Applied Soft Computing 13 (2013) 3646–3656

Contents lists available at SciVerse ScienceDirect

Applied Soft Computing

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

Assessing the visual quality of sanitary ware by fuzzy logic

Mesut Kumru ∗
Department of Industrial Engineering, Dogus University, Kadıkoy 34722, Istanbul, Turkey

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

Article history: Quality control systems focus on maintaining standards in manufactured products. To improve the homo-
Received 5 March 2012 geneity of batches received by final users and to detect manufacturing defaults, a visual control stage is
Received in revised form integrated into production line before the packing operation. At this stage, the products are inspected,
12 November 2012
mostly by human eye, for their obvious external defects; e.g., chips, cracks, scratches, holes and pitting,
Accepted 12 March 2013
lumps, spots, notches, and glazing. This task is often referred to as visual inspection which is important
Available online 22 April 2013
to categorize the final products into quality-constant batches. Prototypes are of invaluable help while
inspecting the visual attributes of products. Visual quality of products is assessed with respect to these
Keywords:
Quality control
standard units and some quality ratings are made there on the results. However, assessing visual qual-
Product audit ity is somewhat an ambiguous and troublesome work. Therefore, it will be helpful to utilize a decision
Visual inspection making technique, such as fuzzy logic, to facilitate and improve the process. This paper addresses first to
Quality assessment the significance of visual inspection and assessment of visual quality of industrial products, and second,
Fuzzy logic gives a unique application of visual quality assessment of vitreous china ceramic sanitary wares by using
fuzzy logic method.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction products from various sources. There is not a standard rat-

ing system in place and the manufacturer typically determines
In today’s market, quality is a minimal requirement to stay- what qualifies a product for each level. Some of the differences
ing in business. Customers have no tolerance for products that are between ratings involve levels of material and esthetic (appear-
built outside of specifications. In efforts to maintain a highly satis- ance) qualities, life of the product, number of uses, thickness of
fied customer base, manufacturers are implementing systems that coatings/finishes, and quality of components. Many times the man-
ensure products of highest quality. As recognized by well accredited ufacturer can mix components from different grades to match the
international institutions, standards and regulations on the quality requirements of the market, consumer, or grade [4].
of products have imposed more accurate controls on final products Looking at the number of functions a product is expected to
and on all phases of production [24]. fulfill, it is possible to discern a varying amount of sub-qualities
The product audit is one of the oldest methods, which gives a which can be divided into functional and psychological qualities.
statement on the quality of products prior to their delivery to the The functional qualities consist of properties of the material, work-
customer. It assesses the effectiveness of quality assurance through manship and of the technical and economic nature. These are called
the examination of a small number of products and/confirms the the “objective” characteristics of the product. The psychological
quality capability of the production process based on the quality qualities, such as beauty, sportsmanship, prestige, are formed by
of these products. Thereby, it is verified whether the product com- the symbolic meaning the consumer associates with the product.
plies with the given specifications and/or special customer/supplier These are “subjective” characteristics [6], and the subjective charac-
agreements. The results of product audit appear in the form of the teristics play a considerable role in the perception of visual quality
presence or absence of defects, failures, etc. A continuing score or of the product. This is true for ceramic products too.
“rating” of quality is then prepared based on the audit results. Some of the manufacturing companies develop their own qual-
Many manufacturers utilize a quality rating system for their ity rating systems. Vitrified ceramic product manufacturers, for
products. This rating system can give the customers a gauge example in Turkey, use a quality grading system which desig-
for products’ quality and also a better understanding of reasons nates distinct visual qualities for acceptable quality products as
for product cost. This is very important when comparing similar extra, first class, second class, and ordinary. Visual quality levels
of these products substantially dominate the overall product qual-
ity levels. Of course, these different quality levels correspond to
∗ Tel.: +90 216 544 55 55x1432; fax: +90 216 544 55 35. different prices in the market. This necessitates making quality rat-
E-mail address: [email protected] ing decisions as correct as possible. Whereas, due to the subjective

1568-4946/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
https://dx.doi.org/10.1016/j.asoc.2013.03.012
 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656 3647

