Documente Academic
Documente Profesional
Documente Cultură
Eric A. Grulke, Stephen B. Rice, Jincheng Xiong, Kazuhiro Yamamoto, Tae Hyun
Yoon, Kevin Thomson, Meghdad Saffaripour, Greg Smallwood, Joshua W. Lambert,
Arnold J. Stromberg, Ryan Macy, Nico Briot, Dali Qian
PII: S0008-6223(18)30030-7
DOI: 10.1016/j.carbon.2018.01.030
Reference: CARBON 12771
Please cite this article as: E.A. Grulke, S.B. Rice, J. Xiong, K. Yamamoto, T.H. Yoon, K. Thomson, M.
Saffaripour, G. Smallwood, J.W. Lambert, A.J. Stromberg, R. Macy, N. Briot, D. Qian, Size and shape
distributions of carbon black aggregates by transmission electron microscopy, Carbon (2018), doi:
10.1016/j.carbon.2018.01.030.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
EP
C
AC
ACCEPTED MANUSCRIPT
Size and shape distributions of carbon black aggregates by transmission electron microscopy
Eric A. Grulkea*, Stephen B. Riceb, Jincheng Xiongb, Kazuhiro Yamamotoc*, Tae Hyun Yoond,
PT
Strombergf, Ryan Macya, Nico Briotg, and Dali Qiang
RI
a
Chemical & Materials Engineering, College of Engineering, University of Kentucky, Lexington,
SC
b
Cabot Corporation, 157 Concord Rd, Billerica, MA, 01821, USA
U
c
National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 5,
AN
Higashi 1-1-1, Japan
d
Department of Chemistry, College of Natural Sciences, Hanyang University, Seoul, 133
M
K1A 0R6
TE
f
Department of Statistics, College of Arts and Sciences, 313 Multidisciplinary Science Building,
g
Electron Microscope Center, College of Engineering, 004 ASTeCC Building, University of
C
Yamamoto)
1
ACCEPTED MANUSCRIPT
Abstract
Carbon black aggregate size and shape affects its performance in many applications. In this
interlaboratory comparison, an industry reference carbon black, SRB8, was analysed with a
PT
black aggregates using electron microscopy. Multiple descriptor types (size, elongation,
ruggedness, plus those of ASTM D3849-14a) were assessed for repeatability, reproducibility,
RI
and measurement uncertainties. Carbon black aggregates have been characterized using
SC
descriptor correlations: two important such correlations are affinity coefficients and fractal
exponents. SRB8 aggregates appear to be self-affine, i.e., their width and length descriptors scale
U
anisotropically. ASTM D3849-14a derived descriptors have low interlaboratory reproducibilities
AN
and high measurement uncertainties. When these descriptors are used for projected area-based
fractal analysis, the estimated fractal exponents do not have realistic values. However, the use of
M
measurement uncertainties of about 9%. Carbon black aggregates can be categorized using shape
D
descriptors into the categories: spheroidal, ellipsoidal, branched, and linear. These shape
TE
categories contribute non-uniformly to descriptor values across their data ranges and lead to
EP
multimodal distributions. These findings illustrate the importance of assessing data quality and
1. Introduction
Carbon black (CAS Registry number 1333-86-4) and fumed silica are examples of industrial
aciniform aggregates (grape-like clusters of aggregated particles). World-wide carbon black sales
2
ACCEPTED MANUSCRIPT
were expected to reach 15 million metric tons in 20151, growing by a factor of two over the
previous decade[1]. Commercial products are available in a wide variety of morphologies, each
of which contributes to the economic value and applications of the carbon black grade. The
particle size distribution of a carbon black has been considered its single most important property
PT
[2], affecting its degree of blackness, rubber reinforcement, and UV protection. For example, the
RI
aggregate size and shape affect the viscosity and degree of dispersion of carbon black in
SC
There are alternative methods for characterizing carbon blacks for composites, such as ASTM
U
D6556 – 16, which is based on multipoint nitrogen adsorption inclusive of micropores less than
AN
2 nm and the external surface area (based on the statistical thickness method)[3] . Other test
methods include sieve residue, surface area, structure (dibutyl phthalate adsorption number),
M
aggregate shape (spherical, ellipsoidal, linear, and branched), ash content, toluene extraction,
D
Carbon black aggregate characteristics have been measured using electron microscopy [5, 6],
EP
including area-equivalent diameter of the aggregates, perimeter, area, fractal dimensions, and
branching coefficients [7-13]. Over the past decade, a number of researchers have measured
C
TEM-based aggregate descriptors using automated image analysis methods and linked these to
AC
carbon black performance in a variety of applications, particularly for tire compounds [1] [14-
19]. For example, Lu and Chung [20] related aggregate size and shape, as expressed by its
1
CBS, world wide carbon black sales, 2015. http://carbonblacksales.com/worldwide-carbon-
black-market-12-million-metric-tons-2015/
3
ACCEPTED MANUSCRIPT
descriptors measured by electron microscopy are in common industrial use. Several derived
descriptors, such as equivalent circular aggregate diameter, area to perimeter ratio and
roundness, are popular choices for correlating carbon black morphology with its performance in
elastomer compounds. However, the measurement uncertainties of these descriptors have not yet
PT
been reported in the literature.
RI
ASTM D3849, initially published in 2007 and updated in 2014[2], characterizes carbon black
SC
morphology y using descriptors from TEM images. Aggregate area (nm2) and aggregate
perimeter (nm) are used to derive the area-equivalent aggregate diameter, an aggregation factor,
U
an average nodule size for a single aggregate, an aggregate volume, a nodule volume (often
AN
referred to as a primary particle), and the number of particles in the aggregate. Average values
and sample standard deviations are computed for each discrete distribution. Several round-robin
M
comparisons have been held for ASTM D3829-14a, apparently with little concurrence on
D
results.2. A focus of this work was to identify potential causes of the apparent low reproducibility
TE
of this protocol.
This project is part a series of interlaboratory comparisons (ILCs) to develop a generic protocol
for measuring size and shape distributions by TEM, coordinated through ISO/TC229
C
Nanotechnologies. The test samples include: unimodal, discrete spheroidal nanoparticles (gold)
AC
nanoparticles with different shapes (gold nanorods), amorphous aciniform aggregates (carbon
black), and aggregates of primary crystallites (titania) [22]. These studies conform to guidelines
established by the Versailles Project on Advanced Materials and Standards (VAMAS) [23] and
2
Stephen B. Rice, private communication.
4
ACCEPTED MANUSCRIPT
ISO 5725 [24]. International Standards stakeholders and users expect: 1) measurement of ‘real
life’ materials, 2) highly automated protocol steps, including image acquisition, particle capture,
uncertainty assessments for evaluations by different laboratories, and 5) data visualization tools
PT
to compare methods, procedures, and descriptors. Many co-authors are members of Joint
RI
Working Group 2, Measurement and Characterization, of ISO/TC229 and IEC/TC113.
U SC
AN
M
D
TE
C EP
AC
5
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
C EP
Figure 1 Image of carbon black aggregates, SRB 8, with good contrast between particles
and background. Courtesy of Cabot Corporation.
AC
complexities, and extents of aggregation. Figure 1 shows typical carbon black aggregates from
the sample used in this study. Aggregate size and shape distributions affect the performance of
carbon blacks in their applications. The typical aggregate is composed of nodules that are fused
6
ACCEPTED MANUSCRIPT
together in the gas phase synthesis process[1]. In this study, one laboratory prepared and
distributed mounted test samples to the collaborating laboratories. Typical properties of SRB8
carbon black are shown in Table 1. Iodine numbers relate to the carbon black surface area.
Nitrogen adsorption methods (ASTM D6556) can be used to measure total Nitrogen Surface
PT
Area (NSA) and Statistical Thickness Surface Area (STSA), which is the external surface area
RI
accessible to rubber. Alternative methods include Compressed sample Oil Adsorption Number
(COAN) and Oil Adsorption Number (OAN). Tinting strength relates to the ability of the carbon
SC
black to reduce the reflectance of white pigments.
