Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Hindawi

Advances in Materials Science and Engineering


Volume 2022, Article ID 4528264, 11 pages
https://doi.org/10.1155/2022/4528264

Research Article
Modelling of Modulus of Elasticity of Low-Calcium-Based
Geopolymer Concrete Using Regression Analysis

Ali A. Khalaf and Katalin Kopecskó


Department of Engineering Geology and Geotechnics, Faculty of Civil Engineering,
Budapest University of Technology and Economics, 1111 Budapest, Hungary

Correspondence should be addressed to Ali A. Khalaf; ali.abdulhasan.khalaf@emk.bme.hu

Received 14 September 2021; Revised 19 April 2022; Accepted 25 April 2022; Published 19 May 2022

Academic Editor: Ismail DEMIR

Copyright © 2022 Ali A. Khalaf and Katalin Kopecskó. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

Despite the unremitting efforts to model the modulus of elasticity of low-calcium-based geopolymer concrete, the state-of-the-art
models need much improvement to reduce the error signals and increase the reliability. This study represents a comprehensive
regression analysis to model the modulus of elasticity of low-calcium-based geopolymer concrete in terms of its compressive
strength. The proposed model’s assumptions are based on taking into account the chemical composition and compressive strength
class and considering the normal density of concrete. The modelling is based on 67 different mix-design samples collected from
peer-reviewed literature, which are divided into two groups. The first group consists of 59 samples that are used to construct the
proposed model, while the second group consists of 8 samples that are used to test the validity of the proposed model. The analysis
showed that the proposed model gives the root mean squared error value (RMSE) of 3.122 GPa and the mean absolute percentage
error value (MAPE) of 15.0%. Therefore, the proposed model gives 41% and 52.2% reductions in RMSE and MAPE, respectively,
from the state-of-the-art model in the literature. Furthermore, other statistical parameters to evaluate the goodness of fitness of the
proposed model have been considered, such as the relative root mean squared error (RRMSE) (15.5%), the coefficient of de-
termination (R-squared) (0.773), and the coefficient of correlation (R) (0.88); all of which indicated that the goodness of fitness is
good and the proposed model has a high correlation to the actual values. Applying the proposed model in future applications will
help reduce the time and cost of geopolymer production, as the proposed model has significantly reduced the error signals.

1. Introduction utilised in the process of geopolymer production, which help


to produce sustainable construction materials [7–13].
Geopolymers are three-dimensional networks of silico- Therefore, geopolymer binders can be an efficient alternative
aluminate structures produced by reactivating amorphous to Portland cement in the concrete industry [1, 2, 5].
aluminosilicate (precursor or raw material) in the presence Machine learning techniques have been proven to be
of alkali metal cation (alkali activation solution) through a valuable and beneficial in civil engineering applications for
process that is so-called geopolymerization [1–4]. Geo- the prediction of the mechanical properties of geopolymer
polymer concrete is considered an environmentally friendly concrete due to the fact that these techniques, such as neural
construction material because its manufacturing emits much networks, genetic programming, linear regression, and
less carbon dioxide (CO2) gas than the manufacturing of nonlinear regression, have the ability to capture the complex
Portland cement concrete [5, 6]. Studies stated that 1-ton behaviour [14–17]. In practical applications of concrete
production of cement produces approximately 1 ton of structures, the most important feature to evaluate concrete
carbon dioxide (CO2). Moreover, many industrial wastes strength is compressive strength. All other mechanical
and by-product materials such as red mud, slag, mine properties, such as the modulus of elasticity, are expressed as
tailings, and volcanic pumice ash have been successfully functions of compressive strength [18]. Machine learning
2 Advances in Materials Science and Engineering

techniques help establish a functional approximation so- 2. Research Significance