characteristic of visual attributes, distinguishing the visual quality are easily made due to human subjectivity and eye fatigues. In some
levels of the products by precise (classical) reasoning may lead to cases, human inspection is not even possible when the part to be
incorrect separation of product quality. To avoid this, the fuzzy logic inspected is very small or the production rates are very high. There-
using approximate reasoning will be useful when applied properly. fore, automated inspection is required and it is inherently more
Product audit programs often make use of seriousness classifica- reliable than human inspection. As a result of upgraded and more
tion of defects. Defects are classified in terms such as critical, major, automated technology, manufacturing processes increasingly use
minor A, minor B, each with some “weight” in the form of demer- automation to reduce labor requirements and costs, reduce facti-
its [15], or in some other format such as functional, dimensional, tious error and raise production capacity. Surface defects of modern
and visual (esthetic). Functional and dimensional defects of prod- frangible or small products can be identified using automated
ucts are of variable type and can be detected by laboratory tests and inspection technology [35,37]. Automated system determines the
technical measurements. Visual defects of products, however, are of properties of products using visual information and is most often
attribute type and can be detected mostly by human eye and inter- automated by employing machine vision techniques [31].
preted linguistically. Therefore, functional and dimensional quality Automated visual inspection (AVI) is the automation of the qual-
characteristics are easier to assess when compared to visual quality ity control of manufactured products, normally achieved using a
characteristic of a product. camera connected to a computer. AVI system involves the con-
Assessing visual quality is somewhat an ambiguous and trou- struction of explicit and meaningful descriptions of physical objects
blesome work. Determination of defects varies depending on the based on images. The technology aims to duplicate human vision
inspectors’ knowledge and experience, which results in a large by electronically perceiving and understanding an image [27].
variance and inconsistency in defect reporting. Visual inspectors AVI techniques for defects principally fall into two categories.
normally do their product quality classifications using two-valued Approaches of the first category try to detect defects by comparing
logic (the product is either separated to any one of the quality the testing image with a pre-stored faultless item through a pixel-
classes or not). In-between cases are not principally taken into to-pixel or template-to-template matching process. Approaches of
account. They are implicitly resolved in favor of the nearest cat- the second category, however, select and compute a set of textural
egory. This practice leads to negative concerns and lack of accuracy features in a sub-image, and then compare the feature values with
in the classification of product quality. That is why, it would be those of the faultless image to find significant differences [8]. Both
helpful to utilize fuzzy logic to facilitate and improve the process. approaches have their own disadvantages [40,41,38,39].
Fuzzy logic could be used during the product auditing process in Visual inspection technologies have successful application fields
order to include (assess) in-between cases into the quality classifi- ranging from the electronics devices inspection to the power plants
cation of products in a correct manner. It could offer a more realistic monitoring, from the modern industrial products inspection to
framework for human reasoning than the traditional two-valued the fruits and vegetable classification. With recent advances in
logic. This was the motivation behind this paper. The purpose is to computer technology, modern manufacturers have turned their
demonstrate how to utilize the fuzzy logic method in visual qual- attention to machine-vision inspection systems, and these modern
ity assessment of an industrial product, uniquely for the case of devices now inspect metal products, textile fabrics, nuclear power
ceramic sanitary ware production. fuel elements, trash separation processes, and coal, etc., even while
In the subsequent sections, visual quality inspection is empha- these products are on conveyors. To meet with these wide fields
sized thoroughly along with a short introduction to fuzzy logic of applications, many progresses (wide dynamic range TV cameras,
concept, and an original application of visual quality assessment sophisticated optical-electronical large scale integration to get 3-D
of vitreous china ceramic sanitary wares by fuzzy logic method is shapes of components, new kinds of image processing systems, new
explained. software technologies, etc.) have been introduced into the practical
visual inspection technologies [17].
A visual inspection task typically consists of visual search and
2. Visual quality and its inspection decision making functions which are essentially cognitive activi-
ties and have been shown to be the most important determinants
Visual quality is the degree to which the visual attributes of of inspection performance [22]. If inspection is to be successful, it is
a product join together to form a coherent, harmonious level of critical that these functions be performed effectively and efficiently.
quality. Visual quality is inherently subjective; therefore objective For this purpose, automated inspection is considered the solution to
evaluations are used to quantify the visual assessment of any prod- remove any possible errors from the system. Designers have auto-
uct quality. Rapid prototypes are of invaluable help in testing the mated the functions of the inspection task but have overlooked the
visual (esthetic) attributes of a product during its development innate abilities of humans. In spite of the advantages of automation,
cycle. these inspection systems often fail to meet expectations primarily
Lots of criteria are used to perform the operative assessment of for the reason that they ignore humans’ ability in pattern recog-
the visual quality of an industrial product. Depending on the type nition, as rational decision makers and their flexibility to adapt to
of the product, physical/external defects (corrosion, burrs, han- new situations. Jiang et al. [22] described a system that facilitates
dling damage, cracks, scratches, etc.), chemical/surface appearance the conducting of controlled studies to address issues related to
defects (opaqueness, roughness, spots, etc.), and plating/coating human machine system design and function allocation in visual
heterogeneity (poor/insufficient plating – if plating applied) are inspection. It is known that humans are intelligent and flexible to
checked in visual quality assessments. adapt to changing situations, and at least for simple tasks, they
Visual inspection is to identify non-conformity on a surface of behave as rational decision makers who take into account prob-
an object, in particular assess one or more quantitative or qual- abilities and cost/value structure. Moreover, human information
itative parameters on a surface. Advantageously, the assessment processing is automated and quick. Humans are good at detecting
comprises the determination of at least one parameter and evalu- signals in noise and can make inductive decisions in new situations.
ation of this parameter against a pre-determined standard. Many However, they are limited in computational ability and short-term
humans are engaged in visual inspection tasks but due to factors memory. On the other side, computers are good at computation,
such as tiredness and boredom, their performance is often less memory storage and retrieving, but are poor at detecting signals in
than satisfactory and sometimes inconsistent. Inspection of surface noise and have very little capacity for creative or inductive func-
defects by human eye is quite difficult because wrong judgments tions. Therefore, neither an entirely human nor a purely automated
3648 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656

