Construction and Building Materials 273 (2021) 122019

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Determination of specific surface area of irregular aggregate by random

sectioning and its comparison with conventional methods
Qiang Ren a,b, Luchuan Ding b, Xiaodi Dai b, Zhengwu Jiang a,⇑, Guang Ye b,c, Geert De Schutter b,⇑
a
Key Laboratory of Advanced Civil Engineering Materials of Ministry of Education, School of Materials Science and Engineering, Tongji University, 4800 Cao’an Road, Shanghai
201804, China
b
Magnel-Vandepitte Laboratory, Department of Structural Engineering and Building Materials, Ghent University, Technologiepark-Zwijnaarde 60, Ghent 9052, Belgium
c
Microlab, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Stevinweg 1, 2628 CN Delft, The Netherlands

h i g h l i g h t s

An easy-to-perform method is proposed to determine SSA of irregular aggregate.

The proposed method is reliable compared to CT.

Difference among various methods is analyzed.

Suggestions are provided for selecting proper testing method of SSA.

a r t i c l e i n f o a b s t r a c t

Article history: A two-dimensional (2D) image-based methodology was proposed to measure the specific surface area
Received 24 March 2020 (SSA, specified as the surface area per unit volume) of irregular aggregate by random sectioning.
Received in revised form 12 November 2020 Conventional methods including spherical assumption, Brunauer-Emmett-Teller (BET) and computed
Accepted 12 December 2020
tomography (CT) tests were used and compared in this study. Results show that spherical assumption
provides the lowest SSA among these methods since the feature of anisotropy in dimension is not con-
sidered. SSA by BET test has one order of magnitude higher value than others, which is attributed to
Keywords:
the fact that BET method measures each position of particles that nitrogen molecule can be adsorbed
Aggregates
Specific surface area
on during the applied relative pressure, based on the ‘pixel’ of nitrogen molecule. The proposed random
Random sectioning sectioning method presents very similar SSA result compared to CT method, indicating that it can be con-
Spherical assumption sidered as a reliable method. To improve the estimation of SSA by random sectioning method, factors that
BET may influence SSA result were analyzed. Results indicate that the number of samples should be high
Computed tomography enough to reach a constant result and the thresholding algorithm should be adequate. Besides, a higher
resolution of pixel provides a higher SSA value. The comparison among these methods demonstrate that
it is necessary to determine the scale at which the features of the surface are supposed to be captured
before selecting the optimal testing method.
Ó 2020 Elsevier Ltd. All rights reserved.

1. Introduction [6,7] and durability [8,9] of concrete. SSA determination of aggre-

gates is thus a topic of great interest for concrete research.
Among the various properties of aggregate, specific surface area The original estimation of SSA was based on empirical relation-
(SSA) is a comprehensive evaluation of dimension, shape, angular- ships between SSA and certain easy-to-measure properties. For
ity and surface roughness of aggregate [1,2]. From the viewpoint of instance, particle size distribution (PSD) is widely adopted to cal-
mix design, SSA of aggregates influences the amount of surface to culate SSA by dividing the size distribution of aggregate into sev-
be wetted, hence the requirement of paste in concrete mixture, eral bins and assuming equivalent spherical or polyhedral
which influences the workability [3–5], mechanical properties particles that have the equivalent size of each bin [10,11]. Besides,
an equation of the form of SA ¼ eV 0:667 was proposed by Erdogan
[12] where SA is surface area and V is volume while e is a factor
⇑ Corresponding authors. related to the dimensions of the particles. However, aggregates
E-mail addresses: [email protected] (Z. Jiang), [email protected] are 3D random particles with irregular and diverse shapes. This
(G. De Schutter).

https://doi.org/10.1016/j.conbuildmat.2020.122019
0950-0618/Ó 2020 Elsevier Ltd. All rights reserved.
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