lution with acceptable error from the unknown relationship
based on experimental observations to eventually express an Many researchers have introduced prediction models for the
approximate mathematical model between the dependent modulus of elasticity of low-calcium-based geopolymer
and independent variables [17, 19]. The nonlinear regression concrete in terms of its compressive strength
analysis has the advantages of expressing more complex [27, 30, 31, 34, 37, 38]. However, the error percentage is still
relationships and produces a visualised discrete relationship relatively high; for example, ACI 318 [21] determined the
that is easy to comprehend and apply [19]. MAPE in the range of ±20.0%, while the best performance
Young’s modulus of elasticity plays a pivotal role in model in the literature, Prachasaree et al. [27], has an MAPE
evaluating the load-deformation behaviour of concrete, value of 31.4%. Thus, this research aimed to establish a
which describes the reversible part of the stress-strain curve reliable regression model for the modulus of elasticity, Ec, of
[20]. In engineering practices such as ACI 318 [21] and the low-calcium-based geopolymer concrete in terms of its
CEB FIP model [22], prediction models are developed to compressive strength, fc′, by reducing the error signals and
calculate the modulus of elasticity of concrete instead of increasing the correlation between the proposed model and
testing each time; these models are established based on the actual values. The modelling process was carried out on
empirical equations depending on historical tests [21, 22]. all applicable data collected from peer-reviewed literature,
Since Portland cement concrete practice manuals and design and the proposed model was built with a 95% confidence
codes have been instituted over decades, most cement al- interval based on explicit and firm assumptions. It is re-
ternatives in the primary stage are compared to cement ported that increasing the compressive strength increases the
properties to examine their validation [23]. Despite that, quality of other properties of concrete and, more impor-
these prediction models of the modulus of elasticity of tantly, the modulus of elasticity [18]. Unfortunately, with an
Portland cement concrete overestimate the results when increase in compressive strength, there is a corresponding
used for geopolymer concrete [24]. Moreover, the results of increase in the cost of geopolymer production [39].
the modulus of elasticity of these models for geopolymer Therefore, by reducing the error signals of the proposed
concrete are smaller than the results for Portland cement model of modulus of elasticity, its application increases the
concrete, even when the compressive strength results of safety margins and controls the cost management.
geopolymer concrete are higher [25]. Consequently, it has
been found that high-strength concrete prediction models 3. The Proposed Model Assumptions
are in agreement margins when applied to high-strength
geopolymer concrete [24, 26]. However, this finding did not As a first step prior to analysis, the margins and charac-
solve the problem because the predicted modulus of elas- teristics of the population and samples should be clearly
ticity values will be far from the experimental values when determined and described to ensure that the prediction
the compressive strength decreases [27]. model will be representative. Therefore, the following as-
The other approach to research work is to build a new sumptions are constructed:
model to predict the modulus of elasticity of geopolymer
concrete. Most researchers have constructed their models (1) Geopolymers can behave differently because of the
only based on their experiment results [24, 28–33]. In chemistry of the binding material. Basically, geo-
return, these models seem to be fit only for a limited range polymers can be divided into two main groups,
of compressive strength. Some researchers considered the namely, low-calcium-based geopolymers, CaO <8%
previous results to build the prediction models [27, 34–37]. [37], and high-calcium-based geopolymers, CaO
However, these models also have flaws. The prediction ≥8% [37]. This division is due to the difference in the
model proposed by Prachasaree et al. and Xie et al. had reaction mechanism. In low-calcium-based geo-
neither been tested against experimental data nor provided polymers, the reaction mechanism is classified as
statistics to assure their reliability. In contrast, the pre- pozzolanic, while high-calcium-based geopolymers
diction model proposed by Prachasaree et al. [27] improved undergo hydraulic reactions as well [37]. The gel type
their previous model [35]; however, the new model pro- produced in the high-calcium geopolymers, such as
vided only a small difference in statistics compared to the ground granulated blast furnace slag based geo-
prediction model proposed by Nath and Sarker [34]; the polymers, is C-A-S-H. On the other hand, low-
error percentage is still considerably high. In the con- calcium geopolymers such as class F fly ash based
struction of prediction models proposed by Nath and geopolymers, the gel type of N-A-S-H [40, 41], are
Sarker and Cui et al. [34, 36], only a few data were selected produced. Therefore, it should be noted that only
without clarifying why those few data particularly have low-calcium-based geopolymers will be dealt with in
been chosen and leaving the rest of the previous data, which the formulation of this model.
made these models subjected to bias. Therefore, it is (2) The type of concrete should be normal unit weight
necessary to build a prediction model based on firm and and normal compressive strength concrete. High-
explicit assumptions by incorporating all the data of in- strength concretes are very rigid materials with high
terest, giving engineers a clear idea when applying this modulus of elasticity values and with very low elastic
model in practical applications. deformation. When the specified design compressive
Advances in Materials Science and Engineering 3

strength is 55 MPa or more, concrete is considered respectively, while the standard deviation values are
high-strength and exhibits different material prop- 12.15 MPa and 6.55 GPa, respectively.
erties [42]. Therefore, the samples with a specified
design compressive strength in the range of 55 MPa 5. Results and Discussion
or more will be excluded from the proposed model’s
formulation and the future applications of this 5.1. Initial Modelling. The first step of the modelling is to
model. investigate the proper relationship that governs and fits the
modulus of elasticity, Ec, in terms of the corresponding
(3) Compressive strength and modulus of elasticity
compressive strength, fc′. This step can be accomplished by
values are the mean values of specimens [21]. In cases
running the analysing data in the SPSS software. The analysis
where the authors give the values of specimens for
is performed as shown in the flowchart in Figure 2. The
compressive strength and modulus of elasticity, the
results of the analysis are shown in Table 2. The results show
sample means are taken.
that the highest value of the coefficient of determination, R-
(4) The cylindrical compressive strength, fc′, will be squared, corresponds to the power relationship, equal to
related to the modulus of elasticity, Ec, in the for- 0.809, with initial values of constants equal to 1.094 and
mulation of the proposed model. In cases where the 0.809. The p values are ≤ 0.001; thus, the variables are
authors gave the cubic compressive strength and statistically significant and related. Therefore, the governing
performed the modulus of elasticity on cylindrical equation of the modulus of elasticity is mathematically
specimens, a correlation factor for the normal expressed as
compressive strength should be considered [43]. All
samples should be at 28 days of age, or they are Ec � a ∗ f′b
c ,
(1)
reported to have an equivalent hardening rate to 28
where Ec is the modulus of elasticity in GPa, fc′ is the cy-
days of age [44].
lindrical compressive strength in MPa, and a and b are
(5) The confidence interval that can be seen in the lit- constants. This relationship complies with the form of ACI
erature is from 99.9% to 90.0%. However, it is tra- 318-14 and most of the other previous models in the lit-
ditionally constructed as 95.0% [45]. This value, erature [21].
95.0%, will be set as a confidence interval in the
construction of the proposed model.
5.2. Final Modelling. After selecting the proper governing
(6) The software SPSS version 23 will be employed to equation and its initial values, the final model can be ob-
perform the statistical work of this research. tained by running these results in SPSS software by following
the flowchart shown in Figure 3. The results of the analysis
4. Data Collection, Classification, and Testing are shown in Table 3. The final values of model parameters, a
and b, are 1.088 and 0.815, respectively. The determination
The data collected from peer-reviewed literature consist of coefficient, R-squared, of the model is 0.773.
67 different mix-design samples, in accordance with the The final modelling formula is expressed as
aforementioned assumptions, which are displayed in Ta-
ble 1. Most of the data (59 samples, analysing data) are used Ec � 1.088 ∗ f′0.815
c , (2)
to construct the proposed model. In comparison, some data
(8 samples, testing data) are selected to be testing samples to where Ec is the modulus of elasticity in GPa and fc′ is the
examine the goodness of fitness and compare the proposed cylindrical compressive strength in MPa.
model to the common and most recent models. Both confidence bands and prediction bands are con-
structed by applying equations (3) and (4), respectively, with
a 95% probability [54, 55].
4.1. Testing the Analysing Data for Normality. The data of the 􏽳������������ 􏽳�������������
2
samples are considered representative of the population 􏽐 Y − YPred 􏼁 1 (X − X)2 (3)
YPred. ± t0.05 · + ,
when they follow the same distribution. According to the n−2 n 􏽐(X − X)2
central limit theorem [52], most natural and physical
􏽳������������ 􏽳����������������
phenomena follow a normal distribution. As a result, the 2
⎢ 􏽐 Y − YPred 􏼁 ⎥⎥⎤ 1 (X − X)2
data should be checked for normal distribution. The best YPred. ± t0.05 ⎢⎡⎣1 + ⎦· 1+ + ,
technique to perform the normality test is the quantile- n−2 n 􏽐(X − X)2
quantile plot [53]. The outcome of the normality check is (4)
shown in Figure 1 for both the compressive strength, fc′, in
MPa and the modulus of elasticity, Ec, in GPa. Since the data where YPred is the predicted value of the dependent variable,
adhered to the straight line, the data are normally distributed t is the critical value for 95% of probability, Y is the true value
for the independent variable, the compressive strength, and of the dependent variable, n is the number of samples, and X
the dependent variable, the modulus of elasticity. Therefore, is the mean value of the independent variable of X.
these samples are considered representative of the pop- The relationship between the compressive strength, fc′,
ulations. The mean values of compressive strength and and modulus of elasticity of low-calcium-based geopolymer
modulus of elasticity are 36.79 MPa and 20.284 GPa, concrete, Ec, is described in Figure 4 with the corresponding
4 Advances in Materials Science and Engineering

