Construction and Building Materials 392 (2023) 131925

Contents lists available at ScienceDirect

Construction and Building Materials

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

Hydration and microstructure of cement paste mixed with seawater – An

advanced investigation by SEM-EDS method
Peiran Li a, Wengui Li a, *, Kejin Wang b, Hanbing Zhao a, Surendra P. Shah c
a
School of Civil and Environmental Engineering, University of Technology Sydney, NSW 2007, Australia
b
Department of Civil, Construction and Environmental Engineering, Iowa State University, IA 50011, USA
c
Center for Advanced Construction Materials, Department of Civil Engineering, The University of Texas at Arlington, TX 76019, USA

A R T I C L E I N F O A B S T R A C T

Keywords: This study aimed to investigate the microstructural hydration features of Portland cement incorporating with
Seawater natural seawater (SW). Special attention was paid to characterizing the pore structure, hydrated and anhydrous
Cement phases using microscopy techniques, such as secondary electron (SE), backscattered electron (BSE) imaging and
Hydration
energy dispersive spectrometry (EDS) mapping. Through the utilization of algorithms, information extracted
Characterization
Microstructure
from these methods underwent self-corrected, denoising, and imaging calibration, which significantly contrib­
Quantitative analysis uted to the calculation of pore volume, identification and segmentation of various phases, and quantitative
characterization. By leveraging the chemical and spatial information from EDS mapping, highly detailed dis­
tribution maps of the different cement phases could be obtained, surpassing the conventional BSE grayscale
segmentation method in phase identification. Statistical analysis revealed that the deviation of the phase
quantification was reduced when using a magnification of 250 times. Higher content of hydrate phases with
lower porosity can be observed in SW samples exhibited a higher content of hydrate phases with lower porosity,
indicating a stimulated hydration effect. However, accurately determining hydrate phases with small crystal
sizes, such as ettringite, necessitated an increase in mapping resolution. Therefore, the use EDS mapping for
quantification of various cement phases in submicroscopic and microscopic characterization is both feasible and
promising.

1. Introduction inevitably change, affecting its strength, permeability and durability.

Current research hotspots on SW concrete mainly focus on coping with
Many countries are experiencing freshwater crises due to population the reinforcement’s corrosion risk. The related studies have shown
growth and infrastructure construction aligned with the enormous prospects of SW concrete for components that are not heavily reinforced
freshwater demand. Using seawater (SW) for concrete manufacturing with rebar or coupled with anti-chloride-corrosion materials [9–12].
promisingly provides significant economic and environmental benefits, However, due to the lack of sufficient studies on the phase assemblage
particularly in coastal areas where freshwater is more precious while SW and microstructure development for the seawater-cement system,
is more accessible [1]. However, the chemical ions in SW have caused divergent opinions still exist against using seawater in concrete [13].
significant concerns for being used in concrete. Chloride is the most Putting SW effects aside, for cement-based material itself, the macro
influential ion in seawater on concrete properties. In addition to the performance is hopefully deduced if the microstructure can be well
possible risk of rebar corrosion [2], the excessively stimulated hydration characterized, which may help find effective ways for performance
effect [3,4] brings adverse influences that may limit its practical appli­ improvement. Therefore, the microstructure characterisation may
cation, such as lower workability, intensive heat release and significant contribute to SW concrete’s development.
autogenous shrinkage [5]. Furthermore, some other ions in SW, Many advanced characterization techniques have been used for the
including sulfate, carbonate, sodium and magnesium, can also affect the study of the degree of cement hydration and microstructure. Scanning
cement hydration and phase assemblage (i.e., brucite, calcium oxy­ electron microscopy (SEM) techniques are typical ones used. BSE im­
chloride and M− S− H) [6–8]. The microstructure of concrete will also aging technique is routinely used to identify cement hydration. As a

* Corresponding author at: School of Civil and Environmental Engineering, University of Technology Sydney, NSW 2007, Australia.
E-mail address: [email protected] (W. Li).

https://doi.org/10.1016/j.conbuildmat.2023.131925
Received 3 March 2023; Received in revised form 17 May 2023; Accepted 22 May 2023
Available online 31 May 2023
0950-0618/© 2023 Elsevier Ltd. All rights reserved.
P. Li et al. Construction and Building Materials 392 (2023) 131925

Table 1
Chemical composition of the OPC (oxide composites).
Chemical CaO SiO2 Al2O3 Fe2O3 MgO SO3 Na2O K2O CO2 LOI*

Component (wt.%) 62.9 20.3 4.5 4.6 1.2 2.6 0.3 0.3 2.0 4.0

Note: LOI* presents loss on ignition.

hypermaps recording chemical and spatial information, which can be

Table 2
used to study hydration [15] and durability issues [16] from a micro
Mineral composition of OPC and its physical properties.
perspective. However, it is not easy to interpret the extracted large-scale
Minerals C3S C2S C3A C4AF Limestone Gypsum information, much less the long time required for processing high-
Content (wt.%) 49.03 21.21 4.14 13.00 3.32 4.55 accuracy maps, which are severe limitations for SEM-EDS analysis. In
Note: Specific gravity is 2800–3200; bulk density is 1200–1600 kg/m3 and
the current research, a Python [17] code framework developed by
average particle size (d50) is 8 μm. Georget et al. [18] enabled the mapping data to be visually processed. It
provides interfaces adapting to the optimizing clustering algorithms for
phase identification [19,20]. The framework significantly simplifies the
Table 3 process for increasing the signal-to-noise ratio and additional analysis.
Major chemical composition of SW. Based on the background above, this study aims to characterise the
Dissolved ions Na+ K+ Ca2+ Mg2+ Cl– SO2–
4
microstructure of cement paste with deionized water (DW) and SW.
Impacts caused by SW will be determined from the nano and micro-scale
Concentration (mol/l) 0.47 0.01 0.01 0.05 0.54 0.03
perspectives [60–66]. The characterization combines the latest BSE and
SEM-EDS mapping techniques for obtaining two-dimensional digital
function of element composition, the grey level from BSE images can be images of Portland cement paste in which all major phases and pores are
attributed to different phases, separated and quantified, thus deter­ identified. Numerical treatments of maps are conducted to allow intui­
mining the hydration degree of cement clinkers [14]. Another reliable tive visual identification. Statistical results from different mapping
method, X-ray imaging, acquires the full-chosen range of elemental techniques are compared and discussed. Overall, this paper not only
composition as hyperspectral mappings, which can be processed to characterizes the effect of SW on the microstructure of cement but also
discriminate and quantify phases. A particular use of SEM-EDS provides evaluates the prospect of EDS mapping methods for cement phase

Fig. 1. The elemental atomic ratio mapping result of 12-hr cement paste with DW (data acquired with 40 % deadtime, quantified at resolutions of the same ac­
quired hypermap).

Fig. 2. The elemental atomic ratio mapping result of 12-hour cement paste with DW (data acquired with 7 % deadtime, quantified at resolutions of the same ac­
quired hypermap).

