Upload
shadi-riahi
View
216
Download
2
Embed Size (px)
Citation preview
ORIGINAL ARTICLE
ANFIS-based prediction of the compressive strengthof geopolymers with seeded fly ash and rice husk–bark ash
Ali Nazari • Gholamreza Khalaj • Shadi Riahi
Received: 10 July 2011 / Accepted: 27 September 2011 / Published online: 21 October 2011
� Springer-Verlag London Limited 2011
Abstract In the present work, compressive strength of
geopolymers made from seeded fly ash and rice husk–bark ash
has been predicted by adaptive network-based fuzzy inference
systems (ANFIS). Different specimens, made from a mixture
of fly ash and rice husk–bark ash in fine and coarse forms and a
mixture of water glass and NaOH mixture as alkali activator,
were subjected to compressive strength tests at 7 and 28 days
of curing. The curing regimes were different: one set of the
specimens were cured in water at room temperature until 7 and
28 days and the other sets were oven-cured for 36 h at the
range of 40–90�C and then cured at room temperature until 7
and 28 days. A model based on ANFIS for predicting the
compressive strength of the specimens has been presented. To
build the model, training and testing using experimental
results from 120 specimens were conducted. The used data as
the inputs of ANFIS models are arranged in a format of six
parameters that cover the percentage of fine fly ash in the ashes
mixture, the percentage of coarse fly ash in the ashes mixture,
the percentage of fine rice husk–bark ash in the ashes mixture,
the percentage of coarse rice husk–bark ash in the ashes
mixture, the temperature of curing, and the time of water
curing. According to these input parameters in the ANFIS
models, the compressive strength of each specimen was pre-
dicted. The training and testing results in ANFIS models
showed a strong potential for predicting the compressive
strength of the geopolymeric specimens.
Keywords Geopolymer � Compressive strength �Fly ash � Rice husk–bark ash � ANFIS � FTIR �Curing regime
1 Introduction
Geopolymer that was developed by Davidovits contains both
silica and alumina and can be used as a binder to produce
geopolymer. Various alkali activators also play a key role in
producing geopolymers by dissolving silica and alumina from
the raw material and forming aluminosilicate structures.
Geopolymer is used for a variety of applications such as
sculpture, building, repairing, and restoration. Numerous
research publications related to geopolymers have been
released, with some reporting on chemical composition
aspects or reaction processes while others present results
related to mechanical properties and durability [1]. The gen-
eral formula to describe the chemical composition of these
mineral polymers is Mn[–(SiO2)z–AlO2]n�wH2O, where z is 1,
2, or 3, M is an alkali cation (such as potassium or sodium),
and n is the degree of polymerization. Accordingly, in order to
better describe the geopolymeric structures, a terminology has
been proposed: poly(sialate) (–Si–O–Al–O–), poly(sialate-
siloxo) (–Si–O–Al–O–Si–O–), and poly(sialatedisiloxo)
(–Si–O–Al–O–Si–O–Si–O–) [2]. The main properties of
geopolymers are quick compressive strength development,
low permeability, resistance to acid attack, good resistance to
freeze–thaw cycles, and tendency to drastically decrease the
mobility of most heavy metal ions contained within the geo-
polymeric structure [3]. Such properties make them interest-
ing structural products, such as concrete replacements in
various environments, and immobilization systems for heavy
metal containment [4].
The compressive strength of an inorganic polymer depends
on both the ratio of Si/Al and the types of the utilized raw
material. Fly ash (FA) is recently used as a source material to
produce geopolymer because of its suitable chemical com-
position along with favorable size and shape. Fly ash is a
by-product of coal-fired electric power stations. Literature
A. Nazari (&) � G. Khalaj � S. Riahi
Department of Materials Engineering, Science and Research
branch, Islamic Azad University, Tehran, Iran
e-mail: [email protected]
123
Neural Comput & Applic (2013) 22:689–701
DOI 10.1007/s00521-011-0751-y
survey specifies that fly ash is primarily composed of SiO2,
Al2O3, and Fe2O3. Since the quality of fly ash depends on the
type and the quality of coal along with the performance of the
power plant, difficulties sometimes remain to control its
chemical composition. In order to achieve a suitable chemical
composition to produce geopolymers, the preferred method is
to blend fly ash with another high silica source [1].
Rice husk–bark ash (RHBA) is a solid waste generated by
biomass power plants using rice husk and eucalyptus bark as
fuel. The power plant company providing RHBA for this
research reported that about 450 tons/day of RHBA are
produced and discarded. The major chemical constituent of
RHBA is SiO2 (about 75%) [5, 6]. Therefore, blending FA
and RHBA can adjust the ratio of Si/Al as required.
Several works have addressed utilizing of computer-
aided prediction of engineering properties including those
done by the authors [7–12]. Adaptive network-based fuzzy
inference systems (ANFIS) is the famous hybrid neuro-
fuzzy network for modeling the complex systems [13].
ANFIS incorporates the human-like reasoning style of
fuzzy systems through the use of fuzzy sets and a linguistic
model consisting of a set of IF–THEN fuzzy rules. The
main strength of ANFIS models is that they are universal
approximators [13] with the ability to solicit interpretable
IF–THEN rules. Nowadays, the artificial intelligence-based
techniques like ANFIS [14] have been successfully applied
in the engineering applications. However, there is a lack of
investigations on metallurgical aspects of materials.