Table 1: Data collected to generate the regression analysis and test the proposed model.
Sample ID Reference fc′ (MPa) Ec (GPa) Remarks
M26 [24] 44.0 23.00
Solution N 31.7 11.90
[25] The sample means have been taken
Solution W 41.5 17.10
4 40.4 28.60
5 47.6 29.48
6 46.7 29.36
7 46.8 28.52
8 46.1 26.46
10 47.4 25.64
11 12.2 7.04
12 12.8 6.81
13 20.7 7.96
14 [30] 10.3 7.46
15 46.6 28.74
16 5.5 4.62
19 43.4 25.61
26 20.2 11.27
27 52.6 28.09
28 32.5 22.81
29 23.4 13.18
30 22.7 12.31
32 36.8 24.73
U1 52.2 29.13
U2 [46] 54 32.10
U3 53.8 30.62
GPC0 [47] 45.3 24.00
GPC1–60 50.8 25.10
GPC1–90 54.4 25.40
GPC2–60 52.04 21.90
GPC2–90 [48] 53.6 25.50
GPC3–60 46.34 22.80
GPC3–90 49.73 22.60
GPC3–120 52.43 28.00
A40S00 25.6 17.40
A40S10 38.3 22.60
A40S15 46.6 24.60
A35S00 32.5 19.80
A35S10 33.3 19.20
A40P06 [34] 43.2 23.20
A40P08 34.4 20.60
A35P06 35.3 21.40
A40C02 42 22.40
A40C03 41.5 21.60
A35C02 36.8 22.20
M-0 26.8 11.84
M-1 29.4 14.79 A correlation factor has been used to
[49]
M-2 30.8 15.44 convert cubic strength to cylindrical
M-3 32.8 16.93
1 35.4 24.29
2 29 21.25
3 46.2 21.19
4 36.1 18.47
5 40.5 18.95
6 41.4 18.87
[36] The sample means have been taken
7 33.5 16.74
8 39.1 18.2
9 35 16.97
10 37.1 17.78
11 37.7 16.99
12 37.2 17.9
Advances in Materials Science and Engineering 5

Table 1: Continued.
Sample ID Reference fc′ (MPa) Ec (GPa) Remarks
M1 12.4 13.36 A correlation factor has been used to
[33]
M2 13.7 14.3 convert cubic strength to cylindrical
GP-1 [50] 15.024 7
P1 43.76 25.8
A correlation factor has been used to
P2 27.92 21.5
[51] convert cubic strength to cylindrical
P3 40.08 24.6
P4 38.16 23

70 40

60
Quantiles of Input Sample

Quantiles of Input Sample


30
50

40
20
30

20
10
10

0 0
-2 -1 0 1 2 -2 -1 0 1 2
Standard Normal Quantiles Standard Normal Quantiles
(a) (b)

Figure 1: Normality distribution check: (i) compressive strength, fc′ (MPa), and (ii) modulus of elasticity, Ec (GPa). (a) Normal Q-Q plot of
fc′(MPa). (b) Normal Q-Q plot of Ec (GPa).

Analysing data Analyse Regression

Define variables
Run the Curve estimation
analysis Select all
the available
relationships

Figure 2: The initial modelling process in SPSS.