system may fully achieve the desired performance in an inspection any image some methods were established and placed at different
task. Given this situation, it is logical to explore the potential ben- levels [33].
efits of the pairing of human and machine-vision inspectors. It is In the previous years, some defect detection methods have been
possible to achieve superior performance by a system in which proposed to find out the image defects on ceramic tiles. Those of
certain machine characteristics are dependent on tasks humans prominent ones can be described briefly as follows: Gonzales et al.
are better at executing [22]. The machine vision system is used [14] developed an accurate automatic system to inspect and classify
to inspect a large number of products rapidly. The human inspec- ceramic tiles according to the integrity defects at the tile corners.
tor can then perform slower but more detailed inspection on In order to do this a general segmentation methodology and a geo-
objects that the machine vision system considers to be borderline metric classifier has been developed. In their study, Elbehiery et al.
cases. [11] dealt with the visual inspection of ceramic tiles’ surfaces for
It is seen that when we compare human inspection with auto- the purpose of detecting flaws using a wavelet approach and intro-
mated inspection using machine vision, machine vision generally duced a hierarchical wavelet-based (wave-like oscillation-based)
comes out on top in reliability and repeatability. People win at framework for modeling patterns in digital images. Hocenski et al.
tolerating variations in what constitutes a good or a bad image. [18] presented a method for failure detection in ceramic tile pro-
Until now, there has not been any machine vision system that duction using computer visual diagnostic. They used Canny edge
could replace human completely. However, there have been sys- detector to detect edges and faults on ceramic tiles. Boukouvalas
tems which were quite good approximation of them, and could be et al. [5] concerned about the problem of automatic inspection of
good enough to do the job with more or less success. Most manufac- ceramic tiles using computer vision. They applied techniques for
turers choose between the techniques by translating those specific pinhole and crack detectors for plane tiles based on a set of sepa-
advantages into economic terms. Depending on the case, human rable line filters. Rahaman and Hossain [33] proposed an efficient
inspection may still be favorable against automated inspection, as defect detection and classification technique which is able to find
was in the ceramic sanitary ware industry, or they may be used out image defects at a high rate within a very short time. Ahmady-
conjunctively. fard et al. [2] developed a novel approach for detecting defects of
Even with precisely defined visual quality standards, a certain random textured tiles using Gabor wavelet.
degree of subjective judgment by the person making the inspection To the best of our knowledge, machine vision-based quality
is unavoidable. It will therefore be beneficial to use a tool/prototype inspection systems have not been used in sanitary ware produc-
that minimize the subjectivity of the judgment. tion. Final inspection of finished sanitary ware products have been
performed only by inspectors using their inspection criteria that
require full visual and an occasional physical touch by hands to
3. Literature survey determine the quality of finish. These products are large and heavy
and require full concentration of the inspector(s) to evaluate their
To meet consumer expectation, products should possess condition. This process is slow due to the high data collection
trustworthy appearance properties; maintaining sufficient color requirements. Voice inspection systems (VIS) have been integrated
uniformity, unevenness, and assuring the reproducibility of preci- to this stage. Before voice, the data from inspection was not avail-
sion. Several measurement equipments and approaches have been able real-time or sometimes even the same day due to data entry
utilized for non-uniformity defects. Jiang et al. [23] used lumi- requirements. Keypunch data entry added delays and keypunch
nance measurement equipment to collect data of non-uniformity errors to the data entry process. VIS improves the inspection pro-
defects on LCD panels and applied analysis of variance and expo- cess, allowing the inspector to focus, eyes and hands, on the product
nentially weighted moving average techniques to develop an being inspected.
automatic inspection procedure. Lu and Tsai [26] proposed a global Using voice, the inspector(s) wearing a headset(s) and portable
approach for AVI of micro defects (pinholes, scratches, particles data collection terminal(s) “speaks-in” the inspection data into VIS.
and fingerprints) on TFT panel surfaces using the Singular Value This allows them to more accurately and completely inspect the
Decomposition. Lin utilized discrete cosine transform decomposi- products while freeing the keyboard resources currently used to
tion and cumulative sum techniques to inspect tiny surface defect enter the same data in the computer. Additionally, the data will be
of electronic passive components. Chiu and Lin [8] suggested a immediately available for others to review and take actions.
global approach for AVI of tiny surface defects in surface barrier In this paper, the fuzzy logic is first to be used in product audit-
layer chips, whose random surface texture contains no repetitions ing. Based on the demeriting concept, sanitary ware products are
of basic texture primitives. Bračun et al. [7] demonstrated identifi- qualified through visual assessment by human inspectors using
cation of the typical surface defects on the transmission belts using fuzzy logic approach in decision making.
laser profilometry. In their research, Kung et al. [21] combined the
artificial neural network perceptrons and structure similarity char-
acteristics to create different types of images. Moorthy and Bovik
[30] discussed their vision for the future of visual quality assess- 4. Fuzzy logic
ment research.
Regarding the ceramic industry there are two main groups Fuzzy logic is a form of multi-valued logic derived from fuzzy
of methods developed in the literature for machine vision based set theory to deal with reasoning that is approximate rather than
ceramic tile quality inspection systems. These are reference product precise. The fuzzy logic variables may have a membership value of
comparison methods and self-learning methods [18]. Referenced not only 0 or 1, but a value inclusively between 0 and 1. In fuzzy
product comparison methods are based on comparison of the ref- logic the degree of truth of a statement can range between 0 and 1
erence ceramic product with the analyzed one. This is often used and is not constrained to the two truth values {true (1), false (0)} as
where the texture is the same on every tile. Other methods do not in classic propositional logic [19]. Thus, the fuzzy logic provides a
regard the reference product and are based on learning algorithms basis for approximate reasoning, that is, a mode of reasoning which
and tile feature analysis. Digital image processing is done by com- is not exact nor very inexact. It offers a more realistic framework
puters automatically without or with little human intervention. for human reasoning than the traditional two-valued logic.
One of the most important operations on digital image is to iden- The term “fuzzy logic” emerged as a consequence of the devel-
tify and classify various kinds of defects. To detect the defects from opment of the theory of fuzzy sets by Lotfi Zadeh. In 1965, Zadeh
 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656 3649

Crisp Fuzzificaon Rules Fuzzy Defuzzificaon

Input Output
Fuzzy
Logic/
Inference
Fuzzy Crisp
Input Output

Fig. 1. Fuzzy logic process.

proposed fuzzy set theory [42], and later established fuzzy logic inference. The results of individual rules can be combined in differ-
based on fuzzy sets. The process of fuzzy logic is given in Fig. 1. ent ways (maximum, bounded sum, normalized sum). Depending
Fuzzy logic has an algorithm that is described in the following on the form of the consequent, two main types of rule-based fuzzy
steps [1]: models are distinguished [3]:

1. Define the linguistic variables and terms (initialization). 1. Mamdani linguistic fuzzy model: Both the antecedent and the con-
2. Construct the membership functions (initialization). sequent are fuzzy propositions.
3. Construct the rule base (initialization). 2. Takagi–Sugeno fuzzy model: The antecedent is a fuzzy proposi-
4. Convert crisp input data to fuzzy values using the membership tion; the consequent is a crisp function.
functions (fuzzification).
5. Evaluate the rules in the rule base (inference).
The most commonly used fuzzy inference technique is the so-
6. Combine the results of each rule (inference).
called Mamdani method [28].
7. Convert the output data to non-fuzzy values (defuzzification).
After the inference step, the overall result is a fuzzy value. This
result should be defuzzified to obtain a final crisp output. This is
Linguistic variables are the input or output variables of the
the purpose of the defuzzifier component of a FLS. Defuzzifica-
system whose values are words or sentences from a natural lan-
tion is performed according to the membership function of the
guage, instead of numerical values. A linguistic variable is generally
output variable. There are different algorithms for defuzzification
decomposed into a set of linguistic terms.
as well. These are center of gravity, center of gravity for single-
Membership functions are used in the fuzzification and defuzzi-
tons, center of area, left most maximum, and right most maximum
fication steps of a fuzzy logic system (FLS), to map the non-fuzzy
(see [29,43]). Among these algorithms the most popular one is the
input values to fuzzy linguistic terms and vice versa. A membership
center of gravity (centroid) technique. It finds the point where a
function is used to quantify a linguistic term. There are differ-
vertical line would slice the aggregate set into two equal masses.
ent forms of membership functions such as triangular, trapezoidal,
Mathematically this center of gravity (COG) can be expressed as:
piecewise linear, Gaussian, or singleton.
Fuzzy logic concept can be expressed mathematically as follows. b
a
A (X)xdx
Let X be a nonempty set. A fuzzy set A in X is characterized by its COG = b (3)
membership function A : X → [0, 1] and A (x) is interpreted as the A (X)dx
a
degree of membership of element x in fuzzy set A for each x∈X. It
is clear that A is completely determined by the set of tuples A = ((u, Centroid defuzzification method finds a point representing the
A (u))/u∈X). Frequently A(x) is used instead of A (x). The family of center of gravity of the fuzzy set, A, on the interval, ab.
all fuzzy sets in X is denoted by F(X). Fuzzy logic has been applied to diverse fields, from control the-
If X = (x1 , . . ., xn ) is a finite set and A is a fuzzy set in X then the ory to artificial intelligence. It has a wide range of application areas.
following notation is often used. However, the number of applications in quality control is lim-
1 n ited. Regarding the fuzzy logic applications in quality control, some
A= + ··· + . (1) exemplary studies can be given here. There have been applications
x1 xn
in statistical process control [34,44,12,9,36,16], in sampling [13],
where the term i /xi , i = 1, . . ., n signifies that i is the grade of in QFD [25]. It has been also used for quality-team formation [10].
membership of xi in A and the plus sign represents the union. Perrot et al. [32] gave a review about fuzzy logic application for
From among various membership functions the triangular one food quality control. Without going further on fuzzy logic method
is to be exemplified below.
A fuzzy set A is called triangular fuzzy number with peak (or
center) a, left width ˛ > 0 and right width ˇ > 0 if its membership
function has the following
1
⎧
⎨ 1 − (a − t)/˛ if a − ˛ ≤ t ≤ a
A(t) = 1 − (t − a)/ˇ if a ≤ t ≤ a + ˇ (2)
⎩
0 otherwise

