Abstract

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Forecasting women's apparel sales using mathematical modelingCelia Frank; Ashish Garg; Amar Raheja; Les SztanderaInternational Journal of Clothing Science and Technology; 2003; 15, 2; ABI/INFORM Globalpg. 107

Fuzzy AHP-based decision support system for selecting ERP systemsin textile industry by using balanced scorecard

Ufuk Cebeci*

Department of Industrial Engineering, Istanbul Technical University, Macka, Istanbul 34367, Turkey

a r t i c l e i n f o

Keywords:ERP systemFuzzy AHPTextileBalanced scorecardRequest for proposal

a b s t r a c t

An enterprise resource planning system (ERP) is the information backbone of a company that integratesand automates all business operations. It is a critical issue to select the suitable ERP system which meetsall the business strategies and the goals of the company. This study presents an approach to select a suit-able ERP system for textile industry. Textile companies have some difficulties to implement ERP systemssuch as variant structure of products, production variety and unqualified human resources. At first, thevision and the strategies of the organization are checked by using balanced scorecard. According to thecompany’s vision, strategies and KPIs, we can prepare a request for proposal. Then ERP packages thatdo not meet the requirements of the company are eliminated. After strategic management phase, the pro-posed methodology gives advice before ERP selection. The criteria were determined and then comparedaccording to their importance. The rest ERP system solutions were selected to evaluate. An externalevaluation team consisting of ERP consultants was assigned to select one of these solutions accordingto the predetermined criteria. In this study, the fuzzy analytic hierarchy process, a fuzzy extension ofthe multi-criteria decision-making technique AHP, was used to compare these ERP system solutions.The methodology was applied for a textile manufacturing company.

� 2008 Elsevier Ltd. All rights reserved.

1. Introduction

ERP systems are becoming more necessary for almost everyfirm to improve the competitiveness. According to the success ofthe implementation of ERP system; companies can obtain a com-petitive advantage in the global market rapidly. Over the past dec-ade, many ERP projects have resulted in substantial tangible andintangible improvements in a variety of areas for the organizations(Davenport, 2000; Umble, Haft, & Umble, 2003; Yusuf, Gunaseka-ranb, & Abthorpe, 2004). However, there are a number of exampleswhere organizations were not successful in reaping the potentialbenefits that motivated them to make large investments in ERPimplementations (Davenport, 2000; Umble et al., 2003).

Implementations of ERP systems are one of the most difficultinvestment projects because of the complexity, high cost andadaptation risks. Companies have spent billions of dollars andused numerous amounts of man-hours for installing elaborateERP software systems (Yusuf et al., 2004). A successful ERP projectinvolves selecting an ERP software system and co-operative ven-dor, implementing this system, managing business processeschange and examining the practicality of the system (Wei &Wang, 2004). Karsak and Özogul (2009) presented a novel deci-

sion framework for ERP software selection, employing qualityfunction deployment, fuzzy linear regression and zero–one goalprogramming. Teltumbde (2000) proposed a methodology basedon the nominal group technique and the AHP for evaluating ERPsystems. Chang et al. (2008) proposed a neural network evalua-tion model for ERP performance from SCM perspective. The sur-vey data was gathered from a transnational textile firm inTaiwan (Table 4).

Determining the best ERP software that fits with the organiza-tional necessity and criteria, is the first step of tedious implemen-tation process. Hence, selecting a suitable ERP system is anextremely difficult and critical decision for managers. An unsuit-able selection can significantly affect not only the success of theimplementation but also performance of the company. However,many companies install their ERP systems hurriedly without fullyunderstanding the implications for their business or the need forcompatibility with overall organizational goals and strategies(Hicks & Stecke, 1995). The result of this hasty approach is failedprojects or weak systems whose logic conflicts with organizationalgoals. This paper aims:

� to manage the early stages of ERP selection according to thevision and strategies by using balanced scorecard and

� to provide an analytical tool to select the most suitable ERP soft-ware for textile industry.

0957-4174/$ - see front matter � 2008 Elsevier Ltd. All rights reserved.doi:10.1016/j.eswa.2008.11.046

* Tel.: +90 212 2931300; fax: +90 212 2407260.E-mail addresses: [email protected], [email protected]

Expert Systems with Applications 36 (2009) 8900–8909

Contents lists available at ScienceDirect

Expert Systems with Applications

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

mailto:[email protected]


http://www.sciencedirect.com/science/journal/09574174

http://www.elsevier.com/locate/eswa

Robotics and Computer-Integrated Manufacturing 24 (2008) 174–186

Fuzzy logic path planning for the robotic placementof fabrics on a work table

G.T. Zoumponos, N.A. Aspragathos�

Mechanical & Aeronautics Engineering Department, University of Patras, 26500 Patra, Greece

Received 2 January 2006; received in revised form 2 August 2006; accepted 2 October 2006

Abstract

In this paper, an innovative fuzzy logic approach for the robotic laying of fabrics on a work table and based on fuzzy sets is presented.

Through handling experiments the solution domain for the path of the robotic gripper is determined, the handling parameters are

identified and implicit knowledge is accumulated. Then a proper scheme for the data acquisition is formed and a path-planning algorithm

based on fuzzy logic is developed. Due to conflicts and inaccuracies of the acquired data, a subtractive clustering algorithm is used, to

identify the proper clusters for the two developed fuzzy systems, with the first employing the clusters as rules and the second a neuro-

fuzzy system initialised by the implicit knowledge and trained via back-propagation. Finally, the effectiveness of the two path-planning

systems is investigated in an experimental stage where the robot successfully places on a table fabrics of a variety of materials and sizes.

r 2006 Elsevier Ltd. All rights reserved.

Keywords: Fabric handling; Fuzzy systems; Subtractive clustering; Motion planning

1. Introduction

In the industries where non-rigid materials are processed,like cloth or automobile seats making, the degree ofautomation is still low. The labour-intensive processes inmanufacturing increase the production cost significantly inthe developed countries, so it is very urgent to developflexible manufacturing systems by advancing the robotintelligence control. Saadat and Nan [1] conducted a surveyof 96 published key research papers regarding themanipulation of flexible materials underlining the impor-tance of the automation. A 76% of those publications dealswith sheet materials and 58% of this percentage is relatedto the garment industry, whereas the rest is distributedalmost equally among the aerospace, automotive and shoe/leather industries. This survey shows the importance ofautomation for the manipulation of non-rigid materials ingeneral and particularly the handling of highly flexiblesheet materials in new relevant processes emerging in the

aircraft and automotive industry. Since the apparelindustry is still labour-intensive, automated solutionsshould be found for manually performed tasks andoperations and the operators-workers’ contribution shouldbe restricted to supervising the progress of the productionand thus reduce significantly the production cost.The placement of a ply of a non-rigid material, such as

fabric or leather, on a work table is nowadays a worker’stask, despite some attempts made to automate this task.The placement task includes the laying of a single ply ontop of a work surface, its laying on top of another ply to besewn with and the folding of the ply on to itself, withaccuracy and without the appearance of wrinkles andloops. Such tasks being carried out by a robot presentssome complications, due to the nature of the materials tobe handled. Non-rigid materials, as the term denotes, arematerials whose bending rigidity is quite low, and thuslarge deformations appear even when low bending forcesare applied such as their own weight. In particular, a textilefabric is a very complex non-linear mechanical systemwhose shape, orientation, physical and mechanical proper-ties vary almost unpredictably. It is impossible to obtain‘‘closed form’’ mathematical solutions for the behaviour of

ARTICLE IN PRESS

www.elsevier.com/locate/rcim

0736-5845/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.

doi:10.1016/j.rcim.2006.10.001

�Corresponding author. Tel.: +302610 997268; fax: +30 2610 997212.

E-mail addresses: [email protected] (G.T. Zoumponos),

[email protected] (N.A. Aspragathos).


dx.doi.org/10.1016/j.rcim.2006.10.001



ARTICLE IN PRESS

Genetic optimization of fabric utilization in apparel manufacturing

W.K. Wong �, S.Y.S. Leung

Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong


Article history:

Received 22 May 2006

Accepted 14 February 2008Available online 25 March 2008

Keywords:

Evolutionary strategies

Optimization

Decision support

Resource utilization

a b s t r a c t

In apparel manufacturing, cut order planning (COP) plays a significant role in managing

the cost of materials as fabric usually occupies more than 50% of the total manufacturing

cost. Following the details of retail orders in terms of quantity, size and colour, COP seeks

to minimize the total manufacturing costs by developing feasible cutting order plans with

respect to material, machine and labour. In this paper, a genetic optimized decision-

making model using adaptive evolutionary strategies is proposed to assist the production

management of the apparel industry in the decision-making process of COP in which a

new encoding method with a shortened binary string is devised. Four sets of real

production data were collected to validate the proposed decision support method. The

experimental results demonstrate that the proposed method can reduce both the material

costs and the production of additional garments while satisfying the time constraints set

by the downstream sewing department. Although the total operation time used is longer

than that using industrial practice, the great benefits obtained by less fabric cost and extra

quantity of garments planned and produced largely outweigh the longer operation time

required.

& 2008 Elsevier B.V. All rights reserved.

1. Introduction

In today’s apparel industry, fashion products require asignificant amount of customization due to differences inbody measurements, diverse preferences on style andreplacement cycle. It is necessary for apparel supplychains to be responsive to the ever-changing fashionmarkets by producing smaller jobs in order to providecustomers with timely and customized fashion products.In apparel supply chains, fabric is the single largestmaterial in the cost of a garment; approximately 50–60%of the manufacturing cost can be attributed to fabric.Apart from the fabric material, labour and factoryoperation costs have also been continuously increasingwhile the selling price of apparel merchandise have beendecreasing. Adopting quick response strategies to manu-facture and deliver apparel products to the retailers while

maximizing the fabric utilization rate (in other words,minimizing the material cost) and minimizing the labourand manufacturing cost becomes a great challenge toapparel manufacturers.

1.1. Cut order planning

Cut order planning (COP) is the first stage in theproduction workflow of a typical apparel manufacturingcompany, as shown in Fig. 1. It is a planning process todetermine how many markers are needed, how many ofeach size of garment should be in each marker and thenumber of fabric plies that will be cut from each marker.Marker is the output of the process of marker planning,which is the operation following the COP. Fig. 2 illustratesa marker planning process using commercial computingto arrange all patterns of the component parts of one ormore garments on a piece of marker paper, as shown inFig. 3. Following marker planning, the third operation isfabric spreading, which is a process by which fabric piecesare superimposed to become a fabric lay on a cutting


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

Int. J. Production Economics

0925-5273/$ - see front matter & 2008 Elsevier B.V. All rights reserved.

doi:10.1016/j.ijpe.2008.02.012

� Corresponding author. Tel.: +852 2766 6471; fax: +852 2773 1432.

E-mail address: [email protected] (W.K. Wong).

Int. J. Production Economics 114 (2008) 376– 387

www.sciencedirect.com/science/journal/proeco

www.elsevier.com/locate/ijpe

dx.doi.org/10.1016/j.ijpe.2008.02.012


YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

Investigating the developmentof digital patterns forcustomized apparel

Yunchu Yang and Weiyuan ZhangFashion Institute, Donghua University, Shanghai, People’s Republic of China, and

Cong ShanFashion Institute, Donghua University, Shanghai, People’s Republic of China and

Shanghai University of Engineering Science, Shanghai,People’s Republic of China

Abstract

Purpose – The paper aims to provide an overview of the area of digital pattern developing forcustomized apparel.

Design/methodology/approach – The paper outlines several methods of digital patterndeveloping for customized apparel, and discusses the principles, characters and applications.Digital pattern developing process has two paths. One path develops apparel according to traditional2D pattern-making technology. There are three methods: parametric design, traditional gradingtechnique, and pattern generating based on artificial intelligence (AI). Another path develops patternthrough surface flattening directly from individual 3D apparel model.

Findings – For parametric method, it can improve greatly the efficiency of pattern design or patternalteration. However, the development and application of parametric Computer-Aided-Design (CAD)systems in apparel industry are difficult, because apparel pattern has fewer laws in graphicalstructure. For grading technique, it is the most practical method because of its simple theory, withwhich pattern masters are familiar. But these methods require users with higher experience. Creatingexpert pattern system based on AI can reduce the experience requirements. Meanwhile, a great deal ofexperiments should be conducted for each garment with different style to create their knowledgedatabases. For 3D CAD technology, two methods of surface flattening have been outlined, namelygeometry flattening and physical flattening. But many improvements should be done if the 3D CADsystems are applied in apparel mass customization.

Originality/value – The paper provides information of value to the future research on developing apractical made-to-measure apparel pattern system.

Keywords Customization, Clothing, Parametric measures, Artificial intelligence, Flatness measurement

Paper type Viewpoint

IntroductionIn today’s apparel market, most of consumers desire to personalize the style, fit and color ofthe clothes. They require high-quality customized products at low prices with faster delivery.With this sort of consumer interest in mind, the concept of mass customization emerged inthe late 1980s (Seung-Eun and Chen, 2000). Mass customization is a hybrid of massproduction and customization and is a new manufacturing trend. It is an effective competingstrategy for maximizing customers’ satisfaction and minimizing inventory costs. In thebook, Mass Customization, Pine (1993) defined mass customization as “the mass productionof individually customized goods and services”. Pine stated that information technology andautomation are prerequisite of implementing mass customization because they constitute

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0955-6222.htm

The developmentof digitalpatterns

167

International Journal of ClothingScience and Technology

Vol. 19 No. 3/4, 2007pp. 167-177

q Emerald Group Publishing Limited0955-6222

DOI 10.1108/09556220710741632

ECIS 2002 • June 6–8, Gdańsk, Poland — First — Previous — Next — Last — Contents —

1596

E-BUSINESS IN APPAREL RETAILING INDUSTRY –

CRITICAL ISSUES

Virpi Kristiina Tuunainen

Matti Rossi

Helsinki School of Economics P.O. Box 1210

FIN-00101 Helsinki Finland

email: {tuunaine|rossi}@hkkk.fi

ABSTRACT

The apparel industry has, like most other industries quickly started using the Internet to gain

improvements in the efficiency and effectiveness of operations and marketing. In this report we briefly

overview the developments of electronic commerce in apparel industry. We try to develop a framework

for choosing the right technology and development options based on the business model and business

orientation chosen. We illustrate the framework by four case companies, which have adapted different

basic strategies and business models. The cases include companies with traditional operations with

also physical retail outlets, as well as companies operating only on the Internet. There are still a

number of unresolved problems related both to consumer-oriented e-commerce in general and to

apparel industry in particular. Nevertheless, consumers are increasingly using the Internet to do

extensive amount of research on products and fashion trends before purchasing through any media,

also making more and more online purchases

1. INTRODUCTION

The apparel industry has started using the Internet in an attempt to improve the efficiency and effectiveness of marketing, provide customers access to information about products and their availability, build brand value, and to offer customers a convenient medium to make purchases online. The most valuable aspects of Internet shopping, as compared to store-based ad catalog shopping, are typically perceived to be competitive pricing, one-source shopping, convenience and time-savings (Corral, 2000). In addition to increasing brand loyalty among consumers, the goals of a manufacturer might include increasing opportunities for collaboration with suppliers and customers. A retailer, in turn, might have goals such as increasing sales or revenue by accepting orders through an Internet storefront, getting more customers to come into traditional bricks-and-mortar stores, and reducing customer service costs by allowing customers to view order-tracking information over the Web. (Machan, 2000) According to Xceed Intelligence, the apparel industry has traditionally been slow to adopt new business practices, and the outdated practices have consequently slowed down adoption of e-commerce (Masters, 2000). Apparel has, nonetheless, become the third-largest retail sales category on the Internet and Forrester Research1 expects online sales of apparel to reach $20.2 billion in 2003 - which accounts, however, for just over 7% of total apparel sales. Often heard argument behind the slow take-off lies in the fact that the consumers perceive clothing as products that have to be seen, touched and tried of before the purchase. In addition to the inability to touch, feel or try on clothing on the Internet, consumers have been concerned with returns, security and costs (Kelly, 2000) - worries

1 http://www.forrester.com

Mathematical model and genetic optimization for the jobshop scheduling problem in a mixed- and

multi-product assembly environment: A case study basedon the apparel industry q

Z.X. Guo *, W.K. Wong, S.Y.S. Leung, J.T. Fan, S.F. Chan


Received 4 September 2005; received in revised form 6 March 2006; accepted 23 March 2006Available online 11 Juyl 2006

Abstract

An effective job shop scheduling (JSS) in the manufacturing industry is helpful to meet the production demand andreduce the production cost, and to improve the ability to compete in the ever increasing volatile market demanding multi-ple products. In this paper, a universal mathematical model of the JSS problem for apparel assembly process is construct-ed. The objective of this model is to minimize the total penalties of earliness and tardiness by deciding when to start eachorder’s production and how to assign the operations to machines (operators). A genetic optimization process is then pre-sented to solve this model, in which a new chromosome representation, a heuristic initialization process and modifiedcrossover and mutation operators are proposed. Three experiments using industrial data are illustrated to evaluate the per-formance of the proposed method. The experimental results demonstrate the effectiveness of the proposed algorithm tosolve the JSS problem in a mixed- and multi-product assembly environment.� 2006 Elsevier Ltd. All rights reserved.

Keywords: Job shop scheduling; Mathematical model; Optimization; Genetic algorithm; Apparel industry

1. Introduction

Today’s enterprises are confronted with ever increasing global competition and unpredictable demand fluc-tuations. These pressures compel enterprises to continuously improve the performance of their productionprocesses in order to deliver the finished product within the most approximate period of time and at the lowestproduction cost. The apparel industry is one which is necessary to operate their assembly systems using mixed-and multi-product scheduling method due to rapid market changes.

0360-8352/$ - see front matter � 2006 Elsevier Ltd. All rights reserved.doi:10.1016/j.cie.2006.03.003

q This manuscript was processed by Area Editor Maged Dessouky.* Corresponding author. Tel.: +852 27666465; fax: +852 27731432.

E-mail address: [email protected] (Z.X. Guo).

Computers & Industrial Engineering 50 (2006) 202–219

www.elsevier.com/locate/dsw


ELSEVIER Int. J. Production Economics 54 (19981 65 76

international journal of

production economics

Genetic algorithm approach to earliness and tardiness production scheduling and planning problem

Y. Li a'*, W.H. Ip a, D.W. Wang b

aDepartment of Electronic Engineering, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong bDepartment o['System Engineering, School o/'lnJormation Science and Engineering, Northeastern University. Shenyang Liaoning,

People's Republie of China 110006

Received 8 August 1996; accepted 29 August 1997

Abstract

A genetic algorithm (GA) approach is proposed to address the problem of earliness and tardiness production scheduling and planning (ETPSP) in this paper. The proposed method includes lot-size consideration as well as the conflicting issue of capacity balancing. The common problem of large-scale discrete problem where the restriction of linearity, convexity and differentiability is in the cost function is new one which is completely relaxed by this approach. This paper outlines the fundamental issues of the manufacturing design in a genetic algorithm formulation. Both simulation and comparison results indicate that this new scheduling scheme is an effective and efficient technique to tackle the problem.

Keywords: Production scheduling and planning (PSP); Production and inventory management; GA; Manufacturing resource planning (MRP-II); Just-in-time (JIT)

1. Introduction

MRP-II and JIT are two methods used worldwide for modern production and inventory management. Although they provide many advantages, the high in-process inventory, the nervousness of production planning in MR P - I I manufacturing systems, the shock of bottleneck, sensitiveness of imbalance and uncertainty in JIT manufacturing

*Corresponding author. Tel.: 852-27889956; fax: 852-27887227; e-mail: [email protected].

systems (Ho, 1989; Mazzola et al., 1989; Sugimori et al., 1977; Wang and Xu, 1993) are difficult problems and they remain open.

To overcome these problems and achieve the best result of production and inventory management, since 1980s (Gunasekaran, 1993; Sarker and Fitzsimmons, 1989) more and more researchers have focused on integrating MRP-I I with JIT philosophy. The presented researches on miscellaneous MRP- I I and JIT mainly focus on the control level of production in manufacturing systems (Hodgson and Wang, 1991a,b). Our research interest focuses on using JIT philosophy to improve the production planning approach of MRP- I I by an efficient

0925-5273/98/$19.00 Copyright '.(~ 1998 Elsevier Science B.V. All rights reserved Pll S0925- 5273{97)00 1 24-2

International Journal of Production Research,Vol. 44, No. 21, 1 November 2006, 4465–4490

Determination of fault-tolerant fabric-cutting schedules in a just-in-time

apparel manufacturing environment

C. K. KWONG*y, P. Y. MOKz and W. K. WONGz

yDepartment of Industrial and Systems Engineering, The Hong Kong Polytechnic University,

Hunghom, Kowloon, Hong Kong

zInstitute of Textiles and Clothing, The Hong Kong Polytechnic University,

Hunghom, Kowloon, Hong Kong

(Revision received December 2005)

In apparel manufacturing, accurate upstream fabric-cutting planning is crucialfor the smoothness of downstream sewing operations. Effective and reliablefabric-cutting schedules are difficult to obtain because the apparel manufacturingenvironment is fuzzy and dynamic. In this paper, genetic algorithms and fuzzy-settheory are used to generate fault-tolerant fabric-cutting schedules in a just-in-timeproduction environment. The proposed method is demonstrated by two caseswith production data collected from a Hong Kong-owned garment productionplant in China. Results of the two cases preliminarily show that the geneticallyimproved fault-tolerant schedules effectively satisfy the demand for downstreamproduction units, guarantee consistent and reliable system performance, and alsoreduce production costs through reduced operator idle time. More cases will beconducted in order to further validate the effectiveness of the proposed method.

Keywords: Genetic algorithms; Fuzzy set theory; Parallel machine scheduling;Fabric cutting

1. Introduction

Rapid response to customer demand, and a short product development and

production lead time are essential for the survival of today’s apparel manufacturers.

Advances in information technology and the development of new computation

techniques have made real-time apparel design possible. Research on virtual garment

simulation to interactive fashion design has received much attention recently (Yang

et al. 1992, Cordier et al. 2002, Fuhrmann et al. 2003). In addition to real-time

fashion design, new technologies have also been employed for productivity

improvement in the apparel-manufacturing process (Wong et al. 2000, Mok et al.

in press). Apparel production is a type of assembly manufacturing that involves a

number of processes including fabric spreading, cutting, sewing, and finishing. The

fabric-cutting operation is done in a cutting department, which usually serves several

downstream sewing assembly lines. Ineffective upstream planning causes chaos in the

*Corresponding author. Email: [email protected]

International Journal of Production Research

ISSN 0020–7543 print/ISSN 1366–588X online � 2006 Taylor & Francis

http://www.tandf.co.uk/journals

DOI: 10.1080/00207540600597047

Downloaded By: [Hong Kong Polytechnic University] At: 05:17 26 August 2009

Pattern Recognition 32 (1999) 1049—1060

Artificial neural networks for automated quality controlof textile seams

Claus Bahlmann*, Gunther Heidemann, Helge Ritter

AG Neuroinformatik, Technische Fakultat, Universitat Bielefeld, Universitatsstr. 25, D-33615 Bielefeld, Germany

Received 9 January 1998; received in revised form 3 August 1998

Abstract

We present a method for an automated quality control of textile seams, which is aimed to establish a standardizedquality measure and to lower costs in manufacturing. The system consists of a suitable image acquisition setup, analgorithm for locating the seam, a feature extraction stage and a neural network of the self-organizing map type forfeature classification. A procedure to select an optimized feature set carrying the information relevant for classification isdescribed. ( 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.

Keywords: Neural networks; Self-organizing feature maps (SOFM); Textile seams; Quality control; Feature selection

1. Introduction

Reliable and accurate quality control is an importantelement in industrial textile manufacturing. For manytextile products, a major quality control requirement isjudging seam quality. Presently, this is still accomplishedby human experts, which is very time consuming andsuffers from variability due to human subjectivity. Conse-quently, investigations about automated seam qualityclassification and an implementation of an automatedseam classificator are highly desirable. Such a systemwould be useful not just to objectify quality control oftextile articles but it can also provide a basis to performonline adjustment of sewing machine parameters toachieve smoother seams.

Previous approaches to automated classification oftextile seams were made by Dorrity [1] and Clapp et al.

*Corresponding author. E-mail: [email protected]

[2]. Using piezoelectric sensors, Dorrity [1] measuresthe ratio of thread motion and a sewing machine cycleand compares it to an optimal value. Clapp et al.[2] determine fabric density using beta-rays. Fromdensity variation, a quality measure can be derived.However, an optical control method appears to benot only easier to realize from a technical point ofview, but also more appropriate, since humans also judgevisually.

In this contribution we present a system that can judgeseam quality from greyvalue images. An overview ofthe approach is shown in Fig. 1. The first stage is animage acquisition system, which can record the structureof the seams and map it onto a greyvalue image (step ‘‘a’’in the figure, Section 3). As a next step, an algorithm forlocating the seam is applied (b, Section 4). This allows tonormalize the position of the acquired image. Next, a setof appropriate features is extracted from the normalizedseam images, which have to code information about thequality of the respective seam (c, Section 6). We divide theimages into two sets: the first (training set) is used to train

0031-3203/99/$— See front matter ( 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.PII: S 0 0 3 1 - 3 2 0 3 ( 9 8 ) 0 0 1 2 8 - 9

Analytica Chimica Acta 595 (2007) 72–79

Genetic algorithm optimisation combined with partial least squaresregression and mutual information variable selection procedures in

near-infrared quantitative analysis of cotton–viscose textiles

A. Durand, O. Devos, C. Ruckebusch ∗, J.P. HuvenneLaboratoire de Spectrochimie Infrarouge et Raman (LASIR) UMR CNRS 8516, Bat.C5, Universite des Sciences et

Technologies de Lille (USTL), 59655 Villeneuve d’Ascq, France

Received 12 October 2006; received in revised form 2 February 2007; accepted 13 March 2007Available online 18 March 2007

Abstract

In this work, different approaches for variable selection are studied in the context of near-infrared (NIR) multivariate calibration of textile. First, amodel-based regression method is proposed. It consists in genetic algorithm optimisation combined with partial least squares regression (GA–PLS).The second approach is a relevance measure of spectral variables based on mutual information (MI), which can be performed independently of anygiven regression model. As MI makes no assumption on the relationship between X and Y, non-linear methods such as feed-forward artificial neuralnetwork (ANN) are thus encouraged for modelling in a prediction context (MI–ANN). GA–PLS and MI–ANN models are developed for NIRquantitative prediction of cotton content in cotton–viscose textile samples. The results are compared to full-spectrum (480 variables) PLS model(FS-PLS). The model requires 11 latent variables and yielded a 3.74% RMS prediction error in the range 0–100%. GA–PLS provides more robustmodel based on 120 variables and slightly enhanced prediction performance (3.44% RMS error). Considering MI variable selection procedure,great improvement can be obtained as 12 variables only are retained. On the basis of these variables, a 12 inputs ANN model is trained and thecorresponding prediction error is 3.43% RMS error.© 2007 Elsevier B.V. All rights reserved.

Keywords: Near-Infrared Spectroscopy; Textile; Multivariate calibration; Genetic algorithm; Mutual information; Artificial neural network

1. Introduction

Determining the composition of textile is an essential topicdue to the wide range of applications in production control,custom check or textile waste sorting and, in the near future,rapid measurement methods such as on-line systems or sensorsare expected for this purpose [1–5]. In most applications, theproperties of textile samples are derived from the knowledgeof the chemical composition. The quantitative measurement ofthe composition of textiles is thus a critical issue in the indus-try, more especially as it is framed by the European directiveCE 96/74 [6]. The usual analytical methods for textile blendsanalysis depend on the nature of the considered textiles. Inall the cases, the methods are time consuming, not mention-ing that harmful chemicals are required for dissolution [7]. On

∗ Corresponding author. Tel.: +33 3 20436661; fax: +33 3 20436755.E-mail address: [email protected] (C. Ruckebusch).

the contrary, near-infrared (NIR) spectroscopy provides a rapidand direct measurement and, combined with multivariate cali-bration, it enables accurate determination of textile propertiessuch as raw fibres, finished products or fabrics [8–12]. But amajor difficulty in NIR quantitative analysis remains the largesample-to-sample variation in the reflectance spectra of textilesof identical chemical composition. Indeed, spectra are sensitiveto many non-chemical factors such as weaving, geographicalorigin or industrial supplier [8,9,13].