Table 1 Typical values of iodine number, surface area, and oil adsorption measurements
U
for SRB8. Courtesy of Cabot Corporation3
AN
Properties Iodine Number NSA STSA COAN OAN Tint Strength
Note 3: values in the 3rd row reflect the specific method for making the measurement, e.g., the specific surface area
EP
There are a number of methods for characterizing the morphologies of carbon blacks, including:
AC
affinity exponents that scale aggregates in the longitudinal and transverse dimensions and fractal
analysis, and 3) analysis of carbon black aggregate descriptors after they have been separated by
3
Jincheng Xiong, private communication
7
ACCEPTED MANUSCRIPT
shape (spheroidal, ellipsoidal, branched, and linear). All of these methods can lead to improved
PT
Researchers have evaluated size and shape descriptors for carbon black aggregates for some time
[11-13, 25]. Size descriptors for two-dimensional images can be based on length (Category 1) or
RI
area (Category 2). Shape descriptors are often derived from size descriptors and can describe
SC
image elongation or image boundary irregularity (ruggedness). A minimal set of descriptors for
carbon black aggregates might include size, elongation, and ruggedness. Using powder data,
U
Hentschel and Paige [26] proposed a specific minimal set of shape descriptors: aspect ratio to
AN
represent elongation and form factor to represent ruggedness (boundary irregularities). Table 2
shows size, elongation, and ruggedness descriptors are based on ISO definitions [27] [21, 22].
M
Table 2 Size and shape descriptor table based on ISO 9276-6 definitions [27]
TE
8
ACCEPTED MANUSCRIPT
∙
of the particle is similar to a circle,
considering the overall form of the
particle with the maximum Feret
4∙
diameter
PT
Roundness 2 Square of the compactness / [28]
∙
Extent 2 Ratio of the particle area to the product
∙
RI
of the Feret and minimum Feret
diameters
Shape/ruggedness
4∙ ∙
SC
Circularity 1 Degree to which the projected area of
the particle is similar to a circle, based
4∙ ∙
on its perimeter
Form factor 2 Square of the circularity
Solidity 2
U
Ratio of the projected area, A, to the
AN
area of the particle’s convex hull
a
– category 1 = length or length ratios; category 2 = area or area ratios
M
A key question for the community is: how repeatable and reproducible are the descriptors that
we have been using to report aggregate size and shape? Measurement uncertainties of carbon
D
black aggregate descriptors do not seem to be reported. Since the quality of the size and shape
TE
descriptors were not established, an expanded set of size, elongation, and ruggedness descriptors
were evaluated to determine which of these might have lower uncertainties in intra- and
EP
interlaboratory tests. Descriptors with low uncertainties might provide a good basis for preferred
size and shape distribution data. Carbon black manufacturers, customers, researchers, and
C
relevant regulatory agencies all have an interest in the repeatability, reproducibility, and
AC
measurement uncertainties of descriptor data. This interlaboratory comparison addresses not only
how much variation there might be in descriptor means as they are evaluated by a number of
9
ACCEPTED MANUSCRIPT
The area-equivalent diameter, or equivalent circular diameter (ECD), is often used to report
carbon black aggregate size. For cases in which the aggregate is treated as an ellipse, the major
and minor ellipse axes can be used for size [25]. Elongation parameters include the aspect ratio,
compactness, roundness, extent (bulkiness), and the affinity exponent [29]. Roundness is the
PT
square of compactness and is considered not as robust as compactness [27]. Ruggedness
RI
parameters include circularity, form factor, and solidity. Prior researchers have used the ellipse
axes for size, the ellipse ratio for elongation, and bulkiness for ruggedness [25], or ECD for size,
SC
aspect ratio and roundness for elongation, and circularity, form factor, and solidity for
ruggedness [30]. All of these descriptors are evaluated for measurement uncertainties.
U
AN
1.3.2 Descriptor-descriptor correlations
Carbon black aggregates are known to grow anisotropically. Affinity exponents and fractal
M
analyses provide information on particle anisotropy and both have previously been applied to
carbon black aggregate data. Both methods are, at their basic level, descriptor-descriptor
D
correlations. As such, it is important to report the values of fitting parameters as well as their
TE
standard errors. The affinity exponent, H, describes the ratio of scaling coefficients for length
and width of aggregates, where nW and nL are the scaling factors in the transverse and
EP
!" #$%$&
= ≈
#$%$&
Equation 1
C
ln = ∙ ln
AC
Equation 2
Equation 2 suggests that the natural logarithm of the minimum Feret diameter is linearly
correlated with the natural logarithm of the Feret diameter. The affinity exponent is not a
descriptor reported by ISO[27]but has been applied to carbon blacks [29, 31]. If H ~ 1, the
aggregate is considered to be self-similar, i.s., the scaling factors for length and width are the
10
ACCEPTED MANUSCRIPT
same. If H < 1, the aggregate is considered to be self-affine. The various shapes of carbon black
aggregates may have different values of H, depending on how their width and length dimensions
scale. Neimark, Koylu, and Rosner [29] estimated the values of H using the approximation
shown in Eq. 2, finding the slope, H, of a plot of the minimum Feret diameter logarithm (as the
PT
y-axis) against the Feret logarithm (as the x-axis). In this case, the reference lengths for width
RI
and length were both set to 1 (W0 = L0 = 1 nm), forcing the correlation through the origin in
logarithmic space. The size units need to be set such that width and length logarithms are
SC
positive. In addition to providing elongational information, the affinity exponent provides
U
AN
Analyses of fractal dimensions have been applied to carbon black aggregates and soots for some
time [32-40]. A number of methods have been proposed to evaluate the fractal dimensions of
M
carbon blacks and soot particles[30, 40-43]. Methods include: 1) projected area-based scaling,
radius of gyration scaling, and mobility diameter scaling[43] and 2) minimum bounding
D
rectangle [40] and 3) box counting techniques[41]. The projected area-based scaling method is
TE
convenient to use with TEM size distribution data [31, 44, 45]. For projected area-based scaling,
EP
the number of nodules (some references use the term, primary particle) in the aggregate, N, is
related to the ratio of the projected areas of the aggregate divided by that of a nodule by a power
C
-
) = *+ ∙ , /
+.
Equation 3
where ka is the projected area prefactor, A is the projected area of the aggregate, ap is the average
area of a nodule, and Df is the fractal exponent. While ASTM D3849-14a does not compute a
fractal exponent, its derived parameters can be used to do so. In this study, nonlinear regression
11
ACCEPTED MANUSCRIPT
methods of SYSTAT© were used to determine values of the prefactor, ka, and fractal exponent,
Df, of Eq. 3. The aggregate area in Eq. 3, A, is measured directly with TEM images, but the
values of N and dp can be estimated in several ways. ASTM D3849-14a estimates N and dp via
derived parameters based on the measured aggregate areas and perimeters (see Table 3). Its
PT
workflow uses a correlation for the average nodule size, estimates the aggregate volume and
RI
nodule volumes, and computes a value for N. In this method, N is the ratio of the estimated
aggregate volume divided by the estimated nodule volume. When the scheme of ASTM D3849-
SC
14a is followed, the average nodule size in the aggregate is computed for each measured particle.
An alternative approach is to determine an average value for dp for the entire aggregate sample,
U
and use this value to estimate the volume of a nodule and the cross-sectional area of a nodule.