To the authors’ knowledge, there are no works on utilizing
a mixture of FA and RHBA with seeded distribution of par-
ticles to produce geopolymers. In addition, since the concept
of geopolymers is completely new and there are few works on
their properties, application of computer programs like neural
networks to predict their properties is rarely reported. The aim
of this study is to investigate the compressive strength of
geopolymers produced form seeded FA and RHBA mixture
experimentally and presenting suitable model based on AN-
FIS to predict their compressive strength. Both FA and RHBA
with two different particle size distributions have been mixed
with different amounts to produce four classes of geopoly-
mers. Compressive strength of the produced specimens has
been investigated after specific times of curing. Totally, 120
data of compressive strength tests in different conditions were
collected, trained, and tested by means of ANFIS. The
obtained results have been compared by experimental meth-
ods to evaluate the software power for predicting the com-
pressive strength of the geopolymer specimens.
2 Experimental procedure
The cementitious materials used in this work were FA and
RHBA. Their chemical composition has been illustrated in
Table 1. In addition, Fig. 1 shows SEM micrograph of the
cementitious materials, respectively. The as-received ashes
were sieved and the particles passing the finesses of 150
and 33 lm were grinded using ball mill for 30 and
180 min, respectively, which yielded two different samples
for each of FA and RHBA. The average particle sizes
obtained for FA were 75 lm (coarser FA named cF in this
study) and 3 lm (finer FA named fF in this study) with the
BET specific surface of 31.3 and 38.9 m2/g, respectively.
The average particle sizes obtained for RHBA were 90 lm
(coarser RHBA named cR in this study) and 7 lm (finer
RHBA named fR in this study) with the BET specific
surface of 26 and 33.1 m2/g, respectively. The four
Table 1 Chemical composition of FA, RHBA, and WG (wt%)
Material SiO2 Al2O3 Fe2O3 CaO SO3 Na2O Loss on
ignition
FA 35.21 23.23 12.36 20.01 2.36 0.36 0.24
RHBA 81.36 0.4 0.12 3.23 0.85 – 3.55
WG 34.21 – – – – 13.11 –
Fig. 1 SEM micrograph of a FA and b RHBA used in this study
690 Neural Comput & Applic (2013) 22:689–701
123
produced samples were used in the experiment. Figure 2
shows the particle size distribution of the four produced
samples.
Sodium silicate solution or water glass (WG) and
sodium hydroxide (NaOH) were used as the solution part of
the mixture. WG was used without following modification,
but the sodium hydroxide was diluted to different con-
centrations before using. The chemical composition of the
utilized WG is also given in Table 1.
Totally 4 series of geopolymer specimens each con-
taining 2 different mixtures of FA and RHBA as illustrated
in Table 2 were prepared for compressive strength tests.
The mixed alkali activator of sodium silicate solution and
sodium hydroxide solution was used. Sodium hydroxide
was diluted by tap water to have concentrations of 4, 8, and
12 M. The solution was left under ambient conditions until
the excess heat had completely dissipated to avoid accel-
erating the setting of the geopolymeric specimens. The
sodium silicate solution without preparation was mixed
with the sodium hydroxide solution. The ratio of the
sodium silicate solution to sodium hydroxide solution was
2.5 by weight for all mixtures because this ratio demon-
strated the best properties for fly ash–based geopolymer
[15, 16]. For all samples, the mass ratio of alkali activator
to FA–RHA mixture was 0.4. Pastes were mixed by
shaking for 5–10 min to give complete homogenization.
The mixtures were cast in ø30 mm 9 60 mm polypropyl-
ene cylinders. The mixing was done in an air-conditioned
room at approximately 25�C. The molds were half-filled,
vibrated for 45 s, filled to the top, again vibrated for 45 s,
and sealed with the lid. The mixtures were then precured
for 24 h at room temperature (this precuring time has been
found to be beneficial to strength development [17]). Pre-
curing time before application of heat induces significant
dissolution of silica and alumina from fly ash and forma-
tion of a continuous matrix phase, increasing, therefore, the
homogeneity of the geopolymeric materials [17, 18]. After
the precuring process, the samples and molds were placed
in a water bath to prevent moisture loss and the carbonation
of the surface. One batch of these samples was placed in an
air-conditioned room at 25�C. The other batch was put in
the oven at the elevated temperatures of 50–90�C for 36 h.
To determine the most effective alkali concentration on
compressive strength, one set of the specimens cured at
80�C for 36 h were subjected to compressive strength tests.
Afterward, the other sets of samples were tested at 7 and
28 days of curing (for the specimens cured at elevated
temperature, the time of oven-curing was also considered).
The compressive strength results of the produced spec-
imens were measured on the cylindrical samples. The tests
were carried out triplicately and average strength values
were obtained.
The microstructural characteristic of geopolymeric
specimens, which was made at the optimum condition and
had a high compressive strength, was analyzed using
Fourier transform infrared spectroscopy (FTIR). The
powder samples were mixed with KBr at a concentration of
0.2–1.0 wt% to make the KBr disks. Then, the disks were
evaluated using a Perkin Elmer FTIR microscope.