2
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 3. Flowchart of the image and data treatment process corresponding to sections.

identification and quantification. moulds of 50 × 100 mm were used for casting. The cast paste samples
were sealed and placed at 20 ℃ for 12 hr, 7 days, 28 days, and 63 days
2. Experimental program until testing. Two drying methods were applied afterwards, which
depended on the curing time of the samples. For the 12-hour cement
2.1. Raw materials paste samples, a vacuum freeze dryer was used to prevent brittle fracture
of the early-stage samples. Before drying, samples were crushed into 2 –
Ordinary Portland cement (OPC) used in the study was measured by 3 mm particles. The drying process lasted 2 days at − 55 ℃ in vacuumed
X-ray fluorescence (XRF), and the results are summarized in Table 1. The condition to remove the free water. For the other longer-cured samples,
mineral compositions of OPC were calculated using Bogue calculation particles with a diameter of about 2 mm were collected from the crushed
[21], and the results are shown in Table 2. Two types of mixing water: cement paste. Isopropanol was used as the solvent exchange to dry the
(1) DW; and (2) natural SW, were applied. The natural SW was obtained particles for 7 days. During the drying, the isopropanol was renewed at 1
from Congwong Beach on the Sydney coast of Australia. Table 3 shows hr, 1 day and 3 days. The dried particles were transferred into a vacuum
the chemical composition of SW determined by inductively coupled desiccator with CO2 and H2O absorbing agents to allow the residual
plasma mass spectrometry (ICP-MS). isopropanol to evaporate. In these two drying ways, the chemically
bound water in hydrate phases was preserved as possible, thus keeping
2.2. Sample preparation the hydrates unaltered [22]. However, it should be noted that the freeze-
drying method was not the condition encountered in conventional
Cement paste mixing was performed with DW or SW at a w/b ratio of cement hydration, especially at the stage of initial structure formation,
0.5 using a Hobart mixer according to ASTM C305. Cylindrical plastic and the pore structure in freeing-drying samples may be changed.

3
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 4. A typical BSE image treatment for region segmentation in S63d_1.

Samples were tested within 1 day after the drying process to ensure
Table 4 the reliability of experimental results. To inspect the morphology of the
Threshold values for hydrate phases segmentation of cement paste. hydrate phases, the dried samples were broken to reveal a fresh surface,
Cement paste Pore Hydrated phase Unhydrated phase cleaned with nitrogen to clean the investigated surface, and then coated
D12h_1 0–53 54–140 141–255 with Au/Pt alloy (80/20) with a thickness of 5 nm. Carbon Conductive
D12h_2 0–48 49–145 146–255 tape was also used to bond the sample to the pin stubs, to improve
S12h_1 0–117 118–192 193–255 conductivity and reduce the charge-up effect. As for further quantifi­
S12h_2 0–106 107–182 183–255
cation of hydrate phases, preparing a smooth surface was significant to
D7d_1 0–125 126–200 201–255
D7d_2 0–109 110–186 187–255 acquire reliable experimental results on delicate microstructure. The
S7d_1 0–131 132–205 206–255 dried samples were crushed to collect 2 mm slices. To obtain a flat
S7d_2 0–118 119–197 198–255 surface, silicon carbide paper #1200 was used for further polishing.
D28d_1 0–107 108–207 208–255 Resin mounting was conducted using Leco 812-522TM epoxy, followed
D28d_2 0–108 109–182 183–255
S28d_1 0–124 128–187 188–255
by vacuum impregnation. The embedded samples were then polished by
S28d_2 0–146 147–204 205–255 an SS-1000 polishing machine with polishing cloth leclothTM discs.
D63d_1 0–119 120–191 192–255 Water-free diamond polishing suspensions were used as abrasives, with
D63d_2 0–130 131–192 193–255 diamond lapping oil as a lubricant to prevent further hydration during
S63d_1 0–130 131–212 213–255
the polishing process. An ideal mirror-like surface can be attained by a
S63d_2 0–123 124–216 217–255
20-minute polishing under a pressure of 15 N with a 9 µm diamond
Nomenclature: using “D28d_1” as an example, D denotes deionized water, 28d suspension, followed by a 30-minute polishing at 20 N with a 3 µm
denotes curing for 28 days, and 1 represents the sample’s number. diamond suspension and a 1-hour-polishing at 20 N with a 1 µm dia­
mond suspension.
During the polishing process, the lapping oil was added continuously

4
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 5. Determination of threshold values for testing samples at different curing ages.

to cool down the surface’s temperature, protecting AFt from thermal investigated area, which can quantify various phases based on their
decomposition. Isopropanol ultrasonic bath was used to clean the sam­ chemical composition.
ples before changing the polishing abrasives. The samples were coated It takes nearly 1 hr to obtain a high-quality EDS mapping database
with carbon with a thickness of 15 nm to avoid the interference effect of (hypermap). Therefore, during the long scanning time, the conductivity
Au to EDS analysis after dried in the vacuum desiccator. of the samples was critical to avoid image shifting due to the beam
drifting effect. In addition, the EDS mapping scanning process syn­
2.3. Test methods chronizes with BSE imaging to ensure the two imaging areas are
consistent. 15 kV was applied as the accelerating voltage to access K
The morphology analysis of hydrate phases at high magnification lines of Fe. Whereas clear BSE images with high contrast gradients can
was conducted with a Field emission scanning electron microscope be obtained from a scanning beam current of 2nA, the deadtime value
(FESEM) of Zeiss SUPRA 55-VP equipped with an Energy Dispersive X- should be carefully considered. As proposed in the existing literature on
ray Spectrometer (EDS) of Bruker Quantax 400. EDS point-scan analysis quantitative EDS analysis, the deadtime was 30–35 % from the study of
was performed to identify hydrate phases. The microscope hydrate Georget [18], who used Aztec software to acquire and quantify hyper­
phases were expected to be quantitatively characterized by analyzing maps. However, as proposed by Newbury, the deadtime was recom­
polished samples through Zeiss EVO SEM with EDS. The quantification mended to be lower than 10 % to minimize pulse coincidence (sum
analysis of pores (500 nm–100 µm) applied statistical map image anal­ peaks) [23]. By using Esprit software to acquire and process EDS
ysis, combining SE and BSE imaging. The grey-level statistics analysis of hypermaps in this study, if the deadtime was over 40 % or more, the
BSE images was also conducted to roughly separate the region of hy­ input counts per second are too high to be handled optimally by the
drated and unhydrated phases. Furthermore, EDS quantitative mapping current amp time and may lead to excessive summed peaks, and causing
technology was applied to obtain data on the atomic proportion of the high dispersion after when processing hypermaps, as indicated in Fig. 1.

5
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 6. Delineation of pore boundary processed by SE image.

Fig. 7. Correction of pore area by combining SE and BSE image.

The significant discrepancy may be due to the different software used. nearly 40 min to acquire a quantified hypermap.
Therefore, based on the performance of EDS detection equipment, the The quantitative data about the sample composition is derived from
optimum deadtime was 5 %− 9 %, and the beam current was adjusted the different peak intensities by an extensive mathematical process. In
from 300 to 600 A, for the different types of polished samples this study, the standard peak-to-background ZAF correction (P/B ZAF)
investigated. was used as the self-calibrating method. The data of standards for all
The EDS hypermaps were acquired over regions of 440 µm by 330 µm elements of interest in this study was from the built-in standards library:
at a resolution of 1024 × 768 pixels. The dwelling time (the time the wollastonite for Ca, corundum for Al, anhydrite for S, quartz for Si, NaCl
electron beam stayed on a pixel while acquiring hypermaps) was 256 µs, for Cl, hematite for Fe and MgO for Mg. Due to the limited standards
with the line average (controls the number of times a horizontal line was database, the quantification of Na and K was performed by the stan­
scanned before proceeding to the following line) of 10, wherein it took dardless P/B ZAF method, which can also provide reliable and adequate

6
P. Li et al. Construction and Building Materials 392 (2023) 131925

phases and pores are identified. A sequential approach to the study is

presented in Fig. 3, which will discuss the following issues in separate
sections: 1) Region segmentation of BSE image and its defects; 2)
Correction of pore area by combining SE and BSE images; 3) Processing
EDS mapping data for phase identification and regional quantification;
and 4) Obtaining two-dimensional digital images for microscopic
characterization.

