Construction and Building Materials 152 (2017) 192–205

Contents lists available at ScienceDirect

Construction and Building Materials

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

Simultaneous effects of microsilica and nanosilica on self-consolidating

concrete in a sulfuric acid medium
Ali Hendi a,⇑, Hamid Rahmani b, Davood Mostofinejad c, Ali Tavakolinia d, Mohsen Khosravi e
a
Civil Engineering Department, Islamic Azad University-Khomeinishahr Branch, Isfahan, Iran
b
Civil Engineering Department, University of Zanjan, Zanjan, Iran
c
Department of Civil Engineering, Isfahan University of Technology (IUT), Isfahan, Iran
d
Civil Engineering Department, Islamic Azad University-Estahban Branch, Shiraz, Iran
e
Department of Nanotechnology, University of Isfahan, Isfahan, Iran

h i g h l i g h t s g r a p h i c a l a b s t r a c t

 Seven percent substitution of micro-

silica has the same effect as 2 percent
of nano-silica.
Fresh concrete tests and samples

 For the first time RS is used in the

Curing
preparation

field of concrete technology.

 In contrast to micro-silica, nano-silica
has no effect on residual compressive Different tests on exposed
concrete at specified ages
Submerging in sulfuric acid
medium

strength.
 Concrete performance and durability
are considerably improved by using
micro- and nano-silica Experimental results and
discussions

simultaneously. ANN and PSO analysis on the

experimental results

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

Article history: Combination of microsilica and nanosilica (colloidal silica) are considered to design a high strength self-
Received 8 November 2016 consolidating concrete to resist in the sulfuric acid medium. Artificial intelligence was used to predict and
Received in revised form 24 June 2017 compare the behavior of these two pozzolans in a sulfuric acid medium. Contour plots were used to
Accepted 28 June 2017
investigate the products combination better. Thermogravimetric analysis (TGA) was also used to find
the calcium hydroxide range while using these two pozzolans. TGA revealed that colloidal silica did
not contribute to cement hydration within seven days of curing while a combination of them boosted cal-
Keywords:
cium hydroxide consumption. The results show that more substitution of the pozzolans could lead to
Microsilica
Nanosilica
lower mass loss while nanosilica has marginal effect on the residual compressive strength. The results
Sulfuric acid also revealed that 7 percent substitution of microsilica showed the same effect as 2 percent nanosilica
Artificial intelligence replacement.
Thermogravimetric analysis Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

⇑ Corresponding author at: #20, Shahid Moslem Esmaeili Alley, Sheikh Safi St.,
Bozorgmehr Ave., Isfahan, Iran.
E-mail addresses: ali_hendi@ymail.com, ali.hendi@iaukhsh.ac.ir (A. Hendi),
hrahmani@znu.ac.ir (H. Rahmani), dmostofi@cc.iut.ac.ir (D. Mostofinejad),
tavakolinia_ali@yahoo.com (A. Tavakolinia), m.khosravi@ast.ui.ac.ir (M. Khosravi).
Vibration is a substantial need to compact a normal concrete
(NC) and labors skill can significantly impact the final compaction.
Self-compacting concrete (SCC) has been introduced to take out the
labors hand from the compacting and the consequent effects on

http://dx.doi.org/10.1016/j.conbuildmat.2017.06.165
0950-0618/Ó 2017 Elsevier Ltd. All rights reserved.
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 193

concrete durability. The SCC necessity was first proposed by Oka-
mura in 1986 and a prototype of this kind of concrete was made
by Ozawa in 1988 [1]. The fresh mortar properties, volume of
coarse aggregates as well as the plasticizer are three important
characteristics in the SCC production [2]. Therefore, investigations
on SCC mix designs started on packing density field for aggregates
[3–5], based on continuous packing models [6] and/or discrete
ones [7,8] to produce powder type of SCC [9]. As SCC microstruc-
tures are different from NC, they define their distinct durability cri-
teria and thus their durability should be investigated separately.
Three main sources of sulfuric acid are depots containing sulfur
(usually FeS2) [10] in stockpiles, crown of sewage pipes and acidic
rains which could have deleterious effects on infrastructures.
Investigations on the microbiological sulfuric acid attack started
in 1945 when Parker introduced corrosion process with Thiobacil-
lus bacteria [11]. After Parker’s discovery, many efforts were car-
ried out to understand corrosion process and the results revealed
the metabolic effects of thiobacilli [12].
Generally, three techniques of visual inspection, mass-loss and
compressive strength-loss are used to assess the corrosion process
[13–16]. Microsilica is a widely used pozzolan whose effects on
concretes exposed to a sulfuric acid medium does not follow a gen-
eral pattern. For instance, some studies [10,16–22] have shown
either enhancing or neutral effects while some other studies [23–
25] have revealed contradictory results. Such behaviors have been
shown in changing parameters, such as the water to cement ratio
[26–28] and the use of different types of Portland cement
[10,20,21,29]. Nanoparticles are widely used in studies [30,31] to
enhance the concrete properties where nanoparticles act as a cen-
ter of crystallization and also consuming Ca(OH)2 [32]. Nanosilica
can reduce the corrosion of concretes under sulfuric acid medium
(pH = 2) [33,34]. The main concern in using nanosilica is its uni-
effects. Third, the non-linear regression (NLR) as a multi-
dimensional analysis was implemented to distinguish every
parameter effect with preliminary assumptions. Therefore, analyt-
ical data were predicted using recommender systems (RS) for a
certain amount of substitution. Finally, to reduce the amount of
preliminary assumptions, an artificial neural network (ANN) was
used to predict the mass-loss values and the root-mean-square-
error (RMSE) was calculated to determine the accuracy of the
model. The ANN and NLR are widely used in different aspects of
concrete properties [16,42,43] and the accuracy of these two pre-
diction methods was compared to find the more reliable one.
2. Materials
The following materials were used in this study:
2.1. Cement
Ordinary Portland cement Type I in accordance with ASTM C 150 [44] was used
to make the samples. The chemical composition of the cement and its physical
properties are illustrated in Table 1.
2.2. Aggregates
Two types of calcic sand (Sand 0–4 and sand 0–8) and one type of calcic coarse
aggregate (CA) were chosen to have a wide variety of particle size distribution (PSD)
in the aggregates (Sand 0–8 was out of ASTM C33 [45] limitations). Physical prop-
erties of the aggregates are provided in Table 2. ASTM C 136-01 [46] was imple-
mented for aggregate grading. Sample preparation was consistent with ASTM C
702 [47] using B and C methods. Specific gravity of the coarse and fine aggregates was
obtained in accordance with ASTM C 127 [48] and ASTM C 128 [49], respectively.
2.3. Additives
Two types of additives were used as a partial cement replacement. The first one
was Microsilica (MS) with the average particle size of 0.25 lm and the relative

form dispersion. The agglomeration of nanoparticles causes a lot

Table 2
of defects especially in high pressure [35]. Nanosilica as a low- Physical properties of the aggregates (passing percentage).
cost nanoparticle results in the higher pozzolanic activity and the
compressive strength and the lower concrete permeability even Sieve number (size) Sand 0–4 Coarse aggregate Sand 0–8

by low contents of substitution [36]. Colloidal nanosilica has been 3/4 (19.06) 100 100 100
reported more effective than microsilica [37–40], except that some 1/2 (12.7) 100 98.32 100
3/8 (9.53) 100 58.65 99.44
researchers ([41,37]) believe that nanosilica particles result in a 1/4 (6.35) 99.97 6.18 92.11
lower compressive strength and a higher capillary pores. 3/16 (4.76) 99.96 3.44 80.18
The behavior of concrete in an acidic solution is related to mul- 8 (2.36) 87.65 2.38 59.45
tiple factors and drawing a conclusion based on one parameter is 16 (1.18) 54.97 1.97 39.48
30 (0.6) 32.75 1.94 25.53
fallible. Interpreting all of the factors together with usual statistical
50 (0.3) 22.89 1.83 18.35
analyses is not possible. In this study, four methods for interpreting 100 (0.15) 8.58 0.8 7.08
results were used. First, an analysis of variance (ANOVA) and 150 (0.11) 0.47 0.26 2.22
Spearsman’s correlation were used to identify the effective param- 200 (0.08) 0.37 0.11 1.26
eters. Second, the related plots were drawn to show the correla- Specific gravity 2690 2710 2680
*
tions. These two analyses cover the relationship between SSD (%) 3 1 3
separated input parameters and cannot clarify the combining *
Saturated surface dry.

