Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

Contents lists available at SciVerse ScienceDirect

Chemometrics and Intelligent Laboratory Systems


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

A TOPSIS-based Taguchi optimization to determine optimal mixture


proportions of the high strength self-compacting concrete
Barış Şimşek a, Yusuf Tansel İç b,⁎, Emir H. Şimşek a
a
Department of Chemical Engineering, Faculty of Engineering, Ankara University, 06100, Tandoğan, Ankara, Turkey
b
Department of Industrial Engineering, Faculty of Engineering, Baskent University, 06810, Baglica, Etimesgut, Ankara, Turkey

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

Article history: In general, the optimization problems contain more than one response, which often conflict with each other.
Received 18 October 2012 This paper proposes the TOPSIS-based Taguchi optimization approach to determine the optimal mixture
Received in revised form 22 March 2013 proportions of the high strength self-compacting concrete (HSSCC) in a ready-mixed concrete plant. The
Accepted 25 March 2013
performance criteria are identified for the following: the average convective heat transfer coefficient, the per-
Available online 29 March 2013
centage of air content, the slump flow, the T50 time, the water absorption, the compressive strength, the split-
Keywords:
ting tensile strength, and the production cost. Five factors having three control levels and one factor having
High strength self-compacting concrete two control levels affect these identified performance criteria. The data of the HSSCC quality criteria are
(HSSCC) obtained by running scenarios that combine factor levels in Taguchi design, while signal to noise (S/N) ratios
Multi-response optimization are calculated for the data. After a decision matrix is generated by the S/N ratios, the TOPSIS (Technique for
TOPSIS Order Preference by Similarity to an Ideal Solution) method is then used to transform the multi-response
Taguchi method problem into a single-response problem. The anticipated improvement rate is also determined by finding
the levels of the factors in order to optimize the system which uses Taguchi's single response optimization
methodology.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction et al. [14] analyzed the influence of the composition and the curing con-
ditions on the bending strength of the polyester and the epoxy concrete
In the literature several optimization methods have been proposed by using the Taguchi method.
to investigate the optimal mixture proportions for many concrete The practical applications usually include some features of a
types. In order to investigate the effects of the variables on the concrete multi-response optimization problem [15,16]. When multiple re-
quality, there are two general approaches most commonly used in the sponses should be optimized, the decision maker usually determines
literature which are the Taguchi method and the response surface the optimal factor levels based on his experiences. On the other hand,
methodology (RSM) [1–7]. some multi-response optimization applications use complex mathe-
Several researchers studied the mechanical properties of the con- matical models [2]. The multi attribute decision making (MADM)-based
crete by using response optimization methodology [8–14]. Hinişlioğlu Taguchi models are proposed for Taguchi applications in which the
et al. [8] studied the optimization of the early flexural strength of pave- multi-response is involved [16]. The TOPSIS-based Taguchi optimiza-
ment concrete with the silica fume and the fly ash by using the Taguchi tion is an appropriate procedure for the multi-response optimization
method. Muthukumar et al. [9] optimized the mechanical properties of studies. The TOPSIS-based Taguchi method is quite simple and adapt-
the polymer concrete and recommended the mix-design based on the able compared to other MADM-based Taguchi methods, such as
design of the experiment. Tan et al. [10] investigated the effect of ben- VIKOR (Vlse Kriterijumska Optimizacija I Kompromisno Resenje) and
tonite, fly ash and silica fume on the bleeding of cement based grouts. GRA (Grey relational analysis). These methods are known in the litera-
Turkmen et al. [11], Fauzan et al. [12], Simon et al. [13] and Ribeiro ture as the methods which have increasing difficulties for the user with
et al. [14] investigated the most influential factor of concrete mixing an increasing number of responses. Recently, the TOPSIS based Taguchi
on the properties of concrete by using the Taguchi design of the exper- method has been utilized to solve the multi-response optimization
imental method [3]. Moreover, Ozbay et al. [3] proposed the optimum problems in some fields, such as the evaluation of the microinjection
mix proportions and the ranking of the effective parameters of the molding process parameters of the light guide plate (LGP) [17], the op-
fresh and hardened properties of the high strength self-compacting timization of the melt-spinning process parameters which are used to
concrete (HSSCC) by using the Taguchi method. Additionally, Ribeiro produce polyester (PET)/titanium dioxide (TiO2) UV-resistant fibers
[18], the improvement of the CNC turning process [19], the improve-
⁎ Corresponding author. Tel.: +90 312 2466666/1316. ment of the quality of the submerged arc welding process [20], the iden-
E-mail address: ytansel@baskent.edu.tr (Y.T. İç). tification of an optimal hybrid push/pull production system [21], and

0169-7439/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.chemolab.2013.03.012
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 19

the simulation optimization in the production systems [15]. The authors Table 2
are not aware of any literature that discusses the multi-response ready Properties of the SP at 20 °C.

mixed dosage optimization problem by using the TOPSIS-based Taguchi Properties Super plasticizer
method.
Chemical description Polycarboxylic type polymer
This paper proposes a multi-response optimization approach to Color Brown
determine the optimal mixture proportions of the HSSCC. The main Specific gravity (kg/L) 1.08–1.14
contribution of this article is to demonstrate the application of the Chlorin content % (EN 480-10) b0.1
Alkaline content% (EN 480-12) b3
hybrid TOPSIS and the Taguchi method to determine the optimal
State Liquid
mixture proportion of an HSSCC. Today, the HSSCC is a well-known
concrete technology [3]. It has been established on the basis of vari-
ous civil engineering applications that the main benefit that can be
expected from this kind of material is the high speed of casting [3]. The TOPSIS procedure is used to integrate all determined perfor-
First of all, the criteria to determine the quality of the concrete, the re- mance values of the system into a single value that can then be
sponse, the factors and their levels that affect these performance used as a single performance in the multi-variable optimization
criteria are defined to obtain the optimal mixture proportions. The problems.
data of the determined performance criteria are obtained by running The TOPSIS-based Taguchi application is very simple, adaptable
scenarios that combine the factor levels in the Taguchi design. Then, and easy to perform compared to the other multi-response simula-
the S/N ratios will be calculated for the data. After a decision matrix tion optimization methods such as the GRA-based Taguchi method
is generated with the S/N ratios, the TOPSIS method will be used to and the RSM. The application of the TOPSIS-based Taguchi method
transform the multi-response problem into a single-response prob- does not require complicated statistical tools, and reduces the com-
lem. As a result, the best factor levels will be identified according to plexity of the mathematical calculations [15].
the Taguchi method principles for a single-response problem. The proposed TOPSIS-based Taguchi method includes the idea of a
The remainder of this paper in which the mixture proportion multi-response robust design method to solve the multi-response
optimization of the HSSCC at a ready-mixed concrete plant in Turkey optimization problem. The robust design provides a systematic and
is analyzed, is organized as follows: a brief introduction to the mate- efficient methodology for determining the optimum combination of
rials in the plant is provided in Section 2. The details of the solution the design factors such that the product is effective and has high
methodology are discussed in Section 3. Identified performance opti- performance, and also is robust to the noise factors [15,16]. The
mization conditions for the HSSCC are given in Section 4. Section 5 noise factors are known as the uncontrollable factors. In the tradition-
provides the empirical results. Finally, Sections 6 and 7 present the al approach, the products that do not meet the tolerances lead to a
conclusions. quality loss on the manufacturer, and a loss arises in the form of a
waste material in the plant, which the producer adds to the cost of
the product. Moreover, the products that meet the tolerance also
2. Materials
lead to a quality loss that the customer incurs which can adversely
affect the profit and the reputation of the company [15].
The cement used in this research is a CEM I 42.5 R. It has a specific
The idea of the quality loss is applicable to the study when we trans-
gravity of 3.08 and Blaine fineness of 3742 cm2/g. Fly ash with a specific
form the performance measures into the signal to noise (S/N) ratios.
gravity of 2.46 is used. The chemical composition of the cementitious
Also, the S/N ratio has some benefits over the multi-objective approach.
materials is given in Table 1. A polycarboxylic type superplasticizer
The optimization steps are simplified when the S/N ratio is used for the
(SP) is used in all concrete mixtures and physical properties of SP are
process optimization and the determination of the adjustment factors.
given in Table 2 [3]. Crushed sands (with a size smaller than 4 mm
Additionally, the design or the development processes also become
(number I) as fine aggregate and with a size between 4 mm and
more reliable, especially when the various different responses can be
11.2 mm as the coarse aggregate (number II)) are used in concrete
treated as the dynamic characteristics [22]. The S/N ratio, (η) is a useful
mixtures. Fig. 1 presents aggregate sieve analysis. The fine and coarse
way to obtain significant factors by evaluating the minimum variance
aggregates have specific gravities of 2.65 and 2.70 and mean water
[15,16]. The S/N ratio can be used to measure the performance of a sys-
absorptions of 1.5% and 0.9%, respectively. Well water is used for the
tem. The-smaller-the-better and the-larger-the-better responses are
test. Temperature measurements are made with a ±0.1 °C precision
suitable for the ten performance measures determined for the HSSCC.
thermometer.

