3 and 4, where the lower mixing speed of experiment 3
resulted in smaller values for D2; although the modes
were identical at the three different times. Apparently,
the slower mixing speed allows less dense aggregates to
form.
From Table 1, it can be seen that experiments 1 and 5
produced the lowest D2 values. These experiments both
had higher alum dose and lower shear rate. In addition,
upon comparing paired experiments (1 and 2, 3 and 4, 5
and 6, or 7 and 8), the lower shear rate (the first
experiment in each of these pairs) always produced a
lower value of D2; particularly for the 30-min measure-
ments. It appears then, that the lower mixing intensity
was more favorable for producing less compact aggre-
gate structures (lower D2), though alum dose and
particle concentration also affected D2: Similarly, anexamination of data in Table 1 shows that higher alumT
able2
Calculationof
D2basedonsub-ranges
foraggregatesize,forexperiments1and2;
n=number
ofparticlesin
each
sub-range
10min
20min
30min
Majoraxisrange(l)(mm)
Sample
Majoraxisrange(l)(mm)
Sample
Majoraxisrange(l)(mm)
Sample
10–18
18–25
10–18
18–25
25–45
10–18
18–25
25–45
>45
Ex
per
imen
t1
D2
1.8270.12
1.8370.15
1.8370.10
1.6870.30
1.6770.20
1.6870.30
1.6870.10
1.5570.40
1.5270.30
1.5170.30
1.5270.30
1.5170.10
R2
0.80
0.60
0.89
0.60
0.70
0.52
0.91
0.80
0.42
0.80
0.63
0.89
N50
36
86
27
32
18
77
630
10
19
65
10min
20min
30min
Majoraxisrange(l)(mm)
Sample
Majoraxisrange(l)(mm)
Sample
Majoraxisrange(l)(mm)
Sample
10–18
18–25
10–18
18–25
11–25
25–50
>50
Ex
per
imen
t2
D2
1.8270.12
1.8170.20
1.8270.10
1.7070.09
1.6970.30
1.7070.10
1.6670.15
1.6370.26
1.6570.50
1.6470.10
R2
0.64
0.72
0.87
0.81
0.53
0.91
0.71
0.58
0.54
0.90
n73
24
97
72
23
95
48
27
984
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883 879
Page 8
dose generally produced lower D2 values (and larger
aggregates). This is true for both the higher (experiments
2 and 8 or 4 and 6) and lower (experiments 1 and 7
or 3 and 5) shear rates. There is no significant difference
in results for the different particle concentrations
used, but that is likely because only a relatively
small variation in particle concentration was tested
here.
Fig. 5. Images from the mixing jar for multiple aggregates after (a) 10 min, (b) 20 min, and (c) 30 min.
0
5
10
15
20
25
30
35
0.50 1.00 1.50 2.00
Log l
Rel
ativ
e F
requ
ency
(%)
10 min
20 min
30 min
0
5
10
15
20
25
30
35
0.5 1 1.5 2 2.5
Log l
Rel
ativ
e F
requ
ency
(%
)
10 min
20 min30 min
(a)
(b)
Fig. 6. Particle size distributions at three sampling times for experiments (a) 1, and (b) 2.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883880
Page 9
In order to develop relationships between fractal
dimensions and aggregate growth, it should first be
noted that aggregates in natural systems do not
generally follow the theoretical scaling laws developed
for ideal fractals, as discussed previously. In theory,
fractal dimensions are scale-invariant; that is, they do
not change as a function of aggregate size. However, this
concept is expected to hold only when aggregates
become sufficiently large relative to the primary particle
[19]. This idea is illustrated by considering the basic
approach of the diffusion-limited aggregation (DLA)
model [27] or one of its several variations (reaction-
limited, cluster-particle, cluster–cluster, etc.), which has
been used to develop theoretical values of fractal
dimensions for aggregates. In these models, aggregation
occurs as primary particles enter a sample space and
follow a random walk until they join a collector, which
is another particle or cluster of particles. The primary
particles are assumed to be solid, so their fractal
dimensions correspond to Euclidean values.
Now consider the changes in size and fractal dimen-
sion that may be expected in an aggregation experiment
starting with a monodisperse suspension of spherical
primary particles, as in the present experiments. In this
case the aggregates grow over time, at least initially, and
increasing time corresponds with increasing size. Chemi-
cal conditions and mixing speed are assumed to be held
constant. Initially, there are primary particles only and
D2 ¼ 2: As particles and clusters start to join, average
aggregate size grows and fractal dimensions decrease, as
illustrated in Fig. 7. Also shown in Fig. 7 are average
values for D2 and mode of particle size distribution
obtained at three different times in the experiments.
Eventually, D2 values are expected to level off (point A in
Fig. 7), or may start increasing slightly with longer times,
as restructuring and rearrangement result in more
compact and stable floc structures.