and we use the notation A = (a, ˛, ˇ). A triangular fuzzy membership

function is depicted in Fig. 2.
A triangular fuzzy number with center a may be seen as a fuzzy
quantity “x is approximately equal to a”. a-α a a+β
In a FLS, a rule base is constructed to control the output variable.
A fuzzy rule is a simple IF–THEN rule with a condition and a conclu- Triangular fuzzy number
sion. After evaluating the result of each rule, these results should
be combined to obtain a final result. This process is called (fuzzy) Fig. 2. Triangular fuzzy membership function.
3650 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656

at this stage, the reader is supposed to have a basic understanding (such as glaze flow, unglazed region, blob, and blur) and the other
thereon. For further reading see [20] and [43]. surface defects (such as spots/blemishes, holes/pitting, pinholes,
To the best of our knowledge, no research has been conducted and blisters). Among these defect types, the spotted defects are
for its application in product auditing in general, and specifically for quite widespread, and can be defined as relatively high contrasted
visual inspection. Therefore, this study is the first one on this issue. groups of pixels against to intensity of their surroundings. This kind
In the proposed approach, a fuzzy rule base is used to rank the visual of defects appear during ceramic ware baking process (dark spots),
quality attributes of the sanitary ware products. The approach then irregularity during casting (lumped piles, bumps), or because of
extends the analysis to the quality assessment of the products using dust presence on product surface during its glazing (light or dark
defuzzified linguistic terms. spots). The most common defect types are defined shortly below.
Crack Break down of the corner, edge, or any part of the ware
Scratch Generally scratch on surface
5. Industrial application Chips Irregularity on edges and corners
Pinhole Scattered isolated black–white pinpoint spot
5.1. The company, its products and manufacturing process Blob Water drop spot on ware surface
Spot Discontinuity of color on surface, kiln/firing/baking
spots/blemishes
Serel Ceramic Company is one of the Turkey’s leading manu- Glaze Blurred surface on ware, glaze flow, unglazed region
facturing companies famous with its high quality vitreous china Holes/pitting Slight/small lumps and notches on the surface
sanitary ware products (wash basin, pedestal wash basin, water Deformity Irregular shape of any part of the ware, especially around
closet, urinal, bidet, squatting pan WC, shower tray, kitchen sink, holes and edges

and bathroom accessories). To meet a variety of consumers’ needs, Exemplary prototypes with distinct defect magnitudes cor-
sanitary wares come in a wide range of styles. Most of the products responding to different quality levels are used during visual
are glazed in different colors. inspections in the factory. Visual quality of products is assessed
The production process of sanitary ware includes milling and by the inspectors with respect to these standard units and some
slip preparation section, casting section, green ware section, glaz- quality ratings (superior quality, first quality, second quality, third
ing section, kiln section, and final inspection and packing section. quality, and scrap) are made there on the results. Visual inspec-
The shape and the color of the ceramic products affect the manufac- tors rate each defect as low or average quality and give demerit
turing process. The finished sanitary wares are sorted in prescribed points of −1 for each of the low quality level defects and −0.5
grades and checked for any damages before packing. Today, in the for each of the average quality level defects, and then, sum up all
plant, almost all production stages are more or less automated. these points and assess the visual quality level of the product with
The final inspection stage in the production chain involves the respect to the following demerit classification: from 0 up to −5
activities of final product quality control and classification. This points → super quality; from −5.5. up to −10 points → first quality;
stage includes visual inspection of sanitary ware surface defects from −10.5 up to −15 points → second quality; from −15.5 up to
for integrity in chromatic abnormality, edge and corner defects, −20 points → third quality; and for ≤−20.5 points → scrap. Classi-
cracks and scratches presence, dot shaped and blob defects, glazing cal reasoning is used during assessments, and the inspectors’ skills
defects and printed texture anomaly. This stage is still least auto- and experiences have a direct impact on the quality classification
mated in the whole production chain. Human resources are used of the products. Time by time, inconsistencies occur in decision
here because of the complexity of the quality control process, and making which leads to the search for improvements in assessment
the demand for changeability in quality classification during the process. Accuracy in quality ratings is very important because dif-
production. ferent prices are imposed by the company for the distinct quality
Final product audits are applied for each lot ready to deliver, products in the market.
during which overall visual inspections are carried out by humans. The standard operation process for visual inspection and rating
scale of the sanitary ware is given below.

5.2. Visual quality audits

Surface defects in the ceramic wares are viewed as in- 1. Take a unit of ceramic product from the final stage of operation
homogeneities in regularity and orientation fields. To improve the process.
homogeneity of batches received by final users and to detect manu- 2. By thoroughly handling the product, visually inspect it according
facturing defaults, the company has integrated into the production to the quality attributes, defects and rating standard.
line a visual control stage before the packing operation. It is an 3. Classify the defects with respect to their sizes as low, medium,
important stage to categorize the ceramic wares after production and high.
based on surface defects. The goal of the inspection is not to give a 4. Give demerit points (−1) for each of the low size of defects and
statistical analysis of the production, but to classify every product (−0.5) for each of the medium size of defects.
into quality-constant batches. These tasks are referred to as visual 5. Enter the total demerit points as to each attribute/defect group.
inspection where several visually identifiable failures on product 6. Find the corresponding composite score.
are tried to be distinguished and evaluated regarding the exemplary 7. Determine the quality level of the product/lot according to the
specimens. composite score.
After coming out of the kiln, the finished ware is inspected. The
inspector records a number of parameters, including the model
(plate) number, the date the piece was cast, the person who cast,
sprayed and finished the piece, and the final grade evaluation (good, Inspectors fill in a standard form during the visual inspection
scrap or rework). process (Fig. 3). This form involves sufficient places to record in
Vitreous china sanitary ware surface failures can be of struc- and also the instructions necessary to guide inspectors during the
tural or external type. Structural types of failures mainly refer to process.
the material of the product and cover the defects of cracks, flaws, In the next section, as an improvement to current quality assess-
chips, scratches, and deformity. External type of failures can be ment method, visual quality of a selected product (ceramic bidet)
divided into two parts. Those related with the glazing operation is to be assessed as an experiment by using fuzzy logic method.
 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656 3651

………… Ceramic Co. VISUAL QUALITY CLASSIFICATION No …..

