Construction and Building Materials 106 (2016) 480–499

Contents lists available at ScienceDirect

Construction and Building Materials

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

Prediction of the mechanical properties of structural recycled concrete

using multivariable regression and genetic programming
Iris González-Taboada a, Belén González-Fonteboa a,⇑, Fernando Martínez-Abella a,
Juan Luis Pérez-Ordóñez b
a
Department of Construction Technology, University of A Coruña, E.T.S.I. Caminos, Canales, Puertos. Campus Elviña s/n, 15071 La Coruña, Spain
b
Department of Construction Technology, University of A Coruña, E.U. Arquitectura Técnica, Campus Zapateira s/n, 15071 La Coruña, Spain

h i g h l i g h t s

A database about recycled concrete (RC) has been developed with 1831 mixes from 81 papers.

Prediction of some of the most important properties of RC has been analyzed.

Code expressions regarding fsp and Ec prediction have been adjusted using correction coefficients.

Specific expressions used to predict these properties have been optimized.

Multivariable regression and genetic programming have been used for predictions optimization.

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

Article history: This study focuses on the prediction of some of the most important properties of structural recycled con-
Received 27 October 2015 crete (compressive strength, modulus of elasticity and splitting tensile strength) taking into account, not
Received in revised form 16 December 2015 only the recycled percentage and the quality of the recycled aggregates used, but also the production
Accepted 17 December 2015
method. For said purpose, a database has been developed with 1831 mixes obtained from 81 papers.
Firstly, in this manner, these properties have been compared with those of conventional concrete. Then,
the need to adapt the prediction code expressions (adjusted for conventional concretes) was analyzed to
Keywords:
take into account the use of recycled concrete, developing, if finally necessary, the correction coefficients
Structural recycled concrete
Database
which allow engineers to predict the recycled properties with the same approximation degree as in con-
Mixing procedure ventional concretes. These correction coefficients have been adjusted using multivariable regression, and
Mechanical properties have been analyzed using different statistical indexes.
Multivariable regression Lastly, specific expressions used to predict these properties in structural recycled concretes have been
Genetic programming optimized. Two different tools have been used to develop these expressions: multivariable regression and
genetic programming. The proposed expressions have been analysed using statistical parameters which
have been compared with those obtained using the expressions proposed by other authors. In this regard,
and finally, the best prediction expressions for the modulus of elasticity and the splitting tensile strength
of structural recycled concretes have been proposed.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction where the higher absorption of recycled aggregates (which influ-

The greatest difference between a recycled aggregate and a con-
ventional aggregate is the adhered mortar of the original concrete.
This is the cause of the main differences between the physical
properties of recycled aggregates and conventional aggregates,
⇑ Corresponding author.
E-mail addresses: iris.gonzalezt@udc.es (I. González-Taboada), bfonteboa@udc.
es (B. González-Fonteboa), fmartinez@udc.es (F. Martínez-Abella), jlperez@udc.es (J.
L. Pérez-Ordóñez).
ences fresh concrete properties, particularly its workability) should
be noted. Both this and the moisture state, affect the effective
water to cement ratio which finally influences the properties of
fresh and hardened concrete [1,2].
When recycled concrete is used, the authors propose three pro-
duction methods to compensate the high water absorption of recy-
cled aggregates. In the first method, aggregates are added after
being immersed in water for a pre-established time (pre-soaking),
generally 10 min [3–6]. In the second, aggregates are added with
their natural moisture, compensating the aggregate absorption

http://dx.doi.org/10.1016/j.conbuildmat.2015.12.136
0950-0618/Ó 2015 Elsevier Ltd. All rights reserved.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499
Nomenclature
Proc,CC,cal calculated property of conventional concrete using code
expressions
Proc,CC,exp experimental property of conventional concrete
Proc,RC,cal calculated property of recycled concrete using code
expressions
Proc,RC,exp experimental property of recycled concrete
%RCA replacement ratio (recycled coarse aggregate percent-
age)
WA recycled coarse aggregate water absorption
fc,CC conventional concrete compressive strength
fc,RC recycled concrete compressive strength
Ec,CC conventional concrete elasticity modulus
Ec,RC recycled concrete elasticity modulus
by adding extra water (extra water), generally the water necessary
to reach 80% saturation. Lastly, other authors produce concrete
without taking into account the high absorption of the recycled
aggregate (dry), and control workability with supplementary quan-
tities of superplasticizer.
Therefore, essentially, three alternatives have been proposed
[7–87]: (a) to pre-soak the recycled aggregates (pre-soaked-PS)
for a fixed time, (b) to work with air-dry aggregates increasing
the amount of water incorporated in the mix according to a fixed
percentage of the saturated surface dry state (air-dry with extra-
water-ADwEW), or (c) to work with air-dry aggregates without
any extra water but using a significant amount of superplasticizer
to maintain workability (air-dry without extra-water-AD).
It is obvious that the selected production method (i.e. the mix-
ing procedure) influences the recycled concrete properties. In fact,
recycled concrete presents two ITZ (interfacial transition zones),
one located between the original aggregate and the old adhered
mortar and the other between said adhered mortar and the new
mortar matrix [58] (Fig. 1). The strength of this concrete is nor-
mally influenced by the weakest one. It is agreed that the selected
production method modifies the ITZ quality between the recycled
coarse aggregate and the new paste [28].
In general terms, authors conclude that pre-saturated recycled
aggregates cause damage in the ITZ between the recycled coarse
aggregate and the new cement paste [12], which has a negative
effect on the concrete compressive strength [8,14,19]. In the case
of using recycled aggregates in the saturated surface dry state,
Old transition zone
Natural aggregate
New transition zone
Recycled aggregate
Old Paste
New Paste
Fig. 1. The recycled concrete interfacial transition zones (ITZ) [58].
481
fsp,CCconventional concrete splitting tensile strength
fsp,RCrecycled concrete splitting tensile strength
X adjusted coefficient (linear regression)
CF correction coefficient for the Eurocode expression (mul-
tivariable regression)
r Pearson correlation coefficient
MSE mean squared error
COV coefficient of variation
MAE mean average error
AD air dry without extra-water
ADwEW air dry with extra-water
PS pre-soaked
the high water content inside the aggregate particles may result
in bleeding. The water inside recycled aggregate particles can
move to the cement paste, creating a relatively high water to
cement ratio sited in the proximity of the particles. This process
may weaken the link between the recycled aggregate and the
cement matrix. Consequently, a high degree of bleeding leads to
a lower recycled concrete strength [30].
Current trends suggest that the ADwEW method prevents the
bleeding effect and hence, leads to better behavior than the PS
method [8]. Lastly, the AD method avoids, in general, the damage to
the ITZ although, nowadays, this method has almost been given up
because very high superplasticizer quantities are required to main-
tain workability and, in some cases, it may be impossible to achieve.
Lastly, some authors have confirmed that the moisture state of
recycled aggregates also introduces changes in the behavior of
recycled concrete [8,9,28,34]. According to Poon et al. [9], the ini-
tial slump of recycled concrete is strongly dependent on the initial
free water content.
Regarding all of the aforementioned, the production method
must be considered when analysing any property of recycled con-
crete. Hence, in this work, the main objective is to predict the basic
mechanical properties of structural recycled concrete taking into
account, not only the recycled percentage and quality of the recy-
cled aggregates used, but also the production method. Further-
more, this prediction must conclude with the proposal of simple
expressions which allow engineers to estimate properties of recy-
cled concrete similarly to how those of conventional concrete are
estimated.
2. Objectives
This study focuses on the influence of the recycled concrete pro-
duction method, and recycled coarse aggregate quantity and qual-
ity, on some of the most significant properties of recycled concrete
(compressive strength, modulus of elasticity and splitting tensile
strength). To achieve this objective, a database has been developed
[7–87] with 1831 mixes of 81 papers selected from over 250 inter-
national references related to the field of recycled concrete.
The database was developed considering works where the recy-
cled coarse aggregate used was obtained from concrete waste
(recycled concrete coarse aggregate). In this regard, all recycled
coarse aggregates have presented a water absorption capacity
under 8.5%. Furthermore, as different mechanical properties are
achieved as a function of the method used to compensate the high
water absorption capacity of recycled aggregates, the works have
been classified considering the mixing procedure (PS, ADwEW
and AD).
482 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499
The properties studied are as follows: compressive strength,
modulus of elasticity and splitting tensile strength. In the first case,
this study focuses on comparing the compressive strength of con-
ventional concrete with that of recycled concrete. The objective is
to discover how much this property can decrease when recycled
concrete coarse aggregates are used, by analysing different incor-
poration ratios and mixing procedures. The works used to develop
this analysis have to use a control concrete (conventional concrete)
and different recycled concretes made with the same dosage and
materials except for the coarse aggregate, which will be replaced
with recycled concrete coarse aggregate (by volume) at different
percentages.
Regarding the modulus of elasticity and splitting tensile
strength, there are three objectives: firstly, to compare these prop-
erties with those of conventional concrete. The only goal of this
method is to obtain information, as this way of analysing the prop-
erties of recycled concrete is not useful for predicting them as a
conventional concrete reference is needed. The second objective
is to analyze whether or not it is necessary to adapt the prediction
code expressions (adjusted for conventional concretes) to take into
account the use of recycled aggregates, developing, if finally neces-
sary, correction coefficients that allow engineers to predict recy-
cled properties with the same approximation degree as with
conventional concretes. These correction coefficients have been
adjusted using multivariable regression. Lastly, the third objective
is to optimize specific expressions to predict these properties in
structural recycled concretes. Two different tools have been used
to develop the expressions: multivariable regression and genetic
programming. Both the correction coefficients and the new specific
expressions must take into account the recycled concrete compres-
sive strength, production procedure, replacement ratio and recy-
cled concrete coarse aggregate quality.
The proposed expressions and also the correction coefficients
have been analyzed using different statistical indexes: the Pearson
correlation coefficient (r), the mean squared error (MSE), the coef-
ficient of variation (COV) and the mean average error (MAE).
Regarding the proposed expressions, the statistical parameters
have been compared with those obtained using expressions pro-
Fig. 2. Algorithm example [99].
posed by other authors. In this regard, and lastly, the best predic-
tion expressions for the modulus of elasticity and the splitting
tensile strength of structural recycled concretes have been
proposed.
3. Methodology
3.1. Genetic programming methodology
Genetic programming (GP) is a subset of solution search techniques based on
evolutionary computation. In GP, an analogy between the solution set to a problem
and the individual set in a natural population is established, codifying the informa-
tion of each solution using a tree-shaped structure. In this codification two types of
nodes are differentiated. The first type are non-terminal nodes or functions where
the algorithm operators are lodged (for example addition, subtraction, etc.). They
are characterized as always having one or more children. The second type are ter-
minal nodes or tree leaves, where the constant values and the previously defined
variables are located. These nodes have no children [99].
The most extensive GP application is probably to develop mathematical expres-
sions. Its codifying manner enables them to be easily represented and they have
been applied in many different engineering related fields [100]. The results that
were found have been very beneficial and, on a great number of occasions, the
expressions achieved were an improvement on previously existing ones [101].
In this article, the algorithm developed is based on classic genetic programming
techniques and has been specialized in the optimization of mathematical formulas
(Fig. 2). Therefore, the algorithm development has been focused on the symbolic
regression technique for learning patterns. Furthermore, a module has been added
to enable the incorporation of expert knowledge and to orientate the search
process.
The first step involved debugging the database. It may seem trivial but it is
essential to obtaining coherent and unbiased results. Having a complete and well
verified database is an essential condition for the process. Refinement was carried
out until a representative set was achieved. Once a good dataset had been obtained,
this was divided into two subsets to see whether the results were correctly gener-
alized: one for training and the other for testing (Fig. 3).
Definition of the initial equation was important in guiding the new expression
search process. From this moment, an iterative process began, consisting of the def-
inition of restrictions, execution and results analysis.
3.2. Conventional concrete property vs. recycled concrete property
As mentioned, the first objective is to compare the conventional concrete prop-
erty with that of recycled concrete. The aim is to discover how much one property
can decrease when recycled concrete coarse aggregates are used, by analysing dif-
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 483