3. Proposed multi-response optimization methodology

There are 8 flow steps in the determination of the optimal mix


proportions of the HSSCC. This flow diagram is given in Fig. 2.

Table 1
Chemical composition of cement and fly ash.

Chemical analysis CEM I 42.5 R (%) Fly ash (%)

CaO 66.25 4.76


SiO2 21.79 56.21
Al2O3 5.98 23.1
Fe2O3 2.51 6.51
SO3 1.54 0.73
MgO 1.15 2.11
K2O 0.61 2.53
Na2O 0.15 0.27
Cl 0.0071 0.0018
Loss of ignition 3.71 2.24
Fig. 1. Aggregate grain-size curve.
20 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

STEP 1 6.1
• Determine concrete performance • Experimental design
optimization objectives
STEP 2 6.2
• Determine criteria and constraints of
• Signal to noise ratio calculation
mixture proportions
STEP 3 6.3
• Determination of factors and their levels
• Multi-response optimization problem
STEP 4
• Determine test conditions 6.4
STEP 5 • TOPSIS Method
• Performance Optimization of High • Determination of decision matrix
Strength Self Compacting Concrete • Determination of weighted normalization decision
• Determine average convective heat transfer matrix
coefficient • Identify positive and negative ideal solution
• Estimation of production cost • Calculate seperation measures
• Obtain experiment results • Calculate similarities to ideal solution
• Rank reference order
STEP 6
• TOPSIS basedTaguchi optimization 6.5

STEP 7 • Single response optimization problem


• Determination of optimal levels
6.6
STEP 8
• Verification of model • Taguchi optimization

Fig. 2. Proposed performance optimization framework.

The S/N ratios for the smaller-the-better and the-larger-the-better re- iii. Determination of weighted normalized decision matrix (V):
sponses are given by Eqs. (1) and (2) respectively [15].
V ¼ ⌊vij ⌋mn i ¼ 1; 2; … ::; m; j ¼ 1; 2; …:; n ð5Þ
" #
1X n
2
ηij ¼ −10 log10 y ð1Þ and,
n k¼1 ijk

" # υij ¼ wj r ij i ¼ 1; :::::; m j ¼ 1; :::::; n: ð6Þ


1X n
1
ηij ¼ −10 log10 : ð2Þ
n k¼1 y2ijk Weight of the response j is represented by wj in Eq. (6) and
X
n
wj ¼ 1.
In Eqs. (1)–(2); ηij is the S/N ratio for the response j of experimen- j¼1
tal number i, and yijk is the experiment result for the response j of the iv. Identification of positive ideal and negative ideal solutions
experiment i, in the kth replication; n is the total number of replica- (A* and A −):
tions [15,16]. The positive ideal solution, A* (Ai⁎; i = 1,2,…,m), is made of all
The multiple-response problem is converted to a single-response the best values (maximum S/N ratio) and the negative-ideal
problem with the following steps by using the TOPSIS methodology solution, A − (Ai −; i = 1,2,…,m), is made of all the worst values
[15]: (minimum S/N ratio) at the responses in the weighted normal-
i. Determination of the decision matrix: ized decision matrix (V). They are calculated by using Eqs. (7)
In the TOPSIS application, the characteristic values of alterna- and (8). In these equations, I is the set of cost type criteria
tives (experimental numbers) at attributes (calculated S/N and I′ is the set of benefit type criteria [15,21].
ratios for responses) (ηij; i = 1,2,…, number of experiment    
(m), j = 1,2,…. number of responses (n)) are inputs and placed A ¼ max vij jj∈J ; min vij jj∈J ′ ð7Þ
i i
in matrix form as shown in Eq. (3).

2 3    
η11 η12 ⋯ η1n −
A ¼ min vij jj∈J ; max vij jj∈J ′ : ð8Þ
6 η21 η22 ⋯ η2n 7 i i
6
D¼4 7: ð3Þ
⋯ ⋯ ⋯ ⋯ 5
ηm1 ηm2 ⋯ ηmn v. Calculation of the separation measures: the distance of an
alternative (experimental number) i to the positive ideal solu-
ii. Calculation of normalized ratings by the vector normalization: tion (Si*), and the distance from the negative ideal solution
(Si −) are calculated by using Eqs. (9) and (10).
ηij vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ij ¼ − sffiffiffiffiffiffiffiffiffiffiffiffiffi i ¼ 1; …; m and j ¼ 1; …n: ð4Þ uX
X m  u n  
 2
2
ηij Si ¼t vij −vj ð9Þ
i¼1 j¼1
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 21

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uX
− u n  − 2
 For each concrete mix, the compressive strength of three 15 cm
Si ¼t vij −vj : ð10Þ cubes is determined for 2, 7, 28 days according to the “TS EN
j¼1 12390/3” [29]. Splitting tensile strength tests are conducted on
three 150 mm cube samples for 28 days according to the Turkish stan-
vi. The ranking score (Ci⁎) is calculated using Eq. (11). dard “TS 3129” [3,30,31]. Each compressive strength and split tensile
strength experiment is conducted on an average of the three 150 mm
 S−
i cube specimens. Four criteria are identified as the compressive strength
Ci ¼ : ð11Þ
S−
i þ Si for 2, 7, 28 days and as the split tensile strength for 28 days. The
compressive strength and splitting tensile strength of the HSSCC are
4. Identifying performance optimization conditions of HSSCC in the ranges of 50.9–74.9 N/mm2, and 3.9–5.0 N/mm2, respectively
[3]. Higher compressive and higher split tensile strength means a
4.1. Determination of the concrete performance optimization objectives better concrete quality. The compressive test apparatus can be seen
in Fig. 3.
The assessment of the optimal mixture is an important issue to ob- The other criterion is that the average heat transfer coefficient
tain desired quality [2]. The lower rates of the input used in the pro- is required to be the maximum. Buildings are large consumers of
duction of the quality concrete with the desired characteristics can be energy in all countries [32]. Significant energy savings could be
achieved by the optimization methods. achieved in the buildings if they are properly designed and operat-
The purposes of the optimization of the mixture proportions in ed. As a least cost energy strategy, the energy conservation should
this study are as follows: The first one is the optimization of the be supported in the future [32]. The building designers can contrib-
responses that determine the concrete quality by using MINITAB® ute to solve the energy problem if proper early design decisions are
and the determination of the optimal mixture proportions. The sec- made regarding the selection and integration of the building com-
ond one is the evaluation of the average convective heat transfer ponents [32,33].
coefficient and the production cost which are the two quality criteria In order to improve the thermal resistance of the structures and to
for HSSCC. Therefore, this study provides useful tools for the system reduce the heating loads, some new construction materials such as
performance improvement to many organizations which have the the aerated concretes are used in the constructions [34]. In the pro-
ready-mixed concrete plant. duction of concrete, the aim should be producing a concrete type
In this study, the best possible factor levels of the mix proportions that minimizes the heat loss. Therefore, the production of concrete
are analyzed to maximize the splitting tensile strength and the com- with a low coefficient of the heat transfer should be targeted.
pressive strength in order to minimize the water absorption, the air The surface temperature values of the fresh concrete mixes
content, the production cost, and the total heat transfer coefficient are determined on three 150 mm cube samples during the first
values by using the TOPSIS-based Taguchi method. The optimization 20–30 min after transforming the concrete into the molds (called
of the mixture proportions plays an important role in providing com- early-age concrete). The average temperature values of the twenty
petitive advantage and customer satisfaction. The assessment of the points which belong to each sample are selected as the concrete sur-
optimal mixture is an important issue to obtain desired quality [2]. face temperature. For each sample average, the concrete surface
temperature is the average of the temperature of the three samples.
4.2. Determination criteria and constraints of mixture proportions of The air temperature is used to calculate the convection heat transfer
standard concrete optimization coefficient.
Ten quality criteria are identified for the HSSCC. The first perfor-
The concrete temperature, the air content, the air temperature, the mance criterion is identified as the average convection heat transfer
slump flow and the unit weight of the fresh concrete are measured. coefficient which provides information on energy consumption and
The minimum concrete temperature of fresh concrete has to be
+ 5 °C according to the TS EN 206-1 (Turkish Standard) [23]. The vis-
cosity of the produced concrete is evaluated through the slump flow
test according to the TS EN 12350/5 (Turkish standard) [24]. The
slump flow represents the mean diameter of the mass of the concrete
after the release of a standard slump cone on the flow table. The di-
ameter is measured in two perpendicular directions after spreading
has stopped. According to Nagataki and Fujiwara, the slump flow
range has to be between 500 and 700 mm for a concrete to be consid-
ered as self compacting [25,3]. Moreover, the slump flow range must
be larger than 600 mm for HSSCC [20]. In addition to measuring the
spread diameter, the elapsed time beginning from the moment that
concrete starts to spread from the cone until the diameter of the con-
crete reaches to 500 mm is measured [26]. This elapsed time which is
called T50 time is recorded in seconds [27]. The shortest T50 time is
observed in the case of the fastest compact of HSSCC [28]. The T50
time is recommended to be between 3 and 6 s [28]. The slump flow
test is a kind of a simple and fast implementation experiment both
in the laboratory and the field [28]. Therefore, the two criteria that
have effects on the concrete quality are the slump flow value that
should be the maximum and T50 time that should be the minimum.
Another criterion is that the percentage of the air content of the stan-
dard concrete should be the minimum because it is inversely propor-
tional to the compressive strength [3]. The water absorption is known
as another criterion of the concrete mixtures which also should be the
minimum [3]. Fig. 3. Compressive strength test apparatus.
22 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

Table 3
Quality characteristic and their weights.