3.1. Greyscale segmentation

Fig. 4 illustrates typical greyscale segmentation results from a BSE

image of the 63-day hydrated OPC paste (the sample S63d_1) with SW,
as an example in the following image and data treatment. The decon­
volution procedure of the histogram was performed by Peakfit software,
using a Gaussian response function with a Fourier deconvolution algo­
rithm. Three Gaussian curves can be revealed in Fig. 4(b), which
represent a local maximum. The first Gaussian curve is primarily
assigned to the C-S-H phases, and the second Gaussian curve can be
classified as CH. Since the intersection of the two Gaussian curves was
close to the peak, segmenting the two phases directly by greyscale
discrepancy was not rigorous to some extent; the greyscale of CH and
inner C-S-H phases were hard to distinguish, especially at the late stage.
Therefore, Gauss 1 and Gauss 2 were counted jointly as hydrated phases.
The third Gaussian curve can be attributed to unhydrated phases.
The intersection between Gauss 2 and Gauss 3 was set as the threshold to
divide hydrated and unhydrated phases. It should be noted that no
sharpened peak occurred in the histogram because the BSE imaging can
still detect partly phases inside the larger pores. Therefore, the tangent-
slop method [14,24] was used as the strategy to determine the threshold
to divide the C-S-H phase and pore. Proper area segmentation with other
strategies, such as the “overflow” method [25], was also reasonable in
common. However, it needs to be emphasized that the pore determi­
nation only by the greyscale of BSE was not wholly reliable. Joint
strategies would be used to correct the pore phase determination, which
was illustrated in the subsequent section. Here, the greyscale values of
123 and 216 were adopted as the threshold for the pores, hydrated, and
unhydrated phases of the particular BSE image.
The Polychromatic BSE images can be obtained based on the grey­
scale segmentation above in Fig. 4(c). Table 4 lists the determined
threshold of samples in this study, wherein slight differences in
threshold value can be observed even for the samples from the same
cement paste. This is because the current intensity was adjusted to
Fig. 8. Effect of the denoising algorithm on the Ca map and representative
obtain qualified EDS hypermaps with enough counts per second,
points in the ratio plot.
resulting in a brightness change in the BSE image. Due to the various
hydrated time, the current intensity of microscopy needed to be adjusted
element results for non-light elements. In addition, the stoichiometric of depending on the imaging quality and deadtime for elemental mapping.
O was alternatively carried out during elements processing. Although it Thus, there was an unvoided deviation of brightness and contrast of
may not be accurately quantified, it would correct matrix errors effec­ these images, and threshold values should be determined independently
tively by processing the oxides of the dominant [18]. for each sample. More details about threshold determination for the
The resolution of output quantitative processing results also needed samples were plotted in Fig. 5. Even though the phase segmentation
to be adjusted; the long processing time severely limited the efficiency of might be arbitrary to some extent, it can ensure the reliability of
determination. If the resolution remained unchanged (1024 × 768 revealing regularity by using the identical processing strategy.
pixels) or increased, quantifying a map would typically take 10 h to
more than one day but obtain high-quality maps. A finer quantifying 3.2. Pore determination
calibration could significantly reduce data dispersion. As shown in
Fig. 2, the higher the resolution outputted, the higher the quality of the Following the discussion above, it is defective for only using BSE
results. A compromise of one-second resolution of the acquired maps greyscale histogram to determine pore area. A slight change in pore-
was adopted, which could produce a satisfactory result in an acceptable related threshold can significantly increase or decrease the accounted
one-hour time. In this way, a quantified elemental map was obtained in area, mainly because some hydrate phase detected in the pore was not
about two hours. counted as pore area. Therefore, a method for correcting the pore size
was applied by combining SE images, in which the area of large pores
3. Image and data treatment was easy to determine. The charge-up effect was undesirable in SEM
imaging. However, it can well depict the boundary of the pores. The
The characterization combines the latest image and data processing binarization of the pores and other phases can be obtained by adjusting
technologies, SEM imaging, and EDS mapping methods for obtaining the threshold, as shown in Fig. 6(a) and (b). It must be emphasized that
two-dimensional digital images of cement paste in which all major the charge-up area as the white area in Fig. 6(b) was larger than the real

7
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 9. Components of a composite image.

Fig. 11. Atomic ratio (Si/Ca to Al/Ca) of representative points (in 20,000
fragments) from EDS hypermaps (The pure phases refer to the corresponding
stoichiometries proposed by Taylor [38]).

boundary of the pores. Therefore, a visual adjustment was required to

shrink the white area (generally two pixels inward) after filling the in­
ternal area (for ImageJ software, the Fill Holes and Erode function may
help as Fig. 6(c)), and the outline of pores based on SE image can be
obtained eventually in Fig. 6(d).
Notably, the pore determined should consist of hollow reaction
shells, cracks, and capillary pores [25]. By narrowing treatment, the
finer pores about two or fewer pixels (less than 1 µm in diameter) might
be filtered out inevitably, hence the need for a merged pore boundary.
According to the threshold determined from the BSE image above, the
outlines of finer pores can be well preserved, as depicted in Fig. 7(a) and
(b). The merged pore boundary in Fig. 7(c) can be obtained by
Fig. 10. Composite image generated from the translucent RGB layer over grey-
scale BSE in Fig. 9 and 9(b) Region segmented image using SLIC algorithm. combining the superiority of two pore-boundary determination