Proc;CC;exp Proc;RC;exp Proc;CC;exp Proc;RC;exp

– ) ¼ ð2Þ
Proc;CC;cal Proc;RC;cal Proc;CC;cal Proc;RC;cal  coefficientðf c;RC ; %RCA; WAÞ

Proc;exp Ratio recycled concrete Ratio RC

Ratio ¼ ! Coefficient ¼ ¼ ð3Þ
Proc;cal Ratio conventional concrete Ratio CC

Being:
Proc,CC,cal: calculated property of conventional concrete using code expressions.
Proc,CC,exp: experimental property of conventional concrete.
Proc,RC,cal: calculated property of recycled concrete using code expressions.
Proc,RC,exp: experimental property of recycled concrete.

Using this method, the proposed prediction expressions will be as follows:

Proc;RC ¼ CFðf c;RC ; %RCA; WAÞ  Eurocodeexpression ð4Þ

Being:
Proc,RC = property of recycled concrete.
CF = correction coefficient for the Eurocode expression.
fc,RC = recycled concrete compressive strength.
%RCA = replacement ratio (recycled concrete coarse aggregate percentage).
WA = recycled coarse aggregate water absorption.

3.4. Adjustment of specific expressions

Lastly, in this work, as already mentioned, specific expressions to predict the

studied properties have been adjusted. In this case, conventional concrete proper-
ties are not necessary because the expressions are adjusted using those which best
fit the experimental results for recycled concrete. For said purpose, the Eurocode
expressions have been used as a basis and modified to introduce the replacement
ratio and the recycled concrete coarse aggregate quality (considering it, again,
based on its water absorption). Also in this case, a different expression (Eq. (5))
has been adjusted for each of the different mixing procedures considered, and the
techniques used have been multivariable regression and genetic programming.
With both techniques, part of the data (80%) has been kept for the training process
while the remaining cases (20%) were left for the test process.

Proc;RC ¼ ðf c;RC ; %RCA; WAÞExpression ð5Þ

Fig. 3. Methodology used [99].
ferent incorporation ratios and mixing procedures, which have not been considered
by other authors when developing their expressions but have a well known influ-
ence on the properties of recycled concrete [7,14,79,103].
To develop this kind of analysis, some works were selected from the database:
those that use a control concrete (conventional concrete) and different recycled
concretes made with the same dosage and materials except for the coarse aggre-
gate, which will be replaced with recycled concrete coarse aggregate (by volume)
at different percentages.
With these works, and using linear regression, a coefficient can be adjusted
which allows the estimation of the recycled concrete property as a function of that
of conventional concrete (Eq. (1)). This coefficient will take into account the recy-
cled content and the mixing procedure used.
Regarding genetic programming, Table 1 shows the default settings used in this
work when running the algorithm. These parameters were chosen because they
gave the best results in the initial tests. The input data to the algorithm was not
standardized in the aim of achieving directly applicable formulas.
Eq. (6) shows the fitness function used, where n is the number of datasets used,
xp is the predicted value by the individual and di is the desired value obtained from
the experimental data, a is the level of parsimony and si is the size (number of
nodes) of the individual. In essence, this fitness function uses the mean square error
(MSE) penalized by the size of the individual. Thus, if two individuals have the same
MSE value, the simple individual (less nodes) gets a better fit.
1X n
FitnessðiÞ ¼ ðxp  yp Þ2 þ a  si ð6Þ
n p¼1
In short, Fig. 4 synthesises the methodology of this paper.

Proc;RC ¼ Xð%RCAÞ  Proc;CC ð1Þ 3.5. Adjustment goodness

where:
Some statistical indexes have been used to evaluate the performance of the dif-
Proc,RC: recycled concrete property.
ferent adjustments. In this regard, when linear regression is used, the square of the
Proc,CC: conventional concrete property.
Pearson product-moment correlation coefficient (coefficient of determination, R2)
X: adjusted coefficient.
can be considered a good index of the adjustment goodness.
When multivariable regression or genetic programming is used, the perfor-
3.3. Correction of code expressions mance of the obtained expressions has been evaluated using four statistical
indexes: the Pearson correlation coefficient (r), the mean squared error (MSE), the
This analysis tries to correct the code expressions (Eurocode) to predict the coefficient of variation (COV) and the mean average error (MAE) (Eqs. (7)–(10)).
recycled concrete properties with the same approximation degree as in conven- These indexes allow the author to determine which of the equations (proposed in
tional and recycled concretes. For said purpose, the theoretical value of any prop- this work or obtained from the literature) are the best for predicting the behavior
erty is calculated taking into account the experimental value of the compressive of recycled concrete.
strength. Using these results, the ‘‘experimental property/calculated property” ratios Pn
are obtained. If the ratios of recycled concretes are similar to those obtained with 
i¼1 ðxi  xÞðyi  yÞ

r ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn ffi ð7Þ
the conventional kind, it can be considered that the code expressions provide the  2
2
i¼1 ðxi  xÞ ðyi  yÞ
same approximation degree both in recycled and conventional concretes when cal-
culating the analyzed property and, therefore, it is not necessary to modify these
expressions. However, if these ratios decrease with the increase in the content of 1X n
2
MSE ¼ ðxi  yi Þ ð8Þ
recycled coarse aggregate, the code expressions should be corrected using a coeffi- n i¼1
cient which, in this case, will be adjusted by multivariable regression. As already
mentioned, the variables considered have been the replacement ratio and the recy- rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P n
ðxi xÞ2
cled concrete coarse aggregate quality (considering it based on its water absorp- i¼1
n Actual value
tion) (Eqs. (2) and (3)). A different coefficient has been adjusted for each of the COV ¼ Pn  100; x ¼ ð9Þ
ðxi Þ Predicted value
different mixing procedures considered. i¼1
n
484 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

1X n
to the expression (15) are shown in Table 2. These allow engineers
MAE ¼ absðxi  yi Þ ð10Þ
n i¼1 to estimate the expected compressive strength decrease for struc-
tural recycled concrete as a function of the recycled coarse aggre-
gate content (RCA) and the production method.
4. Compressive strength
f c;RC ¼ X  f c;CC ð15Þ
In general, recycled concrete compressive strength is lower
The square of the Pearson product-moment correlation coeffi-
than that of conventional concrete, and it decreases as the coarse
cient (coefficient of determination, R2) was calculated. In all cases,
aggregate substitution percentage increases. There are several
the value of R2 is between 0.9 and 1, which indicates a very strong
interesting works [97,103–105] which have proposed different
correlation (Table 2).
methods for predicting the compressive strength of recycled con-
In Figs. 5–7, it can be seen that the lower the recycled aggregate
cretes. De Brito and Robles [97] using the data of four authors
percentage the lower the strength loss. Figs. 5–7 show the ‘‘recycled
[18,109–111] have studied the variation of the ratio between the
concrete compressive strength vs. conventional concrete compressive
28 and 90-day compressive strengths of recycled and conventional
strength” relationship for the different recycled percentages
concretes (fcRAC/fcRC), and the ratio between the densities (D), the
obtained using each of the production methods.
water absorptions (WA) of the mixture of aggregates and the 7-
The PS method presents the highest reduction in compressive
day compressive strengths of concretes (Eqs. (11)–(13)).
strength, due to the bleeding effect. In this method, as previously
 
fcRAC WARAC mentioned, the aggregates are pre-soaked for a fixed time before
¼a 1 þ1 ð11Þ
fcRC WARC use. The pre-soaking time is usually adjusted taking into account
the duration of time required for the water absorption of the recy-
  cled coarse aggregates.
fcRAC DRAC
¼b 1 þ1 ð12Þ The ADwEW method, however, employs an extra quantity of
fcRC DRC
water which is added during the mixing process. This quantity of
  extra water is adjusted to a fixed percentage of water absorption
fcRAC fc7RAC
¼c 1 þ1 ð13Þ at 24 h. With this method the reduction in compressive strength
fcRC fc7RC
is lower and there is no need for the dosage of superplasticizer
Recently, Silva et al. [103] suggested a model to predict strength to be increased.
loss based on the recycled aggregate content and quality. These Lastly, the AD method presents a compressive strength decrease
authors show several relationships as follows: lower than that obtained with the PS method and, in general, sim-
ilar to the one obtained with ADwEW. However, with this method
f cm-RAC
¼ a  Replacement level ð%Þ þ 1 ð14Þ the high water absorption of the recycled aggregates reduces the
f cm-Control free water in the mixes. This also leads to a loss in the workability
Other methods are based on artificial neural networks (ANN) of the fresh concrete which has to be compensated with a high
[104,105]. These investigators concluded that the ANN models dosage of superplasticizer.
could accurately predict compressive strength values.
As already mentioned, this study focuses on comparing conven-
5. Modulus of elasticity (Ec)
tional concrete compressive strength with that of recycled con-
crete. The objective was to discover how much this property can
It is well know that the concrete elasticity modulus is related to
decrease when recycled concrete coarse aggregates are used, by
the aggregates and the rigidity of the cement paste. The cement
analysing different incorporation ratios and mixing procedures,
paste modulus is lower than that of the aggregate and the concrete
which have not been considered by other authors when developing
[88]. Furthermore, recycled aggregates (with adhered mortar) pre-
their expressions but have a well known influence on the proper-
sent a modulus of elasticity lower than that of natural aggregates
ties of recycled concrete. The database analysis confirms that the
and, therefore, the recycled concrete modulus will be lower than
reduction in compressive strength depends on the percentage of
that of conventional concrete. This is also due to the weakness of
recycled aggregate and the concrete production method. Using lin-
the new interfaces (new cement mortar – aggregate, old cement
ear regression, the correction coefficients (X) which allow engi-
mortar – aggregate and old cement mortar – new cement mortar)
neers to estimate the decrease in compressive strength according
which can lead to the progressive development of cracks, affecting
Table 1 the concrete deformability [89].
Parameters used. Most of the codes and standards suggest equations to predict
Configuration parameters Values
the modulus of elasticity as a function of compressive strength. It
is well known that these expressions are not suitable when recy-
Population size 1000
Stopping criteria (maximum 40,000, 2000
cled concrete is used. Therefore, many authors have proposed dif-
generation, epoch without ferent expressions for its prediction (Table 3). However, these
improvement) expressions have been adjusted with the authors’ particular results
Crossing rate 90% and, so, it has been revealed that they present some scatterings
Non-terminal selection rate 90%
when other researchers’ results are analyzed.
Mutation probability 10%
Algorithms Selection: tournament creation: Furthermore, in recent years, as with compressive strength,
intermediate mutation: subtree methods have been developed by different authors [97,98,106] to
Elitist strategy Yes predict the recycled concrete modulus. De Brito and Robles [97]
Parsimony 0/100 using the data of four authors [18,109–111] have studied the vari-
Initial tree height 3/6
Maximum tree height 6/9
ation of the ratio between the 28 and 90-day modulus of elasticity
Maximum mutation tree height 3/6 of recycled and conventional concretes (EcRAC/EcRC), and the ratio
Terminal nodes 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, random between the densities (D), the water absorptions (WA) of the mix-
real numbers [1, 1] ture of aggregates and the 7-day compressive strengths of con-
Non-terminal nodes +,  , *, /
cretes (Eqs. (16)–(18))
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 485

Fig. 4. Methodology of the paper.