In multivariate calibration, variable selection attempts toidentify and remove the variables that penalise the performanceof a model since they are useless, noisy and redundant or cor-related by chance [14,15]. Variable selection procedures are ofparticular interest when dealing with spectroscopic data. Indeed,the number of variables is potentially very large with regard tothe number of samples at disposal for a regression model. Usu-ally, this dimensionality problem is circumvented using methodssuch as partial least squares (PLS) regression [16]. But the PLSlatent variables calculated may also be affected by redundancies

0003-2670/$ – see front matter © 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.aca.2007.03.024


dx.doi.org/10.1016/j.aca.2007.03.024

Int. J. Production Economics 114 (2008) 615–630

Fashion retail forecasting by evolutionary neural networks

Kin-Fan Au�, Tsan-Ming Choi, Yong Yu

Business Division, Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Kowloon, Hung Hom, Hong Kong

Received 28 August 2006; accepted 15 June 2007

Available online 10 March 2008

Abstract

Recent literature on nonlinear models has shown that neural networks are versatile tools for forecasting. However, the

search for an ideal network structure is a complex task. Evolutionary computation is a promising global search approach

for feature and model selection. In this paper, an evolutionary computation approach is proposed in searching for the ideal

network structure for a forecasting system. Two years’ apparel sales data are used in the analysis. The optimized neural

networks structure for the forecasting of apparel sales is developed. The performances of the models are compared with the

basic fully connected neural networks and the traditional forecasting models. We find that the proposed algorithms are

useful for fashion retail forecasting, and the performance of it is better than the traditional SARIMA model for products

with features of low demand uncertainty and weak seasonal trends. It is applicable for fashion retailers to produce short-

term retail forecasting for apparels, which share these features.

r 2008 Elsevier B.V. All rights reserved.

Keywords: Forecasting; Evolutionary neural networks; SARIMA

1. Introduction

In fashion retailing, demand uncertainty isnotorious of creating many big challenges inlogistics management (Hammond, 1990). Followingthe fashion trend and market response, fashionproducts have a highly unpredictable demand. Inorder to avoid stock-out and maintain a highinventory fill rate, fashion retailers need to keep asubstantial amount of safety stock. In order toreduce the inventory burden, fashion retailers haveadopted various measures such as the accurateresponse policy (Fisher and Raman, 1996) and

quick response policy (Iyer and Bergen, 1997; Auand Chan, 2002; Choi et al., 2006; Choi and Chow,2007). Some fashion retailers improve their deci-sions by acquiring market information and revisingtheir forecast in multiple stages (see Donohue, 2000;Gallego and Ozer, 2001; Sethi et al., 2001; Choi etal., 2003, 2004; Tang et al., 2004; Choi, 2007). Byutilizing market information (e.g., the sales of otherclosely related fashion products), fashion retailerscan reduce the forecast error and it is widelybelieved that it can help to reduce inventory cost,and hence improve profit (e.g., see Eppen and Iyer,1997). Undoubtedly, forecasting is one crucial taskin retail supply chains (Luxhoj et al., 1996; Chu andZhang, 2003; Thomassey et al., 2005; Sun et al.,2007) and it can affect the retailer and other channelmembers. We hence propose to investigate in this

ARTICLE IN PRESS


0925-5273/$ - see front matter r 2008 Elsevier B.V. All rights reserved.

doi:10.1016/j.ijpe.2007.06.013

�Corresponding author. Tel: +852 2766 6428;

fax: +852 2773 1432.

E-mail address: [email protected] (K.-F Au).


dx.doi.org/10.1016/j.ijpe.2007.06.013


Pattern Recognition 36 (2003) 1645–1659www.elsevier.com/locate/patcog

Neural network based detection of local textile defects

Ajay Kumar∗

Department of Computer Science, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong

Received 5 April 2002; accepted 28 October 2002

Abstract

A new approach for the segmentation of local textile defects using feed-forward neural network is presented. Every fabricdefect alters the gray-level arrangement of neighboring pixels, and this change is used to segment the defects. The featurevector for every pixel is extracted from the gray-level arrangement of its neighboring pixels. Principal component analysisusing singular value decomposition is used to reduce the dimension of feature vectors. Experimental results using this approachillustrate a high degree of robustness for the detection of a variety of fabric defects. The acceptance of a visual inspectionsystem depends on economical aspects as well. Therefore, a new low-cost solution for the fast web inspection using linearneural network is also presented. The experimental results obtained from the real fabric defects, for the two approachesproposed in this paper, have con2rmed their usefulness.? 2003 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.

Keywords: Defect detection; Machine vision; Automated visual inspection; Quality assurance; Neural networks

1. Introduction

Automated visual inspection of industrial goods for qual-ity control plays an ever-increasing role in production pro-cess as the global market pressures put higher and higherdemand on quality. In most cases, the quality inspectionthrough visual inspection is still carried out by humans.However, the reliability of manual inspection is limited byensuing fatigue and inattentiveness. For example in textileindustry, the most highly trained inspectors can only detectabout 70% of the defects [1]. Therefore, the automation ofvisual inspection process is required to maintain high qual-ity of products at high-speed production.

Some of the most challenging visual inspection prob-lems deal with the textured materials. Three common crite-rion used to measure the quality index of textured materi-als are related to material isotropy, homogeneity and texturecoarseness [2]. While there is a remarkable similarity in theoverall automation requirements for the textured materials,the cost eAective solutions are problem speci2c and require

∗ Tel.: +852-23-58-8384; fax: +852-23-58-1477.E-mail address: [email protected] (A. Kumar).

extensive research and development eAorts. Quality assur-ance in production lines for textured materials such as wood[3], steel-roll [4], paper [5], carpet [6], textile [1,7–21], etc.,have been studied by various researchers. The detection oflocal fabric defects is one of the most intriguing problemsin visual inspection, and has received much of the attentionover the years [7–21]. This paper focuses on this problemand investigates some new techniques to address this prob-lem.

1.1. Prior work

Fabric defect detection using digital inspection imageshas received considerable attention during the past decadeand numerous approaches have been proposed in the lit-erature [7–21]. At microscopic level, the inspection prob-lems encountered in digital images become texture analy-sis problems. Therefore texture features based on statistical,geometrical, structural, model based, or signal-processingapproaches are the potential source of investigation [22].In the approach by Cohen et al. [7] Gauss Markov Ran-dom Field (GMRF) model has been used for the charac-terization of fabric textures. The web inspection problem is

0031-3203/03/$30.00 ? 2003 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.doi:10.1016/S0031-3203(03)00005-0


The use of arti®cial neural network (ANN) for modeling ofthe H2O2/UV decoloration process: part I

Yness March Slokar a,*, Jure Zupanb, Alenka Majcen Le Marechal a

aFaculty for Mechanical Engineering, Smetanova 17, 2000 Maribor, SloveniabNational Institute of Chemistry, Hajdrihova 19, 1000 Ljubljana, Slovenia

Received 5 November 1998; accepted 18 January 1999

Abstract

A brief introduction into arti®cial neural networks (ANNs) is given, with emphasis on counter-propagation learning

strategy, as well as their use for the purpose of modeling and optimization of H2O2/UV decoloration process. The useof Plackett±Burman partial factorial design for seven variables on three di�erent levels, for the selection of experi-ments, needed to calculate the signi®cance of variables, is described. Results of learning with Kohonen ANN are

described, and the best prediction assembly suggested. # 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Arti®cial neural network; Decoloration; Variables; Partial factorial design; Ecological parameters; Modeling

1. Introduction

E�uents of dye manufacturers and dye usercompanies are usually highly polluted with di�er-ent types of dyes. Even though there are not manydyes that are proven to be carcinogenic forhumans, the extent of dyes in surface waters isrising, and methods for their removal need to beevaluated. Due to the large number of di�erentdyes that are available on the market at present(over 3000) [1], it is almost impossible to ®nd aperfect method which would satisfactorily purifythe waste-waters, regardless of the chemical natureof the pollutant. Many possibilities have beenreported [2], but these are more or less selective.

Even for the same dye, the decoloration processmay depend on many di�erent factors. For a

decoloration process, one usually has to considerthe time needed for the decoloration to be com-pleted, i.e. the time needed for the dye to be eithercompletely removed or to be removed up to areasonable amount. It is desired that this time is asshort as possible. For these reasons our researchgroup wanted to optimize the decoloration pro-cess, which implies that a model to predict thetime needed for decoloration to conclude had tobe obtained.

Modeling of the decoloration process involvesmany problems, since the process depends onmany factors, i.e. we are dealing with a multi-variate system. Furthermore, the concentration ofthe dye is not the only parameter of interest; eco-logical parameters such as COD, BOD and TOC,are also important. This means our system is alsoa multi-response one. It is also evident that theseproblems cannot be solved by simple linear multi-variate correlation [3]. Recently, arti®cial neural

Dyes and Pigments 42 (1999) 123±135

0143-7208/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.

PII: S0143-7208(99)00022-4

* Corresponding author c/o Tiziana D'Adda, Strada D'Az-

eglio 55, 43100 Parma, Italy.


Engineering the drapability of textile fabricsGeorge K Stylios; Norman J PowellInternational Journal of Clothing Science and Technology; 2003; 15, 3/4; ABI/INFORM Globalpg. 211

Multiple-objective genetic optimization of the spatial designfor packing and distribution carton boxes

S.Y.S. Leung, W.K. Wong, P.Y. Mok *


Received 24 July 2006; received in revised form 20 July 2007; accepted 29 October 2007Available online 6 November 2007

Abstract

Packing and cutting problems, which dealt with filling up a space of known dimension with small pieces, have been anattractive research topic to both industry and academia. Comparatively, the number of reported studies is smaller for con-tainer spatial design, i.e., defining the optimal container dimension for packing small pieces of goods with known sizes sothat the container space utilization is maximized. This paper aims at searching an optimal set of carton boxes for a towelmanufacturer so as to lower the overall future distribution costs by improving the carton space utilization and reducing thenumber of carton types required. A multi-objective genetic algorithm (MOGA) is used to search the optimal design of car-ton boxes for a one-week sales forecast and a 53-week sales forecast. Clustering techniques are then used to study the orderpattern of towel products in order to validate the genetically generated results. The results demonstrate that MOGA effec-tively search the best carton box spatial design to reduce unfilled space as well as the number of required carton types. It isimportant to note that the proposed methodology for optimal container design is not limited to the apparel industry butpractically attractive and applicable to every industry which aims for distribution costs reduction.� 2007 Elsevier Ltd. All rights reserved.

Keywords: Multi-objective genetic algorithms; Clustering technique; Packing and cutting; Container design

1. Introduction

Packing and cutting problem is an active topic of research and has numerous applications in many indus-tries. For example, electronic components are packed into a minimal case to form a device in electronic indus-try. In the case of manufacturing industry, a large sheet of fabric, glass, paper, or woods is usually cut intoseveral smaller pieces of known dimensions. For another instance, loading pallets with goods or filling con-tainers with cargo in distribution industry (Pisinger, 2002). The research on packing and cutting was origi-nated from the seminal work of Gilmore and Gomory (1965). Packing and cutting problem appears inmany related studies such as knapsack loading, assortment, pallet loading, and container loading. In the knap-sack loading of a container, each box has an associated profit and the problem is to choose a subset of the

0360-8352/$ - see front matter � 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.cie.2007.10.018

* Corresponding author. Tel.: +852 2766 4442; fax: +852 2773 1432.E-mail address: [email protected] (P.Y. Mok).

Available online at www.sciencedirect.com


www.elsevier.com/locate/caie


Journal of Materials Processing Technology 140 (2003) 95–99

Constructing an intelligent conceptual designsystem using genetic algorithm

Jeng-Jong LinDepartment of Textile Science, Van Nung Institute of Technology, Chung-Li, Tao-Yuan, Taiwan, ROC

Abstract

Product design is a problem of multi-solution and how to find out the design schema, which is fit for the target demand, is a challengefor a designer to deal with at any moment. The main goal of this paper is focused on developing an intelligent conceptual design system,through whose assistance, the crucial problem for a designer mentioned above can thus be eliminated.© 2003 Published by Elsevier B.V.

Keywords: Conceptual design; Genetic algorithm; Search mechanism; Evaluation mechanism

1. Introduction

The evaluation of an object is through closely comparingit with the others, a specific value of the object to someonecan thus be defined. In other words, the value of an objectto someone can be obtained by evaluation. The evaluationway is set as a forward inductive inference[1], which isof a “1 to 1” mathematic relationship. In other word, thevalue for specific object to a certain individual is unique asonly one kind of value. It never happens there exist two ormore options for the value identification. For an instance,judging from the appearance, the value of a car, which ismanufactured by assembly from various mechanical com-ponents, is always deemed as only one specific value. Thiskind of inference is so called a “forward inductive inferenceof evaluation” (i.e.,A → B), which is illustrated asFig. 1a.On the contrary, while proceeding with reverse inferencebased on the same viewpoint, we can recognize why a cer-tain kind of product style is so as to be represented. Thiskind of reverse inference is a way of “knowledge”. Whileproceeding with product design, firstly, a certain value levelwill be set as target and then start to manufacture the prod-uct, which can live up to the previously set target value, byassembly from a few of appropriate components. It dependson the “forward inductive inference of knowledge”[1], i.e.,the backward deductive inference of “evaluation”, which isof a “1 to n” mathematic relationship. In other words, thereexist many products, which are of the identical value thatcan live up to the customer’s demand. There will be sev-eral different combinations of components to create differ-ent products to meet the desired target value. For example, aconsumer is searching for a preferred car, which is of certain

kind of his expected value. There must exist many differentcombinations for components to assembly as various cars,all of which can meet the customer’s expected value. Theway to find different assembly combinations for a product,which can meet the expected demand, depends on forwardinductive inference of knowledge, i.e., backward deductiveinference of evaluation (i.e.,B → A), which is illustratedasFig. 1b. Product design is a reverse inference by evalua-tion [1–5] (i.e.,Y → X: a forward inference by knowledge)shown asFig. 1d. How to find the attributes combinations tolive up to customer’s conceptual demand is the very prob-lem for a designer to get solved. In this paper, we present anintelligent conceptual design system (ICDS), which is de-veloped using genetic algorithm (GA) to construct a searchmechanism to find various solutions to the product designconstrains. The ICDS can fulfill a computer with intelligenceto create the design inspiration for a designer. The goal ofICDS is to provide a function for the support and explicitcapture of the top-down apparel design process.

2. Design by backward deductive inference

General speaking, the procedure of forming viewpoint onevents for mankind (i.e.,A → B) is firstly by judging thegeneral attributes from a specific object, then the relation-ship between physical mechanics of architecture and psy-chology concept of image for the object can thus be created.On the contrary, despite of proceeding with the judgmentdirectly, the evaluation can be approached through settingup the relationship between various combinations of the at-tributes of various design factors. It is no longer need the

0924-0136/$ – see front matter © 2003 Published by Elsevier B.V.doi:10.1016/S0924-0136(03)00691-5

Robotics and Computer-Integrated Manufacturing 22 (2006) 279–287

Genetic algorithms for the optimal common due dateassignment and the optimal scheduling policy in parallel

machine earliness/tardiness scheduling problems

Liu Min�, Wu Cheng

Department of Automation, Tsinghua University, Beijing 100084, China

Received 23 November 2002; accepted 23 December 2004

Abstract

Earliness/tardiness scheduling problems with undetermined common due date which have wide application background in textile

industry, mechanical industry, electronic industry and so on, are very important in the research fields such as industry engineering

and CIMS. In this paper, a kind of genetic algorithm based on sectional code for minimizing the total cost of assignment of due date,

earliness and tardiness in this kind of scheduling problem is proposed to determine the optimal common due date and the optimal

scheduling policy for determining the job number and their processing order on each machine. Also, simulated annealing mechanism

and the iterative heuristic fine-tuning operator are introduced into the genetic algorithm so as to construct three kinds of hybrid

genetic algorithms with good performance. Numerical computational results focusing on the identical parallel machine scheduling

problem and the general parallel machine scheduling problem shows that these algorithms outperform heuristic procedures, and fit

for larger scale parallel machine earliness/tardiness scheduling problem. Moreover, with practical application data from one of the

largest cotton colored weaving enterprises in China, numerical computational results show that these genetic algorithms are effective

and robust, and that especially the performance of the hybrid genetic algorithm based on simulated annealing and the iterative

heuristic fine-tuning operator is the best among them.

r 2005 Elsevier Ltd. All rights reserved.

Keywords: Parallel machine; Genetic algorithm; Heuristic fine-tuning operator; Simulated annealing; Scheduling; Textile

1. Introduction

With the success of just-in-time (JIT) productionmode in Japan, earliness/ tardiness scheduling problemswith the purpose of JIT production have become anactive research field. In 1990, Baker and Scudder [1]presented the first survey on earliness/tardiness schedul-ing problems, then some papers [2–10] related to thiskind of scheduling problems were published. Currently,most works proposed mainly focus on single machineearliness/tardiness scheduling problems and parallelmachine earliness/tardiness scheduling problems with

fixed due date constraint, but the parallel machineearliness/tardiness scheduling problems with undeter-mined common due date constraint are very typical inindustrial production practice and have wide applicationbackground in textile industry, mechanical industry,electronic industry and so on, in which both the optimalcommon due-date and the optimal scheduling policyused to determine the job number and their processingorder on each machine need determining at the sametime. For the earliness/tardiness scheduling problemwith common due-date constraint, only a few works[4,9,11–14] were proposed. Aiming at single machineproblem, Cheng [11] proposed the method for determin-ing the optimal job sequence, and the result obtainedwas generalized to the identical parallel machine

ARTICLE IN PRESS


0736-5845/$ - see front matter r 2005 Elsevier Ltd. All rights reserved.

doi:10.1016/j.rcim.2004.12.005

�Corresponding author. Tel.: +8610 62793756.

E-mail address: [email protected] (L. Min).


Engineering Applications of Artificial Intelligence 13 (2000) 635–644

Application of interactive genetic algorithm to fashion design

Hee-Su Kim, Sung-Bae Cho*

Department of Computer Science, Yonsei University, 134 Shinchon-dong, Sudaemoon-ku, Seoul 120-749, South Korea

Abstract

In general, computer-aided design support systems have got an approach of traditional artificial intelligence, which statistically

analyzes data such as the behavior of designer, to extract formal design behavior. This approach, however, can neither deal with

continuous change of fashion nor reflect personal taste well, as it just depends on large amount of collected data. To overcome this

sort of problem interactive genetic algorithm (IGA) has been recently proposed, as a new trend of evolutionary computation. IGA

uses human’s response as fitness value when the fitness function cannot be explicitly defined. This enables IGA to be applied to

artistic domains, and we propose a fashion design aid system using it. Unlike the previous works that attempt to model the dress

design by several spline curves, the proposed system is based on a new encoding scheme that practically describes a dress with three

parts: body and neck, sleeve, and skirt. By incorporating the domain-specific knowledge into the genotype, we could develop a more

realistic design aid system for women’s dress. We have implemented the system with OpenGL and VRML to enhance the system

interface. The experiments with several human subjects show that the IGA approach to dress design aid system is

promising. # 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Interactive genetic algorithm; Fashion design; Subjective evaluation; Domain knowledge; User satisfaction; OpenGL

1. Introduction

One of the biggest changes since the industrial revo-lution is on the market economy. Think about clothesmarket. Before the Industrial Revolution, consumershad to make their own clothes or buy one from verysmall producers. Naturally they have few choices on it.However, the Industrial Revolution enables mass-production, and now consumers can make their choicefrom very large amount of clothes. The trend thatconsumers lead the market is now on progress. Perhapsin the future, consumers can order their favorite designto the manufacturer, and then a cloth is producedaccording to that design (Brockman, 1965).

As most consumers are not professional at design,however, a sophisticated computer-aided design systemmight be helpful to choose and order what they want. Itcan be a solution that designer contacts consumers andpercepts their favorite design, but it is not efficient interms of cost and time. Computer-aided fashion designsystems for non-professional may search out user’spreferential design efficiently.

In this paper we develop a fashion design aid systemwith interactive genetic algorithm (IGA) using domain-specific knowledge. We have classified women’s dressdesign into 3 parts, made them as separate 3-D modelswith OpenGL and GLUT library, and producedindividual designs from combination of these models.Through the interaction with user, our system caneffectively suggest the nearest design of what the userprefers to.

This paper is organized as follows: Section 2introduces fashion design and conventional fashiondesign aid systems, and gives an account of IGA.Section 3 describes the overview of the system, genotypeencoding, and genetic operators. Section 4 gives 3-Dmodeling process and system implementation usingOpenGL and GLUT library. Section 5 analyzes someexperimental results.

2. Background

2.1. Fashion design

The word ‘design’ originated from ‘designare’ of theLatin language, which means ‘to symbolize some plan’.

*Corresponding author. Tel.: +82-2-361-2720; fax: +82-2-365-

2579.

E-mail address: [email protected] (S.-B. Cho).

0952-1976/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved.

PII: S 0 9 5 2 - 1 9 7 6 ( 0 0 ) 0 0 0 4 5 - 2

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

Application of artificial neuralnetworks to the prediction

of sewing performance of fabricsPatrick C.L. Hui

Institute of Textiles and Clothing,The Hong Kong Polytechnic University, Kowloon, Hong Kong

Keith C.C. ChanDepartment of Computing,

The Hong Kong Polytechnic University, Kowloon, Hong Kong, and

K.W. Yeung and Frency S.F. NgInstitute of Textiles and Clothing,

The Hong Kong Polytechnic University, Kowloon, Hong Kong

Abstract

Purpose – This paper aims to investigate the use of artificial neural networks (ANN) to predict thesewing performance of fabrics. The purpose of this study is to verify the ANN techniques that could beemulated as human decision in the prediction of sewing performance of fabrics.

Design/methodology/approach – In order to verify the ANN techniques that could be emulatedas human decision in the prediction of sewing performance of fabrics, 109 data sets of fabrics weretested by using fabric assurance by simple testing system and the sewing performance of eachfabric’s specimen was assessed by the domain experts. Of these 109 input-output data pairs, 94were used to train the proposed backpropagation (BP) neural network for the prediction of theunknown sewing performance of a given fabric, and 15 were used to test the proposed BP neuralnetwork.

Findings – After 10,000 iterations of training of BP neural network, the neural network converged tothe minimum error level. The experimental results reveal the great potential of the proposed approachin predicting the sewing performance of fabrics for apparel production.

Originality/value – Generally, the fabric’s performance in the manufacturing process is judgedsubjectively by the operators and/or their supervisors. Current methodologies of acquiring fabricproperty information and predicting fabric sewing performance are still incapable of providing ameans for efficient planning and control for the sewing operation. Further, development of techniquesto predict the sewing performance of fabric is essential for the current apparel productionenvironment. In this paper, the use of ANN to predict the sewing performance of fabrics in garmentmanufacturing is investigated.

Keywords Neural nets, Fabric testing

Paper type Research paper

In apparel production, the sewing process is one of the critical processes in thedetermination of productivity and the quality of the finished garment. Consistency ofsewing quality is essential if the apparel manufacturer is to satisfy the customer. Failingsor variations in sewing quality may be caused by one or both of the following factors:mechanical factor (the sewing machine) and human factor (the operator).



Applicationof artificial

neural networks

291

Received 12 January 2006Accepted 5 February 2007


Vol. 19 No. 5, 2007pp. 291-318


DOI 10.1108/09556220710819500

An intelligent approach to integration andcontrol of textile processes

Sungshin Kim a, George J. Vachtsevanos b,*

a Department of Electrical Engineering, Pusan National University, Pusan, South Koreab School of Electrical and Computer Engineering, Georgia Institute of Technology,

Atlanta, GA 30332-0250, USA

Received 20 September 1998; received in revised form 15 August 1999; accepted 29 October 1999

Abstract

This paper introduces a methodology to integrate and control e�ectively major plant

processes with strong couplings between them. The proposed integration philosophy

consists of cause±e�ect relationships and decides upon control setpoints for the indi-

vidual processes by optimizing a global objective function which aims at improving

process yield. A neuro-fuzzy model and a fuzzy objective function are employed to

address the integration and control tasks. Such models and objective functions are

de®ned and developed using experimental data or an operator's experience. The ob-

jective is to maximize productivity and at the same time, reduce defects in each of the

subsequent operations. A textile plant is considered as a testbed and three major pro-

cesses ± warping, slashing and weaving ± are employed to illustrate the feasibility of the

approach. The supervisory level of the control architecture is intended to continuously

improve the control setpoints depending upon feedback information from the weave

room, slasher operator, and warping data. Ó 2000 Published by Elsevier Science Inc.

All rights reserved.

Keywords: Polynomial fuzzy neural networks; Fuzzy logic control; Integration; Cause±

e�ect relation; Genetic algorithms; Hybrid genetic optimization

Information Sciences 123 (2000) 181±199www.elsevier.com/locate/ins

* Corresponding author.

E-mail addresses: [email protected] (S. Kim), [email protected] (G.J. Vachtsev-

anos).

0020-0255/00/$ - see front matter Ó 2000 Published by Elsevier Science Inc. All rights reserved.

PII: S 0 0 2 0 - 0 2 5 5 ( 9 9 ) 0 0 1 3 0 - 9


An integrated machine vision based system for solving the nonconvex cutting s...Sam Anand; Christopher McCord; Rohit Sharma; Thiagarajan BalachanderJournal of Manufacturing Systems; 1999; 18, 6; ABI/INFORM Globalpg. 396

An implementation of genetic algorithms for rule based machinelearning

S. Settea,*, L. Boullartb

aDepartment of Textiles, University of Ghent, Technologiepark 9, 9052 Zwijnaarde, BelgiumbDepartment of Automation & Control Engineering, University of Ghent, Technologiepark 9, 9052 Zwijnaarde, Belgium

Received 1 July 1999

Abstract

Genetic algorithms have given rise to two new ®elds of research where (global) optimisation is of crucial importance: `GeneticProgramming' and `Genetic based Machine Learning' (GBML). In this paper the second domain (GBML) will be introduced.

An overview of one of the ®rst GBML implementations by Holland, also known as the Learning Classi®er Systems (LCS) willbe given. After describing and solving a well-known basic (educational) problem a more complex application of GBML ispresented. The goal of this application is the automatic development of a rule set for an industrial production process. To thisend, the case study on generating a rule set for predicting the spinnability in the ®bre-to-yarn production process will be

presented. A largely modi®ed LCS, called Fuzzy E�ciency based Classi®er System (FECS), originally designed by one of theauthors, is used to solve this problem successfully. 7 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Genetic based machine learning; Learning classi®er systems; Fuzzy e�ciency based classi®er systems; Textiles; Production process

1. Introduction

Sette et al. (1996) set forth the basic principles ofGenetic Algorithms and some accompanying tech-niques (sharing function, Pareto optimisation, etc.)which were applied in an industial production process.The basic representation scheme of the optimising par-ameter thereby was a simple (mostly 1 byte) string.This `chromosome'-string was subject to the genetic(Darwinist) manipulation, leading from one populationto the next towards ever more ®t o�springs surpassingtheir parents. For many problems this representationscheme is really a key issue, since it directly re¯ects itsfundamental behaviour. A ®xed string may in practicelimit not only the class of suitable problems, but alsothe optimising capabilities in the problem itself.

Among the limitations are: no hierarchical structures,no recursiveness, lack of computational procedures, novariability on the string length, no or poor self-adap-tive mechanisms. Sometimes this can be solved by`smart' encoding techniques, but in many cases there isno outcome. Therefore, there is certainly need formore complexity in the structures undergoing adap-tation by genetic mechanisms. There are two importantways to escape from this rigid ®xed string represen-tation scheme.

The ®rst mechanism is the so-called ``classi®er sys-tems'', which is a cognitive architecture in which thegenetic algorithms allow adaptive modi®cations of apopulation of string based if±then rules: i.e., an archi-tecture with a self learning if±then rule based system,where learning is based on some economical reward/punish principle and genetic algorithms are used toinject `genetic' material. Those classi®ers are describedin the underlying paper.

The second mechanism is the so-called ``genetic pro-graming'', where computer programs are self generatedby the genetic paradigm to perform a speci®c task or

Engineering Applications of Arti®cial Intelligence 13 (2000) 381±390

0952-1976/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.

PII: S0952-1976(00 )00020 -8

www.elsevier.com/locate/engappai

* Corresponding author. Tel.: +32-9264-5744; fax: +32-9264-

5846.

E-mail addresses: [email protected] (S. Sette), boullart@au-

toctrl.rug.ac.be (L. Boullart).

Pergamon

Expert Systems With Applications, Vol. 7, No. 2, p. 337-356, 1994 Copyright © 1994 Elsevier Science Ltd Printed in the USA. All rights reserved

0957-4174/94 $6.00 + .00

AEEF: A Knowledge-Based Framework for Apparel Enterprise Evaluation

SAMBASIVAN NARAYANAN,* L. HO W A RD OLSON, AND SUNDARESAN JAYARAMAN

Georgia Institute of Technology, Atlanta, GA

Abstract-- The practice of subcontracting some or all the operations involved in manufacturing products is prevalent in many industries. The buying organization typically receives bids from several companies offering to carry out these operations. The process of determining whether a manufacturing facility is capable of producing the required quantity of the commodity at the right time and of the specified quality is fairly complex and involved. In this research, a knowledge-based approach has been adopted to identify the major factors that affect the capability of an apparel manufacturing enterprise to perform on a contract. This knowledge-based framework, known as Apparel Enterprise Evaluation Framework ( AEEF), has been developed. In this article, we present the acquisition and representation of the knowledge in a structured hierarchical framework. The article also outlines the selection of the inference mechanism and the decomposition of the high-level abstract enterprise capabilities into low- level observable factors.

