Trends in Supply Chain Design and Management

0 downloads 0 Views 5MB Size Report
WDA Centre of Enterprise in Manufacturing Engineering. University of ...... shows that the sales, service, and marketing aspects of a supply chain can also ...... 2001, Han et al., 2001), BizTalk (URL 1; He and Milosevic 2001; Herring and ...... loading truck picks up and/or drops off goods at every stop like a city bus, where.
Springer Series in Advanced Manufacturing

Series Editor Professor D. T. Pham Intelligent Systems Laboratory WDA Centre of Enterprise in Manufacturing Engineering University of Wales Cardiff PO Box 688 Newport Road Cardiff CF2 3ET UK

Other titles in this series Assembly Line Design B. Rekiek and A. Delchambre Advances in Design H.A. ElMaraghy and W.H. ElMaraghy (Eds.) Effective Resource Management in Manufacturing Systems: Optimization Algorithms in Production Planning M. Caramia and P. Dell’Olmo Condition Monitoring and Control for Intelligent Manufacturing L. Wang and R.X. Gao (Eds.) Optimal Production Planning for PCB Assembly W. Ho and P. Ji

Hosang Jung, F. Frank Chen and Bongju Jeong (Eds.)

Trends in Supply Chain Design and Management Technologies and Methodologies

123

Hosang Jung, PhD Grado Department of Industrial and Systems Engineering Virginia Polytechnic Institute and State University Blacksburg, VA 24061 USA

F. Frank Chen, PhD Grado Department of Industrial and Systems Engineering Virginia Polytechnic Institute and State University Blacksburg, VA 24061 USA

Bongju Jeong, PhD Department of Information and Industrial Engineering Yonsei University Seoul 120-749 Republic of Korea British Library Cataloguing in Publication Data Trends in supply chain design and management : technologies and methodologies. - (Springer series in advanced manufacturing) 1. Business logistics I. Jung, Hosang II. Chen, F. Frank III. Jeong, Bongju 658.5 ISBN-13: 9781846286063 ISBN-10: 1846286069 Library of Congress Control Number: 2006938339 Springer Series in Advanced Manufacturing ISSN 1860-5168 ISBN 978-1-84628-606-3 e-ISBN 978-1-84628-607-0

Printed on acid-free paper

© Springer-Verlag London Limited 2007 AnyLogic™ is a trademark of XJ Technologies, Office 410, 49 Nepokorennykh pr., 195220 St. Petersburg, Russia. http://www.xjtek.com/ Arena is a registered trademark of Rockwell Automation, Inc., Building 2, 2000 Marconi Drive, Warrendale, PA 15086, USA. http://www.arenasimulation.com/ CPLEX is a registered trademark, and OPL Development Studio is a trademark, of ILOG, Inc., 1080 Linda Vista Ave., Mountain View, CA 94043, USA. http://www.ilog.com/ Java™ is a trademark of Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, CA 95054, USA. http://www.sun.com/ Pentium is a registered trademark of the Intel Corporation, 2200 Mission College Blvd., Santa Clara, CA 95052, USA. http://www.intel.com/ ProModel is a registered trademark of the ProModel Corporation, 556 East Technology Ave., Orem, UT 84097, USA. http://www.promodel.com/ RHYTHM is a registered trademark of i2 Technologies, Inc., One i2 Place, 11701 Luna Road, Dallas, Texas 75234, USA. http://www.i2.com/ WBI Modeler is a registered trademark of the IBM Corporation, 1133 Westchester Avenue, White Plains, New York 10604, USA. http://www.ibm.com/ WITNESS is a registered trademark of the Lanner Group, Ltd., The Oaks, Clews Road, Redditch, Worcestershire, B98 7ST, UK. http://www.lanner.com/ Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms of licences issued by the Copyright Licensing Agency. Enquiries concerning reproduction outside those terms should be sent to the publishers. The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore free for general use. The publisher makes no representation, express or implied, with regard to the accuracy of the information contained in this book and cannot accept any legal responsibility or liability for any errors or omissions that may be made. 987654321 Springer Science+Business Media springer.com

Preface

Supply chain management has played an important role in restructuring many global companies and increasing their performance since the mid-1990s. Researchers and practitioners also have published various articles on this research area so far. However, with the fast development of information and network technologies and market requirements, new challenges to supply chain management have been emerging. From the perspective of technologies, three noticeable technologies can affect traditional supply chain management: (1) radiofrequency identification (RFID) technology, (2) mobile information transaction and handling technologies, and (3) multiagent modeling technology. These new technologies enable global companies to change their way of thinking on supply chain design and management for surviving in continuously evolving environments. With the help of the advanced technologies mentioned above, various methodologies for supply chain design and management have been newly devised and investigated. The call for book chapter submissions received a very enthusiastic response from the research community worldwide. After undergoing a stringent technical review and editorial process, the eighteen chapters accepted for this book include the work of researchers in Canada, China, France, Germany, Greece, Korea, New Zealand, Switzerland, Taiwan, United Kingdom, and United States. The book is divided in two parts: trends in technologies, and trends in methodologies. In the first part of the book, Chapters 1–7 review new trends in technologies. Various cases and reviews of RFID applications are presented in Chapters 1 and 2. The value of RFID in supply chains is reviewed in Chapter 3. In Chapter 4, the relationship between RFID and inventory management is discussed. Chapters 5 and 6 deal with the impact and applications of mobile and information technologies. Chapter 7 reviews the agent-based approach to supply chain management. In the second part of the book, Chapters 8–18 report new trends in methodologies. Chapters 8–11 cover topic related to supply chain design. Chapter 8 deals with the design of reverse logistics networks. Chapter 9 shows the emerging business models and enabling technologies of government value chains. Chapter 10 focuses on the lean supply chain beyond traditional lean manufacturing. In addition, recent research results on diverse production and distribution models for a real semiconductor industry are summarized in Chapter 11. Chapters 12–14 discuss decentralized supply chains. Each supply chain participant involves mainly a local optimized plan, collaboration or negotiation among supply chain participants should be considered an important enabler in planning and managing of the decentralized supply chain. Chapter 12 presents two different types of

vi

Preface

decentralized supply chains, and Chapter 13 deals with a multiagent approach for dynamic lot-sizing in decentralized supply chains. In Chapter 14, the vendor managed inventory (VMI) technique is presented with a scenario analysis based on a real grocery sector. Chapters 15–17 provide recent issues regarding supply chain uncertainty. Along with the rapid development of various technologies and methodologies in supply chain management, uncertainty issues have gained lots of attention in both academi and industry. Chapter 15 reviews results of supply uncertainty and diversification. Chapter 16 proposes a quantitative robustness index design for supply chains considering uncertainty. Chapter 17 investigates the impact of reducing uncertainty in European supply chains. Chapter 18 introduces recent research results on available-to-promise. The available-to-promise activity is simply to give delivery date promise to customers for specific orders, and it is related to customer satisfaction and the total cost of supply chain management. The collected chapters present interesting topics on both technologies and methodologies for advanced supply chain management with diverse applications and reviews not published before. Each of the chapters also contains guidelines for practitioners as an independent section to provide them with: (1) key insights and contributions, (2) techniques for applying introduced methodology or technology to real-world situations, and (3) future possible extension of research for practitioners. The editors hope that researchers may gain an insight into the evolving trend of supply chain design and management, and practitioners may find practical use of the technologies and methodologies presented in their companies or organizations from reading this book. In addition, with guidelines for practitioners in each chapter, this book is not intended to be a theoretical handbook but rather it contains useful and practical information for all readers. Finally, the editors are grateful to all contributors who provided their very fine work to this book and to those reviewers who kindly evaluated the manuscripts promptly in response to our requests. The editors also would like to acknowledge the editorial assistance of Springer. Especially, the support and cooperation of Mr. Anthony Doyle, Senior Editor (engineering) of Springer, are greatly appreciated. It would not have been possible without his initial invitation and continuous concern. The first editor, Dr. Jung also wants to thank the Korea Research Foundation (KRF) for their support (KRF-2004-214-D00407) of his research stay in Virginia Tech where this book was devised and prepared for publication. The partial financial support for Dr. Jung and Dr. Chen from the Center for High Performance Manufacturing (CHPM) at Virginia Tech is also gratefully acknowledged. Hosang Jung Blacksburg, Virginia, USA F. Frank Chen Blacksburg, Virginia, USA Bongju Jeong Seoul, Korea

Contents

List of Contributors………………………………………………………xi Part I. Trends in Technologies 1. A Systems Approach to Viable RFID Implementation in the Supply Chain Can Saygin, Jagannathan Sarangapani and Scott E.Grasman ............................................................................3 2. Applications of RFID in Supply Chains Gary M. Gaukler and Ralf W. Seifert ...................................................29 3. A Tool Set for Exploring the Value of RFID in a Supply Chain Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy..........................................................................49 4. The Effect of RFID on Inventory Management and Control Uttarayan Bagchi, Alfred L. Guiffrida, Liam O’Neill, Amy Z. Zeng, and Jack C. Hayya ...............................................................................71 5. Mobile Supply Chain Event Management Using Auto-ID and Sensor Technologies – A Simulation Approach Frank Teuteberg and Ingmar Ickerott .................................................93 6. Impact of Information Technology on Supply Chain Management Enver Yücesan ....................................................................................127 7. An Agent-based Approach to Enhance Supply Chain Agility in a Heterogeneous Environment Chun-Che Huang, Tzu-Liang (Bill) Tseng, Hong-Fu Chuang, and Yu-NengFan ................................................................................149

viii Contents

Part II. Trends in Methodologies Supply Chain Design 8. Design of Reverse Logistics Networks for MultiProducts, Multistates, and Multiprocessing Alternatives Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi ...............181 9. Transforming the Government Value Chain: Emerging Business Models and Enabling Technologies Nikolaos A. Panayiotou, Stavros T. Ponis, and Sotiris P. Gayialis .......................................................................213 10. Beyond Partnerships: The Power of Lean Supply Chains Leonardo Rivera, Hung-da Wan, F. Frank Chen, and Woo Min Lee ...............................................................................241 11. Diverse Production and Distribution Models in Supply Chains: A Semiconductor Industry Case Young Hoon Lee and Kyung Hwan Kang ..........................................269

Decentralized Supply Chains 12. Decentralized Supply Chain Planning for Two Classified Supply Chains Hosang Jung and F. Frank Chen .......................................................289 13. Multiagent System Approach for Dynamic Lot-sizing in Supply Chains Seokcheon Lee and Soundar Kumara ................................................311 14. Integrating Transport into Supply Chains: Vendor Managed Inventory (VMI) Andrew Potter, Denis R. Towill, and Stephen M. Disney ..................331

Contents

ix

Supply Chain Uncertainty 15. Supply Uncertainty and Diversification: A Review M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee .............345 16. Quantitative Robustness Index Design for Supply Chain Networks Ming Dong and F. Frank Chen ..........................................................369 17. Impact of Reducing Uncertainty in European Supply Chains Paul Childerhouse and Denis R. Towill .............................................393

Available-To-Promise 18. Analyzing the Effectiveness of the Availability Management Process Young M. Lee .....................................................................................411

About the Editors ....................................................................................437 Index ........................................................................................................439

List of Contributors

D. Aït-Kadi, Centre de recherche sur les technologies de l’organisation réseau (CENTOR), Département de génie mécanique, Université Laval, Canada

H. Chuang, Department of Accounting Information, Da-Yeh University, Taiwan

U. Bagchi, Department of Information, Risk, & Operations Management, McComb School of Business, The University of Texas at Austin, USA

S. D’Amours, Centre de recherche sur les technologies de l’organisation réseau (CENTOR), Département de génie mécanique, Université Laval, Canada

F. Frank Chen, Grado Department of Industrial and Systems Engineering, Virginia Polytechnic Institute and State University, USA

S. M. Disney, Logistics Systems Dynamics Group, Cardiff Business School, Cardiff University, UK

F. Cheng, IBM Thomas J. Watson Research Center, 1101 Kitchawan Road, Yorktown Heights, NY 10598, USA

M. Dong, Department of Industrial Engineering and Management, School of Mechanical Engineering, Shanghai Jiao Tong University, P.R. China

P. Childerhouse, Management Systems, Waikato Management School, The University of Waikato, New Zealand

Y. Fan, Department of Information Management, National Chi-Nan University, Taiwan

M. Chouinard, Centre de recherche sur les technologies de l’organisation réseau (CENTOR), Département de génie mécanique, Université Laval, Canada

G. M. Gaukler, Department of Industrial and Systems Engineering, Texas A&M University, USA

xii List of Contributors

S. P. Gayialis, Section of Industrial Management and Operational Research, School of Mechanical Engineering, National Technical University of Athens, Greece

K. H. Kang, Department of Information and Industrial Engineering, Yonsei University, Korea

S. E. Grasman, Engineering Management and Systems Engineering Department , University of MissouriRolla, USA

S. Kumara, Department of Industrial and Manufacturing Engineering, Pennsylvania State University, USA

A. L. Guiffrida, Department of Management and Information Systems, Kent State University, USA

C. Lee, Department of Mechanical and Industrial Engineering University of Toronto, Canada

J. C. Hayya, Department of Supply Chain and Information Systems, Smeal College of Business, Pennsylvania State University, USA

S. Lee, Department of Industrial and Manufacturing Engineering, Pennsylvania State University, USA

J. J. Hennessy, IBM Global Business Services, 11 Madison Avenue, New York, NY 10010, USA

W. M. Lee, Grado Department of Industrial and Systems Engineering, Virginia Polytechnic Institute and Stare University, USA

C. Huang, Department of Information Management, National Chi-Nan University, Taiwan

Y. H. Lee, Department of Information and Industrial Engineering, Yonsei University, KOREA

I. Ickerott, Institute for Information Management and Control, University of Osnabrueck, Germany

Y. M. Lee, IBM Thomas J. Watson Research Center, 1101 Kitchawan Road, Yorktown Heights, NY 10598, USA

H. Jung, Grado Department of Industrial and Systems Engineering, Virginia Polytechnic Institute and State University, USA

Y. T. Leung, IBM Almaden Research Center, 650 Harry Road, San Jose, CA 95120, USA

List of Contributors

xiii

L. O’Neill, HMP Department, University of North Texas, USA

M. Mahdi Tajbakhsh, Department of Mechanical and Industrial Engineering University of Toronto, Canada

N. A. Panayiotou, Section of Industrial Management and Operational Research, School of Mechanical Engineering, National Technical University of Athens, Greece

F. Teuteberg, Institute for Information Management and Control, University of Osnabrueck, Germany

S. T. Ponis, Section of Industrial Management and Operational Research, School of Mechanical Engineering, National Technical University of Athens, GREECE

D. R. Towill, Logistics Systems Dynamics Group, Cardiff Business School, Cardiff University, UK

A. Potter, Logistics Systems Dynamics Group, Cardiff Business School Cardiff University, UK

T. (Bill) Tseng, Department of Mechanical and Industrial Engineering, The University of Texas at El Paso, USA

L. Rivera, Grado Department of Industrial and Systems Engineering, Virginia Polytechnic Institute and Stare University, USA

H. Wan, Grado Department of Industrial and Systems Engineering, Virginia Polytechnic Institute and Stare University, USA

J. Sarangapani, Electrical and Computer Engineering Department, University of Missouri-Rolla, USA

E. Yücesan, INSEAD, Technology Management Area, Boulevard de Constance, France

C. Saygin, Engineering Management and Systems Engineering Department , University of Missouri-Rolla, USA

A. Z. Zeng, Department of Management, Worcester Polytechnic Institute, USA

R. W. Seifert, IMD - International Institute for Management Development Chemin de Bellerive 23, PO Box 915, CH-1001 Lausanne, Switzerland

S. Zolfaghari, Department of Mechanical and Industrial Engineering, Ryerson University, Canada

PART I. TRENDS IN TECHNOLOGIES

1 A Systems Approach to Viable RFID Implementation in the Supply Chain Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Abstract:

This article presents a systems approach to RFID deployment in a supply chain from two perspectives: (1) RFID data-based decision-making at all levels of supply chain and (2) communications infrastructure necessary to provide seamless data and information flow. The main contribution of this article is highlighting the interconnection between these two perspectives and the fact that without tackling these issues simultaneously, no viable solution can be developed. Unlike typical articles in the literature, which solely highlight the potential advantages of RFID technologies, this article aims to provide an overview of the basic challenges of deploying RFID systems in supply networks. By providing visibility, effective RFID implementation in a supply chain can bridge the gap between the shop floor and higher level operations. Such a revolutionary concept changes the traditional “linear” structured supply chains into network-centric supply chains. Based on this concept, this article presents a wide range of RFID applications, ranging from shop floor control to inventory management, and discusses technical and research challenges.

1.1 Introduction Radio frequency identification (RFID) has received a great deal of attention for its potential ability to perform noncontact object identification and to provide visibility at the point of use in a variety of different industries (Mills-Harris et al. 2006; Penttlia et al. 2004; Jones et al. 2005; Prater and Frazier 2005; Helo and Szekely 2005; Strassner and Fleisch 2005). Yet many technical and business challenges lie ahead before RFID becomes commonplace. RFID is not a new technology. It dates back to the techniques developed to differentiate “friendly” aircraft from enemy warplanes in World War II. However, recent developments in computer technology and electronics have combined to make the RFID technology potentially viable for commercial purposes.