Table 2 100
Correction coefficients (X) to estimate fc,RC. PS 100%
90 ADwEW
%RCA AD ADwEW PS AD
80
100 Data 147 140 89
X 0.897 0.888 0.828 70
R2 0.940 0.960 0.908 60 y = 0.897x
fc,RC (MPa)

50 Data 91 49 31 R² = 0.940
50
X 0.909 0.937 0.861 AD
R2 0.975 0.911 0.874 40 y = 0.888x
20 Data 83 38 37 30 R² = 0.960
ADwEW
X 0.964 0.969 0.936
20 y = 0.828x
R2 0.991 0.954 0.958 R² = 0.908
10 PS
0
  0 10 20 30 40 50 60 70 80 90 100
EcRAC WARAC
¼a 1 þ1 ð16Þ fc,CC (MPa)
EcRC WARC
Fig. 5. fc,RC vs. fc,CC (100%).
 
EcRAC DRAC
¼b 1 þ1 ð17Þ
EcRC DRC
100
  PS 50%
EcRAC fc7RAC 90 ADwEW
¼c 1 þ1 ð18Þ AD
EcRC fc7RC 80

On the other hand, Behnood et al. [106] developed the M5 tree 70

y = 0.937x
algorithm to model the recycled concrete modulus. They assume 60
fc,RC (MPa)

R² = 0.911
the general expression (19) or (20) and develop the model in the 50 ADwEW
form of logarithms, as the method is only able to produce linear y = 0.909x
40
relationships. The results show that the model can be an accurate R² = 0.975
tool for predicting the modulus of elasticity of recycled aggregate 30 AD
y = 0.861x
concrete. 20
R² = 0.874
wc CAd  f
FA 10 PS
b
ERAC ¼ aðf cu Þ ðrÞe ðSGSSD Þ g ðW a Þh ð19Þ 0
c C TA
0 10 20 30 40 50 60 70 80 90 100

 c0 CAd0  f 0 fc,CC (MPa)

b0 w 0 FA
ERAC ¼ a0 ðf cu Þ ðrÞe ð20Þ Fig. 6. fc,RC vs. fc,CC (50%).
c C TA
486 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499
100
PS 20%
90 ADwEW
AD
80
70
60
fc,RC (MPa)
50
40
30
y = 0.936x
20 R² = 0.958
10 PS
0
0 10 20 30 40 50 60 70
fc,CC (MPa)
Fig. 7. fc,RC vs. fc,CC (20%).
Duan et al. [98] explored the applicability of artificial neural
networks (ANN) in modeling the recycled aggregate concrete mod-
ulus. Their results show that the constructed ANN models can
accurately predict the elastic modulus of concrete made with recy-
cled aggregates derived from different sources. However, these
techniques [98,106] do not provide structural designers with a
clear mathematical expression.
As shown, the prediction of the recycled concrete modulus is a
question that remains unsolved. In this work, using a database
with over 300 datasets, a study of the relationship between recy-
cled and conventional concrete modulus has been carried out. After
which, with over 400 datasets and using multivariable regression
and genetic programming, expressions to predict the structural
recycled concrete modulus have been developed.
5.1. Conventional concrete property vs. recycled concrete property
In this first step, and using linear regression (Figs. 8 and 9), a
coefficient to estimate the recycled concrete modulus as a function
of that of conventional concrete has been adjusted (Eq. (21)). This
coefficient takes into account the recycled content and the mixing
procedure used. The adjustment goodness has been evaluated with
the coefficient of determination, R2 (Table 4).
Ec;RC ¼ X  Ec;CC
As in the compressive strength analysis, the PS method presents
the highest reduction in the elasticity modulus, which is, once
again, due to the bleeding effect. The AD method shows the lowest
reductions and, lastly, the ADwEW method presents an elasticity
modulus decrease lower than that obtained with the PS method
and higher than that with the AD method (Figs. 8 and 9).
Figs. 8 and 9 show the ‘‘recycled concrete elasticity modulus vs.
conventional concrete elasticity modulus” relationship for the differ-
ent replacement percentages and different mixing procedures.
The only goal of this method (conventional concrete modulus vs.
recycled concrete modulus), is to obtain information as the recycled
concrete modulus cannot be predicted with this analysis technique
as a conventional concrete reference is needed.
5.2. Correction of code expressions
The second step is to analyze whether or not it is necessary to
adapt the proposed code expression (Eurocode) to predict the elas-
ticity modulus of recycled concrete with the same approximation
degree as that of conventional concrete. In Figs. 10–12 the ‘‘exper-
imental modulus/calculated modulus” ratio of recycled vs. conven-
tional concretes is shown for each mixing procedure (AD,
ADwEW and PS).
As observed, the ratios of conventional concretes are higher
than those obtained with recycled concretes. This means that the
y = 0.969x code expression will not provide the same approximation degree
R² = 0.954 in recycled concretes as in conventional ones when calculating
ADwEW
the modulus. Therefore, it is necessary to modify it by introducing
y = 0.964x a correction coefficient.
R² = 0.991
AD
Hence, in order to get the same approximation degree as with
the Eurocode expression, i.e. to obtain the same ‘‘experimental mod-
ulus/calculated modulus” ratio in conventional and recycled con-
cretes using said expression, a correction coefficient (CF) has
been adjusted by multivariable regression. The variables consid-
80 90 100 ered have been the replacement ratio and the recycled concrete
coarse aggregate quality (considering it based on its water absorp-
tion). A different coefficient has been adjusted for each of the dif-
ferent mixing procedures considered. More variables affecting
the results may be considered to improve the proposed models.
However, as this occurs with the same relationship for conven-
tional concrete, the cost (complexity of the model)/benefit (relia-
bility of the prediction) was taken into account and it was
decided not to increase the number of variables.
Then, according to Section 3.3, the expression is as follows:
Ec;RC ¼ CFðf c;RC ; %RCA; WAÞ  Eurocodeexpresssion ð22Þ
where:
 0:3
fc
Eurocodeexpresssion ¼ Eccal ¼ 22; 000  ð23Þ
10
Ratio recycled concrete
CFðf c;RC ; %RCA; WAÞ ¼
Ratio conventional concrete
Ecexp;RC =Eccal;RC
¼ ð24Þ
Ecexp;CC =Eccal;CC
In this work, the following general expression has been selected
to adjust the coefficient CF:
 
1  b  %RCA  WA
CF ¼  
100 5
ð25Þ
f c;RC ð0:3a 100 Þ  
%RCA
10
 1  d  %RCA
100
Through multivariable regression, the parameters a, b and d
ð21Þ have been adjusted (Table 5). The adjustment performance has
been analyzed using the Pearson correlation coefficient (r) and
the mean squared error (MSE).
Figs. 13–15 show the ‘‘predicted value vs. actual value” relation-
ship for the CF coefficient.
Figs. 16–18 show the new ‘‘experimental modulus/calculated
modulus” ratios of recycled vs. conventional concretes for each
mixing procedure. In this case the recycled concrete modulus
was predicted using the Eurocode expression corrected with the
adjusted coefficient. As observed, at this stage the ratios of recycled
concretes are similar to those obtained with conventional kinds.
5.3. Adjustment of specific expressions
Lastly, in this work, specific expressions to predict the modulus
of elasticity have been adjusted. For said purpose, the Eurocode
expression has been taken as a basis, but it has been modified to
introduce the replacement ratio and the recycled concrete coarse
aggregate quality (considering it, again, based on its water absorp-
tion). Also in this case, a different expression has been adjusted for
each of the different mixing procedures considered, and the tech-
niques used have been multivariable regression and genetic
programming.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 487

Table 3
Authors’ equations.

Refs. Equation
Ravi [96] 0:33
Ec ¼ 7770  f c
Ravi [96] Ec ¼ 4630  f c
0:5

Ravi [32] 0:5

Ec ¼ 3480  f c þ 13; 050
Tang [38] Ec ¼ 4450  f c
0:5

Bair [89] Ec ¼ a  f c ; a ¼ aðrÞ ¼ 5780—1340  rð0 6 r 6 1Þ

0:5

Dhir [93] Ec = 370  fc + 13,100

Dill [94] Ec = 634.43  fc + 3057.6
Mell [107] Ec = 378  fc + 8242
Xiao [90] Ec ¼ 2:8þ
10 5
40:1
fc

Katz [92] Ec ¼ 0:043  q1:5  f c

0:5

Kaki [95]  q 1:5  f 0:5

Ec ¼ 1:9  105  2300  2000
c

0:5  q   
Li [91] Ec ¼ 5:5  10  f c  2400  1  ar ; a ¼ 2:2876  r þ 0:1288ð0 6 r 6 1Þ
3
qffiffiffiffiffiffiffiffiffiffiffiffiffi
Cori [7]
Ec ¼ 18; 800  0:83f
3 cu
10
Cori [7] Ec = 909  fcu + 8738
Zilc [108] 1=3  q 2
Ec ¼ 9100  ðf c þ 8Þ  2400

Ec = modulus of elasticity (MPa).

fc or fcu = compressive strength (MPa).
q = density (kg/m3).
r = replacement percentage.