1. INTRODUCTION

The practice of subcontracting some or all the operations involved in manufacturing products is prevalent in many industries. The buying organization typically receives bids from several companies offering to carry out these operations. To obtain a good quality product at the right time the buying organization must ensure the bidder's capability to manufacture the product to its requirements. Therefore, there is a need to evaluate the facilities of the bidder's enterprise.

The process of determining whether a manufacturing facility is capable of producing the required quantity of the commodity at the right time and of the specified quality is fairly complex and involved. Also, when more than one contractor bids for manufacturing a product, the buyer needs to evaluate a specific manufacturing facility in comparison to others. The process of selecting the bidder likely to deliver the best value for the in- curred cost is known as source selection. Normally, this process is carried out by experts in the area, and they evaluate bidders according to several criteria, such as manufacturing capability, quality capability, and financial capability.

The ultimate objective of any procurement process is to get the best overall value for the buyer, which is a trade-offbetween the price quoted in the bid and the

* Present address: CAPS LOGISTICS lnc, 2700, Cumberland Park- way Ste. 150, Atlanta, GA, 30339-3321.

Requests for reprints should be sent to Sundaresan Jayararnan, Georgia Institute of Technology, Atlanta, GA 30332-0295.

ability of the bidder to fulfill the performance requirements of the contract. Selecting the lowest bidder may appear to be beneficial at the time of awarding the contract, but it may not necessarily turn out to be the overall best value decision. This is because the total cost involved in the specific lowest bid contract may be higher than the initial bid, as a result of poor quality or failure to fulfill the customer's order on time. So the evaluation of the technological competence of the bidders becomes essential in deciding which bidder should be awarded the contract, and knowledge of the bidders' manufacturing and other capabilities is a prerequisite for performing this evaluation.

In this article, the need for the research, the basic research methodology, the knowledge acquisition for the evaluation framework, and the representation of the acquired knowledge in a structured framework are discussed.

2. COMPUTER-INTEGRATED MANUFACI 'URING (CIM)

Computer-Integrated Manufacturing (CIM) is the philosophy of manufacturing that concentrates on automation of various activities in a manufacturing enterprise with special emphasis on coordination between those activities to achieve integration. CIM involves integrating computers in various functions of an enterprise to produce the fight product at the right time, of the right quality and at the right price (Jayaraman, 1990). Therefore, an important prerequisite for the implementation of CIM is the in-depth knowledge

337

Achieving expected depth of shade in reactive dye applicationusing artificial neural network technique

M. Senthilkumar, N. Selvakumar*

Department of Textile Technology, A.C. College of Technology, Anna University, Chennai 600025, India

Received 25 May 2004; received in revised form 13 August 2004; accepted 20 December 2004


Abstract

Achieving the expected depth of shade in the production of dyed goods is a very important aspect. It requires the termination of

the process at the right time in other words, correct duration of dyeing should be used. Prediction of this duration for the application

of reactive HE dyes on cotton fabric using artificial neural network (ANN) is reported. The results obtained from the network gives

an average training error of around 1% in the prediction of the time duration for achieving the correct depth of shade. The trained

network gives the same average error % when tested with other reactive HE dyes even when the input parameters selected are

beyond the range of inputs, which were used for training the network.


Keywords: Artificial neural network; Neuron; Sigmoid function; Hidden layers; Total dye fixed; Spectral reflectance curve

1. Introduction

Expected depth of shade is one of the very importantqualities to be achieved in the dyed goods. In case, thedepth produced is different from that of the expected,the product has to be either taken for reworking orrejected.

When goods are taken for dyeing, once the recipe andthe conditions of dyeing for a given machine is fixed, theonly parameter which needs attention to achieve theexpected depth of shade is ‘‘the duration of the process’’.The required dyeing duration for a given situation canbe predicted using statistical tools such as multipleregression analysis or computational processors such asartificial neural networks (ANN). Prediction usingANNs is claimed to have better accuracy compared tomultiple regression analysis [1,2].

Neural networks are used for modelling non-linearproblems and to predict the output values for a giveninput parameters from their training values. Most of thetextile processes and the related quality assessments arenon-linear in nature and hence neural networks findapplication in textile technology. Web density control incarding [3], prediction of yarn strength [4], ring androtor yarn hairiness [5], total hand evaluation of knittedfabrics [6], classification of fabric [7] and dyeing [8]defects, tensile properties of needle punched non-wovens[2], quality assessment of carpets [9], dye concentrationsin multiple dye mixtures [1], modelling of the H2O2/UVdecolouration process [10], automated quality control oftextile seams [11], fabric processability in garmentmaking [12] and evaluation of seam puckering ingarments [13] are some of the areas where ANNs havebeen attempted.

An attempt made on the prediction of dyeing timerequired to achieve expected depth of shade in theapplication of reactive HE dyes on cotton fabric usingANN is reported in this paper.

* Corresponding author. Tel.: C91 44 22203564.

E-mail address: [email protected] (N. Selvakumar).

0143-7208/$ - see front matter � 2005 Elsevier Ltd. All rights reserved.

doi:10.1016/j.dyepig.2004.12.016

Dyes and Pigments 68 (2006) 89e94

www.elsevier.com/locate/dyepig


http://www.elsevier.com/locate/dyepig

IJCST12,1

50

International Journal of ClothingScience and Technology,Vol. 12 No. 1, 2000, pp. 50-62.# MCB University Press, 0955-6222

Received January 1998Revised September 1999Accepted September1999

A study of the roll planningof fabric spreading using

genetic algorithmsC.L. Hui Patrick and S.F. Ng Frency

Institute of Textiles and Clothing, The Hong Kong Polytechnic University,Hong Kong, ROC, and

C.C. Chan KeithDepartment of Computing, The Hong Kong Polytechnic University,

Hong Kong, ROC

Keywords Fabric, Garments, Manufacturing

Abstract In the process of fabric spreading, the variance of fabric yardage between fabric rollsmay lead to a difference in fabric loss during spreading. As there are numerous combinations thearrangement of the fabric roll sequences for each cutting lay, it is difficult to construct a rollplanning to minimise the fabric wastage during spreading in apparel manufacturing. Recentadvances in computing technology, especially in the area of computational intelligence, can beused to handle this problem. Among the different computational intelligence techniques, geneticalgorithms (GA) are particularly suitable. GAs are probabilistic search methods that employ asearch technique based on ideas from natural genetics and evolutionary principles. This paperpresents the details of GA and explains how the problem of roll planning can be formulated forGA to solve. The result of the study shows that an optimal roll planning can be worked out byusing GA approach. It is possible to save a considerable amount of fabric when the best rollplanning is used for the production.

IntroductionIn clothing production, the fabric cost alone is about 35-40 percent of the sellingprice of a garment, that is the major cost item in clothing product[1]. In recentyears, the price of fabric has increased continuously, so a certain percentreduction in fabric cost would affect the total manufacturing cost. The fabricspreading and cutting is the major production process that determines thematerial utilization as well as the finished quality of the garment.

Apart from the fabric loss due to the fabric flaws, there are two causes offabric loss in the production process:

(1) marking loss or marker fallout, which is formed because of the gaps andother non-usable areas that take place between the garment panels of amarker; and

(2) spreading loss, which is the fabric loss that exists during the spreadingprocess other than the loss caused by the marker arrangement; theseinclude the end loss, width loss, splicing loss and remnant loss.

The current issue and full text archive of this journal is available athttp://www.emerald-library.com

We are grateful to Mr Lewis Chung for preparing the coding and the results of this experimentin graphic form.

Stochastics and Statistics

A short and mean-term automatic forecastingsystem––application to textile logistics

S�ebastien Thomassey *, Michel Happiette, Jean Marie Castelain

Laboratoire GEMTEX-ENSAIT, 9 rue de l�Ermitage, 59100 Roubaix, France

Received 25 July 2002; accepted 3 September 2002

Available online 13 December 2003

Abstract

In order to reduce their stocks and to limit stock out, textile companies require specific and accurate sale forecasting

systems. More especially, textile distribution involves different forecast lead times: mean-term (one year) and short-term

(one week in average). This paper presents two new complementary forecasting models, appropriate to textile market

requirements. The first model (AHFCCX) allows to automatically obtain mean-term forecasting by using fuzzy

techniques to quantify influence of explanatory variables. The second one (SAMANFIS), based on a neuro-fuzzy

method, performs short-term forecasting by readjusting mean-term model forecasts from load real sales. To evaluate

forecasts accuracy, our models and classical ones are compared to 322 real items sales series of an important ready to

wear distributor.

� 2003 Elsevier B.V. All rights reserved.

Keywords: Forecasting; Fuzzy inference system; Neuro-fuzzy model; Textile logistics

1. Introduction

Textile managers must use forecasting systems,

in order to set up all logistical steps required to

produce and deal with a product. The efficiency ofthe supply management optimization relies on the

forecast accuracy of the finished product sales

(Sboui et al., 2001; Graves et al., 1998).

Sales forecasting in textile industry is very

complex. Indeed, a wide range of textile item ref-

erences exists (about 15 000 per year), and their

historic sale data are often short (104 periods: 2

years on 52 weeks) and particularly perturbed by

numerous factors, which are neither strictly con-

trolled nor identified (De Toni, 2000). These fac-

tors can depend on the item (colors, price,. . .),distributor (number of stores, merchandizing,. . .),customers (fashion,. . .) or external factors

(weather, holidays,. . .). These data are not alwaysavailable and have different influences on sales

(Vroman, 2000).

The various stage durations of a textile items

development (Fig. 1) implies the need for predic-

tion up to one year before the raw materials areordered. Production managers also require item

quantities to manufacture, particularly early in the

case of imported items from far away countries. It

*Corresponding author.

E-mail address: [email protected] (S. Thomas-

sey).

0377-2217/$ - see front matter � 2003 Elsevier B.V. All rights reserved.

doi:10.1016/j.ejor.2002.09.001

European Journal of Operational Research 161 (2005) 275–284


mail to: [email protected]

A new fuzzy approach to improve fashion product development

T.W. Lau a, Patrick C.L. Hui a,*, Frency S.F. Ng a, Keith C.C. Chan b

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, PR Chinab Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, PR China

Received 1 June 2004; accepted 28 April 2005

Available online 26 September 2005

Abstract

This study attempts to use a fuzzy expert system with gradient descent optimization for prediction of fabric specimens in fashion product

development. Compared with the traditional methods used fabric mechanical properties to predict fabric specimens, our advisory system accepts

fabric hand descriptors which are more closely related to the sensory judgments made by individuals during fabric selection. Fifty participants were

selected to evaluate the performance of the proposed fuzzy fabric advisory system. They were asked to express their preferred fabric specimen on

inputs of the 14 bipolar fabric hand descriptors in the system. The fuzzy prediction rules associated with the membership functions of each fabric

specimen were developed from a survey. After fine-tuning of the proposed system, the prediction accuracy is over eighty percent. The outcomes of

this study could help consumers to select the most appropriate fabric and provide field practitioners appropriate suggestions for effective product

development in clothing and fashion industries.

# 2005 Elsevier B.V. All rights reserved.

Keywords: Fuzzy system; Sensory knowledge; Fabric specimen prediction; Fabric hand descriptors

1. Introduction

Traditionally, field practitioners rely on their knowledge and

experience to select an appropriate fabric material for product

development. Automation of the fabric selection becomes an

interesting research area for clothing industries. Abbott [1] and

Howorth and Oliver [6] in 1950 and 1958 were the pioneers to

look at the mechanical properties of fabric closely related to

fabric hand. They extracted those key properties by using the

multiple factor analysis. Sudnik [18] in 1972 applied the laws in

perceptual psychophysics, the Weber-Fechner’s law and the

Steven’s power law in specific, for the selection of fabric

mechanical properties related to fabric hand.

The first prediction model on fabric hand was proposed in

1980 by Kawabata [9] who designed a linear regression model

to predict a total hand value (THV) which is a coarse attribute to

grade fabric hand. Pan et al. [13] in 1988 reformulated the linear

regression model to the nonlinear model. The nonlinear

mapping of fabric mechanical properties to fabric hand was

further extended to intelligent systems.

However, those previous models for predicting the fabric

hand are inefficient. The THV derived from the fabric

properties is coarse to model the goodness in fabric hand. It

is only an objective metric which does not encounter the

different psychological effects from individuals. Individuals

prefer to use their sensory feelings instead of the mechanical

properties to evaluate the fabric performance, because it is more

convenient and direct. In this study, we have developed a fuzzy

fabric advisory system which could provide the most

appropriate fabric satisfying individual desires for fabric hand.

This paper is organized as follows. Section 2 reviews the

literature regarding the fabric hand descriptors and the frame-

work of the standard fuzzy system on the fabric selection. Section

3 outlines the research methodology involving the identification

of the sensory knowledge on the fabric hand descriptors under the

selected woven fabrics and the design of the proposed fuzzy

fabric advisory system with fine-tuning algorithm to refine the

system performance. Section 4 presents the experimental results.

Finally, the last section concludes the findings and describes the

limitations of the current study as well as providing some

suggestions for future research in this area.

2. Literature review

In this section, we review the subjective fabric hand

descriptors and previous works using artificial intelligent

www.elsevier.com/locate/compind

Computers in Industry 57 (2006) 82–92

* Corresponding author. Tel.: +852 27 666 537; fax: +852 27 731 1432.

0166-3615/$ – see front matter # 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.compind.2005.04.003

A neural clustering and classification system for

sales forecasting of new apparel items

Sebastien Thomassey *, Michel Happiette

GEMTEX-ENSAIT, 9 rue de l’ermitage, 59100 Roubaix, France

Available online 28 February 2006

Abstract

The Textile-Apparel-Distribution network actors require a very accurate production and sourcing management to minimize their costs and

satisfy their customers. For a such strategy, distributors rely on sales forecasting system to respond to the versatile textile market. However, the

specific constraints of the textile sales (numerous and new items, short lifetime) complicate the forecasting procedure and distributors prefer to use

intuitive estimation methods of the sales rather than the existing forecasting models. We propose a decision aid system, based on neural networks,

which automatically performs item sales forecasting. Performances of our model are evaluated using real data from an important French textile

distributor.

# 2006 Elsevier B.V. All rights reserved.

Keywords: Sales forecasting; Clustering; Classification; Neural networks

1. Introduction

These last decades, methods based on Supply Chain

Management tools (Manufacturing Requirement Planning,

Distribution Requirement Planning, Enterprise Resource

Planning), have enabled an improvement of the sourcing,

production and distribution of the textile items. However, due to

the competitive environment, the globalization, the irreducible

manufacturing lead times and the uncertainty of the customer’s

demand, the sales forecast is a fundamental success factor of

the supply chain optimization of apparel companies [47]. The

forecasting system must deal with the constraints of the textile

market:

� Large number of items (about 15,000 per year).

� Items with short lifetimes (6–12 weeks).

� Substitution of most of the items for each collection (95%).

� Long lead time of textile items requires considerations of

producing and planning of sourcing at a mid-term horizon

(the forecasting horizon is one season or 1 year).

� Influence of many explanatory variables. These factors can

be: weather data, holiday, marketing action, promotions,

fashion, economic environment.

Several forecasting models, such as regression models

[51,24,7], exponential smoothing and Box & Jenkins models

[52,21,58], neural networks [74,77,50] or fuzzy systems

[15,46,11], have been developed and provide satisfactory

results in different domains [36]. However, their performances

strongly depend of the field of application, the forecasting goal,

the user experience, and the forecast horizon [4,13] and these

methods are not easily usable in the specific textile environ-

ment. Preceding works are enabled us to carry out a global

forecasting system for textile sales forecasting. This system,

based on soft computing techniques, is composed of several

models which automatically performs mid- [64,65] and short-

terms sales forecasting [66]. However, due to the substitution of

most items for each collection, the aggregation of sales by items

families or by clustering procedures to obtain complete

historical data of several years is required.

This paper deals with the mid-term sales forecasting for

items for which we have no historical sales data. Thus, we

propose to improve the system by combining clustering and

classification tools. The uncertainty and the complex relation-

ship between the sales and the descriptive criteria of items lead

us to rely on neural techniques.

Section 2 describes the proposed forecasting system.

Section 3 reports and analyzes the empirical results obtained

with real data supplied by an important French ready-to-wear

distributor.

www.elsevier.com/locate/asoc

Applied Soft Computing 7 (2007) 1177–1187

* Corresponding author. Tel.: +33 3 20 25 64 64; fax: +33 3 20 27 25 97.

E-mail address: [email protected] (S. Thomassey).

1568-4946/$ – see front matter # 2006 Elsevier B.V. All rights reserved.

doi:10.1016/j.asoc.2006.01.005

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

A hybrid model using genetic algorithm and neural networkfor classifying garment defects

C.W.M. Yuen a, W.K. Wong a,*, S.Q. Qian a, L.K. Chan a, E.H.K. Fung b

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Kowloon, Hong Kongb Department of Mechanical Engineering, The Hong Kong Polytechnic University, Kowloon, Hong Kong

Abstract

The inspection of semi-finished and finished garments is very important for quality control in the clothing industry. Unfortunately,garment inspection still relies on manual operation while studies on garment automatic inspection are limited. In this paper, a novelhybrid model through integration of genetic algorithm (GA) and neural network is proposed to classify the type of garment defects.To process the garment sample images, a morphological filter, a method based on GA to find out an optimal structuring element,was presented. A segmented window technique is developed to segment images into several classes using monochrome single-loop rib-work of knitted garment. Four characteristic variables were collected and input into a back-propagation (BP) neural network to classifythe sample images. According to the experimental results, the proposed method achieves very high accuracy rate of recognition and thusprovides decision support in defect classification.� 2008 Elsevier Ltd. All rights reserved.

Keywords: Image segmentation; Morphological filter; Genetic algorithms; Neural network; Garment inspection

1. Introduction

Although clothing manufacturers have devoted a greatdeal of effort and investment to implement systematic train-ing programs for sewing operatives before they are assignedto work on the production floor, the sizing, stitching andworkmanship problems can still be found during the on-lineand final inspections. Quality inspection of garments is animportant aspect of clothing manufacturing. However,defect detection is usually done by human inspectors, andresults are greatly influenced by their mental and physicalconditions. Therefore, automatic inspection systems (AIS)are becoming fundamental to advanced manufacturing.

There have been a lot of studies of fabric inspectiontechniques to detect defects since the last two decades. Shi-mizu, Ishikawa, and Kayama (1990) used an expert systemto recognize fabric defects. A Gaussian Markov random

field was also used to inspect fabric defects (Cohen, Fan,& Attai, 1991). Jasper and Potlapalli (1995) reported thatthe wavelet transform gave better results than the Sobeledge operating and the fast Fourier transform in terms offabric defect detection. Chen, Liang, Yau, Sun, and Wang(1998) used a BP neural network with power spectra toclassify textiles. Shiau, Tsai, and Lin (2000) classified webdefects with a BP neural network by color image process-ing. A method of laser-based morphological image process-ing was studied to detect fabric defects (Goswami & Datta,2000). A survey of several techniques available for theinspection of textured surfaces could also be found(Kumar, 2001). Now there are three AIS of fabric, namelyBarco Vision’s Cyclops, Elbit Vision Systems’s I-Tex andZellweger Uster’s Fabriscan, which are available on themarket.

Most of the past research was about fabric inspection orgeneral web material inspection, and there were few on gar-ment inspection. The development of automatic garmentinspection to replace manual inspection in the clothing


doi:10.1016/j.eswa.2007.12.009

* Corresponding author. Tel.: +852 2766 6471; fax: +852 2773 1432.E-mail address: [email protected] (W.K. Wong).

www.elsevier.com/locate/eswa



Expert Systemswith Applications


YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

A Hybrid Fuzzy Knowledge-Based Expert System and Genetic Algorithmfor efficient selection and assignment of Material Handling Equipment

S. Hamid L. Mirhosseyni *, Phil WebbSchool of Mechanical, Materials and Manufacturing Engineering, The University of Nottingham, University Park, Nottingham NG7 2RD, UK


Keywords:Material Handling EquipmentFuzzy Knowledge-Based Expert SystemGenetic AlgorithmArtificial intelligence

a b s t r a c t

Material Handling (MH) is one of the key issues for every production site and has a great impact on man-ufacturing costs. The core concern in the design of a MH system is selecting the most suitable equipmentfor every MH operation and optimising them totally in order to attain an optimum solution. This paperpresents a hybrid method for the selection and assignment of the most appropriate Material HandlingEquipment (MHE). In the first phase, the system selects the most appropriate MHE types for every MHoperation in a given application using a Fuzzy Knowledge-Based Expert System consisting of two setsof rules: Crisp Rules and Fuzzy Rules. In the second phase, a Genetic Algorithm (GA) searches throughoutthe feasible solution space, constituting of all possible combinations of the feasible equipment specifiedin the previous phase, in order to discover optimum solutions. The validity of the methodology developedin this paper is proved through the use of a real problem. Finally a comparison of the method with theother available publicised methods reveals the effectiveness of this hybrid approach.


1. Introduction

A Material Handling (MH) system is responsible for transport-ing materials between workstations with minimum obstructionand joins all workstations and workshops in manufacturing sys-tems by acting as a basic integrator (Sujono & Lashkari, 2007).According to our definition, ‘‘MH is the art of implementing move-ment economically and safely” (Apple, 1972). The key role of a MHsystem in industry is apparent simply because without it themovement of materials between processes is impossible and pro-duction therefore could not be accomplished.

Additionally, the MH cost is a substantial component of the to-tal costs in manufacturing. Tompkins et al. (1996) estimated that ina typical manufacturing operation, 25% of the number of employ-ees, 55% of all plant area, and 87% of production time are assignedto MH and it accounts for between 15% and 70% of the total cost ofmanufacturing a product.

In summary, an efficient MH system greatly improves the com-petitiveness of a product through the reduction of handling cost,enhances the production process, increases production and systemflexibility, provides effective utilisation of manpower and de-creases lead time (Chan, 2002; Chu, Egbelu, & Wu, 1995). Havingan efficient and cost-effective MH system necessitates designing

the entire MH system at once even though it may comprise severalsubsystems. The selection and configuration of Material HandlingEquipment (MHE) types are the key subsystems in the design ofa MH system (Chan, 2002; Park, 1996).

Since the 1970s research concentrating on the selection andassignment of MHE has been carried out and significant achieve-ments have been attained. The vast majority of the research hasaddressed only the selection problem whilst there only a smallamount aimed at developing methods for the resolution of boththe selection and assignment problem and to reach a comprehen-sive solution for the whole MH problem.

This paper presents a novel two-phase method employing a Hy-brid Fuzzy Knowledge-Based Expert System and Genetic Algorithmto practically solve both the selection and assignment of MHEproblem. In the first phase a Fuzzy Expert System is used to iden-tify the best MHE types for every handling operation with theirappropriateness factors whilst in the second phase a GA investi-gates throughout the feasible solution space to select a numberof optimal solutions.

This paper is organised as follows. Section 2 presents an over-view of some of the related literature. In Section 3 the MHE selec-tion and assignment problem is reviewed in a total view. While themethodology framework is presented in Sections 4–6, respectively,discuss the first and the second phase of the method in detail. Thesoftware developed is outlined in Section 7, and the capacityanalysis together with the comparison of the method with othertechniques is discussed in Section 8. Finally the last sectionconcludes the discussion of this paper.


* Corresponding author. Tel.: +44 1159786157.E-mail addresses: [email protected], [email protected]

(S.H.L. Mirhosseyni).









YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight


A global forecasting support system adapted totextile distribution

S!ebastien Thomassey*, Michel Happiette, Jean-Marie Castelain

GEMTEX-ENSAIT, 9 rue de l’Ermitage, BP 30329, Roubaix cedex 1 F-59056, France

Received 20 February 2003; accepted 8 March 2004

Abstract

Competition and globalization imply a very accurate production and sourcing management of the Textile–Apparel–

Distribution network actors. A sales forecasting system is required to respond to the versatile textile market and the

needs of the distributor. Nowadays, the existing forecasting models are generally unsuitable to the textile industry. We

propose a forecasting system, which is composed of several models and performs forecasts for various horizons and at

different sales aggregation levels. This system is based on soft computing techniques such as fuzzy logic, neural

networks and evolutionary procedures, permitting the processing of uncertain data. Performances of our models are

then evaluated using the real data from an important French textile distributor.


Keywords: Sales forecasting; Textile distribution; Soft computing

1. Introduction

As in other competitive industries, companies inthe Textile–Apparel–Distribution network requirethe rigorous management of sourcing, productionand distribution. Supply chain management(SCM) (Lee and Sasser, 1995), includes all theseprocesses from the expression of the demand untildelivery of the finished products. This SCMconcept uses tools which intervene at various stepsand places in the logistic chain (GPA, MRP, DRP,ERP, etc.) and include different functions, such aspurchasing, sourcing, production planning, inven-tory control and exchanges of information. How-

ever, even if existing methods improve thereactivity of Textile–Apparel–Distribution net-work, many transformations, which are requiredto produce textile item, always impose significantand not easily reducible manufacturing lead times.Globalization, which causes dispersion of networkactors, also increases these lead times. Thus, inorder to deal with the customer’s requests,companies often need to anticipate productionand to produce items for stock.

The main goal of the distributors is to offer theright product, at the right place and the right price,while maintaining the right stock. These con-straints require an appropriate sales forecastingsystem. For the distributor, the correct anticipa-tion of requests from the consumer allows theupstream companies to provision and adjust theirproduction. Thus, the effectiveness of the supply

ARTICLE IN PRESS

*Corresponding author.

E-mail address: [email protected]

(S. Thomassey).


doi:10.1016/j.ijpe.2004.03.001

A genetic algorithm for solving the two-dimensional

assortment problem

Chang-Chun Lin *

Department of Information Management, Kun-Shan University, No. 949, Da-Wan Road, Yung-Kang City, Tainan 710, Taiwan, ROC

Received 11 October 2003; received in revised form 8 March 2006; accepted 8 March 2006

Available online 2 May 2006

Abstract

Assortment problems arise in various industries such as the steel, paper, textiles and transportation industries. Two-

dimensional assortment problems involve finding the best way of placing a set of rectangles within another rectangle whose

area is minimized. Such problems are nonlinear and combinatorial. Current mixed integer programming models give optimal

solutions, but the computation times are unacceptable. This study proposes a genetic algorithm that incorporates a novel

random packing process and an encoding scheme for solving the assortment problem. Numerical examples indicate that the

proposed genetic algorithm is considerably more efficient and effective than a fast integer programming model. Errors with

respect to the optimal solutions are low such that numerous practical industrial cutting problems can be solved efficiently using

the proposed method.

q 2006 Elsevier Ltd. All rights reserved.

Keywords: Assortment problem; Genetic algorithm; Random bottom-left procedure

1. Introduction

Two-dimensional assortment problems arise when a given set of rectangles are to be cut from a larger rectangle

whose area is to be minimized, or when a set of rectangles are to be packed within a enveloping rectangle of minimal

area. A problem that is similar to the assortment problem is the trim-loss problem (Hinxman, 1980). Assortment and

trim-loss problems arise in various industries such as the steel, paper, textiles and transportation industries. One

application of the assortment problem concerns the layout design of a set of departments (Buffa, Armour, & Vollman,

1964; Seehof & Evans, 1967). In a trim-loss problem, the sizes of the material stocks from which smaller pieces are

cut off are predetermined, unlike the size of the enveloping rectangle in an assortment problem. All such problems can

be classified as one-dimensional, 11⁄2 -dimensional and two-dimensional (Hinxman, 1980). However, 11⁄2 -dimensional

problems can be viewed as special cases of two-dimensional problems. Thus, this study focuses mainly on two-

dimensional assortment problems.

The assortment problem has been studied less than the trim-loss problem. Page (1975) first recognized the

assortment problem in relation to the cutting of steel bars and developed a dynamic programming formulation.

Meanwhile, Chambers and Dyson (1976) considered a version of the two-dimensional problem in which possible

stock size widths and lengths are integers in given ranges. Beasley (1985) developed an integer model and a heuristic



0360-8352/$ - see front matter q 2006 Elsevier Ltd. All rights reserved.

doi:10.1016/j.cie.2006.03.002

* Fax: C886 6 273 2726.

