Measurement 46 (2013) 2807–2815

Contents lists available at SciVerse ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Optimizing convexity defect in a tile industry using fuzzy goal

programming
Abbas Al-Refaie a,⇑, Ali Diabat b
a
Department of Industrial Engineering, The University of Jordan, Amman 11942, Jordan
b
Engineering Systems & Management, Masdar Institute of Science & Technology, United Arab Emirates

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

Article history: In most designed experiments, the main focus is to find the factor settings that optimize a
Received 9 July 2012 quality response regardless of engineer’s preferences about factor settings. Further, in tiles
Received in revised form 4 March 2013 industry convexity defects result in huge quality costs as well as production losses. This
Accepted 18 March 2013
research, therefore, aims at optimizing convexity defect while considering process engi-
Available online 9 April 2013
neers’ preferences using fuzzy goal programming (FGP). Three two-level key process fac-
tors are considered, including below-rollers temperature, above-rollers temperature,
Keywords:
direct blow air. Experiments are conducted with two repetitions; in each the convexity
Fuzzy goal programming
Taguchi method
is measured on four tiles. Two optimization techniques are employed to determine the
Convexity defect combination of optimal factor settings, including the Taguchi method and latter technique.
ANOVA The Taguchi approach and FGP approach provide relative improvements of 61.2% and
41.2%, respectively. Although the former technique reduces convexity larger than latter
approach, it failed to satisfy the preferences on the settings of process factors. In contrast,
the optimal factor settings obtained using FGP completely satisfy engineers’ preferences. In
conclusion, FGP successfully optimizes process performance and completely satisfies pro-
cess engineers’ preferences in tiles industry.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction and extended for application in manufacturing environ-

Ceramic tiles are produced by pressing clay and other
ingredients into shape then firing it at high temperatures.
It may then be glazed, or left unglazed depending on its in-
tended use [7]. In ceramic tiles industry, defects result in
huge quality costs as well as production losses. Hayajneh
et al. [5] used a subtractive clustering fuzzy identification
method and a Sugeno-type fuzzy inference system to mon-
itor tile defects in tile manufacturing process. Xianghua [10]
implemented texture analysis techniques for the detection
of abnormalities in color texture surfaces of ceramic tile
on which patterns are regularly of a random nature. Farooq
et al. [4] developed a new methodology to detect defects in
ceramic tile. The concept of photometric stereo is adopted
⇑ Corresponding author.
E-mail addresses: Abbas.alrefai@ju.edu.jo (A. Al-Refaie), adiabat@
masdar.ac.ae (A. Diabat).
ments. High speed inspection of ceramic tiles is utilized
for the analysis of surfaces at production line rates.
Designed experiments have been found effective in
optimizing process performance [8]. In these techniques,
the process factors are assigned specific levels; for exam-
ple, high and low. Then, the optimal settings will be
restricted at each of these levels. The Taguchi method [9]
has been found only effective in optimizing a single re-
sponse [2,3,6]. In some cases, process engineers prefer set-
ting the levels of process factors within specified ranges
depending on setup time and operation cost constraints.
As well as, they may have preferences on the optimal value
of a response. This in return transforms a single-objective
optimization problem to multi- objectives problem. The
goal programming (GP) model is an optimization tech-
nique enables the decision maker to take multiple objec-
tives simultaneously into account. In general, GP models
consider the model parameters and objective values as