In the present analysis, values for D2 are seen to vary
with different experimental conditions and with time
(Table 1 and Fig. 7). In particular, D2 decreases over
time for each of the eight experiments, in some cases
more dramatically than in others. A statistical analysis of
the combined results from all experiments indicated that
D2 values were statistically different (Po0:05) for the
three measurement times. Following the above argu-
ments, a faster rate of decrease in D2 should imply that
the process is farther away from the semi-equilibrium
point (A) identified in Fig. 7. It is not clear at this time
why D2 should decrease with time, but not with size, at
least for aggregates that are not much larger than the
primary particles (Table 2). It is possible that aggregates
are simply joined differently (an aggregate growth
resulting from collision and rearrangement of particles)
after different mixing times due to restructuring effects.
Furthermore, the radius of the whole aggregate also
changes with restructuring and growth and thereby, the
mass contained in the aggregate also changes according
to the aggregate structure (see Fig. 1b). As more and
more particles are collected, the aggregate reaches a stage
where further addition of particles or clusters does not
result in further changes in D2:
5.5. Effect of shear rate on coagulation
Upon evaluating the effects of different parameters
used in this study, it was found that shear rate had the
most significant effect on aggregate growth and asso-
ciated fractal dimension. Fig. 8 illustrates this effect by
comparing the results for experiments 1 and 2, in which
a constant and relatively higher alum dose and particle
concentration were used (Table 1). It can be seen that
the correlation curves for these data points were
significantly separated (slope7standard error in the
regression line for cumulative distributions in experi-
ments 1 and 2 are 1.6270.02 and 1.7570.03, respec-
tively). This separation was the most pronounced,
relative to other comparisons (e.g., high or low alum
or particle concentration) used in this study. Fig. 8 also
shows that a lower shear rate produced a smaller slope
(i.e., a smaller D2 value), whereas the higher shear rate
produced relatively dense flocs with a correspondingly
higher D2 value, closer to D2 ¼ 2:As previously reviewed, existing studies of fractal
aggregation relate various properties of aggregates to
fixed values of fractal dimensions. In other words, given
particular values of D2 and corresponding one- and
three- dimensional fractal dimensions, it is possible to
calculate features such as porosity, settling velocity and
collision frequency. As shown here, however, it is clear
that changes in fractal dimensions, along with changes
in aggregate size, need to be taken into consideration
when describing a dynamic aggregation process. This
will be especially important for aggregation modeling,
where temporal changes in particle sizes and distribution
are simulated. Li and Logan [4] and Chakraborti et al.
[28], for example, showed that collision frequencies
Fig. 7. Temporal changes in aggregate size and fractal dimen-
sion; data points represent ranges of values and averages
obtained from the present experiments.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883 881
Page 10
calculated according to fractal theory could be orders of
magnitude greater than standard values based on a
spherical particle assumption. This suggests that sig-
nificant differences may be obtained using a model that
assumes fractal aggregates, compared with one based on
impermeable spherical particles. In addition, the present
results show that changes in fractal dimensions should
be incorporated into aggregation modeling in order to
simulate the aggregate growth process as realistically as
possible. Models based on the Smoluchowski equation
would require significant modification in order to
incorporate fractal dimension and alternate modeling
approaches may be worth pursuing.
6. Conclusions
In this study, the effects of different process condi-
tions for aggregation have been examined. Results were
reported in terms of particle size distributions and floc
morphology, as represented by the two-dimensional
fractal dimension. An in situ, non-invasive optical
technique was used to obtain the required measure-
ments. The optical sampling and digital image analysis
technique was found to be useful for obtaining and
evaluating time-varying in situ aggregate data. In
combination with the NIH-Image software, the proce-
dure provides a convenient means of obtaining data to
calculate fractal dimensions and size distributions. An
important aspect of the image processing technique is
resolution, and an analysis was conducted to evaluate
the smallest particle size that could be accurately
resolved, based on relative pixel size. A conceptual
framework was also developed to describe changes in
fractal dimension with aggregate growth over time. The
measurements and discussion in this study apply to
relatively early stages of coagulation, at least prior to the
semi-equilibrium state indicated by point A in Fig. 7.
According to the pixel resolution analysis, the area of
a single primary particle in the present experiments may
be overestimated by about 30%. However, the expected
error decreases rapidly for increasing size and most
aggregates analyzed were larger than the primary
particles. In addition, analysis of the fractal dimensions
of the sample divided into several different size ranges
indicated that a possible error in area measurement did
not have an impact on the calculated D2 values.
As may be expected, higher alum dose generally
resulted in larger flocs, while lower shear rate produced
flocs with lower values of D2: There also was
a statistically significant change in D2 values over
time (i.e., as larger flocs were produced). Particle
concentration had only a minor effect, but further
studies should be conducted over a larger range
of concentrations before firmer conclusions can be
made.
The observations reveal a trend in decreasing D2
and increasing l values over time, for coagulation of an
initially monodisperse suspension of homogeneous
spherical particles. Furthermore, the decrease in D2 is
associated with an increase in particle size. Additional
experiments should be conducted to monitor changes
over longer periods of time, which are more closely
representative of natural conditions, to evaluate the
possible effect of restructuring of primary particles.