50000
PS 50-100% y = 0.856x Table 4
ADwEW R² = 0.763
AD Correction coefficients (X) to estimate Ec,RC.
40000 ADwEW
Production method <50% 50–100%
AD Data 36 78
30000
Ec,RC (MPa)

X 0.948 0.856
y = 0.856x R2 0.830 0.510
R² = 0.510
20000 AD ADwEW Data 19 73
X 0.882 0.856
R2 0.931 0.763

10000 y = 0.792x
PS Data 39 68
X 0.886 0.792
R² = 0.579
PS R2 0.437 0.579
0
0 10000 20000 30000 40000 50000
Ec,CC (MPa)
1.2
Fig. 8. Ec,RC vs. Ec,CC (50–100%).
AD

50000 1.1
PS <50%
ADwEW y = 0.948x
AD R² = 0.830
40000 1.0
AD

30000
Ec,RC (MPa)

y = 0.882x
0.9
R² = 0.931
Rao RC

ADwEW
20000 y = 0.886x
R² = 0.437 0.8
PS
10000

0.7
0
0 10000 20000 30000 40000 50000
Ec,CC (MPa)
0.6
Fig. 9. Ec,RC vs. Ec,CC (<50%).

0.5
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
The following general expression has been selected as a basis Rao CC
for the prediction of the recycled modulus of elasticity using mul-
tivariable regression (Eq. (26)). Fig. 10. Ratio RC vs. ratio CC (AD method).
488 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

1.3 where:
ADwEW    
%RCA WA %RCA WA
1.2 Branch1 ¼ f ; and Branch2 ¼ f ;
100 5 100 5
The rules for generating the branches have been as follows:
1.1
both include two variables, %RCA and water absorption (WA),
and the maximum number of occurrences of these variables, equal
1.0 in both branches, has been restricted to 2.
Apart from these limitations, the equation has been generated
Rao RC

by the terminal and non-terminal nodes shown in Table 1. At the

0.9 end of the genetic programming, one final expression was obtained
for each mixing procedure. The ‘‘shape” of each branch of the equa-
0.8 tion has been determined and the final results are as follows:
For the AD method:

0.7 WA
Branch1 ¼   5
 ð28Þ
5%RCA
19  5:09  %RCA
100
 100
WA
0.6 5

1 1
0.5 Branch2 ¼ %RCA þ ð29Þ
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 100
þ 10 5  %RCA
100
Rao CC
For the ADwEW method:
Fig. 11. Ratio RC vs. ratio CC (ADwEW method).  
0:0207  0:0443
%RCA þ 0:2003
Branch1 ¼ 100
ð30Þ
1.2 2  WA
5
 0:8552
PS
0:2211
1.1 Branch2 ¼ %RCA
100
þ 0:2078
0:003067
1.0
þ WA %RCA  WA  þ 0:0442 ð31Þ
5
 100
þ 0:2003  5  0:8518

And for the PS method:

0.9
Rao RC

 
WA
Branch1 ¼ 0:0521   0:0521
0.8 5
!
0:2027 0:52
 %RCA  þ %RCA ð32Þ
WA
5
 100
 0:2027 100
0.7

WA 0:004188 0:13
0.6 Branch2 ¼ 0:1202  þ WA þ %RCA  0:03346 ð33Þ
5 5
 0:9616 100

The obtained equations have been analyzed comparing their

0.5 behavior with those obtained in the literature. In all cases, the
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
equations obtained with multivariable regression show r, MSE
Rao CC
and MAE values better than the best ones in the literature. The
Fig. 12. Ratio RC vs. ratio CC (PS method). COV value is the only index that does not improve the accuracy
of the other authors’ expressions. With the genetic programming
  0:3 1a%RCA   technique, the equations obtained always show the best statistical
f c;RC ½ ð 100 Þ
1  b  %RCA  WA indexes (Appendix, Tables A1–A3).
Ec;RC ¼ 22; 000    100 5
ð26Þ
10 1  d  %RCA
100 Lastly, on Figs. 19–24 the ‘‘predicted value vs. experimental value”
relationship can be observed for the group of data used to adjust
Table 6 shows the parameters a, b and d adjusted using multi-
the equations (training set) and the group of data used to validate
variable regression.
them (testing set). Both relationships are shown for the equation
When genetic programming is used, the first step is the selec-
obtained using multivariable regression and the equation opti-
tion of the type of equation to be optimized. In this case, after dif-
mized using genetic programming.
ferent tests it was decided to work with an equation that guides
Both techniques provide an expression that can be suggested to
the search process (Eq. (27)). Therefore, Eq. (27) has been estab-
calculate the modulus of recycled aggregate concrete. The adjust-
lished to be optimized using genetic programming.
ment is quite similar in both cases. However, the equation opti-
  1 1%RCABranch1 Þ  
f c;RC ½3 ð 100 %RCA mized using genetic programming has shown better statistical
Ec;RC ¼ 22; 000   1  Branch2 ð27Þ indexes whereas the equation obtained using multivariable regres-
10 100
sion is simpler.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 489

Table 5
Adjustment of the correction coefficient (CF) to calculate the RC modulus.

Production method Data Adjustment r MSE

Training Testing a b d Training Testing Training Testing
AD 90 22 0.1268 0.0037 0.2649 0.5332 0.4146 0.004051 0.003528
ADwEW 164 42 0.0697 0.0677 0.1044 0.4019 0.4936 0.003843 0.003038
PS 73 17 0.3433 0.1842 0.1120 0.5106 0.7252 0.007868 0.004857

1.2
1.2
Training set (AD) Training set (PS)
Tesng set (AD) Tesng set (PS)
1.1
1.1

1.0
1.0
CF actual value

0.9

CF actual value
0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
CF predicted value CF predicted value

Fig. 13. CF predicted value vs. CF actual value. Training and testing performance Fig. 15. CF predicted value vs. CF actual value. Training and testing performance (PS
(AD method). method).

1.2 1.2
Training set (ADwEW)
AD
Tesng set (ADwEW)
1.1 1.1

1.0 1.0
Rao RC (corrected)
CF actual value

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
CF predicted value Rao CC
Fig. 14. CF predicted value vs. CF actual value. Training and testing performance
Fig. 16. Ratio RC corrected with CF vs. ratio CC (AD method).
(ADwEW method).
490 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

1.3 For both the compressive strength and the modulus Ec, there are
ADwEW several interesting works [97,102] which have tried to predict the
splitting tensile strength of recycled aggregate concretes. Again, De
1.2
Brito and Robles [97] using the data of two authors [18,111] stud-
ied the variation of the ratio between the 28 and 90-day splitting
1.1 tensile strength of recycled concrete and that of conventional con-
crete (EcRAC/EcRC), and the ratio between the densities (D), the
water absorptions (WA) of the mixture of aggregates and the 7-
1.0
Rao RC (corrected)

day compressive strengths of concretes.

 
0.9
fspRAC WARAC
¼ a 1 þ1 ð34Þ
fspRC WARC
 
0.8 fspRAC DRAC
¼b 1 þ1 ð35Þ
fspRC DRC
0.7  
fspRAC fc7RAC
¼c 1 þ1 ð36Þ
fspRC fc7RC
0.6
These authors also stated that the results of the splitting tensile
strength graphs obtained, confirm the scatter of the test results for
0.5 this property in different campaigns.
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
Rao CC
Recently, Silva et al. [102] suggested that the tensile strength is
also affected, like the compressive strength [103], by the recycled
Fig. 17. Ratio RC corrected with CF vs. ratio CC (ADwEW method). aggregate content, size, type and quality. They have found that
the relationship between tensile and compressive strengths
1.2 appears to have been unaffected by the use of recycled aggregate.
They conclude that the inclusion of recycled aggregates causes a
PS proportional decrease in both tensile and compressive strengths,
1.1 at a rate that follows the same relationship as observed in conven-
tional concrete.
However, other authors suggest that the equations proposed by
1.0 codes and standards to predict the splitting tensile strength as a
function of the compressive strength are not suitable when recy-
cled concrete is used. In this regard, these authors [90–92], based
Rao RC (corrected)

0.9 on their experimental campaigns, have proposed specific expres-

sions adjusted to predict the recycled concrete splitting tensile
strength (Table 7).
0.8 Again in this work, using a database with over 330 datasets,
studies on the relationship between recycled and conventional
concrete splitting tensile strength have been analyzed. Then, with
0.7 over 480 datasets, the ‘‘experimental property/calculated property”
ratios have been obtained. If these are similar in conventional
and recycled concretes, it can be considered that code expressions
0.6 provide the same approximation degree in both recycled and con-
ventional concretes and, therefore, it is not necessary to modify
these expressions. Otherwise, a correction coefficient to predict
0.5 structural recycled concrete splitting tensile strength should be
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Rao CC
proposed. Lastly, with multivariable regression and genetic pro-
gramming, the relationship between splitting tensile and compres-
Fig. 18. Ratio RC corrected with CF vs. ratio CC (PS method). sive strength in recycled concretes has been optimized, taking into
account the replacement percentage, the quality of the recycled
aggregate and the mixing procedure used.
Table 6
Parameters a, b and d adjusted using multivariable regression – Ec,RC expression. 6.1. Conventional concrete property vs. recycled concrete property
a b d
AD 0.6378 0.4449 1.4893
In this first step, and using linear regression (Figs. 25 and 26), a
ADwEW 0.7539 0.1403 1.0534 coefficient to estimate recycled concrete splitting tensile strength
PS 0.4143 0.1636 0.2700 as a function of that of conventional concrete has been adjusted
(Eq. (37)). This coefficient takes into account the recycled content
6. Splitting tensile strength (fsp) and the mixing procedure used. The suitability of the adjustment
has been evaluated by the coefficient of determination, R2
The same analysis for the modulus of elasticity has been carried (Table 8).
out for the splitting tensile strength, although the results obtained
with this parameter show higher scatterings than those obtained f sp;RC ¼ X  f sp;CC ð37Þ
for the modulus.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 491

45 45
Mulvariable Mulvariable
Genec Prog Genec Prog
40 40

35 35
Ec,RC experimental value (GPa)

Ec,RC experimental value (GPa)

30 30

25 25

20 20

15 15

10 10
10 15 20 25 30 35 40 45 10 15 20 25 30 35 40 45
Ec,RC predicted value (GPa) Ec,RC predicted value (GPa)

Fig. 19. Ec,RC predicted value vs. Ec,RC experimental value. Training performance (AD Fig. 21. Ec,RC predicted value vs. Ec,RC experimental value. Training performance
method). (ADwEW method).

45 45
Mulvariable Mulvariable
Genec Prog Genec Prog
40 40

35 35
Ec,RC experimental value (GPa)

Ec,RC experimental value (GPa)

30 30

25 25

20 20

15 15

10 10
10 15 20 25 30 35 40 45 10 15 20 25 30 35 40 45
Ec,RC predicted value (GPa) Ec,RC predicted value (GPa)

Fig. 20. Ec,RC predicted value vs. Ec,RC experimental value. Testing performance (AD Fig. 22. Ec,RC predicted value vs. Ec,RC experimental value. Testing performance
method). (ADwEW method).

