Construction and Building Materials 189 (2018) 157–169

Contents lists available at ScienceDirect

Construction and Building Materials

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

Development of empirical models for chloride binding in cementitious

systems containing admixed chlorides
David Trejo a, Mahmoud Shakouri b,⇑, Naga Pavan Vaddey a, O. Burkan Isgor a
a
School of Civil and Construction Engineering, Oregon State University, Corvallis, OR, USA
b
Department of Industrial Technology, University of Nebraska Kearney, Kearney, NE, USA

h i g h l i g h t s

Binding behavior of admixed chlorides in various cementitious systems are studied.

Free chlorides were measured using water-soluble and pore press testing procedures.

An interaction between w/b, SCM, SCM level, and admixed chloride level exists.

Three empirical binding isotherms are developed.

Chloride testing method influences the concentration of measured bound chlorides.

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

Article history: The binding of admixed chlorides in the concrete can delay the corrosion of reinforcing steel and extend
Received 12 June 2018 the service life of reinforced concrete structures. Although extensive research has been conducted to
Received in revised form 23 August 2018 study the chloride binding capacity of cementitious systems subject to external chlorides, there is little
Accepted 28 August 2018
information available on the chloride binding capacity of cementitious systems containing admixed chlo-
Available online 6 September 2018
rides. In this paper, the chloride binding capacity of several cementitious systems containing ordinary
portland cement (OPC), Type C and F fly ashes, slag, and silica fume with different replacement levels
Keywords:
is studied using acid-soluble, water-soluble, and pore-pressed testing procedures. In addition, the influ-
Chloride binding
Admixed chlorides
ence of water-to-binder ratio and admixed chloride levels on the percentages of bound chlorides is
Concrete durability explored and discussed. The experimental results indicate that, on average, the concentration of pore-
Corrosion pressed chlorides is significantly lower than the concentration of water-soluble chlorides. Furthermore,
the estimated bound chlorides using the pore-pressed chlorides are significantly higher than the bound
chloride estimated using water-soluble chlorides. Results show that the Langmuir isotherm can be used
to explain the relationship between free and bound chlorides in OPC paste samples containing admixed
chlorides. Results of this study also suggest that chloride testing procedures can be very influential in
quantifying the concentration of bound chlorides. Therefore, there is a need to standardize the process
used for measuring the chlorides in cementitious systems for assessing chloride binding.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction of concrete structures. In the context of chloride-induced corro-

Chlorides can be introduced into the fresh concrete by the con-
stituent materials or by intentionally adding salts (i.e., admixed
chlorides). Regardless of how chlorides enter the concrete, in suffi-
cient quantities, these chlorides can lead to early corrosion of the
reinforcing steel [1–5]. Corrosion can lead to cracking, spalling,
and reduction in the cross-section of reinforcing steel, which in
turn can result in the loss of serviceability and reduced service life
⇑ Corresponding author.
E-mail address: shakourim2@unk.edu (M. Shakouri).
sion, the service lives of a structure can be divided into two phases,
namely, corrosion initiation and propagation. The corrosion initia-
tion phase is the time required for chlorides to reach the surface of
the reinforcing steel in sufficient quantities. The quantity of chlo-
rides that can initiate the corrosion of reinforcing steel is known
as the critical chloride threshold, CT [6]. It is important to note that
there are uncertainty and debate on a single CT value or distribu-
tion of CT values. A comprehensive review of the challenges in mea-
suring CT is given in [7,8].
One of the critical factors that can delay the corrosion initiation
of the reinforcing steel in concrete is the chloride binding capacity