8
P. Li et al. Construction and Building Materials 392 (2023) 131925

An ideal denoising result should remove hot pixels while preserving

the sharp edges [18]. Otherwise, the fuzzy edges of elemental maps
would cause mutual encroachment in adjacent areas, distorting the
original volume. The left column of Fig. 8(b) to (d) displayed the pro­
cessed Ca map by three denoising algorithms: the total variation algo­
rithm [30], the Gaussian filter [31], and the mean filter. The other maps
(such as Si and Al) were denoised using consistent algorithms, plotting
the element ratio on the right column. The denoising effect of each al­
gorithm was controlled by its parameter. The mean filter depends on a
radius parameter (the value here was 10.0); the Gaussian filter depends
on a scale parameter (the value here was 0.5); The total variation al­
gorithm relies on a weight parameter (the value here was 0.1). For each
denoising algorithm, the optimum value of the parameter was not a
constant (the larger the parameter value was, the further blurred the
map was), and here it can be used as a reference for a reasonable range.
In addition, adjusting the parameter value cannot optimize the algo­
rithm, and thus the denoising algorithm selection should be prioritized.
By comparing the deionizing methods, a better effect can be obtained by
using the total variation algorithm, even converging the dispersed
scatters into two strips. The mean filter can obtain a similar result, but
Fig. 12. Definition of representative point clouds for phase separation.
the edge in the map was significantly blurred. As for the Gaussian filter,
a compromise should be reached between the texture reduction and the
fuzzing edge by adjusting the parameter, which limits the denoise
Table 5
Strategy for separating phases in cement pastes. extent. Based on the Gaussian function, a Joint bilateral filter [32–34]
was developed to be the non-linear, edge-preserving denoising algo­
Phase Si/Ca Al/Ca Others
rithm for images. It was also a competitive method to reduce the need for
C3S 0.33 ± ε <0.06 — compromise while becoming more sensitive to parameters [18].
C2S 0.5 ± ε <0.06 —
Denoising is not an objective process-it inevitably leads to participant
CxAF <0.2 >0.25–0.5 —
AFm <0.2 <0.25–0.5 S/Ca, Cl/Ca bias. Nonetheless, the information distributed over the many composi­
CH/CC <0.2 <0.1 — tion ratios was still quantitative on average [35]. Because of the best
C-S-H matrix Variable Variable — denoising effect achieved from the total variation algorithm, it was used
Quartz >1 <0.1 High value in Si to process the elemental maps.
Gypsum <0.2 <0.1 S/Ca
Kaolinite >0.8 >0.8 Si/Ca = Al/Ca
3.3.2. Merging composite maps
Note: ε represents a tolerance dependent on the distribution of representative The main elements of interest in cement composition are Al, Si and
points in the atomic ratio plot. Ca, by which most of the phases can be identified. A composite map with
the three elemental distributions can be obtained by merging the
methods, resulting in higher reliability in characterizing pores by image elemental maps as the three channels of an RGB image: Si map for the
analysis. The corrected pore distribution is depicted in Fig. 7(d), which red channel, Al map for the green channel, and Ca map for the blue
can be further investigated for quantification. channel. To enhance the definition of the grains’ outlines in the image,
the corresponding grey-scale BSE map can be used as the background
3.3. Phase quantification with a translucent layer of the composite map covered. Fig. 9 shows the
four components of a composite image (a greyscale BSE image and three
Since the hydrated and unhydrated phases can be roughly deter­ quantified element maps). The composite image is displayed in Fig. 10
mined by image analysis, it should be worth considering the identifi­ (a). Specific chemical compositions of phases were displayed as colours
cation and quantification of various phases. A new framework with the corresponding RGB values; thus, they can be identified. The
developed by Georget et al. [18], edxia, was used to identify phases by background of the legend was set as the average colour of outer C-S-H
the ratio plots. This program was coded in Python [17], using developed for a better distinction by visual intuition.
libraries of scipy [26,27], matplotlib [26] and scikit-image [28]. Based The composite map was suitable for fast phase identification, which
on its open-sourced framework, visual operating interface and linked was adequate for those phases with significant differences in chemical
maps with ratio plots, the edxia method significantly simplified the data composition, for example, the CH and AFm phases. However, when it is
processing. necessary to consider the other elements (such as sulphate, chloride and
magnesium), the composite map cannot merge another component un­
3.3.1. Denoizing of elemental maps less replacing the original channel or displayed as a grey channel. In
Due to the limited scanning time for the EDS detector, the number of addition, the threshold in the composite map was offered by the hue of
characteristic x-ray counts collected per pixel is usually lower than the three quantized elements, not the chemical composition; For example, it
threshold needed for dictating a high accuracy [29], which causes maps was intractable to distinguish the AFm phases and AFt. Therefore, phase
to be noisy. The noise was featured with the brightness variations as the separation based on chemical composition was necessary to go future.
densely dotted textures on the elemental maps. For instance, a calcium
map with noise texture can be observed in raw EDS data displayed in 3.3.3. Regions segmentation in image
Fig. 8(a). The noise texture significantly caused a higher spatial The segmentation can also apply to BSE images. However, without
dispersion in the pixels’ element ratio distribution and brought incon­ information on chemical compositions, it typically failed to distinguish
venience to the later processing while identifying various phases. Thus, hydrate phases through the region segmentation based on only greyscale
image denoising was necessary. Although it may cause a few pixels to be distribution. Therefore, region segmentation by chemical composition
unreliable, most can still provide competent information to distinguish was more beneficial for hydrate phase identification. A simple linear
more representative areas readily. iterative clustering (SLIC) algorithm was used to perform segmentation

9
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 13. Boundary definition for specific phases.

Fig. 14. Identified phases by stoichiometries as overlays of the phase masks on Fig. 15. Composite two-dimensional digital image by ternary microstructure
the BSE image. characterizing results.

[36], which adapted a k-mean clustering approach [37] to generate 20,000 fragments is recommended [18]. The geometric centre of these
superpixels (a group of adjacent pixels that share common characteris­ segmented regions would be used as the representative points with the
tics). Here in EDS’s hypermaps, the similar characteristics were similar chemical information to be analyzed.
chemical compositions. By giving similar pixels the same colour
belonging to a superpixel, it would provide perceptual meaning to 3.3.4. Identification and division of mixtures
facilitate phase identification. The segmentation result is displayed in For identifying cementitious phases, the atomic ratio was commonly
Fig. 10(b), which constituted 5000 chromatic fragments. The dots were adopted [35]. A dataset of Al/Ca to Si/Ca plots can be obtained in Fig. 11
the geometrical centre of each fragment. The main details were retained based on the atomic ratio of the representative points. The circle regions
compared with the composite image in Fig. 10(a). The segmentation was representing some pure phases were marked corresponding to stoichi­
unnecessary while significantly reducing the data processing time and ometries from Taylor [38]. It can be observed that most of the points
increasing the signal-to-noise ratio. Meanwhile, the loss of image in­ were not located ideally in the regions; they were distributed as mixtures
formation was minimized by using the SLIC algorithm. In a full scan instead of pure phases. The reason may result from the limitation of
image (with 1024 × 768 = 786, 432 pixels), a segmentation in 5000 to scanning resolution. The C-S-H matrix contains various small particles,

10
P. Li et al. Construction and Building Materials 392 (2023) 131925

corresponding to the C-S-H matrix was depicted by the atomic ratio

points from the pure C-S-H gels and the sub-resolution phases. The
“impurity effect” occurred in other identifiable hydrate phases (such as
Layered double hydroxide (LDH) particles of AFms and other solid so­
lutions), resulting in the mixture lines between the C-S-H matrix and the
pure phases. In the Glue interface, the region of interest (ROI) can be
separated by setting the thresholds along these lines, which correspond
to certain phases by their stoichiometries. Conversely, the ROI can be
corrected graphically using the grey-scale BSE image. Finally, after
multiple adjustments, a finer ROI can be obtained through the manual
selection approach.