3 Experimental results and discussion
3.1 Compressive strength
To determine the effect of NaOH concentration on com-
pressive strength of the specimens, one series of the
specimens (fF-fR series) were randomly selected, produced
by different concentrations of NaOH (4, 8 and 12 M) and
subjected to compressive strength tests after 36 h oven-
curing. Figure 3 shows the effects of NaOH concentration
Fig. 2 Particle size distribution pattern of the different ashes used in
this study
Table 2 Mixture proportioning of the utilized FA and RHBA to
produce geopolymeric specimens
Sample
designation
Weight
percent
of fine
FA (fF
wt%)
Weight
percent
of coarse
FA (cF
wt%)
Weight
percent
of fine
RHBA
(fR wt%)
Weight
percent of
coarse
RHBA
(cR wt%)
SiO2/
Al2O3
ratio
fF-fR-1 60 0 40 0 3.81
fF-fR-2 70 0 30 0 2.99
fF-fR-3 80 0 20 0 2.38
fF-cR-1 60 0 0 40 3.81
fF-cR-2 70 0 0 30 2.99
fF-cR-3 80 0 0 20 2.38
cF-fR-1 0 60 40 0 3.81
cF-fR-2 0 70 30 0 2.99
cF-fR-3 0 80 20 0 2.38
cF-cR-1 0 60 0 40 3.81
cF-cR-2 0 70 0 30 2.99
cF-cR-3 0 80 0 20 2.38
Ratio of alkali activator (WG ? NaOH) to FA–RHBA mixture is
0.4
Neural Comput & Applic (2013) 22:689–701 691
123
on compressive strength of the fF-fR series specimens
cured for 36 h at 80�C. As Fig. 3 shows, the compressive
strength by the geopolymers synthesized using the most
concentrated alkaline solution (12 M NaOH) was the
highest for mixtures. Zuhua et al. [19] reported that the use
of high molarities NaOH (such as 12 M) could accelerate
dissolution and hydrolysis but obstruct polycondensation.
Thus, 12 M NaOH can be considered as the suitable
solution for preparing geopolymer specimens. Therefore,
as discussed in the experimental section, the specimens
were produced by 12 M NaOH solution and were cured at
the mentioned time and temperatures.
The compressive strength of the produced specimens
has been illustrated in Figs. 4 and 5 for 7 and 28 days of
curing. Figures 4 and 5 show that the best strength has been
achieved for fF-fR2 specimen cured at 80�C for 36 h in
both 7 and 28 days curing regimes. As Figs. 4 and 5 show,
the optimum curing condition for all the mixtures is at
80�C. Curing temperature has a significant effect on the
compressive strength development because it affects
specimens’ setting and hardening. Synthesized products are
known to be very sensitive to experimental conditions [20].
However, compressive strength begins to decrease after
curing for a certain period of time at higher temperature.
Prolonged curing at higher temperatures can break down
the granular structure of geopolymer mixture. This results
in dehydration and excessive shrinkage due to contraction
of the gel, which does not transform into a more semi-
crystalline form [21].
On the whole, samples made with the fine RHBA and
FA particles (fF-fR series) showed considerably higher
strength than the other series. This may be due to pro-
duction of more compacted specimens. Fine particles are
capable of filling the vacancies and producing more den-
sified specimens, which make them stronger to the applied
loads. This has been confirmed in some works done on
concrete specimens [22], but to the authors’ knowledge,
there are not any reports that confirm this matter in
geopolymers.
3.2 FTIR results
The FTIR spectra of the inorganic polymers are shown
in Fig. 6. In Fig. 6, the IR bands are identified as follows:
Si–O stretching is located in the range 1,000–1,200 cm-1,
Si–O bending is found at 800 cm-1 and between 890 and
975 cm-1. The band at approximately 1,100 cm-1 is
assigned to the Si–O stretching of Q4 units and the band atFig. 3 The effect of NaOH concentration on compressive strength of
fF-fR series specimens cured at 80�C for 36 h
Fig. 4 Seven days compressive
strength of a fF-fR, b fF-cR,
c cF-fR, and cF-cR series
specimens
692 Neural Comput & Applic (2013) 22:689–701
123
1,050 cm-1 is assigned to Q3 units with a non-bridging
oxygen (Si–O–NBO) per SiO4 tetrahedron [23].
From Fig. 6, it is seen that an increase in the fine fly ash
particles shifts the position of the maximum absorbance of
Si–O bands toward lower wave numbers, indicating the
transformation of Q4 units to Q3 units.
4 Architecture of ANFIS
The architecture of an ANFIS model with two input vari-
ables is shown in Fig. 7. Suppose that the rule base of
ANFIS contains two fuzzy IF–THEN rules of Takagi and
Sugeno’s type as follows:
Rule 1: IF x is A1 and y is B1, THEN f1 = p1x ?
q1y ? r1.
Rule 2: IF x is A2 and y is B2, THEN f2 = p2x ?
q2y ? r2.