AN
This second method was previously reported by Tian et al. [45], justified by the modest size
distribution range for nodules compare to the large size range for aggregates. Nodule sizes for
M
the SRB8 sample varied from 7 to 52 nm, with an average of 21.6 ± 11 nm, while the Feret
D
4∙
D 1 Area-equivalent aggregate diameter, nm
C
;<.=
= 13.092 ∙ 9 :
2∙ ∙
>? =
dp 1 Average particle (nodule) size for a single
aggregate, nm
12
ACCEPTED MANUSCRIPT
8
@- = , / ∙
VA 3 Aggregate volume, nm3
3
∙ >? C
@B =
VP 3 Particle (nodule) volume, nm3
6
@-
) =
PT
3 Number of particles (nodules) in the aggregate
@B
N
a
– category 1 = length or length ratios; category 2 = area or area ratios; category 3 = volume or
RI
volume ratios
SC
1.3.3. Separating aggregates by shape
Carbon black aggregates are known to have several characteristics shapes. Therefore, it seems
U
appropriate to separate aggregate datasets by shape as a way to interpret the size and shape
AN
distributions of the various descriptors. Hess and coworkers proposed classification systems for
carbon black morphologies [7, 8]. Shishido and coworkers [46] developed ranges of descriptors
M
that were used to categorize carbon black aggregates into spheroidal, ellipsoidal, branched and
linear configurations. Ono et al [47] used this classification system to evaluate aggregates shapes
D
as affected by temperature and residence time during vapour phase synthesis of carbon black.
TE
Figure 2 shows the aggregate classification scheme is based on an elongational shape (aspect
ratio) and two ruggedness descriptors, form factor and solidity. This set of descriptors are
EP
independent of each other: aspect ratio is based on Feret and minFeret, the form factor is the
C
projected area divided by the area of a circle with the same perimeter, and the solidity is the ratio
AC
13
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
Figure 2 Each aggregate was sorted into a linear, branched, ellipsoidal, or spheroidal
TE
morphology category using this workflow[46], which is based on ISO descriptor definitions
[27].
Data analysis and morphology evaluations were focused on three core areas: data repeatability
EP
fitted parameters (relevant for industrial quality control and regulatory use), and additional
C
black aggregates).
reproducibility would be preferred for characterizing size and shape distributions. Such
descriptors can be identified by comparing sample means for intralaboratory data frames
14
ACCEPTED MANUSCRIPT
similarity). Common statistical protocols [21, 22] were used to reduce variability in analysing
and reporting data. Such tools can be critical to stakeholders wishing to optimize protocols for
specific materials and applications. P-values reported from ANOVA or bivariate analysis tests
PT
provide qualitative comparisons of datasets. Energy values (E-values) of bivariate analyses can
RI
be used to rank dataset pairs with respect to their similarity. However, neither the p-value or the
E-value provide information on the scale or the width of a descriptor distribution. Arithmetic
SC
means are often used to compute an average descriptor value.
U
Measurement uncertainty estimates help reveal how well models and correlations describe size
AN
and shape distribution data. For example, descriptor distributions are fitted to reference models,
requiring a minimum of two parameters, scale and width, to define them. When descriptor
M
distributions are unimodal, it is straightforward to estimate the scale and width parameters of
reference models fitted to the data and calculate their measurement uncertainties. Normal,
D
lognormal, and Weibull models are considered empirical distributions and are used to find values
TE
of scale and width parameters. Direct comparisons of the measurement uncertainties of these
EP
scale and width parameters can help identify which ones have the highest repeatability and
reproducibility for the sample. Such descriptors might be preferred for developing additional
C
analysis, and determining the contribution of different shape configurations to the observed
descriptor distributions. In addition, measurement uncertainty values can provide a basis for
improving size and shape distribution measurement protocols, as has been done with raw data
15
ACCEPTED MANUSCRIPT
2. Experimental
2.1 Sample preparation
A typical carbon black sample, ASTM reference material SRB8, was supplied by and mounted
on TEM grids by Cabot Corporation. The protocol used for this study is based on that of ASTM
PT
Table 4 Sample preparation for ASTM Reference Series 8 Sample B (SRB8)
RI
Step Procedure
1 Acquire approximately one quarter spatula (approximately 10 mg) of carbon black, and
SC
place into a vial (3 cm X 5.5 cm), add 15cc chloroform. This quantity of carbon black
gives a good representation of the entire black.
2 Immerse the vial into an ice-water bath set in a medium size plastic container. The ice
bath minimizes the heat introduced by ultra-sonic energy.
U
3 Insert the microtip probe to a depth of approximately 2.5cm into the mixture.
AN
4 Set timer on control box of W-385 Ultrasonic Processor for 10 minutes @ level 5 for and
output value of approximately 45 watts (see Notes).
5 Withdraw suspension with disposable pipette, and release (3) or more drops of the heavy
concentrated suspension into a clean vial, then add 15ml of chloroform (wipe probe tip)
M
suspension against a white background, (only a few drops on the filter paper). The
dispersion should be relatively transparent.
TE
7 Repeat the ultrasonic sonication for about 3 minutes at same power level. The volume of
carbon black and chloroform mixture should always be maintained during the dispersion
process. If considerable further dilution is required, the excess volume is disposed of into
EP
the grid as close to the center as possible, from a height of about 12mm.
10 Allow specimen to dry for about 1 minute.
AC
11 Check the concentration of the diluted dispersion for agglomeration. If necessary, adjust
the concentration by adding more solvent as required and sonicate for three minutes.
Notes: Specific to the probe sonicator unit used at Cabot, Misonix (Qsonix) Sonicator
Ultrasonic Processor XL. The ultrasonic horn should be housed in a soundproof box set
inside the hood. Utilizing the ½” tip, output being drawn at level 5 is approximately 45
watts. Clean the ultrasonic probe by immersing the tip into a vial (3cm x 9cm) containing
30 ml of ethyl alcohol with output control at level (5) on ultrasonic control box. Subject
to sonic energy for 1 minute. Wipe tip with Kimwipes®.
16
ACCEPTED MANUSCRIPT
All labs received mounted samples for their analysis. One experienced laboratory prepared the
samples, leading to high quality grids and low fractions of touching particles (generally to be
avoided). It is assumed that all particles in an image (Fig. 1, for example) are aggregates rather
PT
than agglomerates. As the protocol is one used by a manufacturer using TEM to size carbon
RI
black aggregates, it is thought to be the best available. The assessments of repeatability,
reproducibility, and measurement uncertainty help define the data quality available with this
SC
protocol.
set-ups. Rather than specify magnification, which can change between instruments, the protocol
M
suggested a magnification and labs reported the image resolution, pixels/nm. Laboratories
reported their calibration methods, which were shown by ANOVA analysis not to relate to
D
ASTM D3849-14a provided guidelines on TEM image digital resolution setting, which depends
EP
on the nodule size range of the carbon black. The recommended image resolution setting is from
1.5 to 2.0 nm/pixel for SRB8B with its average nodule size of 18 nm. On the one hand, low
C
resolution setting allows the possibility of characterizing high number of aggregates on the same
AC
frame, reducing the cost of image capture and the bias of the TEM operator to pick out aggregate
fields. Conversely, high resolution allows better aggregate edge detection and improved accuracy
of the aggregate length and perimeter measurements. The ASTM recommended resolution
settings provide a balance between efficiency of image capture and precision of the image
17
ACCEPTED MANUSCRIPT
reduce the error of the area measurement to less than 5%. Aggregates that did not meet this
criterion were removed from the datasets. At the resolution setting of 1.59 pixel/nm, the image
has adequate resolution for determining the aggregate size of SRB8B with its average area
equivalent diameter of 54 nm. In addition, at this digital resolution, the TEM image with 3072 X
PT
3670 pixels covers 4.9 µm by 5.8 µm area, leading to sufficient numbers of well-dispersed CB
RI
aggregates (~ 200-300) captured in a single image when appropriate CB concentration is used for
preparing the TEM sample (see the new TEM image at the end). At this setting, 10 images are
SC
enough for getting sufficient numbers of CB aggregates (over 2000) for the image analysis.
Standard bright-field TEM images with mass-thickness contrast are used for CB aggregate
U
analysis. After achieving best image resolution with minimal contrast through focus adjustment,
AN
a minimal level of defocus was added to enhance the contrast at the edge of the particles so that
the particles can be reliably segmented from background during image analysis.
M
software is used industrially, but software choice was not an element of the protocol design. In
TE
the carbon black industry, a norm has been to report ~ 2,000 aggregates.