3.3.5. Phase separation

According to the above approach, Fig. 12 shows the boundary defi­
nition of point clouds and the corresponding false-colour ROI layers
displayed on the BSE image. Based on the atomic ratio, the strategy for
separating the main phases is listed in Table 5. A tolerance ε can be set
up out of the scope of the stoichiometry of ideal pure phases dependent
on the ratio distribution. The regions of anhydrous particles in the grey-
scale BSE image can refine the typical value of ε. Because the compo­
sition of the cement and the sub-resolution phases mixed in the matrix
may be variable, the boundary of the C-S-H matrix should be determined
according to the specific situation rather than a fixed scope.
Compared to the outer C-S-H (the phases formed between clinker
grains), the composition of the inner C-S-H (the phases formed in clinker
particles) was more homogeneous due to the fewer other phases glued.
The related point cloud was adjacent more to the boundary of C2S. It
seems reasonable because the inner C-S-H was more likely to form after
the surface of anhydrous particles was completely covered by hydrated
phases [39]. Therefore more inner C-S-H was formed around C2S due to
its slower hydration rate. The definition of these boundaries can also, or
perhaps should, refer to the BSE image. Thus, the scope listed in Table 5
should not be taken as a strict guideline, but it can be considered an
empirical rule to mark the phases readily.
For the threshold definition of specific aluminium-bearing phases,
such as AFt and Friedel’s salt (FS), the Si/Ca vs. Al/Ca plot was insuf­
ficient to determine the boundaries; it should correspond to the ratio of
sulfate and chloride. The typical crystal size of AFt was approximately
100 nm (refer to Fig. 16 and Appendix A), which was significantly lower
than the mapping resolution (430 nm for one pixel). Therefore, it was
not strange to hardly find a representative point in the region of pure AFt
phases, as shown in Fig. 13(a). Enhancing the resolution of EDS
hypermaps was surely an effective measure to identify the sub-
resolution phases. However, it can still be detected with the aid of a
S/Ca vs. Al/Ca ratio plot. Based on the stoichiometry of AFt, the region
rich in AFt of the C-S-H matrix would lead to a higher S/Ca ratio. Some
representative points within the stochiometric range, thus, can be
assigned as the geometric centres of segmented regions representing Aft
phases. It would be, to some degree, disputed in phase identification.
One is that, for example, in the regions of AFm phases, some points with
a lower sulfate ratio might be caused by a small amount of AFt, not the
AFms. Second, as shown in Fig. 13(b), it can be observed that most of the
selected AFt points were distributed in the regions of the C-S-H matrix,
indicating that the sub-resolution AFt phase was detected integrally
mixed with C-S-H. It seems hasty to assign the whole region as the AFt
phases. Therefore, a compromise strategy was needed. For instance, the
Fig. 16. EDS identification of OPC paste with SW at 28 days. S ratio of C-S-H close to AFt phases was not assigned to the C-S-H
matrix—assigning the AFt phase was a higher priority. Similar strategies
such as anhydrous grains and other hydrate phases. These fine particles were adopted to separate the C-S-H matrix with other detectable sub-
were even smaller than the interaction volume of electron beams (the resolution phases. The compromise strategy was not rigorous for
size of one pixel) that cannot be identified to a specific phase and, thus, quantifying these sub-resolution phases, but conducting qualitative
would be considered “impurities” in the matrix. comparison was adequate if the points assignment was performed ac­
The “impurity effect” inevitably caused a dispersion of point clouds cording to the unified compromise strategy. SW increased the chloride
in the result from the EDS mapping on the matrix unless a large amount ratio in the cement matrix. The consistent compromise strategy allows
of pure phase formed within a specific volume. Therefore, the region representative points for FS and Kuzel’s salt (KS) to be selected by Cl/Ca
vs. S/Ca plot outside the C-S-H matrix in Fig. 13(c). Similarly, these

11
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. 17. Statistical results of various phases (in volume) by the two methods of microstructure characterization.

points were not considered pure phases; they represented AFms with a Although it was time-consuming research, it was an exploratory attempt
higher chloride ratio on average. The Cl/Ca vs. S/Ca plot also provided for future research.
valuable information. For example, it can be observed that most of the
AFm phase identified contains lower S and Cl, indicating the formation 4. Experiment results
of carbo-AFm phases.
Fig. 14 displays the separated regions corresponding to various 4.1. Morphology of hydrate phases
phases. The inner C-S-H phase was found surrounding the C2S and C3S
particles. Some of them were observed in the shape of particles formed Fig. 16 displays the SE micrographs of the cement paste after 28-day
by the completely hydrated C3S and C2S phases. Combining with the BSE hydration with SW. Specific phases in the images can be identified by
image, most CH clusters were found in the pores, which can be distin­ their typical micro morphologies and using the information from EDS
guished from CC particles with smooth surfaces and neat edges. A group points measurements. For instance, some crystals with high purity can
of small AFm particles precipitated adjacent to the anhydrous CxAF be determined, such as CH as hexagonal platelet morphology (Plot 4)
phase or scattered around, formed in the space left from clinker disso­ and AFt as hexagonal and prismatic crystals (Plot 10). The growing CH
lution. Some of them can be identified as FS and AFt. The outer C-S-H as crystal thickened as a function of time (Plot 15). Although sealed curing
the matrix glued various phases as a composite mixture. was conducted to insulate CO2 exposure, the carbonates can still be
observed due to the self-contained CC in cement. Since the chloride was
3.4. Quantitative data analysis more competitive than carbonate to bind AFm phases, the initially dis­
solved carbonate would carbonate the CH phase eventually. Early signs
Based on the image treatments mentioned above, a composite image of nucleation of carbonized CH [40] can be observed in Plot 1; the
can be obtained to display the detailed distribution of pores and phases crystal nuclei were evenly distributed on the CH matrix. The elongated
in Fig. 15. However, extracting quantitative data from only one image clusters morphology can be observed from the growing carbonate cal­
may cause a great error due to contingency; a very large pore or anhy­ cium in Plot 2. The magnesium from SW can be precipitated as the
drous grain would result in a remarkable difference in value. Reducing M− S− H hydrate phase [41–43] mixed with C-S-H phases (Plot 3) and
the scanning magnification can decrease the error effectively, but more Mg-Al hydrotalcite-like phase [44] (Plot 5 and 6). As the main hydrate
sub-resolution phases became hard to be identified. Geogret et al. [18] phase, C-S-H (Plot 7) can link other ions, such as aluminate or alkalis,
proposed a feasible method for images at high magnification-to due to its poorly crystalline structure [45,46].
renormalize the big particles according to the normalized result from Cement with a higher content of aluminate may lead to the alumi­
remaining regions. In this study, images with a 1.57 times larger scan­ nate incorporation by C-S-H to form C-A-S-H and change the chain
ning area were used for quantitative analysis. In addition, more maps for length [47–49]. The phase can be determined on Plot 13 with a looser
one sample were used for statistical analysis (listed in Appendix B). surface than C-S-H. In Plot 8 and 9, the AFt phase was converted to carbo

12
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. A1. FESEM images of OPC paste at various curing ages.

AFm phases. The higher chloride or sulfate content in some AFm phases these phases had no remarkable differences in morphology, and no
indicated the formation of FS and sulfo-AFm phases, as shown in Plots obvious boundary can be determined. This was analogous to a sponge
11, 12, and 14. The coexisted AFm phases indicated that the sulfate, with absorbed fresh water on one side and salt water on the other.
carbonate (mainly from gypsum and CC in cement) and chloride (mainly Segmentation of area could only be performed by ion concentration.
from SW) compteted for the formation of AFms. Like solid solution, Therefore, it confirms the feasibility of characterizing cement phases by

13
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B1. Image and data treatment results for D12h_1.

Fig. B2. Image and data treatment results for D12h_2.

element ratio from EDS approach. by accelerating the hydration, noticeable differences in phase
In terms of the evolution of morphology, more details are listed in morphology cannot be observed at the late stage. However, issues
Appendix A. Although SW boosts the earlier microstructure formation should be considered regarding its stability. Since the microstructure of

14
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B3. Image and data treatment results for S12H_1.

Fig. B4. Image and data treatment results for S12H_2.