Table 1
Chemical compositions of the cement and microsilica and physical properties of the
cement.

Chemical SiO2 Al2O3 Fe2O3 CaO MgO Na2O K2O All of alkali SO3 Insoluble residue
Composition content
Cement (%) 26.5 5.70 2.90 61.3 2.60 – – 0.75 2.20 0.40
Microsilica (%) 95.4 1.32 0.87 0.49 0.97 0.31 1.01 – 0.1 –

Physical Characteristics of cement Results

Autoclave expansion (%) 0.05
Compressive strength (MPa) 3 Days 15.0
7 Days 30.0
28 Days 46.0
Specific gravity 3.10
Blaine fineness (m2/kg) 320
Time of setting Initial, minutes 160
Final, minutes 215
194 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205

25

20

Frequency (%)
15

10

0
10 11 13 15 18 20 22 25 30 35 40 45 50
Diameter (nm)

Fig. 1. Particle size distribution of NS.

160

140

120

100
Counts/Au.

80

60

40

20

0
10 20 30 40 50 60 70 80 90
2Theta/Degree

Fig. 2. XRD analysis of NS.

specific gravity of 2.34. The chemical composition of the microsilica is presented in
Table 1. The second one was colloidal silica (nanosilica) (NS) solution with particle
size distribution (PSD) as shown in Fig. 1 and amorphous intrinsic (in accordance to
X-ray diffraction (XRD) which as shown in Fig. 2. The pH for this solution was equal
to 9 and the solid content was 7 percent.
2.4. Superplasticizer
Polycarboxylate based superplasticizer was used in the concrete mixtures with
relative specific gravity of 1.15 and chloride ion content below 0.1%. The values of
solid content and pH were equal to 33% and 7, respectively.
3. Experimental program
3.1. Mixture proportioning
With respect to reference [50], passing contents of sieves num-
bers 4, 8 and 3/8 are between the ranges of 50–60, 35–55 and 65–
80 percent, respectively. Considering the mid-range of these val-
ues, the aggregates proportioning would be 44, 32 and 24 percent
for sand 0–4, sand 0–8 and coarse aggregate (CA), respectively. As
the first attempt, these proportions showed good conformity to
Fuller curve (with the maximum nominal size of 1 cm and RMSE
equal to 2.29). The water to cement ratio was chosen with respect
to modified Feret formula [51] and ACI 237 [52] equal to 0.45. The
Superplasticizer (SP) content was chosen in a way that the changes
in the slump flow value reach the least tolerance by SP variation
and even no segregation [53] taking place. When there are signs
of bleeding on the surface of the paste, the segregation occurs.
Thus, the optimum value was obtained between 1.6 and 2 percent.
The volume percentages of CA, sand, mortar and paste were
obtained equal to 26.3%, 36.7%, 70.2% and 33.5%, respectively.
The first batch of this concrete was made but did not satisfy the
SCC considerations [54] perfectly. Therefore, the water to cement
ratio decreased and the powder phase increased to reach
83 ± 3 cm slump flow with 0.23 water to cement ratio. It should
be noted that the slump flow for normal SCC mixture was 83 cm
and adding other admixtures reduced the value to 80 cm. The final
composition of the concrete and its properties are given in Table 3.
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 195

Table 3
Final mix design properties.

Slump flow T500 Viscosity Stability Index L-box V-funnel CA Sand Mortar Paste
(centimeter) (Second) (VSI) (Second) (%)
Values [52] 83 3.4 0 1 22.8 29 35% 69% 34%
Acceptable range More than 65 More than 2 Less than 3 More than Less than 25 + 2 27– 48– – 30–
[54,55] 0.8 36 55 38

The difference between the parameters in Table 3 are all accept-
able because the separating sieve between coarse and fine aggre-
gates was different from ACI [52] and other references [54,55].
The current mix design had good conformity to ACI 237 [52].
3.2. Mix designs and test program
All of the specimens were prepared using ASTM C 192 [56] but
in 10  10  10 (cm) cubic molds and without vibration. Six
10  10  10 (cm) specimens were prepared for the compressive
strength test at the age of 14, 28 and 90 days; 8 for sulfuric acid
test; two 15  15  15 (cm) samples to investigate volume of per-
meable pores according to ASTM C 642 [57]; and four for Thermo-
gravimetric (TG) test. Colloidal nano-silica (NS) was stirred with
water and superplasticizer (SP) for ten minutes with hand before
adding to the concrete mixture for the later five-minute mixing.
The mix designs are illustrated in Table 4. It should be noted that
colloidal silica is a transparent liquid and adding superplasticizer
changes it to the cloudy substance (agglomeration indication).
Ninety-nine percent purified industrial sulfuric acid was added
to the water to reach a constant pH equal to 1.0 (1% concentration).
All of the specimens were weighed before submerging in the sulfu-
ric acid solution and then were weighed at the ages of 7, 14, 21, 30,
44, 60, 90, 120 and 150 days after being ‘‘machine-brushed” and
‘‘not-being-brushed” states in the saturated surface dry (SSD) con-
dition. The mean dimensions of the specimens were measured and
reported after 44, 60, 90, 120 and 150 days of exposure. The
specimens were capped in accordance with ASTM C 617 [58] for
the compressive strength test after 30, 75 and 150 days of expo-
sure to investigate their compressive strength loss.
4. Results and discussions
The SCC specific tests were similarly carried out on the SCC mix
designs with different amounts of the additives. The constant PSD
of the aggregates regarding the Fuller curve created similar packing
density for the mixtures. Therefore, the concrete class [54] did not
change for all of the mixtures. Variations for the slump, V-funnel
and T500 were between 80–83 cm, 15–25 s and 2–3.5 s, respec-
tively. The properties of the concretes including density, water
absorption, and volume of the permeable pores in hardened state
were examined in accordance with ASTM C 642 [57] and are shown
in Table 5. Figs. 3 and 4 provide an overview of the mass loss per-
centage of the brushed and not-brushed specimens exposed to
H2SO4 solution, respectively. The average residual compressive
strengths (RCS) of the three specimens after exposing to sulfuric
acid solution for 30, 75 and 150 days are illustrated in Table 6.
Dimension loss for the specimens are provided in Table 7. The fluc-
tuating behavior of these results suggests that the dimension loss
of the specimens and consequently visual inspections would not
supply enough evidence to make a decision for the corrosion pro-
cess. Fig. 5 provides clear evidence to prove this claim.
Correlations between distinct concrete characteristics are drawn
out by applying ’’Spearsman’s correlation coefficient’’ using SPSS
software. Henceforth, all of the relevant characteristics are

Table 4
Mixtures proportions (kg).

Mixture Name Cement Water SP NS (%) MS (%)

N 580 174 10.44 0 (0) 0 (0)
NS0.3 578.26 174 10.44 1.74 (0.3) 0 (0)
NS1 574.2 174 10.44 5.8 (1) 0 (0)
NS2 568.4 174 10.44 11.6 (2) 0 (0)
MS7 539.4 174 10.44 0 (0) 40.6 (7)
MS6.7NS0.3 539.4 174 10.44 1.74 (0.3) 38.86 (6.7)
MS5NS2 539.4 174 10.44 11.6 (2) 29 (5)
MS5NS1 545.2 174 10.44 5.8 (1) 29 (5)
MS7NS1 533.6 174 10.44 5.8 (1) 40.6 (7)
MS9NS1 522 174 10.44 5.8 (1) 52.2 (9)

Table 5
Hardened concrete properties.