non-equidimensional feature is more remarkable for crushed consideration that aggregates with various lithology have different
aggregates like manufactured sand, which is attracting increasing apparent densities.
attention as fine aggregate in concrete [13]. Certain correction fac-
tors like flakiness & elongation index [14] or surface area factor 2. Conventional methods
[15–17] were used to include the influence of particle shape during
SSA calculation. However, these methods consider the overall 2.1. Spherical assumption
shape influence on SSA but fail to involve the influence of proper-
ties like angularity and roughness, both of which increase the tor- PSD can be used to approximate SSA by dividing the size distri-
tuosity of particle surface. In addition to the semi-empirical bution of aggregate into several bins and assuming equivalent
estimations, indirect method was proposed by coating aggregates spherical or polyhedral particles that have the equivalent size of
with a liquid or a powder and comparing the amount of coated each bin. The SSA based on spherical assumption can be deter-
substance on aggregate particles with that on the reference having mined as follows.
a known surface area [18–20]. However, this method is operator-
sensitive and the result is influenced by many factors like aggre- F X 6f d
SSA ¼ ð1Þ
gate surface features given that different aggregates may have q D
diverse adsorbing performances. Similarly, gas was also used to where q is the average mass density of particle, f d is the volume
coat particles to approximate the surface area of solids while fraction of particles with diameter of D, F is an empirical correction
Brunauer-Emmett-Teller (BET) method is a most widely used and factor 1.13 to consider the particle surface shape [16,17]. In prac-
standardized one [21,22] SSA can be calculated based on the tice, PSD is determined by sieving. D is determined as the geomet-
amount of adsorbate on the surface of particles. However, BET ric mean of maximum and minimum diameters of each bin.
method shows higher value since the gas adsorption occurs not
only on the external surface but also on the surface of open internal 2.2. BET analysis
pores. BET result is thus highly influenced by the porosity of aggre-
gates, which may arise from factors like weathering and lithology 2.2.1. BET model
[23]. The principle of BET test is to quantify the amount of an adsor-
Image processing technology is another tool that can be used to bate (nitrogen in this study) required to form a closely packed
approximate the properties of particles based on their two- monomolecular layer on the whole surface of solid. This amount
dimensional (2D) or three-dimensional (3D) images. For instance, can be determined in the low partial pressure region of the adsorp-
scanning electron microscope (SEM) images were used to deter- tion isotherm. In previous investigations, the specific surface area
mine the SSA [17], particle shape and size [24], as well as mineral was expressed as the amount of area per mass of particles, shown
compositions [25] of cementitious materials based on image anal- as follows [31,32].
ysis. However, these 2D methods cannot be directly used for mea-
suring the SSA of irregular aggregates. In terms of 3D image V m NAm
SSA ¼ ð2Þ
methods, computed tomography (CT) is a well-developed technol- 22400M
ogy with considerable value in many fields [26–28]. The most where Am is the projection area of one vapor molecule which is
remarkable advantage of CT is its ability to quickly image the inte- determined based on the hexagonal close packing model, N is Avo-
rior of solid in three dimensions without any destruction [29]. A 3D gadro constant, and M is the mass of particles.
view of the sample can be obtained by stacking 2D slices. Then the For comparison with other methods, the specific surface are of
SSA or other geometric features can be measured based on the BET method is then presented as the amount of are per volume
reconstructed 3D sample. CT can be regarded as the current ulti- of particles, shown below.
mate testing method for SSA determination of aggregates. In spite
V m NAm q
of the advantages it offers, CT equipment is relatively costly, its SSA ¼ ð3Þ
usage for routine analyses is not practical yet. Therefore, CT plays 22400M
a better role as a benchmark for estimating new testing methods where q is the apparent density of measured particles.
in term of SSA.
It can be summarized that improvement on SSA determination 2.2.2. BET test
is achieved due to imaging techniques by eliminating human BET test was carried out using a BET multi-point nitrogen
errors. However, 3D image method like CT for the moment seem- physisorption equipment (BEISHIDE 3H-2000PS2) [33,34]. Samples
ingly has limited use for daily analyses owing to its capital inten- were degassed at 105 ℃ under vacuum for 6 h before adsorption
siveness and complicated operation requirement. As a result, it is test. The nitrogen adsorption capacity was determined from a
of great interest to develop a cheaper and easy-to-perform quan- six-points adsorption isotherm at 77.3 K in the relative pressure
tification method of SSA with high accuracy, which can be widely ranging from 0.05 to 0.30. Each sample was measured for three
used for general analyses. times.
If particles are randomly distributed, 2D sections could be used
to estimate the 3D properties based on the principles of stereology 2.3. CT analysis
[30]. SSA of aggregates is possibly determined based on 2D images
if the number of images is high enough given that CT result is 2.3.1. Principle of CT test
obtained from numerous 2D slices as well. In this paper, a 2D During CT testing, X-rays irradiate a 3D sample from various
image-based methodology is proposed to measure the SSA by ran- angles and a detector evaluates the resulting intensity of a known
dom sectioning of aggregate. The validation of the SSA determina- unidirectional (x-axis) X-ray beam intensity due to the absorption
tion approach is then accomplished by comparing SSA result with by the sample, and for different directions of irradiations [35,36].
that from spherical assumption, BET and CT methods. Besides, the According to attenuation measurements over many views (ray
reasons behind the diverse results were discussed. Finally, factors paths), a 2D cross-sectional image can be mathematically recon-
that may influence SSA result of random sectioning method are structed. Then a computer-based reconstruction technique can be
analyzed. It needs to be mentioned that SSA in this paper refers used to produce gray images. Each image represents one slice of
to the surface area per unit volume of a particle for the the sample and the contrast in gray levels is attributed to the
2
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

difference of constituents in the sample regarding X-ray absorption 20

capacity. A 3D view of the sample can be obtained if these images
4
are successively stacked together. The SSA can be finally calculated

Relative error (%)

based on the reconstructed 3D structure. 18 2

0
2.3.2. CT test
16

SSA (mm /mm3)

-2
Tomographic data were obtained using Hector CT scanner at the
Center for X-ray Tomography, Ghent University (Belgium) [37]. The -4
raw data (sonograms) were reconstructed into 2D slice images 14

2
60 70 80 90 100 110 120
using the in-house developed software package Octopus [38].
The reconstructed voxel dimension is 73 mm3 based on a source-
to-object distance of 41.4 mm and a source-to-detector distance 12
of 1166.0 mm. The reconstructed images comprise 2000  2000
pixels and there are 1500 reconstructed images for one sample.
3D reconstruction was accomplished by using the software
10
Blob3D. The process of 3D reconstruction is illustrated in Fig. 1.
Segment procedures like thresholding, median smoothing, Gaus- 8
sian filtering and islands/holes removing were adopted to define 0 20 40 60 80 100 120
pixels that belong to the object of interest, indicated by the trans- Number of particles
formation of gray images into binarized images. After the forma-
tion of 3D structure by stacking the successive 2D binarized Fig. 2. SSA results measured from CT method based on various number of
images, 3D particles contact with adjacent ones. Thus, the proce- reconstructed 3D particles.
dure of separation was performed, like size thresholding, erosion/
dilation and plane definition, to separate the reconstructed and
connected particles. It is to be noted that erosion/dilation opera- 60. Thus 60 reconstructed particles can be considered sufficient for
tions may slightly influence the morphology of a particle, thus ero- SSA determination based on CT test, compared to 30 particles in
sion/dilation operation was used as less as possible. Finally, SSA Ref [2].
can be calculated based on the isosurface surrounding the object
voxels. 3. Random sectioning method