E-mail address: [email protected].

http://www.elsevier.com/locate/dsw


PERGAMON Computers & Industrial Engineering 37 (1999) 375-378

o o r n p u ~ induata-lal

engirmm, lng

A Genetic Algorithm for a 2D Industrial Packing Problem E. Hopper, B. Turton

University of Wales, Cardiff, School of Engineering, Electronic Division, Newport Road, Cardiff, CF2 3TD, UK, tel. +44-1222-874425, fax: +44-1222-874716,

HopperE @ cf.ac.uk, Turton @ cf.ac.uk

Cutting and packing problems are encountered in many industries, with different industries incorporating different constraints and objectives. The wood-, glass- and paper industry are mainly concerned with the cutting of regular figures, whereas in the ship building, textile and leather industry irregular, arbitrary shaped items are to be packed. In this paper two genetic algorithms are described for a rectangular packing problem. Both GAs are hybridised with a heuristic placement algorithm, one of which is the well-known Bottom-Left routine. A second placement method has been developed which overcomes some of the disadvantages of the Bottom-Left rule. The two hybrid genetic algorithms are compared with heuristic placement algorithms. In order to show the effectiveness of the design of the two genetic algorithms, their performance is compared to random search. © 1999 Elsev ie r Science Ltd. All rights reserved. Keywords: two-dimensional orthogonal packing problem, nesting, combinatorial optimisation, genetic algorithms, random search, heuristics, simulation

I N T R O D U C T I O N

Packing problems are optimisation problems that are concerned with finding a good arrangement of multiple items in larger containing regions (objects). The usual objective of the allocation process is to maximise the material utilisation and hence to minimise the "wasted" area. This is of particular interest to industries involved with mass-production as small improvements in the layout can result in savings of material and a considerable reduction in production costs. The development of an algorithm to solve an industrial packing problem clearly must consider the complexity of the problem, determined by the geometry of the objects and the constraints imposed. In addition the algorithm must be easy to adapt to the present competitive market with frequent product introductions, changing product designs and "shorter time to market" strategy. The flexibility achieved by manual packing is no longer a competitive solution due to high labour and liability costs. Conventional automated packing methods do not offer this flexibility, since they are mostly tailored to a particular packing task (Hinxman, 1980; Satin, 1983; l-I~sler and Sweeney, 1991). This calls for a new approach to packing problems, which implements automation, but also maintains the flexibility which is offered by manual composition of packing layouts. One of the main aspects in the development of flexible packing systems is the integration of intelligent search processes in order to find good packing patterns. Intelligent search processes such as genetic algorithms are highly flexible since they describe the packing problem in the form of general search principles rather than a set of special placement rules. Our work is concerned with a two-dimensional packing problem frequently encountered in the wood-, glass- and paper industry. The problem consists of packing rectangular items onto a rectangular object while minimising the used object space. The packing process has to ensure that there is no overlap between the items, which are allowed to rotate by 90 ° . So far only a few researchers have applied genetic algorithms to this problem type. Genetic algorithms for packing problems mainly concentrate on guillotineable packing problems (KrOger, 1995; Andr~ls, 1996) and bin-packing (Hwang, 1992; Falkenauer, 1994). Smith (1985) developed an order-based genetic for a rectangle packing problem, where the orientation of the items is fixed. The genetic algorithm by Kr6ger et al. (1991) includes rotation and is based on a tree structure to encode the problem. Since its performance is compared to well-known packing heuristics, a relative comparison with our technique is possible. 0360-8352/99 - see front matter © 1999 Elsevier Science Ltd. All rights reserved, PII: S0360-8352(99)00097-2

A GA methodology for the scheduling of yarn-dyed textile production

Hsi-Mei Hsu a,*, Yai Hsiung b,1, Ying-Zhi Chen a,2, Muh-Cherng Wu a,3

a Department of Industrial Engineering and Management, National Chiao Tung University, Hsin-Chu, Taiwan, ROCb Department of Information Management, Ta Hwa Institute of Technology, Hsin-Chu, Taiwan, ROC


Keywords:SchedulingSequence-dependent setupMulti-stageTextileGenetic algorithmGroup-delivery

a b s t r a c t

This paper presents a scheduling approach for yarn-dyed textile manufacturing. The scheduling problemis distinct in having four characteristics: multi-stage production, sequence-dependent setup times, hier-archical product structure, and group-delivery (a group of jobs pertaining to a particular customer ordermust be delivered together), which are seldom addressed as a whole in literature. The scheduling objec-tive is to minimize the total tardiness of customer orders. The problem is formulated as a mixed integerprogramming (MIP) model, which is computationally extensive. To reduce the problem complexity, wedecomposed the scheduling problem into a sequence of sub-problems. Each sub-problem is solved bya genetic algorithm (GA), and an iteration of solving the whole sequence of sub-problems is repeateduntil a satisfactory solution has been obtained. Numerical experiment results indicated that the proposedapproach significantly outperforms the EDD (earliest due date) scheduling method—currently used in theyarn-dyed textile industry.


1. Introduction

Yarn-dyed textiles are distinct in their manufacturing processesin which yarn must be dyed before weaving, while most other tex-tiles are first woven and then dyed. A yarn-dyed textile product, forexample a shirt, contains several patterns cloths. A pattern clothmanifests itself by a particular pattern of colors. In a colorful shirt,its sleeve may be a single-color pattern while its pocket may be athree-color pattern. A three-color pattern is composed of three dif-ferent color yarns, with each color yarn being individually dyed.Only when the three different color yarns have been dyed, theycould be weaved into the three-color pattern cloth.

Group-delivery is an essential characteristic in the dyeing pro-cess. Referring to the shirt shown in Fig. 1, we have five differentcolor yarns to be dyed in the dyeing stage. To weave each patterncloth, all its composing yarns have to be delivered to the weavingmachine in a group manner. That is, only when all the composingyarns of a particular pattern cloth arrive at the weaving machine,can the weaving of the pattern cloth be carried out.

Likewise, group-delivery is also an essential characteristic in theweaving process. See the shirt shown in Fig. 1, we have three pat-tern-cloths to be woven. For effectively making the shirt, the three

pattern-cloths also have to be delivered in a group manner. That is,only when all the three pattern-cloths have shipped to the down-stream shirt-maker, can the shirt-maker starts to manufacture theshirt.

In addition, the dyeing process is distinct in having a setupdependency characteristic. Before dyeing a yarn, we need to cleanthe dyeing tank—the machine that processes the yarn to be dyed.The clean time (setup time) required to prepare for dyeing a com-ing job can be different, dependent upon the colors of the comingyarn and the one just finishing dyeing. Consider two consecutivedyeing jobs. If the preceding job is dark-color (e.g. black) and thefollowing one is light-color (e.g. yellow), then we need a thoroughcleaning for the dyeing tank. That is, before dyeing the light-colorjob, the dark-coloring agent in the tank should be completely re-moved. In contrast, if the preceding job is light-color and the fol-lowing one is dark-color, then we need only a rough cleaning forthe dyeing tank. The time required for a thorough cleaning is muchlonger than that for a rough cleaning. This feature indicates thatthe dyeing process is sequence-dependent in setup time.

In summary, the manufacturing of the yarn-dyed textiles essen-tially involves two consecutive production processes—dyeing andweaving. These two processes are distinct in three points: (1)group-delivery in the dyeing process, (2) group-delivery in theweaving process, and (3) sequence-dependent in the dyeing pro-cess. To our knowledge, scheduling problems concerning thesethree features as a whole have not been examined in literature.

This paper formulated the scheduling problem for the yarn-dyed textile manufacturing process as a mixed integer program,and developed a genetic algorithm based approach to solve the


* Corresponding author. Tel.: +886 3 5731761.E-mail addresses: [email protected] (H.-M. Hsu), [email protected]

(Y. Hsiung), [email protected] (Y.-Z. Chen), [email protected](M.-C. Wu).

1 Tel.: +886 3 5927700x2751; fax: +886 3 5923957.2 Tel.: +886 3 5731761.3 Tel.: +886 3 5731913.

Expert Systems with Applications xxx (2009) xxx–xxx




ARTICLE IN PRESS

Please cite this article in press as: Hsu, H.-M., et al. A GA methodology for the scheduling of yarn-dyed textile production. Expert Systems with Applications(2009), doi:10.1016/j.eswa.2009.04.075







YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

Computers and Mathematics with Applications 53 (2007) 1840–1846www.elsevier.com/locate/camwa

A fuzzy neural network model for predicting clothing thermalcomfort

Xiaonan Luoa, Wenbang Houa,∗, Yi Lib, Zhong Wangb

a Computer Application Institute, Sun Yat-sen University, Guang Zhou 510275, Chinab Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hong Kong

Received 3 November 2005; received in revised form 29 June 2006; accepted 11 October 2006

Abstract

This paper presents a Fuzzy Neural Network (FNN) based local to overall thermal sensation model for prediction of clothingthermal function in functional textile design system. Unlike previous experimental and regression analysis approaches, this modeldepends on direct factors of human thermal response — body core and skin temperatures. First the local sensation is predicted by aFNN network using local body part skin temperatures, their change rates, and core temperature as inputs; then the overall sensationis predicted. This is also performed by a FNN network. The FNN networks are developed on the basis of the Feed-Forward Back-Propagation (FFBP) network; the advantage of using fuzzy logic here is to reduce the requirement of training data. The simulationresult shows a good correlation between predicted and the traditional experimental data.c© 2007 Elsevier Ltd. All rights reserved.

Keywords: Thermal sensation; Functional textile design; Fuzzy neural network; Feed-forward back-propagation; Clothing thermal comfort

1. Introduction

Today numerous consumers consider thermal comfort to be one of the most significant attributes when purchasingtextile and apparel products, so there is a need to develop a functional garment CAD system. In recent years, manytextile thermal function models and simulation systems have been developed [1–3]. They can simulate human thermalphysiological status and clothing heat and moisture transfer processing for designated arbitrary garment constructionsand thermal environments.

Because the human–clothing environment is a transient and non-uniform thermal environment, up to now there hasbeen no appropriate thermal comfort model to evaluate clothing thermal comfort. The existing literatures on humanthermal sensation and comfort are generally focused on steady-state and uniform conditions. Representatives areFanger’s PMV (Predicted Mean Vote) model [4] and Gagge’s two-node model with its indices of TSENS (ThermalSensation) and DISC (Thermal Discomfort) [5]. They are the basis of ASHRAE Standard 55-1992 and ISO EN 7730Standard. There are also works addressing transient and non-uniform conditions separately [6,7]. The above modelsare usually aimed at representing relationships between environment conditions and human thermal responses. In

∗ Corresponding author. Tel.: +86 20 3402 2313.E-mail addresses: [email protected] (X. Luo), [email protected] (W. Hou), [email protected] (Y. Li).

0898-1221/$ - see front matter c© 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.camwa.2006.10.035

http://www.elsevier.com/locate/camwa




http://dx.doi.org/10.1016/j.camwa.2006.10.035


A framework of E-SCM multi-agent systemsin the fashion industry

Wei-Shuo Loa,�, Tzung-Pei Hongb, Rong Jengc

aDepartment of Public Finance, Meiho Institute of Technology, Ping-Tung, TaiwanbDepartment of Computer Science and Information Engineering, National University of Kaohsiung, Kaohsiung, Taiwan

cDepartment of Information Management, I-Shou University, Kaohsiung, Taiwan

Received 7 September 2006; accepted 1 September 2007


Abstract

The fashion industry’s supply chain is full of uncertainty and unpredictability. Thus, building an intelligent system to

effectively capture the requirements of customers and help manage the supply chain is very important. Typical quick

response (QR) systems have been broadly used in the fashion industry to serve as a way of maintaining an efficient supply

chain management (SCM). The original functions of a QR system cannot, however, completely overcome the challenge of

quickly satisfying the requirements of customers with effective customer relationship and quality of service. In this paper,

we have integrated the typical management information system (MIS) development procedure with that of an e-fashion

SCM multi-agent system. Some related research and reports from different countries have been thoroughly surveyed in

order to find possible IT and non-IT methods for use in the SCM of fashion retailers. This paper thus provides an

electronic fashion SCM system by adopting the techniques of the Semantic Web and multiple agents. The proposed system

can integrate different information technologies to make its behavior more intelligent and to catch more useful information

from customers. Its implementation also considers some practical issues in the fashion retailing SCM.


Keywords: Fashion industry; E-SCM system; Semantic web; Multiple agents

1. Introduction

The fashion industry has faced more and fasterchanges in recent years due to the differentrequirements of customers and the variations ofglobal economic environments. This industry usual-ly needs to produce or provide various, complex,

and fashionable textile products, such as fashionouterwear, fashion wear, indoor and outdoorsportswear, fashion textiles, interior textiles, textiledesign, working wear, and so on.

The different requirements of textile productsmay arbitrarily appear at any fashion market. Theserequirements and the order information are thendelivered from each sale company of textileproducts to its upstream sale company or manu-facturers. The upstream sale companies or manu-facturers then refer to the specifications, prices,colors, and time the orders were received to finish

ARTICLE IN PRESS



doi:10.1016/j.ijpe.2007.09.010

�Corresponding author. Tel.: +886 8 7799821x8500;

fax: +8868 7788118.

E-mail addresses: [email protected] (W.-S. Lo),

[email protected] (T.-P. Hong), [email protected] (R. Jeng).


dx.doi.org/10.1016/j.ijpe.2007.09.010




A fashion mix-and-match expert system for fashion retailers usingfuzzy screening approach

W.K. Wong a,*, X.H. Zeng b, W.M.R. Au a, P.Y. Mok a, S.Y.S. Leung a

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kongb College of Information Engineering, Dong Hua University, Shanghai 200051, China

Abstract

In today’s fashion retailing business, providing ‘‘fashion mix-and-match” or ‘‘fashion coordination” recommendations is a ‘must’ strat-egy to enhance customer service and improve sales. In this study, a fashion mix-and-match expert system is developed to provide custom-ers with professional and systematic mix-and-match recommendations automatically. The system can capture the knowledge and emulatethe decisions of fashion designers on apparel coordination and its knowledge base can store the literal form of information. A set of attri-butes of the apparel for coordination are identified and formulated; their corresponding importance is also defined with designers’ opin-ions using ordered weighted averaging operators. The Fashion Coordination Satisfaction Index is devised and computed using the fuzzyscreening approach to represent the satisfaction degree of the coordinating pairs of apparel product items. The experimental results dem-onstrate that the proposed system can generate effective mix-and-match recommendations and is now integrated with a smart dressingsystem used effectively in a fashion chain store company in Hong Kong.� 2008 Elsevier Ltd. All rights reserved.

Keywords: Expert systems; Fuzzy screening; Multi-criteria decision-making

1. Introduction

Customer service plays a vitally important role in today’sfashion retailing business. Strategies like private sales andVIP memberships are used by nearly every fashion retailerto enhance customers’ brand loyalty. Providing mix-and-match recommendations is therefore a ‘must’ strategy forretailers to enhance customer service and improve sales.Mix-and-match recommendations are traditionally givenby individual sales personnel based on their experienceand/or designers’ suggestions, by either showing customersphotos in the product catalogues or locating the matchingproducts on racks. With the recent invention of a smartdressing system by the authors, fashion items carried by cus-tomers can automatically be detected with mix-and-matchrecommendations shown in real-time (Hkpolyu, 2006; Itc,2007; Wong, Leung, & Mok, 2006a, Wong, Leung, &

Mok, 2006b). Such an invention revolutionizes fashionretail operations and enhances customer service. The smartdressing system makes use of the technology of radio fre-quency identification (RFID) to detect items brought intoa fitting room or placed in front of a dressing mirror. Eachproduct item bears a RFID tag. When an item is picked intothe fitting room or placed in front of a dressing mirror, theproduct will be immediately detected and transmitted to thesystem through the antennae and reader. The mix-and-match database of the invention will then deliver recom-mendations to the customer through a touch-screen LCDmonitor or projected screen (For details, please see refer-ences Hkpolyu, 2006; Itc, 2007; Wong et al., 2006a,2006b.). The recommendations stored in this mix-and-match database are provided by the proposed fashionmix-and-match expert system in this paper for emulatingfashion designers to generate and export the mix-and-matchrecommendations to the database.

In the past, fashion designers create a collection of fash-ion products and also determined how to coordinate and


doi:10.1016/j.eswa.2007.12.047

* Corresponding author. Tel.: +852 2766 6471; fax: +852 2773 1432.E-mail address: [email protected] (W.K. Wong).






A decision support tool for apparel coordination throughintegrating the knowledge-based attribute evaluation expert system

and the T–S fuzzy neural network

W.K. Wong a,*, X.H. Zeng b, W.M.R. Au a

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kongb College of Information Engineering, Dong Hua University, Shanghai 200051, China

Abstract

In today’s competitive fashion retailing business, providing ‘‘mix-and-match” or ‘‘fashion coordination” recommendations canenhance customer service, brand loyalty and improve sales. In this study, we propose a decision support tool for fashion coordinationthrough the integration of the knowledge-based attribute evaluation expert system and the Takagi–Sugeno fuzzy neural network(TSFNN). A set of attributes of the apparel items for coordination are identified and formulated. The evaluation of these attributescan be accomplished by a knowledge-based expert system which can handle the difficulty of processing linguistic and categorical infor-mation effectively. A fuzzy clustering technique and a new hybrid learning algorithm combining the PSO and GA techniques are pro-posed to reduce the coordination rules and the training time for the TSFNN. The experimental results show that rules reduction canshorten the TSFNN training time while keeping a very satisfactory and low MSE value. The proposed hybrid algorithm outperformsthe Back Propagation, the Genetic Algorithm, and the Particle Swarm Optimization. The apparel pairs recommended by the decisionsupport tool are now integrated with a smart dressing system of a fashion retailing company in Hong Kong and practically used.� 2008 Elsevier Ltd. All rights reserved.

Keywords: Fuzzy neural networks; Particle swarm optimization; Genetic Algorithm; Decision support; Back Propagation

1. Introduction

Customer service plays a vitally important role intoday’s fashion retailing business. Strategies like privatesales and VIP membership are used by nearly every retailerto enhance customers’ brand loyalty. Providing apparelcoordination, also known as ‘‘mix-and-match” recommen-dations, is therefore a ‘must’ strategy for retailers toenhance customer service and improve sales. The word‘coordination’ means ‘‘harmonious combination or inter-action, as of functions or parts” (Dictionary.com, 2006).We can further say that apparel coordination means tocombine in a harmonious or interesting way, as articles

of clothing in an ensemble. Mix-and-match recommenda-tions are traditionally given by individual sales personnelbased on their experience and/or designers’ suggestions,by either showing customers photos in the product cata-logue or locating the coordination products on racks. Withthe recent invention of a smart dressing system by theauthors (Wong, Leung, & Mok, 2006a,b) fashion items car-ried by customers can automatically be detected with mix-and-match recommendations shown in real time. Such aninvention revolutionizes fashion retail operation andenhances customer service (Hkpoly, 2006; Itc, 2007). Thesmart dressing system makes use of the technology ofRadio Frequency Identification (RFID) to detect itemsbrought into a fitting room or placed in front of a dressingmirror. Each product item bears an RFID tag. When anitem is brought into a fitting room or placed in front of adressing mirror, the product will be immediately detected


doi:10.1016/j.eswa.2007.12.068

* Corresponding author.E-mail address: [email protected] (W.K. Wong).






YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

IJCST13,2

106

International Journal of ClothingScience and Technology,Vol. 13 No. 2, 2001, pp. 106-114.# MCB University Press, 0955-6222

Received September 1999Revised January 2001Accepted January 2001

Total handle evaluation fromselected mechanical propertiesof knitted fabrics using neural

networkShin-Woong Park, Young-Gu Hwang and Bok-Choon Kang

Department of Textile Engineering, Inha University,South Korea, and

Seong-Won YeoDepartment of Electrical Engineering, Inha University, South Korea

Keywords Mechanical properties, Fuzzy logic, Neural networks, Knitwear, Simulation

Abstract This paper concentrated on the objective evaluation of total hand value in knittedfabrics using the theory of neural networks and the comparison of two methods. For the objectiveevaluation of overall hand feeling in knitted fabric, 47 kinds of weft-knitted and warp-knittedfabrics were manufactured. The optimum construction of neural networks was investigatedthrough the change of layer and neuron number. For the comparison of the two methods, asubjective test was carried out. Two techniques, KES-FB system and neural network appliedsimulator, were compared using nine randomly selected knitted fabrics. These fabrics were usedto show that the neural network adapted simulation method was in good agreement withsubjective test results.

1. IntroductionFabric hand has been considered as one of the most important performanceattributes of textiles intended for use in garments. Methods for predicting knittedfabrics in apparel manufacture from its physical, mechanical and dimensionalproperties have been investigated (Kawabata, 1980; Gong, 1995; Park and Hwang,1999). In previous papers (Park and Hwang, 1999; Park et al., 1996; 1997; 1998), wehave published data regarding fuzzy predicting model of woven, warp-knittedand double weft-knitted fabrics and have studied a fuzzy applied method and aneural network applied simulation. A subjective test gave a good agreement withresults of the fuzzy model and the neural network prediction simulator rather thanthat of KES-FB system. But fuzzy method is not a simulator, but a mathematicalmodeling equation. We have investigated the neural network applied total handleevaluator, which is a method setting fuzzy mathematical results as a targetoutput. Therefore, it was recognized that there is a need to establish an exactautomatically hand evaluation system, being based on subjective test results.

In this paper, we extend our investigations to the objective hand evaluationof knitted fabrics used for fall and winter, which included various type of fibersand their constructions of single, double and warp-knitted fabrics.

The current issue and full text archive of this journal is available athttp://www.emerald-library.com/ft

The authors wish to thank the Inha University and Industrial Technology Research InstituteFoundation for the financial support provided for this study.

477

Predicting the Performance of Fabrics in Garment Manufacturing withArtificial Neural Networks

R. H. GONG

Department of Textiles, University of Manchester Institute of Science and Technology, Manchester M60 1QD,United Kingdom

Y. CHEN

Department of Silk Textile Engineering, Suzhou Institute of Silk Textile Technology, Suzhou,People’s Republic of China

ABSTRACT

Neural networks are used to predict the performance of fabrics in clothing manufac-turing. The predictions are based on fabric mechanical properties measured on the KES-FBsystem. The influence of the number of input and hidden nodes on the convergence speedand the prediction accuracy are investigated. Tests indicate that these artificial neural networks are effective for predicting potential problems in clothing manufacturing.

In recent years, garment manufacturing processes havebecome more and more automated, the consumer market is

increasingly sophisticated, demanding more choices, andthere is an expanding variety of fabrics that manufacturershave to process into different styles of garments. Qualitycontrol in garment manufacturing is therefore becomingmore difficult. In the last decade, greater attention has been

paid to the influence of fabric properties in garment pro-cessing. In order to obtain good quality products with highefficiency production lines, clothing companies have estab-lished advanced laboratories to measure fabric propertiesfor controlling fabric quality, production processes, andgarment quality [2, 6]. Many researchers have also beenworking on the relationships between fabric properties andperformance in clothing production in order to predict afabric’s performance on the basis of its properties, espe-cially mechanical properties under low stress as measuredby the KES-FB system [3.4,5,8,9,10.11,16,17,181. TableI summarizes the fabric property parameters measured bythe tcES-~ system and the areas where these properties areexpected to be influential in clothing manufacturing. Sev-eral equations have also been developed for calculatingtailorability, but they can only provide guidelines and can-not offer specific predictions of how a fabric might performin garment manufacturing. The relationships between fabricperformance and properties are very difficult to describequantitatively by traditional mathematics or mechanics dueto the nonlinearity of the parameters and the large numberof variables involved. This, however, seems to be an idealsituation for the application of artificial neural networks(ANN), which are developed to tackle problems with largenumbers of nonlinear variables.

An artificial neural network is one of the new intelli-

gence technologies for data analysis. It imitates the be-

havior of biological neural networks to &dquo;learn&dquo; a subjectfrom the data provided to it. The ANN has been success-fully used in areas where a large number of factorscontribute to the eventual outcome. but precise relation-ships between these various factors and their outcomescannot be defined, for example, medical diagnoses andcredit evaluation in banking. Attempts have recentlybeen made to apply the ANN technique to textiles.

Ramesh et al. used an ANN to predict yarn tensile prop-erties based on yarn processing and material variables[ 14]. Pynckels et al. used an ANN to determine the spin-ning performance of fibers from fiber properties [ 13].Cheng and Adams predicted yam strength according tofiber properties with ANNS [ 1 ]. Sette et al. applied thistechnique in the assessment of fabric set marks and

carpet wear [ 15]. The applications of ANNS in these areasshow great promise because an ANN can deal with thenonlinearity of problems, detect patterns and relation-ships in the data, and interpret information from tens ormore variables.

-

’

In this paper, we investigate the use of artificial neural

networks to predict fabric performance in garment manufx-ture and the appearance of the made-up gatment. The purposeof this work is to verify the possibility of using ANN techniquesin this area and the effects of different ANN architectures on

&dquo;

their training speed and prediction accuracy.

Experimental ~ ,

We selected 32 fabrics with a variety of fiber compo-sitions and fabric weaves, all made by_the industrial

at The Hong Kong Polytechnic University on October 6, 2009 http://trj.sagepub.comDownloaded from

http://trj.sagepub.com

Ž .Computers in Industry 43 2000 1–10www.elsevier.nlrlocatercompind

Optimization of spreading and cutting sequencing modelin garment manufacturing

W.K. Wong a,), C.K. Chan a, W.H. Ip b

a Institute of Textiles and Clothing, The Hong Kong Polytechnic UniÕersity, Hunghom, Kowloon,Hongkong, People’s Republic of China

b Department of Manufacturing Engineering, The Hong Kong Polytechnic UniÕersity, Hunghom, Kowloon,Hongkong, People’s Republic of China

Received 1 January 1999; received in revised form 1 November 1999; accepted 1 March 2000

Abstract

Many researches on the machine scheduling and flowshop sequencing problem have been conducted by using geneticŽ .algorithms GA . Recently, GAs have been applied to the scheduling and line balancing problem in garment manufacturing.

These applications have only been confined to the sewing operations. This paper presents a spreading and cutting sequencingŽ .SCS model using GA to solve the sequencing problem of the computerized cutting system used in the garment industry.The comparison results obtained between the actual production cycle and the proposed model using GA indicate that GA isan appropriate and effective technique to solve the problem. q 2000 Elsevier Science B.V. All rights reserved.

Keywords: Optimization; Genetic algorithms; Sequencing; Computerized fabric-cutting system

1. Introduction

Many studies have been made on the applicationsŽ .of genetic algorithms GA to the flowshop problems

w x w x w x2 , workshop problems 3 , line balancing 11 andw xtravelling salesman problems 9 of different indus-

w xtries. Goldberg 6 also stated that GA is a robustapproach, that is, a specific method better than GAsmay exist for a particular instance, but on averageGAs are never bad for solving a wide variety ofproblems. Within recent years, researches have been

) Corresponding author. Tel.: q852-27666471; fax: q852-27731432.

E-mail address: [email protected]Ž .W.K. Wong .

emphasized specifically on solving the schedulingand line balancing problem of sewing lines in thegarment manufacturing industry. Many methods tosolve these problems are exact methods, e.g. tradi-tional linear programming, the branch and boundapproach, which are time and place consuming. Ap-proximate methods, such as heuristic methods, local

w ximprovement, etc. are also employed. Chen et al. 1presented a heuristic solution procedure based on thesimulated annealing concept to solve the dailyscheduling problem in the make-to-order garment

w xindustry. Dessouky et al. 10 proposed the schedul-ing of multi-stage flowshops with identical jobs of asewing line of a garment manufacturer by using

w xbranch and bound approach. Chan et al. 7 presentedthe line balancing problem of a sewing line by using

w xGA. Lo 8 presented the scheduling problem of a

0166-3615r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0166-3615 00 00057-9

DOI 10.1007/s00170-004-2161-0

O R I G I N A L A R T I C L E

Int J Adv Manuf Technol (2005) 27: 152–158

W.K. Wong · C.K. Kwong · P.Y. Mok · W.H. Ip · C.K. Chan

Optimization of manual fabric-cutting process in apparel manufactureusing genetic algorithms

Received: 28 November 2003 / Accepted: 1 March 2004 / Published online: 26 January 2005© Springer-Verlag London Limited 2005

Abstract In apparel manufacturing, experience and subjectiveassessment of production planners are used quite often to planthe production schedules in their fabric-cutting departments. Thequantities of cut-pieces produced by fabric-cutting departmentsbased on these non-systematic schedules cannot fulfil the cut-piece requirements of the downstream sewing lines and mini-mize the makespan. This paper proposes a genetic algorithms(GAs) approach to optimize both the cut-piece requirementsand the makespan of the conventional fabric-cutting departmentsusing manual spreading and cutting methods. An optimizationmodel for the manual fabric cutting process based on GAs wasdeveloped. Two sets of production data were collected to vali-date the performance of the model and the experimental resultswere obtained. From the results, it can be found that both themakespan and cut-piece fulfilment rates are improved in whichthe latter is improved significantly.

Keywords Fabric-cutting · Genetic algorithms ·Production scheduling

Nomenclature

X Job (fabric lay)N Maximum number of jobsi Job setup (spreading) order and i = 1, 2, . . ., Nj Job processing (cutting) order and j = 1, 2, . . ., Nσi, σj Setup and processing sequence of jobsφ Production order of job X and φ = 1, 2, . . ., PO

W.K. Wong (�) · C.K. ChanInstitute of Textiles and Clothing,The Hong Kong Polytechnic University,Hunghom, Kowloon, Hong KongE-mail: [email protected].: +852-27666471Fax: +852-27731432

C.K. Kwong · P.Y. Mok · W.H. IpDepartment of Industrial System and Engineering,The Hong Kong Polytechnic University,Hunghom, Kowloon, Hong Kong

χ Quantity of garments of job Xϕ Length of fabric lay of job Xs(Xi) Setup (spreading) time of job Xi

c(Xj ) Processing (cutting) time of job Xj

m Number of spreading tables in the fabric-cuttingdepartment

1 Introduction

In apparel manufacturing, fabric cutting is done before assembly.The performance of the cutting department, which is generallyneglected by manufacturers, is a critical factor on the smoothnessof downstream operations in sewing lines and hence the overallefficiency of the apparel manufacturing plant. Since the late 80s,some apparel manufacturers have implemented the computerizedfabric-cutting systems in their apparel manufacturing process.The demands on fabric-cutting departments for greater accuracy,faster throughput, larger fabric and labour savings have driventhe adoption of computerized cutting systems.