Mixture Name Water absorption% (2 days) Water absorption% (90 days) Volume of permeable pores (voids)% Compressive Strength (MPa)
14 Days 28 Days 90 Days
N 3.13 3.63 8.96 58.4 62.3 80.5
NS0.3 3.53 5.14 9.62 51 60.1 86
NS1 3.25 3.99 8.45 60.3 68.7 82.7
NS2 3.45 3.6 8.9 58.5 62.8 83.7
MS7 3.72 3.8 9.85 59.9 63.9 83.7
MS6.7NS0.3 3.25 3.5 8.44 48.9 55.6 82.8
MS5NS2 3.35 4.06 8.38 65.7 72.7 83.9
MS5NS1 3.82 3.76 9.54 66.7 69.9 84.2
MS7NS1 3.78 4.01 9.2 67.1 70 84.3
MS9NS1 3.49 3.85 8.57 59 66.7 83.2
196 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205

0.2

0.15

Mass-loss (%)
0.1

N Ns2 MS7
0.05
MS5-NS2 NS1 NS0.3
MS6.7-NS0.3 MS5-NS1 MS9-NS1
MS7-NS1
0

-0.05
0 30 60 90 120 150
Exposure time (days)

Fig. 3. MassLoss versus time (brushed). Fig. 4. Mass-loss versus time (not brushed).

abbreviated according to Table 8. The related parameters are distin-
guished from others by ‘‘TRUE” term in Table 9. The results desig-
nated by ‘‘TRUE” in this Table do not necessarily mean a logical
and definitive relationship between them. The parameters with no
specific relations to each other are designated by ‘‘FALSE” instead.
In what follows, the results reported in Table 9 are analyzed and
interpreted. The combined effect of NS (nano-silica) and MS
(microsilica) cannot be taken into account while using Spearsman
and using two-way ANOVA (analysis of variance) is one of the best
ways to analyze normal data. Correlations are investigated first;
then ANOVA and two-way ANOVA are used to find the most effective
combination of the mixtures. Finally, contour plots were drawn to
investigate the combined effect of NS and MS.
4.1. Effects of microsilica and nanosilica
There is no relationship between MS and CS90 (Compressive
Strength after 90 days of curing) as shown in Table 9. It means
that the compressive strength of other specimens grew up to
reach the compressive strength of MS specimens. According to
Fig. 6, the combination of MS and NS reaches the highest com-
pressive strength at 28 days of curing, but substitution of only
0.3% NS has the most effect after 90 days of curing. The low
strength behavior of the samples containing 0.3% NS at the early
ages can be related to some possible factors. First, the volume of
the nucleuses is less than its critical value and this may result in
discrete formation of CASAH [59]. Also, the agglomerated NS
content form loose masses result in instability at transitional
zones. 0.3% NS is not enough to fill the spaces between the
cement particles, therefore the porous matrix formed at the early
ages.
The effect of MS and NS on RCS (Residual compressive strength),
RCS-loss and Mass change are illustrated in Fig. 7. The mass change
and RCS-change were used instead of MassLoss and RCS-loss
deliberately to show the minimums with warmer colors
(because mass-loss is the negative format of the mass change).
Interpretation about the effect of NS and MS are taken into account
after ANN analysis. It is, however, obvious that the combined effect

Table 6
Residual compressive strength.

Mixture 30 Days Compressive Strength Loss 75 Days Compressive Strength Loss 150 Days Compressive Strength Loss
Name (MPa) percentage (RCS-Loss) (MPa) percentage (RCS-Loss) (MPa) percentage (RCS-Loss)
N 41.6 33.2 33 47.1 26.3 57.9
NS0.3 39 35.2 39.8 33.8 41 31.9
NS1 47.8 30.5 38.9 43.4 38 44.7
NS2 49.7 21 39 38 33.8 46.2
MS7 55 13.9 49.5 22.6 44.1 31
MS6.7NS0.3 47 15.6 47.4 14.8 47.4 14.9
MS5NS2 50.1 31.1 44 39.5 39 46.4
MS5NS1 64.2 8.2 50.7 27.5 47.6 31.9
MS7NS1 66.6 4.9 63 10.1 60.5 13.6
MS9NS1 56.4 15.5 53 20.6 51.8 22.4

Table 7
Dimension-loss (%).

Name Exposure time (days)

44 60 90 120 150
N 0.00% 0.30% 5.30% 5.40% 6.40%
NS2 0.00% 0.00% 2.80% 2.90% 5.90%
MS7 0.00% 1.80% 2.60% 1.30% 5.10%
MS5NS2 0.00% 1.20% 2.60% 3.80% 4.90%
NS1 0.00% 0.30% 3.40% 5.70% 4.00%
NS0.3 0.00% 0.40% 2.70% 6.70% 5.10%
MS6.7NS0.3 0.00% 3.20% 4.90% 5.90% 7.30%
MS5NS1 0.00% 0.60% 1.80% 5.30% 4.20%
MS9NS1 0.00% 0.50% 4.50% 6.50% 8.40%
MS7NS1 0.00% 1.20% 4.70% 3.70% 11.10%
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 197

Fig. 5. Visual inspection of specimens.

Table 8 null hypothesis, using one-way ANOVA. This could be a direct

Abbreviation descriptions.
result of combination of ‘‘DOE” and ‘‘type of the mixtures” at the
Abbreviation Description same time in the ANOVA analysis which is related to concentration
term of data in the initial days of exposure. Therefore, a two-way-
ANN Artificial Neural Network ANOVA was used to show the differences more clearly according
ANOVA Analysis of variance to post hoc Tukey analysis [60] (see Table 10). It should be noted
B Brushed that all of the specimens are compared to each other and the sig-
CS14 Compressive Strength after 14 days of curing
CS28 Compressive Strength after 28 days of curing
nificant differences are shown by ‘‘TRUE” statement and ‘‘FALSE”
CS90 Compressive Strength after 90 days of curing for the identical behavior between the specimens. This two-way
DOE Days Of Exposure ANOVA shows that NS and MS both together are effective param-
MassLoss Percentage of Mass-Loss ((Mass loss/Initial weight) * 100) eters in resistance to sulfuric acid. This analysis also shows that
MS Microsilica replacement ratio (%)
some different specimens have the same behavior. It is obvious
Name Concrete mixture
NB Not Brushed that NS0.3, NS1 and MS7NS1 show similar behaviors and NS2
NLR Non-Linear Regression and MS7 were also the same. It should be noted that the same
NS Nanosilica replacement ratio (%) results just limited to the trend of mass-loss and other parameters
PI Predictor Importance are not involved in this conclusion. Therefore, more comprehensive
RCS Residual compressive strength
RCS-Loss Percentage of compressive strength loss (((CS28 RCS)/
analysis as ‘‘recommender system” is needed to define the best
CS28) * 100) combination of additives and also derive the unexamined data
RS Recommender system for artificial neural networks analysis.
SP Super-Plasticizer ratio (%)
TGA Thermogravimetric analysis
VOPP Volume Of Permeable Pores (%) after 28 days of curing 4.3. Recommender systems
WAR48 Water Absorption ratio after 48 h (%)
WAR90 Water Absorption ratio after 90 days (%) It is necessary to determine the best combined substitution when
a constant cumulative substitution is investigated. With 7 percent
of NS and MS is much more significant than what each one of them substitution of MS, the best combination of MS and NS can be found
does separately. The minimum RCS-loss occurred for simultaneous using recommender systems (RS) [61]. With respect to the values of
substitution of MS and NS about 6 to 8 percent and 0.5 to 1.5 per- MS and NS for the current mixtures, the mixture properties with ini-
cent, respectively. tial values of MS and NS can be reliably predicted using RS. MS6NS1
with 6 percent of MS and 1 percent of NS is the mixture that can be
4.2. ANOVA located in the ranges of MS5NS2 and MS6.7NS0.3. An OCTAVE pro-
gram has been written with collaborative filtering algorithm to solve
ANOVA analysis is used to see if there is a significant difference this problem. Figs. 8 and 9 show the compressive strength, WAR and
between the mixtures in mass-loss or not. First, no difference was VOPP results derived by RS. It can be concluded that more substitu-
found between statistical populations with respect to not-rejected tion of NS in the combined MS and NS mixtures results in a more

Table 9
Correlations.