Item Code………………. Item Name……………… Date……………..

Attribute Rating
Item No Surface Glazing Material Total Quality Level
Low Medium Low Medium Low Medium Demerit Sup. 1st 2nd 3rd Scr.
1
2
3
4
5
6
7
8
9
10
Total
Explanation

Inspector’s Signature QC Approval

Name
Notes:
1) Attribute Rating: a) For easily visible surfaces (upper); ≤0.1 cm² Medium, ≤0.2 cm² Low
b) For hardly visible surfaces (side and bottom); ≤ 0.5 cm² Medium, ≤ 1.0 cm² Low.
2) Demeriting: Use -1.0 point for each Low quality attribute, -0.5 for each Medium quality attribute.
3) Quality Rating: a) For individual items; ≥ -0.5 Super, ≥ -1.0 First, ≥-1.5 Second, ≥ -2.0 Third, <-2.0 Scrap.
b) For group of items: The above scale is multiplied by the number of items in the group.

Fig. 3. Visual quality classification sheet.

5.3. Visual quality assessment by fuzzy logic Matlab 6.5 software was used in the study to analyze visual
inspection results by fuzzy logic method. Triangular membership
While visually inspecting the ceramic products, it is very difficult functions were employed for the fuzzy subsets. The Mamdani fuzzy
to take into account all of the above defects one by one and make rules relating the input variables to the output variable were cre-
a decision thereabout. To simplify the quality assessment process, ated. The rating scale for input variables (surface, glazing, and
it was decided to classify the foregoing defects as in the following material) was three-level as of high (10), medium (5), and low
way. (0). The rating scale for output variable (quality) was, however,
five-level as of super quality (10), first quality (7), second quality
Surface defects: Spots/blemishes, holes/pitting, and pinholes. (5), third quality (3), and scrap (0). The model was run based on
Glazing defects: Glaze flow, unglazed region, blob, and blur. the 27 decision rules created during the application (Appendix 1).
Material defects: Crack, scratch, chip, and deformity. The fuzzy logic model, the membership function of input variables,

The severity of these three groups of visual defects was iden- Table 1
Sampling results of visual inspection.
tified with respect to their positioning on the product and their
magnitude as well. Surface and glazing defects were classified as Number of sampling: 1
to their size (cm2 ) while materialistic defects were classified as Quality attribute Low Medium High
Surface 2 3 5
to their size (cm2 ) and length (cm). Thus, low, medium, and high
Glazing 1 2 7
quality product classes were determined and their exemplary spec- Material 0 2 8
imens were prepared to be used during the visual inspection.
Number of sampling: 2
The three criteria used in the factory to perform the operative
Quality attribute Low Medium High
appraisal of the visual (esthetic) quality of the products were those Surface 3 1 6
attributes of surface, glazing, and material. Expert evaluation based Glazing 2 2 6
on these three criteria has been proven to be good predictors of Material 1 2 7
visual quality which consisted of the average performance of these Number of sampling: 3
three criteria. Each of the three criteria is assumed to be indepen- Quality attribute Low Medium High
dent and is intended to evaluate one aspect of visual quality as Surface 1 3 6
Glazing 0 2 8
defined before.
Material 0 1 9
The experimental analysis has been performed on five sets of Number of sampling: 4
ceramic bidets, different in size and surface morphology. The fuzzy Quality attribute Low Medium High
logic method was applied on the data retrieved from random sam- Surface 6 2 2
pling results of 50 bidets taken from 5 different lots (10 units from Glazing 5 3 2
Material 4 2 4
each lot) produced in the 48th week of the year 2010. Sampled Number of sampling: 5
units were assessed by visual inspectors as low, medium, and high Quality attribute Low Medium High
quality with respect to the associated quality attributes of surface, Surface 1 5 4
glazing, and material. The results of sampling inspection are given Glazing 2 3 5
Material 2 2 6
in Table 1.
3652 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656

Fig. 4. Fuzzy logic model.

Table 2
Quality levels of the lots (fuzzy approach).

Number of sampling Attribute rating scale Composite score Quality level of the lot

Surface Glazing Material

1 6.5 8.0 9.0 8.79 Super quality

2 6.5 7.0 8.0 7.54 First quality
3 7.5 9.0 9.5 8.86 Super quality
4 3.0 3.5 5.0 5.61 Second quality
5 6.5 6.5 7.0 7.29 First quality
Average 6.0 6.8 7.7 7.40a First quality
a
It is not the arithmetic average of composite scores, but a score corresponding to the attribute averages.

5.4. Results and discussion

5.4.1. Results of the fuzzy approach

As to each of the sampling results, pertinent quality levels of
the lots were determined by the model as given in Table 2. Accord-
ing to the results; two of the lots were of super quality, one lot
was of second quality, while the rest were of first quality. When
we compare the average points of the defect groups, the material
quality seems the best (7.7), the surface quality seems the worst
Fig. 5. Membership function of input variables “Surface”, “Glazing”, and “Material”. (6.0), and the glazing quality takes part in between (6.8). The com-
posite visual quality score of 7.40 corresponds to first quality level
in general. Thus, visual quality levels of these five lots of ceramic
and the membership function of output variable are displayed in
bidets subjected to sampling were assessed properly by the method
Figs. 4–6.
discussed.
The surface, glazing and material values (weights) of the fail-
ures were identified with the help of expert opinions and by
using a database of 27 decision rules (see Appendix) determined 5.4.2. Comparison with ongoing application results
specifically. The rules were designed to take into account all pos- If we were to apply the ongoing visual inspection approach to
sible situations. The membership function was also viewed as an the same set of data, we would get the following visual quality
expert’s opinion. Mamdani min/max method of inference mecha- assessments of the lots (Table 3). Though the overall quality level
nism (input method: min; aggregate method: max) was used and of the lots is the same (first quality) with that of the fuzzy approach,
the results were defuzzified by center of gravity method. the assessment of two lots are different. While the fuzzy approach
Visual quality assessments were derived from the analysis and assessed the first lot as super quality and the fourth lot as second
consequently visual quality levels of the lots were determined quality, the ongoing approach assessed these lots as first and third
(Table 2). The results are discussed below.
Table 3
Quality levels of the lots (ongoing approach – visual inspection).