In Fig. 7 fuzzy reasoning is illustrated and also the
corresponding equivalent ANFIS architecture is shown in
Fig. 8. The functions of each layer are described as follows
[13, 14, 24, 25]:
Layer 1—Every node i in this layer is a square node
with a node function:
O1i ¼ lAi
ðxÞ ð1Þ
where x is the input to node i, and Ai is the linguistic label
(fuzzy sets: small, large, …) associated with this node
function.
Layer 2—Every node in this layer is a circle node
labeled P, which multiplies the incoming signals and sends
the product out. For instance,
Wi ¼ lAiyð Þ � lBi
yð Þ; i ¼ 1; 2 ð2Þ
Each node output represents the firing weight of a rule.
Layer 3—Every node in this layer is a circle node
labeled N. The ith node calculates the ratio of the ith rule’s
firing weight to the sum of all rule’s firing weights:
Wi ¼ Wi= W1=W2ð Þ; i ¼ 1; 2 ð3Þ
Layer 4—Every node in this layer is a square node with
a node function:
O4i ¼ �wi pixþ qiyþ r1ð Þ ð4Þ
where �wi is the output of layer 3, and {pi, qi, ri} is the
parameter set.
Fig. 5 Twenty-eight days
compressive strength of a fF-fR,
b fF-cR, c cF-fR, and cF-cR
series specimens
Fig. 6 FTIR results of the selected specimens: a fF-fR-2, b fF-cR-2,
c cF-fR-2 and d cF-cR specimens
Neural Comput & Applic (2013) 22:689–701 693
123
Layer 5—The signal node in this layer is a circle node
labeled R that computes the overall output as the summa-
tion of all incoming signals, that is,
O5i ¼
X
i
�wifi ¼X
i
wifi=X
i
wi: ð5Þ
The basic learning rule of ANFIS is the back-propagation
gradient descent, which calculates error signals recursively
from the output layer backward to the input nodes. This
learning rule is exactly the same as the back-propagation
learning rule used in the common feed-forward neural
networks [26, 27]. Recently, ANFIS adopted a rapid
learning method named as hybrid-learning method that
utilizes the gradient descent and the least-squares method
to find a feasible set of antecedent and consequent
parameters [26, 27]. Thus, in this paper, the later method is
used for constructing the proposed models.
4.1 ANFIS model structure and parameters
The structure of proposed ANFIS networks consisted of six
input variables including the percentage of fine fly ash in
the ashes mixture (fF), the percentage of coarse fly ash in
the ashes mixture (cF), the percentage of fine rice husk–
bark ash in the ashes mixture (fR), the percentage of coarse
rice husk–bark ash in the ashes mixture (cR), the temper-
ature of curing (T), and the time of water curing (t). The
value for output layer was compressive strength (fS). The
input space is decomposed by three fuzzy labels. In this
paper, for comparison purposes, two types of membership
functions (MFs) including the triangular (ANFIS-I) and
Gaussian (ANFIS-II) were utilized to construct the sug-
gested models. The ANFIS models were trained by 96
input–target pairs and tested by 26 data from testing pairs.
Moreover, up to 1,000 epochs were specified for training
process to assure the gaining of the minimum error
tolerance.
In this study, the Matlab ANFIS toolbox is used for AN-
FIS applications. To overcome optimization difficulty, a
program has been developed in Matlab, which handles the
trial-and-error process automatically [28–31]. The program
tries various functions and when the highest RMSE (root
mean squared error) of the testing set, as the training of the
testing set, is achieved, it was reported [28–31].
Fig. 7 The reasoning scheme of ANFIS [14]
Fig. 8 Schematic of ANFIS
architecture [8]
694 Neural Comput & Applic (2013) 22:689–701
123
The IF–THEN rules in this study were achieved as
follows. Suppose that the rule base of ANFIS contains
two fuzzy IF–THEN rules of Takagi and Sugeno’s
type:
Rule 1: IF fF is A1, cF is B1, fR is C1, cR is D1, T is E1
and t is F1 THEN f1 = p1fF ? q1cF ? r1fR ?
s1cR ? t1T ? u1t ? v1.
Rule 2: IF fF is A2, cF is B2, fR is C2, cR is D2, T is E2
and t is F2 THEN f2 = p2fF ? q2cF ? r2fR ?
s2cR ? t2T ? u2t ? v2.
The corresponding equivalent ANFIS architecture is
shown in Fig. 9. The functions of each layer are described
as follows:
Layer 1—Every node i in this layer is a square node
with a node function:
O1i ¼ lAi
ðfFÞ i ¼ 1; 2 ð6Þ
O1i ¼ lBi
ðcFÞ i ¼ 1; 2 ð7Þ
O1i ¼ lCi
ðfRÞ i ¼ 1; 2 ð8Þ
O1i ¼ lDi
ðcRÞ i ¼ 1; 2 ð9Þ
O1i ¼ lEi
ðTÞ i ¼ 1; 2 ð10Þ
O1i ¼ lFi
ðtÞ i ¼ 1; 2 ð11Þ
where fF, cF, fR, cR, T and t are inputs to node i, and Ai, Bi,
Ci, Di, Ei and Fi are the linguistic label (fuzzy sets: small,
large, …) associated with this node function.