2.3.3. CT result analysis 3.1. Theoretical background

SSA calculated from various numbers of reconstructed 3D parti-
cles is shown in Fig. 2. It is observed that SSA significantly fluctu- Random sectioning method was proposed by Smith and Gutt-
ates for the first 10 particles. Then a sluggish variation of SSA can man to measure the internal boundaries in 3D structures [39]. This
be seen when more particles are involved. The coefficient of varia- method is employed to calculate the SSA of particles in this study.
tion of SSA after 60 particles is 1.4% while the maximum relative The derivation of equations for calculating SSA are shown as fol-
error of SSA from more than 60 particles to that from 60 particles lows based on references [30,39].
(superposed in Fig. 2) is 2.0%, where such small values indicate that Fig. 3 describes an irregular solid body with closed surface,
a relatively constant SSA is available when particle number reaches which is intersected by a stack of parallel planes with a distance

Fig. 1. 3D reconstruction process of CT analysis.

3
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

R R
dx dzlðx; zÞ
l¼ R R ð7Þ
dx dz
If the orientation is fixed,ldz is surface element of the figure pro-
jection on a certain plane that is normal to the stack of planes and
includes the intersection.

ldz ¼ dSsinh ð8Þ

where h is the angle between the normal direction of the figure
and the normal direction of the parallel planes. Thus, the integra-
tion over z presents the total area of the surface projected on the
vertical plane. The orientation angle x can be expressed as follows.

dx ¼ sinhdh ð9Þ
The range (0, p/2) of h covers all possible orientations. Thus, Eq.
(7) can specified as Eq. (10).
Fig. 3. A solid body intersected by a set of parallel planes. R p=2
S sin hdh
2
pS
l ¼ R p=20 Rh ¼ ð10Þ
0
sinhdh 0 dz 4h
h. The position of the surface of the object can be defined by the
distance z of a fixed point on the aggregate surface to the intersec- This result is not confined to plane figures given that any sur-
tion plane. The orientation of this surface can be described by angle face is constituted by a large number of plane elements. Besides,
x for simplicity. The area (A) of the intersection is thus a function these elements take random positions and orientations when the
figure as a whole dose so. It needs to be mentioned that the contri-
of x and z. The average area (A) for all possible positions and ori-
butions to the surface and to the length of intersection are additive.
entations of the object, can be determined based on Eq. (4).
As a result, Fig. 4 holds for any surface. Therefore, the average
R R
dx dzAðx; zÞ length (l) indicates the average perimeter of all possible sections
A¼ R R ð4Þ
dx dz of the object.
By combining Eqs. (6) and (10), the following equation can be
If the orientation is fixed, Aðx; zÞdz is the volume element of the obtained.
object.
Z l pS
d
¼ ð11Þ
Aðx; zÞdz ¼ V ð5Þ A 4V
0
This result does not depend on h, which indicates that the set of
where V is the volume of the object.
parallel planes are not needed in essence. It is practicable to aver-
Therefore, Eq. (4) can be simplified as Eq. (6), which indicates
age the line length and area of the intersections of any solid by a
that the average area of intersection for all possible positions and
randomly-oriented and -placed plane.
orientations of the object is the volume of the body divided by
Eq. (11) can be presented as the following one, from which the
the distance of the parallel planes.
SSA of a particle can be determined based on the ratio between
V average perimeter and average area of sections at random posi-
A¼ ð6Þ tions and orientations. This equation is supposed to be valid for
h
particle mixtures if the number of intersections is large enough.
Now, we consider a surface element of the object (red area in
Fig. 3), which intersects one of the parallel planes. This surface ele- S 4 l
ment can be taken as a plane shown in Fig. 4. The length of the ¼ ð12Þ
V pA
intersection is l and its average for all positions and orientations
can be expressed as Eq. (7) [39].
3.2. Random sectioning test

3.2.1. Sample preparation

Manufactured sand particles between 0.50 mm and 1.00 mm
sieve sizes were used in this study (Fig. 5(a)) in view of its signif-
icantly irregular geometrical features among aggregates. Sand par-
ticles were washed and dried first. Then they were impregnated
into epoxy resin (EpoFix ResinTM) in a silicon mold with an inner
diameter of 20 mm. In order to increase the contrast between
objects and background, epoxy resin was blended with 0.5% (mass
ratio) fluorescent dye (EpoDyeTM) and homogenized beforehand.
The sample was cured at 40 ℃ for 24 h to harden. The hardened
sample was then ground and polished by a polishing machine
(LaboPol-5, Struers) at a rotational speed of 200 rpm using SiC
papers #180, #320, #1200 and #2000 successively, during which
one-quarter of a turn was made every half a minute for a total of
2 min for each SiC paper. Water was used as the cooling medium.
Finally, a smooth surface with abundant exposed particles can be
Fig. 4. A unit surface intersected by a set of parallel planes. obtained after these procedures, seen in Fig. 5(b). Afterwards, the
4
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

Fig. 5. Sand particles (a) and sample for image acquisition (b) under optimal microscope (c).

optical microscope (Leica S8 APO, Fig. 5(c)) with a fluorescent light 0.035
was used to capture the section image of particles. Probability density
0.030 Gauss fitting
3.2.2. Image processing
The objective of image processing is to measure the perimeter 0.025