Number of Attribute rating scale Composite Quality level

sampling score of the lot

Weak Average

1 3 7 −6.50 First quality

2 6 5 −8.50 First quality
3 1 6 −4.00 Super quality
4 15 7 −18.50 Third quality
5 5 10 −10.00 First quality
Average 6.0 7.0 −9.50 First quality
Fig. 6. Membership function of output variable “Quality”.
 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656 3653

Table 4 Table 7
Sampling results of measurement-based inspection. Coefficients of variations of the methods.

Number of sampling: 1 Measurement-based Ongoing visual Fuzzy logic visual

Quality attribute Low Medium High inspection inspection inspection
Surface 1 3 6
Absolute mean 7.80 9.50 7.40
Glazing 1 1 8
score
Material 0 2 8
Standard 3.99 5.51 1.35
Number of sampling: 2
deviation
Quality attribute Low Medium High
Coefficient of 0.51 0.58 0.41a
Surface 3 1 6
variation
Glazing 1 3 6
Material 1 1 8 a
Adjusted to the quality range scale of the other applications.
Number of sampling: 3
Quality attribute Low Medium High
Surface 1 3 6 composite scores. The quality classification of the lots is the same
Glazing 0 3 7
with that of the fuzzy approach. The results are exhibited in
Material 0 1 9
Number of sampling: 4 Table 5.
Quality attribute Low Medium High
Surface 3 4 3
5.4.4. Overall evaluation
Glazing 4 4 2
Material 2 3 5
The values (composite scores and quality level of the lots) of
Number of sampling: 5 ongoing and fuzzy approaches are given in comparison with the
Quality attribute Low Medium High measurement-based inspection values in Table 6. Quality rating
Surface 1 4 5 differences of the ongoing approach is underlined in the table. As
Glazing 1 3 6
a normalized measure of dispersion of a quality distribution, the
Material 1 2 7
coefficient of variation was calculated for all of the three meth-
ods and given in Table 7. For comparability reasons, the coefficient
quality, respectively. Thus, both methods have performed quite of variation of the fuzzy approach was calculated by adjusting its
different in quality rating of the product. We cannot prefer any- quality range scale to those of the other applications. The lowest
one to another at this stage for better performance. In order to dispersion (0.41) is seen in the values of fuzzy approach, while the
do this we need to have real measurement values on hand as highest dispersion (0.58) is for the ongoing approach, and the coef-
benchmark. ficient of variation for the measurement-based application takes
place in between (0.51). Regarding the quality rating scales, the rat-
5.4.3. Measurement-based inspection results as benchmark ing confidences of these three methods were calculated separately
Sampling results of measurement-based inspection applied for each quality classification. For example, for the first sampling
to the same units selected previously are given in Table 4. All of of the ongoing approach the absolute difference between the com-
the sampled units were measured one by one according to the posite value and the lower bound of the associated class (10.0–6.5)
defectiveness scale (attribute rating: (a) for easily visible surfaces was divided by the class range (4.5), and 0.78 was attained as the
(upper); ≤0.1 cm2 medium, ≤0.2 cm2 low (b) for hardly visible confidence level of the method. It means that the ongoing approach
surfaces (side and bottom); ≤0.5 cm2 medium, ≤1.0 cm2 low), supports the first quality rating of the first sampling at 0.78 level.
and made classification of quality attributes as low, medium, For the fuzzy approach, however, calculating the confidence level
and high. Using the standard demeriting system applied in was a little bit different. Here, the difference between the lower
the factory, the sampled units were classified again as to their bound of the associated class and the composite value (8.00–8.79)

Table 5
Quality levels of the lots (measurement-based inspection).

Number of sampling Attribute rating scale Composite score Quality level of the lot

Weak Average

1 2 6 −5.00 Super quality

2 5 5 −7.50 First quality
3 1 7 −4.50 Super quality
4 9 11 −14.50 Second quality
5 3 9 −7.50 First quality
Average 4.0 7.6 −7.80 First quality

Table 6
Comparison of the methods.

Number of sampling Composite scorea Quality level of the lota Composite scoreb Quality level of the lotb Composite scorec Quality level of the lotc

1 −5.00 Super quality −6.50 First quality 8.79 Super quality

2 −7.50 First quality −8.50 First quality 7.54 First quality
3 −4.50 Super quality −4.00 Super quality 8.86 Super quality
4 −14.50 Second quality −18.50 Third quality 5.61 Second quality
5 −7.50 First quality −10.00 First quality 7.29 First quality
Average −7.80 First quality −9.50 First quality 7.40 First quality
a
Measurement-based inspection.
b
Ongoing approach – visual inspection.
c
Fuzzy approach.
3654 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656

Table 8
Confidence levels of the methods.

Number of sampling Measurement-based inspection Ongoing visual inspection Fuzzy logic visual inspection

1 0 (border) 0.78a 0.40

2 0.56 0.33 0.77
3 0.11 0.20 0.43
4 0.11 0.33a 0.81
5 0.56 0 (border) 0.65
Average 0.27 0.33 (0.17b ) 0.61
Aggregate 7.80 → 0.49 9.50 → 0.11 7.40 → 0.70
a
Accepted (0) for adjusted average value in order to have the same ordering of quality classification.
b
Adjusted average value.