0263-2241/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.measurement.2013.03.013
2808 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815
precise and certain. However, in most real applications
these values are not known precisely. Under such a cir-
cumstance, fuzzy GP (FGP) models can be employed
[1,11,12]. Therefore, this research aims at improving the
performance of a tiles manufacturing process by reducing
convexity defect using the Taguchi method and FGP
approach. The remainder of this paper is outlined in the
following sequence. Section 2 maps tiles manufacturing
processes and explores tiles defects. Section 3 proposes
FGP optimization model for convexity reduction. Section 4
conducts experimental design. Section 5 summarizes
research results. Finally, conclusions are presented in
Section 6.
2. Tiles process mapping and main defects
2.1. Tiles process mapping
The production process of ceramic tiles is illustrated
briefly in Fig. 1.
The production process starts from batching deter-
mined amounts of raw materials. The raw materials are
tabulated with their chemical compounds in Table 1.
Then, the mixture is fed inside large mills of about
50 tons weight, where the milling operation takes place,
where ceramic stones are placed in. After milling, the mix-
ture looks like a milky slip called slip, stored in large tanks
underground. The slip is pumped up an atomizer for the
spray drying operation to produce a very fine powder with
humidity between 5.5% and 6%. Now, the fine powder is fed
into for pressing, where the tiles take their initial shape.
Tiles are fed into a dryer for increasing their strength by
loosing water content from their body, where the humidity
after drying is almost 0%. During tiles production, painting,
decorating and surface treatments are carried out on the
upper surface of the tiles. At the end of the production line,
there exists a short dryer. This dryer causes the tempera-
ture of the tiles bodies to rise up. The purpose of rising
up the tiles temperature is to avoid thermal shocks that
may occur during the firing process in the kiln due to high
water content. The final stage of tiles production is the fir-
ing process, in which physical, chemical and structural
changes take place. Due to the lack of proper techniques
for monitoring, controlling, and correcting the kiln some
defects may be produced.
2.2. Ceramic tiles defects
In the production process of tiles, the most common
defects are the planarity defects that occur in the firing
Fig. 1. Production process of ceramic tiles.
Table 1
Ceramics tiles raw material with their chemical compound.
Raw material Chemical compound
Kaolin Al2O32SiO22H2O
Silica SiO2
Feldspar K2OAl2O36SiO2 + Na2OAl2O36SiO2
Calcium carbonate (CaCO3) CaCO3
process; especially the convexity defects of the fired tiles.
To experimentally investigate the convexity defect of fired
ceramic tiles, a factory specialized in the production of
wall and floor ceramic tiles is selected, where the average
production rate is 900,000 m2 of tiles/year and the aver-
age of defective tiles is 8% from the annual production,
40% of the average of defective tiles result from the firing
process. The types of defects result during the firing pro-
cess are planarity defects, holes (pin holes, holes, or bub-
bles in the glaze), crazing, cooling breakage, breakage
during preheating, black core, and tile bursts in the pre-
kiln. Fig. 2 illustrates the percentages of these defects, in
which the planarity defects are the most common defects
that occur during the firing process. Fig. 3 illustrates the
percentages of the various kinds of the planarity defects,
where it is clear that the company is facing a real problem
in the form of the tile’s convexity defect (50% of the total
number planarity defects). Fig. 4 depicts a tile with
convexity defect. Two hundred samples are randomly se-
lected and tile convexity is measured then shown in Fig. 5.
It is noted that there are many points exceeding the upper
limit value for convexity of 10 mm; the threshold for
acceptable tile. The annual average production of final
fired ceramic tiles that contain convexity defect is about
27,000 m2 of tiles/year. This large quantity of tiles defects,
increases the wastage and decreases the company finan-
cial returns.
3. Proposed fuzzy satisfaction model
This research adopts a weighted additive GP model for
optimizing convexity as follows:
Step 1: let y represents the value of tile convexity defect
and xj is the value of the jth process variable. Formulate
the multiple linear regression function, f(x1, x2, . . ., xj),
between the y and process controllable factors using
regression model; that is,
y ¼ f ðx1 ; x2 ; . . . ; xj Þ; ð1Þ
 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815 2809

Step 2: for convexity, which is the smaller-the-better

(STB) type response, the process engineer aims at
reducing the response value to a value below the upper
acceptable limit, gy, which is the imprecise fuzzy value
(aspiration level). This corresponds to region I in Fig. 6,
in which the process engineer is completely satisfied at
any point with membership function, lys ; equals one.
Let dþ
ys represents the positive deviation of y from gy,
which falls in the range [0,Dþy ].
Let the Dþ y is the maximal positive admissible violation
from gy. Then, when dþ y is greater than zero; i.e., the y falls
in region II, the ly decreases linearly to zero and conse-
Fig. 2. The defect percentages during the firing process. quently the process engineer satisfaction decreases. Final-
ly, when the y falls above (g y þ Dþ y ) the solution is
unacceptable. Consequently, the ly is utilized for describ-
ing STB response and is represented as:
8
>
> 1; 0 6 y 6 g y
< yg y þ
ly ¼ 1  Dþy ; g y < y 6 g y þ Dy ð2Þ
>
>
:
0; y > g y þ Dþy

The corresponding goal constraints of ys can be ex-

pressed as:
y  dþy 6 g y ð3Þ

dþy
Fig. 3. The percentages of various kinds of the planarity defects. ly þ ¼1 ð4Þ
Dþy

0 6 dþy 6 Dþy ð5Þ

Step 3: since the process engineer prefers setting a pro-

cess factor level within a continuous interval, then the
trapezoidal membership function (MSF), lxj ; shown in
Fig. 7, is the most appropriate, which is defined as:
8
>
> 0; xj < g lxj  Dxj
>
>
>
> g lx xj
>
> 1  Dj  ; g lxj  Dxj 6 xj < g lxj
>
>
>
< xj

lxj ¼ 1; g lxj 6 xj 6 g uxj ð6Þ

>
>
>
> xj g ux
>
> 1  Dþ j ; g uxj < xj 6 g uxj þ Dþxj
>
>
>
> xj
>
:
Fig. 4. A tile with convexity defect. 0; xj > g uxj þ Dþxj

Fig. 5. Chart representing convexity sample for current settings.