Table 2: Model summary and parameter estimates (the dependent variable is Ec_GPa, and the independent variable is Fc_MPa).
Equation R-squared p value Constant, a b1 b2 b3
Linear 0.772 ≤0.001 2.977 0.471
Logarithmic 0.723 ≤0.001 −21.513 11.869
Inverse 0.531 ≤0.001 26.480 −183.139
Quadratic 0.773 ≤0.001 1.997 0.544 −0.001
Cubic 0.775 ≤0.001 4.754 0.177 0.012 0.000
Compound 0.769 ≤0.001 6.204 1.031
Power 0.808 ≤0.001 1.094 0.809
S 0.687 ≤0.001 3.394 -13.425
Growth 0.769 ≤0.001 1.825 0.030
Exponential 0.769 ≤0.001 6.204 0.030
Logistic 0.769 ≤0.001 0.161 0.970
6 Advances in Materials Science and Engineering

Analysing data Analyse Regression

Define variables
Run the Define parameters Nonlinear
analysis
Define model
expression

Figure 3: The process of the final modelling in SPSS.

Table 3: Results of final modelling for the dependent variable Ec_GPa.


Iteration historyb
Parameters
Iteration no.a Residual sum of squares
a b
1.0 581.108 1.094 0.809
1.1 575.086 1.088 0.815
2.0 575.086 1.088 0.815
2.1 575.086 1.088 0.815
3.0 575.086 1.088 0.815
3.1 575.086 1.088 0.815
Parameter estimates
95% confidence interval
Parameter Estimate Std. error Lower
Upper bound
bound
a 1.088 0.308 0.470 1.705
B 0.815 0.076 0.662 0.967
ANOVAc
Source Sum of squares df Mean squares
Regression 23632.371 2 12946.106
Residual 523.877 57 10.089
Uncorrected total 24156.249 59
Corrected total 2294.661 58
Derivatives are calculated numerically. aMajor iteration number is displayed to the left of the decimal, and the minor iteration number is to the right of the
decimal. bRun stopped after 6 model evaluations and 3 derivative evaluations because the relative reduction between successive residual sums of squares is at
most SSCON � 1.000E − 8. cR-squared � 1 − (residual sum of squares)/(corrected sum of squares) � 0.773.

95% confidence bands and 95% prediction bands. It can be respectively. Therefore, the proposed model gives a 52.2%
seen that the 95% confidence bands are tight and narrow, reduction in MAPE from the least error model, Prachasaree
confining the proposed model through a long distance with et al. [27].
only a slight expansion at the edges. This means that the true Moreover, the data have been run by neural networks in
regression line is very close to the proposed model line with a the SPSS software. The optimised neural network archi-
probability of 95% [55]. The proposed model also performs tecture was one hidden layer with two neurons for the
well in terms of prediction with a 95% probability. The 95% multilayer perceptron neural network type. The activation
prediction bands confine all the experimental data with function for the hidden layer was the hyperbolic tangent. The
much inside space between the outer experimental data and results of the neural network analysis have yielded an R-
the prediction bands. squared value of 0.752 and an RMSE value of 4.764 MPa. As
The root mean squared error (RMSE) of the proposed a result, the neural network results are comparable to the
equation is equal to 3.122. This value is much lower than the proposed model but with less accuracy since the RMSE and
results of the most recent findings, Lee and Lee [31], Nath R-squared values are 3.122 MPa and 0.773 for the proposed
and Sarker [34], and Prachasaree et al. [27], which are equal model in this study.
to 8.900, 6.01, and 5.260, respectively. Therefore, the pro-
posed model gives a 41% reduction of RMSE from the least
error model [27]. The mean absolute percentage error 5.3. Error Evaluation. In addition to the RMSE, R-squared,
(MAPE) of the proposed equation is equal to 15.00%, which and MAPE criteria, other error evaluation criteria such as
is much lower than the results in the most recent findings, the correlation coefficient (R), mean absolute error (MAE),
Lee and Lee [31], Nath and Sarker [34], and Prachasaree relative squared error (RSE), and relative root mean squared
et al.[27], which are equal to 40.2%, 35.0%, and 31.4%, error (RRMSE) have been calculated as recommended by
Advances in Materials Science and Engineering 7

Ec = 1.088(f′c)0.815 Table 4: Statistical parameters to evaluate the goodness of fitness of


40 R2 = 0.773 the proposed model.
Statistical parameter Value
Elastic modulus, Ec, in GPa.

30 R 0.880
R-squared 0.773
RMSE 3.122
20
MAE 2.632
RSE 0.290
10 MAPE 0.150
RRMSE 0.155

0
0 10 20 30 40 50 60
8.00
Compressive strength, f′c, in MPa.
6.00
Experimental data 95% prediction bands
Proposed model 95% confidence bands 4.00

Figure 4: The relationship between compressive strength, fc′, and 2.00

Residuals
modulus of elasticity, Ec, of geopolymer concrete. R2 = 0.0002
0.00
0.00 5.00 10.00 15.00 20.00 25.00 30.00
-2.00
Khan et al. [56] and Li et al. [57], and the results are shown in
-4.00
Table 4. The R and R-squared are indications of the cor-
relation between the proposed model and the actual values of -6.00
the modulus of elasticity of low-calcium geopolymer con-
-8.00
crete. The closer the values are to 1.0 of the R-squared for a Fitted values
model, the higher the variation by the independent variable Figure 5: Distribution of the residuals against fitted values.
is reflected [56]. When the R of a model is greater than 0.8,
then the model is highly correlated [56]. The proposed
model showed R and R-squared values equal to 0.88 and 5.4. Testing the Proposed Model against Overfitting and
0.773, respectively. As a result, the proposed model has a Comparison with the Previous Models. The testing data, 8
high correlation and represents a high percentage of vari- samples separated earlier from the collected data, are
ation. On the other hand, RMSE, MAPE, RSE, and RRMSE employed to evaluate the proposed model against overfitting
are evaluation criteria for the error of a model, and they and its advantage over the previous models described below:
reflect the deviation of the predicted value from the actual
Hardjito et al. [38]:

values [58]. The proposed model is very close to the actual
values, as it has only 2.632 MPa and deviates only 3.122 MPa
Ec  2707 fc′ + 5300, (5)
on average in terms of MAE and RMSE, respectively. The
model accuracy is also expressed by the RRMSE criterion.
When RRMSE <10%, a model is classified as excellent, where Ec and fc′ are in MPa.
10 < RRMSE < 20 is good, 20 < RRMSE < 30 is fair, and Diaz and Eleazar [30]:
RRMSE >30 is poor [58]. Interestingly, the proposed model
Ec  580 fc′, (6)
demonstrated 15.5% of RRMSE, which is classified as good.
Moreover, the error of the proposed model can be
where Ec and fc′ are in MPa.
evaluated by plotting the residuals against the fitted values of
the dependent variable. The residuals are the differences Lee and Lee [31]:
between the experimental values and the predicted values 
3
Ec  5300 fc′, (7)
obtained by the proposed model. The fitted values are the
values predicted by the proposed model. It can be seen from
Figure 5 that the residuals are scattered in a symmetrical and where Ec and fc′ are in MPa.
balanced way about the horizontal axis. Moreover, the outer Nath and Sarker [34]:
points of the residuals perform almost a circle shape. This 
distribution confirms that the residuals are independent of Ec  3.51 fc′, (8)
each other and randomly distributed around the centreline
[54]. The determination coefficient, R-squared, is calculated where Ec is in GPa and fc′ is in MPa.
to ensure this situation. It is found to be almost equal to zero Xie et al. [37]:
(0.0002). That means that the residuals are independent of 
each other. Therefore, the goodness of fitness of the pro- Ec  3.65 fc′, (9)
posed model is very well.
8 Advances in Materials Science and Engineering

Table 5: The results of the comparison between the previous models and the proposed model.
Sample fc′ Ec (Gpa) Equation (5)/ Equation (6)/ Equation (7)/ Equation (8)/ Equation (9)/ Equation Equation (2)/
ID (Mpa) exp. exp. exp. exp. exp. exp. (10)/exp. exp.
19 43.40 25.61 0.90 0.98 0.73 0.90 0.94 1.02 0.92
26 20.20 11.27 1.55 1.04 1.28 1.40 1.46 1.31 1.12
U2 54.00 32.10 0.78 0.98 0.62 0.80 0.84 0.96 0.88
GPC160 50.80 25.10 0.98 1.17 0.78 1.00 1.04 1.18 1.06
A35P06 35.30 21.40 1.00 0.96 0.81 0.97 1.01 1.05 0.93
M-2 30.80 15.44 1.32 1.16 1.08 1.26 1.31 1.31 1.15
3 46.2 21.19 1.12 1.26 0.90 1.13 1.17 1.30 1.17
4 36.10 18.47 1.17 1.13 0.95 1.14 1.19 1.24 1.10
Max. 1.55 1.26 1.28 1.40 1.46 1.31 1.17
Min. 0.78 0.96 0.62 0.80 0.84 0.96 0.88

where Ec is in GPa and fc′ is in MPa.


35.0
Prachasaree et al. [27]:
Ec  1.55  fc′ ,
3/4
(10)

where Ec is in GPa and fc′ is in MPa. 25.0


The predicted values of modulus of elasticity are cal- Ec (GPa)
culated by the models from equations (2) and (5)–(10), and
the results are presented in Table 5.
The results show that equation (5) overpredicts the 15.0
modulus of elasticity. The predicted value of the modulus
of elasticity reached 55% of overprediction. The authors
did not take into account the compressive strength class in
the development of their model [38]. Equation (6) gives
only a 26% overprediction. This result is because the 5.0
15.0 25.0 35.0 45.0 55.0
author developed equation (6) depending on relatively
f′c (MPa)
large data. However, the error percentage value did not
reach the best possible value because the author did not Ec_Exp. (Gpa)
take into account the chemical composition. The author Proposed_Model
treated the class F fly ash and class C fly ash the same, (±) 15% bands
while the first mentioned is low-calcium fly ash and the
Figure 6: The proposed model and ±15% bands against the testing
latest is high-calcium fly ash [30]. Equation (7) shows 38%
data and Ec_Exp (GPa).
of underprediction. The authors only included their ex-
perimental data, which were relatively a few. Moreover,
their experimental data were mixed with low-calcium and
high-calcium precursors [31]. Equation (8) overpredicts considerably low percentage of error. The maximum
the testing data by up to 40%. Although the authors in- overprediction is +17%, which is very close to the MAPE
cluded some data from previous studies, the error per- value (15%). On the other hand, the minimum prediction
centage was still relatively high. It was noticed that the is −12% of the error percentage, which is very close to the
authors did not consider the compressive strength class MAPE as well; therefore, the proposed model gives
[34]. However, their data were not sufficient to generate a overprediction and underprediction of 15% on average.
model with a low percentage of error. Although equation This means that the proposed model evidently shows
(9) was introduced after equation (8), it did not reduce the consistency in terms of overprediction and under-
error percentage. It overpredicts the testing data by up to prediction. Therefore, the upper and lower bands of ±15%
46% [37]. Equation (10) shows a 31% overprediction. The are constructed versus the testing data. As shown in
authors took into account a wide range of data from the Figure 6, all testing data were confined between the upper
literature to generate their model, but they included low- and lower bands. Therefore, the proposed model is valid
calcium based geopolymer and high-compressive strength for the application of future data and it is not overfitted.
data [27]. Therefore, they did not come up with the best
possible value of the error percentage. 6. Conclusions
All previous models present a considerably high
percentage of error. Moreover, they tend to either over- A comprehensive regression analysis of the modulus of
predict or underpredict the modulus of elasticity values. elasticity of low-calcium-based geopolymer concrete has
On the contrary, the proposed model shows a been conducted on 59 different mix-design samples and
Advances in Materials Science and Engineering 9