4

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Information is the fuel that drives the economy and the society today (Wyld 2005). As manufacturing operations go increasingly global, proper coordination among business and manufacturing units can be provided by sharing information in a timely manner (Mills-Harris et al. 2006). Similarly, market and other uncertainties can be reduced and better managed by sharing information instead of building up inventories (Shaw 2000). The value of sharing information and its impact on inventory management is well studied, including Cachon and Fisher (2000), Lee and Whang (2000), Yao and Carlson (1999), Yu et al. (2001), and Moinzadeh (2002). From supply chain level operations to shop floor level manufacturing execution, deploying RFID technologies can help facilitate information sharing and provide visibility in the processes (Brewer et al. 1999; Lee et al. 2004; Michael and McCathie 2005). Further, with the existence of proper infrastructure, e.g., hardware, software, and networking, RFID technology may improve the real-time exchange of data between locations and entities in a logistics network, facilitating better and more accurate information flow. In contrast to other technologies, e.g., bar codes, information regarding the location, amount of inventory, and realized demands at each location in the network may be more easily made transparent and shared with other members in the network, thus enabling better decision-making. Based on this motivation, leading organizations, such as the U.S. Department of Defense and Wal-Mart, have set goals for their suppliers to begin using RFID on shipments to their organizations. Nevertheless, while the physics of RFID is relatively simple, there are two fundamental problems that must be addressed prior to widespread adoption: (1) to make readers and tags effectively and efficiently communicate to achieve the primary goal of seamless information flow and (2) to redesign information-driven robust business processes that can effectively use not only RFID data but similar sensor data. Though business cases for RFID technology have been discussed in many papers (McFarlane 2002; Alexander et al. 2002; Chappell et al. 2003; Lee et al. 2004), a well-rounded implementation of such a system to demonstrate the actual cost savings has not been demonstrated due to limited hardware and software capabilities and technical limitations. The Computing Technology Industry Association revealed that slightly more than half of the 500 North American organizations studied have either completed RFID implementations or plan to do so in the next 12 months, i.e., they are evaluating, pilot testing, or currently using RFID. Experimental and pilot implementations of RFID technologies in “controlled” settings represent the best possible results that can be achieved in a laboratory-like environment. Upon technology transfer to an industrial environment, unforeseen problems, such as readability, arise. Although every industry and service could potentially benefit from the enhanced decision-making capabilities provided by RFID, it has not been completely demonstrated in practical implementation, nor have effective strategies for implementing RFID been fully explored. Most importantly, many companies have been reluctant to adopt RFID because they do not see immediate benefits. Despite these problems, companies continue to invest heavily in RFID technology and are likely to continue to do so in the future. In

A System Approach to Viable RFID Implementation in the Supply Chain

5

doing so, one must identify possible solutions to these problems and provide guidance on ways to implement these solutions. This article approaches RFID deployment in a supply chain from two perspectives: (1) RFID data-based decision-making at all levels of a supply chain and (2) communications infrastructure necessary to provide seamless data and information flow. The main contribution of this article is highlighting the interconnection between these two perspectives and the fact that without tackling these issues simultaneously, no viable solution can be developed. Unlike typical articles in the literature, which solely highlight the potential advantages of RFID technologies, this article aims to (1) provide an overview on the fundamental difficulties and oversights associated with deployment of RFID systems in supply networks from decision-making and communications perspectives and (2) discuss technical and research challenges.

1.2 RFID: The Basics RFID represents a noncontact method for data transfer in object identification. RFID dates back to the 1940s when the British Air Force used the IFF (Identify Friend or Foe) system, which was based on radio-frequency technology, in World War II to identify enemy aircraft on its coasts (Landt 2001). In his paper entitled “Communication by Means of Reflected Power,” Stockman was the first to explain the theory behind RFID in 1948 (Stockman 1948). The first patent for RFID was filed by Charles Walton, a former IBM researcher, in 1973; his patent was for a radio-operated door lock (Takahashi 2004). A typical RFID system consists of three components: (1) an electronic data carrying device, called a transponder or tag, (2) antennas and readers that facilitate tag interrogation, and (3) software, called middleware, that controls the RFID equipment, manages the RFID data, and distributes information to other remote data processing systems by interfacing with enterprise applications. An RFID system can be considered a wireless communication system since the reader communicates with the tags by using electromagnetic waves at radio frequencies (Keskilammi et al. 2003). RFID systems can be categorized as active and passive systems. In an active system, the tag (i.e., active RFID tag) has its own power source, which is a battery, enclosed in the transponder housing. In a passive system, the tag does not have its own power source; instead, it draws power from the reader’s radio signals. Passive tags are inexpensive compared to active tags.

1.3 Network-centric Supply Chains Although the focus of this article is on RFID, it is important to note that RFID is only one of many possible sensors that can be “embedded” in business processes to achieve seamless information flow. Most importantly, as will be described in detail, RFID on its own cannot facilitate the desired visibility; a variety of sensors

6

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

and technologies, such as global positioning systems, should be integrated with RFID technologies to gain full benefits. The potential benefits of RFID, when implemented properly, are as follows (Mills-Harris et al. 2006; Soylemezoglu et al. 2006): x

x x

x

Instantaneous operator-free data entry and monitoring: RFID readers and antennas communicate voluminous RFID data in milliseconds and can scan multiple items simultaneously. This significantly facilitates automation of many supply chain processes that are typically labor-intensive. Effective use of labor: Many repetitious tasks can be automated via RFID, so labor can be used for more effective tasks. Visibility: RFID provides real-time visibility for products across the supply chain providing accurate and detailed information, which can be used to improve efficiency, productivity, and quality. In addition to product-level visibility, an organization can track its valuable assets by tagging them. RFID technology also provides benefits for product recalls. Mobile databases: Active RFID tags can be used as mobile databases; such tags can be updated dynamically as parts move across the supply chain.

From this perspective, we envision that future supply chains, from shop floor to top floor, will be built on a sensor-based, distributed architecture and the business processes will be driven by real-time information. We define a network-centric supply chain (NCSC) to characterize such a supply network as follows: A network-centric supply chain incorporates a dynamic network of selforganizing, autonomous assets and entities that operate, collaborate, cooperate, and compete upon basic principles of decentralization, participation, and coordination through rich information exchange. Without RFID or other similar sensors, it is not possible to extract and harvest data in a timely manner, to share information, and to be proactive in control of materials, products, processes, and systems in a supply network. In Figure 1.1, a network-centric supply chain model, that has been developed at the auto-ID Testbed in the University of Missouri – Rolla (UMR), is shown (Soylemezoglu et al. 2006). This application demonstrates a bottom-up RFID implementation for RFID data-driven supply chain management. As opposed to typical applications where the focus is on supply chain operations, the objective is to investigate how RFID data can be used to streamline supply chain-level business processes integrated with shop floor operations. At its current stage, there are three types of actors in the environment: suppliers, manufacturer, and customers. Each actor is provided with hypertext preprocessor (PHP) based Web pages that facilitate interaction with each other. The supplier receives orders for raw materials automatically from the manufacturer when the baseline inventory level drops below a certain value. The customer site can be used by several customers, similar to a typical Internet-based shopping site. Each customer can place an order, which includes an assembly of a certain number of raw materials stored in the automated storage/retrieval systems (AS/RS).

A System Approach to Viable RFID Implementation in the Supply Chain

7

Figure 1.1. Network-centric supply chain model (UMR)

The manufacturer’s facility includes industry-size production equipment: Dock doors, representing its warehouse; a closed-loop conveyor that transports raw materials to/from an AS/RS, and an assembly area. The dock door and the conveyor are equipped with RFID antennas, which are connected to an RFID reader. A PLC is used to control the shop floor operations, including the conveyor, AS/RS; and the motion sensors on the dock door to detect inbound and outbound material. In a network-centric supply chain, RFID can facilitate the implementation of new business processes due to the visibility it provides. One such application is vendor managed inventory (VMI), which is a centralized supply chain initiative where the supplier is authorized to manage inventories and to make decisions such as when and how much inventory to ship to the manufacturer or retailer. Applications in the grocery industry are among the earlier adopters of VMI, and VMI has been adopted by many organizations, including major retailers such as Wal-mart, K-mart, Sears, target, Office Depot, and Dillards. VMI is seen as an effective means of managing inventory through the strategic use of information technology built on auto-ID technologies, and it leverages advanced technology and trading-partner relationships to enable the flow of information and inventory throughout the entire supply chain. The tangible benefits of successful VMI implementation include improved forecasting by reducing uncertainty due to increased visibility across the supply chain, reduced inventory levels with higher inventory turns, and reduced costs. Driven by the business practice of VMI, integrated inventory and transportation systems have received much attention recently. Traditional inventory routing is concerned with the repeated distribution of products from a centralized distribution center to locations in the supply chain. Routes are generated based on partial information about the inventory requirements at each location, and vehicles start

8

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

from the distribution center and follow the prescribed route, fulfilling the demand at each node either partially or completely. In this scenario, potentially beneficial inventory routing changes cannot be identified due to a lack of real-time information. However, in a logistics network enabled by RFID technology, information exchange between all enabled entities in the network provides information for evaluating possible inventory routing changes. Although it is currently not practical (or desirable) to tag all entities in the supply chain, it is reasonable to assume that entities associated with inventory routing from a distribution center are adequately enabled, i.e., the product on the trucks is tagged either by item, carton, or pallet, and the distribution center, locations, and vehicles are equipped to read/write and transmit/receive information. Jaragumilli and Grasman (2005, 2006), provide a methodology for dynamic control of RFID-enabled inventory routing problems, where a set of locations is served by a single distribution center using a vendor managed inventory approach. In this scenario, RFID technology enables efficient control of inventory distribution by exchanging real-time information upon arrival at each location. The paper develops mathematical models that illustrate the substantial potential savings from implementation of this technology and provides a discussion of practical impediments to full application. In a related paper, Jaragumilli et al. (2006) propose a simulation methodology as a framework for the real-time control of inventory and routing decisions. Inventory sharing is another concept that can be facilitated by RFID technologies. The idea of sharing inventories is based on the premise that units needed by one location can be transshipped from another location to fulfill customer demand. In many cases, there are nearby locations that could easily share units. Although, this activity is normally considered nonvalue added and subsequently avoided, much research has shown that lateral transshipment may lead to significant cost savings (Bertrand and Bookbinder 1998; Grahovac and Chakravarty 2001; Herer and Tzur 2001; and Rudi et al. 2001). Grasman and Vasquez (2002, 2006) present a model of inventory and information sharing among rental locations and the option of offering the customer a substitute unit. The objective is to develop a strategy for offering units to customers that will allow the locations to improve inventory management while satisfying customer demand. The successful implementation of the model depends on the customer decision of accepting or not accepting the offer, a network of nearby locations that share information through a decision support system, and profit/cost sharing agreements. Above all, RFID provides products with a “voice” in the supply chain. Such products, which can be considered intelligent products, can possess a unique identity, store information about themselves, effectively communicate with their environment, and can make and trigger decisions relevant to their own destiny. The concept of intelligent products has been discussed by Zaharudin et al. (2002) and Wong et al. (2002) and requires investigation of new product design, production planning, and shop floor control models that are driven by products (McFarlane et al. 2003).

A System Approach to Viable RFID Implementation in the Supply Chain

9

1.4 Technical Challenges Overall, RFID is still financially, technically, and operationally infeasible for many organizations, whose supply chain, manufacturing, and logistics operations are not rationalized and standardized. Similarly, an organization experiencing difficulty in carrying out the four essential tasks discussed in Section 1.3 would first need to “clean up” its business processes and practices before rushing through an RFID implementation. RFID deployment without conducting business process analysis and (re)engineering could solve some minor problems, but it will certainly give rise to a new set of potential problems, which will be far more complicated to sort out. Without reaching a lean perspective on operations and workflow in an organization, RFID cannot bring visibility out of a chaotic environment. In other words, it is not the deployment of RFID tags, antennas, and readers by itself that will push the organizations in a supply network ahead of their competitors, but it is the way an organization uses the fine-grained and real-time RFID data to improve its business processes that will determine the extent of potential benefits to be obtained from RFID (Asif and Mandviwalla 2005). RFID hardware and software are a significant expenditure. Tagging materials and products is not sufficient; many applications require special equipment, such as forklifts, conveyors, dock-doors, equipped with RFID hardware to fully integrate RFID into operations. To keep the cost low, most small businesses adopt a “slapand-ship” approach to satisfy mandates such as those of Wal-Mart and the U.S. Department of Defense. Current RFID technology is still not mature. For instance, the presence of products or packaging material containing metal components that block the RFID signal, conveyor belts made of static producing nylon, or glass fiber that produces radio noise may necessitate expensive changes in the physical infrastructure (Margulius 2004). Such consequences caused by RF-unfriendly environments are not only costly to avoid, but also very difficult to anticipate even after completing extensive pilot studies, thus making the overall technology very risky. The complexity of implementation varies depending on the level of tagging (i.e., nested tags). Item-level versus pallet-level tagging is a critical business issue. Between the two levels, there is a significant difference in cost and number of tags to be handled, potential interference among tags, accuracy and read-rate problems, the amount of labor required to place tags, and the amount of data that needs to be handled properly. However, implementing pallet-level tagging to avoid those difficulties cannot deliver all the strategic benefits of RFID technologies. Supply chain systems require interoperability for seamless information flow. However, there is no standard related to RFID technology that meets the needs of all users. The development of standards has progressed through the formation of the EPC global network, which is a member-based organization, comprised of numerous large firms providing funding. EPC global started at MIT in 2000 as the auto-ID Center. EPC global’s operation must be backed by the International Organization for Standardization (ISO) to develop a widely accepted standard that can facilitate interoperability. Therefore, the EPC (Electronic Product Code) standard is of critical importance to the success of RFID in the supply chain.

10

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Information sharing is proposed to counteract the bullwhip effect, which is a phenomenon where distorted information from one end of the supply chain leads to excessive fluctuation in demand projections further up the chain. The most promising benefit of RFID is that it facilitates visibility by providing real-time information on identity and location. On the other hand, having tens of thousands of tags can very easily lead to a very high volume of data, which can (1) cause congestions and data packets dropped (i.e., lost) on the network, (2) lead to storage of unused data, and (3) require high computing power to sort through such voluminous data. Overall, having “too much visibility” (i.e., very high volume of data flowing at very high speed) is in effect equivalent to having no visibility if an infrastructure to use the data effectively is not available.

1.5 Research Challenges Automatic identification (auto-ID) represents a broad category of technologies that are used to identify objects without human intervention. RFID is a type of auto-ID technology. In general, auto-ID technologies, advanced sensing capabilities, and recent developments in the area of mobile wireless Ad Hoc networking provide the potential to establish a data-rich supply network. Such technological capabilities provide real-time visibility of each single entity in the supply chain; nevertheless they can be effective only if the real-time data can be integrated into the necessary business processes, such as manufacturing execution, production planning, and scheduling systems for improved decision-making. Myopic approaches, such as using auto-ID technologies for asset tracking but without integrating it with scheduling, lead to a disconnect among a variety of islands of information in a supply chain and do not improve the system performance (Soylemezoglu et al. 2006). Therefore, there is a need for comprehensive data models that facilitate intelligent decision-making among the organizations in a supply chain. RFID imposes challenges at three layers: (1) physical layer: It represents the application environment that is equipped with a variety of auto-ID technologies, advanced sensing capabilities, handheld tools, and mobile wireless networks to facilitate timely communication; (2) decision-making layer: This layer consists of effective decision-making models founded on efficient data harvesting, processing, and sharing schemes so that the performance goals at the physical layer (i.e., the application environment) can be met; and (3) networking layer: handles collecting, scheduling, and routing of voluminous data, which provides timely data and information flow for the decision-making layer. The literature shows that most studies focus either on manufacturing-specific decision-making (manufacturing engineering and industrial engineering) or on networking (electrical and computer engineering) in isolation from each other. In other words, the solutions provided in this area are focused only on a particular layer and are isolated from the other layers by making various assumptions, such as “the read-rates on RFID readers are 100%” or “no data packets are dropped at the networking layer.” Such studies fall short of being realistic or complete. Therefore, integrative architectures that tie performance metrics at various levels in a supply chain with networking-level routing and scheduling protocols are required.

A System Approach to Viable RFID Implementation in the Supply Chain

Decision-Making Module • Assess current practice • Develop alternative decisionmaking models • Validate these models

11

Networking Module 1 2

• Mimic major hardware, software, networking problems • Develop alternative solutions, such as protocol, topology, etc. • Validate protocol and assess network performance

3 Hardware-in-the-loop Testing 4 VIABLE SOLUTIONS “Cost Savings & Improved performance”

Figure 1.2. solutions

UMR’s auto-ID Research Group’s approach to providing viable auto-ID

The approach adopted by the auto-ID Research Group at UMR for generating viable auto-ID solutions is depicted in Figure 1.2 (Soylemezoglu et al. 2006). It includes two major activities so that the research and development work can be carried out in realistic and accurate conditions. In the decision-making module, the current business process is analyzed. After initial data collection about the process, a simulation model is developed to carry out “what-if” scenarios. As a result of this approach, various alternative decision-making models that rely on a certain level of auto-ID data are developed. The information flow required by each model is then communicated to the networking module, which mimics, through network simulation, the possible load on the network. In this module, various data scheduling and routing protocols are developed and tested using simulation. These two modules, decision-making and networking, provide an in-depth analysis of all possible solutions. The most reliable networking scheme and the most promising decision-making model are then combined for hardware-in-the-loop testing. Finally, a viable solution emerges after hardware-in-the-loop testing is successfully completed. In general, RFID adds a new level, which consists of fine-grained data, to the traditional research and development methodologies. This new level imposed by RFID necessitates more sophisticated schemes for process modeling and data management. 1.5.1 Data Management and Effective Decision-making RFID applications create terabytes of real-time data. Collecting such voluminous data at a very high speed, processing it and storing it impose severe challenges on current data management and storage strategies. Timely communication of RFID

12

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

data through business processes among organizations in a supply chain provides potential for seamless integration. In this area, research is essential to identify the issues, develop alternative solutions, and conduct cost-benefit analysis. RFID systems are too complex to allow evluating realistic models analytically. Therefore, simulation-based approachs are essential for modeling and analyzing of such systems, where a simple closed-form analytical solution through the use of a mathematical model is not possible due to the highly stochastic natures of these systems.