2810 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815

! !
dþy XJ dxj dþxj
Z ¼ wy þ wxj  þ þ ð12Þ
Dþy j¼1
Dxj Dxj

where the first term denotes the weighted dþ y of the con-

vexity defect, the second represents the sum of the
weighted d þ
xj and dxj of the J process variables.
Step 5: build a linear programming (LP) model for
Minimize Z ð13Þ

subject to

y ¼ f ðx1 ; x2 ; ::::; xj Þ;
Fig. 6. The MSF for convexity defect.

y  dþy 6 g y ;
0 6 dþy 6 Dþy ;

dþy
ly þ ¼ 1;
Dþy

xj þ dxj P g lxj ; j ¼ 1; :::; J

xj  dþxj 6 g uxj ; j ¼ 1; :::; J

dxj dþxj
Fig. 7. The membership functions for the jth process variable. lxj þ þ ¼ 1; j ¼ 1; :::; J
Dxj Dþxj

0 6 dxj 6 Dxj ; j ¼ 1; :::; J

where g lxj and g uxj are the lower and upper limit of xj,
respectively. The D þ
xj and Dxj are the maximal negative
and positive admissible violations from g lxj and g uxj , 0 6 dþxj 6 Dþxj ; j ¼ 1; :::; J
respectively.
Then, the goal constraints of xj will be formulated as. y; ly P 0;
xj þ dxj P g lxj ð7Þ
xj ; lxj P 0; j ¼ 1; :::; J

xj  dþxj 6 g uxj ð8Þ Solve the LP model and find the values of the decision
variables xj and y, and Z.
dxj dþxj
lxj þ  þ ¼1 ð9Þ 4. Experimental design
Dx j Dþxj
4.1. Selection of factors affecting convexity defect
0 6 dxj 6 Dxj ð10Þ
In practice, several factors thought effecting convexity
0 6 dþxj 6 Dþxj ð11Þ defects. Technical knowledge about the factors affecting
convexity deformation suggests that the most important
where d þ
xj and dxj represent the negative and positive devi- factors affecting the convexity defect are listed in Table 2.
ations from g xj and g uxj , respectively. As shown in Fig. 7, re-
l
The factors’ working ranges are selected appropriately so
gion I (III) indicates that xj is smaller (larger) than g lxj (g uxj ), that the firing process functions without kiln failure. To en-
the lxj decreases. However, when the xj falls between its sure that optimization of the firing process has higher de-
two limits (region II), both deviations have zero values, gree of validity, a wide range of experimental region
and consequently, the lxj is at its maximal level. determined by the levels of the three selected factors is
used. The list of factor working ranges is also mentioned
Step 4: formulate the objective function, Z, of the model in Table 2.
as a weighted additive GP model to minimize the sum For the ceramic tiles manufacturers, the cost of experi-
of the weighted deviations of convexity defect and pro- ments and the time needed to complete them is extraordi-
cess controllable factors. These weights are preferably narily high. Thus, each factor is assigned at two levels.
be normalized such that their sum is equal to one. Hence, a full factorial experiment requires a total of
Accordingly, the Z is to minimize 8 = (23) experimental runs. Two replicates for each treat-
 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815 2811

Table 2
Controllable process factors and their ranges.