tested by other 8 different mix-design samples; all are col- References


lected from the peer-reviewed literature based on firm as-
sumptions such as classifying the geopolymers as low- [1] J. Davidovits, “Geopolymers,” Journal of Thermal Analysis,
calcium and specifying the strength class and density of vol. 37, no. 8, pp. 1633–1656, 1991.
[2] P. Duxson, S. W. Mallicoat, G. C. Lukey, W. M. Kriven, and
concrete. Therefore, the following conclusions are drawn:
J. S. J. van Deventer, “The effect of alkali and Si/Al ratio on the
(i) The proposed model performs very well in pre- development of mechanical properties of metakaolin-based
dicting the modulus of elasticity when taking into geopolymers,” Colloids and Surfaces A: Physicochemical and
account the chemical composition in terms of low- Engineering Aspects, vol. 292, no. 1, pp. 8–20, 2007.
and high-calcium contents and compressive [3] A. L. Almutairi, B. A. Tayeh, A. Adesina, H. F. Isleem, and
strength class. It reduced RMSE from ±5.260 GPa to A. M. Zeyad, “Potential applications of geopolymer concrete
in construction: a review,” Case Studies in Construction
±3.122 GPa and MAPE from 31.4% to 15.0%
Materials, vol. 15, Article ID e00733, 2021.
compared to the best performance model found in [4] S. M. S. Taher, S. T. Saadullah, J. H. Haido, and B. A. Tayeh,
the literature described by Prachasaree et al. [27]. “Behavior of geopolymer concrete deep beams containing
Therefore, the proposed model gives a 41% and waste aggregate of glass and limestone as a partial replacement
52.2% reduction of RMSE and MAPE, respectively. of natural sand,” Case Studies in Construction Materials,
(ii) The proposed model has a high correlation with the vol. 15, Article ID e00744, 2021.
actual values since the value of R (0.88) > 0.8. [5] J. Davidovits, “Geopolymer cements to minimize carbon-
dioxide greenhouse-warming,” Ceramic Transactions, vol. 37,
(iii) The goodness of fitness of the proposed model is pp. 165–182, 1993.
characterised as good since its RRMSE (15.5) [6] M. S. Saif, M. O. R. El-Hariri, A. I. Sarie-Eldin, B. A. Tayeh,
< 20.0%. and M. F. Farag, “Impact of Ca+ content and curing condition
on durability performance of metakaolin-based geopolymer
7. Future Recommendations mortars,” Case Studies in Construction Materials, vol. 16,
Article ID e00922, 2022.
This comprehensive regression analysis study has yielded a [7] M. Amin, A. M. Zeyad, B. A. Tayeh, and I. S. Agwa, “Effect of
reliable proposed model to predict and visualise the rela- high temperatures on mechanical, radiation attenuation and
microstructure properties of heavyweight geopolymer con-
tionship between the modulus of elasticity of low-calcium-
crete,” Structural Engineering & Mechanics, vol. 80,
based geopolymer concrete and its compressive strength. It pp. 181–199, 2021.
is recommended to utilise the proposed model in the pre- [8] S. M. A. Qaidi, B. A. Tayeh, A. M. Zeyad, A. R. G. de Azevedo,
diction of the modulus of elasticity if and only if the basic W. Emad, and W. Emad, “Recycling of mine tailings for the
assumptions of deriving the proposed model equation, geopolymers production: a systematic review,” Case Studies in
which are stated in Section 3, are met. Construction Materials, vol. 16, Article ID e00933, 2022.
[9] S. M. A. Qaidi, B. A. Tayeh, H. F. Isleem, A. R. G. de Azevedo,
W. Emad, and W. Emad, “Sustainable utilization of red mud
Data Availability waste (bauxite residue) and slag for the production of geo-
polymer composites: a review,” Case Studies in Construction
The data used to support the findings of this research study Materials, vol. 16, Article ID e00994, 2022.
are included within the article. [10] B. A. Tayeh, A. Hakamy, M. Amin, A. M. Zeyad, and
I. S. Agwa, “Effect of air agent on mechanical properties and
Conflicts of Interest microstructure of lightweight geopolymer concrete under
high temperature,” Case Studies in Construction Materials,
The authors declare that there are no conflicts of interest vol. 16, Article ID e00951, 2022.
regarding the publication of this paper. [11] B. A. Tayeh, A. M. Zeyad, I. S. Agwa, and M. Amin, “Effect of
elevated temperatures on mechanical properties of light-
weight geopolymer concrete,” Case Studies in Construction
Acknowledgments Materials, vol. 15, Article ID e00673, 2021.
[12] M. M. Ahmed, K. A. M. El-Naggar, D. Tarek et al., “Fabri-
The Stipendium Hungaricum Scholarship Programme is cation of thermal insulation geopolymer bricks using ferro-
highly acknowledged for supporting the PhD study and silicon slag and alumina waste,” Case Studies in Construction
research work (SHE-09408-004/2019). The authors ac- Materials, vol. 15, Article ID e00737, 2021.
knowledge the support of the Hungarian Research Grant [13] A. M. Zeyad, H. M. Magbool, B. A. Tayeh, A. R. Garcez de
NVKP 16-1-0019: “Development of Concrete Products with Azevedo, A. Abutaleb, and Q. Hussain, “Production of
Improved Resistance to Chemical Corrosion, Fire or Freeze- geopolymer concrete by utilizing volcanic pumice dust,” Case
Studies in Construction Materials, vol. 16, Article ID e00802,
Thaw.” They also acknowledge the support of the Stiftung
2022.
Aktion Österreich-Ungarn in the framework of the bilateral [14] F. Aslam, M. A. Elkotb, A. Iqtidar et al., “Compressive
research cooperation project no. 108öu3 between TU Wien strength prediction of rice husk ash using multiphysics ge-
and the Budapest University of Technology and Economics. netic expression programming,” Ain Shams Engineering
This research was partially funded by the Stipendium Journal, vol. 13, no. 3, Article ID 101593, 2022.
Hungaricum Scholarship Programme as part of the PhD [15] M. A. Khan, S. A. Memon, F. Farooq, M. F. Javed, F. Aslam,
study of the first author. and R. Alyousef, “Compressive strength of fly-ash-based
10 Advances in Materials Science and Engineering