However, many manufacturers still rely on the manualmethod for the fabric-spreading and cutting operations in theirfabric-cutting department. Before daily spreading and cuttingoperations start, the production planners of cutting departmentsneed to plan the production (spreading and cutting) schedule soas to minimize the idle time of operatives and fulfil the fabriccut-piece requirements from different sewing production lines.The production planning is normally based on their experienceand subjective assessment which is not a systematic methodand an optimal schedule cannot be obtained. As a result, idletimes occur on the spreading and cutting operatives which in turnincreases the overall makespan of cutting departments. The cut-piece quantities produced cannot fulfil the different requirementsof each downstream sewing production line.

As most of the apparel manufacturers and researchers em-phasize the importance of sewing process, research has beendone to improve the operation of sewing lines. However, the pro-ductivity of cutting departments, which plays a significant role

ELSEVIER European Journal of Operational Research 88 (1996) 165-181

EUROPEAN JOURNAL

OFOPERATIONAL RESEARCH

T h e o r y a n d M e t h o d o l o g y

On genetic algorithms for the packing of polygons

S t e f a n J a k o b s

RWTH Aachen, Lehrstuhl C J~r Mathematik, Templergraben 55, 17-52062 Aachen, Germany

Received June 1993

Abstract

A genetic algorithm for placing polygons on a rectangular board is proposed. The algorithm is improved by combination with deterministic methods.

Keywords: Optimization; Genetic algorithms; Mathematical programming; Adaptive processes; Packing problems

1. Introduction and motivation

In the steel industry problems frequently occur when the need to stamp polygonal figures from a rectangular board arises. The aim is to maximize the use of the contiguous remainder of the board. Similar problems exist in the textile industry, when clothes are cut out of a rectangular piece of material.

In order to solve these problems let us consider the following simpler approach. Given a finite number of rectangles ri, i = 1 , . . . , n, and a rectangular board, an orthogonal packing pattern requires by definition a disjunctive placement of the rectangles on the board in such a way that the edges of r i are parallel to the x- and y-axes, respectively. The computation of the orthogonal packing pattern with minimal height is called orthogonal packing problem (OPP).

Baker, Coffman and Rivest propose an heuristic for the orthogonal packing problem; in addition they present an upper bound for the height of the packing pattern [2]. A recent survey on packing problems and their respective heuristics

is given in [16]. The extension from rectangles to polygons can be realized in several ways. The first method places the polygons directly on the board and then the algorithm optimizes locally by means of shifts and rotations [23]. A second approach places two or three polygons in a cluster. The clusters are then placed on the board [1].

In this article we use another approach, namely an evolutionary algorithm. There are three main classes in this approach, each of which is independently developed. The first class is called evolutionary programming (EP). L.J. Fogel, Owens, and Walsh were the first to develop the EP-algorithms [5]. D.B. Fogel has recently improved this approach [6]. The second class was developed by Rechenberg and Schwefel. They called their approach evolutionary strategies (ES) [17-20]. Fi- nally, Holland developed the so called genetic algorithm (GA) [12]. The genetic algorithm has been perfected by De Jong [13] and Goldberg [9].

The paper is organized as follows. It begins by explaining the problem and its complexity. In the next section the data structure and its transformation into a packing pattern are described. Sec-

0377-2217/96/$9.50 © 1996 Elsevier Science B.V. All rights reserved SSDI 0377-2217(94)00166-A

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 31, NO. 6, NOVEMBERDECEMBER 1995 , 1371

New Developments for Seam Quality Monitoring in Sewing Applications

J. Lewis Dorrity, Member, IEEE

Abstract-The automation of sewing machinery requires that sensing be added to insure quality of the stitching. This paper describes the use of piezoelectric technology to indirectly monitor sewing thread consumption which allows an inference of seam quality. An inexpensive microprocessor system is employed to monitor the sensor. The technology developed is reliable and applicable to a wide range of sewing machinery.

I. INTRODUCTION

UTOMATION of any process always requires that more A sensing of the process be done electronically or mechan- ically in order to replace the observations made by a human operator. Too often engineers begin to consider replacing the actions of an operator by robotics or fixed automation without first considering the inspection duties inherently included in the job. These duties are frequently taken for granted and are difficult or very expensive to replace with sensor technology. The vision of an operator, for example, can be replaced but the image analysis is often complex and difficult to replace.

Research by the author in this area was begun several years ago under a research contract with the Defense Logistics Agency [I] which is responsible for production of uniforms for the military services. The purpose of the research was to promote automation in apparel manufacturing. The work was to develop new technologies which would permit automation of sewing operations and thus make apparel production in the United States less labor intensive and therefore more competitive. The initial interest was in the common lockstitch machine and the sewing of denim woven materials. More recently, the National Textile Center has funded research to expand the knowledge and applicability of this new technology. The focus has shifted to other types of stitches and other materials including knits.

Stitch formation consists of the interlacing of one or more threads which penetrate multiple layers of textile fabric, thereby holding them together. The simple single needle chainstitch requires only one thread supply while a double needle chainstitch requires three thread packages. Other stitches which are combinations of the fundamental stitches

Paper PID 95-23, approved by the Textile, Fiber and Film Industry Committee of the IEEE Industry Applications Society for presentation at the 1995 IEEEDAS Annual Textile, Fiber and Film Industry Technical Conference, Charlotte, NC, May 2 4 . Manuscript released for publication May 14, 1995.

The author is with the School of Textile & Fiber Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0295 USA.

IEEE Log Number 9414388.

require even more yarn packages. The machines designed to form the stitches are ingenious and highly mechanical. Newer machines do employ electronics to monitor and control various aspects of machine operation, but the stitch formation remains mechanical.

To automate the textile sewing operation, some sensing of whether the machine is performing properly is in order and necessary. Research of others such as Matthews & Little [2] and Murray [3] have worked to that end. Without such on-line monitoring for quality, one must rely on inspection further down the process stream to catch problems. In fast modern processing, a great amount of off-quality product may be produced before the inspection process finds and stops it.

11. THREAD MOTION RATIO

Early research done for the Defense Logistics Agency on the single-needle lockstitch machine [4] showed that the time of thread motion changed proportionally with the machine cycle time. In order to reduce the effect of speed as a source of error, the normalized ratio TMR was calculated as follows:

TMR = tthread motionltmachine cycle X 100%.

TMR is actually the equivalent of the ratio of the average thread velocity divided by the average thread velocity during the intermittent motion. The thread is in motion from 15% to 35% of the cycle time depending on the stitch type and the particular thread in that stitch.

The thread consumption (G) is calculated as follows:

c = io V(t)dt

where { t o : t l } is a thread motion interval and { t o : t 2 } is a machine cycle. If the velocity, vi is assumed constant over aby interval {t , : t b }

Let vint be the average velocity on the interval { t o : t l } and vaVg be the average velocity over the machine cycle {to : t z } . These velocities are related to time and thread consumption

0093-9994/95$04.00 0 1995 IEEE

Authorized licensed use limited to: Hong Kong Polytechnic University. Downloaded on September 25, 2009 at 02:58 from IEEE Xplore. Restrictions apply.

31

Neural Network Predictions of Human Psychological Perceptionsof Clothing Sensory Comfort

A. S. W. WONG, Y. LI, AND P. K. W. YEUNG

Institute of Textiles and Clothing, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

P. W. H. LEEDepartment of Psychiatry, University of Hong Kong, Pokfulam, Hong Kong

ABSTRACTThe objective of this’paper is to investigate the predictability of clothing sensory

comfort from psychological perceptions by using a feed-forward back-propagation net-work in an artificial neural network (ANN) system. In order to achieve the objective, aseries of wear trials is conducted in which ten sensory perceptions (clammy, clingy, damp,sticky, heavy, prickly, scratchy, fit, breathable, and thermal) and overall clot hing comfort(comfort) are rated by twenty-two professional athletes in a controlled la ratory. Theyare asked to wear four different garments in each trial and rate the sensations above duringa 90-minute exercising period. The scores are were input into five different eed-forwardback-propagation neural network models, consisting of six different numbe rs of hiddenand output transfer neurons. Results showing a good correlation between redicted andactual comfort ratings with a significance of p < 0:001 for all five models ind icate overallcomfort performance is predictable with neural networks, particularly models with logsigmoid hidden neurons and pure linear output neurons. Models with a single log sigmoidhidden layer with fifteen neurons or three hidden layers, each with ten log sigmoid hiddenneurons, are able to produce better predictions than the other models for is particulardata set in the study.

In order to survive in the quickly changing, highlycompetitive clothing market, companies in textile and

clothing industries are searching for competitive advan-tage by understanding and meeting consumer needs anddesires. Various consumer research groups have reportedthat modem consumers consider comfort one of the most

important attributes in their purchase of textile and ap-parel products, so there is a need to develop a soundscientific understanding of the psychological perceptionof clothing comfort sensations.Up to now, there has been no one clear definition of

comfort, since this subjective feeling differs from personto person, but a lot of researchers have investigatedcomfort over the past years. For example, LaMotte( 1977) stated that physical comfort might be greatlyinfluenced by tactile and thermal sensations arising fromcontact between skin and the immediate environment [3].Slater ( 1986) defined comfort as &dquo;a pleasant state of

physiological, psychological and physical harmony be-tween a human being and the environment&dquo; 1101. Li

( 1986) defined comfort as a holistic concept, which is astate of multiple interactions of physical, physiological,

and psychological factors [7]. owever. these definitionsonly,, identify the factors influ cing human sensory per-ceptions; the relationships these factors and

overall comfort have not yet detennined.10 the past, many research rs of thermal and tactile

comfort have used traditional tatistical methods such as

cluster analysis and factor ana ysis [5, 6]. In earlier work( Wong et al. [ 1 ? ]. we develo d a linear model based ontraditional statistics to simulate huA~an psychologicalperception; of clothing senso comfort t!2j. We usedstati~tical factor analysis to i ntify independent factorsand Itheir relative contribution from ten sensory percep-tions. We identified three jor factors of moisture.tactile. and thermal-fit comfo . and constructed a linear

m(*l using these three factor’ and their contributions asweights to predict overall co fort perceptions. Compar-ing the predictions with the ctual comfort ratings, we

observed good agreement belween the two. indicatingthat overall clothing comfort be predicted by indi-vid441 sensory perceptions. IThe application of statistic methods in this research

has

~a number of limitations, however, including diffi-

at The Hong Kong Polytechnic University on October 6, 2009 http://trj.sagepub.comDownloaded from


375

Neural Network Prediction of Human PsychologicalPerceptions of Fabric Hand

C. L. HUI, T. W. LAU, AND S. F. NG

Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

K. C. C. CHAN

Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

ABSTRACT

Fabric hand is commonly adopted for assessing fabric quality and prospective perfor-mance in a particular end use. In general, fabric hand is primarily assessed subjectively.Subjective assessments treat fabric hand as a psychological reaction obtained from thesense of touch, based on the experience and sensitivity of humans. It is very difficult topredict such psychological perceptions of hand based on fabric properties. In this paper,we identify reliable sensory fabric hand attributes with correlated attributes of fabricproperties, and we attempt a novel approach for predicting sensory hand based on fabricproperties using a resilient back-propagation neural network. In this study, we assess fortywoven fabrics to determine twelve significant fabric properties and fourteen reliable

attributes of sensory hand. Our proposed system performs at a very low mean square errorafter fine tuning. Five extra woven fabrics are used to show that the performance of sucha prediction system closely agrees with subjective test results. Our proposed system canallow field practitioners to evaluate their fabrics more closely to match with customers’expectations.

Fabric hand is a generic term for the tactile sensationsassociated with fabrics that influence consumer prefer-ences [ 12]. It is basically a reflection of overall quality,consisting of a number of individual physical properties[24], and is the human response to touching, squeezing,rubbing, or otherwise handling a fabric [ 16]. It is com-

monly adopted for assessing fabric quality and prospec-tive performance in a particular end use.

In general, fabric hand can be assessed by subjectiveand objective methods. Subjective assessments treat fab-ric hand as a psychological reaction obtained by thesense of touch. It is a primary descriptive method basedon the experience and sensitivity of human beings. Ob-jective assessments attempt to predict fabric hand usinginstrumental data and sensory-instrumental relationships.

Theoretical approaches to subjective fabric hand sen-sory assessments have recently aroused great interest inthe area of clothing and textiles, and many researchershave looked for &dquo;world-famous&dquo; methodologies to trans-form subjective hand properties to objective measure-ments [38]. The motivation behind these works is due to

the different fabric sensory perceptions of individuals.Brand [2] is one of several researchers who commentedon differences between vocabularies of experts and un-trained judges of textile hand. Wauer [36] concluded thatthese differences are great enough to interfere with com-munications between experts and consumers. They mayuse the same adjective, say, &dquo;harsh,&dquo; to describe a handthat differs among individuals. Moreover, Brand [2]stated that, &dquo;Aesthetic concepts are basically people’spreferences and should be evaluated subjectively by peo-ple.&dquo; This differentiation has initiated much researchfocused on how to model subjective fabric hand objec-tively. Although many older techniques for evaluatingfabric hand did not use standards or proper psychologicalmethods, more recent approaches certainly do use stan-dard scales and measures. For example, the SpectrumMethod of Descriptive Hand Evaluation (Civille andDus) [5] is based on a set of fifteen-point intensity scalesfor twenty-one different attributes of fabric hand. Each ofthese intensity scales is anchored at several points byspecific fabric standards, i.e., physical references, so that

at The Hong Kong Polytechnic University on September 24, 2009 http://trj.sagepub.comDownloaded from


Computing, Artificial Intelligence and Information Technology

Neural network forecasting for seasonal and trend time series

G. Peter Zhang a,*, Min Qi b

a Department of Management, J. Mack Robinson College of Business, Georgia State University, 35 Broad Street, NW,

Atlanta, GA 30303, USAb Department of Economics, College of Business Administration, Kent State University, Kent, OH 44242, USA

Received 19 October 2001; accepted 8 August 2003

Available online 18 November 2003

Abstract

Neural networks have been widely used as a promising method for time series forecasting. However, limited em-

pirical studies on seasonal time series forecasting with neural networks yield mixed results. While some find that neural

networks are able to model seasonality directly and prior deseasonalization is not necessary, others conclude just the

opposite. In this paper, we investigate the issue of how to effectively model time series with both seasonal and trend

patterns. In particular, we study the effectiveness of data preprocessing, including deseasonalization and detrending, on

neural network modeling and forecasting performance. Both simulation and real data are examined and results are

compared to those obtained from the Box–Jenkins seasonal autoregressive integrated moving average models. We find

that neural networks are not able to capture seasonal or trend variations effectively with the unpreprocessed raw data

and either detrending or deseasonalization can dramatically reduce forecasting errors. Moreover, a combined detr-

ending and deseasonalization is found to be the most effective data preprocessing approach.

� 2003 Elsevier B.V. All rights reserved.

Keywords: Neural networks; Box–Jenkins method; Seasonality; Time series; Forecasting

1. Introduction

Many business and economic time series exhibit

seasonal and trend variations. Seasonality is a

periodic and recurrent pattern caused by factors

such as weather, holidays, repeating promotions,

as well as the behavior of economic agents (Hyl-

leberg, 1992). Although seasonal variations areperhaps the most significant component in a sea-

sonal time series, a stochastic trend is often ac-

companied with the seasonal variations and can

have a significant impact on various forecasting

methods. A time series with trend is considered to

be nonstationary and often needs to be made sta-

tionary before most modeling and forecasting

processes take place. Accurate forecasting of sea-

sonal and trend time series is very important for

effective decisions in retail, marketing, production,inventory control, personnel, and many other

business sectors (Makridakis and Wheelwright,

1987). Thus, how to model and forecast seasonal

and trend time series has long been a major re-

search topic that has significant practical implica-

tions.

* Corresponding author. Tel.: +1-404-651-4065; fax: +1-404-

651-3498.

E-mail address: [email protected] (G.P. Zhang).

0377-2217/$ - see front matter � 2003 Elsevier B.V. All rights reserved.

doi:10.1016/j.ejor.2003.08.037




Pergamon

Computers ind. EngngVol. 29, No. 1-4, pp. 513-517, 1995 Copyright © 1995 Elsevier Science Ltd

Printed in Great Britain. All rights reserved 0360-8352/95 $9.50 + 0.00

0360-8352(95)00126-3

Minmax Earliness/Tardiness Scheduling in Identical Parallel Machine System Using Genetic Algorithms

Runwei Cheng* Mitsuo Gen t Tatsumi Tozawa*

* Graduate School of Engineering Utsunomiya University, Utstinomiya 321, Japan

t Department of Industrial and Systems Engineering Ashikaga Institute of Technology, Ashlkaga 326, Japan

A b s t r a c t : In this paper, we address an earliness/tardiness scheduling problem in identical paral- ]el machine system with an objective of minimizing the maximum weighted absolute lateness. Genetic algorithms are applied to solve this problem. The performance of proposed procedure is compared with ex~ng heuristic procedure on randomly generated test problems. The results show that the prbposed approach performs well for this problem.

K e y words : Genetic algorithms, earll- ness/tardiness scheduling, identical parallel machine system and minmax optimization.

1 IntroduCtion

In this paper we discuss the application df genetic algorithms to earliness/tardiness scheduling problem in identical parallel machine system with an objective of minimizing the maximum weighted absolute lateness. This problem was firstly considered by Li and Cheng as follows [1]: there are a set of jobs associated with known processing times and weights, several parallel and identical machines, and a common due date that is not too early to constrain the scheduling decision. The objective is to find an op- timed job schedule so as to minimize the maximum weighted absolute lateness.

This kind of objective function is known as one of non-regular performance measures. In recent years, scheduling research involving non-regular performance measures has received much attention from practitioners as well as researchers to respond to the increasing competitive pressure in domestic and in- ternationul market. Baker and Scudder [2] have presented a comprehensive survey of earliness/tardiness scheduling. Recent overviews on parallel machine scheduling research are given by Cheng and Sin [3]. There are two non-regular performance measures commonly used in earliness/tardiness scheduling: reinsure and minmax. A reinsure problem attempts to minimize the sum of weighted absolute deviation of job completion time about the due date, i .e., to reduce customers' aggregate disappointment; CAIE 29:1/4-1I

while a mlnmax problem attempts to mlnlmile the m~ximum weighted absolute deviation of job completion time about the due date, i .e. , to reduce a customers maximum disappointment.

Li and Cheng have shown that minmax scheduling problem is NP-complete even for single machine system. Due to the intrinsic difficulty of the problem, search methods based upon heuristics are most promising for solving such problem. Li and Cheng have proposed two greedy heuristic based procedures to solve this problem.

In recent-years, a growing body of literature sug- gests the use of genetic algorithm as one of powerful heuristic search methods to solve combinatorial optimization problems [4]. Gupta at el. have solved the mlnsu.m scheduling problem in a single machine system using genetic algorithms [5]. Single machine schudeling problem just considers how to find out a best permutation of jobs with respect to someper- formance measures. As we know that there are two essential issues to be dealt for all kind of multiple machine scheduling problems:

• partition jobs to machines • sequence jobs for each machine

Because genetic algorithms are very effective at performing global search for combinatorial optimization problems, in this paper, we investigate how to apply genetic algorithms to solve minmax multiple machines scheduling problem. An extended permutation representation is adopted as the coding scheme for multiple machines scheduling problem, crossover and mutation are defined to adjust job partition among machines and job permutation within each machine. The performance of proposed procedure is compared with Li and Cheng's greedy heuristic on randomly generated test problems. The results show that the proposed approach performs well for this problem.

2 Problem and Assumpt ions

We consider the following multiple machines scheduling problem:

513


Fashion retail forecasting by evolutionary neural networks

Kin-Fan Au�, Tsan-Ming Choi, Yong Yu

Business Division, Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Kowloon, Hung Hom, Hong Kong

Received 28 August 2006; accepted 15 June 2007


Abstract

Recent literature on nonlinear models has shown that neural networks are versatile tools for forecasting. However, the

search for an ideal network structure is a complex task. Evolutionary computation is a promising global search approach

for feature and model selection. In this paper, an evolutionary computation approach is proposed in searching for the ideal

network structure for a forecasting system. Two years’ apparel sales data are used in the analysis. The optimized neural

networks structure for the forecasting of apparel sales is developed. The performances of the models are compared with the

basic fully connected neural networks and the traditional forecasting models. We find that the proposed algorithms are

useful for fashion retail forecasting, and the performance of it is better than the traditional SARIMA model for products

with features of low demand uncertainty and weak seasonal trends. It is applicable for fashion retailers to produce short-

term retail forecasting for apparels, which share these features.


Keywords: Forecasting; Evolutionary neural networks; SARIMA

1. Introduction

In fashion retailing, demand uncertainty isnotorious of creating many big challenges inlogistics management (Hammond, 1990). Followingthe fashion trend and market response, fashionproducts have a highly unpredictable demand. Inorder to avoid stock-out and maintain a highinventory fill rate, fashion retailers need to keep asubstantial amount of safety stock. In order toreduce the inventory burden, fashion retailers haveadopted various measures such as the accurateresponse policy (Fisher and Raman, 1996) and

quick response policy (Iyer and Bergen, 1997; Auand Chan, 2002; Choi et al., 2006; Choi and Chow,2007). Some fashion retailers improve their deci-sions by acquiring market information and revisingtheir forecast in multiple stages (see Donohue, 2000;Gallego and Ozer, 2001; Sethi et al., 2001; Choi etal., 2003, 2004; Tang et al., 2004; Choi, 2007). Byutilizing market information (e.g., the sales of otherclosely related fashion products), fashion retailerscan reduce the forecast error and it is widelybelieved that it can help to reduce inventory cost,and hence improve profit (e.g., see Eppen and Iyer,1997). Undoubtedly, forecasting is one crucial taskin retail supply chains (Luxhoj et al., 1996; Chu andZhang, 2003; Thomassey et al., 2005; Sun et al.,2007) and it can affect the retailer and other channelmembers. We hence propose to investigate in this

ARTICLE IN PRESS



doi:10.1016/j.ijpe.2007.06.013

�Corresponding author. Tel: +852 2766 6428;

fax: +852 2773 1432.

E-mail address: [email protected] (K.-F Au).


dx.doi.org/10.1016/j.ijpe.2007.06.013


238

Using a Neural Network to Identify Fabric Defects in DynamicCloth Inspection

CHUNG-FENG JEFFREY KUO, CHING-JENG LEE, AND CHENG-CHIH TSAI

Intelligence Control and Simulation Laboratory, Department of Fiber & Polymer Engineering,National Taiwan University of Science and Technology, Taipei, Taiwan, Republic of China

ABSTRACT

In this research, an image system is used as a tool for dynamic inspection of fabrics, andthe inspection sample is a piece of plain white fabric. The four defects are holes, oil stains,warp-lacking, and weft-lacking. The image treatment employs a high-resolution linearscan digital camera. Fabric images are acquired first, then the images are transferred to acomputer for analysis. Finally, the data are adopted as input data for a neural network,which is obtained from readings after treating the images. In this system, there are threefeedforward networks, an input layer, one hidden layer, and an output layer. Because it hasthe ability to cope with the nonlinear regression property, this method can reinforce theeffects of image identification.

In recent years, the technology of image processingsystems has been applied to inspections in the fabric

industry, for instance, cotton cloth evaluations, fiber

characteristics, structural characteristics of nonwovens,evaluations of carpet forming structures, and fabric de-fects, etc. Shin [8] used a texture-tuned mask method toidentify defects, but oil stain results were poor. Wang[ 10] applied a skeleton method to identify surface defectsbut wasted too much time making the identifications.Ribolzi [6] made an optical electron analysis of warp-lacking and weft-lacking, and Konda [4] employed im-age analysis on woolen balls.

In recent years, neural networks have been adopted forimage analyses. Neubauer [5] used image segmentationtechnology in conjunction with a neural network in iden-tifying fabric defects. Sanby [7] employed a line-scandigital camera to inspect lace. Vangheluwe [9] usedimage analysis and a neural network to measure set

marks. Barrett [ 1 ] applied Fourier transformation and aneural network in a stitching system for on-line classifi-cation. Chen [3] applied a back propagation neural net-work to Fourier analysis to inspect fabric defects such aslack of yarns and oil stains. Actually, dynamic fabricinspection is more difficult than static fabric inspectionbecause moving situations are more complicated, and thespeed of computer processing is another critical issue.Bradshaw [2] reported that no computer has a detectionefficiency greater than 60% when used for fabrics, andtheir use is therefore restricted to inspection for low tomedium quality production. In this article, after acquir-ing an image from the media generated from a VC+ +program, the computer uses this image to process the

accumulations of image values from longitudinal andtransverse directions in order to find the length, width,and gray level of a fabric defect. Not only can it find adefect in high-speed performance, it can also preciselycalculate the length, width, and gray level of that defect.

System Scheme

This research is intended to set up a dynamic inspec-tion tum-key system for fast image acquisition with alinear scan digital camera. The essential factors consid-ered here include whether or not the light-source condi-tions of the fabric images during the image-acquiringprocess by the line-scan digital camera are consistent.We also need to be aware of whether or not the clothitself vibrates along with the conveyor belt, so we use ahelium neon direct light source with strong brightnessand stability. The stability of the entire module willincrease with the best external environmental conditions.The camera DASA CL-C7-4096 is adapted for highspeed linear scanning because it can pick up 7200 linesper second, and it can obtain high-precision images on adynamic fabric. The major specifications are 7 X 7 ¡Lmpixel pitch, 28.7 mm X 7 ¡Lm lens diaphragm, and 7.2kHZ maximum camera line rate.

ANALYSIS OF DEFECT FORMATION

The reasons for weft-lacking generally are too great apick force strength and too high a tension [8]. Warp-lacking is due to poor original yarn, too short a warproute, or too high a tension. Because warp- or weft-



Pergamon Expert Systems With Applications, Vol. 9, No. 2, pp. 237-246, 1995

Copyright © 1995 Elsevier Science Ltd Printed in the USA. All rights reserved

0957-4174/95 $9.50 + .00 0957-4174(94)00065-4

Expert System Support in the Textile Industry: End Product Production Planning Decisions

E NELSON FORD

Department of Management, College of Business, Auburn University, Auburn, AL 36849, U.S.A.

JARICK RAGER

Department of Textile Engineering, College of Engineering, Auburn University, Auburn, AL 36849, U.S.A.

Abstract--The textile industry is slowly developing expert system applications to increase production, improve quality, and reduce costs. Such systems are surfacing in a variety of areas throughout the textile manufacturing process. This paper describes an expert system developed to support an important decision scenario in the textile industry. The scenario concerns a sequence of production planning decisions necessary to produce a specific category of end product. This sequence is described as follows: given the decision to produce a particular type of end product, the appropriate fiber type is chosen; next, the appropriate yarn count group is chosen; next, the appropriate spinning system is chosen; and finally, the appropriate preparation method is chosen. Each decision in the sequence depends on the combination of decisions made in the preceding stages. The resulting system is described and its application is illustrated through the presentation of a sample consultation. The integration of the expert system into a broader environment for textile manufacturing decision support is also discussed.

1. INTRODUCTION

EXPERT SVSa'EMS have emerged from research labs of leading universities and major corporations into business and industry for everyday use. The textile industry is slowly developing expert system applications to increase production, improve quality, and reduce costs. Expert systems function as intelligent assistants, serving any number of individuals needing help and guidance to solve problems and make sound decisions. Expert systems have the potential to increase machine efficiency, decrease maintenance down-time, and improve managerial decisions. The textile industry is beginning to realize the advantages of computerizing the perishable knowledge of the experts in the industry. Expert systems offer the potential of being the first practical method available to preserve the intellectual property of an organization (Demers, 1989).

I.I. Sample Expert Systems in Textiles

Expert systems developed for textile manufacturers are surfacing in a variety of areas throughout the textile

Requests for reprints should be sent to F. Nelson Ford, Department of Management, College of Business, Auburn University, Auburn, AL 36849, U.S.A..

manufacturing process. Some of these are described briefly below.

1.1.1. North Carolina State University Expert System. Expert system technology is used in the design of industrial fabrics and made-to-measure clothing. An expert system that aids in the structural design process for woven industrial fabrics was developed at North Carolina State University. Designers use the system to integrate knowledge concerning the structural properties of yams and fabrics with a customer's desired fabric characteristics. The system searches an industrial fabric data base to attempt to find a match that meets the customer's requirements. If an exact match is not found, the system will attempt to redesign a similar fabric found in the data base to meet the customer's requirements. If the system is still unable to make a match, the system will attempt to synthesize the entire design process to produce the fabric (Demers, 1989).