Factor A: below-rollers temperature in the initial section of the firing zone. Range: 960–980 °C

Factor B: above-rollers temperature in the final section of the firing zone. Range: 1120–1130 °C

Factor C: direct air blowing below rollers at the start of the rapid cooling zone, i.e. the temperature gradient. Range: 50–75 mbar

ment combination are conducted. The experimental design 4.3. Analysis of designed experiments
is shown in Table 3.
Two methods will be used for analyzing the experimen-
tal data; the Taguchi method and FGP approach.
4.2. Conducting designed experiments

4.3.1. Using the Taguchi method

To make sure that the experiments are conducted
according to scientific rules, the following conditions are
considered:
1. Randomness: experiments are run in a random order to
ensure unbiased and more accurate conclusions.
2. Frequency: every treatment combination was conducted
two times so as effect of uncontrollable factors shows
up on the experimental results and an estimate of the
experimental error could be obtained.
3. Duration: duration time for each experiment is 2 h.
4. The production rate: the production rate is constant dur-
ing the run of the experiments (130 m2 of tiles/hour).
5. The instrument used: The instrument used to measure
the tiles convexity is shown in Fig. 8. This gauge is
divided into ten millimeters (i.e. one full round of the
indicator measures 10 mm), and each millimeter is
divided into ten divisions. In other words, the smallest
division of this gauge is 0.1 mm.
Experiments are conducted according to the design
layout shown in Table 3. To experimentally measure tile
convexity, the tile is placed on a reference surface shown
in Fig. 4. The gauge is placed between the lower surface
of the reference and the upper surface of the tile, and then
the thickness of the reference table and the thickness of the
tile are subtracted from the reading taken to measure the
convexity value. Four tiles are selected at each experimen-
tal run taken randomly. Each experiment is repeated twice.
The experimental convexity values are also listed in
1i and y
2i denote the average of the four obser-
Table 3. Let y
vation of convexity for experimental run i; i = 1, . . ., 16, for
the first and second replicate, respectively. Let y i denotes
the average of convexity averages for each experimental
run i.
The Taguchi method is widely applied approach for
optimizing product/process performance to improve prod-
ucts quality and productivity at a relatively low cost. The
three two-level process factors with their corresponding
physical level values are displayed in Table 4.
Taguchi method uses signal-to-noise (S/N) ratio as a
performance measure. For tile convexity, which the smal-
ler-the-better, response, the S/N ratio, gi, for each experi-
mental run i is calculated using the following equation:
X 
gi ¼ 10 log y2ij =2 ; i ¼ 1; . . . ; 16 ð14Þ
Table 3 also displays the obtained values of S/N ratio for
all eight experiments. The average S/N ratios for all factor
levels are shown in Table 5.
From Table 5, the combination of factor levels that opti-
mize convexity is A1B1C1.
4.3.2. Applying the FGP approach
The proposed FGP approach is utilized for optimizing
tiles convexity in the following steps:
Step 1: let x1, x2, x3 denotes process factors A, B, and C,
respectively. Let xixj and xixjxk denotes the two-way and
three-way interaction terms, respectively. The linear
multiple regression (R2adjusted = 94.66%) between
convexity and the coded three process factors is esti-
mated as:
^ ¼ 0:4419 þ 0:0234x1 þ 0:0972x2 þ 0:0344x3
y
þ 0:0469x1 x2  0:0378x1 x3 þ 0:0109x2 x3
 0:0231x1 x2 x3
2812 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815

Table 3 The MSFs for x2 and x3 are constructed in the same

Experimental convexity results. manner.
Experiment Factor setting y1i y2i Sum of yi S/N
i
A B C
averages ratio gi Step 4: the relative importance of the deviations of
(°C) (°C) (mbar) responses, wy and process variables, wxj, are assigned
1.9 960 1120 50 0.3250 0.335 0.660 9.629
weights of 0.7 and 0.1, respectively. Thus, the objective
5.13 980 1120 50 0.2750 0.350 0.625 10.041 function, Z, is expressed as
3.11 960 1130 50 0.3250 0.400 0.725 8.768
7.15 980 1130 50 0.6250 0.625 1.250 4.082 X
3
2.10 960 1120 75 0.3875 0.425 0.812 7.432 Z ¼ 0:7=2  dþy þ 0:1=2  ðdxj þ dþxj Þ
6.14 980 1120 75 0.3500 0.310 0.660 9.614 j¼1
4.12 960 1130 75 0.5750 0.575 1.150 4.807
8.16 980 1130 75 0.5875 0.600 1.187 4.524
Step 5: utilizing steps 1–4, the complete model is using
the coded process factor levels then formulated as
X
3
Minimize 1:4  dþy þ 0:05  ðdxj þ dþxj Þ
j¼1

subject to

y ¼ 0:44188 þ 0:02344x1 þ 0:09719x2 þ 0:03437x3

þ 0:04688x1 x2  0:03781x1 x3 þ 0:01094x2 x3
 0:02312x1 x2 x3

y  dþy  0:5;

ly þ 2dþy ¼ 1;

0  dþy  1;

x1 þ dx1  1;

x1  dþx1  1;