Figure 1.3. Hardware-in-the-loop simulation methodology: sample application

Due to the limitation and uncertainty of RFID technology, simulation models that simply focus on RFID data-based decision-making, in isolation from actual hardware, can provide only limited insights. Prototypes that integrate RFID hardware, as well as decision-making, are essential to generate more realistic research results. Driven by this need, a hardware-in-the-loop simulation methodology has been developed by the auto-ID Research Group at UMR, as shown in Figure 1.3 (Soylemezoglu et al. 2006). The fundamental motivation behind such an approach has stemmed from the fact that to demonstrate the potential benefits of integrating auto-ID technologies, a testbed needs to be equipped with a variety of industry-grade production equipment and auto-ID hardware and software tools, which is not a realistic expectation for two reasons: Such equipment (1) has a very high overall cost and is not usually flexible enough to be used in a variety of different configurations for a variety of different industrial applications, therefore it is not cost effective; and (2) do not allow for expedited testing (i.e., testing alternative models in a very short time). Therefore, there is a need to develop “virtual” hardware that can mimic real hardware in a cost-effective way and also allow for developing simulated production and logistics scenarios for expedited testing. An additional advantage of such virtual models is that they can be run on distributed computers, similar to a realistic industrial setting, which then allows for networking and communications related testing. UMR’s auto-ID Testbed is equipped with virtual models that can communicate with real hardware, which

A System Approach to Viable RFID Implementation in the Supply Chain

13

leads to the concept of hardware-in-the-loop simulation models. Such an approach provides a dynamic, controlled testbed environment for developing, testing, and evaluating auto-ID systems, and it enables development of guidelines for technology transfer. Virtual models can be used to mimic actual hardware and they can generate a large amount of data, which are essential for developing realistic solutions to industrial problems. For example, data generated by using such a combination of actual hardware and virtual models can provide a basis for low-cost experimentation on network overloading. To make effective decisions with the data generated in a hardware-in-the-loop simulation environment, data must be properly communicated among decision-makers (i.e., controllers). Therefore, reliable data scheduling and routing schemes are required for reliable data transfer. 1.5.2 Networking The networking layer acts as a bridge between the decision-making layer and the physical layer. It consists of two sublayers: medium access control (MAC) and link. Effective routing and scheduling of RFID data and control packets pertaining to decision variables is critical. Due to the nature of auto-ID technologies and abundant data communication needs, a mobile wireless sensor network is essential to provide flexibility, scalability, and self-reconfigurability since voluminous RFID data must be transported through the network to a variety of business processes in a supply chain. To make timely decisions, data to/from the decision-making layer must be communicated in real-time satisfying the network quality of service (QoS) requirements. Note that sensor data, in general, from a transmitter during wireless communication would face interference from other transmitters. Due to mobility of RFID tag-embedded products and other entities with radio-frequency motes, even operators with their electronic badges, the networking layer is treated as a mobile multihop network. Under such uncertain mobile environments, a proactive data routing scheme is necessary. Moreover, energy efficiency must be considered in routing since these mobile resources are powered by batteries. Transmitting wirelessly will improve safety and facilitate mobility since wiring is eliminated. Existing wireless sensor networking protocols cannot be applied to RFID-based environments since they are not optimal and fail to handle large quantities of data generated by RFID tags and wireless sensor-based equipment. Analytical results (Ge et al. 2002) ensuring the network performance objectives are not currently available with available networking protocols. To make timely decisions and to meet certain performance measures in a supply chain, route discovery and maintenance for the networking layer is an important issue in mobile wireless sensor networks. And it is typically performed using routing-related control packets. Within the field of mobile wireless networks, the optimized routing problem can be defined as finding a distributed routing scheme so that under the connectivity assumption, any mobile host can transmit/receive information to/from any other host in the network while minimizing overhead and resources. At the same time, it should be capable of rapidly adapting to link failures and additions caused by node movements. Based

14

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

on past studies (Ge et al. 2002), a proactive protocol appears to be more suitable for Ad Hoc networks with high node mobility and channel fading conditions so that QoS provisioning can be performed. Several routing protocols (Ryu and Cho 2001; Chang and Tassiulas 2000; Gerla et al. 2000; Pursely et al. 2000; Gomez et al, 1999) attempt to reduce the control overhead and some rely on constructing energy efficient trees (Torgerson and Leeuwen 2001) for connectivity. Others use the load (traffic) with either the shortest path or minimum hops to find a route. Recently, an optimized link state routing (OLSR) protocol, a proactive QoS routing protocol (Clausen 2001; Ge et al. 2002) was introduced to optimize control overhead by selectively flooding control messages via multipoint relay selectors and transmitting periodic control messages in the event of link failures and additions. It was shown in (Ge et al. 2002) that the OLSR results in better performance compared to DSR (Johnson and Maltz 1995) and AODV (Clausen 2001) and it scales better with network size. OLSR uses the number of hops as a metric when deciding an optimal route; however, selecting a route using the number of hops is not optimal in terms of energy used to transmit a packet and end to end delay encountered. The energy efficiency or transmitter power, end to end delay, and channel state have to be considered during route selection. An RFID network renders a distributed sensor network. Similar to all other wireless systems, interference is one of the major issues faced by RFID networks and needs to be investigated. Interference can prevent readers from achieving the desired read rate and interrogation range. For instance, in a warehouse, interference can lead to uncertainty in terms of location of tagged items, where they digitally “disappear,” or a tagged item can appear to be in two locations at the same time. Such false reads deteriorate effective decision-making in a supply chain. Further, in passive RFID systems, tags harvest energy from the carrier signal, which is obtained from the reader to power internal circuits. Moreover, passive tags do not initiate any communication, but they respond only to reader commands through backscatter communication (Rao 1999). In a backscatter process, information from tags is modulated onto the reader carrier signal and reflected back to the reader. When multiple readers are deployed in a working environment, signals from one reader may reach others and cause interference, which prevents a reader from decoding tag information. This RFID interference problem is described in Engels (2002) as reader collision, which is frequently encountered with RFID systems and is classified as frequency and tag interference. Frequency interference occurs when readers operating in the same frequency channel introduce high noise levels and jam the on-going communication with tags. Tag interference occurs when tags are being read by multiple readers simultaneously regardless of the differences in frequency. Consequently, reader collision causes unreadable tags and disturbs the normal operation of the readers, lowering the overall read rates. Tag interference can be avoided when reader interrogation zones do not overlap or the readers access the tags in different time slots. Frequency interference can be avoided if the colliding readers operate at different frequencies. Else, noise power

A System Approach to Viable RFID Implementation in the Supply Chain

15

level at each reader must fall well below the tag backscatter power level to decode tag information successfully. 1.5.3 Examples In this section, four examples are presented. The examples have been selected to emphasize different aspects of RFID implementation within a supply chain. The first example is related to logistics. The second presents a warehouse application with emphasis on dock doors. The third example is a shop floor inventory management application. The last example ties RFID with shop floor control by introducing conveyors, automated storage and retrieval systems, controllers, and real-time decision-making. Example 1) RFID-Enabled Inventory Routing: In this scenario, a methodology for dynamic control of RFID-enabled inventory routing problems is considered (Jaragumilli and Grasman 2005, 2006). The problem includes a set of locations served by a single distribution center using a vendor managed inventory approach. It is assumed that all inventory items are tagged and the distribution center and locations have RFID-equipped gates that read the content of trucks. The inventory routing decision flowchart is shown in Figure 1.4. Within this framework, there are two decision epochs: (1) generation of initial feasible (optimal) delivery quantities and routing using classical inventory and routing models and (2) determination of feasible delivery quantity and the next location to visit based on updated information. This problem can be modeled using stochastic dynamic programming, where the state space is defined as the current location of the delivery vehicle, the predicted and actual inventory levels at each location, and the amount of inventory remaining on the vehicle. The decision variables are the quantity to deliver at each location and the choice of the next route. Thus, the objective function is given by (1.1):

~ f t (k , I kt , I kt 1 , Q t )

~

t t 1 t 1 t 1 min ^ ckj  Yk ( Dk )  f t 1 ( j, I j , I j , Q )`,

(1.1)

jk c 0d Dkt dQt

where



Yk D

t k

Dkt  I kt

hk

¦ ( D d 0

t k

f



 I kt )  d ) O ( d )  bk d

¦ (d  ( D Dkt

t k

 I kt ))) O ( d ).

(1.2)

 I kt

Equation (1.2) provides the inventory control policy cost component of (1.1), including inventory carrying cost and shortage cost associated with the delivery quantity.

16

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Start

Obtain initial inventory levels at each location in the distribution network

RFID-equipped Gate

Generate partitioned set of locations to be visited

Generate initial feasible (optimal) delivery quantities and routing

Update information about all locations in the network on reaching the location through RFID-equipped gates (SEE LEFT)

Determine delivery quantities and the routing decisions based on the objective function given below:

~ f t ( k , I kt , I kt 1 , Q t )

~

t t 1 t 1 t 1 min ^c kj  Yk ( Dk )  f t 1 ( j, I j , I j , Q )`

jk c 0 d Dkt dQ t

Where Yk Dkt hk

Dkt  I kt

¦ ( D

t k

 I kt )  d ) O ( d )  bk

f

¦ (d  ( D

t k

 I kt ))) O (d )

d Dkt  I kt

d 0

Update vehicle inventory

No

All locations in the network are visited or inventory on vehicle is zero? Yes Return to distribution center

Stop

Figure 1.4. Inventory routing decision flowchart

Table 1.1 shows the list of notation for (1.1) and (1.2). It note than any setup costs may be included in the operations cost, cij. The model relies on the robustness of the RFID-equipped gates to read RFID tags on inventory items. One major problem is the read-rate problem at the gate, which could lead to incomplete or false data for the contents of trucks passing through. The same problem also exists at warehouses, where a number of RFIDequipped dock doors are used for receiving and shipping operations. This problem is elaborated in the following example.

A System Approach to Viable RFID Implementation in the Supply Chain

17

Table 1.1. List of notation for RFID-enabled inventory routing model

kK cij dij tij ak bk hk pk ĭkȜ(d) șk

Vk (R,s,S)k t Ikt Ƭkt Qt Dkt xijt

Network representation k=0 represents the distribution center k=1, 2, 3…K represent the wholesaler locations cost of operating between node i and node j in the network distance between node i and node j in the network travel time from node i to node j in the network Inventory modeling parameters for location k set up cost for each order shortage cost per unit Holding cost of inventory unit cost Distribution of demand during time period Ȝ expected demand during lead time standard deviation of demand during lead time R = review interval, s = reorder point, S = order up to quantity State notation Index of node visitation Actual inventory at location k upon arrival at the tth node predicted inventory at location k upon arrival at the tth node inventory on vehicle upon arrival at the tth node Decision variables inventory delivered to location k while at the tth node binary variable associated with selection of the route associate with selection of the next route = 1 if t = i and t + 1 = j, 0 otherwise

Figure 1.5. RFID-equipped dock doors in a warehouse

Example 2) RFID-Enabled Dock Doors: Dock doors can be equipped with RFID antennas, light stacks, and motion sensors to fully use the benefits of RFID technology for shipping and receiving operations in a warehouse, as shown in Figure 1.5 (Soylemezoglu et al. 2006). Two major problems exist for this scenario: (1) the read-rate problem, which is inherent in any RFID application and (2) interference and cross-reading of material, which refers to the case where several dock doors read each other’s material due to the range of

18

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

the RFID antennas, and location and identity information cannot be matched. These problems are caused by tag interference and frequency interference. To avoid tag interference, readers must operate in different time slots if their interrogation zones overlap. To avoid frequency interference, readers should operate in different time slots or different frequency channels. The basic dock door operation involves the following steps. Each material has an RFID tag and all materials are placed on industrial pallets. First, the motion sensors detect the direction of the pallet (i.e., inbound or outbound material flow). The information is sent by wirelessl to an RFID server (i.e., middleware). It turns on the red light on the light stack for the forklift to stop at the dock door. Then, the server initiates the RFID tag reading process and passes the readings to the inventory control computer, which compares the order list with the pallet content to determine if the right materials in the right amounts have been received. After reading the tags, there can be a mismatch between the order list and the pallet content, which can be due to two possibilities. First, the pallet content may be correct but the readings, due to read-rate problems, can be wrong. Second, the pallet may include an incomplete shipment. In either case, the material flow is not disrupted at the dock door, but the inventory database is updated to reflect this anomaly. When the pallet reaches its destination, inventory items on the pallet are placed in RFID-equipped storage spaces. During this process, the content of the pallet is checked to find whether the anomaly detected at the dock door still exists. Human intervention is necessary if the problem is not resolved. This scenario has been tested at UMR’s auto-ID Testbed. The dock doors use Class-0 900 MHz UHF RFID readers and tags, which are known to be prone to interference. High-power devices, such as RFID readers, placed in close proximity, cause interference among themselves and for other RF devices (i.e., a wireless transmitter for a motion sensor). In dense networks, detection range and read rates of RFID readers suffer severely and other lower power RF devices become completely unusable. Similarly, dock doors located next to each other form a dense RFID reader network, and therefore it is not possible to obtain reliable identification data. In addition, when a pallet, loaded with tagged materials, is moved parallel to the dock doors, the antennas on both dock doors read the tags. In other words, it is not possible to detect whether the material is being moved through a dock door or it is being maneuvered in front of the dock doors. To solve both problems, a “read it when you need it” concept has been developed, which is implemented using an on/off power control protocol (Soylemezoglu et al. 2006). The protocol enables turning on and off RFID antennas intelligently to avoid communication signal interference depending on the status of the motion sensors on dock doors. The antennas are turned on and off whenever needed by making sure that only one dock door is active at any one time, so that cross-reading is avoided. The power control on antennas and management of forklifts are synchronized by the use of light stacks (i.e., similar to traffic lights). The interference problem can be expanded to a distribution center environment that has RFID-equipped gates, similar to Example 1. For such an application, the power on the antennas must be controlled properly to maximize read rates. Due to the large amount of inventory on a truck passing through an RFID-equipped gate,

A System Approach to Viable RFID Implementation in the Supply Chain

19

the inventory content can be stored in the form of nested tags. An active tag that stores the current content of the truck can be used to verify the amount of inventory items that have passive tags. A second RFID-equipped gate can also be used to avoid the probability of a false read. Regardless of the application, the process of reading/writing RFID tags must be effectively designed so that the process itself becomes visible. A graph theory based approach can be developed to avoid the interference problem. A reader collision problem is usually presented as a graph G = (V, E) where V is the set of nodes in a reader network and E, the set of edges representing unique unordered pairs of nodes colliding with each other. Let dt and df be minimal distance for tag interference and frequency interference respectively; let D(vi, vj) be the distance between two reader stations vi and vj. Tag interference will occur if dt is greater then D(vi, vj). The scenario demonstrated in Figure 1.6 describes a tag interference problem where dt = D1 + D2 > D(vi, vj).

Figure 1.6. Tag Interference

A reader collision graph, whose connected edges satisfy the requirement dt > D(vi, vj), can now be obtained. The task becomes obtaining the minimal number of groups in which no two nodes from the same group have connected edges. In graph theory, this can be described as a graph coloring problem. Coloring of a simple graph is the assignment of a color to each vertex of the graph so that no two adjacent vertices are assigned the same color. Using the graph coloring algorithm from Rosen (1998), the minimal number of groups with no tag interference can be found if the geographic information for each node and collision distance dt are given. Figure 1.7 demonstrates a graph representation of a reader collision problem along with the solutions of minimal coloring. Using the graph coloring algorithm, the minimum number of colors used in this graph is four. These colors now can be assigned to different time slots to avoid tag interference.