EP
aggregate), minimum Feret diameter (minFeret, the minimum width of the aggregate) and
C
particle area. Some labs reported more, including perimeter, circularity, form factor, and solidity.
AC
There are literature references for using ImageJ for carbon black [19].
4
https://imagej.nih.gov/ij/
18
ACCEPTED MANUSCRIPT
provide similar data analysis outputs. We have developed three statistical methods5, programmed
in R and implemented as Shiny Apps®, which were used to analyse all datasets. The approach
PT
has helped ensure that all statistical methods and reports were standardized. Distribution scale
and width parameters were determined by fitting reference models to the empirical cumulative
RI
distribution data. Two fitting techniques, maximum likelihood estimates and nonlinear regression
SC
estimates, were used for each distribution. One descriptor distribution was fitted with a bimodal
distribution using a nonlinear regression tool in Systat®. A bimodal fit requires two sets of
U
scales and widths, one set for each peak, plus a weighting factor, i.e., five parameters rather than
AN
two.
Descriptor distributions of each laboratory were fitted to distributions and the best values for
M
their scale and width parameters were reported; Measurement uncertainty values for the study
were determined using the scale and width parameters for all laboratories. For each parameter of
D
each descriptor, the average and the standard deviation of all reported parameter estimates were
TE
Equation 4
IJK = * ∙ ∙ L1 + 1/
AC
E Equation 5
where UILC is the measurement uncertainty (%), k is taken as 2, and n is the number of
observations (4 laboratories in this case). Equation 5 and its variations are often used to report
5
https://shiny.as.uky.edu/anova-app/; https://shiny.as.uky.edu/curve-fitting-app/;
https://shiny.as.uky.edu/bivariate-fitting-app/; developed by Lambert and Stromberg.
19
ACCEPTED MANUSCRIPT
the measurement uncertainty for reference materials; some examples are found in recent
Certification Reports [48, 49] [50]. When descriptor distributions are not unimodal, the
measurement uncertainty estimates may have large % differences. Eq. 5 can be modified to
include additional uncertainty factors related to the measurement protocol. Here, the UILC simply
PT
shows the parameter’s uncertainty across the ILC.
RI
3. Results
3.1 Raw data triage
SC
Raw data triage includes assessing image quality, assessing data quality, measuring
U
taken by laboratories had insufficient contrast between background and particle, leading to poor
AN
image quality. Poor thresholding can be detected by evaluating outlined images for the following
defects: poorly replicated aggregate outlines, particles that have fragmented after outlining,
M
creation of artefact particles, and gradients in background shading across the image. This was
corrected by providing examples of TEM images with good contrast to illustrate to the individual
D
operator the image quality they should be aiming to achieve. In other cases, some of the
TE
aggregates had less than 100-200 pixels, a recommended number to keep the area error less than
EP
5%[51]. Table 5 shows the image resolution and number of aggregates removed for each
reduce the error of the area measurement to less than 5%. With image resolution varying from
AC
1.0 pixel/nm to 2.3 pixel/nm, the minimum area for aggregates to be counted varied from 200 to
460 nm2.
Table 5 Criteria for removing aggregates below the minimum area, labs L1-L4
Factor L1 L2 L3 L4
20
ACCEPTED MANUSCRIPT
PT
Average particle area for revised datasets, nm2 3576 4900 3261 3549
RI
Intralaboratory repeatability was good, with many frames having the same mean as the complete
SC
dataset for most descriptors. ANOVA analysis of interlaboratory reproducibility showed that
some dataset pairs did not have similar means. In particular, ANOVA pairwise comparisons with
U
the size data of laboratory L4 had p values less than 0.05, indicating that their means were
AN
different. The aspect ratio descriptor had similar means for all four laboratories. Only one of six
M
laboratory pairs had statistically similar ruggedness descriptors. Details of these findings are
scale and width parameters with low uncertainties. However, shape distributions were
multimodal, which generally had greater measurement uncertainties and were fitted either by
C
normal or Weibull distributions. In the case that a specific descriptor was multi-modal, bivariate
AC
analysis of descriptor cumulative distributions6 was used to determine whether a pair of labs had
similar results. A similar workflow was used to analyse the descriptors specifically cited in
6
https://shiny.as.uky.edu/bivariate-fitting-app/; developed by Lambert and Stromberg
21
ACCEPTED MANUSCRIPT
Figures 3-5 show distributions of three classes of descriptors: size, elongation, and ruggedness.
The left hand side of each figure shows a smoothed density distribution of the raw data (black
curve) compared to the density distributions of the maximum likelihood (MLE: red curve) and
nonlinear regression (NLS: blue curve) fits. The right hand side of each figure shows the
PT
empirical cumulative distribution with the maximum likelihood and nonlinear regression fits.
RI
The lognormal distribution appears to be the preferred reference model for the Feret diameter
distribution (Figure 3). Bimodal fits can be done with the Feret diameter distributions, one of
SC
which is shown in Section 2 of the Supplementary Material. A different approach toward
understanding the multimodal nature of these distributions is to deconvolute them with respect to
U
aggregate shapes, which is done later in the Results Section.
AN
M
D
TE
C EP
AC
Figure 3 The lognormal distribution provides a good fit to the Feret data of Lab L1. Black
curve = smoothed data, blue curve = model parameters fitted using nonlinear regression,
red curve = model parameters fitted using maximum likelihood.
Even though aspect ratio data had similar means across all labs, these elongational shape
distributions show multimodal peaks (Figure 4). The density distributions in the left hand plot
show that the highest peak of the data coincided with peaks of the unimodal models. However,
22
ACCEPTED MANUSCRIPT
the data shows evidence of minor peaks, to the left and right of the major peak, which are not
captured by unimodal models. Differences are also apparent in the cumulative distribution plot
in the upper right hand side. The other elongational descriptor, compactness, showed similar
multimodal peaks.
PT
RI
U SC
AN
M
D
TE
Figure 4 Normal distribution fits to the aspect ratio data of Lab L1. Both fitted curves give
maximum density values near that of the data; however, the smoothed density data (black
curve) shows three possible peaks near its maximum. Blue curve = fitted parameters using
nonlinear regression, red curve = fitted parameters using maximum likelihood.
EP
Figure 5 shows the distribution of the form factor, a ruggedness descriptor. It also has
multimodal peaks, and is better described by a normal or Weibull distribution. In this case, the
C
data peak is offset from those of the unimodal models, significantly increasing the measurement
AC
uncertainty of the model scale parameters. The solidity distribution data have a lower
23
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
Figure 5 Normal distribution fits to the form factor data of Lab L1. The maxima of the
fitted models are shifted to higher values of form factor than the smoothed data (black
curve), which shows multiple possible peaks above and below its maximum value. Blue
M
curve = fitted parameters using nonlinear regression, red curve = fitted parameters using
maximum likelihood.
D
descriptors are reported: size, elongational, and ruggedness descriptors plus descriptors
specifically used for ASTM D3849-14a. The aggregate size descriptors appear to be best
EP
represented by log normal distributions rather than normal or Weibull reference models. The
C
coefficients of variation for the scale parameters of these descriptors varying from 1.5% to 2.0%,
AC
with the exception of the aggregate equivalent circular diameter, which is a factor of 2 smaller.
The coefficients of variation for the widths of the size distributions vary from 2.5% to 4.0%,
uncertainty of 9.1%, perimeter would seem to be the least accurate choice for aggregate size. The
24
ACCEPTED MANUSCRIPT
reference model with UILC = 1.6%, compares with favourably with the typical reported value of
54 nm for SRB8 (Table 1). ECD has the lowest measurement uncertainty of the size descriptors;
this value confirms that it is preferred choice for reporting aggregate size.
Table 6 Scale and width measurement uncertainties of descriptors across all data sets. Eq.