geopolymer concrete by gene expression programming and blast furnace slag blended geopolymer concrete,” Materials
random forest,” Advances in Civil Engineering, vol. 2021, Today Proceedings, vol. 27, pp. 718–723, 2020.
Article ID 6618407, 17 pages, 2021. [34] P. Nath and P. K. Sarker, “Flexural strength and elastic
[16] M. A. Khan, F. Farooq, M. F. Javed et al., “Simulation of depth modulus of ambient-cured blended low-calcium fly ash
of wear of eco-friendly concrete using machine learning based geopolymer concrete,” Construction and Building Materials,
computational approaches,” Materials, vol. 15, no. 1, vol. 130, pp. 22–31, 2017.
pp. 58–28, 2021. [35] W. Prachasaree, S. Limkatanyu, A. Hawa, and
[17] A. A. Khalaf, K. Kopecskó, and I. Merta, “Prediction of the A. Samakrattakit, “Development of equivalent stress block
compressive strength of fly ash geopolymer concrete by an parameters for fly-ash-based geopolymer concrete,” Arabian
optimised neural network model,” Polymers, vol. 14, no. 7, Journal for Science and Engineering, vol. 39, no. 12,
p. 1423, 2022. pp. 8549–8558, 2014.
[18] A. M. Neville, Properties of concrete, Pearson Education India, [36] Y. Cui, K. Gao, and P. Zhang, “Experimental and statistical
Essex, England, 5th edition, 2011. study on mechanical characteristics of geopolymer concrete,”
[19] R. Davidson and J. G. MacKinnon, Econometric Theory and Materials, vol. 13, no. 7, p. 1651, 2020.
Methods, Oxford University Press, Oxford, UK, 2002. [37] T. Xie, P. Visintin, X. Zhao, and R. Gravina, “Mix design and
[20] K. Dhir, S. Ghataora, and J. Lynn, “Concrete-Related Ap- mechanical properties of geopolymer and alkali activated
plications,” Sustainable Construction Materials, Woodhead concrete: review of the state-of-the-art and the development
Publishing, Duxford, UK, 2017. of a new unified approach,” Construction and Building Ma-
[21] ACI 318-14, Building Code Requirements For Structural terials, vol. 256, Article ID 119380, 2020.
Concrete and Commentary, American Concrete Institute, [38] D. Hardjito, S. E. Wallah, D. M. J. Sumajouw, and
Farmington Hills, MI, USA, 2014. B. V. Rangan, “The stress-strain behaviour of fly ash-based
[22] Fib (International Federation for Structural Concrete), Model geopolymer concrete,” in Development in Mechanics of
Code 2010, Wilhelm Ernst & Sohn, Lausanne, Switzerland, Structures and Materials, A.J. Decks and H. Hao, Eds.,
2013. pp. 831–834, A.A. Balkema Publishers, Leiden, South Holland
[23] “Portland cement association types and causes of concrete Australia, 2004.
deterioration,” Portl. Cem. Assoc. - Concr. Inf.vol. 1–16, 2002. [39] S. You, S. W. Ho, T. Li, T. Maneerung, and C.-H. Wang,
[24] D. Hardjito and B. V. Rangan, “Development and Properties “Techno-economic analysis of geopolymer production from
of Low-Calcium Fly Ash-Based Geopolymer concrete,” Re- the coal fly ash with high iron oxide and calcium oxide
search Report GC, Faculty of Engineering Curtin University contents,” Journal of Hazardous Materials, vol. 361,
of Technology, Perth, Australia, 2005. pp. 237–244, 2019.
[25] A. Fernandez-Jimenez, A. Palomo, and C. Lopez-Hombrados, [40] J. L. Provis and J. S. J. van Deventer, “Geopolymerisation
“Engineering properties of alkali-activated fly ash concrete,” kinetics. 2. Reaction kinetic modelling,” Chemical Engineering
ACI Materials Journal, vol. 103, 2006. Science, vol. 62, no. 9, pp. 2318–2329, 2007.
[26] M. Albitar, P. Visintin, M. S. Mohamed Ali, and M. Drechsler, [41] J. L. Provis and J. S. J. van Deventer, “Geopolymerisation
“Assessing behaviour of fresh and hardened geopolymer kinetics. 1. In situ energy-dispersive X-ray diffractometry,”
concrete mixed with class-F fly ash,” KSCE Journal of Civil Chemical Engineering Science, vol. 62, no. 9, pp. 2309–2317,
Engineering, vol. 19, no. 5, pp. 1445–1455, 2014. 2007.
[27] W. Prachasaree, S. Limkatanyu, A. Hawa, P. Sukontasukkul, [42] ACI 363R, State-of-the-Art Report on High-Strength Concrete
and P. Chindaprasirt, “Manuscript title: development of Reported by ACI Committee 363, Vol. 92, American Concrete
strength prediction models for fly ash based geopolymer Institute, Farmington Hills, MI, USA, 2005.
concrete,” Journal of Building Engineering, vol. 32, Article ID [43] BS En 206, BSI Standards Publication Concrete — Specification,
101704, 2020. Performance , Production and Conformity, British Standards
[28] M. Sofi, J. S. J. van Deventer, P. A. Mendis, and G. C. Lukey, Institution, London, UK, 2014.
“Engineering properties of inorganic polymer concretes [44] D. Hardjito, S. E. Wallah, D. M. J. Sumajouw, and
(IPCs),” Cement and Concrete Research, vol. 37, no. 2, B. V. Rangan, “On the development of fly ash-based geo-
pp. 251–257, 2007. polymer concrete,” ACI Materials Journal, vol. 101, no. 6,
[29] B. Tempest, Engineering Characterization of Waste Derived pp. 467–472, 2004.
Geopolymer Cement Concrete for Structural Applications, The [45] A. F. Siegel, “Confidence intervals,” Practical Business Sta-
University of North Carolina at Charlotte, Ann Arbor, USA, tistics, vol. 101, no. 6, pp. 219–247, 2012.
2010. [46] J. R. Yost, A. Radlińska, S. Ernst, and M. Salera, “Structural
[30] L. Diaz and I. Eleazar, “Development of Approximating behavior of alkali activated fly ash concrete. Part 1: mixture
Functions to Model and Predict the Properties of Fresh and design, material properties and sample fabrication,” Materials
Hardened Fly Ash-Based Geopolymer concrete,” Disserta- and Structures, vol. 46, no. 3, pp. 435–447, 2013.
tions, ProQuest Dissertations Publishing, Ann Arbor, [47] F. U. A. Shaikh, “Mechanical and durability properties of fly
Michigan, USA, 2011. ash geopolymer concrete containing recycled coarse aggre-
[31] N. K. Lee and H. K. Lee, “Setting and mechanical properties of gates,” International Journal of Sustainable Built Environment,
alkali-activated fly ash/slag concrete manufactured at room vol. 5, no. 2, pp. 277–287, 2016.
temperature,” Construction and Building Materials, vol. 47, [48] K. Neupane, “Investigation on modulus of elasticity of
pp. 1201–1209, 2013. powder-activated geopolymer concrete,” International Jour-
[32] R. J. Thomas and S. Peethamparan, “Alkali-activated concrete: nal of Structural Engineering, vol. 7, no. 3, p. 262, 2016.
engineering properties and stress-strain behavior,” Con- [49] Y. Wang, S. Hu, and Z. He, “Mechanical and fracture
struction and Building Materials, vol. 93, pp. 49–56, 2015. properties of fly ash geopolymer concrete addictive with
[33] R. R. Bellum, K. Muniraj, and S. R. C. Madduru, “Investi- calcium aluminate cement,” Materials, vol. 12, no. 18, p. 2982,
gation on modulus of elasticity of fly ash-ground granulated 2019.
Advances in Materials Science and Engineering 11