https://doi.org/10.1016/j.conbuildmat.2018.08.197
0950-0618/Ó 2018 Elsevier Ltd. All rights reserved.
158 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169
of the cementitious systems. As a result of chloride binding, some
of the chlorides can be removed from the concrete pore solution,
thereby reducing the concentration of available chlorides to initi-
ate the corrosion process [9]. Respectively, chlorides in the cemen-
titious systems can exist in the form of free chlorides in the
concrete pore solution and bound chlorides in the hydrated cement
paste, where the latter chlorides can be further categorized as
either chemically bound chlorides or physically bound chlorides.
Yuan et al. [9] reported that more than 80% of admixed chloride
binding takes place in less than 5 h after water is introduced to
the concrete. The type of chloride binding, to a large extent,
depends on whether chlorides are present in the concrete during
the early or late stages of cement hydration (i.e., fresh concrete
vs. hardened concrete). Some researchers attribute the corrosion
initiation of the reinforcing steel to only the concentration of free
chlorides in the pore solution at early ages [10,11]. However,
research shows that when the pH within the concrete pore solution
drops below 11.5 (e.g., in case of carbonation and sulfate attack),
the chloride binding mechanism can be reversed and more than
90% of bound chlorides can unbind and release into the pore solu-
tion in the form of free at later ages, increasing the risk of corrosion
[12,13].
In the case where chlorides are present in the concrete from the
early stages of hydration (i.e., in the form of admixed chlorides),
chlorides can bind chemically with the primary cement com-
pounds such as C3A and C4AF to form Friedel’s salt and its iron ana-
logue. In addition, depending on the type and concentration of
chlorides in the cementitious system, other products such as
Kuzel’salt and oxychlorides may be formed [14]. However, oxy-
chlorides typically only occur at high chloride exposure concentra-
tions of CaCl2 or MgCl2 salts, [15,16], well above the typical
admixed chloride levels. Some of the admixed chlorides that are
not chemically bound to the hydrated cement compounds can be
adsorbed to the surface of the hydration products, such as the cal-
cium silicate hydrates. This form of binding is of a physical nature
[17]. It should be noted that there is currently no established
method to determine what fraction of bound chlorides are physical
or chemical in nature and information about the distribution of
chloride binding in admixed chloride systems is scant and not well
understood.
Researchers often use chloride binding isotherms to express the
relationship between free and bound chlorides at a given temper-
ature. Three common chloride binding isotherms that have been
used extensively in the literature are the linear, Langmuir, and Fre-
undlich isotherms as shown in Eq. (1) through (3).
Clb;linear ¼ kClf
acf
Clb; Langmuir ¼ ð2Þ
1 þ bClf
b
Clb;Freundlich ¼ aClf
where Clb and Clf are bound and free chlorides, respectively. The
coefficients k,a, and b are empirical values determined by fitting
Eqs. (1) through (3) to the measured data. It is important to note
that Clf is often expressed in terms of the number of moles of chlo-
rides in one liter of concrete pore solution (mol/l) and Clb is
expressed in term of unit mass of chlorides per unit mass of con-
crete, cement, or cementitious materials (e.g., kg/m3 concrete, mg/
g sample, percent by weight of cement or cementitious materials).
Conventionally, the rate of change in bound chlorides as a function
of free chlorides in the pore solution is referred to as the binding
capacity of the cementitious system. This approach, however, is
faced with several challenges. For example, to predict the concen-
tration of bound chlorides, one must first determine the concentra-
tion of free chlorides in the concrete pore solution using different
testing methods. These tests are time-consuming and often require
a well-equipped laboratory. In addition, different test procedures
could produce different estimates of ‘‘free” chlorides which adds
to the uncertainty of the results.
There is currently debate over the test method used to extract
the free chlorides from a cementitious sample. Free chlorides are
commonly defined as the unbound and unabsorbed portion of
chlorides in the concrete pore solution [18]. Using this definition,
a review of the literature shows that there are three common test
methods, namely equilibrium method, water-soluble, and pore
press testing procedure that have been used to assess the concen-
tration of free chlorides and determine the binding capacity of
cementitious systems.
In equilibrium method, a sample (e.g., a thinly sliced disk or
powdered concrete sample) is added to a chloride solution with
a known concentration of chlorides and once the equilibrium is
reached, the concentration of chlorides in the solution is measured
and used as an indicator of free chlorides [9,19–22]. This method
measures the binding capacity of cementitious systems exposed
to external chlorides and therefore, ignores the possible impacts
of admixed chlorides during the hydration process. Another draw-
back to the equilibrium method is that it ignores the influence of
chloride binding products (e.g., Friedel’s salt) on the structure of
pores in concrete. Research shows that the products of chloride
binding can reduce the porosity of concrete and as a result, can
slow down the transport of chlorides into the concrete [9,23]. In
addition, the equilibrium method disregards the variation in the
concentration of free chlorides in the pore solution and its possible
impacts on the transport properties of chlorides in concrete. It is
expected that an increase in the concentration of free chlorides
in the pore solution can result in a reduction in the chloride con-
centration gradient between the pore solution and the exposure
environment which in turn can reduce the diffusion rate of chlo-
rides in concrete.
The literature on chloride binding also shows that some
researchers have used the water-soluble testing method to esti-
mate the concentration of free chlorides and develop binding iso-
therms based on the water-soluble chlorides [24]. Because
water-soluble test involves boiling the powdered concrete samples
in water, the water-soluble test has been reported to release some
of the physically bound chlorides, and therefore, the obtained val-
ues from this test may not be equivalent to the actual concentra-
tion of free chlorides. It is also important to note that there is a
wide range of water-soluble tests (e.g., ASTM C1218 [25], RILEM
TC 178-TMC [26], AASHTO T260 [27], ACI 222.1-96 [28]). In fact,
ð1Þ the amount of unbinding would be expected to be very much
dependent on the test procedures used.
Research indicates that among all different test methods to
measure the concentration of free chlorides in the concrete, the
pore press test can be the most representative test procedure for
ð3Þ assessing these free chlorides [22]. In this test, the pore solution
is extracted by applying high pressures to the hardened cementi-
tious material samples [21]. Applying high pressures to the hard-
ened cementitious materials could unbind some of the bound
chlorides. Therefore, there may be some degree of error associated
with this test per the definition of free chlorides in the pore solu-
tion. In fact, some research indicates that the pore press technique
may overestimate the concentration of free chlorides by up to 20%
[29]. One drawback to using pore press testing procedure is that
extracting the pore solution from cementitious systems requires
high pressure and high-capacity equipment, which are not com-
monly available. In addition, the volume of the extracted pore solu-
tion is dependent on the magnitude and the duration of the applied
force to the crushed cementitious samples. Based on the authors’
experience, pore press method is more suitable for young samples
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169
Table 1
The proportion of common oxides in OPC and SCM (% by mass).
SiO2 Al2O3 Fe2O3 CaO MgO SO3
OPC 20.2 4.6 3.4 63.5 0.8 3.1
FF 48.4 16.4 6.3 14 5.3 0.9
FC 38.4 20.4 6.9 21.6 4.7 0.7
SF 92.1 0.9 1.3 0.9 1.7 0.5
GGBS 31.4 15.7 0.4 37.7 8.6 2.5
LOI = Loss of ignition.
with high water-to-binder (w/b) values. With time, the chances of
obtaining sufficient pore solution for testing becomes significantly
smaller. There is currently no standardized procedure available
regarding how much pressure and how long the pressure should
be applied to the cementitious samples. In fact, research labs fol-
low different protocols, which could result in higher variability
and uncertainty associated with the results of the pore press test.
Standardized testing is needed to provide a realistic representation
of free chlorides such that realistic service lives can be predicted.
Based on the discussion above, there is a pressing need to stan-
dardize the testing methods used to determine the concentration
of free chlorides in the cementitious systems. There is also a need
to further investigate the chloride binding behavior of the cemen-
titious systems containing admixed chlorides. The objectives of
this study are to first, identify the relationship between different
chloride testing methods; second, determine what fraction of
admixed chlorides can bind in the selected cementitious systems;
third, study the effects of w/b, supplementary cementitious mate-
rials (SCM) type, SCM level, and admixed chloride level on binding
capacity of paste specimens containing different admixed chloride
level; fourth, compare the existing binding isotherm models with
the binding behavior of admixed chlorides; and finally, develop
multivariate models for predicting the binding capacity of admixed
chlorides using the generated empirical data in this study. The pro-
posed binding models can reduce or eliminate the need for per-
forming time-consuming and costly tests. These models can also
be used in the future studies to assess the risk of corrosion associ-
ated with the release of bound chlorides due to carbonation and/or
sulfate attacks.
2. Experimental program
Extensive research has been conducted to investigate the effects
of chloride concentration [30,31], cement type and cement compo-
sition [14,32–36], the presence of SCMs [15,37–40], carbonation
[39,41–44], the presence of sulfate ions [45–47], the cation type
of the chloride salt [20,30,48–50], temperature [51], the presence
of surfactants [52], wetting and drying frequencies [53], and pH
of the pore solution [29,54–56] on the binding behavior of cemen-
titious systems exposed to external chlorides.
In this study, a full factorial design was used to study the effects
of SCM type and SCM replacement level, the concentration of
admixed chlorides expressed as percent by weight of the binder,
and w/b on the binding behavior of cementitious systems contain-
ing admixed chlorides. Three test methods were implemented in
this study to measure the concentration of chlorides: ASTM
C1152 [57] to measure acid-soluble chlorides. Acid-soluble chlo-
rides are often used to represent the concentration of ‘‘total chlo-
rides” in a cementitious system; ASTM C1218 [25] to measure
water-soluble chlorides. These chlorides may be used to represent
‘‘free chlorides”; and pore-press testing method to measure the
concentration of free chlorides in the pore solution.
In addition to OPC, the SCMs used in this study included type C
fly ash (FC), type F fly ash (FF), silica fume (SF), and blast furnace
159
slag (GGBS). Table 1 presents the oxide composition of OPC and
SCMs used in this study.
LOI The considered SCM replacement levels for FC, FF, and GGBS
2.7 samples were 25 and 50 percent by weight of cement. For SF sam-
0.3 ples, the replacement levels were 5 and 10 percent by weight of
0.3 cement. Paste samples were prepared with w/b = 0.35, 0.45, and
0.2 0.55. Admixed chlorides were added to the paste samples in four
0.6
levels of 0, 0.05, 0.25, and 1 percent by weight of the binder. Table 2
shows the summary of mixtures used in this study.
2.1. Specimen preparation and testing
In this study, two groups of paste samples were prepared. The
first group of samples were used for acid-and water-soluble chlo-
ride testing, and the second group of samples were used for pore
press chloride testing. The first group comprised of three hundred
and twenty-four cylindrical paste samples of 34.3 by 52.1 mm
(diameter  height) [1.35 by 2.05 in]. The samples size for each
mixture was three. The samples were prepared in a mortar mixer
following ASTM C305-14 [58]. Before mixing, sodium chloride
(NaCl) was dissolved in the mixing water (i.e., deionized water)
at the predefined levels. The mixtures were then poured into
34.3 by 52.1 mm [1.35 by 2.05 in] vials and kept sealed for 24 h.
During this period and to minimize bleeding in the samples with
w/b = 0.45 and 0.55, the vials were turned continuously on a mill
with 40 revolutions per minute. After 24 h, the vials were stripped
off and samples were kept in a moist room at 23 °C [73.4 °F] and
100% relative humidity for 56 days. Upon the completion of the
curing period, samples were removed from the moist room and
air dried for 24 h. The samples were then ground into powder
using a customized lathe. On average, 40 g [1.7 oz] of ground pow-
der was obtained from each sample. The obtained powder for each
sample was then divided into parts, half of which was used for
acid-soluble testing and the other half was used for water-
soluble testing. The concentrations of acid-soluble and water-
soluble chlorides were determined by potentiometric titration
with silver nitrate as a titrant following a modified version of ASTM
C1152 [57] and ASTM C1218 [25], respectively. The modifications
included performing auto-titration testing as opposed to manual
titration. Trejo and Ahmed [59] verified that auto-titration results
in similar mean values but with lower variability. Note that both
water- and acid-soluble chloride results were obtained from the
same hardened paste specimen.
The second group of samples in this study included 540 cylin-
drical paste samples of 51 by 102 mm [2 by 4 in]. It is important
to note that the first and second group of samples were made from
the same batch and underwent the same curing regime. The sam-
ple size for each mixture in the second group was five. Two of these
samples were used to determine the volume of evaporable water in
the samples. Evaporable water content is commonly used to esti-
mate the volume of the pore solution. For this purpose, after
56 days of moist curing, two samples from each mixture were
weighted and then placed in an oven at 105 °C [221 °F] until the
samples reached a constant weight. The percentage difference
between the weights before and after drying was used as the
amount of evaporable water.
The other three samples for each mixture were crushed into
pieces and used to determine the concentration of free chlorides
in the pore solution using a custom-built pore press equipment
as shown in Fig. 1. The crushed pieces were placed in the sleeve
of the pore press and loaded in a 700 Kip Forney compression test-
ing machine at the rate of approximately 136 kg/s [300 lb/s] up to
890,000 N [200,000 lbs]. There is currently no standard available
regarding the amount of required force for extracting pore solution
from hardened cementitious samples. The values used in this study
were found empirically and gave the best results in terms of
160 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169

Table 2
Details of paste mixtures.