MS NS MassLoss DOE CS14 CS28 CS90 WAR48 WAR90 VOPP RCS

MS FALSE FALSE FALSE TRUE TRUE FALSE TRUE TRUE FALSE TRUE
NS FALSE FALSE FALSE TRUE TRUE TRUE FALSE FALSE TRUE FALSE
MassLoss FALSE FALSE TRUE FALSE FALSE FALSE FALSE FALSE FALSE TRUE
DOE FALSE FALSE TRUE FALSE FALSE FALSE FALSE FALSE FALSE TRUE
CS14 TRUE TRUE FALSE FALSE TRUE FALSE TRUE TRUE FALSE TRUE
CS28 TRUE TRUE FALSE FALSE TRUE TRUE TRUE FALSE FALSE TRUE
CS90 FALSE TRUE FALSE FALSE FALSE TRUE TRUE TRUE TRUE FALSE
WAR48 TRUE FALSE FALSE FALSE TRUE TRUE TRUE TRUE TRUE TRUE
WAR90 TRUE FALSE FALSE FALSE TRUE FALSE TRUE TRUE TRUE FALSE
VOPP FALSE TRUE FALSE FALSE FALSE FALSE TRUE TRUE TRUE FALSE
RCS TRUE FALSE TRUE TRUE TRUE TRUE FALSE TRUE FALSE FALSE
198 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205

Fig. 6. Effect of MS and NS on the compressive strength. Fig. 7. Effects of NS and MS on RCS, RCS-loss and.

compressive strength. The RS can be also used to anticipate the miss-

ing data for NLR and ANN analyses. To the best of authors’ knowl-
edge, there is no previous use of RS in concrete technology to find
the missing data and even predicting the concrete behavior. This
could be the turning point in the field of concrete technology using
this algorithm to design concretes in different circumstances and
different durability aspects.
4.4. Non-linear regularized regression and artificial neural networks
analyses
To find the optimum contents of MS and NS under sulfuric acid
attack, NLR or/and ANN could predict concrete behavior, consider-
ing all aspects together beyond the other analysis capacity. An
OCTAVE program was written to calculate non-linear regularized
regression. Applied inputs were considered as B or NB, MS, NS,
CS28, CS90, WAR, VOPP and DOE (inputs) to predict RCS and Mas-
sLoss (outputs). The parameter of Lambda was used to regularize
the equation [62], choosing automatically to minimize the proba-
bility of both overfitting and underfitting conditions. All data were
randomly separated into three data sets including Training data set
(60% of all data), Cross-validation data set (20% of all data) and Test
data set (20% of all data). The adequacy of the data size was
checked by monitoring the Cross-validation data set error and
was evaluated with the Test data set error which showed
non-linear regression (NLR) was reliable to execute on the data
set. The NLR was implemented in accordance with the least possi-
ble integer power and real number power of the variables to
reduce the Cross validation data set error which was evaluated
with the Test data set error. In order to minimize the error of
NLR, gradient descent method was applied. The outputs showed
that the second power of input variables led to the least of the
Root-Mean-Squared-Error (RMSE) equal to 0.97 and 0.89 for
MassLoss and RCS, respectively. An arithmetic sequence was
chosen with initial term of 1 and common difference of successive

Table 10
Two way ANOVA results.

N NS2 MS7 MS5NS2 NS1 NS0.3 MS6.7NS0.3 MS5NS1 MS9NS1 MS7NS1

N TRUE TRUE FALSE TRUE TRUE TRUE TRUE FALSE FALSE
NS2 TRUE FALSE TRUE TRUE TRUE FALSE FALSE FALSE TRUE
MS7 TRUE FALSE TRUE TRUE TRUE FALSE FALSE TRUE TRUE
MS5NS2 FALSE TRUE TRUE TRUE TRUE TRUE TRUE FALSE FALSE
NS1 TRUE TRUE TRUE TRUE FALSE TRUE TRUE TRUE FALSE
NS0.3 TRUE TRUE TRUE TRUE FALSE TRUE TRUE TRUE FALSE
MS6.7NS0.3 TRUE FALSE FALSE TRUE TRUE TRUE FALSE TRUE TRUE
MS5NS1 TRUE FALSE FALSE TRUE TRUE TRUE FALSE TRUE TRUE
MS9NS1 FALSE FALSE TRUE FALSE TRUE TRUE TRUE TRUE TRUE
MS7NS1 FALSE TRUE TRUE FALSE FALSE FALSE TRUE TRUE TRUE
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 199

10

WAR or VOPP (%)

7
WAR48 WAR90 VOPP

3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
NS (%)

Fig. 8. Predicted value for MS6NS1.

85

80

75
Compressive strength (%)

CS90 CS28 CS14

70

65

60

55

50

45
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
NS (%)

Fig. 9. MS6NS1 prediction value for the compressive strength.

members equal to 0.1 in order to find the optimum real number
power for every variable leading to the maximum accuracy. The
outputs showed that the 1.3 and 1.4 power of input variables could
lead to the least of the RMSE equal to 0.88 and 0.89 for MassLoss
and RCS respectively. To find the optimum values for input vari-
ables, the NLR function should be minimized.
Particle swarm optimization (PSO) [63] is one of the most
recent used algorithms which is widely used in the literature
[16,64] for multivariable functions. An Octave program was writ-
ten to use PSO and NLR simultaneously (with 30 to 30 as all of
the variables lower and upper bonds). The solving iteration was
fixed to 20 with population of 10,000 particles. Fig. 10 shows the
iterations convergence to global optimum and Table 11 shows
the optimum values for the inputs.
With respect to Table 11, the optimum values tend to reach the
borders in positive or negative ways. The results show the NLR
function has unlimited extremums which show the impossibility
of minimization. To solve this problem, it is necessary to normalize
the inputs and then derive NLR function once again and add up
variable coefficients to find out the predictors significance men-
tioned by Predictor importance (PI). For the succinctness the values
of the power multiplications for real numbers were omitted and
integer power multiplications are shown in Table 12. In this table
the value of ‘‘predictor importance” (PI) shows the influence of
the related variable to predict the results. The Variable of ‘‘B” is
related to the brushed specimens that would be equal to 1.0 when-
ever brushing process performed. This variable, thus, behaves like a
constant and the high value of PI is related to this behavior. The
sign ( ) is related to the variables that have negative effect on
the results. As shown in Table 12, the values of DOE and WAR90
increased by increasing MassLoss. On the one hand, NS and
WAR90 have the highest effects on MassLoss in positive and nega-
tive ways respectively; but on the other hand, NS marginally has a
negative effect on RCS. The VOPP and MS are the most influential
parameters enhancing RCS. To wrap up, using NS causes MassLoss
reduction while MS enhances RCS.
If the RMSE value of ANN was lower than NLR, that would be
better to consider ANN results. The ANN models ‘‘learn” the pat-
terns of the underlying process from the past data and generalize
the knowledge gained (or mathematical relationships between
200 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205

-850

-860

-870

Value of NLR function -880

-890

-900

-910

-920

-930

-940

-950
0 2 4 6 8 10 12 14 16 18 20
Iteration

Fig. 10. Iterations with PSO to minimize NLR function.