The water demand of recycled aggregates in air-dry condition
without any adjustment dismisses the effective water to cement
ratio, which leads to obtaining the lowest reductions in the
mechanical properties of recycled concretes made using the AD
method. This fact also explains that some recycled concretes made
using this method present higher splitting tensile strength than
those obtained with their control concretes (Fig. 25). It can be
observed that the ADwEW method presents recycled concretes
with a very similar behavior to those produced using the AD
method.
As in the compressive strength analysis, the PS method presents
the highest reduction in the splitting tensile strength, higher than
that of the ADwEW method (Figs. 25 and 26). This can be
explained, again, by the bleeding effect and, according to Silva
et al. [102] and Ferreira et al. [28], by the ‘‘nailing effect”, which
is caused by the cement paste filling the surface pores of aggregate
particles, achieving an improved bond strength.
6.2. Correction of code expressions
As in the elasticity modulus section, the second step is to ana-
lyze whether or not it is necessary to adapt the proposed Eurocode
expression for the splitting tensile strength when recycled con-
492 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

45 Table 7
Mulvariable Authors’ equations.

Genec Prog Refs. Equation

40 0:65
Xiao [90] f sp ¼ 0:24  f cu
pffiffiffiffiffi
Katz [92] f sp ¼ 0:59  f c
Li [91] 2=3
f t ¼ ð0:24  0:06  rÞ  f cu
35
Ec,RC experimental value (GPa)

30 the prediction of the splitting tensile strength is enough to get

the same approximation degree in recycled concretes as in conven-
tional ones.
25 In the case of the PS method (Fig. 29), high scatterings can be
observed (points are not as close as in the AD and ADwEW meth-
ods). However, they do not show a clear trend in their position,
20 being around the midline. In this regard, it can be considered once
again, that a correction coefficient is not necessary.

15
6.3. Adjustment of specific expressions

Lastly, as in the modulus analysis, specific expressions to pre-

10
10 15 20 25 30 35 40 45 dict the splitting tensile strength have been adjusted. For said pur-
Ec,RC predicted value (GPa) pose, the Eurocode expression has been taken as a basis, but it has
been modified to introduce the replacement ratio and the recycled
Fig. 23. Ec,RC predicted value vs. Ec,RC experimental value. Training performance (PS concrete coarse aggregate quality (considering it, again, based on
method).
its water absorption). Also in this case, a different expression has
been adjusted for each of the different mixing procedures consid-
45 ered, and the techniques used have been multivariable regression
Mulvariable and genetic programming.
Genec Prog The following general expression has been selected as a basis
40 for the prediction of recycled splitting tensile strength using mul-
tivariable regression.
 
1 ½2=3ð1a%RCA
100 Þ
1  b  %RCA  WA
35 f sp;RC ¼  f c;RC   100 5
ð38Þ
3 1  d  %RCA
Ec,RC experimental value (GPa)

100

Thus, the parameters a, b and d have been adjusted (Table 9)

30
using multivariable regression.
In the case of genetic programming, Eq. (39) has been estab-
lished to guide the search process and to be optimized using this
25
technique.
 
1 ½2=3ð1%RCA Branch1 Þ %RCA
f sp;RC ¼  f c;RC 100
 1  Branch2 ð39Þ
20 3 100
where:
   
15 %RCA WA %RCA WA
Branch1 ¼ f ; and Branch2 ¼ f ;
100 5 100 5

10
10 15 20 25 30 35 40 45
Ec,RC predicted value (GPa)
7
PS 50-100% y = 0.913x
Fig. 24. Ec,RC predicted value vs. Ec,RC experimental value. Testing performance (PS
method). 6 ADwEW R² = 0.924
AD AD
5
crete is used. In Figs. 27–29 the ‘‘experimental splitting tensile
fsp,RC (MPa)

4
strength/calculated splitting tensile strength” ratio of recycled and y = 0.902x
conventional concretes is shown for each mixing procedure (AD, R² = 0.931
3
ADwEW and PS). ADwEW
It can be observed (Figs. 27–29) that the points (ratio CC vs. 2 y = 0.888x
ratio RC) are noticeably closed and around (up and down) the mid- R² = 0.642
line in the cases of the AD and ADwEW methods. This means that 1 PS
the code expression provides the same approximation degree in
recycled concretes as in conventional ones when calculating split- 0
0 1 2 3 4 5 6 7
ting tensile strength. Therefore, it is not necessary to modify the fsp,CC (MPa)
proposed Eurocode expression with any correction coefficient. In
these cases, the use of the experimental compressive strength for Fig. 25. fsp,RC vs. fsp,CC (50–100%).
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 493

7 1.0
PS <50% AD
6 AD
ADwEW
5 y = 0.946x 0.9
R² = 0.959
fsp,RC (MPa)

4 ADwEW
y = 0.933x
3 0.8
R² = 0.962
AD
2 y = 0.919x

Rao RC
R² = 0.635
PS 0.7
1

0
0 1 2 3 4 5 6 7
0.6
fsp,CC (MPa)

Fig. 26. fsp,RC vs. fsp,CC (<50%).

Table 8 0.5
Correction coefficients (X) to estimate fsp,RC.

Production method <50% 50–100%

AD Data 47 135 0.4
0.4 0.5 0.6 0.7 0.8 0.9 1.0
X 0.933 0.913
R2 0.962 0.924 Rao CC

ADwEW Data 11 78 Fig. 27. Ratio RC vs. ratio CC (AD method).

X 0.946 0.902
R2 0.959 0.931
PS Data 24 45
X 0.919 0.888
1.0
R2 0.635 0.642 ADwEW

Again, the rules for generating the branches have been as fol- 0.9
lows: both include two variables, %RCA and water absorption
(WA) and the maximum number of occurrences of the variables,
equal in both branches, has been restricted to 2.
0.8
Apart from these limitations, the equation has been generated
by the terminal and non-terminal nodes shown in Table 1. At the
end of the genetic programming, one final expression was obtained
Rao RC

for each mixing procedure. The ‘‘shape” of each branch of the equa- 0.7
tion has been determined and the final results are as follows.
For the AD method:

0:001261  WA 0.6
Branch1 ¼  5  ð40Þ
%RCA
100
 WA
1
 0:9568
5

WA  0.5
0:1450 0:1515  3
Branch2 ¼ 0:1515  WA
 %RCA
5
ð41Þ
5 100

For the ADwEW method: 0.4

0.4 0.5 0.6 0.7 0.8 0.9 1.0
WA 
WA
 1 Rao CC
Branch1 ¼ 5 5
ð42Þ
28 Fig. 28. Ratio RC vs. ratio CC (ADwEW method).

2
Branch2 ¼ %RCA  WA  ð43Þ
100
 2 5 þ6 et al. [90] generally shows the best adjustment. The equations
obtained with multivariable regression do not manage to improve
For the PS method: the statistical indexes of the other authors’ expressions. However,
WA
þ %RCA  1:5271 with the genetic programming technique, the equations obtained
Branch1 ¼ 5 100
%RCA
ð44Þ always show the best statistical indexes. They show r, MSE, COV
100
þ8
and MAE values better than the best ones from the literature. Fur-
  thermore, they improve the accuracy of other authors’ expressions.
0:001574  0:1255
%RCA þ 3 Lastly, in Figs. 30–35 the ‘‘predicted value vs. experimental value”
Branch2 ¼  %RCA WA 100
 ð45Þ relationship can be observed for the group of data used to adjust
100
 5  1:1255
the equations (training set) and for the group of data used to vali-
The obtained equations have been analyzed comparing their date them (testing set). Both relationships are shown for the equa-
behavior with those obtained in the literature (Appendix, Tables tion obtained using multivariable regression and for the equation
A4–A6). Regarding the authors’ expressions, the equation of Xiao optimized using genetic programming.
494 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

1.0 7
PS Mulvariable
Genec Prog
6
0.9

fsp,RC experimental value (MPa)

0.8

4
Rao RC

0.7

0.6
2

0.5
1

0.4 0
0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7
Rao CC fsp,RC predicted value (MPa)

Fig. 29. Ratio RC vs. ratio CC (PS method). Fig. 31. fsp,RC predicted value vs. fsp,RC experimental value. Testing performance (AD
method).

using genetic programming has shown better statistical indexes

Table 9 whereas again, the equation obtained using multivariable regres-
Parameters a, b and d adjusted using multivariable regression – fsp,RC expression.
sion is simpler.
a b d In Fig. 36, the final research proposal is shown with the objec-
AD 0.3372 0.6645 4.0387 tive of supporting the applicability of results.
ADwEW 0.2662 0.1752 2.2450
PS 0.6545 0.1803 0.7792
7. Conclusions

In this work, the main objective was to predict the basic

mechanical properties of structural recycled concrete taking into
7 account, not only the recycled percentage and the quality of the
Mulvariable
recycled aggregates used, but also the production method. Lastly,
Genec Prog
simple expressions that allow engineers to estimate the recycled
6
concrete properties in a similar way to those of conventional con-
crete have been proposed, with the main conclusions as follows:
5

In general, compressive strength, modulus of elasticity and
fsp,RC experimental value (MPa)

splitting tensile strength of recycled concretes decrease as the

4
recycled concrete coarse aggregate percentage increases. Taking
into account the production methods (to pre-soak the recycled
aggregates (pre-soaked-PS), to work with air-dry aggregates
3 increasing the amount of water (air-dry with extra-water-
ADwEW), or to work with air-dry aggregates without any extra
water (air-dry without extra-water-AD)), it has been concluded
2 that ADwEW during the mixing shows the best results. With this
method the reductions in compressive strength, splitting tensile
strength and modulus of elasticity are lower (especially when
1 compared with the PS method) and, moreover, there is no need
to increase the dosage of superplasticizer.

Regarding the modulus of elasticity, it has been seen that its
0 prediction code expression (according to Eurocode) has to be
0 1 2 3 4 5 6 7 corrected to get the same approximation degree in recycled
fsp,RC predicted value (MPa)
concretes as in conventional ones. In this regard, using multi-
Fig. 30. fsp,RC predicted value vs. fsp,RC experimental value. Training performance variable regression, a correction coefficient has been adjusted
(AD method). providing good statistical indexes. This correction coefficient
takes into account the recycled concrete compressive strength,
production procedure, replacement ratio and recycled concrete
As in the modulus analysis, both techniques provide an expres- coarse aggregate quality (considering it based on its water
sion that can be suggested to calculate the splitting tensile strength absorption). Different coefficients have been adjusted for each
of recycled aggregate concrete. However, the equation optimized of the different mixing procedures.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 495

7 7
Mulvariable Mulvariable
Genec Prog Genec Prog
6 6

5 5
fsp,RC experimental value (MPa)

fsp,RC experimental value (MPa)

4 4

3 3

2 2

1 1

0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
fsp,RC predicted value (MPa) fsp,RC predicted value (MPa)

Fig. 32. fsp,RC predicted value vs. fsp,RC experimental value. Training performance Fig. 34. fsp,RC predicted value vs. fsp,RC experimental value. Training performance
(ADwEW method). (AD method).

7
Mulvariable
7
Genec Prog
Mulvariable
6
Genec Prog
6

5
fsp,RC experimental value (MPa)

5
fsp,RC experimental value (MPa)

0
0 1 2 3 4 5 6 7
fsp,RC predicted value (MPa) 0
0 1 2 3 4 5 6 7
Fig. 33. fsp,RC predicted value vs. fsp,RC experimental value. Testing performance fsp,RC predicted value (MPa)
(ADwEW method).
Fig. 35. fsp,RC predicted value vs. fsp,RC experimental value. Testing performance (AD
method).