was divided by the class range of 2, and thus, 0.40 was attained for Appendix A. [{(Appendix 1)}]
the first sampling. In the same way, the confidence levels of the
methods concerned was calculated for all the ratings. When the Decision rules
composite score matches the class border values (lower limit) then
the confidence was accepted as (0). Aggregate confidence figures
of the methods were calculated regarding the average composite 1. If (Surface is Low) and (Glazing is Low) and (Material is Low)
scores for the same set of quality rating. But, here, the first and the then (Quality is Scrap).
fourth confidence values of the ongoing approach were accepted 2. If (Surface is Low) and (Glazing is Medium) and (Material is
(0) in order to compute the adjusted average value for the same Low) then (Quality is 3rd quality).
ordering of quality classification. When we look at the final con- 3. If (Surface is Low) and (Glazing is Low) and (Material is
fidence figures we see that the fuzzy approach offers the highest Medium) then (Quality is 3rd quality).
confidence (0.70), the ongoing approach offers the lowest confi- 4. If (Surface is Low) and (Glazing is Medium) and (Material is
dence (0.11), and the measurement-based method brings a control Medium) then (Quality is 2nd quality).
value in between (0.49). Because of using classical reasoning the 5. If (Surface is Medium) and (Glazing is Low) and (Material is
measurement-based method had a confidence level lower than that Low) then (Quality is 3rd quality).
of the fuzzy approach. When compared to the ongoing approach, 6. If (Surface is Medium) and (Glazing is Low) and (Material is
the fuzzy approach here again presented a more reliable result in Medium) then (Quality is 2nd quality).
quality classification (Table 8). 7. If (Surface is Medium) and (Glazing is Medium) and (Material
We can infer from these comparative results that the fuzzy is Low) then (Quality is 2nd quality).
approach performed better in visual quality assessment of sanitary 8. If (Surface is Medium) and (Glazing is Medium) and (Material
ware. It also made available to have knowledge about the defect is Medium) then (Quality is 1st quality).
categories and their weights on the quality level of the products 9. If (Surface is High) and (Glazing is Low) and (Material is Low)
while the ongoing approach only dealt with the number of defects. then (Quality is 2nd quality).
10. If (Surface is High) and (Glazing is Low) and (Material is
Medium) then (Quality is 1st quality).
11. If (Surface is High) and (Glazing is Low) and (Material is High)
6. Conclusion then (Quality is 1st quality).
12. If (Surface is High) and (Glazing is Medium) and (Material is
Visual quality is one of the determinants of overall product Low) then (Quality is 1st quality).
quality. When determined properly, the level of visual quality can 13. If (Surface is High) and (Glazing is Medium) and (Material is
influence significantly the quality rating of the product. Different Medium) then (Quality is 1st quality).
quality levels of a product mean different segments and prices in 14. If (Surface is High) and (Glazing is Medium) and (Material is
the market High) then (Quality is Super quality).
Assessing visual quality is somewhat an ambiguous and trou- 15. If (Surface is High) and (Glazing is High) and (Material is Low)
blesome work, so it may not be possible to make a clear quality then (Quality is 1st quality).
distinction among the product lots by classical logic. Therefore, uti- 16. If (Surface is High) and (Glazing is High) and (Material is
lizing the fuzzy logic would be helpful to facilitate and improve Medium) then (Quality is Super quality).
the process. The fuzzy model developed in this study was quite 17. If (Surface is High) and (Glazing is High) and (Material is High)
satisfactory from this point of view. It performed better than the then (Quality is Super quality).
ongoing approach with respect to the measurements of coefficient 18. If (Surface is Low) and (Glazing is Low) and (Material is High)
of variation and confidence level. then (Quality is 2nd quality).
This method can be used not only for visual quality assess- 19. If (Surface is Low) and (Glazing is Medium) and (Material is
ment of finished products, but also for visual quality assessment of High) then (Quality is 1st quality).
detail parts, subassemblies, and materials as well. Different mem- 20. If (Surface is Low) and (Glazing is High) and (Material is Low)
bership functions can be utilized, and several other attributes can then (Quality is 2nd quality).
be added to the list of visual quality attributes for future applica- 21. If (Surface is Low) and (Glazing is High) and (Material is
tions. As a further research area, all these studies can be extended Medium) then (Quality is 1st quality).
to human–machine combined inspection systems as well. Clearly, 22. If (Surface is Low) and (Glazing is High) and (Material is High)
it is important to test the feasibility of using humans and machines then (Quality is 1st quality).
together in inspection systems. If the result is to be non-trivial, it is 23. If (Surface is Medium) and (Glazing is Low) and (Material is
also necessary to demonstrate that human inspectors and machine- High) then (Quality is 1st quality).
vision systems have better performance than either humans alone 24. If (Surface is Medium) and (Glazing is Medium) and (Material
or machine-vision systems alone. is High) then (Quality is 1st quality).
 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656 3655

25. If (Surface is Medium) and (Glazing is High) and (Material is

Low) then (Quality is 1st quality).
26. If (Surface is Medium) and (Glazing is High) and (Material is
Medium) then (Quality is 1st quality).
27. If (Surface is Medium) and (Glazing is High) and (Material is
High) then (Quality is Super quality).

Appendix B. [{(Appendix 2)}]

Sample pictures

References

[1] Anonymous, A Short Fuzzy Logic Tutorial https://www.cs.bilkent.edu.tr/

∼bulbul/depth/fuzzy.pdf (viewed 10.06.12).
[2] A. Ahmadyfard, H. Alimohamadi, A. Shariati, S. Moghtader, A novel
approach for detecting defects of random textured tiles using Gabor
wavelet, World Applied Sciences Journal 7 (9) (2009) 1114–1119.
[3] R. Babǔska, Fuzzy Systems, Modeling and Identification https://www.dcsc.
tudelft. nl/∼babuska/transp/fuzzmod.pdf (viewed 15.07.12).
[4] D. Blankenship, Quality Rating Systems – Single A Through 5A https://www.
sourcejuice.com/2007/12/21/quality-rating-systems-%E2%80%93-single-a-
through-5a/ (accessed 18.05.09).
[5] C. Boukouvalas, J. Kittler, R. Marik, M. Mirmehdiand, M. Petrou, Ceramic Tile
Inspection for Colour and Structural Defects, under BRITE-EURAM, project no.
BE5638, University of Surrey, 6, 2006.
[6] J.M.F. Box, Product quality assessment by consumers: the role of product infor-
mation, Industrial Management and Data Systems 83 (3/4) (1983) 25–31.
[7] D. Bračun, B. Perdan, J. Diaci, Surface defect detection on power transmis-
sion belts using laser profilometry, Strojniski Vestnik/Journal of Mechanical
Engineering 57 (3) (2011) 257–266.
[8] Y.S.P. Chiu, H.D. Lin, A hybrid approach based on hotelling statistics for auto-
mated visual inspection of display blemishes in LCD panels, Expert Systems
with Applications 36 (10) (2009) 12332–12339.
[9] M. Delgado, P. Olavarrieta, P. Vergara, Fuzzy set based protocols for process
quality control, International Journal of Uncertainty, Fuzziness and Knowledge-
Based Systems 14 (1) (2006) 61–76.
[10] T. Dereli, G. Baykasoglu, S. Das, Fuzzy quality-team formation for value added
auditing: a case study, Journal of Engineering and Technology Management 24
(4) (2007) 366–394.
[11] H. Elbehiery, A. Hefnawy, M.M. Elewa, Surface defects detection for ceramic
tiles using image processing and morphological techniques, Proceedings of
World Academy of Science, Engineering and Technology 5 (April) (2005)
158–160.
[12] S.M. El-Shal, A.S. Morris, A fuzzy rule-based algorithm to improve the perfor-
mance of statistical process control in quality systems, Journal of Intelligent
and Fuzzy Systems 9 (3/4) (2000) 207.
[13] O. Engin, A. Celik, I. Kaya, A fuzzy approach to define sample size for attributes
control chart in multistage processes: an application in engine valve manufac-
turing process, Applied Soft Computing 8 (4) (2008) 1654–1663.
[14] J.M.V. Gonzales, F.A. Linuesa, L. Garcia, An Automatic Visual Inspection
System for Ceramic Tile Manufacturing Defects https://www.disca.upv.es/
articulos/docs/congresos/A012 98.PDF (accessed 10.02.10).
[15] F.M. Gryna, Quality Planning and Analysis, Mc-Graw Hill, Inc., New York,
2001680.
3656 M. Kumru / Applied Soft Computing 13 (2013) 3646–3656