20

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Figure 1.7. A reader network presented as a graph with 10 nodes spanning an area of 10 m โ 10 m; collision distance equals 3 (Due to the black and white print of this book, different “shapes” have been used instead of “colors”)

The graph coloring algorithm (Rosen 1998) can be given as follows: 1) Assign color 1 to the vertex with the highest degree. 2) Also assign color 1 to any vertex that is not connected to this vertex. 3) Assign color 2 to the vertex with the next highest degree that is not already colored. 4) Also assign color 2 to any vertex not connected to this vertex and is not already colored. 5) If uncolored vertices remain, assign color 3 to the uncolored vertex with the next highest degree and other uncolored, unconnected vertices. Proceed in this manner until all vertices are colored. It is important to note that the minimal distance for frequency interference exceeds the requirement imposed by the tag interference, which is dt < df. Therefore, it is still possible for frequency interference to occur within groups. If the above graph coloring algorithm using df > D(vi, vj) as the edge requirement is applied, then a complete interference-free graph can be obtained. However, the number of groups (time slots) will increase, which would cause long waiting times and read-rate problems. It is obvious that an equilibrium point, where the number of time slots and the interference probability both become acceptable, must lie between dt and df. Example 3) RFID-Enabled Inventory Management: Although RFID provides visibility in terms of products and inventory items, certain processes do not allow direct RFID or any other auto-ID implementation, thus they lack visibility. For instance, manufacturing lead times can be difficult to

A System Approach to Viable RFID Implementation in the Supply Chain

21

predict when they involve several stochastic parameters. Similarly, purchasing lead times can vary drastically depending on the manufacturing technology of the items to be purchased. Therefore, expecting “full visibility” upon deploying RFID is not realistic; RFID-based implementations require an effective combination of traditional probabilistic models and deterministic approaches. Consider inventory management of time-sensitive materials. One primary concern is to ensure that the required materials are available at all times to the operators. The lack of time-sensitive materials results in loss of production and in turn loss of profits. On the other hand, those materials that are not used in production within their lives expire and become another cost factor. In addition, disposal of time-sensitive materials to prevent their usage on a product, once they reach their shelf life, is also major concern. UMR’s auto-ID research group conducted a study on this topic for a manufacturing company (Mills-Harris et al. 2005). The primary objective of the manufacturing company was to meet production demand at the highest possible rate (i.e., maximize service level) while avoiding excessive waste due to expired materials. After several site visits to the manufacturing company and conducting interviews with operators, shop floor expeditors, and managers, three inventory management scenarios were developed. The scenarios, including the current practice at the manufacturing company, are as follows: SCManual,BL: This scenario includes the existing manufacturing environment and directly mimics the current inventory management practice, which is heavily manual. The current practice relies on baseline (BL) inventory levels that have been fixed prior to production for each storage area. On a weekly basis, the inventory level at each storage area is checked and an order is placed to bring the inventory level back to its baseline level. SCManual,BL is used as a reference for comparing the other three scenarios. SCRFID,BL: In this scenario, it is assumed that the inventory items are tagged (i.e., RFID tags); therefore, there is theoretically 100% visibility in terms of inventory levels. Fixed BL values, similar to SCManual,BL, are used for ordering purposes. The objective is to determine how much value RFID data can add if it is implemented only for monitoring purposes, not for dynamically adjusting the inventory levels. SCRFID,BL/2: Similar to the first two scenarios, this scenario is also built on fixed baseline inventory levels. The only difference is that the baseline inventory levels are cut in half to investigate the effect of reduced inventory level. For instance, this scenario should allow tighter inventory control due to real-time product visibility since all inventory items are tagged but would require frequent reallocations among storage areas since inventory levels are reduced. SCRFID,Įȕ: In this scenario, an adaptive inventory scheme, which is based on a forecasting model, has been integrated into a decision-making framework to replace manual inventory operations and to avoid the inflexibility of fixed baselines. Since RFID, in theory, provides 100% inventory visibility, the only two stochastic parameters within the problem definition are purchasing lead time and production demand. The objective is to use the RFID data in the best possible way to handle uncertainties in purchasing lead time and adapt to fluctuations in production demand. Therefore, a trend-adjusted exponential smoothing algorithm

22

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

has been adopted. It uses two parameters, Į and ȕ, as coefficients for the average production demand and its trend, respectively. The adaptive inventory scheme looks at the difference between the current inventory level of a particular inventory item in a storage area and the associated forecast (i.e., predicted demand) to determine the amount of material that must ordered. The equations for the forecasting algorithm are given below:

At ,i , j

DDt ,i , j  (1  D )( At 1,i , j  Tt 1,i , j ) ,

(1.3)

Tt ,i , j

E ( At ,i , j  At 1,i , j )  (1  E )Tt 1,i , j ,

(1.4)

Ft 1,i , j

At ,i , j  Tt ,i , j ,

(1.5)

where Dt,i,j = demand in period t for material i at storage area j. At,i,j = exponentially smoothed average of the series in period t for material i in storage area j. Tt,i,j = exponentially smoothed average of the trend in period t for material i in storage area j. Į = smoothing parameter for the average (0 < Į < 1). ȕ = smoothing parameter for the trend (0 < ȕ < 1). Ft+1,i,j = predicted demand for material i at storage area j in period t + 1. A simulation study was carried out to compare the performance of the four scenarios. The results validate the advantage of using the proposed adaptive inventory management model (SCRFID,Įȕ) by proving the ability to lower manufacturing costs, reduce inventory levels, and prevent excessive waste in typical manufacturing environments where RFID technologies can be used. In addition, the study shows that RFID technology facilitates the vendor managed inventory (VMI) concept, which is a centralized supply chain initiative where the supplier is authorized to manage inventories at the manufacturer’s site and to make decisions, such as when and how much inventory to ship to the manufacturer. In this way, the manufacturer does not “own” the inventory until it is used in production, similar to vending machines. On the other hand, the supplier, due to visibility, can manage the inventory better and make decisions such that both the supplier and manufacturer benefits. Example 4) RFID Data-Driven Shop Floor Control: At UMR’s auto-ID Testbed, a shop floor environment that consists of dock doors, automated guided vehicles, a conveyor system, an automated storage and retrieval system (AS/RS), and an assembly area has been equipped with RFID systems to test a variety of decision-making models and networking protocols in the presence of RFID data, as shown in Figure 1.1. A programmable logic controller (PLC) and three computers, which are a cell control computer, an RFID middleware computer, and a production planning and control (PP&C) computer, are used to run the system.

A System Approach to Viable RFID Implementation in the Supply Chain

23

This application is the first step toward bottom-up RFID implementation for RFID data-based supply chain management. The ultimate objective is to demonstrate how RFID data can be used to streamline supply chain-level business processes integrated with shop floor operations. At its current stage, there are three actors in the environment: Supplier, manufacturer, and customer. Each actor is provided with php-based Web pages that facilitate interaction with each other. The supplier receives orders for raw materials automatically from the manufacturer when the baseline inventory level drops below a certain value. The customer site can be used by several customers, similar to a typical Internet-based shopping site. Each customer can place an order, which includes an assembly of a certain number of raw materials stored in the AS/RS. The focus of this section is on the manufacturer’s facility, which includes industry-size production equipment: Dock doors, representing its warehouse, a closed-loop conveyor that transports raw materials to/from an AS/RS, and an assembly area. The dock door and the conveyor are equipped with RFID antennas, which are connected to an RFID reader. A PLC is used to control the shop floor operations, including the conveyor, AS/RS, and the motion sensors on the dock door to detect inbound and outbound material. A cell controller computer is used for the following purposes: (1) Act as an HMI (human-machine interface) for the operator(s), (2) execute the control logic that governs the production cell, (3) store local events in a database, and (4) handle communication to/from the PLC by an OPC server and XML-DA gateway. On another computer, an RFID middleware package resides, which serves as a filter of RFID data and facilitates forwarding the data to the correct recipient (system/software). A production planning and control computer is used to place orders to the suppliers, receive orders from customers, and monitor material flow within the facility. The various computers in this arrangement communicate over the Intranet, which is typical in many industrial settings due to the geographic locations of facilities.

1.6 Conclusions This article presents an overview of the technical and business implications of adopting RFID in supply chains from an operations perspective. The literature shows that the sales, service, and marketing aspects of a supply chain can also benefit from RFID. In general, potential RFID applications include shipping tracking, warehouse management, delivery confirmation, routing, lot traceability, cycle counting, asset tracking, product life cycle management, recalls, and billing to list a few. Although the year 2004 marked a significant shift toward RFID due to mandates by Wal-Mart and the U.S. Department of Defense, technological barriers, such as lack of standards, compatibility, interference, lack of efficient decisionmaking models and integration road maps, and inflated expectations, as well as cost, have worked against widespread adoption of RFID technologies. Our experience suggests that the primary roadblock to successful of RFID implementation is starting without a comprehensive analysis of existing business rules and practices, which leads to ill-structured and misleading business cases.

24

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

The second mistake is having a myopic perspective by focusing only on RFID and ignoring integration of other sensors to facilitate meaningful data and information flow; RFID, when implemented alone, can provide limited visibility, which sometimes could be misleading, as well. The third mistake is the expectation to do business as usual without investigating new business rules that are driven by realtime RFID data. Last, but not least, lack of an information flow model founded on robust networking, efficient data routing and scheduling schemes, and flexible middleware also hinder achieving the potential benefits of RFID technologies.

1.7 Guidelines to Practitioners The experiments conducted at the auto-ID Testbed in UMR suggest that there are four essential tasks that need to be executed in an integrated manner for a successful RFID implementation: x

x

x

x

A comprehensive analysis, similar to a SWOT (strengths-weaknessesopportunities-threats) analysis of the work and information flow, existing business rules and practices, and current decision-making models used by the organization in the supply chain must be carried out. Such an analysis will reveal the dark spots in the processes and help prioritize the areas that can benefit from RFID. Based on the results of the comprehensive analysis, a technical feasibility study that investigates the integration of RFID and other technologies in those high-priority areas must be conducted. The technical feasibility study must focus on the following: (a) What information must be made available to make the processes more efficient and visible? (b) What are the potential interference and incompatibility issues associated with the RFID and other sensor technologies, including RFID middleware and other supplementary software?, and (c) What kind of a networking infrastructure, along with data routing and scheduling schemes, is essential to facilitate the necessary information flow in the most reliable and efficient way? An integrated product-process-system (re)design approach must be adopted to investigate, design, develop, and implement new business processes that rely on RFID and other sensor technologies. At the product level, proper tag placement alternatives that improve read rates must be investigated. At the process level, location of RFID antennas/readers must be properly determined so that the interference among antennas is minimized. At the system level, the flow of tagged items must be synchronized with RFID tag reading frequencies so that the right information at the right time at the right location is captured and used to improve the overall performance of the organization. Such an integrated product-process-system (re)design approach must present an in-depth analysis of performance measures, such as cost savings, to yield viable alternative solutions for the organization. Pilot implementations in actual production and logistics environments must be carried out to verify the comprehensive and technical feasibility analyses and to validate the integrated product-process-system (re)design.

A System Approach to Viable RFID Implementation in the Supply Chain

25

These four steps require very strong managerial commitment to make the whole effort worthwhile for a company. “Islands of RFID applications” developed within departments but lacking integration at the company level will lead to a waste of resources. To achieve an integrated RFID implementation first company-wide then across the supply chain, the main driver must be a strategic plan. Therefore, RFID implementation is more of a managerial decision than a technical one. RFID is an enabling technology; it does not automatically bring business solutions to a supply chain. RFID technology simply facilitates visibility in a process. Tools, standards, and road maps that lead to effective utilization of such visibility to improve performance, reduce cost, and expedite decision-making are crucial for a successful RFID implementation.

1.8 Acknowledgements This research was partially funded by the Air Force Research Lab (FA8650-04-C704) through the Center for Aerospace Manufacturing Technologies (CAMT) at the University of Missouri-Rolla. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Air Force Research Lab or the Center for Aerospace Manufacturing Technologies.

1.9 References Alexander K, Gilliam T, Gramling K, Kindy M, Moogimane D, Schultz M, Woods M, (2002) Focus on the Supply Chain: Applying Auto-ID within the Distribution Center. MIT Auto ID Center, IBM-AUTOID-BC-002. Asif Z, Mandviwalla M, (2005) Integrating the Supply Chain with RFID: An Technical and Business Analysis. Communications of the Association for Information Systems 15:393–427. Bertrand L, Bookbinder, J. (1998) Stock Redistribution in Two-Echelon Logistics Systems. Journal of Operational Research Society 49(9): 966–975. Brewer A, Sloan N, Landers T.L, (1999) Intelligent Tracking in Manufacturing. Journal of Intelligent Manufacturing 10(3-4): 245–250. Cachon G, Fisher M, (2000) Supply Chain Inventory Management and the Value of Shared Information. Management Science 46(8): 1032–1048. Chang JH, Tassiulas L, (2000) Energy Conserving Routing in Wireless Ad Hoc Networks. INFOCOM, March. Chappell G, Durdan D, Gilbert G, Ginsburg L, Smith J, Tobolski J, (2003) Auto-ID on Delivery: The Value of Auto-ID Technology in the Retail Supply Chain. MIT Auto ID Center, ACN-AUTOID-BC-004. Clausen T, Jacquest P, Laouiti A, Muhlethaler P, Qayyum A, Viennot L, (2001) Optimized Link State Routing Protocol. IEEE National Multi-Topic Conference INMIC. Engels DW, (2002) The Reader Collision Problem. MIT Auto ID Center, MIT-AUTOIDWH-007.

26

Can Saygin, Jagannathan Sarangapani, and Scott E. Grasman

Ge Y, Kunz T, Lamont L, (2002), Quality of Service Routing in Ad Hoc Wireless Networks Using OLSR. Proceedings of the 36th Hawaii International Conference on Systems Sciences. Gerla M, Pei G, Lee S, (2000) Wireless Mobile Ad Hoc Network Routing. Proceddings of the MOBICOM. Gomez J, Campbell AT, Naghshineh M, Bisdikian C, (1999) PARO: Power-aware Routing in Wireless Packet Networks. Proceedings of the 6th IEEE International Workshop on Mobile Multimedia Communications. Grahovac J, Chakravarty A, (2001) Sharing and Lateral Transshipment of Inventory in a Supply Chain with Expensive Low-Demand Items. Management Science 47(4): 579– 594. Grasman SE, Vasquez I, (2002) Strategic Inventory Allocation for Vehicle Rental Industries. DSI Annual Conference, San Diego. Grasman SE, Vasquez I, (2006) Strategic Inventory Sharing and Allocation: A Case of the Vehicle Rental Industry, Working Paper. Helo P, Szekely B, (2005) Logistics Information Management – An Analysis of Software Solutions for Supply Chain Co-ordination. Industrial Management and Data Systems 105(1-2): 5–18. Herer Y, Tzur M, (2001) The Dynamic Transshipment Problem. Naval Research Logistics 48(5): 386–408. Jarugumilli S, Grasman S.E, (2005), An Integrated Inventory and Routing Decision Framework in One-Warehouse Multi-Retailer Systems Using RFID Technology. International Conference on Operations Research Applications in Infrastructure Development (Convention of Operation Research Society of India), Bangalore, India. Jarugumilli S, Grasman, SE, (2006) RFID-Enabled Inventory Routing Problems. International Journal of Manufacturing Technology and Management – Special Issue on Connective Technologies and Their Impact on Manufacturing and Logistics (to appear 2006). Jarugumilli S, Grasman SE, Ramakrishan S, (2006) A Simulation Framework for Real-Time Management and Control of Inventory Routing Decisions. Proceedings of the 2006 Winter Simulation Conference. Johnson D, Maltz DA, (1994) Dynamic Source Routing in Ad Hoc Wireless Networks. Mobile Computing, Kluwer Academic. Jones P, Clarke-Hill C, Comfort D, Hillier D, Shears P, (2005) Radio Frequency Identification and Food Retailing in the UK. British Food Journal 107(6): 356–360. Keskilammi M, Sydanheimo L, Kivikoski M, (2003) Radio Frequency Technology for Automated Manufacturing and Logistics. Part 1: Passive RFID Systems and the Effects of Antenna Parameters on Operational Distance. International Journal of Advanced Manufacturing Technologies 21: 769–774. Landt J, (2001) Shrouds of Time: The History of RFID. White paper from AIM Global, October. Lee H, Whang S, (2000) Information Sharing in a Supply Chain. International Journal of Technology Management 20(3-4): 373–387. Lee YM, Cheng F, Leung YT, (2004) Exploring the Impact of RFID on Supply Chain Dynamics. Proceedings of the 2004 Winter Simulation Conference 2: 1145–1152. Margulius DL, (2004) The rush to RFID. InfoWorld, April(15): 36–41. McFarlane D, (2002) Auto ID-based Control Systems: An Overview. IEEE International Conference on Systems, Man and Cybernetics 3: 6–11. McFarlane D, Sarma S, Chirn JL, Wong CY, Ashton K, (2003) Auto ID Systems and Intelligent Manufacturing Control. Engineering Applications of Artificial Intelligence 16: 365–376.

A System Approach to Viable RFID Implementation in the Supply Chain

27

Michael K, McCathie L, (2005) The Pros and Cons of RFID in Supply Chain Management. International Conference on Mobile Business 623–629. Mills-Harris MD, Soylemezoglu A, Saygin C, (2006) Adaptive Inventory Management Using RFID Data. International Journal of Advanced Manufacturing Technology (to appear). Mills-Harris MD, Soylemezoglu A, Saygin C, (2005) RFID Data-based Inventory Management of Time-Sensitive Materials. The 31st Annual Conference of the IEEE Industrial Electronics Society (IECON’05) Special Session: Integrated Manufacturing and Service Systems, Raleigh, North Carolina, Nov. 6–10. Moinzadeh K, (2002) A Multi-Echelon Inventory System with Information Exchange. Management Science 48(3): 414–426. Penttlia KM, Engels DW, Kivikoski MA, (2006) Radio Frequency Identification Systems in Supply Chain Management. International Journal of Robotics and Automation 19(3): 143–151. Prater E, Frazier GV, (2005) Future Impacts of RFID on e-supply Chains in Grocery Retailing. Supply Chain Management-An International Journal 10(2): 134–142. Pursely MB, Russel HB, Wysocarski J.S, (2000) Energy-efficient Transmission and Routing Protocols for Wireless Multihop Networks and Spread Spectrum Radios. EUROCOMM 1–5. Rao KVS, (1999) An Overview of Back Scattered Radio Frequency Identification Systems (RFID). Proc. of the IEEE Microwave Conference 3: 746–749. Rosen KH, (1998) Discrete Mathematics and Its Applications. McGraw-Hill Science/Engineering/Math, 4th edi. Rudi N, Kapur S, Pyke D, (2001) A Two-Location Inventory Model with Transshipment and Local Decision-making. Management Science 47(12): 1668–1680. Ryu J-H, Cho D-H, (2001) A New Routing Scheme Concerning Energy Conservation in Wireless Home Ad Hoc Networks. IEEE Transactions on Consumer Electronics 47. Shaw MJ, (2000) Information-Based Manufacturing with the Web. International Journal of Flexible Manufacturing Systems 12(2/3):115–129. Soylemezoglu A, Zawodniok MJ, Cha K, Hall D, Birt J, Saygin C, Sarangapani J, (2006) A Testbed Architecture for Auto-ID Technologies. Assembly Automation 26(2): 127–136. Stockman H, (1948) Communication by Means of Reflected Power. Institute of Radio Engineers (IRE). Strassner M, Fleisch E, (2005) The Potential Impact of RFID on Supply Chain Management. Wirtschaftsinformatik 47(1):45–54. Takahashi D, (November 23, 2004) The Father of RFID: Walton’s Work Paved Way for Ubiquitous Tracking Chips. Mercury News (www.siliconvalley.com). Torgerson M, Leeuwen BV, (2001) Routing Data Authentication in Wireless Ad Hoc Networks. Sandia National Laboratories Report SAND2001–3119. Wong C, McFarlane D, Zaharudin A, Agarval, V, (2002) The Intelligent Product Driven Supply Chain. Proc. IEEE International Conference on Systems, Man and Cybernetics, Hammamet, Tunis, 4:6–11. Wyld DC, (2005) RFID: The Right Frequency for Government. IBM Center for Business of Government, e-Government Series. Yao A, Carlson J, (1999) The Impact of Real-Time Data Communication on Inventory Management. International Journal of Production Economics 59(1-3): 213–219. Yu Z, Yan H, Cheng T, (2001) Benefits of Information Sharing with Supply Chain Partnerships. Industrial Management & Data Systems, 101(3): 114–119. Zaharudin A, Wong C, Agarwal V, McFarlane D, Koh R, Kang Y, (2002) The Intelligent Product Driven Supply Chain. Auto-ID White Paper CAM-AUTOID-WH-005.