Regarding splitting tensile strength, it has been seen that its

prediction code expression (according to Eurocode) does not
need to be corrected to get the same approximation degree in again, based on its water absorption). Also in this case, different
recycled concretes as in conventional ones. The use of the com- expressions have been adjusted for each of the different mixing
pressive strength is enough to take into account the use of recy- procedures considered, and the techniques used have been mul-
cled aggregates. tivariable regression and genetic programming. In all cases, the

Lastly, specific expressions to predict the modulus of elasticity expressions adjusted by means of genetic programming provide
and the splitting tensile strength have been adjusted. For said the best statistical indexes compared with the literature’s pro-
purpose, the Eurocode expressions have been taken as a basis, posals. Also, regarding the modulus of elasticity, the multivari-
modifying them to introduce the replacement ratio and the able regression expressions improve the predictions proposed
recycled concrete coarse aggregate quality (considering it, in the literature.
496 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

Fig. 36. Final research proposal.

Table A1
Statistical indexes for AD method – Ec expressions.

AD method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.8080 5.6644 8.9822 1.8012 0.8483 4.4249 8.2336 1.7535
Genetic Prog. 0.8971 2.0882 5.2293 1.0869 0.9048 1.9495 4.7249 1.0407
Ravi (1985a) 0.7388 4.8439 8.2254 1.6443 0.7318 4.3589 7.4571 1.7028
Ravi (1985b) 0.7435 29.6662 9.7691 5.0735 0.7332 31.3949 9.8803 4.8797
Ravi (1987) 0.7435 95.0422 8.1404 9.4776 0.7332 92.6756 7.3940 9.3359
Tang (2010) 0.7435 18.6806 9.7691 3.9542 0.7332 20.0183 9.8803 3.7258
Bair (1993) 0.7970 71.7218 12.1098 7.6389 0.8369 65.8072 11.3012 7.5032
Dhir (1999) 0.7528 25.0932 10.6841 4.4864 0.7325 29.8629 11.3283 4.4123
Dill (1998) 0.7528 96.2984 18.6008 8.5445 0.7325 120.3503 20.4147 8.9567
Mell (1999) 0.7528 10.4539 12.9597 2.4701 0.7325 14.2170 14.0001 3.3344
Xiao (2006) 0.7043 6.5197 8.6883 2.0432 0.7111 6.0185 7.5163 2.1443
Cori (2010a) 0.7389 8.4788 8.2333 2.5301 0.7319 7.2522 7.4800 2.0670
Cori (2010b) 0.7528 862.1747 16.5144 27.8252 0.7325 957.4963 18.0433 28.9787
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499 497

Table A2
Statistical indexes for ADwEW method – Ec expressions.

ADwEW method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.8078 14.4707 14.4482 2.9841 0.7597 17.6657 17.7650 3.2886
Genetic Prog. 0.8671 10.6524 11.3337 2.5508 0.8460 11.8494 12.6730 2.7795
Ravi (1985a) 0.7980 18.9172 14.8641 3.0809 0.7350 23.4009 15.3022 3.6288
Ravi (1985b) 0.8055 27.0215 14.3744 4.4556 0.7504 24.4460 17.5507 4.1943
Ravi (1987) 0.8055 89.0661 14.8382 8.7137 0.7504 70.4493 14.9410 7.2888
Tang (2010) 0.8055 19.7364 14.3744 3.6821 0.7504 20.1191 17.5507 3.6363
Bair (1993) 0.7949 35.6756 15.4812 5.0392 0.7821 37.5993 19.7464 5.4085
Dhir (1999) 0.8155 23.8056 14.0323 4.0408 0.7823 24.0211 17.3830 4.0591
Dill (1998) 0.8155 84.6957 23.7461 7.1277 0.7823 94.7654 31.1998 7.7234
Mell (1999) 0.8155 19.0480 15.9246 3.2249 0.7823 31.4233 20.9082 4.2685
Xiao (2006) 0.7373 26.1691 16.8695 3.5221 0.6365 32.8391 17.0190 4.1565
Cori (2010a) 0.7982 18.7484 14.8227 3.4553 0.7354 18.7913 15.3071 3.4348
Cori (2010b) 0.8155 755.9524 20.4253 24.9357 0.7823 731.0634 27.1253 23.4059

Table A3
Statistical indexes for PS method – Ec expressions.

PS method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.7979 7.0012 9.5912 2.2401 0.7440 10.7998 11.3794 2.5861
Genetic Prog. 0.8616 4.1204 7.5443 1.5561 0.8208 6.3237 8.5317 1.8507
Ravi (1985a) 0.7173 11.7832 10.1134 2.8113 0.5120 12.5912 12.5436 2.8763
Ravi (1985b) 0.7155 10.3339 10.4826 2.6425 0.5066 25.0926 13.1288 3.8661
Ravi (1987) 0.7155 56.3982 10.2386 6.9763 0.5066 85.5450 12.5967 8.5287
Tang (2010) 0.7155 8.4362 10.4826 2.3580 0.5066 18.4200 13.1288 3.0277
Bair (1993) 0.7155 27.7789 13.0451 4.5863 0.7240 41.7745 13.5941 5.5709
Dhir (1999) 0.7049 9.1564 10.6955 2.4212 0.4884 20.5140 13.5128 3.1397
Dill (1998) 0.7049 24.0701 17.1268 3.6791 0.4884 50.7502 18.9928 5.1458
Mell (1999) 0.7049 28.3870 12.0010 4.6077 0.4884 22.0852 14.7788 3.8679
Xiao (2006) 0.7123 13.0556 10.3578 2.9247 0.5324 12.9807 12.6062 3.0439
Cori (2010a) 0.7172 7.8622 10.1009 2.2739 0.5119 13.8623 12.5414 2.6316
Cori (2010b) 0.7049 385.3592 14.9647 18.2214 0.4884 565.0563 17.2976 22.2844

Table A4
Statistical indexes for AD method – fsp expressions.

AD method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.9155 0.1786 17.7369 0.3528 0.9557 0.1467 14.8532 0.3175
Genetic Prog. 0.9497 0.0990 13.6302 0.2451 0.9606 0.0953 11.5626 0.2529
Xiao (2006) 0.9321 0.1417 15.5208 0.3010 0.9520 0.1162 13.4749 0.2841
Katz (2003) 0.9322 1.1052 18.6360 0.9897 0.9485 1.1571 17.8589 1.0201
Li (2008) 0.8556 0.4164 19.6790 0.5117 0.8841 0.3701 17.1291 0.4595

Table A5
Statistical indexes for ADwEW method – fsp expressions.

ADwEW method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.8450 0.3022 17.0106 0.3922 0.8201 0.3585 21.4621 0.4680
Genetic Prog. 0.8611 0.2461 15.3000 0.3568 0.8340 0.3054 18.9409 0.4327
Xiao (2006) 0.8631 0.3712 15.0781 0.4816 0.8230 0.3926 19.3739 0.5338
Katz (2003) 0.8615 0.8436 15.7319 0.7975 0.8173 1.0188 18.7836 0.8715
Li (2008) 0.7927 1.0496 17.1365 0.9027 0.7720 0.9755 19.9632 0.8693

Table A6
Statistical indexes for PS method – fsp expressions.

PS method Training set Testing set

r MSE COV MAE r MSE COV MAE
Multivariable 0.5761 0.2826 17.3990 0.3863 0.7940 0.2506 19.2513 0.4490
Genetic Prog. 0.7844 0.5913 15.6460 0.5795 0.8556 0.2838 16.7421 0.4362
Xiao (2006) 0.7215 0.2117 14.1118 0.3777 0.8738 0.2526 16.7163 0.4165
Katz (2003) 0.7303 0.9120 12.7484 0.8545 0.8733 0.8198 18.6741 0.8060
Li (2008) 0.6689 0.4084 17.9447 0.5438 0.8287 0.4725 17.0003 0.5420
498 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499