Probability density
and surface of particles on the cross section. The captured image
(shown in Fig. 6(a)) was transformed into gray image (Fig. 6(b)) 0.020
for the convenience of thresholding, which is to identify the objects
from the background. There exist several thresholding algorithms 0.015
[40,41], mainly based on the gray tone level histogram of the
image. Among these methods, the minimum error algorithm views 0.010
the gray level histogram as a probability density function (PDF) of
the gray levels of both the object and background (j = 1,2). Besides,
0.005
each of them is considered to follow a normal distribution with a
mean value of mðjÞ, a standard deviation of rðjÞ and a PDF of PðjÞ.
0.000
The PDF of gray level of gray image (Fig. 6(b)) is shown in Fig. 7, 0 50 100 150 200 250
where two normally distributed segments are observed with a fit- Gray level
ted R2 of 0.960, which indicates the feature of the minimum error
algorithm. Thus, the minimum error algorithm is used in this Fig. 7. Probability density of gray level histogram of Fig. 6(b).
study. However, the parameters of mðjÞ, rðjÞ and PðjÞ are usually
unknown. Instead, a criterion function JðjÞ is employed, expressed
X
t
as Eq. (13) [42]. P 1 ðt Þ ¼ pðiÞ ð14Þ
J ðtÞ ¼ 1 þ 2½P1 ðt ÞLnr1 ðtÞ þ P2 ðt ÞLnr2 ðtÞ  2½P1 ðt ÞLnP1 ðtÞ
i¼0

þ P2 ðt ÞLnP 2 ðt Þ ð13Þ X
L1
P 2 ðt Þ ¼ pðiÞ ¼ 1  P1 ðt Þ ð15Þ
where P1 ðtÞ is the cumulative probability of gray values less i¼tþ1
than or equal to t while P2 ðt Þ is the cumulative probability of gray
values higher thant, which can be determined according to Eqs. 1 X t

(14) and (15) respectively. r1 ðt Þ is the standard deviation of gray r1 ðtÞ ¼ 2

½i  m1 ðt Þ  pðiÞ ð16Þ
P1 ðtÞ i¼0
values less than or equal to t while r2 ðtÞ is the standard deviation
of gray values higher than t, calculated as Eqs. (16) and (17)
1 X L1
respectively. r2 ðtÞ ¼ 2
½i  m2 ðtÞ  pðiÞ ð17Þ
P1 ðtÞ i¼tþ1

Fig. 6. (a) original and (b) gray images.

5
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

where m1 ðtÞ is the mean of gray values less than or equal to t be the high-frequency, short-wavelength component of a mea-
while m2 ðt Þ is the mean of gray values higher than t. m1 ðtÞ and sured surface. Theoretically, CT and random sectioning methods
m2 ðt Þ are determined as follows. are supposed to theoretically consider any geometrical shapes,
depending on their voxel or pixel size.
1 X t
There is no doubt that CT is the most accurate method to verify
m1 ðt Þ ¼ i  pðiÞ ð18Þ
P1 ðtÞ i¼0 new methods if the resolution is proper given that there are a lot of
factors like assumptions and degassing conditions, influencing the
1 X L1 result of BET methods [17,43]. From the viewpoint of resolution,
m2 ðt Þ ¼ i  pðiÞ ð19Þ we can take the surface of a particle as the combination of huge
P2 ðtÞ i¼tþ1
numbers of small pixels. The resolution of BET method is approxi-
The optimal threshold t  is then determined by minimizing J ðt Þ, mately the diameter of one N2 molecule (around 0.4 nm) while the
as described in Eq. (20). resolutions of CT and random sectioning methods depend on the
resolution of equipment (CT scanner and optical microscope
Jðt Þ ¼ min JðtÞ ð20Þ respectively) or/and user’s selection, 7 mm and 4 mm respectively
t2GL
in this paper. When the resolution is too low, the roughness is
The gray image can be transformed into a binary image with the unable to be captured and SSA is underestimated. However, an
determined threshold value, which identifies the objects from the overestimated SSA will be provided if the pixel is too small due
background. Then numbers of pixels representing particles and to ‘‘roughness” induced by pixilation. This explains why BET
their boundaries are available by the software MATLAB. Therefore, method shows much higher SSA than other methods. To sum up,
the SSA of the particles can be calculated from (21). it is necessary to determine the scale at which the features of the
S 4 NP surface are intended to be captured before selecting the optimal
¼ ð21Þ
V p NA L0 testing method.
It is considered that 3D image methods are supposed to mea-
where N A and N P are the numbers of pixels corresponding to sure the SSA at a higher accuracy than 2D image methods due to
objects and their boundaries, L0 is the edge length of each pixel. the fact that 3D methods are able to capture the surface features
on the third dimension. However, if the number of 2D sections is
4. Comparison and discussion high enough, this influence can be alleviated. Besides, a lot of test-
ing steps and image processing procedures were performed on
The SSA results from spherical assumption, BET, CT and random samples to obtain data for SSA calculation and a lot of parameters
sectioning methods are exhibited in Table 1. It is clear that SSA were selected during image processing, especially for CT method.
from BET is one order or two orders of magnitude higher than Each procedure and parameter affects the SSA result. From this
SSA from the other methods while the SSA calculated based on viewpoint, more deviations are likely to be employed during image
the spherical assumption shows the lowest value. However, it is processing of CT method. As a whole, random sectioning method
to be noted that CT and random sectioning methods provide very provides similar result with CT test while their resolutions are
similar results. comparable. Since the ‘real’ surface area of the sample is not avail-
Spherical assumption fails to consider the influence of complex able, the accuracy of these two methods cannot be evaluated. How-
geometrical features of particles on SSA. For particles with various ever, in general, CT is well considered to be able to accurately
shapes, centrosymmetric particle with smooth surface (perfect measure the SSA. Thus, random sectioning method can be taken
sphere) has the lowest specific surface area while SSA increases as a reliable method compared to CT.
if the particle is deviating from central symmetry or if the surface
becomes rougher [11,12]. This means that the spherical assump-
5. Influence factors on SSA results of random sectioning method
tion underestimates the SSA of real particles and the employed cor-
rection factor is too small to capture the influence of particle shape.
This section is to analyze the factors that may influence SSA
Even though Platonic shapes like tetrahedron, cube, icosahedron,
results, including the number of particles used for calculation,
dodecahedron and octahedron were ever employed to mitigate
threshold value and the resolution of pixel.
the particle shape influence on SSA evaluation, regular shape
assumption cannot be used to accurately calculate the SSA of irreg-
ular aggregate given that the influence of geometric features on the 5.1. Number of particles
SSA is not predictable due to the irregular feature. In the case of
BET method, the surface of open inner pores of particles is also Number of particles should be large enough to present intersec-
counted in for SSA calculation in addition to their external surface tions at random positions and from random orientations, indicated
area. Thus, BET is considered to overestimate the SSA of particles by a steady SSA result. Fig. 8 shows the SSA results calculated from
and this effect is especially significant for particles with porous various numbers of particles. It is seen that SSA presents rather
structures. fluctuating value within the first few tens of particles and then
CT and random sectioning methods are direct visual analysis shows increasingly steady results as the number of particles
methods while CT is based on 3D reconstructed particle shape increases. To be specific, SSA in the range of 12 mm2/mm3 to
and random sectioning method is based on the statistical data from 13 mm2/mm3 can be noticed if more than 400 particles are used.
numerous 2D sections. It needs to be mentioned that SSA is not In addition, the coefficient of variation of SSA from 800 to 1500
only influenced by the overall shape of particles, but also affected particles is 0.4% while the maximum relative error of SSA from
by surface features like roughness, which is typically considered to more than 800 particles to that from 800 particles (i.e. SSA/