Layer 2—Every node in this layer is a circle node
labeled G, which multiplies the incoming signals and sends
the product out. For instance,
Wi ¼ lAifFð Þ � lBi
cFð Þ � lCiðfRÞ
� lDiðcRÞ � lEi
ðTÞ � lFiðtÞ; i ¼ 1; 2 ð12Þ
Each node output represents the firing weight of a rule.
Layer 3—Every node in this layer is a circle node
labeled N. The ith node calculates the ratio of the ith rule’s
firing weight to the sum of all rule’s firing weights:
Wi ¼ Wi= W1=W2ð Þ; i ¼ 1; 2 ð13Þ
Layer 4—Every node in this layer is a square node with
a node function:
O4i ¼ �wi PifFþ qicFþ rifRþ sicRþ tiT þ uit þ við Þ
ð14Þ
where �wi is the output of layer 3, and {pi, qi, ri, si, ti, ui, vi,
zi} is the parameter set.
Layer 5—The signal node in this layer is a circle node
labeled R that computes the overall output as the summa-
tion of all incoming signals, that is,
N
Nw_
w_
w_
w_
out
Fig. 9 Schematic of ANFIS
architecture utilized in this work
Neural Comput & Applic (2013) 22:689–701 695
123
O5i ¼
X
i
�wifi ¼X
i
wifi=X
i
wi ð15Þ
5 Predicted results and discussion
In this study, the error arose during the training and testing
in ANFIS-I and ANFIS-II models can be expressed as
absolute fraction of variance (R2), which are calculated by
Eq. (16) [32]:
R2 ¼ 1�P
i ti � oið Þ2P
i oið Þ2
!ð16Þ
where t is the target value and o is the output.
All of the results obtained from experimental studies
and predicted by using the training and testing results of
ANFIS-I and ANFIS-II models are given in Fig. 10a, b,
respectively. The linear least square fit line, its equation,
and the R2 values were shown in these figures for the
training and testing data. Also, input values and experi-
mental results with training and testing results obtained
from ANFIS-I to ANFIS-II models were given in Table 3.
As it is visible in Fig. 10, the values obtained from the
training and testing in ANFIS-I and ANFIS-II models are
very close to the experimental results. The result of
testing phase in Fig. 10 shows that the ANFIS-I and
ANFIS-II models are capable of generalizing between
input and output variables with reasonably good
predictions.
The performance of the ANFIS-I and ANFIS-II models
is shown in Fig. 10. The best value of R2 is 96.32% for
training set in the ANFIS-II model. The minimum values
of R2 are 94.34% for testing set in the ANFIS-I model.
All of R2 values show that the proposed ANFIS-I and
ANFIS-II models are suitable and can predict the com-
pressive strength values very close to the experimental
values.
6 Conclusions
From the experimental procedure, the following results
were obtained:
1. The compressive strength of the specimens depends on
the particle size distribution pattern of the ashes, time
of oven-curing, and the time of room condition curing.
The finer ashes particle size results in the denser and
hence the stronger specimen. On the other hand, oven-
curing of the specimens at 80�C was found to be the
optimum temperature of curing in geopolymeric
specimens.
2. In all mixtures, the specimens with the SiO2/Al2O3
ratio equal to 2.99 had the highest strength. On the
other hand, the highest strength was achieved equals
58.9 MPa for the mixture of fine fly ash to fine rice
husk–bark ash of 70:30.
3. ANFIS can be an alternative approach for the evalu-
ation of the effect of seeded mixture of FA and RHBA
on compressive strength values of geopolymer
specimens.
Fig. 10 The correlation of the measured and predicted compressive
strength values of geopolymers in a training and b testing phase for
ANFIS models
696 Neural Comput & Applic (2013) 22:689–701
123
Ta
ble
3D
ata
sets
for
com
par
iso
no
fex
per
imen
tal
resu
lts
wit
hre
sult
sp
red
icte
dfr
om
the
AN
FIS
mo
del
s
Th
ep
erce
nta
ge
of
fin
efl
yas
h