Table 11
NLR function minimizing with PSO.

Iterations Inputs
MS NS B DOE CS28 CS90 WAR48 WAR90 VOPP
1 29.33832 25.64346 29.6495 25.1185 1.0849 14.51646 25.5157 20.2921 27.78194
2 28.02839 28.23646 30 21.8181 14.2132 6.06347 2.92845 14.8579 11.59412
3 30 30 30 27.8181 8.21319 0.06347 0.28199 20.8579 5.594116
4 30 30 30 30 2.21319 6.06347 5.718009 26.8579 11.59412
5 30 30 30 30 0.797591 7.72538 13.32238 30 5.015151
6 30 30 30 30 12.7076 9.76138 22.40687 30 10.08662
7 30 30 30 30 6.7076 15.2239 28.40687 30 7.885137
8 30 30 30 30 10.5893 19.3529 30 30 21.03173
9 30 30 30 30 11.3295 19.8663 30 30 30
10 30 30 30 30 11.18329 28.4465 30 30 30
11 30 30 30 30 17.69354 30 30 30 30
12 30 30 30 30 23.69354 30 30 30 30
13 30 30 30 30 29.69354 30 30 30 30
14 30 30 30 30 30 30 30 30 30
15 30 30 30 30 30 30 30 30 30
16 30 30 30 30 30 30 30 30 30
17 30 30 30 30 30 30 30 30 30
18 30 30 30 30 30 30 30 30 30
19 30 30 30 30 30 30 30 30 30
20 30 30 30 30 30 30 30 30 30

Table 12
NLR results.

Power of variable Inputs multiplications Outputs

Intercept MS NS B DOE CS28 CS90 WAR48 WAR90 VOPP
1 0 0.0546 0.0264 0.0605 0.2837 0.0066 0.0008 0.0031 0.021 0.0005
2 0 0.0017 0.0604 0.0605 0.0011 0.0004 0.0006 0.0222 0.1757 0.0112 MassLoss
1 0.0017 0.0971 0.0085 0.0035 0.3598 0.0721 0.069 0.02 0.0337 0.0031
2 0 0.1421 0.0337 0.0035 0.0016 0.0058 0.0045 0.165 0.2612 0.1743 RCS
Predictor Importance (%) 3.8 6.3 ( )56.9 ( )15.4 0.2 ( )0.1 2.3 ( )14.7 0.3 MassLoss
Predictor Importance (%) 22 ( )2.5 0.4 ( )45 1.2 1.3 5.2 ( )6.1 16.2 RCS
*
( ) sign means the negative effect.
** P
Formula: output = MS * (related multiplication) + MS^2 * (related multiplication) + NS * (related multiplication) + NS^2 * (related multiplication) + . . ..

the input and output data) to predict an output given a new set of
input data from the problem domain [65]. In the field of artificial
intelligence, ANN is the method most widely used in solving engi-
neering problems. It is mainly used as a forecasting model because
it does not require prior knowledge and enjoys a high accuracy
[66]. For the purpose of this study, the feed-forward ANN method
was implemented for predicting and the back-propagation
algorithm was used for error assessment (detailed explanations
of the two may be found in the Ref. No. [67]). To evaluate the accu-
racy of the method as compared to the NLR, RMSE was considered
as a determining criterion. The RMSE value for MassLoss
computation was found to be 0.24, which is much less than the
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 201

corresponding value obtained from NLR analysis. The values for Table 13
RCS and MassLoss were predicted using two hidden layers and a Independent variables importance (%).

12  11  11  1 structure with the hyperbolic tangent and sig- Inputs Importance for MassLoss Importance for RCS
moid activation function. The value of RMSE calculated for RCS pre- B or NB 2.7 0.5
diction by the ANN model was found to be equal to 0.3. Figs. 11 and MS 4.5 13.8
12 show the predicted values of MassLoss and RCS versus their real NS 3.5 1.9
values, respectively. Despite the fact that the best-predicted result CS14 4.5 32
CS28 4.7 11.6
of the ANN model was reached by minimizing it, ANN perception is CS90 11.2 5.9
not limited to real boundaries as what has been observed by NLR. DOE 44.6 2.6
Hence, it would be preferable to consider predictor importance (PI) WAR48 10.2 14.3
rather than minimizing it. Table 13 provides information on the WAR90 6.2 7
VOPP 7.9 10.4
importance of independent variables.
Based on this analysis, an increase in DOE, CS90 and WAR90
leads to an increase in MassLoss values, while an increase in other
variables leads to a decrease MassLoss values. As shown in Table 13, decomposition of cement paste. NS has the slightest effect even
DOE, CS90 and WAR48 are of the paramount importance defining on MassLoss and RCS while MS is much more effective to enhance
MassLoss and CS14 (or CS28), WAR48 and MS play such a role RCS. It shows that the amount of substitution is much more effec-
for RCS prediction. One of the different conclusions derived from tive than the kind of it, considering similar composition of MS and
PI could be minimum DOE participating for RCS prediction. It NS. To wrap up the ANN analysis, compressive strength and water
shows that RCS becomes stable due to parallel hydration and absorption are the most effective variables defining MassLoss.

20

15 R² = 0.9858
Predicted MassLoss (%)

10

-5
-5 0 5 10 15 20 25
Real MassLoss (%)

Fig. 11. Comparison between real Mass loss and predicted Mass loss using ANN.

75

70

65
Predicted RCS (MPa)

60 R² = 0.9765

55

50

45

40

35

30
20 30 40 50 60 70 80
Real RCS (MPa)