PT
4 is used to compute the coefficient of variation for each scale and width parameter, which
is used to compute the measurement uncertainty using Eq. 5. Unimodal models
RI
Grand means Scale Standar Width Standard Cv, Cv , UILC UILC
table [exp(scale), d error Scale width Scale width
nm] error
Descriptor Referenc size descriptors
SC
e
Model
Area, nm2 LRM 7.75 0.154 1.044 0.04101.99 3.92 4.46 8.77
[2310] % % % %
U
Perimeter, LRM 5.52 [249]
0.0859 0.696 0.0231 4.06 12.8 9.09 28.7
nm % % % %
AN
Feret, nm LRM 4.39 [80.9] 0.0698 0.586 0.0170 1.59 2.90 3.55 6.49
% % % %
minFeret, nm LRM 3.94 [51.5] 0.0777 0.570 0.0157 1.97 2.76 4.41 6.18
% % % %
M
ECD, nm LRM 3.97 [52.9] 0.0285 0.522 0.0205 0.72 3.93 1.60 8.79
% % % %
elongational descriptors
Aspect ratio NRM 0.642 0.0076 0.133 0.0042 1.18 3.19 2.64 7.14
D
% % % %
Compactness NRM 0.673 0.0098 0.097 0.0064 1.45 6.59 3.24 14.7
TE
7 % % % %
ruggedness descriptors
Form factor NRM 0.485 0.0496 0.212 0.0163 10.2 7.67 22.9 17.1
% % % %
EP
Solidity NRM 0.766 0.0218 0.112 0.0089 2.85 7.94 6.37 17.8
% % % %
ASTM D3849-14a descriptors
dp, nm LRM 2.89 [18.1] 0.167 0.465 0.055 5.77 12.2 12.9 27.3
C
% % % %
VP, nm3 LRM 8.03 [3077] 0.501 1.369 0.166 6.23 12.2 14.0 27.3
AC
% % % %
N LRM 2.87 [17.6] 0.391 1.695 0.070 13.6 4.15 30.5 9.30
% % % %
LRM = lognormal reference model; NRM = normal reference model; CV = coefficient of variance, % (Eq.
4); UILC = measurement uncertainty, % (Eq. 5); [xx, nm] gives the exponential of the scale estimate for
25
ACCEPTED MANUSCRIPT
The parameters for aggregate shape descriptors are reported for the normal distribution as the
reference model. However, the Weibull model could also be used as the coefficients of variation
of the distribution scales and breadths are quite similar between these two choices. A preferred
choice for elongation descriptor is aspect ratio, which has the lowest scale and width
PT
uncertainties. The elongation scales have lower coefficients of variation than the ruggedness
RI
scales, while the coefficients of variation for the distribution widths are similar for both shape
types. The solidity descriptor might be preferred for ruggedness. Aggregate form factor, which is
SC
based on area and perimeter, has a relatively high measurement uncertainty for its scale
parameter, 22%. As shown in Figure 5, the form factor distributions are multimodal and the
U
computed curves do not represent the data well, in particular, the peaks predicted by the models
AN
are displaced from the actual peak in the data. Therefore, the coefficient of variation and the
measurement uncertainty of the form factor scale parameter is high compared to other scale
M
parameters of shape descriptors, which better match the central peak of their data (see Fig. 3, for
D
example).
TE
The measurement uncertainties of the scale and width for some of the derived descriptors of
EP
ASTM D3849-14a are shown in the lowest section of Table 2, which includes the average nodule
diameter of the aggregates (dp), the nodule volume (Vp), and the number of nodules in the
C
aggregate (N). The uncertainties of scale and width parameters of these descriptors are some of
AC
the highest of this ILC. All three descriptors are derived from different combinations of area and
perimeter, so it is not surprising that their scales have measurement uncertainties higher than any
of the size descriptors. The smallest measurement uncertainty for the scale parameters of this
descriptor trio is nearly 13% and that of N is ~ 30%. Since these descriptors are needed to
26
ACCEPTED MANUSCRIPT
compute the fractal dimension using the projected area approach, it might be expected that the
accuracy of a fractal dimension value based on the ASTM derived descriptors will be low.
Nonetheless, the average nodule diameter reported in Table 5, 18.1 nm based on a lognormal
reference model with UILC = 12.9%, compares favourably with the typical reported value of 18
PT
nm for SRB8 (Table 1).
RI
Prior work by Tian et al. [44] provides a possible explanation for the poor correspondence
SC
between N data and a unimodal model for soot aerosols. They noticed that a lognormal fit to N
did not coincide well with the data, particularly at low values of N. One approach to this issue
U
was to eliminate small values of N and consider only values of N > 5. A second approach was to
AN
fit the data with a bimodal distribution, which implicitly assumes a second particle population for
low values of N. Rather than demonstrate bimodal fit to N distributions, the approach has been
M
bimodal fit clearly reduces the error between model and data.
TE
Three descriptor distributions, Feret (size), aspect ratio (elongation), form factor (ruggedness)
were reconstructed from their spheroidal, ellipsoidal, branched, and linear elements.
C
Generally, the shape morphology categories are non-uniformly distributed across the variable
AC
ranges of size, elongation, and ruggedness descriptors. Figure 6 shows the shape morphology
component contributions to the Feret diameter distribution. Spheroidal particles make modest
contributions to small sizes of the Feret diameter distribution. Ellipsoidal particles contribute to
Feret diameters starting at 25 nm and peaking near 60 nm. The long ‘tail’ is due to the broad
27
ACCEPTED MANUSCRIPT
150
125
PT
100 spheroidal
ellipsoidal
Count
75
RI
branched
linear
50
SC
all
25
U
AN
0
0 50 100 150 200 250 300
Feret Diameter, nm
M
D
Figure 6 The aggregate Feret distribution of Lab 1 assembled from its morphology
category distributions, which are not uniformly distributed across the data range. Red
TE
circles = spheroidal, blue diamonds = ellipsoidal, green triangles = branched, orange bars =
linear, black curve = sum of all morphology categories.
Contributions of shape morphology components to the aspect ratio distribution (Figure 7) are
EP
linked to the workflow for morphology separation, Fig. 2. Linear aggregates contribute to aspect
C
ratios less than 0.533 per the workflow. All nonlinear aggregates have aspect ratios larger than
AC
this value. Ellipsoidal and branched aggregates contribute similar ranges of the aspect ratio; they
are separated by solidity in the Shishido scheme. Spheroidal aggregates make modest
28
ACCEPTED MANUSCRIPT
200
180
160
140
PT
120 spheroidal
Count
100 ellipsoidal
RI
80 branched
linear
SC
60
all
40
U
20
AN
0
0 0.2 0.4 0.6 0.8 1
Aspect ratio
M
Figure 7 The aggregate aspect ratio distribution of Lab 1 assembled from its morphology
category distributions. Linear aggregates have low aspect ratios while spheroidal
D
aggregates had high aspect ratios. Red circles = spheroidal, blue diamonds = ellipsoidal,
green triangles = branched, orange bars = linear, black curve = sum of all morphology
TE
categories.
EP
Low values of the form factor are due solely to linear aggregates while branched and ellipsoidal
aggregates dominate the middle values. High values of the form factor are due mostly to
C
spheroidal particles. Figures 9-11 demonstrate that aggregates with different shapes contribute
AC
characterization cases when aggregate shapes can be linked to application performance, such as
discovered for carbon black viscoelastic properties [18]. These examples are shown for
illustration purposes; individual stakeholders need to evaluate whether the information is useful
to them.