inth
eas
hes
mix
ture
(fF
)
Th
ep
erce
nta
ge
of
coar
sefl
y
ash
inth
e
ash
esm
ixtu
re
(cF
)
Th
ep
erce
nta
ge
of
fin
eri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(fR
)
Th
ep
erce
nta
ge
of
coar
seri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(cR
)
Th
ete
mp
erat
ure
of
curi
ng
(T)
Th
eti
me
of
wat
er
curi
ng
(t)
Co
mp
ress
ive
stre
ng
thv
alu
es
ob
tain
edfr
om
exp
erim
ents
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-I
mo
del
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-II
mo
del
(MP
a)
60
04
00
25
72
7.1
26
.82
7.6
70
03
00
25
73
73
6.1
35
.9
80
02
00
25
73
3.4
32
.53
5.8
60
00
40
25
72
4.7
24
.12
5.4
70
00
30
25
73
3.7
29
.43
3.2
80
00
20
25
73
0.5
30
.43
0.2
06
04
00
25
71
9.7
22
.91
9.8
07
03
00
25
72
6.9
27
.42
6.4
08
02
00
25
72
4.3
24
.42
4.4
06
00
40
25
71
4.8
15
.31
5.1
07
00
30
25
72
0.2
19
.82
0.9
08
00
20
25
71
8.2
18
.51
9.1
60
04
00
40
73
0.4
29
.73
0.8
70
03
00
40
73
9.2
38
.73
7.3
80
02
00
40
73
5.4
34
.53
4.6
60
00
40
40
72
7.7
26
.92
9
70
00
30
40
73
5.8
30
.23
6.7
80
00
20
40
73
2.3
30
.33
0.1
06
04
00
40
72
2.1
23
.22
1.4
07
03
00
40
72
8.5
28
.22
6.4
08
02
00
40
72
5.8
25
.22
6
06
00
40
40
71
6.5
17
16
.9
07
00
30
40
72
1.4
19
.52
2.1
08
00
20
40
71
9.3
19
.31
9.1
60
04
00
60
73
1.9
32
.13
4.9
70
03
00
60
74
1.7
41
.33
9.6
80
02
00
60
73
7.2
37
36
.4
60
00
40
60
72
9.1
31
.33
4.8
70
00
30
60
73
83
4.4
41
.8
80
00
20
60
73
3.9
32
.93
6.1
06
04
00
60
72
3.2
24
.92
4.4
07
03
00
60
73
0.3
30
.93
0.7
08
02
00
60
72
72
7.8
29
.5
Neural Comput & Applic (2013) 22:689–701 697
123
Ta
ble
3co
nti
nu
ed
Th
ep
erce
nta
ge
of
fin
efl
yas
h
inth
eas
hes
mix
ture
(fF
)
Th
ep
erce
nta
ge
of
coar
sefl
y
ash
inth
e
ash
esm
ixtu
re
(cF
)
Th
ep
erce
nta
ge
of
fin
eri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(fR
)
Th
ep
erce
nta
ge
of
coar
seri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(cR
)
Th
ete
mp
erat
ure
of
curi
ng
(T)
Th
eti
me
of
wat
er
curi
ng
(t)
Co
mp
ress
ive
stre
ng
thv
alu
es
ob
tain
edfr
om
exp
erim
ents
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-I
mo
del
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-II
mo
del
(MP
a)
06
00
40
60
71
7.4
19
.11
9.1
07
00
30
60
72
2.7
21
.42
3.1
08
00
20
60
72
0.2
20
.41
9.9
60
04
00
80
73
4.8
32
.83
5.3
70
03
00
80
74
4.2
42
.74
1.2
80
02
00
80
73
9.6
38
.33
7.1
60
00
40
80
73
1.8
31
.93
1.9
70
00
30
80
74
0.3
35
.24
0.4
80
00
20
80
73
6.1
33
.33
8.3
06
04
00
80
72
5.3
26
.42
5.5
07
03
00
80
73
2.1
32
.33
2.1
08
02
00
80
72
8.8
27
.82
9
06
00
40
80
71
91
9.5
21
.6
07
00
30
80
72
4.1
22
23
.8
08
00
20
80
72
1.5
20
.92
1.7
60
04
00
90
73
2.4
32
.33
3
70
03
00
90
74
1.7
42
.14
0.1
80
02
00
90
73
7.7
37
.73
7.6
60
00
40
90
72
9.6
32
.12
9.9
70
00
30
90
73
83
5.1
38
.3
80
00
20
90
73
4.4
32
.63
5.8
06
04
00
90
72
3.6
25
.52
5
07
03
00
90
73
0.3
26
.22
5.6
08
02
00
90
72
7.4
26
.82
7.4
06
00
40
90
71
7.7
20
21
07
00
30
90
72
2.7
22
.22
2.9
08
00
20
90
72
0.5
21
.52
1.9
60
04
00
25
28
36
.13
5.4
36
.5
70
03
00
25
28
49
.34
7.8
49
.1
80
02
00
25
28
44
.54
2.6
46
.3
60
00
40
25
28
32
.53
5.4
27
.8
70
00
30
25
28
44
.43
8.9
35
.6
80
00
20
25
28
40
.13
4.8
40
.3
698 Neural Comput & Applic (2013) 22:689–701
123
Ta
ble
3co
nti
nu
ed
Th
ep
erce
nta
ge
of
fin
efl
yas
h
inth
eas
hes
mix
ture
(fF
)
Th
ep
erce
nta
ge
of
coar
sefl
y
ash
inth
e
ash
esm
ixtu
re
(cF
)
Th
ep
erce
nta
ge
of
fin
eri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(fR
)
Th
ep
erce
nta
ge
of
coar
seri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(cR
)
Th
ete
mp
erat
ure
of
curi
ng
(T)
Th
eti
me
of
wat
er
curi
ng
(t)
Co
mp
ress
ive
stre
ng
thv
alu
es
ob
tain
edfr
om
exp
erim
ents
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-I
mo
del