2 Applications of RFID in Supply Chains Gary M. Gaukler and Ralf W. Seifert

Abstract:

In this chapter, we first give an introduction to radio-frequency identification (RFID) technology. We discuss capabilities and limitations of this technology in a supply chain setting. We then present several current applications of this technology to supply chains to demonstrate best practices and important implementation considerations. Subsequently, we discuss several issues that may hinder widespread RFID implementation in supply chains. We close by deriving several consequences for successful implementation of RFID, and we give guidance on how a company might best benefit from this technology.

2.1 An Overview of RFID Technology At its core, RFID is a contactless interrogation method for identification of objects. Besides the applications in supply chain operations on which this chapter is going to focus, some of the everyday uses of this technology are in ID cards, sports equipment, windshield-mounted toll tags, and gasoline quick-purchase tokens. RFID has also begun to be used in keychain auto antitheft devices and toys (most notably, Hasbro Star Wars figures), and even on paper tickets for the 2006 Soccer World Cup in Germany (Odland 2004; Want 2004). 2.1.1 RFID Hardware An RFID system consists of three parts: the RFID tag itself, the RFID reader device, and a backend IT system. The RFID tag typically consists of a silicon chip that can hold a certain amount of data (such as a unique identification number) and an antenna that is used to communicate with the remote reader device. There are chipless RFID tags as well, which exploit certain RF-reflecting properties of materials. In the case of chipless RFID, the tag’s unique serial number is given by

30

Gary M. Gaukler and Ralf W. Seifert

the properties of the material, e.g., the configuration of RF fibers embedded in paper. The reader device communicates with the RFID tag by sending and receiving radio-frequency waves. The way this communication happens differs between socalled passive and active RFID tags. Passive RFID tags do not have a power supply; the energy stored in the reader device's radio-frequency interrogation scan is enough to wake up the RFID tag and to enable it to send a response (that is, the RFID tag's data) to the reader device by reflection. Active tags contain a battery that allows them to respond to the reader's interrogation with a stronger signal, thus increasing the distance from which the tag can be read. The backend IT system is responsible for cross-referencing the RFID tag's ID number with a database record that describes the object to which the tag is attached. The method of communication by reflection of power that is the basis of passive RFID was described as early as 1948 (Stockman 1948). Related transponder technology was also used by the military as a friend-or-foe means of detection in Great Britain by the Royal Air force in World War II. Regardless of the choice of active vs. passive RFID, the radio-frequency communication between tag and reader may happen on several frequency bands. Low frequency (LF) RFID uses the 125–134 kHz and 140–148.5 kHz channels, high frequency (HF) RFID is located at 13.56 MHz, and ultra high frequency (UHF) RFID uses 868–928 MHz. These frequencies are the ones that are used in the United States of America, European and Asian frequency bands differ slightly. The different frequencies have different characteristics with respect to the speed and the range of the communication. In general, the higher the frequency, the longer the communication range, and the faster the communication, which means that more data can be transmitted. Each of the different frequencies typically requires different tags and reader devices, although for some applications, combination devices that work in multiple frequency bands are available. 2.1.2 RFID and Bar coding In some ways, RFID is similar to bar coding: Both technologies use labels and scanners to read the labels, and they both rely on an IT system that cross-references the ID on the label and relates it to an object or a class of objects using a database system. However, from an application standpoint, there are three major advantages of RFID over bar coding (see, e.g., Clampitt 2006; Sweeney 2005): x No line of sight required x Multiple parallel reads possible x Individual items instead of an item class can be identified x Read/write capability. To illustrate the first two of these advantages, consider a typical receiving operation in a warehouse. A mixed pallet of goods is received. If all individual products on the pallet need to be entered into a computer system to acknowledge receipt, then in the bar coding scenario, a worker needs to break the pallet, open the cases, and scan each product. (If there were bar code labels on each case, these bar codes were linked to the contents of each case, and the recipients were to trust that

Applications of RFID in Supply Chains 31

the cases had not been tampered with, then the number of bar code scans could be much lower.) This is both time-consuming and error-prone, and it means that the flow of goods into the warehouse needs to be interrupted to recognize what was received. In a scenario in which each product has an RFID tag attached to it, the pallet would simply be pulled through an RFID reader portal, and all products on the pallet would be identified almost instantaneously. The third advantage is that an RFID tag can give more information than today's bar code labels. (Advanced bar codes have been devised that can store more information, e.g., 2D-bar codes. However, the size of the bar code limits how much information can be encoded.) Consider, for example, a can of Coca Cola. The bar code on each can of Coke is the same–The bar code describes the class of the product. In an RFID scenario, each can of Coke will have its own unique identifier. This means that complete tracing of the origin of an individual product is possible for the first time. This enables new possibilities, for example, for handling product recalls, warranties, and demonstrating the authenticity of a product based on information about its origin. The read/write capability of RFID tags may have additional benefits when a computer link to a network database cannot always be guaranteed. In this case, data about the item that is tagged with the read/write tag can be stored and changed on the tag itself, without requiring a change in the database record that corresponds to the tag’s serial number. Military applications of RFID make heavy use of read/write tags. 2.1.3 Levels of RFID Tagging RFID tagging can take place at essentially three levels of granularity. In palletlevel tagging, a tag is affixed to a pallet. When the pallet is ready for shipment, a tag ID is programmed into the tag. This tag ID is typically cross-referenced to a purchase order and a list of the inventory on the pallet. At the shipment destination, the tag ID can be cross-referenced again to the database record that contains the pallet information. In case-level tagging, tags are placed on cases. As in pallet-level tagging, the tag typically cross-references the purchase order and inventory information. The primary advantage of case-level tagging over pallet-level tagging is that it allows more detailed tracking. If inventory is moved in case quantities (as opposed to in eaches), then full inventory visibility can be achieved with case-level tagging. Case tagging also saves labor time by automatically reporting case counts and thus making manual counting of cases unnecessary. Item-level tags are usually part of the item packaging. They are placed either inside the product’s box or attached to the item itself. Item-level tagging gives the highest possible granularity of visibility. It is useful if individual products are handled in the supply chain, such as in a retail environment. In virtually all existing item-level RFID implementations, the manufacturer places the RFID tag on the finished good. Item-level tagging seems to hold the most potential for the retailer but is the costliest solution for the manufacturer (Gaukler et al. 2006). Despite some widely publicized trials, the business case for item-level tagging on

32

Gary M. Gaukler and Ralf W. Seifert

low-value/low-margin items remains unproven; the cost of RFID tags is a drawback for an application of this technology to low-value, low-margin items. 2.1.4 RFID Standards The need for standards arises automatically when considering RFID for a supply chain implementation. Tags and readers in different geographic locations (e.g., countries) and from different manufacturers need to communicate. Moreover, care must be taken that the serial numbers on RFID tags are unique – if there were two items with the same identification number in a supply chain, how would one determine which item is which? Hence, the main types of standards that we consider for the purposes of this chapter on supply chain application are data standards and technology standards. Data standards specify what is contained in the memory portion of the RFID tag, as well as the format in which this data is stored. For supply chain applications, the most important data standard is the EPC (electronic product code). This standard was developed by EPC global and UCC/EAN (now GS1), based on data format proposed by the MIT Auto ID Center. The EPC is simply a number, usually 64-256 bits long, that is being given out by EPCglobal. Each company that licenses and subscribes to this standard is given its own “address space” (or number range) for its items, so that no EPC will ever be duplicated worldwide. Technology standards specify the protocols that are used for communication between tag and reader. Important variables here are the frequency and the power at which this communication occurs. For supply chain applications, the most important technology standard is the ISO (International Standards Organization) 18000 standards family. This standards family covers frequency bands from low frequency to HF, UHF, and microwave. In supply chain implementations, typically standards 18000-3 (HF), 18000-4 (microwave), and 18000-6 A/B (UHF) are most widely used. The ISO 18000 standards also specify data structure requirements which are compatible with the EPC. Unfortunately, different “flavors” exist, for example, of the 18000-6 standard (A and B), that make interoperability between tags from different manufacturers challenging. The new EPCGlobal Gen2 standard (passive UHF), also known as ISO 180006 C, promises to be the first truly international and interoperable standard. In addition to interoperability, tags conforming to the Gen2 standard also promise higher and faster read rates, smaller size chips, and encryption capabilities (York 2005). 2.1.5 Limitations of RFID Naturally, RFID has limitations. Some limitations are due to the laws of physics. Metals and liquids, for example, effectively block radio waves. This is particularly true for UHF and microwave frequencies. Thus, it is generally not possible to read RFID tags enclosed in metal or surrounded by liquids. There are some advances in tag and antenna design that allow placing RFID tags on metal objects, as long as the tag is not fully enclosed. But in general, RFID does not work very well in environments where the product is surrounded by metals or liquids.

Applications of RFID in Supply Chains 33

A potential work-around to the problem of fluids and metals blocking RF is to use multiple readers, trying to read a tag from different angles. This will improve read rates in the presence of metals and liquids, and also in the general case. A similar method is employed by some companies to boost read rates of item- or case-level tags on a pallet. In this method, only one stationary reader is used, but the pallet with the cases and items on it is rotated so that the reader “sees” tags in different orientations. This can, for example, be combined with an existing workstation where the pallet is wrapped with protective tape. Other limitations of RFID are due to the fact that the application of the technology in logistics operations is still relatively new. RFID tags can be defective (just like bar code labels that may be unreadable because they are torn, dirty, etc.), and to some extent interference issues between readers may exist that prevent tags from being read. It is expected that some of these limitations will find technological solutions as RFID technology becomes more mature in this application.

2.2 Current RFID Applications In this section, we review some of the more interesting existing RFID applications. RFID implementations discussed here are in diverse sectors such as retail, logistics, etc. The discussion here focuses on implementations in industry, not on RFID research in the operations management community. Related discussions on implementation issues are given, for example, in Angeles (2005) and Jones et al. (2004). Sheffi and McFarlane (2003) discuss the value of RFID in supply chain operations. For an excellent overview of current theoretical research connecting RFID and operations management, see Lee and Ozer (2005). In the following, we will cover applications such as product tracking and identification, preventing misplaced items, enabling counterfeit protection, and connecting RFID tags with environment sensors. 2.2.1 RFID in Retail An interesting application of RFID in the retail sector is the use of smart shelves and item-level tagging. Smart shelves are retail shelves that have RFID readers built-in. The main purpose of smart shelves is to prevent out-of-stock situations (OOS) from occurring at the shelf. An out-of-stock situation occurs at the shelf if a customer wants to buy a certain product, but the shelf is empty. Studies by Accenture and IBM Consulting (Alexander et al. 2002; Kambil and Brooks 2002) suggest that in approximately 30% of cases where the shelf is out of stock, there is actually product available in the backroom stocking location, but it just has not been put on the shelf yet. A timely alert to store personnel from a smart shelf could prevent this out-of-stock situation and thus prevent a sale from being lost. Metro AG, a supermarket chain in Germany, has one of the most advanced RFID implementations in the retail sector to date (Metro AG 2006). When this was writen, it was the only RFID implementation that has been the subject of a Harvard Business School case (Ton et al. 2005). Metro has introduced RFID on

34

Gary M. Gaukler and Ralf W. Seifert

the pallet-level (and increasingly on a case-level as well) throughout its supply chain. Around 100 of Metro’s suppliers, 8 Metro distribution centers, and 250 Metro stores use an RFID-enabled logistics unit to document transportation routes, simplify warehouse picking and receiving, and to check inventories automatically. Metro’s partners in this effort include Procter & Gamble, Henkel, Johnson & Johnson, and Esprit. Metro estimates that at the goods receiving stage in their warehouses and at their stores, manual labor times can be reduced as much as 80%. At the same time, backroom out-of-stock situations have become less common because employees are kept aware of backroom inventory levels continually. This in turn enables them to reorder products in a more timely fashion. In addition to the widespread implementation of pallet- and case-level RFID, Metro is experimenting with item-level RFID at one of its supermarkets in Rheinberg, Germany. This supermarket is known as the Metro Future Store. This Future Store is a fully operational supermarket that serves 2,800 customers per day with a sales area of 4,000 square meters. The store is a working showcase for several technologies that are intended to improve the shopping experience. At this store, item-level RFID is used on four products (Procter & Gamble Pantene shampoo, Kraft Philadelphia cream cheese, Gillette Mach3 razors, and DVDs). These products are displayed on smart shelves (Figure 2.1).

Figure 2.1. Smart shelf with Gillette products (photo courtesy of Metro AG)

Shelf replenishment from the backroom is initiated by notification from the shelf to the store personnel. In addition, there are RFID reader gates at the checkout lanes (for future automated self-checkout purposes), between the inbound loading dock and the store backroom, as well as between the store backroom and the sales floor.

Applications of RFID in Supply Chains 35

By using these RFID gates, it is possible to keep track of all RFID-tagged merchandise within the store. Figure 2.2 shows an RFID gate.

Figure 2.2. RFID gate at the Metro Future Store (photo courtesy of Metro AG)

Apart from the inventory control and efficiency aspects of this RFID implementation, Metro also attempts to add value to the shopping experience for its customers. This is done through RFID information kiosks. Customers can, for example, take DVDs to such a kiosk. The kiosk will read the RFID tag on the DVD and play a preview clip of the movie or display other additional information. In response to privacy concerns, there is also an RFID tag deactivation station, where customers can deactivate the RFID tags on their products after they have purchased them. Further ideas for such one-to-one marketing are developed in Klabjan 2005. 2.2.2 RFID in Logistics, Transportation, and Warehousing The benefits of RFID in logistics, transportation, and warehousing can be broadly categorized into (1) labor and timesaving and (2) benefits from increased visibility. The first benefits category has also been described using the notion of “the uninterrupted supply chain” (TUSC) (Supply Chain Digest 2005). This is the idea that too large a percentage of a product’s traversal time through a supply chain or a logistics system is spent waiting for identification or the completion of some manual process such as counting cases related to identification and documentation. Hence, the flow of goods is interrupted by stopping points. A system that uses automated identification through RFID can potentially remove many of these

36

Gary M. Gaukler and Ralf W. Seifert

stopping points, enabling the product to move through the system faster and at less cost. These purely operational cost savings will tend to be especially large if the product in the supply chain is a serialized product, i.e., a product where each case needs to be identified instead of bulk identification per pallet. Serialized products include computers, computer printers, electronics equipment, etc. These are serialized products because, for example, each computer may have a different configuration. This is different from dealing with a generic, noncustomizable product such as dishwasher detergent, for example. Another area where RFID can be a useful tool is in improving inventory accuracy. Inventory accuracy refers to the difference between logical and physical inventory. Logical inventory is the amount of inventory on record in the computer system – WMS, ERP, etc. Physical inventory is what is really in stock. Ideally, the logical and physical inventory quantities are equal, but for a variety of reasons (shrinkage, input errors, loss of goods, misplacement, etc.), these quantities may be quite different. Typically, logical inventory shown in computer systems is larger than physical inventory. RFID can help improve the logical inventory records due to automation of the scanning process. See, for example, Atali et al. (2005) for academic work on this subject. These relatively simple efficiency and accuracy improvements may be large enough to justify an RFID implementation, out the lion’s share of benefits are typically expected to be realized from increased visibility (MIT CTL 2004). Exact knowledge of the amount of inventory at each location in the supply chain in realtime is going to enable supply chain decision-makers to run a much more efficient supply chain. Knowing what is in the replenishment pipeline and when it is expected to arrive, potentially allows safety stocks to be reduced, while maintaining or increasing customer service levels. Essentially, this RFID visibility puts a supply chain on a new inventory/service-level exchange curve (see Figure 2.3). In addition, new and more flexible inventory control policies can be devised that use the added visibility. Gaukler et al. (2005), for example, describe an inventory control policy that uses of knowledge about the location and timing of replenishment orders in the resupply channel to make decisions on placing additional orders from a different supplier if the existing orders are held up for some reason. It is expected that with the spread of RFID installations, many more sophisticated policies will emerge to control supply chain activities in an automated way based on RFID visibility. 2.2.3 RFID in Assembly, Manufacturing, and Configuration Management Assembly and manufacturing of configurable products are supply chain activities that offer much potential for the use of RFID. RFID tags can be used in manufacturing to identify the product that is being assembled, as well as the constituent parts that are to be installed into the product. At the time of assembly, it is then possible to do an instant check to ascertain which parts need to be installed in the product and whether the parts that are installed are the correct parts. Thus, RFID has a role in assuring the quality of the end product. This benefit is particularly valuable if the product is highly customizable. The benefit from an

Applications of RFID in Supply Chains 37

introduction of RFID in this scenario is twofold: On the one hand, there are the labor savings from automating the scanning/identification of chassis and parts, and on the other hand, there are the savings in rework cost due to fewer assembly errors.