Quality characteristic Symbol Description Type of concrete test Target values Weights Normalized weights

1 R1 Convection heat transfer coefficient (W/m2K) Fresh concrete test Smaller is better 3 0.059
2 R2 Air content (%) Fresh concrete test Smaller is better 4 0.078
3 R3 Slump flow (cm) Fresh concrete test Larger is better 8 0.157
4 R4 T50 time (second) Fresh concrete test Smaller is better 5 0.098
5 R5 Water absorption (%) Hardened concrete test Smaller is better 4 0.078
6 R6 Compressive strength (N/mm2) Hardened concrete test Larger is better 6 0.118
2 days
7 R7 Compressive strength (N/mm2) Hardened concrete test Larger is better 4 0.078
7 days
8 R8 Compressive strength (N/mm2) Hardened concrete test Larger is better 9 0.176
28 days
9 R9 Splitting tensile strength (N/mm2) Hardened concrete test Larger is better 3 0.059
28 days
10 R10 Production cost ($/mm2) Fresh concrete test Smaller is better 5 0.098
Total 51 1.000

thermal damage by convective heat transfer in the structures located and coarse aggregate are first mixed with 1/3 water followed by the ad-
at the coastal regions or the areas under the influence of the strong dition of Portland Cement (PC) and 1/3 water. Finally, pre-mix of water
winds [35]. and superplasticizer admixture is added to the mix. Mixture time is
The second and third performance criteria are identified respective- started to be recorded from this moment. From each concrete mix,
ly as the percentages of the air content and the slump flow. Less air con- three 15 cm cube samples for 7 and 28 days compressive strength
tent provides higher concrete durability and higher slump flow value and three 150 mm cubes specimen for 28 days splitting tensile strength
gives higher concrete workability. The other two performance criteria are collected. Water absorption tests are performed on three 150 mm
are identified respectively as the percentage of the water absorption cube samples according to “TS EN 12390/7” [3,36].
obtained from the hardened concrete test and the unit weight values
obtained from the fresh concrete test. These criteria give an idea about 5. Performance optimization of high strength self
the water resistance of the concrete and amount of gaps in the concrete. compacting concrete
The other criteria are identified as splitting the tensile strength for
28 days and the compressive strength for 2, 7 and 28 days. These 5.1. Convection heat transfer coefficient, hc
criteria provide information on the concrete durability. The last crite-
rion is the production cost that provides cost saving. The weight The heat transfer due to the convection occurs because of the tem-
values that are given out of ten values are presented in Table 3 for perature difference between the ambient air and the concrete surface
nine performance/quality criteria. [35]. In the convective heat transfer problems, the flow is generally
designed as a forced convection flow [35,36]. Under the light of this
information, it is considered that the convective heat transfer from
4.3. Determination of factors and their levels the concrete surface constitutes the form of forced convection [35].
However, it is necessary to evaluate the thermal conditions such as
Five factors which has each three control levels and one factor which the convective heat transfer coefficient which is one of the most im-
has two control levels affect the HSSCC identified quality. Cement dos- portant thermal properties associated with the air convection [35].
age is identified as a two level factor and water to cementitious material Previous studies have demonstrated that the range of the convective
ratio, aggregate mixture ratio, superplasticizer content; fly ash content heat transfer coefficient is between 5 and 35 W/(m 2K) [37–40]. How-
and mixture time of fresh concrete were identified as three level factors. ever, Yamagawa et al. [41] suggested that the coefficient has a value
These factors are symbolized as A, B, C, D, E and F respectively (Table 4). between 9 and 13 W/(m 2 K) for the same range of wind velocity
Current levels which are the estimated mixture levels before Taguchi (2–3 m/s) as used in the study of Ohbayashigawa Dam 14–15 W/
experiments were identified by an underscore. (m 2 K) in Japan [35,42], where, Q is the heat loss of the concrete sur-
face with a convection, hc is the coefficient of the convection heat
4.4. Experiment conditions transfer, A is the surface area, Tw is the temperature of the plate (con-
crete surface) and T∞ is the temperature of the fluid (air) [43].
The mixes are prepared about 5–6 min with a rotating planetary The coefficient of the convection heat transfer of the isothermal
mixer. The total aggregate mixture weight is 1634–1637 kg/m3. Sand flat plate heated over entire length is calculated by using Newton's
law of cooling which expresses the overall effect of convection
(Eq. (12)) [43].
Table 4
Levels of factors that affect quality characteristic.
Q
Factors Description Bounds hc ¼ : ð12Þ
A  ðT w −T ∞ Þ
First Second Third
bound bound bound −
The coefficient of the average convection heat transfer (hc ) which
A Cement dosage (kg) 400 425a N/A
is the average of the coefficients of the convection heat transfer of
B Water to cementitious material ratio 0.35 0.37a 0.39
C Aggregate mixture ratio (I/II) 0.60a 0.65 0.70 three 150 mm cubes given in Table 5, are calculated by using
D Super plasticizer content (kg/m3) 1.00 1.25a 1.50 Eq. (13) for all experiments, where L is the sample length and Nu is
E Fly ash content (kg) 80 100 120a called the Nusselt number given in Eq. (14). Re is the Reynolds num-
F Mixture time (s) 100a 110 120 ber which can be interpreted physically as a dimensionless group that
a
Estimated mixture level. represents the ratio of the inertia forces to the viscous forces in the
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 23

Table 5
Average convection heat transfer coefficient and estimated production costs for all experiments.

Average convection heat transfer coefficient for all experiments Estimated production cost for all experiments ($/m3)

Exp. Tw T∞ Tf ν ∗ 106, m2/sn k ∗ 103, Pr Re Nu h (W/m2K) Fly ash Cement SP amount Water Aggregate Aggregate Production
no. (K) (K) (K) W/mK #1 #2 cost

1 292.2 284.4 288.25 14.239 25.408 0.71064 21,069 43.004 14.569 2.26 27 8.96 0.581 4.184 2.446 45.635
2 291 284.4 287.65 14.198 25.357 0.71083 21,130 43.07 14.562 2.83 27 11.62 0.468 4.696 2.216 49.03
3 292 284.4 288.15 14.232 25.4 0.71067 21,079 43.015 14.568 3.4 27 14.2 0.475 4.971 1.869 52.11
4 291.7 284.4 288 14.222 25.387 0.71072 21,094 43.031 14.566 2.83 27 8.96 0.5 4.184 2.446 46.12
5 291.8 284.4 288.05 14.225 25.391 0.71071 21,089 43.026 14.566 3.4 27 11.62 0.484 4.696 2.216 49.612
6 292.8 284.4 288.55 14.26 25.434 0.71055 21,038 42.971 14.572 2.26 27 14.2 0.516 4.971 1.869 51.019
7 292.5 284.4 288.4 14.25 25.421 0.7106 21,053 42.987 14.57 2.26 27 11.62 0.519 4.184 2.446 48.232
8 292.5 284.4 288.4 14.25 25.421 0.7106 21,053 42.987 14.57 2.83 27 14.2 0.534 4.696 2.216 51.676
9 292.7 284.4 288.5 14.257 25.43 0.71057 21,043 42.976 14.572 3.4 27 8.96 0.5 4.971 1.869 46.895
10 290.9 284 287.4 14.181 25.336 0.71091 21,156 43.098 14.559 3.4 29 14.2 0.484 4.184 2.446 53.61
11 291 284 287.45 14.184 25.34 0.71089 21,151 43.092 14.56 2.26 29 8.96 0.5 4.696 2.216 47.536
12 291.5 284 287.7 14.201 25.362 0.71081 21,125 43.065 14.562 2.83 29 11.62 0.531 4.971 1.869 50.721
13 291.5 284 287.7 14.201 25.362 0.71081 21,125 43.065 14.562 3.4 29 14.2 0.513 4.184 2.446 53.638
14 291.3 284 287.6 14.194 25.353 0.71084 21,135 43.076 14.561 2.26 29 11.62 0.531 4.696 2.216 50.227
15 291.4 284 287.65 14.198 25.357 0.71083 21,130 43.07 14.562 2.83 29 8.96 0.55 4.971 1.869 48.079
16 290.5 284 287.2 14.167 25.319 0.71097 21,176 43.12 14.557 2.83 29 14.2 0.55 4.184 2.446 53.11
17 290.3 284 287.1 14.16 25.311 0.711 21,187 43.131 14.556 3.4 29 8.96 0.563 4.696 2.216 48.73
18 290.4 284 287.15 14.163 25.315 0.71098 21,181 43.126 14.556 2.26 29 11.62 0.544 4.971 1.869 50.167

forced-convection flow system [43] and it can be calculated by using 5.2. Estimated production cost and experiment results
Eq. (15) [43]. Additionally, k, is the thermal conductivity from
Eq. (13), Pr, is the Prandtl number from Eq. (14) and ν, is the kine- The self-compacting concrete (SCC) is a developed version of the
matic viscosity from Eq. (15) which are all taken from the tables of conventional concrete, in which the using of the vibrator for
the air properties at atmospheric pressure according to the average compacting is no more required [44,45]. This property of the
film temperature, Tf, given in Eq. (16) [43]. The subscript “f” indicates self-compacting concrete has made its use more attractive through-
that the properties in the dimensionless groups are evaluated at the out the world [44]. But its higher initial supply cost over the conven-
film temperature [43]. The wind velocity u∞ is identified as 3 m/s. tional concrete has hindered its application to general construction
[44]. Therefore, in order to produce the low cost SCC, it is prudent