Table 1
SSA results from different methods (mm2/mm3).

Methods Spherical assumption BET CT Random sectioning

SSA 9.6 329.7 12.6 12.2

6
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

observed that when threshold is smaller than the critical value

20 4 (Fig. 10a), particles were partially eroded, making boundaries more
tortuous. Erosion decreases the area of objects and increases their

Relative error (%)

2
perimeter at the same time. This influence is supposed to be more
18
0 significant when thresholding value further decreases. As a result,
SSA sharply increases with the decreasing threshold value before
SSA (mm /mm )

-2
3

16 the critical point. However, the increase of threshold value after

-4 the critical point dilates the particles (Fig. 10c), which overesti-
mates the area of objects on one hand. On the other hand, objects
2

14 800 1000 1200 1400

will merge with the adjacent ones, resulting in loss of perimeter.
Thus, the SSA is decreasing with the increase of threshold value.
12 This influence is gentler, indicated by the sluggish decreasing
trend. In addition to the minimum error algorithm, other com-
monly used thresholding methods including tangent-slope method
10
[44], K-means method [40], Otsu’s method [45], fuzzy algorithm
[46] and entropy maximization method [47] were also tried and
8 the determined threshold values are also superposed in Fig. 9. It
0 200 400 600 800 1000 1200 1400 is observed that different thresholding methods provide diverse
Number of particles threshold values, ranging from 27 to 129. For the bimodal his-
togram of gray level in this paper, various algorithms are expected
Fig. 8. SSA results versus the number of particles used for calculation. to provide similar threshold values around the valley between the
peaks of foreground histogram and background histogram. How-
ever, this valley is quite wide and flat according to Fig. 7. Minor dif-
SSA800-1, superposed in Fig. 8) is 1.0%, where such a small error ferences between the algorithms can make the resultant threshold
means a rather steady SSA without obvious variation is reached values vary a lot. As a result, compared with SSA based on the min-
from 800 particles. As a result, we believe that 800 particles are imum error algorithm, the tangent-slope method shows 54.2%
sufficient to approach the steady SSA based on random sectioning higher while K-means method, Otsu’s method, fuzzy algorithm
method for the tested manufactured sand particles. and entropy maximization method respectively present 17.0%,
It needs to be mentioned that the number of particles for a 17.1%, 17.1% and 27.0% lower SSA results than that from the min-
steady SSA result depends on the dispersity of particle shape and imum error algorithm.
particle size. A larger number of particles are required when parti-
cles have higher dispersity in shape and a wider particle size
distribution. 5.3. Resolution of pixel