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-II
mo
del
(MP
a)
06
04
00
25
28
27
28
28
.7
07
03
00
25
28
36
.93
6.2
36
.3
08
02
00
25
28
33
.33
3.4
32
.5
06
00
40
25
28
20
.82
3.7
20
.2
07
00
30
25
28
28
.42
6.5
26
.8
08
00
20
25
28
25
.62
4.6
20
.4
60
04
00
40
28
40
.53
9.5
40
.8
70
03
00
40
28
52
.35
2.7
51
.5
80
02
00
40
28
47
.24
5.7
45
.4
60
00
40
40
28
36
.53
5.7
35
.7
70
00
30
40
28
47
.14
0.7
40
.7
80
00
20
40
28
42
.53
7.7
41
.9
06
04
00
40
28
30
.33
0.3
29
.8
07
03
00
40
28
39
.13
8.4
38
.7
08
02
00
40
28
35
.33
53
4.8
06
00
40
40
28
23
.32
3.8
22
.6
07
00
30
40
28
30
.12
8.5
28
.8
08
00
20
40
28
27
.22
5.6
26
60
04
00
60
28
42
.54
2.6
45
.7
70
03
00
60
28
55
.65
5.5
57
.8
80
02
00
60
28
49
.64
8.2
50
.1
60
00
40
60
28
38
.33
9.4
43
.5
70
00
30
60
28
50
45
49
.3
80
00
20
60
28
44
.64
4.3
46
06
04
00
60
28
31
.83
2.6
31
.9
07
03
00
60
28
41
.54
2.2
41
08
02
00
60
28
37
39
.63
6.6
06
00
40
60
28
24
.52
5.1
24
.9
07
00
30
60
28
32
30
.93
0.8
08
00
20
60
28
28
.52
9.1
27
.7
60
04
00
80
28
46
.44
4.2
46
70
03
00
80
28
58
.95
5.8
57
.7
80
02
00
80
28
52
.85
0.9
51
.6
Neural Comput & Applic (2013) 22:689–701 699
123
Ta
ble
3co
nti
nu
ed
Th
ep
erce
nta
ge
of
fin
efl
yas
h
inth
eas
hes
mix
ture
(fF
)
Th
ep
erce
nta
ge
of
coar
sefl
y
ash
inth
e
ash
esm
ixtu
re
(cF
)
Th
ep
erce
nta
ge
of
fin
eri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(fR
)
Th
ep
erce
nta
ge
of
coar
seri
ce
hu
sk–
bar
kas
h
inth
eas
hes
mix
ture
(cR
)
Th
ete
mp
erat
ure
of
curi
ng
(T)
Th
eti
me
of
wat
er
curi
ng
(t)
Co
mp
ress
ive
stre
ng
thv
alu
es
ob
tain
edfr
om
exp
erim
ents
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-I
mo
del
(MP
a)
Co
mp
ress
ive
stre
ng
thv
alu
es
pre
dic
ted
by
AN
FIS
-II
mo
del
(MP
a)
60
00
40
80
28
41
.84
1.7
41
.4
70
00
30
80
28
53
46
.35
1.9
80
00
20
80
28
47
.54
3.9
47
.2
06
04
00
80
28
34
.73
4.4
34
.6
07
03
00
80
28
44
43
.54
3.6
08
02
00
80
28
39
.44
0.5
38
.7
06
00
40
80
28
26
.72
6.2
25
.6
07
00
30
80
28
33
.93
1.6
32
.8
08
00
20
80
28
30
.32
8.6
29
.5
60
04
00
90
28
43
.24
2.7
42
.8
70
03
00
90
28
55
.65
4.6
54
80
02
00
90
28
50
.25
0.3
49
60
00
40
90
28
38
.94
1.9
38
.7
70
00
30
90
28
50
45
.74
9.6
80
00
20
90
28
45
.24
1.3
45
06
04
00
90
28
32
.33
3.3
32
.4
07
03
00
90
28
41
.54
1.9
41
.1
08
02
00
90
28
37
.53
8.8
36
.9
06
00
40
90
28
24
.92
5.8
24
.1
07
00
30
90
28
32
31
.23
1.5
08
00
20
90
28
28
.92
7.9
29
.2
700 Neural Comput & Applic (2013) 22:689–701
123
4. Comparison between ANFIS in terms of R2 showed
that ANFIS models are capable of predicting suitable
results for compressive strength values of geopolymer
specimens.
References
1. Wongpa J, Kiattikomol K, Jaturapitakkul C, Chindaprasirt P
(2010) Compressive strength, modulus of elasticity, and water
permeability of inorganic polymer concrete, mater. Design
31:4748–4754
2. Davidovits J (1991) Geopolymers: inorganic polymeric new
materials. J Therm Anal 37:1633–1656
3. van Jaarsveld JGS, van Deventer JSJ, Lorenzen L (1997) The
potential use of geopolymeric materials to immobilise toxic
metals. Part I. Theory and applications. Miner Eng 10:659–669
4. Alvarez-Ayuso E, Querol X, Plan F, Alastuey A, Moreno N,
Izquierdo M, Font O, Moreno T, Diez S, Vazquez E, Barra M
(2008) Environmental, physical and structural characterisation of
geopolymer matrixes synthesised from coal (co-)combustion fly
ashes. J Hazardous Mater 154:175–183
5. Sata V, Jaturapitakkul C, Kiattikomol K (2007) Influence of poz-
zolan from various byproduct materials on mechanical properties
of high-strength concrete. Constr Build Mater 21(7):1589–1598
6. Tangchirapat W, Buranasing R, Jaturapitakkul C, Chindaprasirt P
(2008) Influence of rice husk–bark ash on mechanical properties
of concrete containing high amount of recycled aggregates.