Acknowledgments [22] F. Debieb, L. Courard, S. Kenai, R. Degeimbre, Roller compacted concrete with
contaminated recycled aggregates, Constr. Build. Mater. 23 (11) (2009) 3382–
3387.
The study is part of three projects entitled: (a) ‘‘Clean, efficient [23] F. Debieb, L. Courard, S. Kenai, R. Degeimbre, Mechanical and durability
and nice construction along its life cycle (CLEAM)” funded by the properties of concrete using contaminated recycled aggregates, Cem. Concr.
Compos. 32 (6) (2010) 421–426.
Centre for the Technology and Industrial Development (CDTI)
[24] Y.A. Villagrán-Zaccardi, C.J. Zega, Á.A. Di Maio, Chloride penetration and
and led by the Group of Economical Interest CLEAM-CENIT, AIE binding in recycled concrete, J. Mater. Civ. Eng. 20 (6) (2008) 449–455.
comprising by the country’s largest construction companies [25] N.F. Ortega, J.M. Moro, R. Meneses, R.R. Aveldaño, Comportamiento dinámico
de vigas de hormigón reciclado con sus armaduras corroídas, in: VI Congreso
(Acciona, Dragados, Ferrovial, FCC, Isolux Corsán, OHL and Sacyr)
Internacional sobre Patología y Recuperación de Estructuras, Argentina, June
and some PYME (Informática 68, Quilosa and Martínez Segovia y 2010.
asociados); (b) ‘‘Industrial Investigation about Concrete for a Sus- [26] A.A. Di Maio, L.P. Traversa, Evaluation of recycled concrete by means of non
tainable Market (InHorMeS)” funded by the Innovation Galician destructive tests [Evaluación de hormigones reciclados mediante ensayos no
destructivos], Mater. Constr. 53 (271–272) (2003) 37–46.
Agency; (c) ‘‘Hormigones reciclados autocompactantes robustos: [27] M.S. De Juan, Estudio sobre la utilización de árido reciclado para la fabricación
reología en estado fresco y propiedades mecánicas. (Ref.: de Hormigón Estructural (Ph.D. dissertation), Polytechnic University of
BIA2014-58063-R)” funded by MINECO. Madrid, Spain, 2005.
[28] L. Ferreira, J. De Brito, M. Barra, Influence of the pre-saturation of recycled
coarse concrete aggregates on concrete properties, Mag. Concr. Res. 63 (8)
(2011) 617–627.
Appendix A. Statistical indexes results [29] V. Corinaldesi, Structural concrete prepared with coarse recycled concrete
aggregate: from investigation to design, Adv. Civ. Eng. 2011 (2011).
See Appendix Tables A1–A6. [30] C.S. Poon, S.C. Kou, L. Lam, Influence of recycled aggregate on slump and
bleeding of fresh concrete, Mater. Struct./Mater. Constr. 40 (9) (2007) 981–
988.
References [31] M. Chakradhara Rao, S.K. Bhattacharyya, S.V. Barai, Influence of field recycled
coarse aggregate on properties of concrete, Mater. Struct./Mater. Constr. 44
(1) (2011) 205–220.
[1] L. Evangelista, J. de Brito, Durability performance of concrete made with fine
[32] R. Ravindrarajah Sri, Y.H. Loo, C.T. Tam, Recycled concrete as fine and coarse
recycled concrete aggregates, Cem. Concr. Compos. 32 (1) (2010) 9–14.
aggregates in concrete, Mag. Concr. Res. 39 (141) (1987) 214–220.
[2] A.K. Padmini, K. Ramamurthy, M.S. Mathews, Influence of parent concrete on
[33] A. Ajdukiewicz, A. Kliszczewicz, Influence of recycled aggregates on
the properties of recycled aggregate concrete, Constr. Build. Mater. 23 (2)
mechanical properties of HS/HPC, Cem. Concr. Compos. 24 (2) (2002) 269–279.
(2009) 829–836.
[34] V. Corinaldesi, G. Moriconi, Recycling of rubble from building demolition for
[3] X. Zhang, S. Deng, Y. Qin, Additional adsorbed water in recycled concrete, J.
low-shrinkage concretes, Waste Manage. 30 (4) (2010) 655–659.
Cent. South Univ. Technol. (Engl. Ed.) 14 (Suppl. 1) (2007) 449–453.
[35] F. Buyle-Bodin, R. Hadjieva-Zaharieva, Influence of industrially produced
[4] M. Rühl, Water Absorption Capacity of Recycled Demolition Rubbish, <http://
recycled aggregates on flow properties of concrete, Mater. Struct./Mater.
www.b-i-m.de/public/tudmassiv/dacon97ruehl.htm>, 1992.
Constr. 35 (252 SPEC) (2002) 504–509.
[5] A. Akbarnezhad, K.C.G. Ong, M.H. Zhang, C.T. Tam, T.W.J. Foo, Microwave-
[36] A. Domingo-Cabo, C. Lázaro, F. López-Gayarre, M.A. Serrano-López, P. Serna, J.
assisted beneficiation of recycled concrete aggregates, Constr. Build. Mater.
O. Castaño-Tabares, Creep and shrinkage of recycled aggregate concrete,
25 (8) (2011) 3469–3479.
Constr. Build. Mater. 23 (7) (2009) 2545–2553.
[6] A. Nealen, A. Schenk, The Influence of Recycled Aggregate Core Moisture on
[37] J.M. Khatib, Properties of concrete incorporating fine recycled aggregate, Cem.
Freshly Mixed and Hardened Concrete Properties, <http://www.b-i-m.de/
Concr. Res. 35 (4) (2005) 763–769.
public/tudmassiv/dacon13nealenschenk.htm>, 1998.
[38] W. Tangchirapat, R. Buranasing, C. Jaturapitakkul, Use of high fineness of fly
[7] V. Corinaldesi, Mechanical and elastic behaviour of concretes made of
ash to improve properties of recycled aggregate concrete, J. Mater. Civ. Eng.
recycled-concrete coarse aggregates, Constr. Build. Mater. 24 (9) (2010)
22 (6) (2010) 565–571.
1616–1620.
[39] M. Malešev, V. Radonjanin, S. Marinković, Recycled concrete as aggregate for
[8] M.J. Pelufo, A. Domingo, V.A. Ulloa, N.N. Vergara, Analysis of moisture state of
structural concrete production, Sustainability 2 (5) (2010) 1204–1225.
recycled coarse aggregate and its influence on compression strength of the
[40] S. Kou, C. Poon, M. Etxeberria, Influence of recycled aggregates on long term
concrete, in: Proceedings of the International Association for Shell and Spatial
mechanical properties and pore size distribution of concrete, Cem. Concr.
Structures (IASS) Symposium 2009, Valencia, 2009.
Compos. 33 (2) (2011) 286–291.
[9] C.S. Poon, Z.H. Shui, L. Lam, H. Fok, S.C. Kou, Influence of moisture states of
[41] S. Kou, C. Poon, Properties of concrete prepared with PVA-impregnated
natural and recycled aggregates on the slump and compressive strength of
recycled concrete aggregates, Cem. Concr. Compos. 32 (8) (2010) 649–654.
concrete, Cem. Concr. Res. 34 (1) (2004) 31–36.
[42] S. Kou, C. Poon, Properties of concrete prepared with crushed fine stone,
[10] M.B. De Oliveira, E. Vazquez, The influence of retained moisture in aggregates
furnace bottom ash and fine recycled aggregate as fine aggregates, Constr.
from recycling on the properties of new hardened concrete, Waste Manage.
Build. Mater. 23 (8) (2009) 2877–2886.
16 (1–3) (1996) 113–117.
[43] V. Bhikshma, R. Kishore, Development of stress–strain curves for recycled
[11] M. Casuccio, M.C. Torrijos, G. Giaccio, R. Zerbino, Failure mechanism of
aggregate concrete, Asian J. Civ. Eng. 11 (2) (2010) 253–261.
recycled aggregate concrete, Constr. Build. Mater. 22 (7) (2008) 1500–1506.
[44] J.P.B. Vieira, J.R. Correia, J. De Brito, Post-fire residual mechanical properties of
[12] M. Etxeberria, E. Vázquez, A. Marí, M. Barra, Influence of amount of recycled
concrete made with recycled concrete coarse aggregates, Cem. Concr. Res. 41
coarse aggregates and production process on properties of recycled aggregate
(5) (2011) 533–541.
concrete, Cem. Concr. Res. 37 (5) (2007) 735–742.
[45] Y. Sun, X. Xiang, Experiment research on basic mechanic property of recycled
[13] K. Rahal, Mechanical properties of concrete with recycled coarse aggregate,
concrete with different ratio of recycled aggregate, Adv. Mater. Res. 250–253
Build. Environ. 42 (1) (2007) 407–415, http://dx.doi.org/10.1016/j.
(2011) 994–1000.
buildenv.2005.07.033.
[46] Y. Li, J. Tao, T. Lei, J. Xie, Experimental study on compressive strength of
[14] B. González, F. Martínez, D. Carro, S. Seara, Stress–strain relationship in axial
recycled concrete, Adv. Mater. Res. 261–263 (2011) 75–78.
compression for concrete using recycled saturated coarse aggregate, Constr.
[47] M. Safiuddin, U.J. Alengaram, M.A. Salam, M.Z. Jumaat, F.F. Jaafar, H.B. Saad,
Build. Mater. 25 (5) (2011) 2335–2342.
Properties of high-workability concrete with recycled concrete aggregate,
[15] M.L. Berndt, Properties of sustainable concrete containing fly ash, slag and
Mater. Res. 14 (2) (2011) 248–255.
recycled concrete aggregate, Constr. Build. Mater. 23 (7) (2009) 2606–2613.
[48] M. Chakradhara Rao, S.K. Bhattacharyya, S.V. Barai, Behaviour of recycled
[16] S.C. Angulo, P.M. Carrijo, A.D. Figueiredo, A.P. Chaves, V.M. John, On the
aggregate concrete under drop weight impact load, Constr. Build. Mater. 25
classification of mixed construction and demolition waste aggregate by
(1) (2011) 69–80.
porosity and its impact on the mechanical performance of concrete, Mater.
[49] M. Batayneh, I. Marie, I. Asi, Use of selected waste materials in concrete
Struct./Mater. Constr. 43 (4) (2010) 519–528.
mixes, Waste Manage. 27 (12) (2007) 1870–1876.
[17] V. Corinaldesi, V. Letelier, G. Moriconi, Behaviour of beam-column joints
[50] J.O. Castaño, A. Domingo, C. Lázaro, A study on drying shrinkage and creep of
made of recycled-aggregate concrete under cyclic loading, Constr. Build.
recycled concrete aggregate, in: Proceedings of the International Association
Mater. 25 (4) (2011) 1877–1882.
for Shell and Spatial Structures (IASS) Symposium 2009, Valencia, 2009.
[18] J.M.V. Gómez-Soberón, Porosity of recycled concrete with substitution of
[51] Z. Chen, K. Huang, X. Zhang, J. Xue, Experimental research on the flexural
recycled concrete aggregate: an experimental study, Cem. Concr. Res. 32 (8)
strength of recycled coarse aggregate concrete, in: International Conference
(2002) 1301–1311.
on Mechanic Automation and Control Engineering (MACE), Wuhan, China,
[19] M. Barra, E. Vázquez, Particularidades do proceso de carbonatação em
2010, pp. 1041–1043.
concretos de agregado reciclado, in: IV Congreso Iberoamericano de Patología
[52] K.Y. Ann, H.Y. Moon, Y.B. Kim, J. Ryou, Durability of recycled aggregate
de las Construcciones, VI Congreso de Control y Calidad, Brasil, Octubre 1997.
concrete using pozzolanic materials, Waste Manage. 28 (6) (2008) 993–999.
[20] M. Etxeberria, A.R. Marí, E. Vázquez, Recycled aggregate concrete as structural
[53] V. Corinaldesi, G. Moriconi, Influence of mineral additions on the
material, Mater. Struct./Mater. Constr. 40 (5) (2007) 529–541.
performance of 100% recycled aggregate concrete, Constr. Build. Mater. 23
[21] L. Evangelista, J. de Brito, Mechanical behaviour of concrete made with fine
(8) (2009) 2869–2876.
recycled concrete aggregates, Cem. Concr. Compos. 29 (5) (2007) 397–401.
 I. González-Taboada et al. / Construction and Building Materials 106 (2016) 480–499
[54] L. Courard, F. Michel, P. Delhez, Use of concrete road recycled aggregates for
roller compacted concrete, Constr. Build. Mater. 24 (3) (2010) 390–395.
[55] R. Zaharieva, F. Buyle-Bodin, F. Skoczylas, E. Wirquin, Assessment of the