1.1.2. Clothing Design Expert System. A research team at the University of Maryland has developed the Clothing Design Expert System (CDES). CDES contains the Alteration Definition Tool (ADT) and the Pattern Requirement Language (PRL). These two tools provide a

237


Comparison of negotiation protocols in dynamic agent-basedmanufacturing systems

Jihad Reaidya,�, Pierre Massottea,b, Daniel Diepa

aLaboratoire de Genie Informatique et Ingenierie de Production, Ecole des Mines d’Ales, Parc Scientifique George Besse,

30035 Nımes, FrancebIBM Academy of Technology, La Gaude, France

Available online 28 January 2005

Abstract

This paper proposes a negotiation methodology based on multi-agent system for heterarchical and complex

manufacturing control systems. This approach has been selected to implement new paradigms based on ‘‘co-

opetition ¼ co-operation+competition’’ in order to improve the ‘‘production on demand’’ and reaction capabilities of

distributed production systems related to the net-economy. Agents may represent products and resources of the system.

The local scheduling and control functions in dynamic environments is addressed by a new negotiation protocol

between agents based on the ‘‘request session’’ principle for cooperation and on the game theory approach for

competition.


Keywords: Production and management control; ‘‘Co-opetition’’; Multi-agent systems; Negotiation protocol; Game theory

1. Introduction

The application of multi-agent systems based onthe concept of distributed artificial intelligence isconsidered as being one of the most promisingcontrol architectures for next-generation of com-plex production systems, specifically in a dynamicenvironment (failed resources, disturbances, etc.).

In particular, very attractive solutions and efficientissues are expected in the domain of localplanning, and execution control, to improve theconventional supply chain management; here, theusual production system management consists, ina set of separate and heterogeneous applicationsoftware packages, such as Enterprise ResourcePlanning (ERP), Manufacturing Execution System(MES), Supervisory Control And Data Acquisi-tion (SCADA), etc. These tools are not able tocover satisfactorily the constraints required by thenew challenges of the economy such as networkedenterprises, production on demand or mass

ARTICLE IN PRESS



doi:10.1016/j.ijpe.2004.12.011

�Corresponding author. Tel.:+33 4 6638 7028.

E-mail addresses: [email protected] (J. Reaidy),

[email protected] (P. Massotte),

[email protected] (D. Diep).


375

Building a Rule Set for the Fiber-to-Yarn Production Process by Means ofSoft Computing Techniques

S. SETTE,1 L. BOULLART,2 AND L. VAN LANGENHOVE1

University of Ghent, Technologiepark 9, B-9052 Zwijnaarde, Belgium

1 Department of Textiles: e-mail : [email protected] Department of Control Engineering & Automation: e-mail:

[email protected]

ABSTRACT

An important aspect of the spinning process is the ability to predict the spinnability ofa yam and its resulting strength based on the fiber quality and machine settings. Currentlyavailable fiber-to-yarn models are limited to the so-called "black box" approach, gener-ating an output (spinnability) without containing physical, interpretable information aboutthe process itself. This paper presents a method to predict the spinnability and strength ofa yam with a set of IF-THEN rules. The rule set is automatically generated using theavailable data by means of a new learning classifier system called a fuzzy efficiency-basedclassifier system (FECS), which enhances the original learning classifier algorithm ofGoldberg [5] by defining several rule efficiencies and introducing them into the learningstrategy of the system. Furthermore, FECS allows the introduction of continuous (fiber andyarn) parameters, which broaden the application fields considerably in contrast to discreteparameters alone. To this end, the generated rules are expanded to represent fuzzy classeswith corresponding membership degrees toward each fiber-to-yarn data sample. Ruleefficiencies and the reward mechanism are modified to account for the membership degreeof each data sample. The paper demonstrates that the resulting prediction accuracy is goodand, more importantly, also delivers additional qualitative information about the fiber-to-yarn process behavior. The generated rule set allows almost 100% acceptable classifica-tion of yarn strength in three classes. The methodologies described in this paper are

conveniently classified as "soft computing."

One of the important production processes in the textileindustry is spinning. Starting with cotton fibers, yams are(usually) created on a rotor spinning machine. The spinna-bility of a fiber depends on its quality and the settings of thespinning machine. It would be very beneficial to be able topredict the spinnability and resulting strength of a yamstarting from a certain quality and from machine settings.To this end, two totally different modeling approaches canbe considered: so-called &dquo;white&dquo; modeling and &dquo;black box&dquo;modeling. In white modeling, the process is described bymathematical equations based on (theoretical) physicalknowledge of the process. Extensive physical informationabout the process is in this case available through physical,chemical, or mechanical equations, giving the user a thor-ough insight into the operation of the process. However,due to the large input (and output) dimensions of the fiber-

to-yarn process and their complex interactions, no exactmathematical model of a spinning machine is known to

exist, nor is it likely that such a model will ever be con-structed. A black box model, in contrast to white modeling,simply connects input parameters to the output withoutgiving or containing any substantial physical informationabout the process itself. Black box models have been suc-

cessfully constructed by Pynckels et al. to predict the spin-nability [9] and characteristics’ [ 10] of a yam using neuralnetworks with a backpropagation leaming rule. Apart fromthe lack of physical information, these models also have nofault indication or measure of uncertainty about the results.

In this research, we will present a new modelingapproach called the &dquo;efficiency-based classifier system&dquo;or Ecs to the fiber-to-yarn process by using an automatedlearning method to generate rules that allow us to predictthe spinnability and strength of the yarn based on fiberquality and machine settings. This kind of approachcould be called &dquo;grey modeling,&dquo; since not only is the

relationship between input and output parameters estab-



675

Applying Fuzzy Logic and Neural Networks to Total Hand Evaluation ofKnitted Fabrics

SHIN-WOONG PARK, YOUNG-GU HWANG, AND BOK-CHOON KANGDepartment of Textile Engineering, Inha University, Nam-Ku, 402-751, Inchon, South Korea

SEONG-WON YEO

Department of Electrical Engineering. Inha University, Nam-Ku, 402-751, Inchon, South Korea

ABSTRACT

This study of two new total hand simulating methods for knits uses fuzzy theory and neural networks. One method, a neural network system trained with a back-propaga-tion algorithm, performs functional mapping between mechanical properties and the resulting total hand values of the fuzzy predicting method. The second method, afuzzy-neural network system, uses the fuzzy membership function, weighted factorvector, and error back-propagation algorithm. The principal mechanical properties ofstretchiness, bulkiness, flexibility, distortion, weight, and surface roughness of theknitted fabrics are correlated with experimentally determined Kawabata total handvalues and fuzzy transformed overall hand values. Fuzzy and neural networks agreebetter with the subjective test results than the KES-FB system. The mechanical prop-erties are fuzzified by fuzzy membership functions, then trained to predict the totalhand value of outerwear knitted fabrics. In each case, the prediction error is less thanthe standard deviation of experimentation, and the optimum structure is investigated.These two systems, which use the Pascal programming language, produce objectiveratings of outerwear knit fabrics.

In previous papers [ 11-15], we have published data ona fuzzy prediction model for double weft knitted fabrics.To replace traditional subjective fabric hand assessment,we have established an objective measure of quality andperformance on the basis of low-stress mechanical prop-erties. Since the handle of fabrics obtained from touchand appearance is influenced by the mental and mechan-ical properties of the expert, it will be more meaningfulto rate overall hand values with fuzzy theory and neuralnetworks.To date, the KES-FB system is the criterion most

commonly used to evaluate the total hand value offabrics in textile research and industry [5]. With anobjective test system such as the KES-FB for evaluatingthe mechanical properties of fabrics, it is possible toestablish hand evaluation software, which helps to

clarify objective ratings in mutual communicationsbetween different sectors in the industry about thequality of a fabric ( 16]. But primary hand expressionsand the total hand value depend mainly on Japanesehand experts and cannot be correlated to other cultural

backgrounds or to subjective factors. At present,the total hand of knitted fabrics is primarily evaluated

by subjective hand assessment [21, so hand eval-uation systems are somewhat subjective and have sev-eral shortcomings when applied to other countries

[9-11].There are, however, several problems in determining

the total hand value of a knitted fabric, such as thedifficulty of measuring, geographical climate, culturalfactors, and application method [9-11. 17].

In order to overcome these shortcomings of the eval-uation software in the KES-FB system, new theoreticalmethods such as a psychological model based on

Steven’s law [4], total handle evaluation based on theconcept of Euclidean distance [8], an empirical modelbased on fuzzy theory [ 18], and variable clustering anal-ysis methods [7] have been investigated. All of these aremainly objective statistical modeling methods exhibitingneither simulation, programing, nor automatic calcula-tion of total hand value.

In this paper, we describe the objective total handevaluation systems for current outerwear knits developedusing fuzzy logic and neural networks. We begin with asummary of the artificial neural network theory and theprinciple of the back-propagation algorithm.



Discrete Optimization

Application of a mixed simulated annealing-geneticalgorithm heuristic for the two-dimensional

orthogonal packing problem

T.W. Leung a, Chi Kin Chan b, Marvin D. Troutt c,*

a Diocesan Girls� School, Kowloon, Hong Kongb Department of Applied Mathematics, The Hong Kong Polytechnic University, Kowloon, Hong Kongc Department of Management and Information Systems, Kent State University, Kent, OH 44240, USA

Received 23 August 2000; accepted 10 January 2002

Abstract

In this paper a pure meta-heuristic (genetic algorithm) and a mixed meta-heuristic (simulated annealing-genetic

algorithm) were applied to two-dimensional orthogonal packing problems and the results were compared. The major

motivation for applying a modified genetic algorithm is as an attempt to alleviate the problem of pre-mature con-

vergence. We found that in the long run, the mixed heuristic produces better results; while the pure heuristic produces

only ‘‘good’’ results, but produces them faster.

� 2002 Elsevier Science B.V. All rights reserved.

Keywords:Meta-heuristics; Mixed heuristics; Simulated annealing; Genetic algorithm; Two-dimensional orthogonal packing problem;

Difference process strategy

1. Introduction

The two-dimensional orthogonal packing pro-

blem consists of packing rectangular pieces of

predetermined sizes into a large but finite rectan-

gular plate (the stock plate), or equivalently, cut-

ting small rectangular pieces from the large

rectangular plate. We wish to find ‘‘packing pat-

terns’’ that minimize the unused area (trim loss).The problem has obvious relevancy to the textile,

paper, and other industries, and in the three-dimensional case, is related to the problem of

packing boxes into a container.

The problem has been formulated as an integer

program. For that approach and variations, see

Beasley (1985), Tsai (1993) or Christofides (1995).

Recently, different meta-heuristics have been ap-

plied to problems of this kind, for instance, see

Parada et al. (1998), Lai and Chan (1997), Jakobs(1996), Glover et al. (1995), Dowsland (1993,

1996). For an introduction to meta-heuristics, see

Osman and Laporte (1996). Based on the papers of

Jakobs (1996), also Lai and Chan (1997), we have

carried out extensive comparisons of these heu-

ristics (Leung et al., 2000). An important building



*Corresponding author. Tel.: +1-330-672-1145; fax: +1-330-

672-2953.

E-mail address: [email protected] (M.D. Troutt).

0377-2217/03/$ - see front matter � 2002 Elsevier Science B.V. All rights reserved.

PII: S0377 -2217 (02 )00218 -7


An immune algorithm approach to hybrid flow shopsscheduling with sequence-dependent setup times

M. Zandieh *, S.M.T. Fatemi Ghomi *, S.M. Moattar Husseini *

Department of Industrial Engineering, Amirkabir University of Technology, 424 Hafez Avenue, Tehran, Iran

Abstract

Much of the research on operations scheduling problems has either ignored setup times or assumed that setup times oneach machine are independent of the job sequence. This paper deals with the hybrid flow shop scheduling problems inwhich there are sequence dependent setup times, commonly known as the SDST hybrid flow shops. This type of produc-tion system is found in industries such as chemical, textile, metallurgical, printed circuit board, and automobile manufac-ture. With the increase in manufacturing complexity, conventional scheduling techniques for generating a reasonablemanufacturing schedule have become ineffective. An immune algorithm (IA) can be used to tackle complex problemsand produce a reasonable manufacturing schedule within an acceptable time. This paper describes an immune algorithmapproach to the scheduling of a SDST hybrid flow shop. An overview of the hybrid flow shops and the basic notions of anIA are first presented. Subsequently, the details of an IA approach are described and implemented. The results obtained arecompared with those computed by Random Key Genetic Algorithm (RKGA) presented previously. From the results, itwas established that IA outperformed RKGA.� 2006 Elsevier Inc. All rights reserved.

Keywords: Short-term scheduling; Hybrid flow shops; Sequence dependent setup times; Makespan; Heuristics; Immune algorithms

1. Introduction

Several flow patterns can be encountered, depending on the number of operations required to process a joband on the number of available machines per operation. When a job requires only one operation for its com-pletion, we characterize it as single-operation; otherwise, we call it multi-operation. In the latter case, the con-cept of routing may be introduced based on machines, we have single machine shop, flow shop, permutationflow shop, job shop, and open shop scheduling problems. When processing stages are considered instead ofmachines, we have parallel machine shop, hybrid flow shop, job shop with duplicate machines schedulingproblems. The diagram in Fig. 1 illustrates schematically the relationships between the different machine envi-ronments [71].

0096-3003/$ - see front matter � 2006 Elsevier Inc. All rights reserved.

doi:10.1016/j.amc.2005.11.136

* Corresponding authors.E-mail addresses: [email protected], [email protected] (M. Zandieh), [email protected] (S.M.T. Fatemi Ghomi), moattarh@

aut.ac.ir (S.M. Moattar Husseini).

Applied Mathematics and Computation 180 (2006) 111–127

www.elsevier.com/locate/amc




mailto:moattarh@ aut.ac.ir

mailto:moattarh@ aut.ac.ir

Journal of the Chinese Institute of Industrial Engineers, Vol. 21, No. 1, pp. 59-67 (2004)

59

AN ELECTROMAGNETISM ALGORITHM OF NEURAL NETWORK ANALYSIS - AN APPLICATION TO TEXTILE

RETAIL OPERATION

Peitsang Wu* Department of Industrial Engineering and Management

I-Shou University 1, Sec, 1, ShiueCheng RD., DaShu Shiang, Kaohsiung County, 840, Taiwan, R.O.C.

Wen-Hung Yang Department of Industrial Engineering and Management

Yuan-Ze University Nai-Chieh Wei

Department of Industrial Engineering and Management I-Shou University

ABSTRACT

This paper applies a heuristic algorithm, called the “Electromagnetism Algorithm” (EM) [3], for neural network training. We develop a meta-model of the relationships between key inputs and performance measures of an apparel retail operations using neural network technology. This method simulates the electromagnetism theory of physics by considering each weight connection in a neural network as an electrical charge. Through the attraction and repulsion of the charges, weights move toward the optimality without being trapped into local optima like other algorithms such as genetic algorithm and gradient descent method. The computation results show that the EM algorithm not only converges much faster than those of genetic algorithms and back propagation algorithms in terms of CPU time but also saves more memories than those in genetic algorithms and back propagation algorithms. Keywords: neural networks, electromagnetism algorithms, quick response, textile

manufacturing, retail operations

* Corresponding author: [email protected]

1. INTRODUCTION

The textile industry is extremely competitive internationally. Due to the low cost of foreign labor (e.g. China and Southeast Asia competitors), the textile industry has been rapidly losing its market share to overseas competitors. Because the industry is labor intensive, many jobs are threatened. In an effort to curtail the loss of market share to overseas competitors, the development of quick response (QR) methodologies for apparel was undertaken in 1984 under the auspices of Crafted With Pride, Inc. (CWP), U. S. A. [7]. Nuttle et al. [14] have developed a simulation model of an apparel retail store in order to obtain quantitative comparisons of QR and traditional retailing procedures for seasonal apparel in a wide variety of settings and at negligible cost (for simulation). The retail model represents part of an ongoing research effort to model the textile apparel

retail chain [10]. The retail simulation model allows rapid cost/benefit studies of specific retailing situations, permits the buyer to play out specific scenarios, and provides the retail executive with a tool for the development of QR procedures. Experimentation with the model [8,14] has shown the clear advantage of QR over traditional retailing practice, as well as the limitation of QR in terms of selling season length, sales/SKU (stock-keeping-unit), etc. Additional results with the retail model linked to an apparel manufacturing model can be found in [9].

In this paper, a neural network for textile retail operations is presented. It is capable of capturing the essential features of the retail simulation model in multidimensional, mathematical relationships between performance (e.g. service level and lost sales) and key decision parameters (e.g. SKU mix and season length). The simulation model is used to generate the training data. Once trained, the neural network is able to

Knowledge and Information Systems (2002) 4: 257–282Ownership and CopyrightSpringer-Verlag London Ltd. c© 2002

Agents in E-Commerce: State of the Art

Minghua He and Ho-fung LeungDepartment of Computer Science and Engineering, Chinese University of Hong Kong,

Hong Kong, PR China

Abstract. This paper surveys the state of the art of agent-mediated electronic commerce (e-commerce), especially in business-to-consumer (B2C) e-commerce and business-to-business(B2B) e-commerce. From the consumer buying behaviour perspective, the roles of agentsin B2C e-commerce are: product brokering, merchant brokering, and negotiation. Theapplications of agents in B2B e-commerce are mainly in supply chain management.Mobile agents, evolutionary agents, and data-mining agents are some special techniqueswhich can be applied in agent-mediated e-commerce. In addition, some technologies forimplementation are briefly reviewed. Finally, we conclude this paper by discussions on thefuture directions of agent-mediated e-commerce.

Keywords: Agent; Auction; Contract; Electronic commerce; Negotiation; Supply chain

1. Introduction

Today, agents and electronic commerce (e-commerce) are among the most import-ant and exciting areas of research and development in information technology.Combining these two fields offers lucrative opportunities both for business toconduct transactions on-line and for developers of tools to facilitate this trend(Tolle and Chen, 2000). This paper tries to draw a picture of the state of the art ofagents in e-commerce, especially in two popular branches in the current researchfield: business-to-consumer (B2C) and business-to-business (B2B) e-commerce.

1.1. Agents and Multi-Agent Systems

An agent is a hardware or (more usually) software entity with (some of) thefollowing characteristics (Jennings et al., 1998; Ferber, 1999; Shoham, 1999):

Received 14 September 2000Revised 13 January 2001Accepted 27 February 2001

Abstract—Seam pucker grade is one of the most important quality parameters in garments manufacturing industry. At present, seam pucker is usually evaluated by human inspectors, which is subjective, unreliable and time-consuming. Instead of subjective evaluation, this paper presents an objective method by using image analysis and pattern recognition. The evaluation system consists of image acquisition, image normalization, feature extraction and self organizing map classifier. Textural features of seam puckers are studied with a widely used statistical method, the co-occurrence matrix approach. The grades of seam puckers can be obtained from the trained self organizing map classifier and the results are very promising.

Index Terms—Classification, Seam puckers, Self organizing map, Fabrics.

I. INTRODUCTION

Nowadays, garment manufacturing industries are faced with increased pressure to become more competitive by increasing yield whilst reducing costs. The ability to compete mainly depends on productivity and quality. With the advances in electronic technologies, much can be done to improve productivity and quality by using automation as an integral part of manufacturing systems. However, automated vision-based inspection of textile products has been developing at a relative slow pace, and has not been widely studied in the research literature.

Seam pucker is defined as the ridges, wrinkles, and

corrugations running along the seam line of garments, and has been regarded as one of the most serious faults in garment manufacturing. It is usually caused by improper selection of sewing parameters and material properties, which results in unevenness on fabrics being stitched together, thus impairing their aesthetic values. In severe cases, seam pucker could appear like a wave front, originating from the seam, and extending to the entire piece of garment, e.g., when the seam is the center ridge linking the two pieces of fabrics in the back of a man’s suit. In less severe cases, the wave formation is less pronounced, but nevertheless discernible. Indeed, garments

Manuscript received March 22, 2007. K. L. Mak is Professor at the Dept. of Industrial and Manufacturing Systems

Engineering (IMSE), the University of Hong Kong. (phone: 852-28592582; e-mail: [email protected]).

WEI LI is a PhD student at the Dept. of IMSE, HKU (e-mail: [email protected]).

exhibiting pronounced seam pucker are certainly unwelcome by customers.

It has been well recognized that elimination of seam pucker

entirely is almost impossible, and the common practice is to accept a small amount of pucker as normal. Hence, it is essential to be able to grade puckered seams as objectively as possible. For this purpose, a set of photographic standards (Fig. 1) has been produced by the American Association of Textiles Chemists and Colorists (AATCC) which shows five standard classes in descending order of severity, from class 5 (no pucker) to class 1 (the most severe pucker). Using this method, observers compare each seam sample with the standard photographs and classify the sample as similar in pucker severity to one of the standard classes. However, this human inspection process is known to be subjective, unreliable and inconsistent. Since quality control plays a prominent role in garment manufacturing, the ability to evaluate seam puckers and to solve the seam pucker problem in the manufacturing process becomes vital. An objective method to evaluate seam pucker is therefore highly desired.

Fig. 1. Photographic standards for subjective pucker inspection

by the AATCC method [7]. Although some research [1-7] has been conducted over the

years to evaluate seam puckers objectively, the economical and accurate method is still absent. In this paper, an objective evaluation method based on the technique of artificial neural networks is presented to grade seam puckers with high accuracy.

A Neural Network Approach to Objective Evaluation of Seam Pucker

K.L. MAK, WEI LI

Proceedings of the World Congress on Engineering 2007 Vol IWCE 2007, July 2 - 4, 2007, London, U.K.

ISBN:978-988-98671-5-7 WCE 2007

int. j. prod. res., 1998, vol. 36, no. 9, 2543± 2551

A genetic algorithm for scheduling job families on a single machine

with arbitrary earliness/tardiness penalties and an unrestricted

common due date

S. WEBSTER² *, P. D. JOG³ and A. GUPTA§

We propose and investigate a genetic algorithm for scheduling jobs about anunrestricted common due date on a single machine. The objective is to minimizetotal earliness and tardiness cost where early and tardy penalty rates are allowedto be arbitrary for each job. Jobs are classi® ed into families and a family setuptime is required between jobs from two di� erent families. Results from a compu-tational study are promising with close to optimal solutions obtained rather easilyand quickly.

1. Introduction

The study of earliness and tardiness penalties in scheduling models has recentlygenerated widespread interest among researchers. Past research on scheduling hasconcentrated mainly on regular performance measures in which the objective func-tion is non-decreasing in job completion times. Some commonly used regular meas-ures include mean ¯ owtime, maximum lateness and mean tardiness. However,practitioners are increasingly adopting concepts such as just-in-time and zero inven-tory to respond to competitive pressures. These concepts stress that earliness, as wellas tardiness, should be discouraged, and an ideal schedule is one where each jobcompletes on its assigned due date. Consequently, non-regular performance criteriasuch as the completion time variance or sum of earliness/tardiness (E/T) penalties aremore appropriate in these settings.

The importance of meeting due dates is well understood in practice.Traditionally, however, the due date has been assumed to be an external variable,outside the control of the job shop manager. Conway (1965) was the ® rst to formallyintroduce due date selection as part of the scheduling problem. This issue of due dateas a decision variable has since received considerable attention. Overviews of this lineof research can be found in Baker (1984), Ragatz and Mabert (1984), Cheng andGupta (1989), Kanet and Christy (1989), Baker and Scudder (1990), and Christy andKanet (1990).

Most of the literature on E/T problems addresses single machine models wherethe set of jobs to be scheduled is known in advance and all jobs are available forprocessing. A number of authors have studied a model where all jobs share acommon due date, but the due date is allowed to be an unrestricted decision variable.The reader is referred to Cheng and Gupta (1989), and Baker and Scudder (1990) for

0020± 7543/98 $12.00 Ñ 1998 Taylor & Francis Ltd.

Revision received 1997.² School of Management, Syracuse University, Syracuse, NY, USA.³ Motorola, Arlington Heights, IL, USA.§ Andersen Consulting, Northbrook, IL, USA.* To whom correspondence should be addressed.

A genetic algorithm for hybrid flowshops withsequence dependent setup times and machine eligibility

Ruben Ruiz *, Concepcion Maroto

Departamento de Estadıstica e Investigacion Operativa Aplicadas y Calidad, Universidad Politecnica de Valencia,

Camino de Vera S/N, 46021 Valencia, Spain

Received 30 September 2003; accepted 28 June 2004Available online 14 March 2005

Abstract

After 50 years of research in the field of flowshop scheduling problems the scientific community still observes anoticeable gap between the theory and the practice of scheduling. In this paper we aim to provide a metaheuristic,in the form of a genetic algorithm, to a complex generalized flowshop scheduling problem that results from the additionof unrelated parallel machines at each stage, sequence dependent setup times and machine eligibility. Such a problem iscommon in the production of textiles and ceramic tiles. The proposed algorithm incorporates new characteristics andfour new crossover operators. We show an extensive calibration of the different parameters and operators by means ofexperimental designs. To evaluate the proposed algorithm we present several adaptations of other well-known andrecent metaheuristics to the problem and conduct several experiments with a set of 1320 random instances as well aswith real data taken from companies of the ceramic tile manufacturing sector. The results indicate that the proposedalgorithm is more effective than all other adaptations.� 2005 Elsevier B.V. All rights reserved.

Keywords: Scheduling; Hybrid flowshop; Setup times; Genetic algorithm

1. Introduction

The area of flowshop scheduling has been a veryactive field of research during the last 50 yearssince Johnson (1954) seminal work. This researchincludes literally hundreds of papers in exact tech-

niques and also heuristic and metaheuristic algo-rithms for flowshop scheduling problems andsome of its variants. However, the reality of theproduction systems is more complicated and thereis a noticeable gap between the theory and theapplication of the existing methods. Graves(1981) pointed out this problem and proposed sev-eral research directions to aid in bridging this gap.Other studies, like the ones in Ledbetter and Cox(1977) and Ford et al. (1987), show that there is

0377-2217/$ - see front matter � 2005 Elsevier B.V. All rights reserved.doi:10.1016/j.ejor.2004.06.038

* Corresponding author. Tel.: +34 96 387 70 07x74946; fax:+34 96 387 74 99.

E-mail address: [email protected] (R. Ruiz).


www.elsevier.com/locate/ejor


IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 2. NO. 3, AUGUST 1994

Papers A Fuzzy Neural Network and its

Application to Pattern Recognition Hon Keung Kwan, Senior Member, IEEE and Yaling Cai, Student Member, IEEE

Abstract-In this paper, we define four types of fuzzy neurons and propose the structure of a four-layer feedforward fuzzy neural network (FNN) and its associated learning algorithm. The proposed four-layer FNN performs well when used to recognize shifted and distorted training patterns. When an input pattern is provided, the network first fuzzifies this pattern and then computes the similarities of this pattern to all of the learned patterns. The network then reaches a conclusion by selecting the learned pattern with the highest similarity and gives a nonfuzzy output. The 26 English alphabets and the 10 Arabic numerals, each represented by 16x 16 pixels, were used as original training patterns. In the simulation experiments, the original 36 exemplar patterns were shifted in eight directions by 1 pixel (6.25% to 8.84%) and 2 pixels (12.5‘h to 17.68% ). After the FNN has been trained by the 36 exemplar patterns, the FNN can recall all of the learned patterns with recognition rate. It can also recognize patterns shifted by 1 pixel in eight directions with loo%, recognition rate and patterns shifted by 2 pixels in eight directions with an average recognition rate of 92.01%. After the FNN has been trained by the 36 exemplar patterns and 72 shifted patterns, it can recognize patterns shifted by 1 pixel with recognition rate and patterns shifted by 2 pixels with an average recognition rate of 98.61%. We have also tested the FNN with 10 kinds of distorted patterns for each of the 36 exemplars. The FNN can recognize all of the distorted patterns with 100% recognition rate. The proposed FNN can also be adapted for applications in some other pattern recognition problems.

I. INTRODUCTION NEURAL NETWORK (NN) has a massively parallel A structure which is composed of many processing ele-

ments connected to each other through weights [ 11-[3]. Neural networks (NN’s) are built after biological neural systems. A NN stores patterns with distributed coding and is a train- able nonlinear dynamic system. A NN has a faster response and a higher performance than those of a sequential digital computer in emulating the capabilities of the human brain. Recently, NN’s have been used in pattem recognition problems, especially where input patterns are shifted in position and scale-changed. Fukushima et al. [4], [SI have presented the Neocognitron, which is insensitive to translation and deformation of input patterns, and used it to recognize hand- printed characters. However, the Neocognitron is complex

Manuscript received July 2. 1992; revised February 14, 1994; accepted July 5 . 1993. This work was supported in part by the Natural Sciences and Engineering Research Council of Canada

The authors are with the Department of Electrical Engineering. University of Windsor. Windsor. Ontario, Canada N9B 3P4.