Exchange Curves (black = without RFID, yellow = with RFID)

Avg. inventory [units]

8,000 6,000 4,000 2,000 0 85.00

89.00

99.90

99.99

Service level [%]

Figure 2.3. Sample exchange curves with and without RFID visibility

Ford Motor Co. has been using RFID tags for this purpose in their facility in Cuautitlan in Mexico as well as in U.S. facilities for a number of years (Johnson 2002). In Ford’s implementation, an RFID tag is attached to each car’s chassis skid. The tag indicates via its serial number, which parts and options are to be installed on that particular chassis. As the chassis moves from one assembly station to the next, RFID readers read the chassis’ assembly requirements automatically so that the correct parts can be installed without the need for errorprone manual bar code scanning. “Error-prone” here refers, for example, to the case of a worker scanning the wrong tag, or a worker scanning the tag on one part, but then erroneously installing a different part. Gaukler and Hausman (2005) study a similar RFID scenario based on an implementation at a European car manufacturer. They present several net present value evaluations of a move from bar coding technology to an RFID implementation in such an assembly environment and provide an algorithm to determine an optimal RFID rollout schedule. Further analysis of RFID opportunities in the automotive sector is given, for example, in Strassner and Fleisch (2003). It is expected that similar benefits can be realized in other, nonautomotive, complex assembly processes that consist of a large amount of manual labor with a

38

Gary M. Gaukler and Ralf W. Seifert

high probability of assembling incorrect parts. Examples of this are the assembly of medical devices and aircraft. 2.2.4 RFID in Asset Tracking and Locating Objects RFID can be used to prevent misplacement of items or to facilitate locating items. One way this can be done is to keep track of item movements in a database. For example, RFID readers can be installed at doors between rooms in a building. Evaluating the records of the RFID readers would then allow one to deduce in which room the item is located. It is worthwhile to note that this is typically not a good theft prevention method, because RFID readers can easily be inhibited by placing the item in a metal-lined bag for instance. However, this system can work well in an environment where one is more concerned about accidental misplacement of an item than about theft. An example of such tracking is given at Bon Secours Richmond, which has implemented an RFID-based asset-tracking program at four hospitals (Baker 2005). Bon Secours tracks 20,000 pieces of medical equipment in this implementation, ranging from infusion pumps to ventilators and wheelchairs. This asset tracking implementation saves costs at several levels: First, the cost for purchase or rental of new equipment has gone down because utilization of existing equipment has gone from 40% to 70%; second, preventive maintenance costs for the equipment have decreased significantly since, with the tracking program, maintenance personnel do not have to search for the equipment anymore. Moreover, the handling of equipment recalls has become more efficient: Potentially hazardous equipment can be located fast and retired from active duty, thus increasing patient safety. Another way of locating items is to use a mobile handheld RFID reader and RFID tags as a homing device. A variant of this method is to use information stored on RFID tags on a container to find out the contents of that container. The U.S. Army uses this extensively during overseas deployments. Cargo containers that are shipped to U.S. deployment sites contain an active RFID tag that lists the contents of that container. Containers are also tracked during shipment by GPS. During operation Desert Storm, more than 50% of the containers that were shipped to the Gulf region had to be opened to find out what was stored inside. Since logistics officers did not know what materials were already on site, many redundant replenishment orders were released, contributing to a high level of inefficiency. RFID and supporting technologies have drastically increased inventory visibility both on the ground and in transit. Due to this visibility alone, the U.S. Army has managed to decrease on-hand inventory for some pieces of equipment to one-tenth of the level necessary during Desert Storm (Trebilcock 2006). 2.2.5 RFID in Authentication, Counterfeit Protection, and Security There are three ways in which RFID can aid in authentication and counterfeit protection of products: (1) by virtue of the presence of an RFID tag, (2) by proprietary encodings on the RFID chip, and (3) by establishing a chain of custody, also known as an e-pedigree of a product. (1) and (2) are a step up from no

Applications of RFID in Supply Chains 39

counterfeit protection at all, as it makes it harder (and more costly) for counterfeiters to copy the product. However, there is no absolute security because with sufficient expense, any RFID tag can be duplicated. Much better security from counterfeiting is afforded by the e-pedigree approach. In this method, an item’s authenticity is established by looking at the source of the item. As the item traverses the supply chain, its tag is read at multiple locations, including the manufacturer, logistics carrier, distribution centers, and retailer. Thus, an authentic item will have a record of tag reads (which is not saved on its tag, but rather on a supply chain computer system) that proves its progress from the manufacturer down the chain. A nonauthentic item, which is inserted into the supply chain at some other point, will not have this “correct” trace of tag reads. Hence, by looking at an item’s e-pedigree, it is possible to distinguish counterfeit from original items. This scheme is harder to defeat, since defeating it would necessitate changing tag read records in the supply chain’s computer system. One needs to keep in mind, however, that even when using an e-pedigree approach, an RFID tag can be duplicated. In this case, once the item with the duplicated tag is introduced into the supply chain, the system would see two distinct items that point to the identical history record. Although the system cannot decide which item is counterfeit based on this information alone, it can flag both items as suspect and would also be able to pinpoint the supply chain location where the counterfeit item was introduced. Not surprisingly, a major target of counterfeiters is pharmaceuticals. Pharmaceuticals are typically high-value items and are relatively easy to fake. Pfizer is one of the pharmaceuticals companies that have turned to RFID to make its supply chain more secure. In Pfizer’s RFID implementation, item-level RFID tags are put on drug bottles (in this case, Viagra and Lipitor). Pharmacists can check the authenticity of a particular bottle by scanning its RFID serial number and validating its authenticity with Pfizer directly, using their personal authentication account on Pfizer’s Web site. This is an example of authentication through proprietary encoding. In the future, Pfizer plans to move toward an e-pedigree approach (Pfizer 2006). Another area where authentication is required for security reasons is in inbound overseas container shipments. The goal there is twofold: To enable containers from trusted sources (countries) to be routed through customs faster and to prevent containers that have been tampered with from entering the country. Several of the largest seaport operators (Hutchison, PSA, P&O, China Merchants, and SSA) have formed a coalition to explore the use of automated tracking and RFID for containers shipped to the U.S. The idea behind this Smart and Secure Tradelanes Initiative (Savi 2006) is to apply an electronic seal using an RFID tag to the container at the outbound port and track the container through RFID readers and GPS on route to the destination port. At the destination port, the e-pedigree of the container would be checked, along with the integrity of the electronic seal. The result of this is ideally not only a higher level of security, but because of the epedigree, manual inspections may become largely unnecessary, speeding up the port operations considerably. Preliminary studies (Lee and Whang 2003) on the effectiveness of this method reveal large cost savings – for one pilot program, savings were estimated at $1,000 per container (Johnson 2004).

40

Gary M. Gaukler and Ralf W. Seifert

2.2.6 RFID with Environment Sensors Apart from integrating RFID tags with GPS sensors as in the port security example, there are other applications where data from environmental sensors is integrated with and stored on RFID tags. The U.S. Army, for example, uses RFID tags with temperature and shock sensors on cases of munitions. Shock sensors are important because the shelf life of munitions becomes drastically shorter once they have been dropped. The shock sensor records the g-shocks to which the case has been subjected to and writes the data to the RFID tag. When the case is opened, the data can be read out and used to determine the expected shelf life of the content of the case. A similar approach is used for temperature-sensitive material (DeLong 2003; Aitoro 2005). A similar opportunity for individually setting expected shelf lives for product lots exists in perishable products that require a cold chain for transportation and processing. Such products include many food items, some pharmaceuticals, as well as blood bags and donor organs. Karkkainen (2003) discusses the value of RFID for short shelf-life products in detail, including perishable items. The Kroger supermarket chain is running a pilot project to explore the use of RFID and temperature sensors in distributing case-ready meat. Meat temperature will be tracked from packaging the meat to arrival at the retail store. It is envisioned that RFID temperature history of each case of meat will be used by the retailer as a guide to decide on discounts and on optimizing the sequence of restocking the sales floor. Since different cases of meat may have experienced different ranges of temperature during transportation and thus have different remaining shelf lives, it is no longer optimal to replenish the sales floor using a "first-in/first-out" rule. Rather, the replenishment rule would likely be of the type "lowest shelf life first." OATSystems, Inc, which is supplying middleware and integration to this pilot project, estimates that Kroger can potentially cut its losses from spoilage in half (O’Connor 2005).

2.3 Proliferation and Adoption Issues Even with the clear advantages of RFID technology over bar coding and other identification methods, these factors impede the pervasive, large-scale adoption of RFID in many applications: The cost of RFID, the technological maturity of RFID, privacy issues, and the difficulty in allocating the cost of RFID to supply chain partners. 2.3.1 Cost of RFID At first glance, the cost of RFID seems high. The cost of RFID includes the cost of tags, readers, and IT infrastructure. In 2006, several years after the commercialization of RFID, the costs of tags and readers have come down significantly, but they are still high compared to that of the major competitor, bar coding. Tag costs run anywhere between 25 ¨ for short-range passive tags to well over $10 for specialized long-range active tags. Tag prices tend to increase with

Applications of RFID in Supply Chains 41

the amount of information that can be stored on the tag. Reader costs hover between $500–$5000 per device, depending on technology and features. Faced with these cost figures, it is obvious that companies need to look for a significant return on this investment. Typically, one can view reader and infrastructure costs as fixed costs because they tend not to vary significantly with the amount of product that passes through the supply chain. Tag costs, however, are variable costs in that every pallet, box, or item needs a tag. Thus, in popular literature, the cost of tags is usually seen as the most important determinant of RFID profitability. Our view of the cost issue of RFID is that in most cases, the cost of RFID (and the cost of RFID tags) is overemphasized. In all RFID implementations that we have observed and in all the studies that we have conducted, it turned out that the RFID cost could easily be amortized over a few years. This held true even when only the most basic RFID benefits were realized. By basic RFID benefits, we mean "low-hanging fruit" such as labor savings from not having to bar code scan boxes or pallets in a warehouse, etc. It is important to understand that the economies of RFID are fundamentally different from bar coding: With bar codes, the label cost is low, but the incremental cost of each scan is high; because it typically involves human labor and a break in the material flow. With RFID, the tag cost is higher, but the incremental cost of each scan is very low because scans can be performed automatically by stationary readers as goods flow by. Thus, potential benefits from RFID tend to be highest when each tagged good is scanned repeatedly. These RFID “economies of scan” notwithstanding, one needs to be realistic about which items can be RFID tagged and where RFID tagging simply does not make sense economically. RFID-tagging very inexpensive and low-margin items is not going to make sense – the 50-¨ candy bar with a 20-¨ RFID tag is not going to happen outside of pilot studies. This explains why mass-market consumer retail businesses that operate on very thin margins are reluctant to adopt item-level RFID tagging. 2.3.2 Technological Maturity of RFID RFID itself is not a very new technology, but its commercial use is very recent. Thus, companies must deal with a number of issues when implementing RFID. First, it is important to realize that there are several physical limitations of RFID technology, and any radio-frequency-based technology, for that matter. RFID does not work through metal or through liquids. Because metals and liquids inhibit the propagation of electromagnetic waves, which is the medium of communication that RFID uses. This does not necessarily mean that it is impossible to read RFID tags on bottles of orange juice. In fact, if there is line-ofsight between the reader and the tag, the tag will still be read. Matters become more complicated, of course, when trying to read an RFID tag on a bottle of orange juice that is stacked in the middle of a pallet. But even in this case, the tag can be read, provided that there are sufficient air gaps between the bottles on that pallet.

42

Gary M. Gaukler and Ralf W. Seifert

A technique that is often applied to alleviate this shortcoming of RFID is to rotate the pallet while reading tags. That way, the reader will "see" tags at different angles and the tag read rate would increase. Product design and the location of the tag on the product also affect how easy it is to read the tag. Second, several companies using RFID have reported issues with RFID tags that could not be programmed. In 2004, commonly reported RFID tag failures ranged from 10–12% – that is, 10–12% of tags purchased from a tag vendor could not be written and/or read (Brandel 2004). Of course, this is not a fundamental issue with RFID technology itself, but rather a quality control problem of the tag supplier(s). However, it shows that parts of the technology are fairly immature in the sense that, for example, manufacturing yields of RFID tags are low, which poses a potential problem for companies relying on 100% working tags. Third, the reader-and-tag interaction in real-world environments outside of a lab setting is often not very well understood. Other radio-emitting devices such as machines using electromotors, microwaves, or radios and TVs create an environment that is prone to interference with RFID tag reads. Reader and tag orientation are key to extracting high read rates in an RFID setup. In many ways, the hardware of an RFID implementation (for example, placing readers and antennas correctly) is more of an art than a science. Detailed theoretical scientific research on how to select readers and antennas exists (see e.g., Keskilammi et al. 2003), but outside of a laboratory, trial and error prevail. During one consulting project for an electronics manufacturer, we observed the following: A conveyor belt system was set up with an RFID reader such that the reader would scan all incoming boxes. Read rates were low for no apparent reason. Then one engineer moved the stationary reader a couple of inches higher and read rates suddenly increased significantly. Such trial-and-error optimizations are common in first rollouts of RFID. 2.3.3 Privacy Issues Consumer privacy concerns have been linked with RFID since the first commercial implementations of item-level RFID. Kelly and Erickson (2005) discuss privacy concerns related to commercial application of RFID in detail. The concerns have to do mostly with RFID tags on individual products, as opposed to RFID tags on boxes or pallets, simply because RFID tags on boxes or pallets do not typically get to the end consumer. The supermarket chain Tesco in the U.K. was among the first to be targeted by privacy advocates. Tesco and Gillette had set up a theft-prevention system that used RFID-tagged razor blades, a smart shelf, and digital cameras. When razor blades were removed from the shelf, a digital camera would automatically take a picture of the customer. Then, at the cashier, another picture would be taken of the customer as s/he paid for the razor blades. At the end of the day, store detectives would compare both sets of images and conclude that potentially a theft occurred if a customer showed up on the picture at the shelf, but not on the picture at the cashier. The efficacy of such a system is doubtful at best, and privacy advocates claimed that customers were not sufficiently informed about this surveillance practice. The ensuing PR proved disastrous for the image of RFID.

Applications of RFID in Supply Chains 43

However, apart from this scenario, most item-level RFID implementations do not save more data about customers than a regular supermarket loyalty card or a credit card would. Thus, we do not view privacy concerns as a significantly limiting factor to RFID adoption in the long run. 2.3.4 Allocating the Cost of RFID to Supply Chain Partners From a supply chain perspective, it makes sense to introduce RFID upstream in the supply chain to achieve the most benefits. This means that in the case of item-level tagging, it makes sense to apply tags for the finished product at the packaging stage of the manufacturing process. Thus, the manufacturer should place the tag on the finished good. However, experience indicates that manufacturers, logistics providers, and retailers typically see different benefits from RFID (e.g., see Alexander et al. 2002; Kambil and Brooks 2002). Manufacturers are generally most interested in tracking cases or pallets of their product via the transportation channel up to the retail outlets, whereas retailers typically gain most benefit from individual-product tracking on their shelves. Thus looking at the supply chain as a whole, the dilemma becomes clear: Itemlevel tagging seems to hold the most potential for the retailer but is the costliest solution for the manufacturer who needs to put on the tags. Hence, in theory in a competitive environment, the manufacturer will generally need some kind of contractual incentive to incur the tag cost, and downstream supply chain partners will need to share in the cost of the tag (for more discussion of contractual incentives, see Pasternack 1985 and Gaukler 2004). Exactly how this is to be done in practice, however, is not clear. But the question "who pays, and who benefits?" remain the subject of much contention. In the retail sector, Wal-Mart has taken the lead and announced a clear mandate: Any supplier to Wal-Mart must put RFID tags on cases and pallets from January 2005 on. The choice for Wal-Mart's suppliers hence has been to make the investment, or to stop being a supplier to Wal-Mart. In the military/government sector, the U.S. Department of Defense (DoD) has given a similar mandate for case- and palletlevel tagging to DoD suppliers. In both the Wal-Mart and the DoD cases, the market power of the mandating entities is such that suppliers overwhelmingly complied with the mandates. In supply chains where the power structure is such that there is no mandating entity, however, things are not so clear-cut. Manufacturers may hold back on tagging their products, because for their operation – viewed in isolation – there is no positive return on investment that would justify item-level tagging. For the supply chain as a whole – including the downstream stages and the logistics providers – it may well be that item-level RFID can have a positive ROI. Such a supply chain would thus not operate optimally because the investment decision in item-level tagging is based on only one supply chain partner's (in this case, the manufacturer's) cost and benefit structure. In a recent paper, Gaukler et al. (2006) show that in the absence of mandating entities, there exists a unique optimal way of sharing the cost of RFID tags between a manufacturer and a retailer. Sharing the RFID tag cost is optimal in the sense that the total supply chain profits are

44

Gary M. Gaukler and Ralf W. Seifert

maximized. Gaukler et al. compute the optimal way of splitting the cost of itemlevel tags and note that when tag costs are split that way, supply chain coordination can be achieved through a lump-sum payment.