− Nu L  k to look for the alternatives that reduce the SSC cost [44].
hc ¼ ð13Þ
L In this research, the main variables are; the amount of fly ash and
cement, the dosage of superplasticizer for flowability, the amount of
−     water, and the aggregate numbers I and II of the mixture ratio. The av-
1=2 1=3
Nu L ¼ 0:664  ReL  Pr ð14Þ erage costs of the fly ash, the cement, the superplasticizer, the water,
and the aggregate numbers I and II are determined as 28.3 $/tons,
68 $/tons, 2 $/kg, 3.125 $/tons, 4.3 $/tons, 3.7 $/tons on average,
u∞ L
ReL ¼ ð15Þ respectively. The estimated production costs for all experiments are cal-
ν
culated and given in Table 5 (columns 11–17). L18 (21 × 3 7) orthogonal
array is used to implement the experiments and its results are also
T f ¼ ðT w þ T ∞ Þ=2: ð16Þ given in Table 6.

Table 6
L18 experimental design and response table.

Exp. run L18 orthogonal array Responses

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10
2 2 2 2 2
A B C D E F (W/m K) % mm s % (N/mm ) (N/mm ) (N/mm ) (N/mm ) $/mm2

1 1 1 1 1 1 1 14.568 1 560 7 2.05 39.8 63.4 73.44 4.73 45.635


2 1 1 2 2 2 2 14.561 1 610 5 2.17 30.1 58.4 67.66 4.6 49.03
3 1 1 3 3 3 3 14.567 1 710 4 1.79 40.1 66.7 69.85 4.65 52.11
4 1 2 1 1 2 2 14.565 1 600 4 2.03 41.8 62.9 65.53 4.55 46.12
5 1 2 2 2 3 3 14.566 1 740 3 2.22 40.2 68.2 72.19 4.7 49.612
6 1 2 3 3 1 1 14.572 1 690 3 2.14 34.2 58.1 69.69 4.65 51.019
7 1 3 1 2 1 3 14.57 1 660 3 2.58 30.7 59.6 70.4 4.66 48.232
8 1 3 2 3 2 1 14.57 1 730 3 2.73 30.6 58.1 59.51 4.41 51.676
9 1 3 3 1 3 2 14.571 1 730 3 2.46 30.2 57 60.51 4.28 46.895
10 2 1 1 3 3 2 14.559 0 780 3 1.59 27.2 59.4 71.09 4.68 53.61
11 2 1 2 1 1 3 14.559 0 750 4 2.2 37.2 61.5 74.52 4.75 47.536
12 2 1 3 2 2 1 14.562 0 760 4 2.01 33.1 55.5 68.52 4.62 50.721
13 2 2 1 2 3 1 14.562 0 770 4 1.7 28.7 54.5 69.97 4.65 53.638
14 2 2 2 3 1 2 14.561 0 740 3 2.39 30.7 53.4 66.77 4.58 50.227
15 2 2 3 1 2 3 14.561 0 770 3 1.91 31.7 56 74.39 4.75 48.079
16 2 3 1 3 2 3 14.556 0 760 3 3.28 27.1 42.3 64.2 4.52 53.11
17 2 3 2 1 3 1 14.555 0 740 5 2.95 25.2 55 68.26 4.62 48.73
18 2 3 3 2 1 2 14.556 0 750 5 3.46 23.5 48.4 61.64 4.46 50.167
24
Table 7
Calculated S/N ratios, and TOPSIS method application.

Decision matrix (S/N ratios) Weighted normalized decision matrix S⁎i Si− C⁎i

B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32
Response R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 vi1 vi2 vi3 vi4 vi5 vi6 vi7 vi8 vi9 vi10

Weight 0.059b 0.078 0.157 0.098 0.078 0.118 0.078 0.176 0.059 0.098

1 −23.3b 1.938 54.96 −16.4 −6.24 32 36.04 37.32 13.5 −33.2 −0.014b 0.003 0.036 −0.033 −0.016 0.029 0.014 0.019 0.038 −0.018 0.036c 0.013d 0.265e
2 −23.3 3.098 55.71 −14.3 −6.73 29.57 35.33 36.61 13.26 −33.8 −0.014 0.006 0.036 −0.029 −0.017 0.027 0.014 0.018 0.037 −0.018 0.033 0.013 0.278
3 −23.3 3.098 57.03 −12.5 −5.04 32.06 36.48 36.88 13.35 −34.3 −0.014 0.006 0.037 −0.025 −0.013 0.030 0.014 0.019 0.037 −0.019 0.031 0.018 0.368
4 −23.3 0 55.56 −12.3 −6.14 32.42 35.97 36.33 13.16 −33.3 −0.014 0.000 0.036 −0.025 −0.016 0.030 0.014 0.018 0.037 −0.018 0.037 0.015 0.293
5 −23.3 4.437 57.38 −10.4 −6.91 32.08 36.68 37.17 13.44 −33.9 −0.014 0.008 0.037 −0.021 −0.017 0.030 0.014 0.019 0.037 −0.018 0.029 0.018 0.387
6 −23.3 0 56.78 −9.83 −6.62 30.68 35.28 36.86 13.35 −34.2 −0.014 0.000 0.037 −0.020 −0.017 0.028 0.014 0.019 0.037 −0.019 0.036 0.017 0.323
7 −23.3 3.098 56.39 −9.83 −8.22 29.74 35.5 36.95 13.37 −33.7 −0.014 0.006 0.037 −0.020 −0.021 0.027 0.014 0.019 0.037 −0.018 0.032 0.016 0.333
8 −23.3 4.437 57.27 −9.54 −8.73 29.71 35.28 35.49 12.89 −34.3 −0.014 0.008 0.037 −0.019 −0.022 0.027 0.014 0.018 0.036 −0.019 0.030 0.017 0.358
9 −23.3 1.938 57.27 −9.25 −7.81 29.6 35.12 35.64 12.63 −33.4 −0.014 0.003 0.037 −0.019 −0.020 0.027 0.014 0.018 0.035 −0.018 0.034 0.017 0.332
10 −23.3 10.46 57.84 −8.94 −4.03 28.69 35.48 37.04 13.4 −34.6 −0.014 0.019 0.037 −0.018 −0.010 0.026 0.014 0.019 0.037 −0.019 0.017 0.030 0.630
11 −23.3 7.959 57.5 −12.7 −6.87 31.41 35.78 37.45 13.53 −33.5 −0.014 0.014 0.037 −0.025 −0.017 0.029 0.014 0.019 0.038 −0.018 0.024 0.019 0.448
12 −23.3 13.98 57.62 −12.3 −6.05 30.4 34.89 36.72 13.29 −34.1 −0.014 0.025 0.037 −0.025 −0.015 0.028 0.014 0.019 0.037 −0.019 0.014 0.029 0.674
13 −23.3 20 57.73 −10.9 −4.61 29.16 34.73 36.9 13.35 −34.6 −0.014 0.036 0.037 −0.022 −0.012 0.027 0.014 0.019 0.037 −0.019 0.006 0.041 0.876
14 −23.3 13.98 57.38 −10.6 −7.56 29.74 34.55 36.49 13.22 −34 −0.014 0.025 0.037 −0.021 −0.019 0.027 0.014 0.018 0.037 −0.019 0.015 0.029 0.661
15 −23.3 13.98 57.73 −8.63 −5.6 30.02 34.96 37.43 13.53 −33.6 −0.014 0.025 0.037 −0.017 −0.014 0.028 0.014 0.019 0.038 −0.018 0.012 0.032 0.735
16 −23.3 20 57.62 −8.94 −10.3 28.66 32.53 36.15 13.1 −34.5 −0.014 0.036 0.037 −0.018 −0.026 0.026 0.013 0.018 0.037 −0.019 0.016 0.039 0.702
17 −23.3 10.46 57.38 −13.4 −9.38 28.03 34.81 36.68 13.29 −33.8 −0.014 0.019 0.037 −0.027 −0.024 0.026 0.014 0.018 0.037 −0.018 0.024 0.020 0.454
18 −23.3 13.98 57.5 −13.1 −10.8 27.42 33.7 35.8 12.99 −34.0 −0.014 0.025 0.037 −0.026 −0.027 0.025 0.013 0.018 0.036 −0.019 0.023 0.026 0.533
98.7 a 44.0 242.0 48.8 31.1 127.8 149.3 155.6 56.3 144.0 A*= −0.014 0.036 0.037 −0.017 −0.010 0.030 0.014 0.019 0.038 −0.018
A−= −0.014 0.000 0.036 −0.033 −0.027 0.025 0.013 0.018 0.035 −0.019
a
The square root of sum of squares of each element in the columns.
b
From Eqs. (4)–(6) in Section 3: 0.059 ∗ [(−23.3) / (98.7)] = −0.01.
c
From Eq. (9) in Section 3: {[(−0.014) − (−0.014)]2 + [(0.003) − (0.036)]2 + [(0.036) − (0.037)]2 + [(−0.033) − (−0.017)]2 + [(−0.016) − (−0.010)]2 + [(0.029) − (0.030)]2 + [(0.014) − (0.014)]2 + [(0.019) −
(0.019)]2+ [(0.038) − (0.038)]2 + [(−0.018) − (−0.018)]2}1/2 = 0.036.
d
From Eq. (10) in Section 3: {[(−0.014) − (−0.014)]2 + [(0.003) − (0)]2 + [(0.036) − (0.036)]2 + [(−0.033) − (−0.033)]2 + [(−0.016) − (−0.027)]2 + [(0.029) − (0.025)]2 + [(0.014) − (0.013)]2 + [(0.019) − (0.018)]2+
[(0.038) − (0.035)]2 + [(−0.018) − (−0.019)]2}1/2 = 0.013.
e
From Eq. (11) in Section 3: 0.013 / (0.013 + 0.036) = 0.265.
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 25