Mix w/b Admixed chloride level Binder (% by weight if cement)

(% by weight of binder)
Portland cement Fly ash type C Fly ash type F Slag Silica fume
OPC 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 100 – – – –
FC25 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 75 25 – – –
FC50 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 50 50 – – –
FF25 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 75 – 25 – –
FF50 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 50 – 50 – –
GGBS25 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 75 – – 25 –
GGBS50 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 50 – – 50 –
SF5 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 95 – – – 5
SF10 0.35, 0.45, 0.55 0, 0.05, 0.25, 1 10 – – – 10

volume of pore solution. During the pore extraction process, the
authors noticed that obtaining pore solution from samples contain-
ing SCMs was very challenging. For example, in many occasions,
the SCM containing samples got stuck in the pore press equipment
and removing them required considerable time and effort. There-
fore, three samples were tested for OPC mixtures and only one
sample was tested for the mixtures that contained any type of
SCM. The concentration of chlorides in the pore solution was deter-
mined by potentiometric titration with silver nitrate as a titrant.
It is important to note that the background chlorides content
(e.g., chlorides in the mixing water and SCMs) in the first and sec-
ond group of samples were measured by performing acid-soluble
testing procedure per ASTM C1152 [57] on samples with zero per-
cent admixed chlorides. The sum of background and admixed chlo-
rides was then used to represent the concentration of total
chlorides in the samples.
2.2. Chloride binding capacity
In this study, the chloride binding capacity of cementitious sys-
tems is defined as the difference between the concentration of
total and free chlorides expressed as percent by weight of the bin-
der. Using this definition, three different models are proposed. The
generic equation used to quantify the binding capacity is as
follows:
Cly  Clx
ClbðyxÞ ¼  100
Cly
where Clb(y-x) is the percentage of bound chlorides, Cly can be the
total or acid-soluble chlorides (denoted by ASC), and Clx can be
Fig. 1. Pore press equipment used in the study.
the pore-pressed chlorides (denoted by PSC) or water-soluble chlo-
rides (denoted by WSC). Depending on the type of test used, Clb(y-x)
can be expressed as Clb(t-w), Clb(a-w), and Clb(t-p) where t denotes total
chlorides, w denotes water-soluble chlorides, and p denotes pore-
pressed chlorides.
3. Experimental results
Fig. 2 shows the bivariate scatter plots of total chlorides vs. the
concentration of chlorides measured using different test methods
in this study. The dashed lines in Fig. 2a through 1c represent
the 1:1 line (Pearson’s r = 1) and the solid lines are the simple lin-
ear regression fits with the intercept going through the origin.
Fig. 2a shows that there is a significant positive correlation
between the total chlorides and measured ASC (r = 0.99). Results
show that when the concentration of admixed chlorides is low,
ASC shows little variability. However, the variability in ASC mea-
surements increases with an increase in the concentration of
admixed chlorides. The fitted linear regression line in Fig. 2a shows
that on average ASC can account for approximately 94% of the total
chlorides in a sample. This finding suggests that up to 6% of the
admixed chlorides can be considered as ‘‘ very tightly” bound chlo-
rides that will not unbind under the acid-soluble testing conditions
described in ASTM C1152 [57]. Fig. 2b and c show that the concen-
tration of WSC and PSC are significantly lower than the concentra-
tion of total chlorides. The slopes of the regression lines indicate
that on average, WSC and PSC underestimate the total chlorides
ð4Þ
significantly. Similar to ASC results, the variability in the WSC
and PSC measurements increases with an increase in the amount
of admixed chlorides and total chlorides.
Fig. 3a and b show that the WSC and PSC measurements are on
average 59% and 46% of the ASC measurements. Fig. 3c shows that
the PSC measurements are on average 77% of the WSC measure-
ments. These results are in line with previous findings in the liter-
ature that water-soluble testing procedure can release a portion of
bound chlorides and therefore, measurements from the WSC test-
ing procedure are typically higher than the measurements from
the PSC testing procedure. It should be noted that the PSC results
for the samples containing SCMs are based on testing one sample.
Therefore, more research is needed to generalize the PSC results.
The differences in the chloride measurements using different
test methods suggest that the test method can be an influential fac-
tor affecting the estimates of bound chlorides. For example, the
concentration of bound chlorides that are estimated using ASC to
represent ‘‘total chlorides” and WSC to represent ‘‘total” ‘‘free”
chlorides would be expected to be significantly lower than the con-
centration of bound chlorides estimated using the concentration of
actual total (i.e., background chlorides + admixed chloride) and
pore-pressed chlorides. These findings underline the need for stan-
dardization of the test procedure used to estimates the concentra-
tion of bound and free chlorides in the cementitious systems.
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169 161

1.0 a 1.0 b 1.0 c

WSC (% wt. cementitious)

ASC (% wt. cementitious)

PSC (% wt. cementitious)

0.8 0.8 0.8

0.6 0.6 0.6

94
0.
0.4 = 0.4 .56 0.4
=0
.43
=0
0.2 0.2 0.2

Pearson's r = 0.99 Pearson's r = 0.96 Pearson's r = 0.88

0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Total Chlorides (% wt. cementitious) Total Chlorides (% wt. cementitious) Total Chlorides (% wt. cementitious)

Fig. 2. The relationship between total chlorides, acid-soluble chlorides (ASC), water-soluble chlorides (WSC) and pore-solution chlorides (PSC).

1.0 a 1.0 b 1.0 c

WSC (% wt. cementitious)

PSC (% wt. cementitious)

PSC (% wt. cementitious)

.77
=0
0.8 0.8 0.8

0.6 0.6 0.6

0.4 .59 0.4 0.4

=0 .46
=0
0.2 0.2 0.2

Pearson's r = 0.96 Pearson's r = 0.88 Pearson's r = 0.91

0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
ASC (% wt. cementitious) ASC (% wt. cementitious) WSC (% wt. cementitious)

Fig. 3. The relationship between acid-soluble chlorides (ASC), water-soluble chlorides (WSC) and pore-solution chlorides (PSC).

The relationship between different binding models as defined in Table 3 shows the descriptive statistics for the concentration of
this study is shown in Fig. 4. Results show that there is no signifi- pore-pressed chlorides obtained from the OPC samples (sample
cant difference between Clb(t-w) and Clb(a-w). However, Fig. 3b and c size = 3). The small values of standard deviations in Table 3 show
show both Clb(t-w) and Clb(a-w) are significantly lower (about 74% in that pore-pressed chloride measurements have low variability.
both figures) than Clb(t-p). This is because bound chlorides in Clb(t-p) The trend in Table 3 implies that in OPC samples, increasing the
are estimated by taking into account the total chloride that is lar- w/b from 0.35 to 0.55 can lead to an increase in chloride binding
ger than ASC and PSC that is smaller than WSC. Assuming that the and reduction in the concentration of free chlorides in the pore
PSC is the most representative measure of free chlorides, Fig. 4 sug- solution.
gests that using other tests will result in a significant underestima- Fig. 5 shows the plots of the mean Clb(t-p) in the cementitious
tion of bound chlorides. systems with different w/b and admixed chloride levels. The trend

100 a 100 b 100 c

Clb(a-w) (% Total Cl)

Clb(a-w) (% Total Cl)

Clb(t-w) (% Total Cl)

80 80 80

60 60 60

40 40 40

20 20 20
01 4
.7 4
1. .7
=
Pearson's r = 0.97 =0 Pearson's r = 0.51 =0 Pearson's r = 0.55
0 0 0
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
Clb(t-w) (% Total Cl) Clb(t-p) (% wt. binder) Clb(t-p) (% Total Cl)

Fig. 4. The relationship between different binding capacities.