Constr Build Mater 22(8):1812–1819
7. Nazari A, Riahi S (2010) Computer-aided prediction of physical
and mechanical properties of high strength cementitious composite
containing Cr2O3 nanoparticles. Nanotechnology 5(5):301–318
8. Nazari A, Riahi S (2011) Prediction split tensile strength and
water permeability of high strength concrete containing TiO2
nanoparticles by artificial neural network and genetic program-
ming. Com Part B Eng 42:473–488
9. Nazari A, Riahi S (2011) Computer-aided design of the effects of
Fe2O3 nanoparticles on split tensile strength and water perme-
ability of high strength concrete. Mater Des 32:3966–3979
10. Nazari A, Milani AA, Zakeri M (2011) Modeling ductile to brittle
transition temperature of functionally graded steels by artificial
neural networks. Comput Mater Sci 50:2028–2037
11. Nazari A, Milani AA (2011) Modeling ductile to brittle transition
temperature of functionally graded steels by fuzzy logic. J Mat
Sci 46(18):6007–6017
12. Nazari A, Didehvar N (2011) Modeling impact resistance of
aluminum-epoxy laminated composites by ANFIS. doi:10.1016/
j.compositesb.2011.05.043
13. Jang JSR (1993) ANFIS: adaptive-network-based fuzzy inference
system. IEEE Trans Syst Man Cyber 23(3):665–685
14. Sarıdemir M (2009) Predicting the compressive strength of
mortars containing metakaolin by artificial neural networks and
fuzzy logic. Adv Eng Soft 40(9):920–927
15. Pacheco-Togal F, Castro-Gomes J, Jalali S (2007) Investigation
about the effect of aggregates on strength and microstructure of
geopolymeric mine waste mud binders. Cem Concr Res 37(6):
933–941
16. Pacheco-Torgal F, Castro-Gomes JP, Jalali S (2005) Studies
about mix composition of alkali-activated mortars using waste
mud from Panasqueira. In: Proceedings of the engineering con-
ference. University of Beira Interior, Covilha, Portugal
17. Bakharev T (2005) Geopolymeric materials prepared using class
F fly ash and elevated temperature curing. Cement Concrete Res
35:1224–1232
18. Chindaprasirt P, Chareerat T, Sirivivatnanon V (2007) Work-
ability and strength of coarse high calcium fly ash geopolymer.
Cement Concrete Comp 29:224–229
19. Zuhua Z, Xiao Y, Huajun Z, Yue C (2009) Role of water in the
synthesis of calcined kaolin-based geopolymer. Appl Clay Sci
43(2):218–223
20. Fernandez-Jimenez A, Garcia-Lodeiro I, Palomo A (2007)
Durability of alkali-activated fly ash cementitious materials.
J Mater Sci 42(9):3055–3065
21. Van Jaarsveld JGS, Van Deventer JSJ, Lukey GC (2002) The
effect of composition and temperature on the properties of fly
ash–and kaolinite-based geopolymers. Chem Eng J 89(1–3):
63–73
22. Naji Givi A, Abdul Rashid S, Nora A, Aziz F, Mohd Salleh MA
(2010) Assessment of the effects of rice husk ash particle size on
strength, water permeability and workability of binary blended
concrete. Const Build Mat 24(11):2145–2150
23. Zholobenko VL, Holmes SM, Cundy CS, Dwyer J (1997)
Microporous Mater 11:83
24. Ramezanianpour AA, Sobhani M, Sobhani J (2004) Application
of network based neuro-fuzzy system for prediction of the
strength of high strength concrete. Amirkabir J Sci Technol
5(59-C):78–93
25. Ramezanianpour AA, Sobhani J, Sobhani M (2004) Application
of an adaptive neurofuzzy system in the prediction of HPC
compressive strength. In: Proceedings of the fourth international
conference on engineering computational technology. Civil-
Comp Press, Lisbon, Portugal, p 138
26. Topcu IB, Sarıdemir M (2008) Prediction of mechanical prop-
erties of recycled aggregate concretes containing silica fume
using artificial neural networks and fuzzy logic. Comp Mater Sci
42(1):74–82
27. Jang JSR, Sun CT (1995) Nuro-fuzzy modeling and control. Proc
IEEE 83:3
28. Guzelbey IH, Cevik A, Erklig A (2006) Prediction of web crip-
pling strength of cold-formed steel sheeting’s using neural net-
works. J Constr Steel Res 62:962–973
29. Guzelbey IH, Cevik A, Gogus MT (2006) Prediction of rotation
capacity of wide flange beams using neural networks. J Constr
Steel Res 62:950–961
30. Cevik A, Guzelbey IH (2008) Neural Network modeling of
strength enhancement for Cfrp confined concrete cylinders. Build
Environ 43:751–763
31. Cevik A. Guzelbey IH (2007) A soft computing based approach
for the prediction of ultimate strength of metal plates in com-
pression. Eng Struct 29(3):383–394
32. Topcu IB, Sarıdemir M (2008) Prediction of compressive strength
of concrete containing fly ash using artificial neural network and
fuzzy logic. Comp Mater Sci 41(3):305–311
Neural Comput & Applic (2013) 22:689–701 701
123