Main Effects Plot for Means


Data Means
A B C
0.6247
0,6
0.5459
0.5164 0.4941
0,5

0,4 0.4436 0.4520


0.4309
Mean of Means

0.3263
0,3
1 2 1 2 3 1 2 3
D E F
0,6
0.5135 0.5069 0.5066 0.5077 0.4956
0,5 0.4914

0.4544
0,4 0.4211 0.4272

0,3
1 2 3 1 2 3 1 2 3

Fig. 4. Optimum factors' levels and mean plots for factor effects.

5.3. TOPSIS-based Taguchi optimization Finally, Eq. (11) is used to calculate the similarity of the ideal solu-
tions in each scenario, Sc, (Ci⁎). The final results are illustrated in
In this study a Taguchi orthogonal array [46–48] (L18) is selected Table 7, last column.
to record the experiment results. In Table 6, columns 2–6 represent Ci⁎, i = 1, 2, . . ., 18 are the surrogate responses for the proposed
the six control factors and their levels. This model provides the ten multi-response optimization problem. The average responses by
performance measures (responses) simultaneously for solving the factor levels can be determined by using the additive property [2].
multi-response optimization problem (in Table 6, columns 8–17). The Their associated factor effect plots are given in Fig. 4. Since the effect
S/N ratios are calculated by using Eq. (2) for each response. The exper- values are smaller-the-better and larger-the-better types, the nor-
imental design and the S/N ratios are given in Table 7, columns 2–11. malization methods led to the final parameter design of A2 B2 C1 D2
In order to convert the multi-response optimization problem E3F3 (Fig. 4).
into a single response problem, the TOPSIS method [15] is used. In order to predict the anticipated improvement under the select-
In Table 7, columns 2–11 are illustrated as the decision matrix for ed optimum conditions between the estimated mixture levels in the
the first step of the TOPSIS method. The normalized decision matrix parameter design stage, the quality contributions of SNRs for these ten
and then the weighted normalized matrix are determined by using responses were predicted by using an optimal mixture level. The exper-
Eqs. (4) and (5) respectively. The positive ideal solution (A*) and imental results derived from the optimum condition and the significant
the negative ideal solution (A −) could be found by Eqs. (7) and (8). anticipated improvement in the TOPSIS–Taguchi method are given in
Eqs. (9) and (10) are used to determine the separation measures. Table 8.

Table 8
Anticipated improvement in optimum condition.

Responses Description Estimated mixture levels before Optimal mixture levels after Anticipated Anticipated
Taguchi experiments A2B2C1 D2E3F1 Taguchi experiments A2B2C1 D2E3F3a improvement improvement
(dB) (%)

1 Average convection heat transfer coefficient (W/m2K) 14.5624 14.556 0.0064 0.044
2 Air content (%) 0.1 0.1 0.0 0.0
3 Slump flow (cm) 770 790 20 2.6
4 T50 time (second) 3.5 3 0.5 14.28
5 Water absorption (%) 1.70094 1.7004 0.00054 0.032
2
6 Compressive strength (N/mm ) 28.7 42.6 13.6 48.43
2 days
7 Compressive strength (N/mm2) 54.5 67.4 12.9 23.67
7 days
8 Compressive strength (N/mm2) 70 81.2 11.2 16
28 days
2
9 Splitting tensile strength (N/mm ) 4.65 5.09 0.44 9.5
28 days
2
10 Production cost ($/mm ) 53.6381 50.8691 2.769 5.162
a
Verification test results with optimal mixture proportions.
26 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

5.4. Verification of model compounds of both the low and the high molar masses extracted
from the roasted Arabic coffee.
In order to verify the optimum mix-design proportion which is Although RSM is an experimental design modeling tool, it is also
obtained by using the TOPSIS-based Taguchi method, an experimen- an optimization technique. But it is not applicable to the complicated
tal study is performed to check whether the splitting tensile cases such as the non-polynomials, the higher-orders and the
strength, the compressive strength and the slump flow could really multi-response functions [55]. Recently, the desirability function ap-
be maximized and the production cost, the T50 time, the total heat proach (DFA), the non-linear programming (NLP) methodologies, and
transfer coefficient, the air content, and the water absorption could the evolutionary algorithms (EAs) have been preferred to solve the
really be minimized by the proposed optimum mixture proportions multi-response optimization problems because they have some superi-
[3]. orities in solving complex mathematical models [55]. For these reasons,
The results can be seen in Table 8, columns 4–6. The results of the we compare the solution capability of the proposed TOPSIS-based
verification study showed that the proposed results satisfy the Taguchi method to an ideal and controllable size of RSM model. So, in
expected increase to split the tensile strength, the compressive order to test the performance of the TOPSIS-based Taguchi method, a
strength, the slump flow and the expected decrease for the produc- small version of our case study is designed in this section. We apply
tion cost, the T50 time, the total heat transfer coefficient, the air con- the small version of our proposed case study on the RSM-based model
tent, and the water absorption. and compare the obtained results to check the performance of the
TOPSIS-based Taguchi algorithm.
In the application of RSM, a Box–Behnken design is carried out
6. Discussion which consists of at least 54 experiments for only two most impor-
tant HSSCC mixture optimization responses which are the ‘slump
In this section the solution performance of the TOPSIS-based flow’ and the ‘28th day compressive strength’. However, the T50
Taguchi approach is compared with the response surface methodolo- time is between 3 and 6 s in all experiments. The other responses
gy (RSM) to check the accuracy of the proposed approach. Although (R1, R2, R4-R7, R9 and R10) are not evaluated in terms of mathe-
RSM is explained in the following sections, the readers are referred matical computation complexity and the experiments take a lot of
to [6,49–52] for the detailed explanations and the application steps time. The mathematical relationship between the six factors (A, B,
of RSM. The comparison of the TOPSIS-based Taguchi application C, D, E, F) and the two responses (R3, R8) can be approximated
and RSM are discussed for better understanding and they are the by the second order polynomial regression equations.
significant contribution of our study.
RSM is a well known design of experiment (DOE) methodology 6.1. Experiment results and building meta-models
and it has been widely employed in various chemical, biological,
and manufacturing process optimization studies [4,6,53,54] and All 54 experimental runs of the Box–Behnken design are
also chemometrics literature [49,50,52]. For example, Nambiar and performed. As shown in Appendix A — Table A1, the Box–Behnken
Ramamurthy [6] discussed the development of the empirical models design is composed of six independent factors as follows: A (cement
for compressive strength and the density of the foam concrete dosage), B (the ratio of water to binder materials), C (the ratio of
through statistically-designed experiments. They used response sur- the fine aggregate (I) amount to the total aggregate amount), D
face plots to help the visual analysis of the influence of the factors (the ratio of the amount of superplasticizer to the100 kg binder
on the responses. Rahman et al. [53] studied the effect of H2SO4 con- materials), E (fly ash content) and F (mixture time). The model
centration, the reaction temperature and the reaction time for the which is based on RSM is useful for the prediction of ‘slump flow’
production of xylose. They used a rotatable central composite design (R 2(adj) = 90.3) and ‘28th day compressive strength’ (R 2(adj) =
(CCD) in order to fit a second order model, and RSM was utilized to 75.1) responses (see Figs. A1, A2 respectively).
optimize the hydrolysis process in order to obtain high xylose yield. The relations among factors and the regression equations are de-
Habib [54] investigated the development of a comprehensive mathe- termined by the polynomial regression analysis by using MINITAB®
matical model in order to correlate the interactive and the higher Statistical Program Package. The regression equations are as follows
order influences of various electrical discharge machining parameters (in Figs. A1 and A2, the p-values of all significant model terms are
through RSM, by utilizing relevant experimental data as obtained smaller than 0.05):
through experimentation. Brown et al. [49] used RSM in the analysis
of soil samples in order to determine the optimum combination of R3 ¼ 730 þ 12:917A þ 36:25B þ 86:667D
the variables that have direct influence on the atomic absorption 2 2 2
þ 14:583E−66:528B −43:194C −22:361D
signal and to study their interactions as well as their effect on the 2 − þ
total potassium concentration. Ribeiro et al. [50] applied a strategy þ 38:056F −81:25BC−47:5BD þ d1 −d1 ð17Þ
based on the central composite design (CCD) and the principal com-
ponent analysis (PCA) to find the operational conditions (extraction R4 ¼ 64:8583−1:5092B−1:8267C−1:4904D þ 0:6188F
temperature, extraction time, and equilibrium time) of the hyphenat- 2 2þ 2
þ 5:9796D þ 2:2113E 1:4079F þ 1:9887BC þ 1:6963BD: ð18Þ
ed method HS-SPME (headspace-solid phase micro extraction)–
GC-FID 1 which simultaneously optimized the amounts of volatile
6.2. Decision model and optimization