As a final note, aggregation modeling can likely be
improved by incorporating fractal dimensions directly in
the modeling framework. This is particularly evident in
the calculation of collision frequencies, which are very
different when calculated using fractal theory as
opposed to standard techniques based on the assump-
tion that interacting particles are solid spheres. Fractal
dimension is difficult to incorporate directly in the
traditional Smoluchowski equation, so alternative ap-
proaches may be needed.
Acknowledgements
Mr. Chakraborti was supported as a Sea Grant
Scholar by the New York Sea Grant Institute during
this study. Comments and suggestions from anonymous
reviewers are greatly acknowledged.
References
[1] Gregory J. Fundamentals of flocculation. CRC Crit Rev
Environ Control 1989;19(3):185–230.
[2] Amirtharajah A, O’Melia CR. Coagulation processes:
destabilization, mixing, and flocculation. In: Water quality
and treatment. New York: McGraw-Hill, 1990.
[3] O’Melia CR. Coagulation and sedimentation in lakes,
reservoirs and water treatment plants. Water Sci Technol
1998;37(2):129–35.
[4] Li X, Logan BE. Collision frequencies between fractal
aggregates and small particles in a turbulently sheared
fluid. Environ Sci Technol 1997;31(4):1237–42.
Fig. 8. Change in fractal dimension as a response to two levels
of shear rate.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883882
Page 11
[5] Gregory J. The role of floc density in solid-liquid
separation. Filtr Sep 1998;35(4):367–71.
[6] Oles V. Shear-induced aggregation and breakup of
polystyrene latex particles. Colloid Interface Sci 1992;
154:51–358.
[7] Spicer PT, Pratsinis SE. Shear-induced flocculation: the
evolution of floc structure and the shape of the size
distribution at steady state. Water Res 1996;30(5):1049–56.
[8] Wiesner MR. Kinetics of aggregate formation in rapid
mix. Water Res 1992;26(3):379–87.
[9] O’Melia CR, Tiller CL. Physicochemical aggregation and
deposition in aquatic environments. In: Buffle J, Leewen
van HP, editors. Environmental particles, vol. 2. Boca
Raton, FL: Lewis Publishers, 1993. p. 353–86 (Chapter 8).
[10] Elimelech M, Gregory J, Jia X, Williams RA. Particle
deposition and aggregation—measurement, modeling and
simulation. Butterworth-Heinemann Ltd., London, UK,
1995.
[11] Logan BE. Environmental transport processes. New York:
Wiley Inc., 1999.
[12] Chakraborti RK, Atkinson JF, Van Benschoten JE.
Characterization of alum floc by image processing.
Environ Sci Technol 2000;34(18):3969–76.
[13] Jiang Q, Logan BE. Fractal dimensions of aggregates
determined from steady state size distributions. Environ
Sci Technol 1991;25(12):2031–8.
[14] Jiang Q, Logan BE. Fractal dimensions of aggregates
from shear devices. J Am Water Works Assoc
1996;88(2):100–13.
[15] Logan BE, Kilps JR. Fractal dimensions of aggregates
formed in different fluid mechanical environments. Water
Res 1995;29(2):443–53.
[16] Gardner KH. Kinetic models of colloid aggregation. In:
Hsu JP, editor. Interfacial forces and fields: theory and
applications. New York: Marcel Dekker, Inc., 1999.
p. 509–50.
[17] Clark MM, Flora JRV. Floc restructuring in varied
turbulent mixing. J Colloid Interface Sci 1991;
147(2):407–21.
[18] Kaye BH, Leblanc JE, Abbot P. Fractal description of ht
structures of fresh and eroded aluminum shot fine
particles. Part Charact 1985;2:56–61.
[19] Meakin P. Fractal aggregates. Adv Colloid Interface Sci
1988;28:249–331.
[20] Gorczyca B, Ganczarczyk JJ. Image analysis of alum
coagulated mineral suspensions. Environ Technol
1996;17:1361–9.
[21] Kramer TA, Clark MM. The measurement of particles
suspended in a stirred vessel using microphotography
and digital image analysis. Part System Charact 1996;13:
3–9.
[22] Miller BV, Lines RW. Recent advances in particle size
measurements: a critical review. CRC Crit Rev Anal Chem
1988;20(2):75–116.
[23] Cheng CY. Application and development of particle image
technology to study turbulence and particle aggregation.
Ph.D. Dissertation, SUNY, Buffalo, New York, 1997.
[24] Li D, Ganczarczyk JJ. Fractal geometry of particle
aggregates generated in water and wastewater treatment
processes. Environ Sci Technol 1989;23(11):1385–9.
[25] Tambo N, Watanabe Y. Physical characteristic of flocs: I.
The floc density function and aluminum floc. Water Res
1979;13:409–19.
[26] Meakin P. A historical introduction to computer models
for fractal aggregates. J Sol–Gel Sci Technol 1999;15:
97–117.
[27] Witten TA, Sander LM. Diffusion-limited aggregation.
Phys Rev B 1983;27(9):5686–97.
[28] Chakraborti RK, Atkinson JF, Van Benschoten JE. Non-
intrusive characterization of fractal aggregates formed by
alum. AWWA Annual Meeting, Universities’ Forum,
Denver, June, 2000.
R.K. Chakraborti et al. / Water Research 37 (2003) 873–883 883