Fig. 12. Comparison between real RCS and predicted RCS using ANN.
202 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205
Several studies have reported discrepancies between the mass-
loss and the residual compressive strength, indicating no relation-
ship between these two parameters [10,21,68]. In this study, how-
ever, efforts were made to establish a relationship between NS, MS,
DOE and RCS versus MassLoss, using the ANN model. The assump-
tion was that if a satisfactory correlation was found between the
predicted values and the real measurements, then differences
between MassLoss and RCS would be no more reliable. Using the
5  11  11  1 structure, a correlation of 0.96 was obtained sup-
porting the mentioned claim (Fig. 13).
Regarding the combining effect of MS and NS, The PI of NS and
MS on MassLoss was approximately the same and confirm to pre-
vious ANOVA results revealing the same performance of NS2 and
MS7. There are some combination of MS and NS which could not
provide a good resistance to sulfuric acid medium as the same
behavior of MS5NS2 and N. Therefore, receding a trace from opti-
mum values could not guarantee the performance in sulfuric acid
medium and limitations should be defined before using any addi-
tives. To reach this goal the prediction of MassLoss for 150, 450 and
900 days are depicted in Fig. 14 to clear the tolerances using MS
and NS simultaneously. This figure shows more substitution of
MS and NS results in a lower MassLoss and also the tolerance of
using materials in the range of one would bring about higher val-
ues of corrosion.
4.5. Thermogravimetric test
Thermogravimetric tests were performed on N, NS1, MS7 and
MS7NS1 after 7 days of curing in accordance with ASTM E1131
[69]. Thermogravimetric analyses (TGA) were carried out in Argon
atmosphere from room temperature to 600 °C with a rate of 10 °C/
min. As the testing conditions for all of the specimens were the
same, CaCO3 decomposition were not taken into account and all
of the considerations focused on Ca(OH)2 dehydration. Thermo-
gravimetric analysis (TGA) for cement usually consist of four
stages. The first stage is referred to CASAH and ettringite decom-
position usually occurs between 100 °C and 180 °C [70]; the second
stage is related to dehydration of calcium aluminate and alumi-
nosilicate hydrates (CAH and CASH) of different compositions usu-
ally between 145 °C and 330 °C; the third stage is ascribed to
disintegration of Ca(OH)2; and finally the fourth stage is of CaCO3.
The point of interest is to CASAH formation by pozzolans which
consume Ca(OH)2. On the one hand, it could be derived from
Table 14 that NS had no effect on Calcium hydroxide consuming
but MS brought about 5% Calcium hydroxide reduction in 7 days.
70
65
60
Predicted RCS (MPa)
55
50
45
40
35
30
20 30
Fig. 13. Predicting RCS with Mass loss using ANN.
On the other hand, when these two came together in MS7NS1,
the calcium hydroxide reduction boosted by 18 percent. It can be
concluded that the combined effect of MS and NS could enhance
the CASAH formation much more than each one of them
separately.
4.6. Discussions
In this study, some statistical analyses were implemented to
define the behavior of the SCC in the sulfuric acid medium.
’’Spearsman’s correlation coefficient‘‘ was used to define the corre-
lated parameters but it did not contribute to combined effect of MS
and NS. Thus, Two-way-ANOVA was used to find the most resis-
tance mix design under sulfuric acid medium. The results show
some differences between the additives combinations and using
them separately. Therefore, more comprehensive analyses using
NLR and ANN were used to investigate the effects of MS and NS
and also their combined effect. Before using these analyses, the
RS was used to revive the missing data viz. some of RCS values.
To minimize the MassLoss, PSO was used over NLR function, but
after 20 iterations, the optimum results were not derived precisely
and the values tended towards the NLR borders. Instead, with
respect to RMSE values, the predictor importance for ANN was
used as the final criteria.
Previous studies on concrete durability under sulfuric acid
attack revealed that the changes in the water/binder ratio, slump
or slump flow and aggregate content among all of the related spec-
imens had been neglected in most cases [10,13,71,72]. A distin-
guishing feature of the present study is that parameters such as
binder, water/binder ratio, and concrete slump were treated as
constants (same concrete class) to closely investigate the effects
of substitute materials as variables. There are some missing data
which are needed to revive and lack of these data could lead to
missing outputs. The RCS of 7 days of exposure is one of the
instances which should be predicted. For the first time, RS was
used to fill in these numbers and it even showed that more NS is
used the more compressive strength in constant substitution (7
percent substitution of pozzolans). This kind of analysis can be
widely used to give better recommendations for making durable
structures.
Since the ANN model is capable of handling several aspects of
concrete properties simultaneously beyond any human or statisti-
cal abilities, it would be logical to rely on ANN results, even though
there were some inconsistencies between analyses. According to
RMSE values, ANN is much more reliable than the NLR, and
R² = 0.9235
40 50 60 70 80
Real RCS (MPa)
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205 203

Fig. 14. Predicted values of mass change after 150, 450 and 900 days.