29
ACCEPTED MANUSCRIPT
180
160
PT
140
120
RI
spheroidal
Count
100
ellipsoidal
80
SC
branched
60 linear
all
U
40 AN
20
0
0 0.2 0.4 0.6 0.8 1
M
Form Factor
D
Figure 8 Aggregate form factor assembled from its morphology category distributions. Red
circles = spheroidal, blue diamonds = ellipsoidal, green triangles = branched, orange bars =
TE
1 and the SRB8 test sample appears to be self-affine. Carbon black aggregates are known to
AC
grow anisotropically, probably due to flow patterns in the reaction systems. Table 7 shows the
average values of H for each laboratory and the measurement uncertainty of H for the ILC,
30
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
Figure 9 Affinity exponent plot (Eq. 2) for the data of Lab L1; solid line = model equation
with H = 0.895 ± 0.00098
D
It would be useful to learn whether the different aggregate shapes have different affinity
exponents, i.e., different width-length scaling. Using the categories of Shishido [46], the particle
31
ACCEPTED MANUSCRIPT
data of Lab L1 was deconvoluted into the four morphologies using the workflow of Fig. 2. The
affinity exponent for each morphology was determined using plots similar to Fig. 8. For the Lab
L1 dataset, spheroidal aggregates are less than 2% with the other three morphologies distributed
about the same (Table 10). As expected, the affinity coefficient for spheroidal aggregates is the
PT
largest (0.937) and that of linear aggregates is the lowest (0.852). By ANOVA analysis, only the
RI
H values for ellipsoidal and branched morphologies are statistically similar. For comparison
purposes, average aspect ratios, estimated in a similar manner, are shown in Table 8. The rank
SC
order of both the affinity exponent and the aspect ratio average is the same. These data
demonstrate that the self-affine properties of carbon black morphologies vary with shape
U
category. Presumably, carbon black aggregate samples with different distributions of spheroidal,
AN
ellipsoidal, branched, and linear morphologies would be distinguishable by their affinity
exponents.
M
Table 8 Comparison of affinity exponent and aspect ratios for different aggregate
morphology categories; data of Lab L1
D
exponent ratio
Spheroidal 1.72% 0.937 0.896 0.813 0.937
Ellipsoidal 36.8% 0.919 0.944 0.685 0.909
Branched 27.8% 0.922 0.953 0.689 0.943
EP
1.3.2, projected area-based scaling could be done directly with the derived parameters of ASTM
D3849-14a in Table 2. A key assumption for fractal analyses is that the two-dimensional solid
projection is self-similar, that is, as the aggregate size increases, both the length and the width
dimensions increase at similar rates. This may not be the case for soots [29, 41] and other
32
ACCEPTED MANUSCRIPT
aggregates[52]. Kaye [52] showed that evaluation of carbon black texture and growth structure
yielded different fractal dimensions. Wozniak et al [41] point out that correlations like Eq. 11 are
only appropriate for self-similar aggregates, for which the images would have scale-invariant
properties [53]. These researchers noticed that the smallest aggregates of their soot samples had
PT
significantly greater fractal dimension than the largest ones.
RI
The derived descriptors of ASTM 3849-14a had the highest measurement uncertainties of this
SC
study, indicating that their use for computing the fractal characteristics of the aggregates might
be problematic. The fractal plot, as represented by Eq. 3, shows the ratios of aggregate to nodule
U
areas as the x-ordinant and the ratios of aggregate to nodule volumes as the y-ordinant (N). Both
AN
ratios depend on the nodule diameter, which had a scale parameter measurement uncertainty
(UILC) of 12.9%, compared to typical scale parameter measurement uncertainties of ~ 4% for size
M
descriptors. This high uncertainty value is a primary concern. A fractal curve can be fitted to the
D
ASTM D3849-14a descriptor data, but the fractal exponent is Df = 3.7 ± 0.14 with a prefactor of
TE
1.98e-04 ± 1.14e-05. This is not a reasonable value for Df , which should not be greater than 2 for
Using the descriptors of Table 2, the estimate of the average nodule diameter depends on the
value of α = 0.40. In a number of plots, the slope of the fractal curve changes when this
AC
the equation for α in Table 2 appears to be the most significant problem, as it develops
33
ACCEPTED MANUSCRIPT
The method of Tian et al[45] for estimating dp is at least self-consistent and, as the nodule
diameter, area, and volume are all constants in this method, does not lead to unexpected changes
in slope. Table 9 shows the measurement uncertainties of the fractal analysis prefactors and
exponents using the method of Tian for an estimated nodule diameter of 21 nm. Coefficients of
PT
variation for the exponent are less than 0.54% for all labs, while coefficients of variation for the
RI
prefactor are less than 2.3% for all labs. For power law fits, it is typically the case that the
exponent has a lower coefficient of variation than the prefactor. For all datasets, the fractal
SC
exponent has a lower measurement uncertainty, 9.2%, than the prefactor, 35.3%.
U
constant [45])
AN
Fractal
Prefactor Exponent
parameter
Dataset ka Cv, % Df Cv, %
L1 0.907 2.1% 1.355 0.44%
M
Fractal analysis was also performed on the datasets deconvoluted by aggregate shape. The fractal
C
exponents for the spheroidal, ellipsoidal, and linear shapes are statistically distinct from each
AC
other within 95% certainty (using ± two standard deviations from the means). However, the
fractal exponent for the branched aggregates has a relatively large standard deviation, and
overlaps the confidence intervals of fractal exponents for all other shapes. It is interesting to
note that the fractal exponents decrease as the aggregate morphologies become more elongated.
In summary, the use of a constant value for dp [45] and the use of Eq. 3 provides fractal
34
ACCEPTED MANUSCRIPT
exponents and prefactors that appear to be self-consistent, both for complete datasets and for
of the fractal prefactors and exponents for the four shape morphologies across all laboratories.
The measurement uncertainties of the exponents are fairly low, except for the 12.8% value for
PT
the branched morphology.
RI
Table 10 Measurement uncertainties of fractal prefactors and exponents of different
aggregate morphology categories
SC
Shape Prefactor estimate Exponent estimate
morphology
ka UILC Df UILC
spheroidal 0.958 5.54% 1.49 0.57%
U
ellipsoidal 0.921 19.8% 1.41 3.08%
branched 0.898 60.5% 1.34 12.8%
AN
Linear 0.871 47.5% 1.30 4.08%
M
The specific surface area of a carbon black is often critical to its performance in its applications.
The Nitrogen Surface Area (NSA) from Table 1 is 142 m2/g. Using a density of carbon of 2100
D
kg/m3 and representing the nodules as individual spheres, this converts to an average nodule size
TE
of 20.1 nm, similar to both the typical reported value of 18 nm (see Note 2 of Table 1) and the
estimate of 21.6 nm used for the application of Eq. 3 to the data of this study. The contribution of
EP
spheroidal, ellipsoidal, branched, and linear categories to the specific surface area of the sample
can be estimated by converting the number balances of morphology particles to mass balances.
C
The number of nodules for each aggregate is proportional to the mass of the aggregate. The
AC
contribution of each category to the specific surface area is proportional to the total number of
nodules. For the data of Lab 1, the specific surface area contributions of the spheroidal,
ellipsoidal, branched, and linear aggregates is 0.204%, 25.0%, 37.3%, and 37.5%, respectively.
Thus, most of the mass of carbon black aggregates is made up of branched or linear shapes.
35
ACCEPTED MANUSCRIPT
is critical to getting accurate and representative particle images. Example images with good
contrast would help labs identify instrument conditions conducive to quality image acquisition.