162 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169

Table 3
Mean (Standard deviation) of pore-pressed chlorides for the OPC samples.

w/b Free chlorides (mol/l) Bound chlorides (% wgt. of Binder)

Admixed chloride level (% wgt. of Binder) Admixed chloride level (% wgt. of Binder)
0.05 0.25 1 0.05 0.25 1
0.35 0.03 (0.001) 0.07 (0.002) 0.74 (0.003) 0.04 (0.001) 0.20 (0.002) 0.16 (0.004)
0.45 0.01 (0.001) 0.04 (0.003) 0.52 (0.017) 0.05 (0.001) 0.22 (0.003) 0.41 (0.02)
0.55 0.01 (0.001) 0.02 (0.001) 0.36 (0.031) 0.05 (0.001) 0.23 (0.001) 0.48 (0.046)

100 a b c
(% Total chloride)

80
Clb(t-p)

60

40
w/b = 0.35 w/b = 0.35 w/b = 0.35
w/b = 0.45 w/b = 0.45 w/b = 0.45
20
w/b = 0.55 OPC w/b = 0.55 SF5 w/b = 0.55 SF10

100 d e f
(% Total chloride)

80
Clb(t-p)

60

40
w/b = 0.35 w/b = 0.35 w/b = 0.35
w/b = 0.45 w/b = 0.45 w/b = 0.45
20
w/b = 0.55 FC25 w/b = 0.55 FC50 w/b = 0.55 FF25

100 g h i
(% Total chloride)

80
Clb(t-p)

60

40
w/b = 0.35 w/b = 0.35 w/b = 0.35
w/b = 0.45 w/b = 0.45 w/b = 0.45
20
w/b = 0.55 FF50 w/b = 0.55 GGBS25 w/b = 0.55 GGBS50
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Admixed Chloride Admixed Chloride Admixed Chloride
(% wt. binder ) (% wt. binder) (% wt. binder )

Fig. 5. Measured bound chlorides based on the pore-solution chlorides.

in Fig. 5a through i indicates that in some samples, an increase in
the concentration of admixed chlorides results in an initial increase
in the concentration of bound chlorides followed by a reduction
afterward. There were, however, a few exceptions that showed
no initial increase in the concentration of bound chlorides. The
increasing and decreasing trend in Fig. 5 suggests that the selected
cementitious systems have a maximum binding capacity. If the
admixed chlorides exceed this point, the degree to which chlorides
are bound does not increase (or increases minimally). Fig. 5a
through i also show that there is an interaction between w/b and
admixed chloride levels. For example, in the OPC samples
(Fig. 5a) the maximum concentration of bound chlorides at low
admixed chloride levels is observed in samples with w/b = 0.45.
However, when the concentration of admixed chlorides increases,
samples with w/b = 0.55 show to the highest concentration of
bound chlorides. There is also an interaction between the SCMs
admixed level, w/b, and the admixed chloride level. For example,
comparing Fig. 5b and c shows that at low concentrations of
admixed chlorides, SF5 samples with w/b = 0.35 have the lowest
binding capacity. However, when the amount of silica fume in
the paste samples is increased to 10%, the samples with w/
b = 0.45 have the lowest binding capacity.
Fig. 6 compares Clb(t-p), Clb(t-w), and Clb(a-w) in paste samples and
with different SCM, SCM level, w/b, and admixed chloride levels.
Results show that there is an interaction between w/b, SCM
replacement level, and admixed chloride level. However, the gen-
eral trend in the results suggest that Clb(t-p) > Clb(t-w) > Clb(a-w).
Results also show that while there is not a significant difference
between Clb(t-w) and Clb(a-w), in some conditions Clb(t-p) can be sig-
nificantly larger or smaller than Clb(t-w) and Clb(a-w). The variability
in Clb(t-p) can be explained by the small sample size used for deter-
mining the concentration of free chlorides in the pore solution. The
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169 163

120 OPC a SF5 b SF10 c

w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55
100
(% Total chloride)
Bound Chloride
80

60

40

20

0
120 FC25 d FC50 e FF25 f
w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55
100
(% Total chloride)
Bound Chloride

80

60

40

20

0
120 FF50 g GGBS25 h GGBS50 i
w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55 w/b =0.35 w/b =0.45 w/b =0.55
100
(% Total chloride)
Bound Chloride

80

60

40

20

0
0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00 0.05 0.25 1.00
Admixed Chloride Admixed Chloride Admixed Chloride
(% wt. binder) (% wt. binder) (% wt. binder)

Clb(t-p) Clb(t-w) Clb(a-w)

Fig. 6. Comparison of different bound chloride models.

Experimental admixed data

general trend in Fig. 6 suggests that with an increase in the concen- 1.0 Admixed Langmuir isotherm
Bound chloride (% wgt. binder)

tration of admixed chloride level, the concentration of bound chlo- External Langmuir isotherm
rides is reduced in all three binding models. 0.8
Fig. 7 shows an empirical isotherm developed for the OPC sam- 4.2Cl f
ples containing admixed chlorides with w/b = 0.45. The superim- Clb =
1 + 3.2Cl f
posed line above the admixed chloride isotherm is an isotherm 0.6
developed by Glass and Buenfeld [60] for an OPC sample exposed
to external chlorides with w/b = 0.45. Results show that the 0.4
admixed chloride binding isotherm follows a Langmuir isotherm
with parameters a = 12.1 and b = 27.7. Similarly, Glass and Buen- 12.1Cl f
0.2 Clb =
feld [60] reported that a Langmuir isotherm with parameters 1 + 27.7Cl f
a = 4.2 and b = 3.2 best fit the chloride binding measurements for
the OPC sample subject to external chlorides. 0.0
Fig. 7 shows that at lower concentrations of free chlorides, the 0.0 0.1 0.2 0.3 0.4 0.5 0.6
chloride binding capacity of the sample containing admixed chlo-
Free chloride (mol/l)
rides (i.e., the slope of the binding isotherm) is greater than the
chloride binding capacity of the sample subjected to external chlo- Fig. 7. Comparison of admixed with external chloride binding isotherms.
rides. However, with an increase in the concentration of free chlo-
rides, the binding capacity of the OPC sample exposed to external
chlorides exceeds the binding capacity of OPC samples containing 4. Developing predictive models for binding capacity
admixed chlorides. These results show that the presence of
admixed chlorides can affect the binding behavior of cementitious In this section, three models for predicting the binding capacity
systems. of admixed chlorides in cementitious systems are developed using
164 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169