2.4 The Strategic Dimension of RFID RFID is a technology that is positioned to influence drastically the way supply chains are managed. The technology itself is rather simple – after all, it is just another way of identifying things – but the implications are vast. For the first time, there exists a low cost way of identifying anything completely automatically. Limitations of RFID technology exist. Some of these limitations are physical, e.g., the difficulty in penetrating metal or liquids with RF. Some approaches to help alleviate this problem, such as varying the angle and position between reader and tag, have been discussed in this chapter. Another (financial) limitation is usually the cost of RFID. It is projected that part of this problem goes away as time goes by and tag and reader prices fall to more acceptable levels. In the meantime, smart use of, for example, reusable tags instead of one-time-use tags as discussed earlier, can dramatically improve the return on investment. The beauty and the value of RFID lies in automating the identification process: No more does one need to rely on workers to scan items. The economies of RFID are fundamentally different from the economies of bar coding: With bar codes, the label cost is low, but the incremental cost of each scan is high because it typically involves human labor and a break in the material flow. With RFID, the label cost is higher, but the incremental cost of each scan is very low because scans can be performed automatically by stationary readers as goods flow by. Thus, potential benefits from RFID tend to be highest when each tagged good is scanned repeatedly, presumably at different stages of the supply chain. We call this the RFID “economies of scan.” Benefits from RFID can be roughly categorized as evolutionary vs.revolutionary. Evolutionary benefits from RFID are those where business processes are essentially unchanged from the pre-RFID status. The difference and the benefits from RFID in this category stem from cost savings due to faster and potentially more accurate scans under RFID compared to bar coding. Examples of such improvements are labor cost savings, increased throughput, and more accuracy in inventory control. These improvements tend to be fairly easy to attain, but the overall impact is usually limited in magnitude. Revolutionary benefits from RFID, on the other hand, can accrue when RFID data are used to drive new automated processes. An example of this is new enterprise event management software that can drive a supply chain much more efficiently by reacting in realtime to events in a supply chain in an automated fashion. In essence, RFID is positioned to be the key enabler of a new wave of supply chain competition based on information flows. The decision to implement RFID, therefore, needs to be made by considering more than just the single company: Upstream suppliers and downstream customers as well as logistics providers, and their needs and requirements for an RFID implementation – all these elements need to be part of the decision process. Hence,

Applications of RFID in Supply Chains 45

when used to its full extent, RFID is not merely an operational tool that can be used in isolation; rather, it is a strategic technology. The successful supply chains in the future will be those that understand this strategic dimension and embrace it.

2.5 Guidelines for Practitioners Out of the discussion in the previous sections, it becomes clear that for companies that are new to RFID, it is prudent to start small and gain first-hand experience with the technology within the four walls of the company. This will enable the company to learn about the technology without interfering with outside business relations. In this phase, the company will evaluate the hardware as well as the software side of an implementation. Some of the questions that need to be answered at this stage are: x What type of tags and readers will be used? x How many read points will be needed, and where? x How does the RFID system interface with existing MRP, ERP, and warehouse management software? x What level of worker training is necessary? There also exist several RFID testing sites that companies can use, for example, Sun's RFID Test Center (Sun 2006). At this site, Sun has set up a full-fledged warehouse in which customers can pilot RFID equipment and Sun software to simulate an actual RFID rollout. When following this "start small, expand later" approach, it is crucial not to neglect the strategic side of an RFID implementation. Typically, the majority of benefits from RFID are not achieved within the four walls of the company but rather from implementation across the complete supply chain. Part of the strategic side of an RFID implementation is to identify the interfaces with suppliers, logistics providers, and customers. RFID is a cross-cutting technology in that it is used across the stages of the supply chain. As such, a successful implementation of RFID always needs to be viewed as an implementation that takes into account the whole supply chain. Insular implementation at individual companies without regard for what functionality is necessary or beneficial upstream and downstream in the supply chain, cannot extract the maximal benefits from RFID. Taking into account the whole supply chain also means that a company needs to be aware of how its particular RFID hardware (tags and readers) is or is not able to interface with the hardware used by its supply chain partners. As we have highlighted previously in this chapter’s section on RFID standards, this is unfortunately not a trivial problem. In our view, considerable care should be taken to implement RFID hardware solutions that are compatible with today’s international standards. The worst scenario a company can find itself in is to do what has been aptly named "slap and ship": Put the RFID tag on the product, and then forget about it. This is the ultimate insular implementation. "Slap and ship" only incurs cost (the RFID tag and the cost of tag placement), but there is no benefit to the company.

46

Gary M. Gaukler and Ralf W. Seifert

Going one step further and using RFID internally will provide some benefits to the company. These benefits will be mainly in the form of labor savings from not having to bar code scan items, savings from increased assembly quality and less rework, or general process improvements due to the fact that material flows can be visualized and improved. For many companies that are starting to think about implementing RFID, the most intimidating factor is the cost of RFID tags. In many cases, this is a variable cost that cannot be amortized the way a one-time fixed hardware purchase cost or a one-time system installation cost can. It can help tremendously if some serious thought is given to potential ways of using reusable tags instead of one-time-use tags. This applies mostly to RFID implementations within the four walls of one company, but may in some instances also be applicable to supply chain implementations. In one example taken from one of our recent consulting activities, a semiconductor manufacturer wanted to employ RFID to track cans of chemicals on its production-lines. The initial idea was to tag each can with a passive RFID tag once it was received from the supplier. Unfortunately, the usage rate of these cans (and hence the rate at which tags would be needed) appeared to be too high to achieve a positive ROI. The financial side started to look markedly improved when we suggested using reusable (reprogrammable) tags instead. These reusable tags would be attached like key fobs to the cans. Even though these tags are more expensive, fewer of them are needed; essentially the tag cost in this example had been transformed into a fixed cost. A similar case is the Volkswagen example that was briefly mentioned in this chapter’s section on RFID in Assembly, Manufacturing, and Configuration (Gaukler and Hausman 2005). Here, to identify a car chassis on the assembly line, a tag is not embedded in the chassis directly, but rather the tag is embedded in the assembly line conveyance that carries the chassis. The chassis tag can be reprogrammed depending on which car chassis the conveyance carries. Hence, again a variable tag cost is converted into a fixed cost that can easily be amortized. Implementing RFID within the four walls of a company can yield substantial benefits. However, the greatest impact typically occurs when one goes beyond these low hanging fruit: RFID promises highest benefits from integrating the RFID data across a supply chain. RFID gives vast amounts of data that need to be processed to filter out information that can be used to develop insights that can then be used to drive operational actions. The true value of RFID is making data (and hence information) available in an automated fashion at low cost. New software systems can be built using this information to drive a supply chain much more efficiently by reacting to supply chain exceptions (e.g., a late order, a stockout situation, etc.) in real-time in an automated way. A first example of such use of information from RFID is given in Gaukler et al. (2005). As competition among supply chains becomes more intense, RFID is positioned to be a key technology. RFID allows successful supply chains to differentiate themselves from others by the extent to which they manage to use information. This competition on information flows, pre-RFID, is exemplified by the retail practices of Seven-11 Japan, the dominant convenience store chain in Japan (Whang 2003). Seven-11 Japan managed to gather detailed information on the types of customers and their shopping preferences based on the time of day.

Applications of RFID in Supply Chains 47

From this customer demographic information, they derived insights that helped them restock and reconfigure the entire store with different products targeted at different customer demographics three times a day, maintaining an unparalleled degree of freshness of their products and offering a maximum of convenience for their customers. The small size of the Seven-11 stores and the limited variety of products allows them to do this efficiently without RFID. Nevertheless, this preRFID scenario illustrates the impact that managing and using information can have on business operations. In our view, RFID technology is going to be the key enabler of a new wave of supply chain competition based on information flows.

2.6 References Aitoro J, (February 2005) The Government and RFID: What You Need to Know, VARBusiness, http://www.varbusiness.com/sections/news/breakingnews.jhtml?articleId=60403591. Alexander K, Birkhofer G, Gramling K, Kleinberger H, Leng S, Moogimane D, Woods M, (2002) Focus on Retail: Applying Auto-ID to Improve Product Availability at the Retail Shelf, White Paper, www.autoidcenter.com. Angeles R, (2005) RFID Technologies: supply chain applications and implementation issues, Information Systems Management 22(1). Atali A, Lee H, Ozer O, (2005) If the inventory manager knew: Value of RFID under imperfect inventory information, Working Paper, Stanford University, Stanford CA 94305. Baker M, (October 2005) Hospitals Save Costs, Time with Wireless Tags, ExtremeNano, http://www.extremenano.com/article/Hospitals+Save+Costs+Time+with+Wireless+Tag s/162772_2.aspx. Brandel M, (December 2004) The Trouble With Tags, Computer World, http://www.computerworld.com/softwaretopics/erp/story/0,10801,98340,00.html. Clampitt H, (2006) The RFID Handbook, http://rfidhandbook.blogspot.com/2004/11/preface.html. DeLong B, (2003) How the US Military is Achieving Total Asset Visibility and More Using RFID and MEMS, Presentation, Smart Labels USA, Cambridge, MA. Gaukler G, (2004) RFID in the Retail Supply Chain: Benefits, Roll-Out Strategies, and Cost Sharing Agreements, The Supply Chain Connection 10(2), Stanford Global Supply Chain Management Forum. Gaukler G, Hausman W, (November 2005) RFID in Assembly Operations: Process and Quality Savings, Working Paper, Dept. of Management Science and Engineering, Stanford University, Stanford, CA. Gaukler G, Ozer O, Hausman W, (November 2005) RFID and Product Progress Information: Improved Emergency Ordering Policies, Working Paper, Dept. of Management Science and Engineering, Stanford University, Stanford, CA. Gaukler G, Seifert R, Hausman W, (2006) Item-level RFID in the Retail Supply Chain, Production and Operations Management (POM), forthcoming. Johnson D, (November 2002) RFID Tags Improve Tracking, Quality on Ford Line in Mexico, Control Engineering, http://www.manufacturing.net/ctl/article/CA257232?pubdate=11%2F1%2F2002. Johnson J, (July 2004) Safe Harbors, DC Velocity, http://www.dcvelocity.com/articles/july2004/equipmentapplications.cfm.

48

Gary M. Gaukler and Ralf W. Seifert

Jones P, Clarke-Hill C, Shears P, Comfort D, Hillier D, (2004) Radio frequency identification in the UK: Opportunities and challenges, International Journal of Retail and Distribution Management, 32(3). Kambil A, Brooks J, (2002) Auto-ID Across the Value Chain: From Dramatic Potential to Greater Efficiency and Profit, White Paper, www.autoidcenter.com. Karkkainen M, (2003) Increasing efficiency in the supply chain for short shelf life goods using RFID tagging, International Journal of Retail and Distribution Management, 31(10). Kelly E, Erickson G, (2005) RFID tags: commercial applications vs.privacy rights, Marketing Intelligence and Planning, 23(4). Keskilammi M, Sydanheimo L, Kivikoski M, (2003) Radio freuqncy technology for automated manufacturing and logistics control – Part 1, International Journal of Advanced Manufacturing Technologies, 21(10). Klabjan D, (2005) One-to-One Marketing with RFID, Working Paper, University of Illinois, Urbana Champaign. Lee H, Ozer O, (March 2005) Unlocking the Value of RFID, Working Paper, Stanford University, Stanford, CA. Lee H, Whang S, (October 2003) Higher Supply Chain Security with Lower Cost: Lessons from Total Quality Management, Working Paper, Stanford University. Metro AG, (February 2006) Press releases on www.futurestore.org. MIT CTL, (November 2004) Reading RFID, http://ctl.mit.edu/index.pl?id=3612. O’Connor M, (December 2005) Kroger Turning to RFID to Stay Fresh, RFIDJournal, http://www.rfidjournal.com/article/articleview/2055/1/1/. Odland T, (February 2004) RFID and Leisure RFID, Card Technology Today. Pasternack B, (1985) Optimal pricing and returns policies for perishable commodities, Marketing Science 4(2). Pfizer, (February 2006) Press releases at www.pfizer.com. Savi, (February 2006) Press releases at www.savi.com. Sheffi Y, McFarlane D, (2003) The impact of Auto-ID on supply chain operations, International Journal of Logistics Management 14(1). Stockman H, (October 1948) Communication by Means of Reflected Power, Proceedings of the IRE, 1196-1204. Strassner M, Fleisch E, (2003) The promise of Auto-ID in the Automotive Industry, White Paper, www.autoidcenter.com. Sun, (February 2006) Sun RFID Test Center Website, http://www.sun.com/software/solutions/rfid/testcenter/index.xml. Supply Chain Digest, (September 2005) Gillette, HP, and Lockheed Martin Say They Are Finding the RFID ROI, http://www.scdigest.com/assets/newsviews/05-09-15-2.cfm. Sweeney P, (2005) RFID for Dummies, Wiley. Ton Z, Dessain V, Stachowiak-Joulain M, (2005) RFID at the Metro Group, Case Study, Harvard Business School. Trebilcock B, (January 2006) RFID on the Front Lines, Modern Materials Handling, http://www.mmh.com/article/CA6299039.html. Want R, (January 2004) RFID A key to automating everything, Scientific American. Whang S, (2003) Seven Eleven Japan, Case Study, Stanford Graduate School of Business. York C, (January 2005) RFID Strategy – What Does the Gen2 RFID Standard Mean to You?, Industry Week, http://www.industryweek.com/Columns/Asp/columns.asp?ColumnId=1070.

3 A Tool Set for Exploring the Value of RFID in a Supply Chain Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

Abstract:

In recent years, radio-frequency identification (RFID) has emerged as an important technology to facilitate the management of a supply chain. Because the technology is expensive and time-consuming to implement on a large scale, most enterprises require a relatively rigorous business case to support the decision whether or when to adopt the technology. To enable the development of such business cases, a tool set has been developed to quantify the business value of RFID for different participants in a manufacturing-retail supply chain. The tool set consists of two tools which are linked: A business value model, implemented as an in-house developed application using commercial spreadsheet software, and a business process model, implemented using a commercial discrete-event simulation package. The business process model computes and provides certain supply chain performance metrics to the business value model, which are otherwise difficult to obtain. Because it is not trivial to capture the full range of potential benefits of RFID, it is necessary to coordinate two different types of decision support tools (spreadsheets and computer simulation).

3.1 Introduction In the last few years, radio-frequency identification (RFID) technology has become commercially viable for automatic identification of physical materials. The physical size and the manufacturing cost of a simple (e.g., passive, read-only) RFID tag have decreased to a point where large-scale applications in both the enterprise and consumer space are plausible even today. As of 2006, a simple RFID tag costs in the range of US$0.20 – US$0.40 (as reported by the RFID Journal); however, ongoing efforts by vendors are aiming to reduce the cost to US$0.05. At the same time, international standards for the physical characteristics of RFID (such as frequency and coding schemes) are well under way (including

50

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

ISO 18000 by the International Standards Organization and EPC by EPCglobal Inc.) As a result, RFID has become a major topic in the automatic identification community; industry-academia research consortia have been set up (e.g., the global RFID consortium at Massachusetts Institute of Technology that led to the formation of EPCglobal Inc.); and a number of publications dedicated to this subject have been launched, including the RFID Journal (www.RFIDjournal.com), RFID Gazette (www.RFIDgazette.org), and RFID News (www.RFIDnews.org), among others. Many commercial and industrial enterprises are seriously investigating the feasibility of applying RFID in their businesses. Most notably, a few large retailers, such as Wal-Mart and target, are in the process of conducting pilot studies or in phases of limited application of RFID in their supply chains. As the industry has now learned, simply using RFID tags as a high-tech version of bar codes is not very attractive economically because of their price premium. (Regardless of how fast the cost of RFID tags will decrease due to mass production, we can safely predict that they will be more expensive than bar code labels in the foreseeable future.) To make RFID commercially attractive, it is important to take advantage of its unique technical characteristics and derive business value from them. However, it is not widely known how we can do so at this time. Consequently, building a business case for RFID is not as easy as it may seem. Rather than a straightforward application of financial analysis alone, building a business case involves elements of exploration and innovation. The supply chain has been widely identified as one key business application of RFID technology. This chapter focuses on this area, and more specifically on manufacturer-retailer supply chains. To help build a business case for RFID in such supply chains, a tool set, consisting of a spreadsheet based business value model (BVM) and a simulation based business process model (BPM), is developed. The spreadsheet based BVM calculats the direct benefits (e.g., labor savings) that can be captured by traditional financial analysis and also serves as a master model to receive the output from the simulation model. The simulation model calculates indirect benefits (e.g., consequences of labor savings, such as a shorter lead time which leads to smaller amount of inventory stocked; a more detailed discussion of direct and indirect benefits is in Section 3.2) that will be hard to capture otherwise. With the tool set, we can explore how RFID can be applied to deliver business value. The individual approaches of financial modeling and simulation modeling are nothing new, but the combination of them to explore and quantify the value of RFID technology has not appeared elsewhere. This chapter describes the business value modeling tool set. The rest of Section 3.1 reviews the current status of business value modeling and discusses what has been done in analyzing the benefits of RFID in a supply chain. Section 3.2 describes the business value modeling tool set by each module and relates how the modules are linked. Section 3.3 discusses some experience of applying the tool set in practice, and Section 3.4 contains a hypothetical example to illustrate its use. Some guidelines for practitioners of RFID are included in Section 3.5. Using the business value modeling tool set, interesting insights on where and how RFID provides business value have been derived for a typical manufacturerretailer supply chain. They are reported elsewhere (Lee et al. 2005).