IEEE Log Number 9401794.

and needs many cells. Carpenter and Grossberg [6] have proposed a self-organizing system which can classify patterns by adaptive resonance theory. However, a lot of internal exemplars including noise patterns are formed in the network. Martin and Pittman [7] have used a backpropagation (BP) learning network to recognize hand-printed letters and digits. Guyon et al. [8] have designed a system for on-line recognition of handwritten characters for a touch terminal using a time- delay neural network and the BP algorithm. Fukumi et al. 191 have proposed a neural pattem recognition system trained by the BP algorithm which can be used to recognize rotated patterns. In [7]-[9], the major problem lies in the lengthy training time of the BP algorithm which does not exist in the proposed fuzzy neural network. Perantonis and Lisboa 1101 have constructed a pattern recognition system which is invariant to the translation, rotation, and scale of an input pattern by high-order neural networks. However, the number of weights in such a network increases greatly with the order of the network.

On the other hand, fuzzy logic [11]-[15] is a powerful tool for modeling human thinking and perception. Instead of bivalent propositions, fuzzy systems reason with multivalued sets. Fuzzy systems store rules and estimate sampled functions from linguistic input to linguistic output. It is believed that the effectiveness of the human brain is not only from precise cognition, but also from fuzzy concept, fuzzy judgment and fuzzy reasoning. Dealing with uncertainty is a common problem in pattern recognition. Fuzzy set theory has proved itself to be of significant importance in pattem recognition problems [ 121-[19]. Fuzzy methods are particularly useful when it is not reasonable to assume class density functions and statistical independence of features.

Some work have been carried out on fuzzy neural systems for pattern recognition. Kosko [14] has proposed a Fuzzy Associate Memory (FAM) which defined mappings between fuzzy sets. FAM used fuzzy matrices instead of fuzzy neurons to represent fuzzy associations. Yamakawa and Tomoda [ 171 have described a simple fuzzy neuron model and used in a neural network for application in character recognition problems. However, they did not describe the specific leaming algorithm for this network. Takagi et ul. [ 181 have constructed a structured neural network using the structure of fuzzy inference rules. This structured NN has better performance than ordinary NN’s when used in pattem recognition problems. However, it is complicated to train this NN as it is composed of

10634706/94$04.00 0 1994 IEEE

Authorized licensed use limited to: Hong Kong Polytechnic University. Downloaded on September 25, 2009 at 01:00 from IEEE Xplore. Restrictions apply.

675

Applying Fuzzy Logic and Neural Networks to Total Hand Evaluation ofKnitted Fabrics

SHIN-WOONG PARK, YOUNG-GU HWANG, AND BOK-CHOON KANGDepartment of Textile Engineering, Inha University, Nam-Ku, 402-751, Inchon, South Korea

SEONG-WON YEO

Department of Electrical Engineering. Inha University, Nam-Ku, 402-751, Inchon, South Korea

ABSTRACT

This study of two new total hand simulating methods for knits uses fuzzy theory and neural networks. One method, a neural network system trained with a back-propaga-tion algorithm, performs functional mapping between mechanical properties and the resulting total hand values of the fuzzy predicting method. The second method, afuzzy-neural network system, uses the fuzzy membership function, weighted factorvector, and error back-propagation algorithm. The principal mechanical properties ofstretchiness, bulkiness, flexibility, distortion, weight, and surface roughness of theknitted fabrics are correlated with experimentally determined Kawabata total handvalues and fuzzy transformed overall hand values. Fuzzy and neural networks agreebetter with the subjective test results than the KES-FB system. The mechanical prop-erties are fuzzified by fuzzy membership functions, then trained to predict the totalhand value of outerwear knitted fabrics. In each case, the prediction error is less thanthe standard deviation of experimentation, and the optimum structure is investigated.These two systems, which use the Pascal programming language, produce objectiveratings of outerwear knit fabrics.

In previous papers [ 11-15], we have published data ona fuzzy prediction model for double weft knitted fabrics.To replace traditional subjective fabric hand assessment,we have established an objective measure of quality andperformance on the basis of low-stress mechanical prop-erties. Since the handle of fabrics obtained from touchand appearance is influenced by the mental and mechan-ical properties of the expert, it will be more meaningfulto rate overall hand values with fuzzy theory and neuralnetworks.To date, the KES-FB system is the criterion most

commonly used to evaluate the total hand value offabrics in textile research and industry [5]. With anobjective test system such as the KES-FB for evaluatingthe mechanical properties of fabrics, it is possible toestablish hand evaluation software, which helps to

clarify objective ratings in mutual communicationsbetween different sectors in the industry about thequality of a fabric ( 16]. But primary hand expressionsand the total hand value depend mainly on Japanesehand experts and cannot be correlated to other cultural

backgrounds or to subjective factors. At present,the total hand of knitted fabrics is primarily evaluated

by subjective hand assessment [21, so hand eval-uation systems are somewhat subjective and have sev-eral shortcomings when applied to other countries

[9-11].There are, however, several problems in determining

the total hand value of a knitted fabric, such as thedifficulty of measuring, geographical climate, culturalfactors, and application method [9-11. 17].

In order to overcome these shortcomings of the eval-uation software in the KES-FB system, new theoreticalmethods such as a psychological model based on

Steven’s law [4], total handle evaluation based on theconcept of Euclidean distance [8], an empirical modelbased on fuzzy theory [ 18], and variable clustering anal-ysis methods [7] have been investigated. All of these aremainly objective statistical modeling methods exhibitingneither simulation, programing, nor automatic calcula-tion of total hand value.

In this paper, we describe the objective total handevaluation systems for current outerwear knits developedusing fuzzy logic and neural networks. We begin with asummary of the artificial neural network theory and theprinciple of the back-propagation algorithm.



Mathematical and Computer Modelling 46 (2007) 1419–1433www.elsevier.com/locate/mcm

Two storage inventory model with fuzzy deterioration over a randomplanning horizon

Arindam Roya,∗, Manas Kumar Maitib, Samarjit Kara, Manoranjan Maitic

a Department of Engineering Science, Haldia Institute of Technology, Haldia, Purba-Medinipur, W.B, Pin-721657, Indiab Department of Mathematics, Mahishadal Raj College, Mahishadal, Purba-Medinipur, W.B, Pin-721628, India

c Department of Applied Mathematics, Vidyasagar University, Paschim-Medinipur, W.B, Pin-721102, India

Received 14 June 2006; received in revised form 25 January 2007; accepted 7 February 2007

Abstract

An inventory model for a deteriorating item with stock dependent demand is developed under two storage facilities over arandom planning horizon, which is assumed to follow exponential distribution with known parameter. For crisp deterioration rate,the expected profit is derived and maximized via genetic algorithm (GA). On the other hand, when deterioration rate is imprecisethen optimistic/pessimistic equivalent of fuzzy objective function is obtained using possibility/necessity measure of fuzzy event.Fuzzy simulation process is proposed to maximize the optimistic/pessimistic return and finally fuzzy simulation-based GA isdeveloped to solve the model. The models are illustrated with some numerical data. Sensitivity analyses on expected profit functionwith respect to distribution parameter λ and confidence levels α1 and α2 are also presented.c© 2007 Elsevier Ltd. All rights reserved.

Keywords: Fuzzy deterioration rate; Stochastic planning horizon; Possibility; Necessity

1. Introduction

Classical inventory models are usually developed over infinite planning horizon. According to Gurnani [8] andChung and Kim [4], the assumption of an infinite planning horizon is not realistic due to several reasons such asvariation of inventory costs, changes in product specifications and designs, technological changes, etc. Moreover,for seasonal products like fruits, vegetables, warm garments, etc., the business period is not infinite. There are somemodels (cf. [5,2,12], etc.) in which time horizon has been considered as finite. For seasonal products, the planninghorizon varies over years and may be considered as stochastic with a distribution. Moon and Yun [17] developed anEOQ model with a random planning horizon. Recently Moon and Lee [16] presented an EOQ model under inflationand discounting with a random product life cycle. Again for seasonal products deterioration is a real life phenomenonand the rate of deterioration is normally imprecise in nature [6]. Though a considerable number of research papershas been published for deteriorating items [3,1,7] none has considered planning horizon of such products as randomin nature especially when deterioration is imprecise.

∗ Corresponding author. Tel.: +91 03224 52900; fax: +91 03224 52800.E-mail address: [email protected] (A. Roy).

0895-7177/$ - see front matter c© 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.mcm.2007.02.017

http://www.elsevier.com/locate/mcm


http://dx.doi.org/10.1016/j.mcm.2007.02.017

Time-dependent reliability of textile-strengthened RC structuresunder consideration of fuzzy randomness

Bernd Moller *, Michael Beer, Wolfgang Graf, Jan-Uwe Sickert

Institute of Statics and Dynamics of Structures, Department of Civil Engineering, Technische Universitat Dresden, D-01062 Dresden, Germany

Received 9 September 2004; accepted 25 October 2005Available online 18 January 2006

Abstract

The reliability of civil engineering structures is time-dependent. By means of strengthening it is possible to improve the load-bearingcapacity and serviceability of structures and simultaneously to increase structural reliability. In this paper, we focus on the time-depen-dent reliability assessment of RC structures strengthened by textile-reinforced fine-grade concrete layers.

The paper starts with a short introduction concerning textile strengthening of RC structures and the underlying mechanical model.The uncertain material parameters of textile-strengthened structures are then investigated. The uncertain parameters are quantified asfuzzy variables or fuzzy random variables. In order to take account of the latter in the assessment of the time-dependent reliability anew fuzzy probabilistic safety concept is presented. The fuzzy adaptive importance sampling (FAIS) method is introduced. The algorithmis demonstrated with an example. The uncertainty of the input parameters is comprehensively reflected in the uncertainty of thecomputed fuzzy reliability index. The assessment of the uncertain results is discussed.� 2005 Elsevier Ltd. All rights reserved.

Keywords: Structural safety; Uncertainty; Fuzzy randomness; Fuzzy probability; Imprecise probability; Fuzzy adaptive importance sampling (FAIS);Textile-strengthened structures; Textile concrete

1. Data uncertainty and textile-strengthened RC structures

1.1. Textile strengthening of RC structures

In order to strengthen damaged RC structures textile-reinforced fine-grade concrete layers are applied to the sur-face. The textile reinforcement of these layers consists of fil-ament yarns (rovings) connected together by stitching yarn(see Fig. 1). Each roving is comprised of a large number ofsingle filaments. The textile reinforcement may consist ofdifferent fiber materials, e.g. alkali resistant glass (ARG)or carbon.

The strengthening of reinforced concrete (aged concrete)with textile-reinforced fine-grade concrete (textile concrete)results in a composite (Fig. 2). The load-bearing behavior

of this composite is determined by the material propertiesof the steel-reinforced concrete, the textile-reinforced con-crete, and the bond between them. An extended layeredmodel with specific kinematics is used to describe theload-bearing behavior of RC constructions with textilestrengthening [1]. This model is referred to as multi-refer-ence-plane model (MRM).

The MRM consists of concrete layers and steel rein-forcement layers of the aged construction, the strengthen-ing layers comprised of the inhomogeneous materialtextile concrete and of the interface layers (Fig. 3).

This multi-layer continuum has the following kinematicfeatures. Because the modification of the concrete layerthickness is very slight and can be neglected, e33 = 0 holds.Furthermore, the transverse shear stresses in the concretelayers have no significant influence on the deformation,e13 and e23 can be set to zero. The deformation state ofthe concrete layers may thus be described by Kirchhoffkinematics. The independent degrees of freedom are


doi:10.1016/j.compstruc.2005.10.006

* Corresponding author. Tel.: +49 351 463 34386; fax: +49 351 46337086.

E-mail address: [email protected] (B. Moller).

www.elsevier.com/locate/compstruc

Computers and Structures 84 (2006) 585–603


Journal of Hazardous Materials B137 (2006) 1788–1795

The use of artificial neural networks (ANN) for modeling of decolorizationof textile dye solution containing C. I. Basic Yellow 28 by

electrocoagulation process

N. Daneshvar ∗, A.R. Khataee 1, N. Djafarzadeh 1

Water and Wastewater Treatment Research Laboratory, Department of Applied Chemistry, Faculty of Chemistry, University of Tabriz, Tabriz, Iran

Received 5 April 2006; received in revised form 7 May 2006; accepted 8 May 2006Available online 22 May 2006

Abstract

In this paper, electrocoagulation has been used for removal of color from solution containing C. I. Basic Yellow 28. The effect of operationalparameters such as current density, initial pH of the solution, time of electrolysis, initial dye concentration, distance between the electrodes,retention time and solution conductivity were studied in an attempt to reach higher removal efficiency. Our results showed that the increase ofcurrent density up to 80 A m−2 enhanced the color removal efficiency, the electrolysis time was 7 min and the range of pH was determined 5–8. Itwas found that for achieving a high color removal percent, the conductivity of the solution and the initial concentration of dye should be 10 mS cm−1

and 50 mg l−1, respectively. An artificial neural networks (ANN) model was developed to predict the performance of decolorization efficiency byEC process based on experimental data obtained in a laboratory batch reactor. A comparison between the predicted results of the designed ANNmodel and experimental data was also conducted. The model can describe the color removal percent under different conditions.© 2006 Elsevier B.V. All rights reserved.

Keywords: Artificial neural networks; Electrocoagulation; Modeling; Decolorization; C. I. Basic Yellow 28

1. Introduction

Many industries such as plastics, paper, textile and cosmet-ics use dyes in order to color their products. These moleculesare common water pollutants and they may be frequently foundin trace quantities in industrial wastewaters. Textile plants, par-ticularly those involved in finishing processes, are major waterconsumers and the source of considerable pollution. The dis-posal of these colored wastewaters poses a major problem forthe industry as well as a threat to the environment. There aremany processes to remove dyes from colored effluents such asadsorption, precipitation, chemical degradation, photodegrada-tion, biodegradation, chemical coagulation and electrocoagu-lation [1–3]. Adsorption and precipitation processes are verytime-consuming and costly with low efficiency. Chemical degra-

∗ Corresponding author. Tel.: +98 411 3393146; fax: +98 411 3393038.E-mail addresses: nezam [email protected] (N. Daneshvar),

ar [email protected] (A.R. Khataee), [email protected](N. Djafarzadeh).

1 Tel.: +98 411 3393165; fax: +98 411 3393038.

dation by oxidative agents such as chlorine is the most importantand effective methods, but it produces some very toxic prod-ucts such as organochlorine compounds [1]. Photooxidation byUV/H2O2 or UV/TiO2 needs additional chemicals and thereforecauses a secondary pollution. Although biodegradation processis cheaper than other methods, it is less effective because ofthe toxicity of dyes that has an inhibiting effect on the bacterialdevelopment [2,3].

Hence, electrocoagulation (EC) as an electrochemicalmethod was developed to overcome the drawbacks of conven-tional water and wastewater treatment technologies. EC processprovides a simple, reliable and cost-effective method for thetreatment of wastewater without any need for additional chemi-cals, and thus the secondary pollution. It also reduces the amountof sludge, which needs to be disposed [3–5].

EC technique uses a direct current source between metal elec-trodes immersed in polluted water. The electrical current causesthe dissolution of metal electrodes commonly iron or aluminuminto wastewater. The metal ions, at an appropriate pH, can formwide ranges of coagulated species and metal hydroxides thatdestabilize and aggregate the suspended particles or precipitate

0304-3894/$ – see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.jhazmat.2006.05.042




dx.doi.org/10.1016/j.jhazmat.2006.05.042

The single-row machine layoutproblem in apparel manufacturing

by hierarchical order-basedgenetic algorithm

Miao-Tzu LinDepartment of Fashion Design and Management,

Tainan University of Technology, Taiwan

Abstract

Purpose – The purpose of this paper is to address the topic of minimizing the moving distanceamong cutting pieces during apparel manufacturing. Change machine layout is often required forsmall quantity and diversified orders in the apparel manufacturing industry. The paper seeks todescribe a hierarchical order-based genetic algorithm to quickly identify an optimal layout thateffectively shortens the distance among cutting pieces, thereby reducing production costs.

Design/methodology/approach – The chromosomes of the hierarchical order-based geneticalgorithm consist of the control genes and the modular genes to acquire the parametric genes, aprecedence matrix and a from-to matrix to calculate the distance among cutting pieces.

Findings – The paper used men’s shirt manufacturing as an example for testing the results of aU-shaped single-row machine layout to quickly determine an optimal layout and improve effectivenessby approximately 21.4 per cent.

Research limitations/implications – The manufacturing order is known. The machine layout isin a linear single-row flow path. The machine layout of the sewing department is independentlyplanned.

Originality/value – The advantage of the hierarchical order-based genetic algorithm proposed isthat it is able to make random and global searches to determine the optimal solution for multiple sitessimultaneously and also to increase algorithm efficiency and shorten the distance among cuttingpieces effectively, according to manufacturing order and limited conditions.

Keywords Garment industry, Textile machinery and accessories, Programming and algorithm theory,Process planning


IntroductionGood machine layout and shorten moving distance among materials are important forreducing production costs. Tompkins et al. (1996) indicated that moving non-valueadded material often takes up 20-50 per cent of the total manufacturing costs, and anefficient layout can save 10-30 per cent of the total manufacturing costs implying thatan optimal layout can improve manufacturing schedules and therefore efficiency.Apparel manufacturing involves small quantities of diversified items that often requirechanges in machine layouts according to different materials, specifications, andmanufacturing processes and methods. If the machine layout is able to be re-arrangedquickly, then the change time, labor required, and moving distance can be reduced.



Single-rowmachine layout

problem

31

Received 2 February 2008Revised 28 May 2008

Accepted 28 May 2008


Vol. 21 No. 1, 2009pp. 31-43


DOI 10.1108/09556220910923737


The principles of intelligent textile and garment manufacturing systemsGeorge StyliosAssembly Automation; 1996; 16, 3; ABI/INFORM Globalpg. 40

Stitching defect detection and classification using wavelet transform and BPneural network

W.K. Wong a,*, C.W.M. Yuen a, D.D. Fan a, L.K. Chan a, E.H.K. Fung b

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kongb Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong


Keywords:Stitching defectImage segmentationDefect classificationWavelet transformQuadrant mean filterNeural network

a b s t r a c t

In the textile and clothing industry, much research has been conducted on fabric defect automatic detec-tion and classification. However, little research has been done to evaluate specifically the stitching defectsof a garment. In this study, a stitching detection and classification technique is presented, which combinesthe improved thresholding method based on the wavelet transform with the back propagation (BP) neuralnetwork. The smooth subimage at a certain resolution level using the pyramid wavelet transform wasobtained. The study uses the direct thresholding method, which is based on wavelet transform smoothsubimages from the use of a quadrant mean filtering method, to attenuate the texture background andpreserve the anomalies. The images are then segmented by thresholding processing and noise filtering.Nine characteristic variables based on the spectral measure of the binary images were collected and inputinto a BP neural network to classify the sample images. The classification results demonstrate that the pro-posed method can identify five classes of stitching defects effectively. Comparisons of the proposed newdirect thresholding method with the direct thresholding method based on the wavelet transform detailedsubimages and the automatic band selection for wavelet reconstruction method were made and theexperimental results show that the proposed method outperforms the other two approaches.

� 2008 Published by Elsevier Ltd.

1. Introduction

Detection of defects plays an important role in the automatedinspection of fabrics and garment products. Quality inspection ofgarment manufacturing still relies heavily on trained and experi-enced personnel checking semi-finished and finished garmentsvisually. However, manual inspection imposes limitations on iden-tifying defects in terms of accuracy, consistency and efficiency, asworkers are subject to fatigue or boredom and thus inaccurate,uncertain and biased inspection results are often produced. As a re-sult, garment inspection is highly prone to errors and it allows de-fects to go undetected. To tackle these problems, it is necessary toset up an advanced inspection system for garment checking thatcan decrease or even eliminate the demand for manual inspectionand increase product quality.

In automated inspection, it is necessary to solve the problem ofdetecting small defects that locally break the homogeneity of a tex-ture pattern and to classify all different kinds of defects. Varioustechniques have been developed for fabric defect inspection. Mostof the defect detection algorithms tackling the problem use Gauss-

ian Markov random field, the Fourier transform, the Gabor filters orthe wavelet transform.

Cohen, Fan, and Attai (1991), Gupta and Sortrakul (1998) andPyun et al. (2007) used a model-based method such as GaussianMarkov random field to inspect fabric defects and the method iscomputationally intensive. Fourier-based methods characterizethe spatial-frequency distribution of images, but they do not con-sider the information in the spatial domain and may ignore localdeviations (Chan & Pang, 2000; Tsai & Huang, 2003; Zhang & Bre-see, 1995). The Artificial Neural Network (ANN) was developed toassess set marks but the parameter selection was inadequate andthe results were unsatisfactory (Vangheluwe, Sette, & Pynckels,1993). In 1996, Tsai and Hu (1996) classified the inputs of nineparameters obtained from a fabric image’s Fourier spectrum usingthe BP neural network. Nevertheless, the identification rate wasnot satisfactory. Gabor filters have been recognized as a jointspatial/spatial-frequency representation for analyzing texturedimages and detecting defects that contain highly specific frequencyand orientation characteristics (Bovik & Clark, 1990; Coggins &Jain, 1989; Jain & Farrokhnia, 1991). A potential disadvantage ofthe decomposition of the Garbor filters is that they are computa-tionally intensive. The Gabor filter banks are not mutuallyorthogonal, which may result in a significant correlation amongtexture features obtained from the Gabor-filtered images. Wavelet

0957-4174/$ - see front matter � 2008 Published by Elsevier Ltd.doi:10.1016/j.eswa.2008.02.066

* Corresponding author.E-mail address: [email protected] (W.K. Wong).








Sales forecasting using extreme learningmachinewith applications in fashion retailing

Zhan-Li Sun, Tsan-Ming Choi ⁎, Kin-Fan Au, Yong YuInstitute of Textiles and Clothing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

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

Article history:Received 10 August 2007Received in revised form 22 July 2008Accepted 31 July 2008Available online 13 August 2008

Keywords:Fashion sales forecastingExtreme learning machineArtificial neural networkBackpropagation neural networksDecision support system

Sales forecasting is a challenging problem owing to the volatility of demand which depends on many factors.This is especially prominent in fashion retailing where a versatile sales forecasting system is crucial. Thisstudy applies a novel neural network technique called extreme learning machine (ELM) to investigate therelationship between sales amount and some significant factors which affect demand (such as designfactors). Performances of our models are evaluated by using real data from a fashion retailer in Hong Kong.The experimental results demonstrate that our proposed methods outperform several sales forecastingmethods which are based on backpropagation neural networks.

© 2008 Elsevier B.V. All rights reserved.

1. Introduction

Sales forecasting refers to the prediction of future sales based onpasthistorical data. Owing to competition [41,42] and globalization, salesforecasting plays a more andmore prominent role in a decision supportsystem [26] of a commercial enterprise. An effective sales forecastingcan help the decision-maker calculate the production andmaterial costsand determine the sales price. This will result in lower inventory levels,quick response and achieve the objective of just-in-time (JIT) delivery[2,5–7,12]. However, sales forecasting is usually a highly complexproblem due to the influence of internal and external environments,especially for the fashion and textiles industry [25–27]. Thus, nowadays,how to develop more accurate and timely sales forecasting methodsbecomes an important research topic. Some retailers improve theirstocking decisions by acquiring market information and revising theirforecast in multiple stages [8–10]. A good forecasting method can helpretailers reduce over-stocking and under-stocking costs [12]. Thus salesforecasting becomes one crucial task in supply chain managementunder uncertainty and it greatly affects the retailers and other channelmembers in various ways [31,43]. In this paper, we propose a newmethod which employs extreme learning machine (ELM) for salesforecasting in fashion retailing [32].

Recently, artificial neural networks (ANNs) have been appliedextensively for sales forecasting [4,13,34,35,44–46] because they havevery promising performance in the areas of control, prediction, andpattern recognition [15,21,22,30,33,38,40]. Many studies conclude thatANN is better than various conventional methods [1,3,28,29,39]. In

[13], the statistical time-series model and the ANN based model wereinvestigated for forecasting women's apparel sales. Chakraborty et al.[3] presented an ANN approach based on multivariate time-seriesanalysis, which can accurately predict the flour prices in three cities inUSA. Lachtermacher and Fuller [28] developed a calibratedANNmodel.In the model, the Box–Jenkins methods are used to determine the lagcomponents of the input data. Moreover, it employed a heuristicsmethod to choose the number of hidden units. In Kuo and Xue [27], theauthors reported that the ANNs are better than many conventionalstatistical forecasting methods (see [3,16] for more details). However,most ANN based sales forecasting methods use gradient-basedlearning algorithms, such as the backpropagation neural network(BPNN), and problems such as over-tuning and long computation timestill arise. A relatively novel learning algorithm for single-hidden-layerfeedforward neural networks (SLFN) called extreme learning machine(ELM) has been proposed in [20,47] recently. In ELM, the input weights(linking the input layer to the hidden layer) and hidden biases arerandomly chosen, and the output weights (linking the hidden layer tothe output layer) are analytically determined by using the Moore–Penrose (MP) generalized inverse. ELM not only learns much fasterwith a higher generalization performance than the traditionalgradient-based learning algorithms but it also avoids many difficultiesfaced by gradient-based learning methods such as stopping criteria,learning rate, learning epochs, local minima, and the over-tunedproblems [16–18,36].

To the best of our knowledge, the application of ELM for fashion salesforecasting has not been studied in the literature. In this paper, the ELMis selected to analyze fashion sales forecasting on the data provided by aHong Kong fashion retailer. In this method, some design factors (size,color, etc.) and sales factors (price, etc.) of the fashion apparels arechosen as the input variables of the ELM. Although ELM has many

Decision Support Systems 46 (2008) 411–419

⁎ Corresponding author. Tel.: +86 852 27666450; fax: +86 852 27731432.E-mail address: [email protected] (T.-M. Choi).

0167-9236/$ – see front matter © 2008 Elsevier B.V. All rights reserved.doi:10.1016/j.dss.2008.07.009


Decision Support Systems

j ourna l homepage: www.e lsev ie r.com/ locate /dss


http://dx.doi.org/10.1016/j.dss.2008.07.009



RECOT: an expert system for the reduction of environmental cost in the texti...Kostas MetaxiotisInformation Management & Computer Security; 2004; 12, 2/3; ABI/INFORM Globalpg. 218

Prediction of the air permeabilityof woven fabrics using neural

networksAhmet Cay

Ege University, Textile Engineering Department, Izmir, Turkey

Savvas Vassiliadis and Maria RangoussiTechnological Education Institute of Piraeus,

Department of Electronics, Athens, Greece, and

Isık TarakcıogluEge University, Textile Engineering Department, Izmir, Turkey

Abstract

Purpose – The target of the current work is the creation of a model for the prediction of the airpermeability of the woven fabrics and the water content of the fabrics after the vacuum drying.

Design/methodology/approach – There have been produced 30 different woven fabrics undercertain weft and warp densities. The values of the air permeability and water content after the vacuumdrying have been measured using standard laboratory techniques. The structural parameters of thefabrics and the measured values have been correlated using techniques like multiple linear regressionand Artificial Neural Networks (ANN). The ANN and especially the generalized regression ANNpermit the prediction of the air permeability of the fabrics and consequently of the water content aftervacuum drying. The performance of the related models has been evaluated by comparing the predictedvalues with the respective experimental ones.

Findings – The predicted values from the nonlinear models approach satisfactorily the experimentalresults. Although air permeability of the textile fabrics is a complex phenomenon, the nonlinearmodeling becomes a useful tool for its prediction based on the structural data of the woven fabrics.

Originality/value – The air permeability and water content modeling support the prediction of therelated physical properties of the fabric based on the design parameters only. The vacuum dryingperformance estimation supports the optimization of the industrial drying procedure.

Keywords Air, Permeability, Neural nets, Porosity, Drying, Modelling


IntroductionTextile fabrics consist of interlaced yarns or fibers. They are complex materials andtheir structure is porous. They permit the flow of the air through the constitutingmaterials: yarns and fibers. The discussion about the air permeability of the textile



The authors would like to acknowledge Clothing Textile and Fibre Technological DevelopmentSA, in Athens, Greece for the use of the Shirley air permeability tester and Gokhan Tekstil San.Tic. A.S., in Denizli, Turkey, for the kind supply of the woven fabrics. The present work has beenpartially supported by the “Archimedes 1” EPEAEK Research Project in TEI Piraeus,co-financed by the EU (ESF) and the Greek Ministry of Education.