[16] M. Gulbay, C. Kahraman, Development of fuzzy process control charts and fuzzy [29] J.M. Mendel, Fuzzy logic systems for engineering: a tutorial, Proceedings of the
unnatural pattern analyses, Computational Statistics and Data Analysis 51 (1) IEEE 83 (3) (1995) 345–377.
(2006) 434–451. [30] A. Moorthy, A. Bovik, Visual quality assessment algorithms: what does the
[17] S. Hata, Practical visual inspection techniques – optics, micro-electronics and future hold? Multimedia Tools & Applications 51 (2) (2011) 675–696.
advanced software technology, in: 15th International Conference on Pattern [31] J.M. Parker, Z. Hou, A numerical investigation of diffuse images for effective
Recognition (ICPR’00), vol. 4, 2000, p. 4114. defect detection, Proceedings of the Institution of Mechanical Engineers, Part
[18] Z. Hocenski, S. Vasilic, V. Hocenski, Improved Canny edge detector in ceramic B: Journal of Engineering Manufacture 216 (7) (2002) 1073–1079.
tiles defect detection, in: Proceedings of IEEE 32nd Annual Conference on Indus- [32] N. Perrot, I. Ioannou, I. Allais, C. Curt, J. Hossenlopp, G. Trystram, Fuzzy concepts
trial Electronics (IECON), 2006, pp. 3328–3331. applied to food product quality control: a review, Fuzzy Sets and Systems 157
[19] https://en.wikipedia.org/wiki/Fuzzy logic (accessed 05.01.09). (9) (2006) 1145–1154.
[20] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic; Theory and Applications, Prentice [33] G.M.A. Rahaman, M.M. Hossain, Automatic defect detection and classification
Hall, Inc., NY, 1995. technique from image: a special case using ceramic tiles, International Journal
[21] C.-H. Kung, W.-S. Yang, C.-M. Kung, A study on image quality assessment using of Computer Science and Information Security 1 (1) (2009) 22–30.
neural networks and structure similarity, Journal of Computers 6 (10) (2011) [34] H. Rowlands, L.R. Wang, An approach of fuzzy logic evaluation and control in
2221–2228. SPC, Quality & Reliability Engineering International 16 (2) (2000) 91–98.
[22] X. Jiang, I. Bingham, R. Master, A.K. Gramopadhye, B.J. Melloy, A system to [35] K.M. Schmitt, J.R. Riddington, R.C.D. Young, D.M. Budgett, R. Chatwin, Image
understand human–machine function allocation issues in visual inspection, processing applied to brick quality control, International Journal of Advanced
in: Proceedings of the IEA 2000/HFES 2000 Congress, San Diego, 2000, pp. Manufacturing Technology 16 (6) (2000) 434–440.
2-487–2-490. [36] S. Senturk, N. Erginel, Development of fuzzy X–R and X–S control charts using
[23] B.C. Jiang, C.C. Wang, H.C. Liu, Liquid crystal display surface uniformity defect ␣-cuts, Information Sciences 179 (10) (2009) 1542–1551.
inspection using analysis of variance and exponentially weighted moving aver- [37] F.C. Tien, C.H. Yeh, K.H. Hsieh, Automated visual inspection for micro drills in
age techniques, International Journal of Production Research 43 (1) (2005) printed circuit board production, International Journal of Production Research
67–80. 42 (12) (2004) 2477–2495.
[24] A. Lago, F. Pesarin, Nonparametric Combination of Dependent Rank- [38] D.M. Tsai, C.Y. Hsieh, Automated surface inspection for directional textures,
ings with Application to the Quality Assessment of Industrial Products Image and Vision Computing 18 (1) (1999) 49–62.
ftp:https://metron.sta.uniroma1.it/RePEc/articoli/2000-LVIII-1 2-3.pdf (accessed [39] D.M. Tsai, C.H. Chiang, Automatic band selection for wavelet reconstruction in
10.07.09). the application of defect detection, Image and Vision Computing 21 (5) (2003)
[25] C.-T. Lin, A fuzzy logic-based approach for implementing quality function 413–431.
deployment, International Journal of Smart Engineering System Design 5 (1) [40] W.Y. Wu, M.J.J. Wang, C.M. Liu, Automated inspection of printed circuit boards
(2003) 55–65. through machine vision, Computers in Industry 28 (2) (1996) 103–111.
[26] C.J. Lu, D.M. Tsai, Automatic defect inspection for LCDs using singular value [41] C.H. Yeh, D.M. Tsai, A rotation-invariant and non-referential approach for ball
decomposition, International Journal of Advanced Manufacturing Technology grid array (BGA) substrate conducting path inspection, International Journal of
25 (1/2) (2005) 53–61. Advanced Manufacturing Technology 17 (6) (2001) 412–424.
[27] J.J. Lyu, M.N. Chen, Automated visual inspection expert system for multivariate [42] L.A. Zadeh, Fuzzy sets, Information and Control 8 (3) (1965) 338–353.
statistical process control chart, Expert Systems with Applications 36 (3) (2009) [43] H.-J. Zimmermann, Fuzzy Set Theory – and Its Applications, 4th ed., Kluwer
5113–5118. Academic Publishers, Norwell, MA, USA, 2001.
[28] E.H. Mamdani, Application of fuzzy logic to approximate reasoning [44] M.H.F. Zarandi, A. Alaeddin, I.B. Turksen, A hybrid fuzzy adaptive sampling
using linguistic systems, IEEE Transactions on Computers 26 (12) (1977) – run rules for Shewhart control charts, Information Sciences 178 (4) (2008)
1182–1191. 1152–1170.