The goal programming (GP) is a multi-objective decision-making


method that can evaluate multiple objectives simultaneously
1
[56–59]. In this study, we used the GP decision making model
Riberio et al. [50, p. 47] states the following: “The analyses were performed
on a G-6850 GC-FID system (Agilent, Wilmington, USA) and they were fitted with
with the decision variables (factor levels) which are formulated as
an HP-5 capillary column (30 m × 0.25 mm × 0.25 μm). Helium (1 mL min−1) follows:
was the carrier gas. The oven temperature was programmed as follows:
40 °C → 5 °C/min → 150 °C → 30 °C/min → 260 °C. The injection port was equipped Objective function:
with a 0.75 mm i.d. liner and the injector was maintained at 220 °C in the splitless
mode. Under these conditions, no sample carry-over was observed on blank runs − −
conducted between extractions.” Min Z ¼ 200; 000  d1 þ 800; 000  d2 ð19Þ
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 27

Fig. 5. GP solution result.

Subject to: 0:35≤x2 ≤0:39 ð23Þ


R3 ¼ 730 þ 12:917x1 þ 36:25x2 þ 86:667x3
2
þ 14:583x4 −66:528x2 −43:194x3 −22:361x4
2 2
0:60≤x3 ≤0:70 ð24Þ
2 − þ
þ 38:056x6 −81:25x2 x3 −47:5x2 x4 þ d1 −d1 ¼ 500 ð20Þ
1:0≤x4 ≤1:5 ð25Þ
R8 ¼ 64:8583−1:5092x2 −1:8267x3 −1:4904x4 þ 0:6188x6
2 2þ 2
þ 5:9796x4 þ 2:2113x5 1:4079x6 80≤x5 ≤120 ð26Þ
− þ
þ 1:9887x2 x3 þ1:6963x2 x4 þ d2 −d2 ¼ 50 ð21Þ
100≤x6 ≤120 ð27Þ
Constraints:
− þ − þ
400≤x1 ≤425 ð22Þ d1 ; d1 ; d2 ; d2 ≥ 0: ð28Þ

Main Effects Plot (data means) for Means


A B C
0.8

0.7 0.6546 0.7447 0.7033


0.6407
0.6155
0.6
0.5757
0.5 0.5667
Mean of Means

0.4 0.3838
1 2 1 2 3 1 2 3
D E F
0.8
0.7730
0.7 0.6620 0.6597 0.6497
0.6243
0.6065
0.6

0.5 0.5455
0.5225
0.4523
0.4

1 2 3 1 2 3 1 2 3

Fig. 6. Main effects plot (data means of Ci⁎) for means.


28 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

Optimal Factor Levels increasing complexity of the calculations and the possibility of the er-
3 roneous judgments [2].

7. Conclusions

2 Because the HSSCC consists of many conflicting factors; it is critical


to use a systematic methodology in order to determine the optimal
mixes and to analyze the most effective factors under a set of con-
straints [3]. For this reason, a multi-response Taguchi method was
1
used in this study to investigate the ranking of the conflicting factor
A B C D E F
levels and the best possible mix proportions of the HSSCC.
RSM TOPSIS-Taguchi This research proposes a hybrid Taguchi method and TOPSIS to
solve the multi-response optimization problem in a ready-mixed
Fig. 7. Comparison of two multi-response optimization methodology's results. concrete plant. Following the Taguchi procedure, the TOPSIS
method was used to transform the multi-response problem into a
(over achievement of a goal is represented by d + and under- single-response problem. An illustrative example illustrated that
achievement of a goal is shown as d −) the difference in the performance between the optimal conditions
and the estimated conditions is significant. The study provides evi-
P R8 > P R3 ðabsolute dominanceÞ: ð29Þ dence for the efficiency of the proposed methodology. The improve-
ments are quite significant. The results showed that the proposed
The GP model has been solved by using MS Excel® Solver Tool methodology is effective in solving the mixture proportions of the
(Fig. 5) and the model obtained the optimal combination as follows: optimization problem.
A1B1C1D1E1F3 (Fig. 6). The anticipated improvement of the quality characteristic between
the estimate and the optimal conditions were calculated as follows;
6.3. Solution comparison the average convection heat transfer coefficient, %0.044, the slump
flow, %2.6, the T50 time, % 14.28, the water absorption, %0.032, the com-
It can be seen in Fig. 7 that in the results of the RSM model pressive strength in 2 days, % 48.43, the compressive strength in 7 days,
(GP solution results), the optimal factor levels are similar to those %23.67, the compressive strength in 28 days, %16.0, the splitting tensile
derived by the result of the TOPSIS-based Taguchi approach strength, %9.5, and the production cost, %5.162.
(see Appendix B). It is concluded that the TOPSIS-based Taguchi Also, it can be seen from the results of the produced concrete
approach could be employed to determine the optimal mixture samples that they satisfied the expected properties of the HSSCC.
proportions of a production process of the ready mixed concrete The outcomes were shared with the Concrete Manufacturer Company
since it is computationally efficient enough to explore all possible and more studies have been initiated to increase the system
combinations among five primary variables. performance.
The TOPSIS based Taguchi approach which employs multi attri- The application of the TOPSIS–Taguchi method in a wider range of
bute decision making (MADM) perspective in the Taguchi method various chemical, manufacturing, and material processes and in the
has been proposed to effectively solve the optimal mixture problem chemometrics discipline which has a large number of variables in
with multiple responses. The TOPSIS can provide a ranking index the chemical systems or in the other possible application areas, re-
(Ci) to represent the complete performance of all responses. The or- mains as a future research scope of this paper.
thogonal arrays and the S/N ratios are the two main components of
the Taguchi method. An orthogonal array is used to reduce the testing
Nomenclature
time and the experimental costs. By using the Taguchi's L18 orthogo-
A heat transfer surface area, [m 2]
nal array, only 18 mixtures are required to estimate the optimal
°C centigrade degree
mixtures.
L Sample length, [m]
The TOPSIS-based Taguchi application is very simple and easy to
hc Convection heat transfer coefficient, [W/m 2∗ K]
perform compared to the RSM. The RSM is a well-known DOE tech- −
h Average convection heat transfer coefficient, [W/m 2∗ K]
nique and it is known to become increasingly difficult for the practi-
K Kelvin
tioners as the number of the variables of the evaluation increases.
k Thermal conductivity [W/m °C]
It could be an unusable tool as the number of the design factors
Q Total heat loss
and/or the responses increase. Conversely, the implementation of
Re Reynolds number
the TOPSIS can easily be extended to include more than two re-
SP Superplasticizer
sponses. Moreover, the calculation steps of the TOPSIS-based Taguchi
Tw Concrete surface temperature,[K]
application can easily be done by using an Excel sheet. The two main
T∞ Ambient temperature, [K]
problems, namely the large time and the cost requirements of the ex-
Tf Film temperature, [K]
periments and the complex mathematical and statistical calculations
u∞ Air velocity, [m/s]
resulting from the RSM, will be encountered in practice.
υ Kinematic viscosity, [m 2/s]
Both conventional RSM and Taguchi methods can only consider a
single response at a time. But, in practice, the presentation of the
quality ought to be considered in various responses, i.e., in the prob-
lem of optimal mixture assessment [2]. In order to solve the optimal Acknowledgments
mixture problems with the multiple responses by RSM [6] and by
the conventional Taguchi method (single response optimization) The authors would like to thank the Concrete Manufacturer
[3], the optimal mixture for each response is frequently assessed indi- Company, the name of which has been kept confidential upon
vidually, and then the overall optimal mixture is determined by the request, for providing the facilities to carry out the experiments at
engineering experience or the cross analysis. This approach still can- their work site. The authors also would like to thank the laboratory
not deal with too many responses at the same time, because of the personnel for their valuable suggestions that improved the paper.
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 29

Appendix A. Response surface methodology based on Box–Behnken design

Table A1
Findings were obtained in all experiments by RSM experiments.