the empirical data obtained in this study. The primary objective of to prevent collinearity among the nominal variables (i.e., SCM
these models is to estimate the fraction of admixed chlorides that type), the OPC group is used as a reference group and is not
can be bound by the cementitious systems without the need to included as a variable in the models.
perform extensive testing. It should be noted that the models pre- The identified model for Clb(t-p) is given by:
sented in this section are developed within the bounds of this
ClbðtpÞ ¼ h1 þ h2 ðCladmixed Þ þ h3 ðCladmixed  SCM lev elÞ
2
study and based on studying specific SCMs and one type of salt. ð5Þ
Moreover, the w/b, SCM, and salt replacement levels were not þh4 ðCladmixed  w=bÞ þ h5 ðCladmixed  SCM lev el  w=bÞ
exhaustive. Thus, care must be taken when using these models where h1 is the intercept and hi s for i = 2,. . .5 are the unknown regres-
for conditions that contain different types of SCMs or admixed sion coefficients. The Cladmixed is the admixed chloride level (% wt.
salts. Fig. 5 indicates that bound chlorides, as a function of admixed cementitious materials), and SCM level is the replacement level of
chlorides, can be nonlinear. Hence, to accommodate this nonlinear- SCM (% by weight of cement). The Clb(t-p) model in Eq. (5) shows that
ity in the models, quadratic models including second order and the SCM type is not considered to be a significant factor in determin-
third order interactions between the w/b, SCM level, and admixed ing Clb(t-p). However, the interaction of SCM level, admixed chloride
chloride levels along with the squared terms of the main variables level, and w/b does influence the binding of admixed chlorides.
were considered. It is important to note that there is no unique The model for Clb(t-w) can be defined as follows:
way of finding the best model and this process is a part of the
art of model building. One significant difference between the pro- ClbðtwÞ ¼ h1 þ h2 FC þ h3 GGBS þ h4 SCM þ h5 Cladmixed
posed models in this study and the existing binding isotherms (e.g., 2
þ h6 ðSCMÞ2 þ h7 ðCladmixed Þ þ h8 ðSCM  Cladmixed Þ
Freundlich or Langmuir) is that the later determines the concentra-
tion of bound chlorides as a function of free chlorides whereas the þ h9 ðCladmixed  w=bÞ ð6Þ
proposed models in this study determine the fraction of admixed where FC and GGBS are dummy variables taking the values of 0 or 1
chlorides that can be bound as a function mixture design variables depending on the inclusion or exclusion of SCMs in the mixture.
and the concentration of admixed chlorides. Finally, the model for Clb(a-w) is given by:
There is a trade-off between the number of variables included in
2
a model and the bias and variance of the model estimates. Elimina- ClbðawÞ ¼ h1 þ h2 FC þ h3 GGBS þ h4 SCM þ h5 Cladmixed þ h6 ðCladmixed Þ
tion of key variables can damage the prediction power of the mod- þ h7 ðSCM  Cladmixed Þ þ h8 ðCladmixed  w=bÞ ð7Þ
els leading to biased estimates [61]. Moreover, the inclusion of too
many variables in a model may overfit the model resulting in a sig- Tables 4 through 6 present the summary statistics of the model
nificant variance in the estimated parameters. Therefore, it is parameters and model errors. The correlation coefficients in Tables
imperative that a good model must be parsimonious and achieve 4 through 6 show that there are high correlations among some of
the lowest bias and variance in the model estimates with the few- the parameters. For example, in Table 3, h2 with h4, and h4 with
est number of variables. Following Gardoni et al. [62], a stepwise h5 are strongly correlated. In Table 5, h4 with h6 and h5 with h7
model reduction algorithm was applied to reduce the number of are strongly correlated, and in Table 6, h5 is highly correlated with
variables in the model. This stepwise algorithm starts with a full h6. Considering that interaction terms were included in the models,
model (i.e., the model that includes all the explanatory variables, such high correlations between these parameters are not surpris-
interaction and squared terms) and in each step, the variable that ing. As a further simplification, when the correlation coefficient
has the highest coefficient of variation (COV) among all other vari- qij between hi and hj is higher than 0.7, hi can be expressed as a lin-
ables is eliminated. Upon eliminating a variable, the standard devi- ear function of hj through the following relationship:
ation of the model error r is recalculated. This process continues rhi
hi ¼ lhi þ qhi hj ðh  lhj Þ ð8Þ
until a sudden jump in the r is observed. rhj j
Fig. 8 shows the results of the variable selection for three chlo-
ride binding models. The selected models in Fig. 8 are shown with wherelh = mean andrh = standard deviation of h . This process is
solid markers. Fig. 8 shows that the model for Clb(t-p) is achieved by repeated until all high correlations are eliminated.
eliminating ten variables. Therefore, including the intercept, the For the Clb(t-p) model, h4 can be expressed in terms of h2 using Eq.
model for Clb(t-p) has five variables. Similarly, results show that (5) as:
the models for Clb(t-w) and Clb(a-w) are achieved after eliminating ^h4 ¼ 1:015  2:339h2 ð9Þ
six and seven variables, respectively. It is important to note that
Upon re-evaluating the model parameters, it was observed that
the correlation coefficient between h3 and h5 was 0.98. Therefore,
Clb(t-p) h5 can be written as a function of h3 as follows:
Standard deviation of the model error

0.14
Clb(t-w)
^h5 ¼ 0:015  2:135h3 ð10Þ
Clb(a-w)
0.12 Furthermore, it was observed that the correlation coefficient
between h2 and h3 was 0.73. Therefore, h3 is written as a function
of h2 as follows:
0.10
^h3 ¼ 2:269 þ 2:584h2 ð11Þ
The final model form for Clb(t-p) is as follows:
0.08
0 1
ðCladmixed Þ þ 2:584ðCladmixed  SCM lev elÞ
2

B C
0.06 ClbðtpÞ ¼ h2 @ - 2:339ðCladmixed  w=bÞ A
- 5:517 ðCladmixed  SCM lev el  w=bÞ
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 þ 2:269ðCladmixed  SCM lev el Þ - 1:015ðCladmixed  w=bÞ
Step - 4:829ðCladmixed  SCM lev el  w=bÞ þ h1
ð12Þ
Fig. 8. Variation of r in different steps of variable selection.
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169 165

Table 4
Regression coefficients for the Clb(t-p) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h1 h2 h3 h4 h5
h1 0.813 0.019 1.00
h2 0.870 0.103 0.59 1.00
h3 0.021 0.004 0.46 0.64 1.00
h4 1.021 0.252 0.69 0.96 0.63 1.00
h5 0.029 0.010 0.45 0.62 0.98 0.65 1.00

Table 5
Regression coefficients for the Clb(t-w) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h1 h2 h3 h4 h5 h6 h7 h8 h9
h1 0.89 0.02 1.00
h2 0.15 0.01 0.09 1.00
h3 0.08 0.01 0.09 0.47 1.00
h4 0.01 0.00 0.60 0.38 0.38 1.00
h5 1.54 0.08 0.56 0.00 0.00 0.04 1.00
h6 0.00 0.00 0.45 0.30 0.30 0.94 0.00 1.00
h7 0.77 0.06 0.52 0.00 0.00 0.00 0.85 0.00 1.00
h8 0.01 0.00 0.43 0.00 0.00 0.23 0.18 0.00 0.00 1.00
h9 0.51 0.09 0.00 0.00 0.00 0.00 0.48 0.00 0.00 0.00 1.00

Table 6
Regression coefficients for the Clb(a-w) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h1 h2 h3 h4 h5 h6 h7 h8
h1 0.87 0.01 1.00
h2 0.13 0.01 0.05 1.00
h3 0.09 0.01 0.05 0.42 1.00
h4 0.00 0.00 0.59 0.33 0.33 1.00
h5 1.65 0.08 0.63 0.00 0.00 0.12 1.00
h6 0.86 0.06 0.59 0.00 0.00 0.00 0.85 1.00
h7 0.01 0.00 0.48 0.00 0.00 0.67 0.18 0.00 1.00
h8 0.44 0.09 0.00 0.00 0.00 0.00 0.48 0.00 0.00 1.00