PT
Some labs reported particles with less than the suggested number of pixels needed to control the
accuracy of size estimates. Both of these elements need to be explicitly cited in a general
RI
protocol. ANOVA analysis of descriptors showed that interlaboratory reproducibilities were the
SC
best for the Feret and minFeret size descriptors and the aspect ratio shape descriptor. The
lognormal reference model provided the best representation of aggregate size descriptors; the
U
measurement uncertainties of its scale and breadth parameters were lower than those of other
AN
model choices. Shape descriptor distributions appear to be multimodal. SRB8 aggregates have an
affinity exponent of 0.895, suggesting that they are self-affine rather than self-similar. The
M
derived descriptors used for the ASTM D3849 standard have fitted parameters with the highest
measurement uncertainties of the study. In addition, the parameters of this standard do not appear
D
to generate realistic values for the fractal exponents. However, an estimate of the average nodule
TE
diameter can be used with Eq. 3 to generate self-consistent fractal exponents for the complete
EP
datasets as well as those for aggregates separated by shape morphologies. Size, shape, and
by shape, demonstrating that aggregate shapes distribute non-uniformly across descriptor ranges
AC
proposed for carbon black shapes[7, 8, 46, 47], but only one set of descriptor criteria have been
to SRB8 ([47], which is based on a carbon black produced by benzene pyrolysis). It is quite
possible that the set of descriptor value used are not the best for the SRB8 sample. For example,
this study did not examine how well or poorly these descriptor ranges differentiate between the
36
ACCEPTED MANUSCRIPT
four categories. Such analyses would constitute a separate study, which would be facilitated by
the measurement uncertainty methods and data analysis tools reported here. These results
illustrate the importance and utility of determining data quality and measurement uncertainties
PT
Acknowledgements
RI
We acknowledge the support of the University of Kentucky’s College of Arts and Sciences in
hosting the Shiny Apps on their research server. The work by AIST was part of the research
SC
program, Strategic International Standardization Acceleration Projects, supported by the
U
AN
References
[1] Gray CA, Muranko H. Studies of robustness of industrial aciniform aggregates and
agglomerates--carbon black and amorphous silicas: a review amplified by new data. J Occup
Environ Med. 2006;48:1279-90.
M
[2] ASTM. D3849-14a Standard test method for carbon black - morphological
characterization of carbon black using electron microscopy West Conshohocken, PA: American
Society for Testing and Materials 2014.
D
[3] ASTM. ASTM D6556-16 Standard test method for carbon black - total and external
surface area by nitrogen adsorption. West Conshohacken, Pa: American Society for Testing and
TE
Materials 2016, p. 6.
[4] Cancarb-Limited. Physical and Chemical Properties. 2017 October 2, 2017 [cited;
Available from: http://www.cancarb.com/docs/pdf/Physical_Chemical_Properties.pdf
[5] Donnet J-B, Bansal RC, Wang M-J. Carbon Black. Microstructure, morphology and
EP
general physical properties. 2nd ed. Basel, Switzerland: Marcel Dekker, Inc.; 1993.
[6] Wang M-J. Kirk-Othmer Encyclopedia of Chemical Technology. 5th ed. New York:
John Wiley 2001, p. 761-803.
C
[7] Herd CR, McDonald GC, Hess WM. Morphology of carbon-black aggregates: fractal
versus Euclidean geometry. Rubber Chem Technol. 1992;65:107-29.
AC
[8] Herd CR, McDonald GC, Smith RE, Hess WM. The use of skeletonization for the shape
classification of carbon-black aggregates. Rubber Chem Technol. 1993;66:491-509.
[9] Gruber TC, Herd CR. Anisometry measurements in carbon black aggregate populations.
Rubber Chem Technol. 1997;70:727-46.
[10] Hess WM, Herd CR, McDonald GC. Influence of carbon black morphology and surface
activity on vulcanizate properties. Rubber World. 1993;208:26-32.
[11] Hess WM, McDonald GC. Morphological analysis of carbon black. ASTM Spec Tech
Publ. 1974;553:3-18.
[12] Hess WM, McDonald GC. Improved particle size measurements on pigments for rubber.
Rubber Chem Technol. 1983;56:892-917.
37
ACCEPTED MANUSCRIPT
[13] Janzen J, Goodarz-Nia I. Stereological functions for estimating surface areas and
volumes of random flocs. J Colloid Interface Sci. 1979;69:476-85.
[14] Castellano M, Conzatti L, Turturro A, Costa G, Busca G. Influence of the Silane
Modifiers on the Surface Thermodynamic Characteristics and Dispersion of the Silica into
Elastomer Compounds. J Phys Chem B. 2007;111:4495-502.
[15] Castellano M, Falqui L, Costa G, Turturro A, Valenti B, Castello G. Investigation on
elastomer-silica interactions by inverse gas chromatography and image analysis aided
PT
transmission electron microscopy. J Macromol Sci, Phys. 2002;B41:451-71.
[16] Conzatti L, Costa G, Castellano M, Turturro A, Negroni M, Gerard JF. Morphology and
viscoelastic behaviour of a silica filled styrene/butadiene random copolymer. Macromol Mater
RI
Eng. 2008;293:178-87.
[17] Costa G, Dondero G, Falqui L, Castellano M, Turturro A, Valenti B. Investigation into
the interactions between filler and elastomers used for tires production. Macromol Symp.
SC
2003;193:195-207.
[18] Falqui L, Costa G, Castellano M, Turturro A, Valenti B. A morphometric investigation
by TEM/AIA on elastomer-based compounds filled with an untreated precipitated silica. Rubber
Chem Technol. 2003;76:899-911.
U
[19] Jiang Z-H, Jin J, Xiao C-F, Li X, Kong L-X. Characterization of carbon black particles
mixing in polymeric matrix by direct methods. J Test Eval. 2012;40:227-37.
AN
[20] Lu S, Chung DDL. Viscoelastic behavior of carbon black and its relationship with the
aggregate size. Carbon. 2013;60:346-55.
[21] Rice SB, Chan C, Brown SC, Eschbach P, Han L, Ensor DS, et al. Particle size
distributions by transmission electron microscopy: An interlaboratory comparison case study.
M
Metrologia. 2013;50(6):663-78.
[22] Grulke EA, Yamamoto K, Kumagai K, Hausler I, Oesterle W, Ortel E, et al. Size and
shape distributions of primary crystallites in titania aggregates. Advanced Powder Technology.
D
2017 - accepted.
[23] VAMAS. VAMAS guidelines for the design and operation of interlaboratory
TE
[25] Medalia AI, Heckman FA. Morphology of aggregates. II. Size and shape factors of
carbon black aggregates from electron microscopy. Carbon. 1969;7(5):567-82.
[26] Hentschel ML, Page NW. Selection of descriptors for particle shape characterization. Part
Part Syst Charact. 2003;20(1):25-38.
C
[27] ISO. ISO 9276-6 Representation of results of particle size analysis - Part 6: Descriptive
and quantitative representation of particle shape and morphology. Geneva: ISO 2008, p. 23.
AC
[28] ASTM. ASTM F1877-16 Standard practice for characterization of particles. West
Conshohocken, PA: American Society for Testing and Materials International 2016, p. 15.
[29] Neimark AV, Koylu UO, Rosner DE. Extended characterization of combustion-generated
aggregates: self-affinity and lacunarities. J Colloid Interface Sci. 1996;180(2):590-7.
[30] Chakrabarty RK, Moosmuller H, Arnott WP, Garro MA, Slowik JG, Cross ES, et al.
Light scattering and absorption by fractal-like carbonaceous chain aggregates: comparison of
theories and experiment. Appl Opt. 2007;46(28):6990-7006.
38
ACCEPTED MANUSCRIPT
PT
Geophys Res, [Atmos]. 2006;111(D7):D07204/1-D/16.
[34] Ehrburger-Dolle F, Lahaye J, Misono S. Percolation in carbon black powders. Carbon.
1994;32(7):1363-8.
RI
[35] Ehrburger-Dolle F, Tence M. Determination of the fractal dimension of carbon black
aggregates. Carbon. 1990;28(2-3):448-52.
[36] Ferraro G, Fratini E, Rausa R, Fiaschi P, Baglioni P. Multiscale Characterization of Some
SC
Commercial Carbon Blacks and Diesel Engine Soot. Energy Fuels. 2016;30(11):9859-66.
[37] Lapuerta M, Ballesteros R, Martos FJ. A method to determine the fractal dimension of
diesel soot agglomerates. J Colloid Interface Sci. 2006;303(1):149-58.
[38] Li Q, Manas-Zloczower I, Feke DL. A simple methodology for the modeling of carbon
U
black aggregate structure. Rubber Chem Technol. 1996;69(1):8-14.
[39] Richter R, Sander LM, Cheng Z. Computer simulations of soot aggregation. J Colloid
AN
Interface Sci. 1984;100(1):203-9.
[40] Sachdeva K, Attri AK. Morphological characterization of carbonaceous aggregates in
soot and free fall aerosol samples. Atmos Environ. 2008;42(5):1025-34.