Exp. no. Box–Behnken design (un-coded variables) Responses

Controllable factors R3 R8

A B C D E F

1 400 0.35 0.65 1 100 110 440 77.76


2 450 0.35 0.65 1 100 110 480 79.31
3 400 0.39 0.65 1 100 110 630 67.32
4 450 0.39 0.65 1 100 110 640 69.91
5 400 0.35 0.65 1.5 100 110 710 71.52
6 450 0.35 0.65 1.5 100 110 740 73.44
7 400 0.39 0.65 1.5 100 110 690 68.56
8 450 0.39 0.65 1.5 100 110 730 70.13
9 425 0.35 0.6 1.25 80 110 540 75.83
10 425 0.39 0.6 1.25 80 110 700 71.26
11 425 0.35 0.7 1.25 80 110 640 63.34
12 425 0.39 0.7 1.25 80 110 470 67.78
13 425 0.35 0.6 1.25 120 110 540 70.65
14 425 0.39 0.6 1.25 120 110 740 67.06
15 425 0.35 0.7 1.25 120 110 680 64.56
16 425 0.39 0.7 1.25 120 110 560 67.87
17 425 0.37 0.6 1 100 100 610 75.85
18 425 0.37 0.7 1 100 100 660 69.51
19 425 0.37 0.6 1.5 100 100 770 74.32
20 425 0.37 0.7 1.5 100 100 790 71.23
21 425 0.37 0.6 1 100 120 600 75.9
22 425 0.37 0.7 1 100 120 620 73
23 425 0.37 0.6 1.5 100 120 780 74.68
24 425 0.37 0.7 1.5 100 120 790 71.95
25 400 0.37 0.65 1 80 110 550 76.21
26 450 0.37 0.65 1 80 110 610 76.63
27 400 0.37 0.65 1.5 80 110 770 70.21
28 450 0.37 0.65 1.5 80 110 780 70.71
29 400 0.37 0.65 1 120 110 590 76.87
30 450 0.37 0.65 1 120 110 620 77.97
31 400 0.37 0.65 1.5 120 110 790 71.75
32 450 0.37 0.65 1.5 120 110 790 71.97
33 425 0.35 0.65 1.25 80 100 610 69.94
34 425 0.39 0.65 1.25 80 100 750 67.45
35 425 0.35 0.65 1.25 120 100 650 70.14
36 425 0.39 0.65 1.25 120 100 770 67.76
37 425 0.35 0.65 1.25 80 120 620 70.02
38 425 0.39 0.65 1.25 80 120 730 67.51
39 425 0.35 0.65 1.25 120 120 640 70.32
40 425 0.39 0.65 1.25 120 120 750 68
41 400 0.37 0.6 1.25 100 100 720 68.38
42 450 0.37 0.6 1.25 100 100 740 70.21
43 400 0.37 0.7 1.25 100 100 710 64.52
44 450 0.37 0.7 1.25 100 100 730 65.32
45 400 0.37 0.6 1.25 100 120 690 68.44
46 450 0.37 0.6 1.25 100 120 720 70
47 400 0.37 0.7 1.25 100 120 700 69.56
48 450 0.37 0.7 1.25 100 120 720 70.1
49 425 0.37 0.65 1.25 100 110 730 64.5
50 425 0.37 0.65 1.25 100 110 730 64.8
51 425 0.37 0.65 1.25 100 110 720 65
52 425 0.37 0.65 1.25 100 110 730 64.75
53 425 0.37 0.65 1.25 100 110 740 65.1
54 425 0.37 0.65 1.25 100 110 730 65
30 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

Fig. A1. Response surface regression: R3 versus A; B; C; D; E; F. Fig. A2. Response surface regression: R8 versus A; B; C; D; E; F.
B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32 31

Appendix B. TOPSIS–Taguchi application results

See Appendix Table B1.

Table B1
Transformation of multi-response problem to single response problem by using TOPSIS method.

Taguchi's L18 orthogonal array Decision matrix Weighted normalized Calculation of Calculation of
(S/N ratios) decision matrix separation similarities to
measures ideal solution

Exp. no. A B C D E F R3 R8 vij3 vij8 Si ⁎ Si− Ci⁎

Weight

0.2 0.8

1 1 1 1 1 1 1 54.96376 37.31865 0.045 0.192 0.002 0.009 0.792


2 1 1 2 2 2 2 55.7066 36.60664 0.046 0.188 0.005 0.006 0.553
3 1 1 3 3 3 3 57.02517 36.88333 0.047 0.190 0.003 0.007 0.712
4 1 2 1 1 2 2 55.56303 36.3288 0.046 0.187 0.006 0.004 0.418
5 1 2 2 2 3 3 57.38463 37.16954 0.047 0.191 0.001 0.009 0.858
6 1 2 3 3 1 1 56.77698 36.86341 0.047 0.190 0.003 0.007 0.698
7 1 3 1 2 1 3 56.39088 36.95145 0.047 0.190 0.003 0.008 0.730
8 1 3 2 3 2 1 57.26646 35.4918 0.047 0.183 0.010 0.002 0.159
9 1 3 3 1 3 2 57.26646 35.63654 0.047 0.183 0.009 0.002 0.180
10 2 1 1 3 3 2 57.84189 37.03617 0.048 0.190 0.002 0.008 0.798
11 2 1 2 1 1 3 57.50123 37.44546 0.048 0.193 0.000 0.010 0.973
12 2 1 3 2 2 1 57.61627 36.71635 0.048 0.189 0.004 0.007 0.640
13 2 2 1 2 3 1 57.72981 36.89824 0.048 0.190 0.003 0.008 0.729
14 2 2 2 3 1 2 57.38463 36.49163 0.047 0.188 0.005 0.006 0.529
15 2 2 3 1 2 3 57.72981 37.43029 0.048 0.192 0.000 0.010 0.988
16 2 3 1 3 2 3 57.61627 36.1507 0.048 0.186 0.007 0.004 0.377
17 2 3 2 1 3 1 57.38463 36.68333 0.047 0.189 0.004 0.006 0.621
18 2 3 3 2 1 2 57.50123 35.79725 0.048 0.184 0.008 0.003 0.236
A* 0.047802 0.193
A− 0.045423 0.183

References [15] T. Yang, P. Chou, Solving a multi-response simulation–optimization problem