SW samples was developed in a stable condition with various ions results of the areas corresponding to various phases were presented in
involved, the main hydrate phase’s volume, C-S-H, inevitably changed Fig. 17 through BSE greyscale segmentation method and EDS mapping
from element incorporation [49] or increased chemical shrinkage [50]. method. The standard deviation of the data set extracted from the
Once the volume change is reversible, The stability of the structure may composite images was calculated and displayed as the error bar on the
be sensitive to the long-term service environment. top of the columns. The short error bars obtained show a higher con­
centration of values, indicating good reliability of the average results. If
needed, the volume content of phases can also convert to the mass
4.2. Statistical result of microstructure characterization
content based on the density of identified cement phases [51,52].
By comparison of the two methods, the total volume of anhydrous
The volume of all identified phases and pores can be counted from
phases obtained by the EDS method was 30–50 % lower than those from
the ternary composite images obtained by image treatment. Statistic

15
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B5. Image and data treatment results for D7d_1.

Fig. B6. Image and data treatment results for D7d_2.

the BSE method. It was because the EDS method was more sensitive than hydrated cement with SW. It was because using the EDS method can
the BSE method to identify whether the clinker was hydrated or not. This only define the phase with the chloride ratio higher enough to the
difference was significant at the early stage when more remained stoichiometry of FS. For those mixed with other phases, the chloride
anhydrous phase to magnify the effect. content was diluted and thus had to be defined as other AFms. Increasing
As for the sub-resolution phases, it can still be concluded that more the magnification might be an effective way to separate these phases.
AFt formed in samples with SW, although the AFt cannot be accurately Compared to the long-term cured samples, an interesting result can
quantified at this magnification. Freidel’s salt can be observed from the be observed that the 12 h cured sample has the lowest porosity, which
early stage of hydration, but the content hardly increased in the seems illogical. Normally, the pores were from the space left from
following stages. Furthermore, it was found to coexist with other AFm clinker dissolution after the adjacent cement matrix was formed and
phases (Sulfo-AFm and Carbo-AFm, as detected in Fig. 16). However, hardened. The porosity in 12 h-cured DW samples was accordingly
other AFms were not detected in the typical phase identification method lower than the SW samples, considering SW can stimulate cement hy­
XRD [3,53–55] or predicted by thermodynamic modelling [56] from the dration, and the hydration degree of SW samples was higher than that of

16
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B7. Image and data treatment results for S7d_1.

Fig. B8. Image and data treatment results for S7d_2.

DW samples at the same curing age. As the hydration continues, the samples was decreased at the same time. It indicated more AFt could be
formed hydrates begin to fill the pores gradually. Therefore, the porosity retained at the late stage, being affected by ions from SW. By comparing
of cement paste seems to start from low to high and then to low the results of the two characterizing methods, EDS identifying method
throughout the whole hydration process. It can also explain the signif­ can perform a detailed phase separation. Increasing the scanning
icant difference in porosity between the two samples after 12 h and 7 magnification would be conducive to quantifying small-size phases,
days of curing. while the time–cost factor should also be considered.
At the late stage, the porosity of the cement matrix in the SW samples
was comparable to, or lower than, that in the DW samples. It might be 5. Discussions
due to the higher content of AFt or AFms directly shown in the figure
(sub-resolution AFt phases were not fully counted), filling the pores. At 5.1. Characterization methods of pore
the late stage of 63 days, the space left from AFt conversion caused a
slight increase of porosity in DW samples. However, the porosity of SW MIP is routinely used for porosity characterization. Previous studies

17
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B9. Image and data treatment results for D28_1.

Fig. B10. Image and data treatment results for D28_2.

[57] found that the MIP samples required a certain hardness and 5.2. Characterization methods of phases
quantity to obtain accurate results. The image analyzing method, which
directly extracts data with various strategies [14,25,58] on BSE grey­ The TG was a quantitative method with high accuracy for cement
scale segmentation, has been proposed as a feasible alternative. How­ phases. However, since the inevitable overlapping of dehydration tem­
ever, the study found that the segmented area was sensitive to the perature, the reliable quantification phases were limited (typically CH
threshold value. A sudden change can be observed when the solid phase and CC [3]). Rietveld quantitative XRD was widely accepted to char­
inside the pore was selected. Combining SE and BSE images can accu­ acterize low-level phases by calibration and determine peak broadening
rately profile the pore area, which benefited from the two imaging [38]. The accuracy of quantification depends highly on the well-defined
mechanisms. The samples do not require hardness but polishing and content of phases from standard reference. However, the amorphous
conductivity, which can ensure high quality and avoid beam drift. phases cannot be directly determined due to the lack of reference. Other
quantitative methods are required to help verify the accuracy of Rietveld
refinement, or else the quantitative reliability will be significantly

18
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B11. Image and data treatment results for S28d_1.

Fig. B12. Image and data treatment results for S28d_2.

19
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B13. Image and data treatment results for D63d_1.

Fig. B14. Image and data treatment results for D63d_2.

reduced [52]. abundant but, unfortunately, it was hard to tap the value efficiently
Image analysis from BSE-EDS hypermaps has been proposed as a without a programming background. The framework developed by
suitable method with a reasonable degree of precision to characterize Geogret et al. [18] provided an easy-to-use interpreter with a graphical
cement phases [15,59]. The information from these hypermaps is interface, significantly optimizing the quantitative analysis of

20
P. Li et al. Construction and Building Materials 392 (2023) 131925

Fig. B15. Image and data treatment results for S63d_2.

microstructure. Although this method requires a high demand for CRediT authorship contribution statement
sample quality and instrument parameters (they directly affect the ac­
curacy of the data and the workload afterwards), it is a promising Peiran Li: Data curation, Formal analysis, Investigation, Writing –
quantitative characterization alternative to link the microscopic and original draft, Writing – review & editing. Wengui Li: Conceptualiza­
submicroscopic domains. tion, Funding acquisition, Project administration, Supervision, Valida­
tion, Writing – original draft, Writing – review & editing. Kejin Wang:
6. Conclusions Validation, Writing – original draft, Writing – review & editing. Hanb­
ing Zhao: Investigation, Methodology, Writing – original draft, Writing
This study investigated the microstructures of OPC paste formed – review & editing. Surendra P. Shah: Validation, Writing – review &
with DW and SW, which combines the latest microscopy and image and editing.
data processing techniques. The quantitative characterization methods
of pore and cement phases were developed and discussed in detail, as Declaration of Competing Interest
well as the identification and morphology study of the phases.
The hydrate phases were identified by EDS point analysis. A notable The authors declare that they have no known competing financial
result was that the various AFm phases coexisted in SW samples, interests or personal relationships that could have appeared to influence
including FS, monosulfoaluminate and monocarboaluminate. These the work reported in this paper.
hydrocalumite-like phases were similar and unable to be distinguished
by morphology. Therefore, determining atomic ratio by the EDS method Data availability
contributed to identifying these phases accurately. In addition, the effect
of SW, such as stimulating cement hydration, hindering AFt decompo­ Data will be made available on request.
sition and reducing porosity, can also be observed by SEM-EDS tech­
niques, which shows its great applicability for microscopic Acknowledgment
characterization.
The study found that BSE images were more suitable for segmenting The authors appreciate the support from the Australian Research
areas for smaller pores, while SE images were more suitable for larger Council (ARC), Australia (FT220100177; DP220100036; DP220101051;
pores, inside which no hydrate phases can be observed. By combining SE IH200100010), and the University of Technology Sydney Research Ac­
and BSE images, the area for pores in the microstructure can be accu­ ademic Program at Tech Lab (UTS RAPT).
rately segmented with lower data errors. Information from EDS hyper­
maps was of great value for phase identification. After the data Appendix A. Morphology analysis of phase
processing of denoising, region segmentation, manual identification,
and imaging calibration, the region of various phases can be separated Fig. A1.
effectively. Although a higher magnification was required to ensure the
accurate result of finer phases such as AFt, the EDS hypermaps can Appendix B. Phase quantification results
identify most of the cement phases and provide detailed distribution of
the phases on microstructure images with low data error. Therefore, Fig. B1. Fig. B2. Fig. B3. Fig. B4. Fig. B5. Fig. B6. Fig. B7. Fig. B8.
with the advancing EDS detectors and optimizing algorithm, this Fig. B9. Fig. B10. Fig. B11. Fig. B12. Fig. B13. Fig. B14. Fig. B15.
method was promising to achieve efficient and highly adaptable quan­
tification of cement phases.