Similarly, the final models for Clb(t-w) and Clb(a-w) are given in
Eqs. (13) and (14).
ClbðtwÞ ¼ h1 þ h2 FC þ h3 GGBS þ h8 ðCM  Cladmixed Þ
þ ðSCMÞ2 ð0:00004  0:0453h4 Þ þ h4 SCM
2 that the proposed models are unbiased.
þ ðCladmixed Þ ð0:249  0:663h5 Þ þ ðCladmixed  w=bÞ
 ð0:276  0:513h5 Þ þ h5 Cladmixed
ClbðawÞ ¼ h2 FC þ h3 GGBS þ h4 SCM þ h5 Cladmixed
2
þðCladmixed Þ ð0:654h5  0:223Þ þ h7 ðCM  Cladmixed Þ
þðCladmixed  w=bÞð0:507h5  0:396Þ
0:102h5 þ 0:701
The means of the model parameters, the standard deviation of
the model error and the correlation coefficients between parameters
of the final models corresponding to Eqs. (12) through (14) are
shown in Tables 7 through 9, respectively. As shown, the absolute
values of the correlation coefficients are below 0.7, and therefore,
these models can be considered as the final parsimonious models.
Figs. 9 through 11 show the model diagnostics results for the
three final binding models. The diagonal 1:1 lines in Figs. 9 through
11 represents the perfect fit, and the dashed lines represent ±30%
margin of error. To validate the models, the mean squared errors
(MSE) of the training set (containing 60% of data) was compared
with the mean squared prediction errors (MSPR) of the prediction
set (containing the remaining 40% of data). Unbiased models with
an acceptable predictive ability should have a reasonably equal
MSE and MSPR values [61]. For information on MSE and MSPR
see [61]. Comparison of MSE and MSPR in all three models show
Furthermore, the alignment of the observed points versus the
ð13Þ predicted points along the 1:1 line indicates that these models
have adequate predictive power. The normality plots show that
the distribution of residuals fit well to a normal distribution, which
is a necessary condition for the validity of the model. Finally, the
plots of predicted values versus the residuals in all three models
show that the residuals are randomly scattered around zero.
ð14Þ
5. Relevance to industry
Chloride binding capacity is an important factor that can affect
the corrosion of the reinforcing steel and service life of reinforced
concrete structures. In addition, the knowledge of the binding
capacity of different cementitious systems can help policymakers
make educated decisions regarding the durability requirements
of concrete structures. In this section, the application of the devel-
oped models to estimate the admixed chloride binding capacity of
cementitious systems is illustrated through an example. To this
end, several concrete mixtures were designed, and the information
regarding the mixture proportions are provided in Table 10. The
166 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169

Table 7
Summary statistics of the parameters in the final Clb(t-p) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h1 h2
h1 0.813 0.011 1.00
h2 0.870 0.001 0.10 1.00

Table 8
Summary statistics of the parameters in the final Clb(t-w) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h1 h2 h3 h4 h5 h8
h1 0.886 0.013 1.00
h2 0.152 0.012 0.07 1.00
h3 0.081 0.012 0.07 0.44 1.00
h4 0.010 0.001 0.64 0.43 0.43 1.00
h5 1.541 0.083 0.62 0.02 0.02 0.04 1.00
h8 0.006 0.000 0.08 0.11 0.11 0.21 0.21 1.00

Table 9
Summary statistics of the parameters in the final Clb(a-w) model.

Parameter Mean Standard Correlation Coefficient

Deviation
h2 h3 h4 h5 h7
h2 0.132 0.012 1.00
h3 0.089 0.012 0.42 1.00
h4 0.003 0.000 0.53 0.53 1.00
h5 1.651 0.081 0.00 0.00 0.17 1.00
h7 0.006 0.000 0.00 0.00 0.58 0.30 1.00

0.4
1.0 a 0.2 b c
Predicted Clb(t-p) (% wt. binder)

Normal Theoretical Qauantiles

0.3

0.8 MSE = 0.011 0.1 0.2

MSPE = 0.013
Residuals

0.1
0.6
0.0
0.0
0.4 -0.1
-0.1
0.2 -0.2
Training set -0.2
Validation set -0.3
0.0
0.0 0.2 0.4 0.6 0.8 1.0 -0.2 -0.1 0.0 0.1 0.2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Observed Clb(t-p)(% wt. binder) Data Qauntiles Predicted Values

Fig. 9. Model diagnostics for the Clb(t-p) model: (a) cross-validation; (b) normality of residuals; (d) residuals versus predictive values.

0.4
Predicted Clb(t-w) (% wt. binder)

1.0 a 0.2 b c
Normal Theoretical Qauantiles

0.3

0.8 MSE = 0.004 0.2

0.1
MSPE = 0.005
Residuals

0.1
0.6
0.0
0.0
0.4 -0.1
-0.1
0.2 -0.2
Training set
-0.2
Validation set -0.3
0.0
0.0 0.2 0.4 0.6 0.8 1.0 -0.2 -0.1 0.0 0.1 0.2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Observed Clb(t-w) (% wt. binder) Data Qauntiles Predicted Values

Fig. 10. Model diagnostics for the Clb(t-w) model: (a) cross-validation; (b) normality of residuals; (d) residuals versus predictive values.
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169 167

0.4
Predicted Clb(a-w) (% wt. binder)
1.0 a 0.2 b c

Normal Theoretical Qauantiles

0.3

0.8 0.1 0.2

MSE = 0.004

Residuals
MSPE = 0.004 0.1
0.6
0.0
0.0
0.4 -0.1
-0.1
0.2 -0.2
Training set -0.2
Validation set -0.3
0.0
0.0 0.2 0.4 0.6 0.8 1.0 -0.2 -0.1 0.0 0.1 0.2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Observed Clb(a-w) (% wt. binder) Data Qauntiles Predicted Values

Fig. 11. Model diagnostics for the Clb(a-w) model: (a) cross-validation; (b) normality of residuals; (d) residuals versus predictive values.

Table 10
The estimated admixed chloride binding capacity of different concrete mixtures.

No. Concrete mixture w/b Admixed Cl Clb(t-p) Clb(t-w) Clb(a-w)

Level (% wt. binder)
1 70% OPC + 30% FC 0.42 0.14 89% 69% 74%
2 50% OPC + 50% GGBS 0.55 0.14 91% 63% 65%
3 90% OPC + 10% SF 0.60 0.14 89% 65% 67%