Fig. 8. The instrument used to measure the convexity.

x2 þ dx2  1;

Step 2: the MSF for convexity is shown in Fig. 9. It is

x2  dþx2  1;
noted that process engineers are accepting convexity
up to 0.5. However, their satisfaction decreases linearly
x3 þ dx3  1;
till 1.0. After, the tile of convexity more than 1.0 will be
rejected. The values of gy and D y are 0.5 and 0.5, respec-
tively. Then, the ly is expressed as: x3  dþx3  1;
8
< 1; 0 6 y 6 0:5
> lx1 þ 10  dx1 þ 10  dþx1 ¼ 1;
ly ¼ 1  y0:5
0:5
; 0:5 < y 6 1
>
: lx2 þ 5  dx2 þ 5  dþx2 ¼ 1;
0; y > 1

lx3 þ 12:5  dx3 þ 12:5  dþx3 ¼ 1;

Step 3: the trapezoidal MSFs are utilized to describe the
preferences on three process variables. The values of D 0  dþ;
x1  1;
xj
and Dþxj are set equal. The MSFs for x1, x2, and x3 are dis-
played in Fig. 10. For example, the MSF, lx1 , of x1 is
0  dþ;
x2  0:2;
defined as
8 0  dþ;
> 0; x1 < 960 x3  0:08;
>
>
>
>
< x1  960; 960 6 x1 < 962
>
y; ly  0;
lx1 ¼ 1; 962 6 x1 6 978
>
>
>
> 980  x1 ; 978 < x1 6 980
>
>
: xj ; lxj  0; j ¼ 1; :::; 3:
0; x1 > 980
 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815
Table 4
The physical values for factor levels.
Factor/level Level 1
Above roller temperature 960 °C
Below roller temperature 1120 °C
Air blowing 50 mbar
Table 5
The average S/N ratios for all factor levels.
Factor/level Level 1 (1)
Above roller temperature 8.130
Below roller temperature 9.179
Air blowing 7.659
Fig. 9. The MSF for tile convexity.
Solving the above model, the x1 (above roller tempera-
ture), x2 (below roller temperature), and x3 (direct air
blowing) are found equal 973.11 °C, 1126.96 °C,
66.03 mbar, respectively. Utilizing the above values, the
convexity (y) is then calculated using the multiple regres-
sions formula and found equals to 0.50.
4.3.3. Analysis of variance
The analysis of variance (ANOVA) is used to determine
significant factors effects. ANOVA results are shown in
Table 6. The detailed test of factor effects is shown in
Table 7.
From Table 6, the main effect terms, two-way interac-
tions and three-way interactions (p value < 0.05) are signif-
icant. However, at a level of significance (a = 0.01) the
three-way interactions is insignificant (p value = 0.015).
From Table 7 only the effects of factors x2, x3, x1x2, and
x1x3 are significant effects on convexity at a value of
0.01. It is noted also that factor x2 has the largest effect
on convexity of 0.19437. Consequently, more control
should be established on this factor.
Residuals are just the difference between the observa-
tions and the corresponding fitted response. Residuals are
_
also plotted against fitted value, yi , in Fig. 11.
2813
Level 2
980 °C
1130 °C
75 mbar
Level 2 (+1)
6.594
5.545
7.065
Fig. 10. The membership functions for the three process variables.
Fig. 11 does not indicate any serious model inadequa-
cies. The random scatter of the residuals indicates indepen-
_
dence between residuals and fitted values yi .
That is a necessary assumption for the validity of the
regression analysis. Therefore, the regression model can
be used to predict the convexity at any combination of fac-
tor settings.
5. Results
At current factor settings, the convexity average is 0.85.
The main results of this research include:
1. Using Taguchi approach, the combination of optimal
factor settings is A1B1C1 or x1 = 960, x2 = 1120, and
x3 = 50. For coded variables, the combination is
(1, 1, 1). Using multiple regressions formula, the
convexity is 0.33001. Thus, an improvement of 61.2%
reduction can be achieved.
2. Using FGP approach, the coded optimal settings for
factors or x1 = 0.33634, x2 = 0.38500, and x3 = 0.29523.
Then, and the Convexity value at optimal settings is
calculated 0.50. This results in reduction in convexity
of 41.2%.
2814 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815

Table 6
ANOVA results.

df Seq. SS Adj. SS Adj. MS p Value

Main effects 3 0.178822 0.178822 0.0596073 0.0001
2-way interactions 3 0.059947 0.059947 0.0199823 0.0001
3-way interactions 1 0.008556 0.008556 0.0085562 0.015
Residual error 8 0.007256 0.007256 0.0009070
Pure error 9 0.007256 0.007256 0.0009070
Total 15 0.254581

Table 7
Effect analysis.