21
P. Li et al. Construction and Building Materials 392 (2023) 131925

References [34] G. Petschnigg, R. Szeliski, M. Agrawala, M. Cohen, H. Hoppe, K. Toyama, Digital

photography with flash and no-flash image pairs, ACM Trans. Graph. (TOG) 23 (3)
(2004) 664–672.
[1] T. Dhondy, A. Remennikov, M.N. Shiekh, Benefits of using sea sand and seawater in
[35] A. Harrisson, N. Winter, H.W. Taylor, X-ray microanalysis of microporous
concrete: a comprehensive review, Aust. J. Struct. Eng. 20 (4) (2019) 280–289.
materials, J. Mater. Sci. Lett. 6 (1987) 1339–1340.
[2] U. Ebead, D. Lau, F. Lollini, A. Nanni, P. Suraneni, T. Yu, A review of recent
[36] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua, S. Süsstrunk, SLIC superpixels
advances in the science and technology of seawater-mixed concrete, Cem. Concr.
compared to state-of-the-art superpixel methods, IEEE Trans. Pattern Anal. Mach.
Res. 152 (2022), 106666.
Intell. 34 (2012) 2274–2282.
[3] P. Li, W. Li, T. Yu, F. Qu, V.W.Y. Tam, Investigation on early-age hydration,
[37] E.W. Forgy, Cluster analysis of multivariate data: efficiency versus interpretability
mechanical properties and microstructure of seawater sea sand cement mortar,
of classifications, Biometrics 21 (1965) 768–769.
Constr. Build. Mater. 249 (2020), 118776.
[38] H.F. Taylor, Cement Chemistry, Thomas Telford, London, 1997.
[4] A. Younis, U. Ebead, P. Suraneni, A. Nanni, Fresh and hardened properties of
[39] K.L. Scrivener, P. Juilland, P.J. Monteiro, Advances in understanding hydration of
seawater-mixed concrete, Constr. Build. Mater. 190 (2018) 276–286.
Portland cement, Cem. Concr. Res. 78 (2015) 38–56.
[5] P. Li, W. Li, Z. Sun, L. Shen, D. Sheng, Development of sustainable concrete
[40] J. Jiang, Q. Zheng, D. Hou, Y. Yan, H. Chen, W. She, S. Wu, D. Guo, W. Sun, Calcite
incorporating seawater: a critical review on cement hydration, microstructure and
crystallization in the cement system: morphological diversity, growth mechanism
mechanical strength, Cem. Concr. Compos. 121 (2021), 104100.
and shape evolution, PCCP 20 (2018) 14174–14181.
[6] C. Qiao, P. Suraneni, J. Weiss, Damage in cement pastes exposed to NaCl solutions,
[41] D. Nied, K. Enemark-Rasmussen, E. L’Hopital, J. Skibsted, B. Lothenbach,
Constr. Build. Mater. 171 (2018) 120–127.
Properties of magnesium silicate hydrates (MSH), Cem. Concr. Res. 79 (2016)
[7] C. Qiao, P. Suraneni, M.T. Chang, J. Weiss, Damage in cement pastes exposed to
323–332.
MgCl2 solutions, Mater. Struct. 51 (2018) 74.
[42] C. Roosz, S. Grangeon, P. Blanc, V. Montouillout, B. Lothenbach, P. Henocq,
[8] X. Wang, S. Sadati, P. Taylor, C. Li, X. Wang, A. Sha, Material characterization to
E. Giffaut, P. Vieillard, S. Gaboreau, Crystal structure of magnesium silicate
assess effectiveness of surface treatment to prevent joint deterioration from
hydrates (MSH): the relation with 2: 1 Mg–Si phyllosilicates, Cem. Concr. Res. 73
oxychloride formation mechanism, Cem. Concr. Compos. 104 (2019), 103394.
(2015) 228–237.
[9] A. Ahmed, S. Guo, Z. Zhang, C. Shi, D. Zhu, A review on durability of fiber
[43] B. Lothenbach, D. Nied, E. L’Hôpital, G. Achiedo, A. Dauzères, Magnesium and
reinforced polymer (FRP) bars reinforced seawater sea sand concrete, Constr.
calcium silicate hydrates, Cem. Concr. Res. 77 (2015) 60–68.
Build. Mater. 256 (2020), 119484.
[44] M.B. Haha, B. Lothenbach, G. Le Saout, F. Winnefeld, Influence of slag chemistry
[10] X. Hu, J. Xiao, K. Zhang, Q. Zhang, The state-of-the-art study on durability of FRP
on the hydration of alkali-activated blast-furnace slag—Part I: effect of MgO, Cem.
reinforced concrete with seawater and sea sand, J. Build. Eng. 51 (2022), 104294.
Concr. Res. 41 (9) (2011) 955–963.
[11] A. Zhou, R. Qin, C.L. Chow, D. Lau, Structural performance of FRP confined
[45] I. Richardson, G. Groves, The incorporation of minor and trace elements into
seawater concrete columns under chloride environment, Compos. Struct. 216
calcium silicate hydrate (C□ S□ H) gel in hardened cement pastes, Cem. Concr.
(2019) 12–19.
Res. 23 (1993) 131–138.
[12] J.G. Teng, Y. Xiang, T. Yu, Z. Fang, Development and mechanical behaviour of
[46] I.G. Richardson, The calcium silicate hydrates, Cem. Concr. Res. 38 (2008)
ultra-high-performance seawater sea-sand concrete, Adv. Struct. Eng. 22 (14)
137–158.
(2019) 3100–3120.
[47] B. Lothenbach, K. Scrivener, R. Hooton, Supplementary cementitious materials,
[13] T. Nishida, N. Otsuki, H. Ohara, Z.M. Garba-Say, T. Nagata, Some considerations
Cem. Concr. Res. 41 (2011) 1244–1256.
for applicability of seawater as mixing water in concrete, J. Mater. Civ. Eng. 27
[48] R.J. Myers, E. L’Hôpital, J.L. Provis, B. Lothenbach, Effect of temperature and
(2015) B4014004.
aluminium on calcium (alumino) silicate hydrate chemistry under equilibrium
[14] K.L. Scrivener, Backscattered electron imaging of cementitious microstructures:
conditions, Cem. Concr. Res. 68 (2015) 83–93.
understanding and quantification, Cem. Concr. Compos. 26 (8) (2004) 935–945.
[49] E. L’Hôpital, B. Lothenbach, G. Le Saout, D. Kulik, K. Scrivener, Incorporation of
[15] P.T. Durdziński, C.F. Dunant, M.B. Haha, K.L. Scrivener, A new quantification
aluminium in calcium-silicate-hydrates, Cem. Concr. Res. 75 (2015) 91–103.
method based on SEM-EDS to assess fly ash composition and study the reaction of
[50] D. Vafaei, X. Ma, R. Hassanli, J. Duan, Y. Zhuge, Microstructural behaviour and
its individual components in hydrating cement paste, Cem. Concr. Res. 73 (2015)
shrinkage properties of high-strength fiber-reinforced seawater sea-sand concrete,
111–122.
Constr. Build. Mater. 320 (2022), 126222.
[16] U.H. Jakobsen, K. De Weerdt, M.R. Geiker, Elemental zonation in marine concrete,
[51] M. Balonis, F.P. Glasser, The density of cement phases, Cem. Concr. Res. 39 (2009)