A proper resolution is desired for image analysis. Fig. 11 shows

5.2. Threshold value the SSA variation with the resolution of pixel. It needs to be men-
tioned that a smaller pixel size requires more particles for an unbi-
Thresholding is a vital procedure for image processing. Various ased estimation. SSA results in Fig. 11 are obtained based on
threshold values ranging from 20 to 200 were used for image pro- sufficient particles for each resolution of pixel. It is clear that SSA
cessing and SSA were calculated as shown in Fig. 9. In general, SSA exhibits an increasing trend with the increase of resolution (de-
decreases with the increase of gray level. SSA shows a dramatic crease of pixel size). Specifically, SSA shows 24.0%, 35.8%, 54.0%,
decrease before a critical value, followed by a sluggish decrease 55.5% and 56.6% higher values when the resolution increases from
after this critical value. This critical value is considered to be the 128 lm/pixel to 64 lm/pixel, 32 lm/pixel, 16 lm/pixel, 8 lm/
threshold value, which is 40 for the gray image (Fig. 6) according pixel, and 4 lm/pixel, respectively. This trend is attributed to the
to the minimum error algorithm (see above). The binary images fact that a higher resolution captures more details on the geomet-
after thresholding at 30, 40 and 100 are shown in Fig. 10. It is rical shape like surface texture.
It can be seen in Fig. 12 that boundaries information is increas-
80 ingly clear when the resolution is gradually improved. Therefore,
the resolution should be as high as possible for SSA determination
70
when the geometrical features at smaller scale are expected to be
Otsu's method / Fuzzy alogrithm

considered. However, it is also noted that SSA shows a sluggish

Tangent-slope method

Entropy maximization method

60
increase when the resolution is higher than 16 lm/pixel. Specifi-
Minimum error algorithm

cally, SSA increases by 2.8% when the resolution increases from

SSA(mm /mm )

50
3

16 lm/pixel to 8 lm/pixel. Further improvement of resolution

40 from 8 lm/pixel to 4 lm/pixel provides 2.0% higher SSA value.
K-means method
2

Therefore, a resolution higher than 16 lm/pixel provides relatively

30 reliable SSA result while 4 lm/pixel is used in this study if not
specified otherwise. In engineering practice, the balance between
20 pixel size and number of particles should be made to make the
result reliable with both pixel size and number of particles
10
considered.
0
0 20 40 60 80 100 120 140 160 180 200
6. Conclusion
Threshold value
Fig. 9. SSA results and threshold values from several commonly used thresholding Random sectioning method was proposed for the determination
methods. of SSA of irregular-shaped aggregate. For validation, conventional
7
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

Fig. 10. Binary images after thresholding at (a) 30, (b) 40 and (c) 100.

14 Besides, the implementation of random sectioning method is sim-

ple, independent of expensive equipment and complicated
13 operations.
In practice of SSA determination, it is necessary to determine
12 the scale at which the features of surface are supposed to be cap-
tured before selecting the optimal testing method.
SSA (mm /mm )
3

11
2

10 CRediT authorship contribution statement

9 Qiang Ren: Conceptualization, Methodology, Investigation, For-

mal analysis, Writing - original draft. Luchuan Ding: Methodology,
8 Writing - review & editing. Xiaodi Dai: Formal analysis, Writing -
review & editing. Zhengwu Jiang: Conceptualization, Writing -
7 review & editing. Guang Ye: Formal analysis, Writing - review &
0 20 40 60 80 100 120 140 editing. Geert De Schutter: Methodology, Writing - review &
m/pixel editing.
Fig. 11. SSA results under various resolutions.

Declaration of Competing Interest

methods including spherical assumption, BET and CT methods
were used for comparison. Factors that possibly influence SSA
The authors declare that they have no known competing finan-
result were analyzed, like number of particles, threshold value
cial interests or personal relationships that could have appeared
and the resolution of pixel.
to influence the work reported in this paper.
Spherical assumption provides the lowest SSA result due to the
fact that the influence of irregular geometric features on SSA is not
considered. BET method measures the surface that nitrogen mole-
cule can reach and be adsorbed on in the relative pressure range of Acknowledgements
0.05 to 0.30, including open pore surface. This method approxi-
mates the SSA based on the ‘pixel’ of nitrogen molecule. As a result, The authors acknowledge the financial supports provided by
BET method provides one order of magnitude higher SSA than the National Key Research and Development Projects
other three methods. (2018YFC0705404), National Natural Science Foundation of China
Both CT and random sectioning measure the SSA of particles (51878480, 51878479, 51678442, 51878481 and 51878496) and
based on direct visual analysis. Random sectioning method pro- the Fundamental Research Funds for the Central Universities.
vides very similar result with CT test and can be considered as a Authors gratefully acknowledge the Centre for X-ray Tomography
reliable method compared to CT result in terms of accuracy. of Ghent University for its support on CT test.

Fig. 12. Binary images with resolutions of (a) 128 lm/pixel, (b) 64 lm/pixel and (c) 4 lm/pixel.