[50] S. Saravanan and S. Elavenil, “Strength properties of geo-


polymer concrete using msand by assessing their mechanical
characteristics,” ARPN J. Eng. Appl. Sci.vol. 13, pp. 4028–4041,
2018.
[51] A. Albidah, A. S. Alqarni, H. Abbas, T. Almusallam, and Y. Al-
Salloum, “Behavior of Metakaolin-Based geopolymer con-
crete at ambient and elevated temperatures,” Construction
and Building Materials, vol. 317, p. 125910, 2022.
[52] V. Krishnan, Probability and Random Processes, John Wiley &
Sons, Hoboken, NJ, USA, 2006.
[53] V. Grech and N. Calleja, “WASP (write a scientific paper):
parametric vs. non-parametric tests,” Early Human Devel-
opment, vol. 123, pp. 48-49, 2018.
[54] H. J. Motulsky and L. A. Ransnas, “Fitting curves to data using
nonlinear regression: a practical and nonmathematical re-
view,” The FASEB Journal, vol. 1, no. 5, pp. 365–374, 1987.
[55] A. M. Brown, “A step-by-step guide to non-linear regression
analysis of experimental data using a Microsoft Excel
spreadsheet,” Computer Methods and Programs in Biomedi-
cine, vol. 65, no. 3, pp. 191–200, 2001.
[56] S. Khan, M. Ali Khan, A. Zafar et al., “Predicting the ultimate
axial capacity of uniaxially loaded cfst columns using mul-
tiphysics artificial intelligence,” Materials, vol. 15, no. 1,
pp. 39–27, 2021.
[57] P. Li, M. Ali Khan, A. M. Galal et al., “Sustainable use of
chemically modified tyre rubber in concrete: machine
learning based novel predictive model,” Chemical Physics
Letters, vol. 793, Article ID 139478, 2022.
[58] M. Despotovic, V. Nedic, D. Despotovic, and S. Cvetanovic,
“Evaluation of empirical models for predicting monthly mean
horizontal diffuse solar radiation,” Renewable and Sustainable
Energy Reviews, vol. 56, pp. 246–260, 2016.

You might also like