objectives of this example are to first, estimate the binding capac- this case, the ‘‘free” admixed chlorides (determined from total and
ity of each concrete mixture using the developed models, and sec- pore-pressed) meet both the water-soluble and acid-soluble chlo-
ond, considering the estimated binding capacity, determine what ride limits.
level of admixed chlorides would meet the allowable admixed This is a confusing situation that can lead to additional testing,
chloride limits specified by ACI 222R-01. reconstruction costs, maintenance costs, and possible litigation,
The last three columns in Table 10 show the estimated admixed but most importantly, leads to confusion within the industry.
chloride binding capacity of the three concrete mixtures using the These results highlight the need for revising the current specified
models developed in this study. Results show that water- and acid- allowable admixed chloride limits. According to ACI 222R-001,
soluble tests can result in a significant underestimation of the the ratio of allowable water-to-acid-soluble admixed chlorides is
binding capacity. For example, for the mixture containing 30% Type 0.75–0.80 [63]. These ratios suggest that ACI 222R-001 considers
C fly ash, Clb(t-w) and Clb(a-w) are about 22% and 17% less than Clb(t-p), the binding capacity of cementitious systems to be 20%–25%.
respectively. For the mixture containing slag, Clb(t-w) and Clb(a-w) are Results of this study, however, show that the Clb(a-w) can range
approximately 31% and 29% less than Clb(t-p), respectively. Simi- from 23% to 97%. If the binding capacity of cementitious systems
larly, for the mixture containing silica fume, Clb(t-w) and Clb(a-w) is greater than 25%, it can be concluded that the reported allowable
are approximately 27% and 25% less than Clb(t-p). Assuming that admixed chloride limits values in ACI 222-R01 are very
Clb(t-p) represents the actual binding capacity (or as close as it can conservative.
be determined), these results indicate that using water-soluble On the other hand, if the binding capacity of cementitious sys-
and acid-soluble testing procedures can lead to significant error tems is less than 25%, then the allowable admixed chloride limits
in estimating the amount of free and bound chlorides. per ACI 222-R01 are not conservative values. In addition, it is more
ACI 222R-01 specifies the maxim allowable chloride limits, CA, realistic to use the pore pressed chlorides when determining the
based on both water- and acid-soluble testing procedures follow- binding capacity. Considering that the concentration of bound
ing ASTM C1218 and ASTM C1152, respectively. Moreover, ASTM chlorides estimated using the pore-pressed chlorides is higher than
C1152 equates acid-soluble chlorides to ‘‘total” chlorides, which the concentration of bound chlorides estimated using the water-
is not the case according to the results from this research. The soluble testing, care must be taken when water- and acid-soluble
water-soluble CA limits in ACI 222R-01 are 0.15% for dry conditions, tests are used for setting allowable admixed chloride limits (i.e.,
0.08% for wet conditions, and 0.06% for prestressed concrete. Using a factor must be used). Currently, there is no consensus on which
acid-soluble testing, ACI 222R-01 specifies 0.2% for dry conditions, test should be used to measure the concentration of free chlorides.
0.1% for wet conditions, and 0.08% for prestressed conditions. In fact, results from this research show that the estimates of bound
Assuming that the concrete mixtures shown in Table 10 are located chlorides are very dependent on the test method and different val-
in the wet condition, and the concentration of total chlorides in ues of bound chlorides can be estimated using different testing
these mixtures is equal to the concentration of admixed chlorides, procedures. Because water- and acid-soluble testing are the most
it can be concluded that acid-soluble testing of the concrete mix- common tests that are typically performed in the industry, care
tures in Table 10 would yield 0.13%, whereas water-soluble testing must be taken to adjust for the differences between the water-
would yield 0.07%. Based on these results, the amount of admixed soluble and pore-press test results.
chlorides will meet the ACI 222R-01 water-soluble requirement
but not the acid-soluble requirement. On the other hand, by con- 6. Conclusion
sidering the binding capacity of cementitious systems determined
using Clb(t-p), only 10 to 11 percent of admixed chlorides will be in This study investigated the effects of w/b, SCM type, and SCM
the form of free chlorides (i.e., 0.014–0.015 percent wt. binder). In level on the binding capacity of paste samples containing different
168 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169
concentrations of admixed chlorides. The concertation of chlorides
after 56 days was measured using water-soluble, pore-pressed,
and acid-soluble test methods. Results showed that the chloride
testing method is a significant factor influencing the chloride mea-
surement results and the binding capacity of cementitious systems.
The findings of this study indicated that on average the pore-press
and water-soluble chloride testing (per ASTM C1218) can measure
up to 46% and 59% of acid-soluble chlorides, respectively. Moreover,
up to 94% of the total chlorides can be estimated using the acid-
soluble testing per ASTM C1152. Results showed that the Langmuir
isotherm can be used to explain the relationship between free and
bound chlorides in OPC samples containing admixed chlorides.
Statistical analysis of the experimental data suggested a three-
way interaction between w/b, SCM level, and the admixed chloride
level in this study. Using the experimental results, three predictive
models were proposed that estimate the concentration of bound
chlorides in terms of percent by weight of the binder. As opposed
to the existing binding isotherm models that require the concentra-
tion of free chlorides to estimate the concentration of bound chlo-
rides, the proposed models in this study do not require extensive
testing and can be used by researchers and engineers in the field.
Although the models in this study were developed for paste
samples containing specific SCMS at specified levels, the results
of this study can still be of value to researchers and industry. This
study indicates that the current allowable admixed chloride limits
specified by the ACI documents should be revised to accommodate
for differences in the binding capacity of cementitious systems. The
current ACI documents assume that the water-to-acid-soluble
chloride ratio to be 0.75–0.80. However, the results of this study
show that this ratio can vary depending on the type of SCM, SCM
level, and w/b. To reduce the financial and engineering risks asso-
ciated with the corrosion of reinforcing steel, there is a need to
specify one test method for measuring the concentration of chlo-
rides in the concrete pore solution. Standards and specifications
must consider the influential role of admixed chloride binding
when specifying allowable admixed chloride limits.
Conflict of interest
None.
References
[1] M. Shakouri, D. Trejo, A time-variant model of surface chloride build-up for
improved service life predictions, Cem. Concr. Compos. 84 (2017) 99–110.
[2] M. Shakouri, D. Trejo, P. Gardoni, A risk-based model for determining allowable
admixed chloride limits in concrete, in International RILEM Conference on
Materials, Systems and Structures in Civil Engineering: Conference segment on
service life of cement-based materials and structures. 2016: Lyngby, Denmark.
p. 631–640.
[3] M. Shakouri, D. Trejo, P. Gardoni, A probabilistic framework to justify allowable
admixed chloride limits in concrete, Constr. Build. Mater. 139 (2017) 490–500.
[4] D. Trejo, V. Mazarei, J.H. Ideker, O.B. Isgor, Influence of alkali-silica reaction
reactivity on corrosion in reinforced concrete, ACI Mater. J. 144 (5) (2017).
[5] V. Mazarei, D. Trejo, J.H. Ideker, O.B. Isgor, Synergistic effects of ASR and fly ash
on the corrosion characteristics of RC systems, Constr. Build. Mater. 153 (2017)
647–655.
[6] D. Trejo, M. Shakouri, N.P. Vaddey. The Need for Standardized Testing of Input
Variables for Reliable Service Life Prediction of Reinforced Concrete Structures,
in: International Conference on Advances in Construction Materials and
Systems. 2017. Chennai, Inida.
[7] M. Shakouri, D. Trejo, Estimating the critical chloride threshold of reinforcing
steel in concrete using a hierarchical Bayesian model, Sust. Resil. Infra. (2017)
1–21.
[8] U. Angst, B. Elsener, C.K. Larsen, Ø. Vennesland, Critical chloride content in
reinforced concrete–a review, Cem. Concr. Res. 39 (12) (2009) 1122–1138.
[9] Q. Yuan, C. Shi, G. De Schutter, K. Audenaert, D. Deng, Chloride binding of
cement-based materials subjected to external chloride environment–a review,
Constr. Build. Mater. 23 (1) (2009) 1–13.
[10] A. Suryavanshi, J. Scantlebury, S. Lyon, Corrosion of reinforcement steel
embedded in high water-cement ratio concrete contaminated with chloride,