surface permeation properties of recycled aggregate concrete, Cem. Concr.
Compos. 25 (2) (2003) 223–232.
[56] A. Katz, Treatments for the improvement of recycled aggregate, J. Mater. Civ.
Eng. 16 (6) (2004) 597–603.
[57] S.C. Kou, C.S. Poon, C. Dixon, Influence of fly ash as cement replacement on
the properties of recycled aggregate concrete, J. Mater. Civ. Eng. 19 (9) (2007)
709–717.
[58] N. Otsuki, S. Miyazato, W. Yodsudjai, Influence of recycled aggregate on
interfacial transition zone, strength, chloride penetration and carbonation of
concrete, J. Mater. Civ. Eng. 15 (5) (2003) 443–451.
[59] S. Nagataki, A. Gokce, T. Saeki, M. Hisada, Assessment of recycling process
induced damage sensitivity of recycled concrete aggregates, Cem. Concr. Res.
34 (6) (2004) 965–971.
[60] C.S. Poon, Z.H. Shui, L. Lam, Effect of microstructure of ITZ on compressive
strength of concrete prepared with recycled aggregates, Constr. Build. Mater.
18 (6) (2004) 461–468.
[61] Y. Lin, Y. Tyan, T. Chang, C. Chang, An assessment of optimal mixture for
concrete made with recycled concrete aggregates, Cem. Concr. Res. 34 (8)
(2004) 1373–1380.
[62] S.C. Kou, C.S. Poon, D. Chan, Influence of fly ash as a cement addition on the
hardened properties of recycled aggregate concrete, Mater. Struct./Mater.
Constr. 41 (7) (2008) 1191–1201.
[63] C.J. Zega, Á.A. Di Maio, Use of recycled fine aggregate in concretes with
durable requirements, Waste Manage. 31 (11) (2011) 2336–2340.
[64] A. Chen, J. Wang, Z. Ge, Experimental study on the fundamental
characteristics of recycled concrete, Adv. Mater. Res. 295–297 (2011) 958–
961.
[65] R.M. Salem, E.G. Burdette, N.M. Jackson, Resistance to freezing and thawing of
recycled aggregate concrete, ACI Mater. J. 100 (3) (2003) 216–221.
[66] A. Garg, G. Koster,M. Rühl, Implementation of Long Term Measurements at a
Building Made of Concrete with Aggregate Derived From Concrete Rubble,
<http://www.b-i-m.de/public/tudmassiv/dacon13gargkoster.htm>, 1998.
[67] P. Grübl, A. Nealen, Construction of an office building using concrete made
from recycled demolition material, Aus Darmstadt Concr. 13 (1998).
[68] S.M. Kalaiarasu, K. Subramanian, Properties of recycled aggregate concrete
with silica fume, J. Appl. Sci. 6 (14) (2006) 2956–2958.
[69] L. Won, J. Sang, D. Hyun, Shear strength of reinforced concrete beams with
recycled aggregates, in: Tailor Made Concrete Structures, CRC Press, 2008.
225–225.
[70] J. Li, H. Xiao, Y. Zhou, Influence of coating recycled aggregate surface with
pozzolanic powder on properties of recycled aggregate concrete, Constr.
Build. Mater. 23 (3) (2009) 1287–1291.
[71] D. Kong, T. Lei, J. Zheng, C. Ma, J. Jiang, J. Jiang, Effect and mechanism of
surface-coating pozzolanics materials around aggregate on properties and ITZ
microstructure of recycled aggregate concrete, Constr. Build. Mater. 24 (5)
(2010) 701–708.
[72] G.F. Peng, Q.B. Liu, Z.B. Guo, Q.X. Hou, S.Q. Cao, J.F. Zhang, Mechanical
properties and permeability of recycled aggregate concrete at low water/
binder ratio, Key Eng. Mater. 477 (16) (2011) 16–220.
[73] R. Sato, I. Maruyama, T. Sogabe, M. Sogo, Flexural behavior of reinforced
recycled concrete beams, J. Adv. Concr. Technol. 5 (1) (2007) 43–61.
[74] S. Manzi, C. Mazzotti, M.C. Bignozzi, Short and long-term behavior of
structural concrete with recycled concrete aggregate, Cem. Concr. Compos.
37 (1) (2013) 312–318.
[75] H. Mefteh, O. Kebaïli, H. Oucief, L. Berredjem, N. Arabi, Influence of moisture
conditioning of recycled aggregates on the properties of fresh and hardened
concrete, J. Clean. Prod. 54 (2013) 282–288.
[76] K. Kim, M. Shin, S. Cha, Combined effects of recycled aggregate and fly ash
towards concrete sustainability, Constr. Build. Mater. 48 (2013) 499–507.
[77] H.Z. Cui, X. Shi, S.A. Memon, F. Xing, W. Tang, Experimental study on the
influence of water absorption of recycled coarse aggregates on properties of
the resulting concretes, J. Mater. Civ. Eng. 27 (4) (2014).
[78] B. González-Fonteboa, F. Martínez-Abella, M.F. Herrador, S. Seara-Paz,
Structural recycled concrete: behaviour under low loading rate, Constr.
Build. Mater. 28 (1) (2012) 111–116.
[79] S. Seara-Paz, B. González-Fonteboa, J. Eiras-López, M.F. Herrador, Bond
behavior between steel reinforcement and recycled concrete, Mater. Struct./
Mater. Constr. 47 (1–2) (2014) 323–334.
[80] P. Folino, H. Xargay, Recycled aggregate concrete – mechanical behavior under
uniaxial and triaxial compression, Constr. Build. Mater. 56 (2014) 21–31.
[81] M. Gomes, J. de Brito, M. Bravo, Mechanical performance of structural
concrete with the incorporation of coarse recycled concrete and ceramic
aggregates, J. Mater. Civ. Eng. 26 (10) (2014).
[82] A. González, M. Etxeberria, Experimental analysis of properties of high
performance recycled aggregate concrete, Constr. Build. Mater. 52 (2014)
227–235.
[83] D. Pedro, J. De Brito, L. Evangelista, Influence of the use of recycled concrete
aggregates from different sources on structural concrete, Constr. Build. Mater.
71 (2014) 141–151.
[84] M. Pepe, R.D. Toledo Filho, E.A.B. Koenders, E. Martinelli, Alternative
processing procedures for recycled aggregates in structural concrete,
Constr. Build. Mater. 69 (2014) 124–132.
499
[85] A. Barbudo, J. De Brito, L. Evangelista, M. Bravo, F. Agrela, Influence of water-
reducing admixtures on the mechanical performance of recycled concrete, J.
Clean. Prod. 59 (2013) 93–98.
[86] S. Kou, C. Poon, Long-term mechanical and durability properties of recycled
aggregate concrete prepared with the incorporation of fly ash, Cem. Concr.
Compos. 37 (1) (2013) 12–19.
[87] S. Kou, C. Poon, Effect of the quality of parent concrete on the properties of
high performance recycled aggregate concrete, Constr. Build. Mater. 77
(2015) 501–508.
[88] F. López, Influencia de la variación de los parámetros de dosificación y
fabricación de hormigón reciclado estructural sobre sus propiedades físicas y
mecánicas (Ph.D. dissertation), University of Oviedo, Spain, 2008.
[89] N.K. Bairagi, K. Ravande, V.K. Pareek, Behaviour of concrete with different
proportions of natural and recycled aggregates, Resour. Conserv. Recycl. 9 (1–
2) (1993) 109–126.
[90] J.Zh. Xiao, J.B. Li, Ch. Zhang, On relationships between the mechanical
properties of recycled aggregate concrete: an overview, Mater. Struct. 39
(2006) 655–664.
[91] Xuping. Li, Recycling and reuse of waste concrete in china: Part I. Material
behaviour of recycled aggregate concrete, Resour. Conserv. Recycl. 53 (2008)
36–44.
[92] A. Katz, Properties of concrete made with recycled aggregate from partially
hydrated old concrete, Cem. Concr. Res. 33 (5) (2003) 703–711.
[93] R.K. Dhir, M.C. Limbachiya, T. Leelawat, Suitability of recycled aggregate for
use in BS 5328 designated mixes, Proc. Inst. Civ. Eng. 134 (1999) 257–274.
[94] R. Dillmann, Concrete with recycled concrete aggregate, in: Proceedings of
the 5th ACI International Conference on Innovation in Design with Emphasis
on Seismic, Wind and Environmental Loading; Quality Control and
Innovations in Materials/Hot-Weather Concreting, Cancun, Mexico, 1988,
pp. 461–474.
[95] M. Kakizaki, M. Harada, T. Soshiroda, S. Kubota, T. Ikeda, Y. Kasai, Strength
and elastic modulus of recycled aggregate concrete, in: Proceedings of the
2nd International RILEM Symposium on Demolition and Reuse of Concrete
and Masonry, Tokyo, Japan, 7–11 November 1988, pp. 726–735.
[96] R. Ravindrarajah Sri, C.T. Tam, Properties of concrete made with crushed
concrete as coarse aggregate, Mag. Concr. Res. 37 (130) (1985) 29–38.
[97] J. De Brito, R. Robles, Recycled aggregate concrete (RAC) methodology for
estimating its long-term properties, Indian J. Eng. Mater. Sci. 17 (2010) 449–
462.
[98] Z.H. Duan, S.C. Kou, C.S. Poon, Using artificial neural networks for predicting
the elastic modulus of recycled aggregate concrete, Constr. Build. Mater. 44
(2013) 524–532.
[99] J.L. Pérez, A. Cladera, J.R. Rabuñal, F. Martínez-Abella, Optimization of existing
equations using a new genetic programming algorithm: application to the
shear strength of reinforced concrete beams, Adv. Eng. Softw. 50 (2012) 82–
96.
[100] J.L. Pérez-Ordóñez, Metodología para orientar procesos de extracción de
conocimiento basados en Computación Evolutiva. Aplicación al desarrollo de
modelos y formulaciones en el ámbito del hormigón estructural (Ph.D.
dissertation), Department of Information and Communication Technologies,
University of A Coruña, Spain, 2010.
[101] A. Cladera, J.L. Pérez-Ordóñez, F. Martínez-Abella, Shear strength of RC
beams. Precision, accuracy, safety and simplicity using genetic programming,
Comput. Concr. 14 (4) (2014) 479–501.
[102] R.V. Silva, J. De Brito, R.K. Dhir, Tensile strength behaviour of recycled
aggregate concrete, Constr. Build. Mater. 83 (2015) 108–118.
[103] R.V. Silva, J. De Brito, R.K. Dhir, The influence of the use of recycled aggregates
on the compressive strength of concrete: a review, Eur. J. Environ. Civ. Eng. 19
(7) (2015) 825–849.
[104] Z.H. Duan, S.C. Kou, C.S. Poon, Prediction of compressive strength of recycled
aggregate concrete using artificial neural networks, Constr. Build. Mater. 40
(2013) 1200–1206.
[105] A. Trocoli, M. Batista, K. De Jesus, Prediction of compressive strength of
concrete containing construction and demolition waste using artificial neural
networks, Constr. Build. Mater. 38 (2013) 717–722.
[106] A. Behnood, J. Olek, M.A. Glinicki, Predicting modulus elasticity of recycled
aggregate concrete using M50 model tree algorithm, Constr. Build. Mater. 94
(2015) 137–147.
[107] G. Mellmann, Processed concrete rubble for the reuse as aggregate, in:
Proceedings of the International Seminar on Exploiting Waste in Concrete,
Dundee, Scotland, 1999.
[108] Z. Zilch, F. Roos, An equation to estimate the modulus of elasticity of concrete
with recycled aggregates, Civ. Eng. 76 (2001) 187–191.
[109] P. Carrijo, Analysis of the Influence of the Density of Coarse Aggregates from
Construction and Demolition Waste on the Mechanical Performance of
Concrete (master dissertation), University of São Paulo, Brazil, 2005.
[110] M. Leite, Evaluation of the Mechanical Properties of Concrete Made with
Recycled Aggregates from Construction and Demolition Waste (Ph.D. thesis),
Federal University of Rio Grande do Sul, Porto Alegre, Brazil, 2001.
[111] S.C. Kou, C.S. Poon, D. Chan, Properties of steam cured recycled aggregate fly
ash concrete, in: International RILEM Conference on Use of Recycled
Materials in Buildings and Structures, Barcelona, 2004, pp. 590–599.