Element Effect Coefficient Standard error T Value p Value

Constant 0.44188 0.007529 58.69 0.0
x1 0.04687 0.02344 0.007529 3.11 0.014
x2 0.19437 0.09719 0.007529 12.91 0.000
x3 0.06875 0.03437 0.007529 4.57 0.002
x1x2 0.09375 0.04688 0.007529 6.23 0.000
x1x3 0.07562 0.03781 0.007529 5.02 0.001
x2x3 0.02188 0.01094 0.007529 1.45 0.184
x1x2x3 0.04625 0.02312 0.007529 3.07 0.015

Fig. 11. Plot of fitted values versus residuals.

Although the Taguchi method reduces convexity larger
than FGP approach, however, it failed to satisfy the setting
preferences on the process factors. That is, by checking the
optimal settings in the corresponding membership func-
tions, it found that the satisfaction value is zero for all
the three factors. However, the optimal factor settings ob-
tained using FGP completely satisfy engineers’ prefer-
ences; lxj equals one.
6. Conclusions
This paper aims at reducing convexity on tile using two
approaches Taguchi and satisfaction function three
process factors; above roller temperature, below roller
temperature, and air blowing. The Taguchi method and
fuzzy goal programming are employed to optimize con-
vexity. The relative anticipated reductions in convexity
are 61.2% and 41.2% using the Taguchi method and FGP,
respectively. Although Taguchi method reduces convexity
larger than FGP approach, however, it failed to satisfy the
preferences on the settings of process factors. Finally,
analysis of variance is conducted to determine significant
factor effects. In conclusion, the FGP successfully opti-
mizes process performance as well as completely satisfies
process engineers’ settings preferences in tiles industry.
Hence, the proposed approach may provide process engi-
neering a great assistance in optimization process perfor-
mance while considering their preferences on a process
and product.
 A. Al-Refaie, A. Diabat / Measurement 46 (2013) 2807–2815 2815

References [6] M.H. Li, A. Al-Refaie, C.Y. Yang, DMAIC approach to improve the
capability of SMT solder printing process, IEEE Transactions on
Electronics Packaging Manufacturing 31 (2) (2008) 126–133, http://
[1] A. Al-Refaie, M.H. Li, Optimizing the performance of plastic injection
dx.doi.org/10.1109/TEPM.2008.919342.
molding using weighted additive model in goal programming,
[7] G. Mazzacani, G. Biffi, Handbook for the Technician of Ceramics
International Journal of Fuzzy System Applications 1 (2) (2011)
Production, Italforni, Italy, 1997.
43–54.
[8] M.S. Phadke, Quality Engineering Using Robust Design, Prentice Hall,
[2] A. Al-Refaie, M.H. Li, K.C. Tai, Optimizing SUS 304 wire drawing
Upper Saddle River, NJ, 1989.
process by grey relational analysis utilizing Taguchi method, Journal
[9] G. Taguchi, Taguchi Methods, American Suppliers Institute Press,
of University of Science and Technology Beijing 15 (6) (2008) 714.
Dearborn, MI, 1991. Research for Development 1.
[3] A. Al-Refaie, T.H. Wu, M.H. Li, Data envelopment analysis approaches
[10] X. Xianghua, Defect Detection in Random Color Textures, PhD thesis,
for solving the multi-response problem in the Taguchi method,
University of Bristol, 2006.
Artificial Intelligence for Engineering Design, Analysis and
[11] M.A. Yaghoobi, M. Tamiz, A method for solving fuzzy goal
Manufacturing 23 (2009) 159–173.
programming problems on MINMAX approach, European Journal
[4] A. Farooq, M. Smith, L. Smith, et al., Dynamic Photometric Stereo for
of Operational Research 177 (2007) 1580–1590.
on Line Quality Control of Ceramic Tiles, Elsevier, Amsterdam, 2005.
[12] M.A. Yaghoobi, D.F. Jones, M. Tamiz, Weighted additive models for
[5] M.T. Hayajneh, A.M. Hassan, F. Al-Wedyan, Monitoring defects of
solving fuzzy goal programming problems, Asia-Pacific Journal of
ceramic tiles using fuzzy subtractive clustering-based system
Operational Research 25 (5) (2008) 715–733.
identification method, Soft Computing 14 (6) (2009) 615–626.