Cem. Concr. Compos. 20 (4) (1998) 263–281.
[11] O.A. Kayyali, M.N. Haque, The C1/OH ratio in chloride-contaminated
concrete — a most important criterion, Mag. Concr. Res. 47 (172) (1995)
235–242.
[12] J. Geng, D. Easterbrook, Q.-F. Liu, L.-Y. Li, Effect of carbonation on release of
bound chlorides in chloride-contaminated concrete, Mag. Concr. Res. 68 (7)
(2016) 353–363.
[13] J. Xu, C. Zhang, L. Jiang, L. Tang, G. Gao, Y. Xu, Releases of bound chlorides from
chloride-admixed plain and blended cement pastes subjected to sulfate
attacks, Constr. Build. Mater. 45 (2013) 53–59.
[14] H. Hirao, K. Yamada, H. Takahashi, H. Zibara, Chloride binding of cement
estimated by binding isotherms of hydrates, J. Adv. Concr. Technol. 3 (1)
(2005) 77–84.
[15] M. Thomas, R. Hooton, A. Scott, H. Zibara, The effect of supplementary
cementitious materials on chloride binding in hardened cement paste, Cem.
Concr. Res. 42 (1) (2012) 1–7.
[16] L. Sutter, T. Dam, K. Peterson, D. Johnston, Long-term effects of magnesium
chloride and other concentrated salt solutions on pavement and structural
portland cement concrete: Phase I results, Trans. Res. Re.: J. Trans. Res. Board
1979 (2006) 60–68.
[17] M. Florea, H. Brouwers, Chloride binding related to hydration products: part I:
ordinary Portland cement, Cem. Concr. Res. 42 (2) (2012) 282–290.
[18] M.N. Haque, O.A. Kayyali, Free and water soluble chloride in concrete, Cem.
Concr. Res. 25 (3) (1995) 531–542.
[19] T. Luping, L.-O. Nilsson, Chloride binding capacity and binding isotherms of
OPC pastes and mortars, Cem. Concr. Res. 23 (2) (1993) 247–253.
[20] C. Arya, N. Buenfeld, J. Newman, Factors influencing chloride-binding in
concrete, Cem. Concr. Res. 20 (2) (1990) 291–300.
[21] C. Arya, N. Buenfeld, J. Newman, Assessment of simple methods of determining
the free chloride ion content of cement paste, Cem. Concr. Res. 17 (6) (1987)
907–918.
[22] G. Glass, Y. Wang, N. Buenfeld, An investigation of experimental methods used
to determine free and total chloride contents, Cem. Concr. Res. 26 (9) (1996)
1443–1449.
[23] P.F. McGrath, Development of test methods for predicting chloride penetration
into high performance concrete. 1997: National Library of Canada =
Bibliothèque nationale du Canada.
[24] T. Mohammed, H. Hamada, Relationship between free chloride and total
chloride contents in concrete, Cem. Concr. Res. 33 (9) (2003) 1487–1490.
[25] ASTM C1218, Standard Test Method for Water-Soluble Chloride in Mortar and
Concrete. 2015, ASTM International: West Conshohocken, PA.
[26] T. Rilem, 178-TMC. Analysis of chloride content in concrete, Mater. Struct. 35
(2002) 583–585.
[27] AASHTO T260, Standard method of test for sampling and testing for chloride
ion in concrete and concrete raw materials. American Association of State
Highway and Transportation Offi cials.
[28] American Concrete Institute Committee 222, Provisional Standard Test
Method for Water-Soluble Chloride Available for Corrosion of Embedded
Steel in Mortar and Concrete Using the Soxhlet Extractor. 1996, ACI 222.1–96.
[29] J. Tritthart, Chloride binding in cement II. The influence of the hydroxide
concentration in the pore solution of hardened cement paste on chloride
binding, Cem. Concr. Res. 19 (5) (1989) 683–691.
[30] G. Blunk, On the distribution of chloride between the hardening cement paste
and its pore solution, 8th Int’l Cong. Chem. Cement (1986) 85–90.
[31] R. Dhir, M. Jones, M. McCarthy, PFA concrete: chloride-induced reinforcement
corrosion, Mag. Concr. Res. 46 (169) (1994) 269–277.
[32] M.J. Kim, K.B. Kim, K.Y. Ann, The influence of C3A content in cement on the
chloride transport, Adv. Mater. Sci. Eng. 2016 (2016).
[33] S.E. Hussain, S. Al-Saadoun, Effect of cement composition on chloride binding
and corrosion of reinforcing steel in concrete, Cem. Concr. Res. 21 (5) (1991)
777–794.
[34] A. Suryavanshi, J. Scantlebury, S. Lyon, The binding of chloride ions by sulphate
resistant Portland cement, Cem. Concr. Res. 25 (3) (1995) 581–592.
[35] T. Sumranwanich, S. Tangtermsirikul, A model for predicting time-dependent
chloride binding capacity of cement-fly ash cementitious system, Mater.
Struct. 37 (6) (2004) 387.
[36] V.S. Ramachandran, R. Seeley, G. Polomark, Free and combined chloride in
hydrating cement and cement components, Mater. Struct. 17 (4) (1984) 285–
289.
[37] V.G. Papadakis, Effect of supplementary cementing materials on concrete
resistance against carbonation and chloride ingress, Cem. Concr. Res. 30 (2)
(2000) 291–299.
[38] V. Baroghel-Bouny, X. Wang, M. Thiery, M. Saillio, F. Barberon, Prediction of
chloride binding isotherms of cementitious materials by analytical model or
numerical inverse analysis, Cem. Concr. Res. 42 (9) (2012) 1207–1224.
[39] M. Saillio, V. Baroghel-Bouny, F. Barberon, Chloride binding in sound and
carbonated cementitious materials with various types of binder, Constr. Build.
Mater. 68 (2014) 82–91.
[40] J.A. Jain, N. Neithalath, Chloride transport in fly ash and glass powder modified
concretes–influence of test methods on microstructure, Cem. Concr. Compos.
32 (2) (2010) 148–156.
[41] J. Liu, Q. Qiu, X. Chen, F. Xing, N. Han, Y. He, Y. Ma, Understanding the
interacted mechanism between carbonation and chloride aerosol attack in
ordinary Portland cement concrete, Cem. Concr. Res. 95 (2017) 217–225.
[42] H. Ye, X. Jin, C. Fu, N. Jin, Y. Xu, T. Huang, Chloride penetration in concrete
exposed to cyclic drying-wetting and carbonation, Constr. Build. Mater. 112
(2016) 457–463.
 D. Trejo et al. / Construction and Building Materials 189 (2018) 157–169
[43] X. Zhu, G. Zi, Z. Cao, X. Cheng, Combined effect of carbonation and chloride
ingress in concrete, Constr. Build. Mater. 110 (2016) 369–380.
[44] A.K. Suryavanshi, R.N. Swamy, Stability of Friedel’s salt in carbonated concrete
structural elements, Cem. Concr. Res. 26 (5) (1996) 729–741.
[45] Y. Kato, S. Naomachi, E. Kato, Changes in chloride penetration properties
caused by reaction between sulfate ions and cement hydrates, International
Conference on Concrete Sustainability, International Center for Numerical
Methods in Engineering Madrid, Spain, 2015.
[46] K. Byfors, C.M. Hansson, J. Tritthart, Pore solution expression as a method to
determine the influence of mineral additives on chloride binding, Cem. Concr.
Res. 16 (5) (1986) 760–770.
[47] W. Holden, C. Page, N. Short, The influence of chlorides and sulphates on
durability, Corros. Reinforcement Concr. Constr. (1983) 143–150.
[48] Z. Shi, M.R. Geiker, K. De Weerdt, T.A. Østnor, B. Lothenbach, F. Winnefeld, J.
Skibsted, Role of calcium on chloride binding in hydrated Portland cement–
metakaolin–limestone blends, Cem. Concr. Res. 95 (2017) 205–216.
[49] A. Delagrave, J. Marchand, J.-P. Ollivier, S. Julien, K. Hazrati, Chloride binding
capacity of various hydrated cement paste systems, Adv. Cem. Based Mater. 6
(1) (1997) 28–35.
[50] O. Wowra, M. Setzer, M. Setzer, R. Auberg, Sorption of chlorides on hydrated
cements and C3S pastes, Frost Resist. Concr. (1997) 146–153.
[51] A. Dousti, M. Shekarchi, Effect of exposure temperature on chloride-binding
capacity of cementing materials, Mag. Concr. Res. 67 (15) (2015) 821–832.
[52] J. Xu, W. Feng, L. Jiang, Y. Xu, Y. Song, Y. Cao, Q. Tan, Influence of surfactants on
chloride binding in cement paste, Constr. Build. Mater. 125 (2016) 369–374.
[53] H. Ye, N. Jin, X. Jin, C. Fu, W. Chen, Chloride ingress profiles and binding
capacity of mortar in cyclic drying-wetting salt fog environments, Constr.
Build. Mater. 127 (2016) 733–742.
169
[54] C. Page, P. Lambert, P.R. Vassie, Investigations of reinforcement corrosion. 1.
The pore electrolyte phase in chloride-contaminated concrete, Mater. Struct.
24 (4) (1991) 243–252.
[55] M. Roberts, Effect of calcium chloride on the durability of pre-tensioned wire
in prestressed concrete, Mag. Concr. Res. 14 (42) (1962) 143–154.
[56] A.K. Suryavanshi, J.D. Scantlebury, S.B. Lyon, Mechanism of Friedel’s salt
formation in cements rich in tri-calcium aluminate, Cem. Concr. Res. 26 (5)
(1996) 717–727.
[57] ASTM C1152, Standard Test Method for Acid-Soluble Chloride in Mortar and
Concrete. 2004, ASTM International: West Conshohocken, PA.
[58] ASTM C305, Standard Practice for Mechanical Mixing of Hydraulic Cement
Pastes and Mortars of Plastic Consistency. ASTM International, 2014.
[59] D. Trejo, A.A. Ahmed, Assessing the automation of the ASTM C1152 test for
rapid determination of chloride concentrations of ordinary portland cement
systems. Journal of Testing and Evaluation, 2018. Under Review.
[60] G.K. Glass, N.R. Buenfeld, The influence of chloride binding on the chloride
induced corrosion risk in reinforced concrete, Corros. Sci. 42 (2) (2000) 329–
344.
[61] J. Neter, M.H. Kutner, C.J. Nachtsheim, W. Wasserman, Applied linear statistical
models. Vol. 4. 1996: Irwin Chicago.
[62] P. Gardoni, A. Der Kiureghian, K.M. Mosalam, Probabilistic capacity models and
fragility estimates for reinforced concrete columns based on experimental
observations, J. Eng. Mech. 128 (10) (2002) 1024–1038.
[63] P. Vaddey, D. Trejo, M. Shakouri, Factors influencing chloride test results of
cementitious systems, ACI J. Mater. (2018). in press.