IJCST19,1

18

Received March 2006Revised August 2006Accepted August 2006

International Journal of ClothingScience and TechnologyVol. 19 No. 1, 2007pp. 18-35q Emerald Group Publishing Limited0955-6222DOI 10.1108/09556220710717026

Optimisation of garment design using fuzzy logic and sensoryevaluation techniques

Y. Chen a, X. Zeng a,�, M. Happiette a, P. Bruniaux a, R. Ng b, W. Yu b

a Ecole Nationale Superieure des Arts & Industries Textiles, Roubaix 59100, Franceb Institute of Textiles & Clothing, the Hong Kong Polytechnic University, Hong Kong, China


Article history:

Received 22 May 2007

Received in revised form

22 April 2008

Accepted 30 May 2008Available online 9 August 2008

Keywords:

Garment design

Ease allowance

Fuzzy logic

Sensory evaluation

Data aggregation

OWA operator

a b s t r a c t

The ease allowance is an important criterion in garment design. It is often taken into account in the

process of construction of garment patterns. However, the existing pattern generation methods cannot

provide a suitable estimation of ease allowance, which is strongly related to wearer’s body shapes and

movements and used fabrics. They can only produce 2D patterns for fixed standard values of ease

allowance. In this paper, we present a new method for optimizing the estimation of ease allowance of a

garment using fuzzy logic and sensory evaluation. Based on the optimized values of ease allowance

generated from fuzzy models related to different key body positions and different wearer’s movements,

we obtain an aggregated ease allowance using the OWA operator. This aggregated result can further

improve the wearer’s fitting perception of a garment and adjust the compromise between the style of

garments and the fitting comfort sensation of wearers. The related weights of the OWA operator are

determined according to designer’s linguistic criteria on comfort and garment style. The effectiveness of

our method has been validated in the design of trousers of jean type. It can be also applied for designing

other types of garment.

& 2008 Elsevier Ltd. All rights reserved.

1. Introduction

Recently, mass customization has made great benefits in manymanufacturing sectors, including automobile, textile, cosmeticand so on. Classically, this concept is defined as ‘‘producing goodsand services to meet individual customer’s needs with near massproduction efficiency’’ Tseng and Jiao (2001). Mass customizerscan customize products quickly for individual customers or forniche markets at better than mass production efficiency andspeed. Using the same principles, mass customizers can build-to-order both customized products and standard products withoutforecasts, inventory, or purchasing delays. In general, masscustomization is realized by the use of flexible computer-aidedmanufacturing systems to produce custom output. In textile andgarment industry, enterprises also pay great attention to masscustomization and wish to quickly produce a great quantity ofpersonalized garments meeting dynamically changing needs ofconsumers on garment comfort and style with low productionand design cost. The garment design-computer aided systempresented in this paper has been developed in this economic

background. The realization of this system is a multidisciplinaryproject which needs joint efforts of computer scientists, textileresearchers, garment and fashion designers. It is not positioned ina unique scientific community but cover competences of severalscientific and technical fields.

A garment is assembled from different cut fabric elementsfitting human bodies. Each of these cut fabric elements isreproduced according to a pattern made on paper or card, whichconstitutes a rigid 2D geometric surface. For example, a classicaltrouser is composed of cut fabrics corresponding to four patterns:front left pattern, behind left pattern, front right pattern andbehind right pattern. A pattern contains some reference linescharacterized by dominant points which can be modified.

Of all the classical methods of garment design, the drapingmethod is used in the garment design of high level Crawford(1996). Using this method, pattern makers drape the fabricdirectly on the mannequin, fold and pin the fabric onto themannequin, and trace out the fabric patterns. This method leadsto the direct creation of clothing with high accuracy but needs avery long trying time and sophisticated techniques related topersonalized experience of operators. Therefore, it cannot beapplied in a massive garment production. Direct drafting methodis faster and more systematic but often less precise Aldrich (1997).It is generally applied in classical garment industry. Using thismethod, pattern makers directly draw patterns on paper usinga pattern construction procedure, implement in a garment

ARTICLE IN PRESS


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

Engineering Applications of Artificial Intelligence

0952-1976/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.

doi:10.1016/j.engappai.2008.05.007

� Corresponding author at: The ENSAIT Textile Institute, GEMTEX Laboratory,

9 rue de l’Ermitage, 59100 Roubaix, France. Tel.: +33 320258967;

fax: +33 320272597.

E-mail address: [email protected] (X. Zeng).

Engineering Applications of Artificial Intelligence 22 (2009) 272–282

www.sciencedirect.com/science/journal/eaai

www.elsevier.com/locate/engappai

dx.doi.org/10.1016/j.engappai.2008.05.007


Optimisation of fault-tolerant fabric-cutting schedulesusing genetic algorithms and fuzzy set theory

P.Y. Mok a, C.K. Kwong a,*, W.K. Wong b

a Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kongb Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong

Available online 25 January 2006

Abstract

In apparel industry, manufacturers developed standard allowed minutes (SAMs) databases on various manufactur-ing operations in order to facilitate better scheduling, while effective production schedules ensure smoothness of down-stream operations. As apparel manufacturing environment is fuzzy and dynamic, rigid production schedules based onSAMs become futile in the presence of any uncertainty. In this paper, a fuzzification scheme is proposed to fuzzify thestatic standard time so as to incorporate some uncertainties, in terms of both job-specific and human related factors,into the fabric-cutting scheduling problem. A genetic optimisation procedure is also proposed to search for fault-tol-erant schedules using genetic algorithms, such that makespan and scheduling uncertainties are minimised. Two setsof real production data were collected to validate the proposed method. Experimental results indicate that the genet-ically optimised fault-tolerant schedules not only improve the operation performance but also minimise the schedulingrisks.� 2005 Elsevier B.V. All rights reserved.

Keywords: Genetic algorithms; Fuzzy set theory; Parallel machine scheduling; Fabric cutting

1. Introduction

In response to the ever changing fashion mar-kets, quick response to customer demands is a

key philosophy of today’s apparel industry. Inorder to shorten products’ time to market, plan-ning and scheduling of various apparel manufac-turing operations have received large researchattention recently. Algorithms and heuristics devel-oped in cutting and packing problems (Bennellet al., 2001; Burke et al., 2004, in press; Dowslandet al., 2002; Gomes and Oliveira, 2006; Jakobs,1996; Kim et al., 2001; Hifi and M’Hallah, 2005)are being applied in marker planning for fabric

0377-2217/$ - see front matter � 2005 Elsevier B.V. All rights reserved.doi:10.1016/j.ejor.2005.12.021

* Corresponding author. Tel.: +852 2766 6610; fax: +8522362 5267.

E-mail address: [email protected] (C.K.Kwong).


www.elsevier.com/locate/ejor


YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

Optimal reorder decision-making in the agent-based apparel supply chain

A. Pan a, S.Y.S. Leung a,*, K.L. Moon a, K.W. Yeung b

a Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hong Kongb Clothing Industry Training Authority, Hong Kong


Keywords:AgentDecision-makingReorder strategy

a b s t r a c t

The application of agent technology in the apparel supply chain management has gained increasing interest.Agents can help automate a variety of tasks and facilitate decision-making in the supply chain. Comparedwith other industries, there are more uncertainties existing in the fashion industry such as market needs,fashion change and seasonality, which increases the difficulty of managing the apparel supply chain espe-cially in the ordering process. Thus, it is necessary to increase the coordination in the apparel supply chainprocesses and develop optimal decision-making strategy for the apparel supply chain under the dynamicenvironment. In this paper, unified modeling language (UML) is applied to simulate the supply chain pro-cesses and describe the relationships between agents. This paper also applies genetic algorithm (GA) andfuzzy inference theory to the dynamic reorder strategy for the supply chain agent to make optimal decisionabout replenishment quantity and reorder point in order to minimize the inventory cost correspondingly.


1. Introduction

In today’s increasingly global and competitive clothing market-place, it is imperative that apparel enterprises work together toachieve common goals such as minimizing the delay of deliveries,the holding and the transportation costs (Roy, Anciaux, Monteiro,& Ouzizi, 2004). A supply chain can be defined as a network consist-ing of suppliers, warehouses, manufacturers, wholesales, and retail-ers through which material and products are acquired, transformed,and delivered to consumers in markets (Hyung & Sung, 2003). Thus,more and more apparel companies adopt and explore better supplychain management (SCM) to improve the overall efficiency. A suc-cessful SCM requires a change from managing individual functionsto integrating activities into key supply chain processes.

Owing to the high complexity and uncertainty of the supplychain in apparel industry, a traditional centralized decisional sys-tem seems unable to manage easily all the information flows andactions. Thus, a more distributed approach, agent technology is re-viewed in this research in order to achieve better operation and tofacilitate the apparel supply chain management.

Generally speaking, agents are active, persistent (software)components with the abilities of perceiving, reasoning, actingand communicating (Fung & Chen, 2005). The agent may follow aset of rules predefined by the user and then applies them. Theintelligent agent will learn and be able to adapt to the environmentin terms of user requests consistent with the available resources(Papazoglou, 2001). The key aspects of agents are their autonomy

and abilities to reason and act in their environment, as well as tointeract and communicate with other agents to solve complexproblems (Jain, Aparico, & Singh, 1999). Autonomy means thatthe agent can act without the direct intervention of humans orother agents and that it has control over its own actions and inter-nal state. The agent must communicate with the user or otheragents to receive instructions and provide results. An essentialquality of an agent is the amount of learned behavior and possiblereasoning capacity that it has.

As the market needs are extremely various and fashion updatesquickly, the supply chain member usually cannot make decisionsimmediately because of the inaccurate or incomplete information.The decision delay in the supply chain prolongs the process timeand causes a company to lose competence. In order to reduce thisdelay, the supply chain member needs to give quick response.Thus, a supply chain can be characterized as a logistic network ofpartially autonomous decision-makers. Supply chain managementhas to do with the coordination of decisions within the network.Different segments of the network are communicating with oneanother through flows of material and information, being con-trolled and coordinated by the activities of supply chain manage-ment. Since more than one decision-maker is involved, thesupply chain has a typical distributed decision-making situation(Schneeweiss, 2003).

In the apparel supply chain, ordering decision and inventorydecision are two critical decisions supply chain managers have toface. The orders are usually made based on the forecasted cus-tomer demand without considering the uncertain factors in appa-rel industry such as weather and fashion trend. Researches havebeen done on the vendor-managed inventory (VMI) replenishment


* Corresponding author. Tel.: +852 27666467.E-mail address: [email protected] (S.Y.S. Leung).








YIUCHUNGCHI

Highlight

YIUCHUNGCHI

Highlight

ELSEVIER

Dyes and Pigments, Vol. 34, No. 3, pp. 181-193, 1997 © 1997 Elsevier Science Ltd

All rights reserved. Printed in Great Britain P I I : S0143-7208(96)00081-2 0143-7208/97517.00+0.00

Multiple Linear Regression (MLR) and Neural Network (NN) Calculations of some Disazo Dye Adsorption on Cellulose"

Simona Timofei, a* Ludovic Kurunczi, a Takahiro Suzuki, b Walter M. F. Fabian c & Sorel Mure~an d

alnstitute of Chemistry, Romanian Academy, Bul. Mihai Viteazu124, 1900, Timisoara, Romania bResearch Laboratory of Resources Utilization, Tokyo Institute of Technology,

4259 Nagatsuta-cho, Midori-ku, Yokohama, 226 Japan qnstitut fiir Organische Chemie, Kari-Franzens-Universit~it Graz, Heinrichstrasse 28,

A-8010, Graz, Austria apolitehnica University of Timisoara, Faculty of Industrial Chemistry, Pial;a Victoriei Nr. 2, 1900,

Timisoara, Romania

(Received 15 July 1996; accepted 16 August 1996)

A BSTRA CT

Multiple Linear Regression ( MLR) analysis and Neural Network ( NN) calculations are applied to a series of 21 disazo anionic dyes. Three-dimensional QSAR parameters were derived from the Cartesian coordinates of the dye molecules. Low energy conformations were obtained by molecular mechanics and quantum chemical calculations. Electronic and steric effects in the dye- cellulose binding are present. The proposed MLR models are rough approxi- mations of nonlinear models. Good correlation with the dye affinity from the MLR calculations and a significantly improved fitting of the NN over the MLR models are observed. The model validity was checked for two proposed models derived from differet:t sets of structural parameters by the leave-one- out cross-validation procedure. For the first model, a better validity ('cross- validated r 2' value, of 0.622) of the NN model is noticed by leaving out one compound (found as outlier) from the training set, in comparison to that of the MLR model obtained for the same set of compounds (q2= 0.434). The q2 value of a second MLR proposed model is better than that of the corresponding NN model. © 1997 Elsevier Science Ltd

Keywords: Multiple Linear Regression (MLR) analysis, Neural Networks (NNs), dye adsorption, cellulose fibre.

~Presented in part at the Symposium on Computational Chemistry, held on 16-17 May 1996 in Tokyo, Japan *Corresponding author.

181

Modelling of the flow behavior of activated carbon cloths using aneural network approach

Catherine Faur-Brasquet *, Pierre Le Cloirec

Ecole des Mines de Nantes, DSEE-GEPEA, UMR CNRS 6144, 4 rue A. Kastler, BP 20722, 44307 Nantes Cedex 3, France

Received 12 April 2002; received in revised form 30 September 2002; accepted 30 September 2002

Abstract

This work investigates the hydrodynamic and aerodynamic behaviors of recent adsorbents, activated carbon cloths (ACC). A first

part presents their characteristics, a particular attention being given to the properties related to their woven structure. The influence

of these characteristics on air and water pressure drops through ACC is shown by experimental measurements. It is established that

a classic model set up for particular media, the Ergun model, does not enable a satisfying modelling of experimental data. An

artificial neural network (ANN) is then used in order to include as explicative factors the cloths properties. The optimization of the

ANN architecture is carried out, in terms of selection of the input neurons and number of hidden neurons. The generalization ability

of the ANN is evaluated using a test dataset distinct from the training set. The influence of specific characteristics of cloths on their

flow behavior is confirmed by an analysis of inputs sensibility, and the determination of their predictive influence.

# 2003 Elsevier Science B.V. All rights reserved.

Keywords: Neural network; Flow behavior; Activated carbon cloths; Ergun’s model

1. Introduction

Activated carbon (AC) in the form of granules (GAC)

or powder (PAC) is commonly used for air or water

treatment [1,2] and granular activated carbon has been

proved to be effective in removing a large number of

organic molecules [3]. Recently, a new form of AC has

appeared: activated carbon fibers (ACF) which may be

pressed to form a felt or woven as a cloth. This last

material, activated carbon cloth (ACC), has been

extensively studied in terms of adsorption performance

in aqueous and gaseous phase. Some researches have

shown their interesting adsorption properties for micro-

organics [4], metal ions [5], dyes [6] in wastewaters, and

for volatil organic compounds contained in polluted

gaseous streams [7]. Their high specific surface area (up

to 1900 m2 g�1) coupled with the low diameter of fibers

(around 10 mm) and some micropores directly connected

to the external surface area of fibers, enable adsorption

rates 1.2�/20 times greater that those obtained with a

commercial granular activated carbon. High adsorption

capacities are also found, reaching up 400 mg g�1 [8].

Furthermore, the woven form allows new kinds of

reactors to be imagined and designed. However, this

industrial design requires the study of the flow behavior

of ACC, in terms of hydro and aerodynamic properties.

Literature reports few information related to pressure

drops through woven structures. Generally, the works

on the dynamic behavior of fibrous media deal with

random stacking of glass fibers [9,10], wood fibers [11]

or textile fibers [12,13], i.e. some structures which have

little in common with woven media from a geometric

point of view. These studies use some models close to the

Ergun equation [14] whose parameters were empirically

calculated for the fibrous media. Other works were

carried out with fabrics and set a relationship between

the permeability of fabrics to air and their degree of

opening but only in the case of a laminar flow [15,16].

The lack of specific deterministic models, coupled

with the multiparameter dimension of fabrics aerody-

namics led us to consider a neural network approach to

model fluid pressure drops through ACC. Artificial

neural networks (ANN) are algorithmic systems derived

* Corresponding author. Tel.: �/33-251-85-8294; fax: �/33-251-85-

8299.

E-mail address: [email protected] (C. Faur-Brasquet).

Chemical Engineering and Processing 42 (2003) 645�/652

www.elsevier.com/locate/cep

0255-2701/02/$ - see front matter # 2003 Elsevier Science B.V. All rights reserved.

doi:10.1016/S0255-2701(02)00202-7


Modelling of CIELAB values in vinyl sulphone dyeapplication using feed-forward neural networks

M. Senthilkumar*

Department of Textile Technology, PSG College of Technology, Coimbatore 641 004, Tamilnadu, India

Received 20 June 2005; received in revised form 18 August 2005; accepted 7 June 2006

Available online 10 August 2006

Abstract

Artificial neural network (ANN) technology has developed from the experimental stage into real industrial applications. To achieve this sig-nificant transition, careful planning and adjustment are required. This article is concerned with the CIELAB values’ prediction based on a neuralnetwork developed for cotton fabric dyed with vinyl sulphone reactive dye. The neural network developed is a multilayer feed-forward network.In textile dyeing industry, achieving the required depth of colour is the important task. In this paper, to achieve the required depth of colour, theCIELAB values of the fabric to be dyed were predicted using trained feed-forward neural network. The results obtained from the network givesan average error of around 2.0% for vinyl sulphone dyes used for training the network in predicting the LAB values. The trained network bringsout the same error for other dyes as well as for input and output parameters selected beyond the range used for training the network.� 2006 Elsevier Ltd. All rights reserved.

Keywords: Colour measurement; Neural networks; Reflectance; Total dye fixed; Whiteness index

1. Introduction

Today, the highly competitive marketplace requires a strongcommitment of firms to satisfy customer’s expectations. Thistendency is even more pronounced for the product appearance.The textile field is especially sensitive to this phenomenon.One of the most important textile characteristics is undou-btedly colour. Among the many quality parameters to beachieved in the dyed goods, achieving the appropriate depthof shade is a very important one. If the depth of colourproduced is different from what is expected, the material hasto be either taken for reworking or rejection. So to proceedfurther colour of the dyed goods has to be measured.

For the measurement of colour, standard values are usedworldwide, for example as determined by an organisationcalled CIE. The values used by CIE 1976 are called L*, a*

and b* and the colour measurement method is called CIELAB.L* represents the difference between light (where L*¼ 100)and dark (where L*¼ 0), a* represents the difference betweengreen (�a*) and red (þa*) and b* represents the differencebetween yellow (þb*) and blue (�b*). The CIE 1976 L*, a*and b* colour space or CIELAB colour space is defined byquantities L*, a* and b*. This L*, a* and b* values are calcu-lated after dyeing the material and based on this values the ma-terial will be either taken for next processing or reworking.The L* a* b* values for a given situation can be predictedusing statistical tools such as multiple regression analysis orcomputational processors such as artificial neural networks(ANN). Prediction using ANNs is claimed to have better accu-racy compared to multiple regression analysis [1,2].

In recent days, neural networks are used for modelling non-linear problems and to predict the output values for a given in-put parameters from their training values. Most of the textileprocesses and quality assessments are non-linear in natureand hence neural networks find application in textile technol-ogy. Web density control in carding [3], prediction of yarnstrength [4], ring and rotor yarn hairiness [5], total hand

* Tel.: þ91 422 2572177x4169, þ91 9443948513 (Mob); fax: þ91 422

2573833.

E-mail address: [email protected]


doi:10.1016/j.dyepig.2006.06.010

Dyes and Pigments 75 (2007) 356e361www.elsevier.com/locate/dyepig


http://www.elsevier.com/locate/dyepig

Intelligent production control decision support system for flexible assembly lines

Z.X. Guo, W.K. Wong *, S.Y.S. Leung, J.T. FanInstitute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong, China


Keywords:Production controlDecision support systemFlexible assembly linesGenetic algorithmsRadio frequency identificationLearning curves

a b s t r a c t

In this study, a production control problem on a flexible assembly line (FAL) with flexible operationassignment and variable operative efficiencies is investigated. A mathematical model of the productioncontrol problem is formulated with the consideration of the time-constant learning curve to deal withthe change of operative efficiency in real-life production. An intelligent production control decision sup-port (PCDS) system is developed, which is composed of a radio frequency identification technology-baseddata capture system, a PCDS model comprising a bi-level genetic optimization process and a heuristicoperation routing rule is developed. Experimental results demonstrated that the proposed PCDS systemcould implement effective production control decision-making for solving the FAL.

� 2008 Published by Elsevier Ltd.

1. Introduction

Effective production control is useful and necessary to improveproduction and management performances and reduce the run-ning cost of factories. A generic architecture for production controldecision-making is shown in Fig. 1. In a real-life production envi-ronment, production data on production orders, production quan-tities of each workstation and the whole production line, operativeefficiency, etc., are collected from shop floors or assembly lines byusing various types of data capture methods, including the manualrecording method, barcode scanning, and the most updated radiofrequency identification (RFID) technology. Based on the collectedproduction data, the production manager makes production deci-sions to achieve various production objectives.

On shop floors or assembly lines with a low level of automation,it is impossible to obtain real-time production data owing to theabsence of an effective data capture system. Thus, it is also impos-sible to make accurate and real-time decisions for production con-trol. This paper presents an intelligent production control decisionsupport (PCDS) system, which is integrated with an RFID-basedreal-time data capture system, for assisting in the production con-trol decisions on a flexible assembly line (FAL).

1.1. Manufacturing flexibility and flexible assembly lines

To meet the increasingly fierce market competition, more andmore manufacturing enterprises seek benefits from manufacturingflexibility and effective production control. Beach, Muhlemann,Price, Paterson, and Sharp (2000) provided a comprehensive

review on manufacturing flexibility. There are various types ofmanufacturing flexibility such as machine flexibility and routingflexibility. Machine flexibility is measured by the number of oper-ations that a workstation processes and the time needed to switchfrom one operation to another. The more operations a workstationprocesses, the less time switching takes and the higher the ma-chine flexibility becomes. Routing flexibility is the ability of a pro-duction system to manufacture a product using several alternativeroutes in the system and it is usually determined by the number ofsuch potential routes.

The FAL is an increasingly attractive assembly form for small ormid-scale production in many industries. Unlike the traditionalassembly line, some FALs allow flexible operation assignment,where one operation can be assigned to multiple workstationsfor processing and multiple operations can be assigned to the sameworkstation. When one operation is assigned to multiple worksta-tions, the processing of this operation is shared by the assignedworkstations and this operation is taken as a shared operation.Each shared operation of a product should be routed to an appro-priate workstation on a real-time basis. Obviously, the FAL withflexible operation assignment involves machine flexibility androuting flexibility. In practice, this type of FAL is usually used in ap-parel manufacturing.

1.2. Variability of operative efficiency

On a highly automated assembly line, the efficiency to process acertain task is deterministic. Yet on FALs with a low level of auto-mation, e.g., FALs highly relying on manual efforts, the operativeefficiency of each task is seldom constant. The variable operativeefficiency leads to the fluctuation of the actual cycle time and in-creases the complexity of production control.

0957-4174/$ - see front matter � 2008 Published by Elsevier Ltd.doi:10.1016/j.eswa.2008.03.023

* Corresponding author. Tel.: +852 27666471; fax: +852 27731432.E-mail address: [email protected] (W.K. Wong).








YIUCHUNGCHI

Highlight

Genetic optimization of order scheduling with multiple uncertainties

Z.X. Guo, W.K. Wong *, S.Y.S. Leung, J.T. Fan, S.F. Chan

Institute of Textiles and Clothing, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong, China

Abstract

In this paper, the order scheduling problem at the factory level, aiming at scheduling the production processes of each productionorder to different assembly lines is investigated. Various uncertainties, including uncertain processing time, uncertain orders and uncer-tain arrival times, are considered and described as random variables. A mathematical model for this order scheduling problem is pre-sented with the objectives of maximizing the total satisfaction level of all orders and minimizing their total throughput time.Uncertain completion time and beginning time of production process are derived firstly by using probability theory. A genetic algorithm,in which the representation with variable length of sub-chromosome is presented, is developed to generate the optimal order schedulingsolution. Experiments are conducted to validate the proposed algorithm by using real-world production data. The experimental resultsshow the effectiveness of the proposed algorithm.� 2007 Elsevier Ltd. All rights reserved.

Keywords: Order scheduling; Uncertain processing time; Probability theory; Genetic algorithms

1. Introduction

Facing with ever increasing global market competition,today’s manufacturers have to continuously improve theirproduction performance so as to be more competitive inthe market. Effective production scheduling plays a signif-icant role in maximizing the resource utilization and short-ening the production lead time.

A wide literature base has been published on productionscheduling, focussing mostly on the scheduling for varioustypes of production systems at the shop floor or assembly-line level, such as job shop scheduling (Adam et al., 1993;Fayad and Petrovic, 2005; Guo et al., 2006; Kondakciand Gupta, 1991), flow shop scheduling (Ishibuchi, Yamamoto, Murata, & Tanaka, 1994; Iyer & Saxena, 2004;Morita & Shio, 2005; Nagar, Heragu, & Haddock, 1996),machine scheduling (Baek & Yoon, 2002; Dimopoulos &Zalzala, 2001; Fowler, Horng, & Cochran, 2003; Liu &Tang, 1999), assembly line scheduling (Guo et al., 2006;Kaufman, 1974; Vargas et al., 1992; Zhang et al., 2000),

etc. Ashby and Uzsoy (1995) have presented a set of sched-uling heuristic to solve the order release and order sequenc-ing problem in a single-stage production system. Axsater(2005) has discussed the order release problem in a multi-stage assembly network by an approximate decompositiontechnique. Their studies only focused on determining thestarting times for different processes of each productionorder (also called order), where the process should be per-formed has not been considered. Chen and Pundoor (2006)have considered the order assignment and scheduling in thesupply chain level, they focused on assigning orders to dif-ferent factories and finding a schedule for processing theassigned orders at each factory. However, multiple shopfloors and multiple assembly lines are setup in most facto-ries. The order scheduling problem at the factory level,where the production process of each order scheduled tothe appropriate assembly line, has not been reported so far.

The great majority of previous studies on productionscheduling are based on the deterministic estimation ofthe processing time of each production process and thearrival time of each order. In real-life production environ-ment, various uncertainties often occur, such as uncertaincustomer orders and uncertain estimation of processing


doi:10.1016/j.eswa.2007.08.058

* Corresponding author. Tel.: +852 27666471; fax: +852 27731432.E-mail address: [email protected] (W.K. Wong).






YIUCHUNGCHI

Highlight

J Intell Manuf (2006) 17:341–354DOI 10.1007/s10845-005-0007-8

Genetic optimization of JIT operation schedules for fabric-cuttingprocess in apparel manufacture

W.K. Wong · C.K. Kwong · P.Y. Mok · W.H. Ip

Received: November 2004 / Accepted: September 2005© Springer Science+Business Media, LLC 2006

Abstract Fashion products require a significant amount ofcustomization due to differences in body measurements, di-verse preferences on style and replacement cycle. It is nec-essary for today’s apparel industry to be responsive to theever-changing fashion market. Just-in-time production is amust-go direction for apparel manufacturing. Apparel indus-try tends to generate more production orders, which are splitinto smaller jobs in order to provide customers with timelyand customized fashion products. It makes the difficult task ofproduction planning even more challenging if the due timesof production orders are fuzzy and resource competing. Inthis paper, genetic algorithms and fuzzy set theory are usedto generate just-in-time fabric-cutting schedules in a dynamicand fuzzy cutting environment. Two sets of real productiondata were collected to validate the proposed genetic opti-mization method. Experimental results demonstrate that thegenetically optimized schedules improve the internal satis-faction of downstream production departments and reducethe production cost simultaneously.

Keywords Genetic algorithms · Fuzzy set theory ·Parallel machine scheduling · Fabric cutting · Apparel

W.K. Wong (B) · P.Y. MokInstitute of Textiles and Clothing,The Hong Kong Polytechnic University,Hunghom, Kowloon, Hong Konge-mail: [email protected]

C.K. Kwong · W.H. IpDepartment of Industrial and Systems Engineering,The Hong Kong Polytechnic University,Hunghom, Kowloon, Hong Kong

Introduction

Apparel production is a type of assembly manufacture that in-volvesanumberofprocesses.Fabric-cuttingoperationisdonein a fabric-cutting department, which usually serves severaldownstream sewing assembly lines. Effective upstream fab-ric-cutting operation ensures the smoothness of downstreamoperations, and thus is vitally important to the overall systemefficiency. Production scheduling of apparel production is achallenging task due to a number of factors. First of all, fash-ion trend is always unpredictable, thus just-in-time produc-tion is employed to ensure products’ short time-to-market.Moreover, in order to cope with the increasing demand onproduct customization, the quantity of garments per produc-tion order tends to be smaller and thus number of produc-tion order processed by the manufacturer has been becominglarger. In this paper, just-in-time (JIT) production schedulingof manual cutting department operation is investigated.

JIT scheduling

Production scheduling has been extensively studied, and theprevious literature has more focused on some single regularmeasures, such as mean flow-time and mean lateness. Sincethe 1980s, the concept of penalizing both earliness and tar-diness has spawned a new and rapidly developing line ofresearch in the scheduling field (Baker & Scudder, 1990).In a JIT environment, both earliness and tardiness must bediscouraged since early finished jobs increase inventory costwhile late jobs lead to customers’ dissatisfaction and loss ofbusiness goodwill. Thus an ideal schedule is one in which alljobs finish within the assigned due dates. The objectives ofearly/tardy (E/T) scheduling could be interpreted in differentways, for example minimizing total absolute deviation fromdue dates, job dependent earliness and tardiness penalties,

Abstract

Documents

erp projects

suitable erp system

erp systemssuch

erp selection

erp consultants

rest erp system solutions

suitable erp system

implementations of erp