8
Q. Ren, L. Ding, X. Dai et al. Construction and Building Materials 273 (2021) 122019

References [25] P.T. Durdziński, C.F. Dunant, M.B. Haha, K.L. Scrivener, A new quantification
method based on SEM-EDS to assess fly ash composition and study the
reaction of its individual components in hydrating cement paste, Cem. Concr.
[1] M.R. Mitchell, R.E. Link, D.Q. van Lent, A.A.A. Molenaar, M.F.C. van de Ven, M.
Res. 73 (2015) 111–122, https://doi.org/10.1016/j.cemconres.2015.02.008.
van de Ven, Influence treatment in laboratory of stone surface on the surface
[26] R.A. Ketcham, W.D. Carlson, Acquisition, optimization and interpretation of X-
roughness, J. Test. Eval. 37 (5) (2009) 000017.
ray computed tomographic imagery: applications to the geosciences, Comput.
[2] L. Wang, J. Frost David, Quantification of the specific aggregate surface area
Geosci-UK 27 (4) (2001) 381–400, https://doi.org/10.1016/S0098-3004(00)
using X-ray tomography, in: E. Tutumluer;, Y.M. Najjar;, E. Masad (Eds.) 15th
00116-3.
Engineering Mechanics Division Conference, ASCE, New York, United States,
[27] M.A. Taylor, E.J. Garboczi, S.T. Erdogan, D.W. Fowler, Some properties of
2002, pp. 3-17. http://doi.org/10.1061/40709(257)1.
irregular 3-D particles, Powder Technol. 162 (1) (2006) 1–15, https://doi.org/
[3] A.K.H. Kwan, L.G. Li, Combined effects of water film, paste film and mortar film
10.1016/j.powtec.2005.10.013.
thicknesses on fresh properties of concrete, Constr. Build. Mater. 50 (2014)
[28] R. Cepuritis, E.J. Garboczi, C.F. Ferraris, S. Jacobsen, B.E. Sørensen, Measurement
598–608, https://doi.org/10.1016/j.conbuildmat.2013.10.014.
of particle size distribution and specific surface area for crushed concrete
[4] H. Li, H. Zhang, L. Li, Q. Ren, X. Yang, Z. Jiang, Z. Zhang, Utilization of low-
aggregate fines, Adv. Powder Technol. 28 (3) (2017) 706–720, https://doi.org/
quality desulfurized ash from semi-dry flue gas desulfurization by mixing with
10.1016/j.apt.2016.11.018.
hemihydrate gypsum, Fuel 255 (2019) 115783.
[29] W.D. Carlson, T. Rowe, R.A. Ketcham, M.W. Colbert, F. Mees, R. Swennen, M.V.
[5] Q. Ren, M. Xie, X. Zhu, Y. Zhang, Z. Jiang, Role of limestone powder in early-age
Geet, P. Jacobs, Applications of high-resolution X-ray computed tomography in
cement paste considering fineness effects, J. Mater. Civ. Eng. 32 (10) (2020)
petrology, meteoritics and palaeontology, Applications of X-ray Computed
04020289, https://doi.org/10.1061/(ASCE)MT.1943-5533.0003380.
Tomography in the Geosciences, Geological Society of London (2003) 7–22,
[6] C.F. Goble, M.D. Cohen, Influence of aggregate surface area on mechanical
https://doi.org/10.1144/gsl.Sp.2003.215.01.02.
properties of mortar, ACI Mater. J. 96 (6) (1999) 657–662.
[30] P.R. Mouton, Unbiased stereology: A concise guide, The Johns Hopkins
[7] X. Yang, J. Liu, H. Li, Q. Ren, Performance and ITZ of pervious concrete modified
University Press, USA, 2011.
by vinyl acetate and ethylene copolymer dispersible powder, Constr. Build.
[31] S. Brunauer, P.H. Emmett, E. Teller, Adsorption of gases in multimolecular
Mater. 235 (2020) 117532.
layers, J. Am. Chem. Soc. 60 (2) (1938) 309–319, https://doi.org/
[8] K. Wu, H.S. Shi, L.L. Xu, G. Ye, G. De Schutter, Microstructural characterization
10.1021/ja01269a023.
of ITZ in blended cement concretes and its relation to transport properties,
[32] D. Zhang, R. Luo, Modifying the BET model for accurately determining specific
Cem. Concr. Res. 79 (2016) 243–256, https://doi.org/10.1016/j.
surface area and surface energy components of aggregates, Constr. Build.
cemconres.2015.09.018.
Mater. 175 (2018) 653–663, https://doi.org/10.1016/
[9] M. Alexander, S. Mindess, Aggregates in concrete, CRC Press, 2010.
j.conbuildmat.2018.04.215.
[10] Y. Ghasemi, M. Emborg, A. Cwirzen, Exploring the relation between the flow of
[33] Q. Ren, Z. Zeng, Z. Jiang, H. Li, Functionalization of renewable bamboo charcoal
mortar and specific surface area of its constituents, Constr. Build. Mater. 211
to improve indoor environment quality in a sustainable way, J. Clean. Prod.
(2019) 492–501, https://doi.org/10.1016/j.conbuildmat.2019.03.260.
246 (2020) 119028.
[11] Y. Ghasemi, M. Emborg, A. Cwirzen, Estimation of specific surface area of
[34] Q. Ren, Z. Zeng, M. Xie, Z. Jiang, Cement-based composite with humidity
particles based on size distribution curve, Mag. Concr. Res. 70 (10) (2018) 533–
adsorption and formaldehyde removal functions as an indoor wall material,
540, https://doi.org/10.1680/jmacr.17.00045.
Constr. Build. Mater. 247 (2020) 118610.
[12] S.T. Erdogan, Simple estimation of the surface area of irregular 3D particles, J.
[35] T.J. Chotard, M.P. Boncoeur-Martel, A. Smith, J.P. Dupuy, C. Gault, Application
Mater. Civ. Eng. 28 (8) (2016) 10, https://doi.org/10.1061/(asce)mt.1943-
of X-ray computed tomography to characterise the early hydration of calcium
5533.0001575.
aluminate cement, Cem. Concr. Compos. 25 (1) (2003) 145–152, https://doi.