Table 14
TGA results. subsequently PI should be treated with the same manner. With
Type Phase Temperature (°C) Change in mass (%) respect to PI, initial compressive strength and water absorption
N First and second 147 0.88 ratio are the most effective parameters defining MassLoss and
328 2.36 RCS. NS has a marginally negative effect on residual compressive
Third 431 2.82 strength in sulfuric acid medium. One of the most influential
574 3.66 parameters enhancing RCS was the MS content which was similar
NS1 First and second 144 0.13 in the NLR and ANN predictions. The VOPP increment also gave rise
330 1.88 to enhance RCS but its importance was marginally different in the
Third 441 2.63
ANN and NLR. The ANN also shows DOE is not an effective
548 3.48
parameter to define RCS in sulfuric acid medium which is in
MS7 First and second 145 0.70
marked contrast to MassLoss prediction.
344 2.63
Third 449 3.11 In accordance to ANN analysis, typical brushing process did
551 3.90 not considerably change the results. On the contrary, NLR shows
MS7NS1 First and second 122 0.10 the most influential parameter was brushing. This result is
340 2.24 related to binary behavior of this parameter that should be
Third 461 3.05 neglected when comparing PIs. It should be noted that ANN
551 3.74
analysis shows the true significant of this parameter and NLR
204 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205
could not work well with binary variables. There is even an
inconsistency between ANN and counterplots. In accordance to
counter plots, the minimum RCS-loss occurred for combined
MS and NS substitution between 6 to 8 and 0.5 to 1.5 percent,
respectively. On the contrary, NLR found marginally negative
effect of NS on RCS. It should be noted that NLR works with
common results with no borders that means limited predictions
could be in contrast with what in reality. ANN also showed the
least effect of NS on RCS which could be generally concluded
that NS substitution didn’t affect on RCS change.
To reach the most compressive strength after 28 days of curing,
Combination of MS and NS were more effective than each one of
them separately. This behavior was observed by counter plots
between NS, MS and MassLoss after Spearsman’s rank correlation
coefficient analysis. This can be related to the microsilica grinding
act for agglomerated nanosilica which boost their combined influ-
ence. This result was also drawn out using two-way ANOVA, show-
ing combination of MS and NS differ from one of them in sulfuric
acid medium. Combination of these two stimulus calcium hydrox-
ide consumption which is parallel to statistical and artificial intel-
ligence outcomes. Equal rate of corrosion was also observed
between some of the mixtures with respect to two-way-ANOVA
which proved the same rate for NS2 and MS7 specimens. It opened
up that 2% of NS and 7% of MS can have identical behavior in sul-
furic acid medium.
5. Conclusion
Durability of different mixtures subjected to sulfuric acid
attacks are investigated in the present study. Ten mixtures of
self-consolidating concrete containing microsilica and nanosilica
were made and submerged in sulfuric acid medium with a con-
stant pH equal to 1. The mass reduction of the specimens was
recorded in consecutive steps during 150 days. The volume of per-
meable pores, water absorption ratio, and calcium hydroxide con-
sumption were determined to better understand of concrete
behavior in terms of their mass loss and compressive strength
under the specific experimental conditions. The six methods of
Spearsman statistical analysis, one-way-ANOVA, Two-way-
ANOVA, recommender systems, non-linear regression, and artifi-
cial neural networks were used to find the appropriate correlations
among the involved parameters. Based on the results of the exper-
imental program, the following conclusions may be drawn:
1. Combination of colloidal silica with superplasticizer has no
effect on portlandite content; instead, it acts as a filler;
2. Two-way-ANOVA revealed the same behavior of 7 percent
microsilica and 2 percent nanosilica in sulfuric acid
environment;
3. Recommender systems show the effectiveness of nanosilica to
enhance compressive strength in 7 percent substitution of
additives;
4. With respect to artificial neural networks outcomes, exposure
time is not an influential parameter on residual compressive
strength;
5. Nanosilica substitution didn’t have considerable effect on resid-
ual compressive strength results but in contrast microsilica
plays a major role to increase this value;
6. Water absorption ratio and initial compressive strength were of
paramount importance to predict mass-loss value;
7. Combination of microsilica and nanosilica was more effective
than when they used separately. This has proved that compres-
sive strength, mass-loss and calcium hydroxide consuming,
considerably have been improved by using binary pozzolans.
Acknowledgements
This work was partly supported by Yasouj University. Their sup-
port is gratefully acknowledged. The authors would also like to
thank Mr. Oveis Javadein for his valuable assistance.
References
[1] C.I. Goodier, Development of self-compacting concrete, Proc. ICE-Struct. Build.
156 (2003) 405–414.
[2] J. Newman, B. Choo, Advanced concrete technology 3: processes, 2003.
[3] M. Ouchi, H. Okamura, Self-compacting concrete development, present and
future’, Proceeding 1st Int Symp. Self-Compacting Concr, RILEM, Sweden, 1999.
[4] H.J.H. Brouwers, H.J. Radix, Self-compacting concrete: the role of the particle
size distribution, in: First Int Symp. Des. Perform. USe Self-Consolidating
Concr., 2005, pp. 109–118. doi:10.1617/2912143624.01.
[5] Y.X.C. Hwang, Z. Yu, C. Shi, K. Khayat, The effect of aggregate packing types on
engineering properties of self-consolidating concrete, 1st Int. Symp. Des.
Perform. Use Self-Consolidating Concr, RILEM Publ, SARL, 2005.
[6] W. Fuller, S.E. Thompson, The laws of proportioning concrete, Trans. Am. Soc.
Civil Eng. (1907) 67–143.
[7] C.C. Furnas, Flow of Gases Through Beds of Broken Solids, US Government
Printing Office, 1929.
[8] F. de Larrard, Concrete mixture proportioning: a scientifil approach, 1999.
doi:10.1017/CBO9781107415324.004.
[9] M. Liu, Wider application of additions in self-compacting concrete, Engineering
(2009).
[10] M.T. Bassuoni, M.L. Nehdi, Resistance of self-consolidating concrete to sulfuric
acid attack with consecutive pH reduction, Cem. Concr. Res. 37 (2007) 1070–
1084. doi:10.1016/j.cemconres.2007.04.014.
[11] C. Parker, The corrosion of concrete 1. The isolation of a species of bacterium
associated with the corrosion of concrete exposed to atmospheres containing
hydrogen sulphides, Aust. J. Exp. Biol. Med. Sci. 23 (1945) 81–90, http://dx.doi.
org/10.1038/icb.1945.13.
[12] E. Vincke, S. Verstichel, J. Monteny, W. Verstraete, A new test procedure for
biogenic sulfuric acid corrosion of concrete, Biodegradation 421–428 (2000).
[13] H. Rahmani, A.A. Ramazanianpour, Effect of binary cement replacement
materials on sulfuric acid resistance of dense concretes, Mag. Concr. Res. 60
(2008) 145–155, http://dx.doi.org/10.1680/macr.2008.60.2.145.
[14] F. Girardi, R.D. Maggio, Resistance of concrete mixtures to cyclic sulfuric acid
exposure and mixed sulfates: effect of the type of aggregate, Cem. Concr.
Compos. 33 (2011) 276–285, http://dx.doi.org/10.1016/j.
cemconcomp.2010.10.015.
[15] ASTM C267-01, Standard Test Methods for Chemical Resistance of Mortars,
ASTM Int. West Conshohocken, PA. (2012).
[16] A. Hendi, A. Behravan, D. Mostofinejad, S.M. Moshtaghi, K. Rezayi,
Implementing ANN to minimize sewage systems concrete corrosion with
glass beads substitution, Constr. Build. Mater. 138 (2017) 441–454, http://dx.
doi.org/10.1016/j.conbuildmat.2017.02.034.
[17] T.A. Durning, M.C. Hicks, Using microsilica to increase concrete’s resistance to
aggressive chemicals, Concr. Int. 13 (1991) 42–48.
[18] P.K. Mehta, Studies on chemical resistance of low water/cement ratio
concretes, Cem. Concr. Res. 15 (1985) 969–978, http://dx.doi.org/10.1016/
0008-8846(85)90087-0.
[19] D. Roy, P. Arjunan, Effect of silica fume, metakaolin, and low-calcium fly ash on
chemical resistance of concrete, Cem. Concr. Res. 31 (2001) 1809–1813, http://
dx.doi.org/10.1016/S0008-8846(01)00548-8.
[20] J. Monteny, E. Vincke, A. Beeldens, N. De Belie, L. Taerwe, D. Van Gemert, et al.,
Chemical, microbiological, and in situ test methods for biogenic sulfuric acid
corrosion of concrete, Cem. Concr. Res. 30 (2000) 623–634, http://dx.doi.org/
10.1016/S0008-8846(00)00219-2.
[21] Z.T. Chang, X.J. Song, R. Munn, M. Marosszeky, Using limestone aggregates and
different cements for enhancing resistance of concrete to sulphuric acid attack,
Cem. Concr. Res. 35 (2005) 1486–1494, http://dx.doi.org/10.1016/j.
cemconres.2005.03.006.
[22] E. Hewayde, M.L. Nehdi, E. Allouche, G. Nakhla, Using concrete admixtures for