with discrete variables using a multiple-attribute decision-making method,
[1] S. Yan, H.-C. Lin, Y.C. Liu, Optimal schedule adjustments for supplying ready Mathematics and Computers in Simulation 68 (1) (2005) 9–21.
mixed concrete following incidents, Automation in Construction 20 (2011) [16] Y. Kuo, T. Yang, G.W. Huang, The use of a Grey-based Taguchi method for optimiz-
1041–1050. ing multi-response simulation problems, Engineering Optimization 40 (6) (2008)
[2] C.Y. Chang, R. Huang, P.C. Lee, T.L. Weng, Application of a weighted Grey–Taguchi 517–528.
method for optimizing recycled aggregate concrete mixtures, Cement and [17] T.-L. Su, H.-W. Chen, C.-F. Lu, Systematic optimization for the evaluation of the
Concrete Composites 33 (2011) 1038–1049. microinjection molding parameters of light guide plate with TOPSIS-based
[3] E. Ozbay, A. Oztas, A. Baykasoglu, H. Ozbebek, Investigating mix proportions of Taguchi method, Advances in Polymer Technology 29 (1) (2010) 54–63.
high strength self compacting concrete by using Taguchi method, Construction [18] G.-B. Hong, T.-L. Su, Statistical analysis of experimental parameters in character-
and Building Materials 23 (2009) 694–702. ization of ultraviolet-resistant polyester fiber using a TOPSIS–Taguchi method,
[4] F. Bayramov, C. Taşdemir, M.A. Taşdemir, Optimisation of steel fibre reinforced Iranian Polymer Journal 21 (12) (2012) 877–885.
concretes by means of statistical response surface method, Cement and Concrete [19] T.-S. Lan, Taguchi optimization of multi objective CNC machining using TOPSIS,
Composites 26 (2004) 665–675. Information Technology Journal 8 (6) (2009) 917–922.
[5] L. Eroğlu, M. Şahmaran, İ.E. Yaman, M. Toktay, Determine the effect of mixture [20] H.C. Liao, Using PCR–TOPSIS to optimize Taguchi's multi response problem, Inter-
parameters on fresh properties of HSSCC, 2. Chemical Additive Symposium in national Journal of Advanced Manufacturing Technology 22 (2003) 649–655.
Structures, Ankara, Turkey, 2007, (in Turkish). [21] J.-C. Lu, T. Yang, C.-T. Su, Analysing optimum push/pull junction point
[6] E.K.K. Nambiar, K. Ramamurthy, Models relating mixture composition to the location using multiple criteria decision-making for multistage stochastic pro-
density and strength of foam concrete using response surface methodology, duction system, International Journal of Production Research 50 (19) (2012)
Cement and Concrete Composites 28 (2006) 752–760. 5523–5537.
[7] M. Olivia, H. Nikraz, Properties of fly ash geopolymer concrete designed by [22] P. Mandal, Signal-to-noise ratio: a fundamental and broad process performance
Taguchi method, Materials and Design 36 (2012) 191–198. measure, Journal of Engineering Design 23 (12) (2012) 927–944.
[8] S. Hinislioglu, O.U. Bayrak, Optimization of early flexural strength of pavement [23] Turkish Standard TS EN 206-1, Concrete — part 1: specification, performance,
concrete with silica fume and fly ash by the Taguchi method, Civil Engineering production and conformity, Turkish Standardization Institute, Ankara, 2002.
and Environmental System 1 (2) (2004) 79–90. (in Turkish).
[9] M. Muthukumar, D. Mohan, Optimization of mechanical properties of polymer [24] TS EN 12390/8, Testing fresh concrete — part 5, flow table test, Turkish Standard-
concrete and mix design recommendation based on design of experiments, ization Institute, Ankara, 2010. (in Turkish).
Journal of Applied Polymer Science 94 (2004) 1107–1117. [25] S. Nagataki, H. Fujiwara, Self-compacting property of highly-flowable concrete,
[10] O. Tan, A.S. Zaimoglu, S. Hinislioglu, S. Altun, Taguchi approach for optimization 154. American Concrete Institute Special Publication (1995).
of the bleeding on cement-based grouts, Tunnelling and Underground Space [26] C. Kılınc, Effect of chemical admixture content and fresh concrete temperature
Technology 1 (2005) 167–173. on of HSSCC properties, Master Thesis, Istanbul Technical University Institute of
[11] I. Turkmen, R. Gul, C. Celik, R. Demirboga, Determination by Taguchi method Science, Istanbul, 2007 [in Turkish].
of optimum conditions for mechanical properties of high strength concrete [27] P.J.M. Bartos, Testing — SCC: towards new European standards for fresh SCC, First
with admixtures of silica fume and blast furnace slag, Civil Engineering and Envi- International Symposium on Design, Performance and Use of Self Consolidating
ronmental System 20 (2) (2003) 105–118. Concrete SCC, Changsha, Hunan, China, 2005.
[12] K.T. Fauzan, M. Hosino, A. Morita, The influence of mixing techniques on the [28] C.A. Arbeláez Jaramilo, J.W. Rigueira Victor, J.R.M. Vargas, P. Serna Ros, M.P.
properties of concrete by using air entraining agent and high range water reducer Barbosa, Reduced models test for the characterization of the rheologic properties
agent, Jurnal Itenas 7 (3) (2003) 1–10. of self compacting concrete, 3rd International Symposium on Self-compacting
[13] M.J. Simon, E.S. Lagergren, K.A. Snyder, Concrete mixture optimization using Concrete, Reykjavik, Iceland, 2003.
statistical mixture design methods, Proceedings of the PCI/FHWA International [29] TS EN 12390/3, Testing Hardened Concrete — Part 3, Compressive Strength of Test
Symposium on High Performance Concrete, 1997, pp. 230–244. Specimens, Turkish Standardization Institute, Ankara, 2010. (in Turkish).
[14] M.C.S. Ribeiro, C.M.L. Tavares, M. Fieiredo, A.J.M. Ferreira, A.A. Fernandez, Bending [30] TS 3129, Determination of Splitting Tensile Strength of Concrete Specimens,
characteristics of resin concretes, Materials Research 6 (2) (2003) 247–254. Turkish Standardization Institute, Ankara, 2002. (in Turkish).
32 B. Şimşek et al. / Chemometrics and Intelligent Laboratory Systems 125 (2013) 18–32

[31] O. Akalin, K.U. Akay, B. Sennaroglu, M. Tez, Optimization of chemical admixture for [46] M. Oudjene, L. Ben-Ayed, On the parametrical study of clinch joining of metallic
concrete on mortar performance tests using mixture experiments, Chemometrics sheets using the Taguchi method, Engineering Structures 30 (2008) 1782–1788.
and Intelligent Laboratory Systems 104 (2010) 233–242. [47] S.M. Phadke, Quality Engineering Using Robust Design, Prentice Hall, New Jersey,
[32] M.S. Al-Homoud, Performance characteristics and practical applications of common 1989.
building thermal insulation materials, Building and Environment 40 (2005) 353–366. [48] G.J. Besseris, Multi-response multi-factorial master ranking in non-linear replicated-
[33] O.G. Santin, Behavioural patterns and user profiles related to energy consumption saturated DOE for qualimetrics, Chemometr Intell Lab 116 (2012) 47–56.
for heating, Energy and Buildings 43 (2011) 2662–2672. [49] J.H. Brown, M.J. Gomez, Z. Benzo, J.E. Vaz, Application of the response surface
[34] T. Uygunoğlu, A. Keçebaş, LCC analysis for energy-saving in residential buildings methodology for potassium determination in soils by AAS using the slurry tech-
with different types of construction masonry blocks, Energy and Buildings 43 nique, Chemometrics and Intelligent Laboratory Systems 35 (1996) 239–247.
(2011) 2077–2085. [50] J.S. Ribeiro, R.F. Teófilo, F. Augusto, M.M.C. Ferreira, Simultaneous optimization of
[35] Y. Lee, M.S. Choi, S.T. Yi, J.K. Kim, Experimental study on the convective heat the microextraction of coffee volatiles using response surface methodology and
transfer coefficient of early-age concrete, Cement and Concrete Composites 31 principal component analysis, Chemometrics and Intelligent Laboratory Systems
(2009) 60–71. 102 (2010) 45–52.
[36] TS EN 123390/7, Test Method for Determination the Specific Gravity the Absorption [51] R.H. Myers, D.C. Montgomery, Response Surface Methodology: Process and
Water and the Void Ratio in Hardened Concrete, Turkish Standardization Institute, Product Optimization Using Designed Experiment, Second ed. Wiley, New York,
Ankara, 2010. (in Turkish). 2002.
[37] H.A. Mohammed, H.Y. Salman, Free and forced convection heat transfer in the [52] T. Lundstedt, E. Seifert, L. Abramo, B. Thelin, A. Nystrom, J. Pettersen, R. Bergman,
thermal entry region for laminar flow inside a circular cylinder horizontally Experimental design and optimization, Chemometrics and Intelligent Laboratory
oriented, Energy Conversion and Management 48 (2007) 2185–2195. Systems 42 (1998) 3–40.
[38] E. Rastrup, Heat of hydration in concrete, Magazine of Concrete Research 6 (17) [53] S.H.A. Rahman, J.P. Choudhury, A.L. Ahmad, A.H. Kamaruddin, Optimization stud-
(1954) 79–92. ies on acid hydrolysis of oil palm empty fruit bunch fiber for production of xylose,
[39] C. Hsieh, C. Qin, E. Ryder, Development of computer modeling for prediction of Bioresource Technology 98 (3) (2007) 554–559.
temperature distribution inside concrete pavements, Report FL/DOT/SO/90-374, [54] S.S. Habib, Study of the parameters in electrical discharge machining through
Mechanical Engineering Department, University of Florida, Gainesville, 1989. 32–59. response surface methodology approach, Applied Mathematical Modelling 33
[40] A. Chapman, Fundamental of Heat Transfer, Macmillian Inc., New York, 1982. (2009) 4397–4407.
[41] Yamagawa, Kasagi, Gobayashi, Investigation of an experimental method for [55] S. Fallah-Jamshidi, M. Amiri, Synergy of ICA and MCDM for multi-response optimisa-
thermal convective coefficient of massive concrete, Proceedings. Japan Society tion problems, International Journal of Production Research 51 (3) (2013) 652–666.
of Civil Engineers 37 (5) (1982) 27–64, [in Japanese]. [56] H.A. Taha, Operations Research, Prentice-Hall, Inc., N.J., 1997
[42] Sikoku Electric Power Co, A Study on the Early Age Cracking of Massive Concrete [57] J. Lee, S.-H. Kang, J. Rosenberger, S.B. Kim, A hybrid approach of goal programming
and Its Prevention Measure, 1964. 36–70 , (in Japanese). for weapon systems selection, Computers and Industrial Engineering 58 (2010)
[43] J.P. Holman, Heat Transfer, 10th ed. McGraw Hill Company, NY, 2010. 521–527.
[44] T. Akram, A.S. Memon, H. Obaid, Production of low cost self compacting concrete [58] M. Yurdakul, Selection of computer-integrated manufacturing technologies using
using bagasse ash, Construction and Building Materials 23 (2009) 703–712. a combined analytic hierarchy process and goal programming model, Robotics
[45] G. Heirman, L. Vandewalle, D. Van Gemert, V. Boel, K. Audenaert, G. De Schutter, and Computer-Integrated Manufacturing 20 (2004) 329–340.
B. Desmet, J. Vantomme, Time-dependent deformations of limestone powder [59] Y.T. İç, F. Elaldi, F. Pakdil, N.E. İpek, Design of experiment and goal programming
type self-compacting concrete, Engineering Structures 30 (2008) 2945–2956. application for GMAW process, Welding Journal 91 (4) (2012) 106–112.

You might also like