[13] Q. Ren, G. De Schutter, Z. Jiang, Q. Chen, Multi-level diffusion model for
org/10.1016/s0958-9465(01)00063-4.
manufactured sand mortar considering particle shape and limestone powder
[36] E.N. Landis, D.T. Keane, X-ray microtomography, Mater. Charact. 61 (12)
effects, Constr. Build. Mater. 207 (2019) 218–227, https://doi.org/10.1016/
(2010) 1305–1316, https://doi.org/10.1016/j.matchar.2010.09.012.
j.conbuildmat.2019.02.139.
[37] B. Masschaele, M. Dierick, D.V. Loo, M.N. Boone, L. Brabant, E. Pauwels, V.
[14] R.P. Panda, S.S. Das, P.K. Sahoo, An empirical method for estimating surface
Cnudde, L.V. Hoorebeke, HECTOR: A 240kV micro-CT setup optimized for
area of aggregates in hot mix asphalt, J. Traffic Transp. Eng.-Engl. Ed. 3 (2)
research, J. Phys. Conf. Ser. 463 (2013) 012012.
(2016) 127–136, https://doi.org/10.1016/j.jtte.2015.10.007.
[38] J. Vlassenbroeck, M. Dierick, B. Masschaele, V. Cnudde, L. Hoorebeke, P. Jacobs,
[15] J. Anochie-Boateng, J. Komba, E. Tutumluer, 3D laser based measurement of
Software tools for quantification of X-ray microtomography, Nucl. Instrum.
mineral aggregate surface area for south African hot-mix asphalt mixtures,
Methods Phys. Res. Sect. A-Accel. Spectrom. Dect. Assoc. Equip. 580 (1) (2007)
Transportation Research Board 90th Annual Meeting, United States,
442–445, https://doi.org/10.1016/j.nima.2007.05.073.
Washington DC, 2011.
[39] C.S. Smith, L. Guttman, Measurement of internal boundaries in three-
[16] H.F. Taylor, Cement chemistry, Thomas Telford London, 1997.
dimensional structures by random sectioning, J. Met. 5 (1) (1953) 81–87,
[17] T. Wang, T. Ishida, R. Gu, A comparison of the specific surface area of fly ash
https://doi.org/10.1007/bf03397456.
measured by image analysis with conventional methods, Constr. Build. Mater.
[40] U. Gonzales-Barron, F. Butler, A comparison of seven thresholding techniques
190 (2018) 1163–1172, https://doi.org/10.1016/j.conbuildmat.2018.09.131.
with the k-means clustering algorithm for measurement of bread-crumb
[18] P.C. Knodel, V.K. Garga, R. Townsend, D. Hansen, A method for determining the
features by digital image analysis, J. Food Eng. 74 (2) (2006) 268–278, https://
surface area of quarried rocks, Geotech. Test. J. 14 (1) (1991) 35.
doi.org/10.1016/j.jfoodeng.2005.03.007.
[19] L. Xiao, D. Jiang, The study of pervious concrete mix proportion by the method
[41] M. Coster, J.-L. Chermant, Image analysis and mathematical morphology for
of specific surface area of aggregate, IOP Conf. Ser. Mater. Sci. Eng. 231 (2017)
civil engineering materials, Cem. Concr. Compos. 23 (2) (2001) 133–151,
012094.
https://doi.org/10.1016/S0958-9465(00)00058-5.
[20] D. Sun, H. Shi, K. Wu, S. Miramini, B. Li, L. Zhang, Influence of aggregate surface
[42] J. Kittler, J. Illingworth, On threshold selection using clustering criteria, IEEE T.
treatment on corrosion resistance of cement composite under chloride attack,
Syst. Man Cy-S SMC-15 (5) (1985) 652–655.
Constr. Build. Mater. 248 (2020) 118636.
[43] S. Mantellato, M. Palacios, R.J. Flatt, Reliable specific surface area
[21] I. Odler, The BET-specific surface area of hydrated Portland cement and related
measurements on anhydrous cements, Cem. Concr. Res. 67 (2015) 286–291,
materials, Cem. Concr. Res. 33 (12) (2003) 2049–2056, https://doi.org/
https://doi.org/10.1016/j.cemconres.2014.10.009.
10.1016/S0008-8846(03)00225-4.
[44] H.S. Wong, M.K. Head, N.R. Buenfeld, Pore segmentation of cement-based
[22] ISO, 15901, Determination of the specific surface area of solids by gas
materials from backscattered electron images, Cem. Concr. Res. 36 (6) (2006)
adsorption — BET method, International Organization for Standardization
1083–1090, https://doi.org/10.1016/j.cemconres.2005.10.006.
(2010).
[45] N. Otsu, A threshold selection method from gray-level histograms, IEEE T. Syst.
[23] R. Cepuritis, B.J. Wigum, E.J. Garboczi, E. Mørtsell, S. Jacobsen, Filler from
Man Cy-S 9 (1) (1979) 62–66, https://doi.org/10.1109/TSMC.1979.4310076.
crushed aggregate for concrete: Pore structure, specific surface, particle shape
[46] L.-K. Huang, M.-J.-J. Wang, Image thresholding by minimizing the measures of
and size distribution, Cem. Concr. Compos. 54 (2014) 2–16, https://doi.org/
fuzziness, Pattern Recogn. 28 (1) (1995) 41–51, https://doi.org/10.1016/0031-
10.1016/j.cemconcomp.2014.03.010.
3203(94)E0043-K.
[24] E.C. Arvaniti, M.C.G. Juenger, S.A. Bernal, J. Duchesne, L. Courard, S. Leroy, J.L.
[47] J.N. Kapur, P.K. Sahoo, A.K.C. Wong, A new method for gray-level picture
Provis, A. Klemm, N. De Belie, Determination of particle size, surface area, and
thresholding using the entropy of the histogram, Comput.VIsion Graph. Image
shape of supplementary cementitious materials by different techniques, 48
Process. 29 (3) (1985) 273–285, https://doi.org/10.1016/0734-189X(85)
(11) (2015) 3687-3701. http://doi.org/10.1617/s11527-014-0431-3.
90125-2.