Cem. Concr. Res. 85 (2016) 12–27.
733–739.
[17] T.E. Oliphant, Python for scientific computing, Comput. Sci. Eng. 9 (3) (2007)
[52] A.V. Soin, L.J. Catalan, S.D. Kinrade, A combined QXRD/TG method to quantify
10–20.
the phase composition of hydrated Portland cements, Cem. Concr. Res. 48 (2013)
[18] F. Georget, W. Wilson, K.L. Scrivener, edxia: microstructure characterisation from
17–24.
quantified SEM-EDS hypermaps, Cem. Concr. Res. 141 (2021) 106327.
[53] Y. Cai, D. Xuan, P. Hou, J. Shi, C.S. Poon, Effect of seawater as mixing water on the
[19] K.J. Krakowiak, W. Wilson, S. James, S. Musso, F.-J. Ulm, Inference of the phase-to-
hydration behaviour of tricalcium aluminate, Cem. Concr. Res. 149 (2021),
mechanical property link via coupled X-ray spectrometry and indentation analysis:
106565.
application to cement-based materials, Cem. Concr. Res. 67 (2015) 271–285.
[54] W. Li, Z. Jiang, M. Lu, W. Long, F. Xing, J. Liu, Effects of seawater, NaCl, and Na2
[20] B. Münch, L.H.J. Martin, A. Leemann, Segmentation of elemental EDS maps by
SO4 solution mixing on hydration process of cement paste, J. Mater. Civ. Eng. 33
means of multiple clustering combined with phase identification, J. Microsc. 260
(2021) 04021057.
(3) (2015) 411–426.
[55] Y. Cao, L. Guo, B. Chen, J. Wu, Thermodynamic modelling and experimental
[21] R.H. Bogue, The chemistry of Portland cement, LWW1955.
investigation on chloride binding in cement exposed to chloride and chloride-
[22] K. Scrivener, R. Snellings, B. Lothenbach, A Practical Guide to Microstructural
sulfate solution, Constr. Build. Mater. 246 (2020), 118398.
Analysis of Cementitious MATERIALS, CRC Press, Boca Raton, FL, USA, 2016.
[56] M. Balonis, B. Lothenbach, G. Le Saout, F.P. Glasser, Impact of chloride on the
[23] D.E. Newbury, N.W. Ritchie, Performing elemental microanalysis with high
mineralogy of hydrated Portland cement systems, Cem. Concr. Res. 40 (2010)
accuracy and high precision by scanning electron microscopy/silicon drift detector
1009–1022.
energy-dispersive X-ray spectrometry (SEM/SDD-EDS), J. Mater. Sci. 50 (2015)
[57] P. Li, X. Gao, K. Wang, V.W. Tam, W. Li, Hydration mechanism and early frost
493–518.
resistance of calcium sulfoaluminate cement concrete, Constr. Build. Mater. 239
[24] K.L. Scrivener, H. Patel, P. Pratt, L. Parrott, Analysis of phases in cement paste
(2020), 117862.
using backscattered electron images, methanol adsorption and thermogravimetric
[58] G. Fang, M. Zhang, The evolution of interfacial transition zone in alkali-activated
analysis, MRS Online Proceedings Library (OPL) 85 (1986).
fly ash-slag concrete, Cem. Concr. Res. 129 (2020), 105963.
[25] H. Wong, M. Head, N. Buenfeld, Pore segmentation of cement-based materials from
[59] V. Kocaba, E. Gallucci, K.L. Scrivener, Methods for determination of degree of
backscattered electron images, Cem. Concr. Res. 36 (2006) 1083–1090.
reaction of slag in blended cement pastes, Cem. Concr. Res. 42 (2012) 511–525.
[26] E. Jones, E. Oliphant, P. Peterson, SciPy: Open source scientific tools for Python
[60] Y. Cai, D. Xuan, C.S. Poon, Influence of the availability of calcium on the hydration
[http://www. scipy. org/], Go to reference in article, (2019).
of tricalcium aluminate (C3A) in seawater-mixed C3A–gypsum system, J. Am.
[27] E. Moore, J. Vand erPlas, D. Laxalde, J. Perktold, R. Cimrman, I. Henriksen,
Ceram. Soc. 105 (9) (2022) 5895–5910.
E. Quintero, C. Harris, A. Archibald, A. Ribeiro, SciPy 1.0: fundamental algorithms
[61] Y. Huo, J. Xiao, J. Shen, Q. Gao, Effects of distiller’s grains and seawater on
for scientific computing in Python, Nat. Methods (2020).
properties of sintered brick made from construction spoil, J. Build. Eng. 62 (2022)
[28] S. van der Walt, J. Schönberger, J. Nunez-Iglesias, F. Boulogne, J. Warner, N.
105391.
Yager, E. Gouillart, T. Yu, t contributors (2014). scikit-image: image processing in
[62] Y. Cai, Y. Tao, D. Xuan, Y. Sun, C.S. Poon, Effect of seawater on the morphology,
python, PeerJ.
structure, and properties of synthetic ettringite, Cem. Concr. Res. 163 (2023)
[29] J. Rossen, K. Scrivener, Optimization of SEM-EDS to determine the C-A–S–H
107034.
composition in matured cement paste samples, Mater. Charact. 123 (2017)
[63] F. Qu, W. Li, W. Dong, V.W.Y. Tam, T. Yu, Durability deterioration of concrete
294–306.
under marine environment from material to structure: A critical review, J. Build.
[30] A. Chambolle, An algorithm for total variation minimization and applications,
Eng. 35 (2021) 102074.
J. Math. Imaging Vision 20 (2004) 89–97.
[31] R.A. Haddad, A.N. Akansu, A class of fast Gaussian binomial filters for speech and
image processing, IEEE Trans. Signal Process. 39 (1991) 723–727.
[32] C. Tomasi, R. Manduchi, Bilateral filtering for gray and color images, Sixth
international conference on computer vision (IEEE Cat. No. 98CH36271), IEEE,
1998, pp. 839-846.
[33] E. Eisemann, F. Durand, Flash photography enhancement via intrinsic relighting,
ACM Trans. Graph. (TOG) 23 (3) (2004) 673–678.

22
P. Li et al. Construction and Building Materials 392 (2023) 131925

[64] F. Qu, W. Li, Z. Tang, K. Wang, Property degradation of seawater sea sand [66] F. Qu, W. Li, Y. Guo, S. Zhang, J.L. Zhou, K. Wang, Chloride-binding capacity of
cementitious mortar with GGBFS and glass fiber subjected to elevated cement-GGBFS-nanosilica composites under seawater chloride-rich environment,
temperatures, J. Mater. Res. Technol. 13 (2021) 366–384. Constr. Build. Mater. 342 (2022) 127890.
[65] F. Qu, W. Li, K. Wang, V.W.Y. Tam, S. Zhang, Effects of seawater and undesalted
sea sand on the hydration products, mechanical properties and microstructures of
cement mortar, Constr. Build. Mater. 310 (2021) 125229.

23