[41] Wozniak M, Onofri FRA, Barbosa S, Yon J, Mroczka J. Comparison of methods to
M
methods of fractal analysis applied to soot aggregates from wood combustion. J Aerosol Sci.
2006;37(7):820-38.
TE
[43] Eggersdorfer ML, Kadau D, Herrmann HJ, Pratsinis SE. Aggregate morphology
evolution by sintering: Number and diameter of primary particles. J Aerosol Sci. 2012;46:7-19.
[44] Tian K, Thomson KA, Liu F, Snelling DR, Smallwood GJ, Wang D. Determination of
EP
the morphology of soot aggregates using the relative optical density method for the analysis of
TEM images. Combust Flame. 2006;144(4):782-91.
[45] Tian K, Liu F, Thomson KA, Snelling DR, Smallwood GJ, Wang D. Distribution of the
number of primary particles of soot aggregates in a nonpremixed laminar flame. Combust Flame.
C
2004;138:195-8.
[46] Shishido F, Hashiguchi H, Matsushita Y, Morozumi Y, Aoki H, Miura T. An
AC
investigation of primary particle growth and aggregate formation of soot using a numerical
model considering the sintering of primary particles. Kagaku Kogaku Ronbunshu.
2007;33(4):306-14.
[47] Ono K, Yanaka M, Tanaka S, Saito Y, Aoki H, Fukuda O, et al. Influence of furnace
temperature and residence time on configurations of carbon black. Chem Eng J. 2012;200-
212:541-8.
[48] NIST. Guidelines for evaluating and expressing the uncertainty of NIST measurement
results: National Institute of Standards and Technology; 1994.
39
ACCEPTED MANUSCRIPT
[49] NIST. Report of Investigation. Reference Material 8012. Gold nanoparticles, nominal 30
nm diameter. Gaithersburg, MD: National Institute of Standards and Technology; 2007 13
December 2007.
[50] Kestens V, Roebben G. Certification Report - The certification of equivalent diameters of
a mixture of silica nanoparticles in aqueous solution: ERM-FD102. Geel, Belgium: IRMM;
2014.
[51] ISO. ISO 9276-3 Representation of results of particle size analysis - Part 3: fitting of an
PT
experimental cumulative curve to a reference model. Geneva: ISO 2008.
[52] Kaye BH. Chaotic perspectives in powder technology: A review. Part Sci Technol.
2001;19(4):301-26.
RI
[53] Theiler J. Estimating fractal dimension. Journal of Optical Society of America A.
1990;7(6):1055-73.
U SC
AN
M
D
TE
C EP
AC
40
ACCEPTED MANUSCRIPT
PT
4 Sample preparation for ASTM Reference Series 8 Sample B (SRB8)
5 Criteria for removing aggregates below the minimum area, labs L1-L4
6 Scale and width measurement uncertainties of descriptors across all data sets. Eq. 4 is
used to compute the coefficient of variation for each scale and width parameter,
RI
which is used to compute the measurement uncertainty using Eq. 5. Unimodal models
7 Measurement uncertainty of the affinity exponent
8 Comparison of affinity exponent and aspect ratios for different aggregate morphology
SC
categories; data of Lab L1
9 Measurement uncertainties of fractal prefactors and exponents (Eq. 3 with dp =
constant [45])
U
10 Measurement uncertainties of fractal prefactors and exponents of different aggregate
morphology categories
AN
Figures
1 Image of carbon black aggregates, SRB 8, with good contrast between particles and
background. Courtesy of Cabot Corporation
M
3 The lognormal distribution provides a good fit to the Feret data of Lab L1. Black
curve = smoothed data, blue curve = model parameters fitted using nonlinear
regression, red curve = model parameters fitted using maximum likelihood.
TE
4 Normal distribution fits to the aspect ratio data of Lab L1. Both fitted curves give
maximum density values near that of the data; however, the smoothed density data
(black curve) shows three possible peaks near its maximum. Blue curve = fitted
EP
parameters using nonlinear regression, red curve = fitted parameters using maximum
likelihood.
5 Normal distribution fits to the form factor data of Lab L1. The maxima of the fitted
models are shifted to higher values of form factor than the smoothed data (black
C
curve), which shows multiple possible peaks above and below its maximum value.
Blue curve = fitted parameters using nonlinear regression, red curve = fitted
AC
1
Jincheng Xiong, private communication
ACCEPTED MANUSCRIPT
aggregates had high aspect ratios. Red circles = spheroidal, blue diamonds =
ellipsoidal, green triangles = branched, orange bars = linear, black curve = sum of all
morphology categories.
8 Aggregate form factor assembled from its morphology category distributions. Red
circles = spheroidal, blue diamonds = ellipsoidal, green triangles = branched, orange
bars = linear, black curve = sum of all component types. Lab L1.
PT
9 Affinity exponent plot (Eq. 2) for the data of Lab L1; solid line = model equation
with H = 0.895 ± 0.00098
RI
U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
PT
Note 1: aggregate equivalent circular diameter ~ 54 nm
RI
Note 3: values in the 3rd row reflect the specific method for making the measurement,
SC
e.g., the specific surface area values (m2/g) differ because the methods differ.
U
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
PT
Feret 1 Maximum Feret diameter,
corresponds to the maximum
RI
length of parallel tangents
(particle length) / [28]
MinFeret 1 Minimum Feret diameter,
SC
corresponds to the minimum
length of parallel tangents
(particle breadth) / [28]
P 1 Length of the perimeter
U
Shape/elongation
Aspect ratio 1 Ratio of the minimum Feret
AN
diameter to the maximum Feret
diameter
4∙
Compactness 1 Degree to which the projection
∙
M
4∙
D
Shape/ruggedness
4∙ ∙
Circularity 1 Degree to which the projected
area of the particle is similar to a
4∙ ∙
circle, based on its perimeter
C
PT
Model
Area, nm2 LRM 7.75 [2310] 0.154
1.044 0.0410 1.99% 3.92% 4.46% 8.77%
Perimeter, nm LRM 5.52 [249] 0.0859
0.696 0.0231 4.06% 12.8% 9.09% 28.7%
RI
Feret, nm LRM 4.39 [80.9] 0.0698
0.586 0.0170 1.59% 2.90% 3.55% 6.49%
minFeret, nm LRM 3.94 [51.5] 0.0777
0.570 0.0157 1.97% 2.76% 4.41% 6.18%
SC
ECD, nm LRM 3.97 [52.9] 0.0285
0.522 0.0205 0.72% 3.93% 1.60% 8.79%
elongational descriptors
Aspect ratio NRM 0.642 0.0076 0.133 0.0042 1.18% 3.19% 2.64% 7.14%
U
Compactness NRM 0.673 0.0098 0.0977 0.0064 1.45% 6.59% 3.24% 14.7%
ruggedness descriptors
AN
Form factor NRM 0.485 0.0496 0.212 0.0163 10.2% 7.67% 22.9% 17.1%
Solidity NRM 0.766 0.0218 0.112 0.0089 2.85% 7.94% 6.37% 17.8%
ASTM D3849-14a descriptors
M
dp, nm LRM 2.89 [18.1] 0.167 0.465 0.055 5.77% 12.2% 12.9% 27.3%
VP, nm3 LRM 8.03 [3077] 0.501 1.369 0.166 6.23% 12.2% 14.0% 27.3%
N LRM 2.87 [17.6] 0.391 1.695 0.070 13.6% 4.15% 30.5% 9.30%
D
LRM = lognormal reference model; NRM = normal reference model; CV = coefficient of variance, % (Eq. 4); UILC = measurement uncertainty, %
TE
(Eq. 5); [xx, nm] gives the exponential of the scale estimate for lognormal reference models.
C EP
AC
ACCEPTED MANUSCRIPT
PT
branched 0.898 60.5% 1.34 12.8%
Linear 0.871 47.5% 1.30 4.08%
RI
U SC
AN
M
D
TE
C EP
AC