sulphuric acid resistance, Proc. Inst. Civ. Eng. Mater. 160 (2007) 25–35, http://
dx.doi.org/10.1680/coma.2007.160.1.25.
[23] Y. Senhadji, G. Escadeillas, M. Mouli, H. Khelafi, Benosman, Influence of
natural pozzolan, silica fume and limestone fine on strength, acid resistance
and microstructure of mortar, Powder Technol. 254 (2014) 314–323, http://dx.
doi.org/10.1016/j.powtec.2014.01.046.
[24] J. Monteny, N. Belie, L. Taerwe, Resistance of different types of concrete
mixtures to sulfuric acid, Mater. Struct. 36 (2003) 242–249, http://dx.doi.org/
10.1007/BF02479618.
[25] H. Saricimen, M. Shameem, M.S. Barry, M. Ibrahim, T.A. Abbasi, Durability of
proprietary cementitious materials for use in wastewater transport systems,
Cem. Concr. Compos. 25 (2003) 421–427, http://dx.doi.org/10.1016/S0958-
9465(02)00082-3.
[26] D. Jana, R. Lewis, Acid attack in a concrete sewer pipe–a petrographic and
chemical investigation, in: Proc 27th Int Conf Cem Microsc, 2005.
[27] K. Kawai, S. Yamaji, T. Shinmi, Concrete deterioration caused by sulfuric acid
attack, in: Durab. Build. Mater, 2005.
 A. Hendi et al. / Construction and Building Materials 152 (2017) 192–205
[28] V. Boel, K. Audenaert, G. De Schutter, Acid attack of Self compacting concrete,
in: Concr. Repair, Rehabil. Retrofit. – Proc. Int. Conf. Concr. Repair, Rehabil.
Retrofit. ICCRRR 2005. (2006) 37–39.
[29] M. Schmidt, K. Hormann, F.J. Hofmann, E. Wagner, Beton mit erhöhtem
Widerstand gegen Säure und Biogene Schwefelsäurekorrosion, Betonw. Und
Fert. Precast. Plant Technol. 63 (1997) 64–70.
[30] P. Bartos, Nanotechnology in Construction. 2009, doi:10.1007/978-3-642-
00980-8.
[31] A.K. Rana, S.B. Rana, A. Kumari, V. Kiran, Significance of nanotechnology in
construction engineering, Int. J. Recent Trends Eng. 1 (2009) 6–8.
[32] K. Sobolev, I. Flores, L.M. Torres-Martinez, P.L. Valdez, E. Zarazua, E.L. Cuellar,
Engineering of SiO2 nanoparticles for optimal performance in nano cement-
based materials, Nanotechnol. Constr. 3 (2009) 139–148, http://dx.doi.org/
10.1007/978-3-642-00980-8_18.
[33] L. Ranjith, Q. Roger, K. Gowtham, Durability study of concrete using nano-silica
and plaster of Paris, J. Adv. Reserach Appl. Sci. 2016 (2016) 53–64.
[34] A.A. Maghsoudi, M.J. Soheil, A. Darbhenz, Effect of the nano particles in the
new generation of concretes, SCC, Int. J. Nanosci. Nanotechnol. 6 (2010) 137–
143.
[35] F. Sanchez, K. Sobolev, Nanotechnology in concrete – a review, Constr. Build.
Mater. 24 (2010) 2060–2071, http://dx.doi.org/10.1016/
j.conbuildmat.2010.03.014.
[36] M. Stefanidou, I. Papayianni, Influence of nano-SiO2 on the Portland cement
pastes, Compos. Part B Eng. 43 (2012) 2706–2710, http://dx.doi.org/10.1016/
j.compositesb.2011.12.015.
[37] M. Nili, A. Ehsani, K. Shabani, Influence of Nano-SiO2 and Microsilica on
Concrete Performance, in: Proc. Second Int. Conf. Sustain. Constr. Mater.
Technol. 2010, 7.
[38] Y. Qing, Z. Zenan, K. Deyu, C. Rongshen, Influence of nano-SiO2 addition on
properties of hardened cement paste as compared with silica fume, Constr.
Build. Mater. 21 (2007) 539–545, http://dx.doi.org/10.1016/
j.conbuildmat.2005.09.001.
[39] M. Collepardi, S. Collepardi, U. Skarp, R. Troli, Optimization of silica fume, fly
ash and amorphous nano-silica in superplasticized high-performance
concrete, ACI Spec. Publ. 221 (2004) 495–505.
[40] B.W. Jo, C.H. Kim, J.H. Lim, Investigations on the development of powder
concrete with nano-SiO2 particles, KSCE J. Civ. Eng. 11 (2007) 37–42, http://dx.
doi.org/10.1007/BF02823370.
[41] P. Mario, C. Engineering, P. Milan, Influence of amorphous colloidal silica on
the properties of self-compacting concretes, in: Proc. Fifth CANMET/ACI Int.
Conf. ‘‘Superplasticizers Other Chem. Admixtures Concr. SP-173. (1997) 1–21.
[42] N. Deshpande, S. Londhe, S.S. Kulkarni, Modelling compressive strength of
recycled aggregate concrete using neural networks and regression, Concr. Res.
Lett. 4 (2013).
[43] G. Jiang, J. Keller, P.L. Bond, Z. Yuan, Predicting concrete corrosion of sewers
using artificial neural network, Water Res. 92 (2016) 52–60, http://dx.doi.org/
10.1016/j.watres.2016.01.029.
[44] ASTM C150, Standard specification for Portland cement, Annu. B ASTM Stand.
4 (2011) 1–7, http://dx.doi.org/10.1002/jbm.b.31853.
[45] ASTM C33/C33 M – 13, Standard specification for concrete aggregates, ASTM
Int. 11 (2003), http://dx.doi.org/10.1520/C0033.
[46] ASTM Standard C136, Standard test method for sieve analysis of fine and
coarse aggregates, ASTM Int. 14 (2001) 1–5, http://dx.doi.org/10.1520/C0136-
06.2.
[47] Astm:C702/C702M-11, Standard practice for reducing samples of aggregate to
testing size doi:10.1520/C0702, ASTM Int. 98 (2011) 1–5.
[48] ASTM C127-01, Standard Test Method for Density, Relative Density (Specific
Gravity), and Absorption of Coarse Aggregate, ASTM Int. West Conshohocken,
PA. (2001) 3–8.
[49] American Society for Testing and Materials-C128, Standard Test Method for
Density, Relative Density (Specific Gravity), and Absorption of fine Aggregate,
ASTM Int. West Conshohocken, PA. (2001) 1–6. doi:10.1520/C0127-12.1.
205
[50] T. Aarre, P.L. Domone, Reference concretes for evaluation of test methods for
SCC, in: Proc. Third Int. RILEM Symp. Self-Compacting Concr. (2003) 495–505.
[51] T.Y. Lim, C.T. Tam, S.L. Lee, R. Sri Ravindrarajah, Relationship between strength
and volumetric composition of moist-cured cellular concrete, Mag. Concr. Res.
39 (1987) 12–18, http://dx.doi.org/10.1680/macr.1987.39.138.12.
[52] ACI Committee 237, 237 R-07 Self-Consolidating Concrete, Mi, Usa. (2007) 30.
[53] K.H. Khayat, D. Feys, Design, Production and Placement of Self-Consolidating
Concrete, RILEM Bookseries. 1 (2010) 153–162 doi:10.1007/978-90-481-9664-
7.
[54] C. BIBM, E. EFCA, The European guidelines for self-compacting concrete:
specification, production and use, 2005.
[55] P.L. Domone, Department of Civil, Environmental and Geomatic Engineering
Proportioning of self-compacting concrete – the UCL method P Domone
November 2009, Dep. Civ., Environ. Geomat. Eng. (2010) 30.
[56] Astm:C192/C192 M-13, Standard practice for making and curing concrete test
specimens in the laboratory, ASTM Int. 4 (2013) 1–8, http://dx.doi.org/
10.1520/C0192.
[57] ASTM C642, Standard Test Method for Density, Absorption, and Voids in
Hardened Concrete, 2013.
[58] ASTM C617/C617 M 12, Standard practice for capping cylindrical concrete
specimens, ASTM Int. 98 (2012) 1–6.
[59] A. Naji Givi, S. Abdul Rashid, F.N.A. Aziz, M.A.M. Salleh, Experimental
investigation of the size effects of SiO2 nano-particles on the mechanical
properties of binary blended concrete, Compos. Part B Eng. 41 (2010) 673–677,
http://dx.doi.org/10.1016/j.compositesb.2010.08.003.
[60] J.W. Tukey, Comparing individual means in the analysis of variance, Biometrics
5 (1949) 99–114, http://dx.doi.org/10.2307/3001913.
[61] S.K. Gorakala, M. Usuelli, Building a Recommendation System with R, 2015.
[62] D.M. Sima, Regularization Techniques in Model Fitting and Parameter
Estimation (PhD Thesis), 2006.
[63] M. Clerc, Particle Swarm Optimization, 2006 doi:10.1002/9780470612163.
[64] H. Yazd, S. Arabshahi, Optimal designing of concrete gravity dam using Particle
Swarm Optimization Algorithm (PSO), Indian J. Sci. Technol. (2015).
[65] G. Jiang, J. Keller, P. Bond, Z. Yuan, Predicting concrete corrosion of sewers
using artificial neural network, Water Res. (2016).
[66] G.T.N. Zavrtanik, J. Prosen, M. Tušar, The use of artificial neural networks for
modeling air void content in aggregate mixture, Autom. Constr. 63 (2016)
155–161, http://dx.doi.org/10.1016/j.autcon.2015.12.009.
[67] N. Smithing, Supervised Learning in Feedforward Artificial Neural Networks, R.
D. Reed, R.J. Marks. (1999).
[68] N. De Belie, H.J. Verselder, B. De Blaere, D. Van Nieuwenburg, R. Verschoore,
Influence of the cement type on the resistance of concrete to feed acids, Cem.
Concr. Res. 26 (1996) 1717–1725, http://dx.doi.org/10.1016/S0008-8846(96)
00155-X.
[69] A. E1131-08, Standard Test Method for Compositional Analysis by
Thermogravimetry, ASTM Int. West Conshohocken, PA,. (2014) 6.
doi:10.1520/E1131.
[70] J. Payá, J. Monzó, M.V. Borrachero, S. Velázquez, Evaluation of the pozzolanic
activity of fluid catalytic cracking catalyst residue (FC3 R). Thermogravimetric
analysis studies on FC3 R-Portland cement pastes, Cem. Concr. Res. 33 (2003)
603–609, http://dx.doi.org/10.1016/S0008-8846(03)00014-0.
[71] J. Hill, E.A. Byars, J.H. Sharp, C.J. Lynsdale, J.C. Cripps, Q. Zhou, An experimental
study of combined acid and sulfate attack of concrete, Cem. Concr. Compos. 25
(2003) 997–1003, http://dx.doi.org/10.1016/S0958-9465(03)00123-9.
[72] E. Hewayde, M. Nehdi, E. Allouche, G. Nakhla, D. Ph, M. Nehdi, et al., Effect of
mixture design parameters and wetting-drying cycles on resistance of
concrete to sulfuric acid attack, J. Mater. Civ. Eng. 19 (2007) 155–163, http://
dx.doi.org/10.1061/(asce)0899-1561(2007)19:2(155).