A Tool Set for Exploring the Value of RFID in a Supply Chain 51

3.1.1 Business Value Modeling Business value modeling has a long history in business and finance. It usually falls under the subject of investment analysis, which is basically a valuation process that assesses the combined effect of positive and negative cash flows over time (see, e.g., Helfert 2001, Chapter 8). Negative cash flows are those generated by the need to invest resources and positive cash flows represent the benefits obtained as a result of the investments. The negative cash flow items, i.e., the investments, are usually easier to estimate, except in cases of exploratory research where the total investment (and time) from ideation to a product is highly uncertain. The positive cash flow items are sometimes not so straightforward because the benefits provided by, say, a technology investment, may be difficult to quantify. An illustrative example of such benefits is the flexibility to perform certain activities. Measuring and quantifying flexibility has been a research issue since the days of flexible manufacturing systems; most recently new models based on real option analysis have been proposed (e.g., Nembhard et al. 2005). In practice today, such benefits are known as “soft” or “strategic” benefits and are typically considered outside of quantitative analysis. However, increasingly more business cases will be dependent on these “strategic” benefits, as enterprises are modernized and investments that provide simple labor savings are mostly completed. Once the positive and negative cash flow items are calculated, we need to combine their effects over time. A typical way to analyze the combined effects is to treat the cash flows as deterministic and then apply well known calculations (such as net present value, return-on-investment, internal rate of return, or payback period) to obtain summary measures for insights and comparisons among alternative investments. Spreadsheet software is the tool of choice, with many of these summary measures as built-in functions. In reality, cash flows, especially those that represent benefits, are uncertain, and techniques have been developed to model the uncertainty. These techniques include scenario analysis (also called what-if analysis) using a deterministic model or more advanced models, such as decision trees and financial or real options based models that explicitly take uncertainty into account. Scenario analysis is by far the most commonly used in practice and is especially well supported by spreadsheet software. Decision trees is a formal way of doing scenario analysis with specified probabilities of each scenario, so that expected measures, such as expected return, can be calculated. Options based modeling has recently become more popular as a tool for the valuation of technology investments. Not only does it take into account uncertainty, it also models the possibility of deferring parts of the investment decisions until later on, pending the outcome of the initial activities. This setting matches well with that of technology or product development. For example, an options based model has been used by NASA to assess the value of different technologies (Shishko et al. 2004). When multiple assets are available for investing simultaneously, one can form a new investment through a combination of the available assets. This is then a portfolio and techniques have been developed for portfolio analysis, such as meanvariance analysis (see, e.g., Luenberger 1998, Chapter 6). Other specialized

52

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

techniques have also been developed for specific types of investments such as security derivatives. These are beyond the scope of this chapter. This chapter focuses on evaluating the investment for a single technology, namely, RFID, in a specific application area, the supply chain. To that end, a combination of traditional value modeling and simulation modeling is applied. A similar approach has been used to model the business value of after-sales service outsourcing in Bagchi et al. (2005). 3.1.2 Analyzing the Value of RFID in Supply Chains RFID tags allow real-time tracking of capital equipment (such as trucks) and products (such as consumer and industrial goods) at the container, pallet, case, and/or item-level. Analyzing the benefits of RFID amounts to determining how enterprises can use such knowledge to streamline their operations. Most of the published materials on the benefits of RFID in supply chains are qualitative studies. For example, IBM Business Consulting Services published a series of papers, e.g., Alexander et al. (2003), discussing the impact of RFID technology on supply chain performance with a focus on consumer goods and retail value chains. Other reports of a similar nature include Agarwal (2001) and Kambil and Brooks (2002). A number of consumer product (CP) manufacturers commissioned business case studies, some with key retail customers, to get a better understanding of the overall benefits of the technology. Such work has resulted in the publication of several business-case focused white papers by industry groups, most recently by a consortium consisting of Food Marketing Institute, Grocery Manufacturing Association of America, and National Association of China Drug Stores (A.T. Kearney and Kurt Salmon Associates 2004), and by GCI (Global Commerce Initiative and IBM 2005). Model-based quantitative studies are sparse today but have drawn increasing attention. They usually focus on some specific aspects of RFID benefits and develop analytical or simulation models to address these aspects. For example, Fleisch and Tellkamp (2005) quantify the direct benefits (mostly) of RFID using a simulation model; Gaukler et al. (2004) develop an analytical model to study and compare the scenarios of a centralized and a decentralized supply chain under a possible range of store operating efficiencies (the most efficient case being the one using item-level RFID tagging); Joshi (2000) uses a simulation approach to evaluate the value of information visibility through the use of RFID. Lee et al. (2005) provide a more comprehensive study, identify where existing works in different technical literature may be applied to study the benefits, and, using a simulation model, develop insights on benefits that had not been investigated quantitatively in the past. For example, simulation results demonstrate that in some situations a supplier can simultaneously reduce warehouse inventory and customer back-orders because of process transformations enabled by the visibility of inventory levels at the retailer’s warehouse and store. In a recent effort to measure the potential benefits of RFID deployment, WalMart Stores commissioned a study to measure the impact of RFID on the amount of out-of-stock (Wal-Mart 2005). From February 14 to September 12, 2005, the

A Tool Set for Exploring the Value of RFID in a Supply Chain 53

inventory of selected items was examined daily in 24 Wal-Mart stores (12 RFIDenabled stores, 12 control stores) representing all store formats. Preliminary results presented in a research report by Hardgrave et al. (2005) suggest that RFID is making a difference: It helped the selected stores increase revenue by reducing out-of-stock merchandise by 16% in the past year, and RFID-tagged items can be restocked three times as fast as nontagged items.

3.2 The RFID Business Value Modeling Tool Set 3.2.1 The Business Value Model Cost-benefit analysis is a key component of deciding whether or when to deploy RFID technology in an enterprise. Using such analysis, the business owner can make a go/no-go decision, structure the project in a way to capture immediate benefits first, or even modify the scope of the project to maximize benefits. Many factors play a role in determining RFID technology's costs and benefits. For example, costs, such as investment in new tools and processes to install and test tags, or can be fixed recurring, such as the cost of RFID tags or the cost of applying them on cases and testing them. Benefits can be direct or indirect, as mentioned above. Clearly, building business cases for an RFID deployment based on realistic cost-benefit analysis is an important and complex task. To enable a rigorous cost-benefit analysis, a business value modeling approach developed in the last few years (Grey et al. 2003a, b) is applied. This approach goes beyond traditional return-on-investment (ROI) analysis to consider the nonfinancial aspects of an investment, such as improved customer satisfaction or shortened customer response times, translating such operational impacts into financial value. By augmenting traditional ROI analysis, it helps provide more complete information on the value of an investment. A second feature of this approach is that it considers the impact of the investment at the enterprise level, in addition to the return of that particular investment treated singly. The performance of a business enterprise is characterized by its profit and loss (P&L) statement and balance sheet. This approach calculates any differences in the P&L statement and balance sheet as a result of the investment. The business value model consists of three key parts: The benefit model, the cost model, and the data set as input to the benefit and cost models. The cost model represents the necessary financial investment over the planning horizon to install, operate, and maintain the RFID technology. The investment items depend on the configuration of the supply chain and the volume of the business but are relatively easy to estimate. Developing the benefit model is the most challenging and critical part of the value modeling exercise. Ideally, all impacts of the investment, no matter how indirect, have to be captured at an appropriate level of detail such that their consequences are clearly seen and yet their data requirements are reasonable in a real-life setting. Direct benefits are relatively straightforward to calculate, whereas indirect benefits may not be obtained using closed form equations. In some cases, we may not even understand fully what indirect benefits there might be. The latter

54

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

is especially true for RFID in a supply chain; Lee et al. 2005 explores this issue in detail. The RFID benefit model is based on a hierarchical decomposition or “tree” of key performance measures (KPIs) of a business, starting from the top level P&L items for the business in question – revenue and major cost items such as cost of goods sold (COGS) and selling, general, and administrative expenses (SG&A). An investment will impact a subset of these KPIs within this tree, the effects of which are then propagated upwards to the root of the tree using the mathematical relationships defined in the model. Figure 3.1 provides a sample illustration for such a decomposition of RFID benefits. At the leaves of the value tree structure in Figure 3.1, we have a list of benefits that may be realized by RFID. Each benefit is characterized by a set of performance measures. The values of the performance measures after RFID is deployed is estimated by an expert user from experience or from published studies, or is an output of another model or analysis (such as our process model). These changes in performance measures are then propagated up to the root of the tree by translating the measures into one of the major cost items or revenue. The translation is done through a set of equations custom built for the context, the supply chain in this case. Many, but not all, of these equations are relatively straightforward. Section 3.2.1.1 below discusses the list of benefits provided by RFID, and Section 3.2.1.2 describes the calculation of costs required by an RFID implementation. We use the BVM to analyze three possible initiatives in RFID investment: 1. 2. 3.

Pallet-level tagging, Case-level tagging, and Item-level tagging.

Each initiative has its own set of benefits, each is characterized by the performance measures impacted by RFID. Generally speaking, item-level tagging can cover all benefits provided by case-level tagging which in turn can cover all benefits provided by pallet-level tagging. 3.2.1.1 BVM Benefit Calculations RFID tags, at their basic level, provide several technical advantages over today’s most commonly used automatic identification technology, bar codes. The most important advantages are that RFID tags can be read with no human operation, with practically zero delay, at little to no variable cost, and any time when a reader is present; they rely less on the environment (e.g., line of sight or surface condition of the label) for reading accurately; many tags can be read simultaneously; and they have larger data capacity than bar codes. (Further, advanced tags present the possibility of writing additional information onto the tag over time, which we will not delve into.) We exploit these advantages to develop the following set of direct benefits.

A Tool Set for Exploring the Value of RFID in a Supply Chain 55

Revenue Incremental revenue due to reduction in stock-out rate Incremental revenue due to Improved visibility of stocks Operating Margin COGS Reduced deductions Reduced expired product write-offs Reduced product shrinkage

RFID Benefits

Reduced product shrinkage SG&A Reduced labor costs in shipping, receiving, managing returns, etc. Reduced inventory carrying cost Capital Efficiency Property, Plant, & Equipment Eliminated bar-code equipment Inventory

Reduced safety stock due to improved inventory accuracy & reduced lead times Reduced cycle stock due to improved visibility

Figure 3.1. Value decomposition or value tree for RFID

56

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

If bar codes are manually scanned at different points in a supply chain (e.g., the shipping and receiving docks of warehouses and stores) at present, the application of RFID will provide a direct benefit of eliminating those labor costs. Computation of such labor savings is relatively straightforward – the average time spent in each relevant manual operation can be collected and the average number of such operations can be estimated from the volume of goods moved in the supply chain and knowing how the goods are handled. In the BVM, we include a list of common, manual operations that may be saved (at least partially) by using RFID as an automation device: 1. 2. 3. 4. 5. 6. 7. 8. 9.

Production reporting at the end of production-lines Shipping pallets, cases, or items at the factory and distribution center Receiving pallets, cases, or items at the distribution center and retailer Handling inventory flow-through or cross-docking Physical or cycle counting Inventory auditing Printing and handling pallet license plates and case labels, or manual entry of information on labels Inventory reconciliation of damaged products Reconciliation and handling, at the factory and distribution center, of returned products, shipment errors, and subsequent claims.

The ability to be automatically read without delay will reduce the cycle time in the movement of goods. Besides labor savings described above such time reduction will result in: 1. 2. 3.

Reduced loading/unloading time that reduces trailer detention and carrier costs Reduced delays in shipping and receiving that reduce supply lead times, directly decreasing the stock one has to carry while waiting for delivery The promotional product component of the distribution center inventory that can be reduced as delivery can be phased closer to the promotional event.

Another important advantage of an automation device captured in the BVM is data accuracy. Today’s bar codes have an extremely high read accuracy in a laboratory environment and very good accuracy in normal use. The only key advantages of RFID over bar codes in reading accuracy is that bar codes can get dirty or damaged relatively easily and that bar code reading needs a clear line of sight (and some minor orientation requirements). If an enterprise faces significant issues because of these reasons, RFID will be one (but not the only) potential solution. The direct benefits due to a reduction in inventory read errors include 1.

The inventory value, the inventory carrying and handling costs to cover extra stock in the supply chain because of such errors (a fraction of the inventory is not usable because the record is incorrect, so the supply chain

A Tool Set for Exploring the Value of RFID in a Supply Chain 57

2. 3. 4. 5.

in time will carry extra inventory to satisfy customer service level requirements) A reduction in shortage claims by the retailer and less overages shipped (and kept) by the retailer A reduction in retailer charge-back because of better delivery compliance A reduction in transportation costs by reducing the transfer of stocks to cover those that are shipped or received in error An increase in claims recovery from carriers.

RFID can carry more information than a bar code, such as a serial number. The direct benefits of the presence of a serial number are as follows: 1.

2.

3.

Often, the actual selling price of a returned item is unknown so a “default price” is assumed when crediting the customer. Additionally, discrepancies between retailer and manufacturer counts for returned merchandise can lead to unnecessary deductions for the manufacturer and the extra labor on both the manufacturer and retailer sides to resolve the deductions. Visibility of product by serial number through the supply chain will allow manufacturers and retailers to better control the winding down of discontinued or about-to-expire products, reducing price markdowns. It will aid in managing recalls of products, saving time and labor. Recalls are rare but are very labor intensive.

Another advantage of RFID captured by the BVM is the ability to detect the presence or absence of tags very frequently at almost no marginal cost. The direct benefits are the following. 1.

2.

We can detect where and when (up to a certain resolution in space and time) material losses are incurred. We can then investigate the sources of such losses and devise action plans to remedy them. In this way, RFID can prevent shrinkages, even though they themselves do not prevent breakage or thefts. The direct benefits are similar to those of inventory reading accuracy discussed above. Actual store inventory can be reduced as a result of eliminating the practice of zeroing out inventory that cannot be found.

Related to shrinkage and theft, the use of RFID tags at the item-level will eliminate fixtures, tags, and labor dedicated to theft prevention. Moreover, eliminating some defensive merchandising strategies, such as holding products in locked display cases, can increase sales. Because of the electronic nature of RFID, its encryption capability, and its audit trails, it is virtually impossible to duplicate the tags. They can be used to distinguish between legitimate and counterfeit products. This can potentially recover significant revenue lost by the manufacturer and help protect the product brand.

58

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

3.2.1.2 BVM Cost Calculations The investment cost of RFID in a supply chain is rather substantial. We illustrate the costs using an example from a case study. Major system costs are from establishing the reader network and the variable cost of the tags themselves. In most cases, pallet and case tagging scenarios require readers at the same points in the supply chain so the infrastructure cost is the same for both. Item tagging requires more readers, mostly in the store, to take advantage of item-level information. Though much of the benefit opportunity is associated with labor savings, there is a new labor cost incurred as a result of deploying the system. This relates to applying and verifying tags on pallets and cases. In some instances, additional labor is not required for case-level tagging if automated in-line application is used in the packaging process. Costs per annum for tags will depend on the level of tagging and the volume of goods flowing through the supply chain. Once installed, the cost of maintenance and operational support for the reader network is estimated at 10–15% of the initial capital investment. 3.2.2 The Business Process Model There are two types of benefits provided by RFID. The first type of benefits are the direct benefits discussed above. These benefits are quantified by the business value model. The second type of benefits is indirect benefits. There are two sources of indirect benefits. The first kind of indirect benefits are those from the dynamic effects of small changes brought about by RFID in one area of the supply chain. Because a supply chain is a complex set of activities connected with each other and connected in time, small changes in one area may lead to unpredictable and/or significant consequences elsewhere in the supply chain or later in time. The well known “bullwhip effect”, first studied by Forrester (1958) and later by Lee et al. (1997) and many others, is an example of small changes propagated and amplified through the dynamic behavior of a supply chain. A representative example of this kind of indirect benefit of RFID is that due to its impact on inventory shrinkage. It is described in Section 3.4. The second source of indirect benefits that might be overlooked by a traditional ROI analysis is the need for a business process transformation to take advantage of the information now available from RFID tags. A simple example is the commonly used periodic inventory replenishment process – most retailers replenish their stores once a week based on a predetermined decision-making cycle for each product. To take full advantage of the potential real-time inventory information provided by RFID, this periodic replenishment process needs to be redesigned – perhaps as simply as reducing the cycle to a day rather than a week, if other considerations (such as the workload of the planners) allow. Such a change, simple it may seem, may have significant impact on the performance of the enterprise. However, the impact may not be easily estimated in a spreadsheet or closed form computation. Therefore, a BPM is used in computing the impact of

A Tool Set for Exploring the Value of RFID in a Supply Chain 59

process transformation by simulating the transformed supply chain, and the relevant results are communicated back to the BVM. Besides the inventory shrinkage and replenishment frequency mentioned, other indirect benefits identified by our BPM include 1.

2.

Visibility of inventory information across the supply chain – When inventory data across the supply chain are used in decision-making in production and inventory planning, substantial benefits can be gained in terms of average inventory in the supply chain and customer service level (i.e., out-of-stock performance). A numerical comparison of three scenarios using or not using inventory data in supply chain planning is contained in Lee et al. (2005). Time delay of inventory data – RFID can potentially eliminate the time delay in updating the inventory data in the information system. There are two types of delays: Delay in updating the system after the physical goods change their place, e.g., after they are received into a distribution center or a store, and delay in using data to make decisions after the data have been collected, e.g., point-of-sale data are uploaded to a central database at midnight and the buyer uses the data to make a purchase order decision the following afternoon. A more detailed discussion is contained in Lee et al. (2005).

The BPM developed is limited in scope to the logistics of the supply chain. Although more specialized to individual business situations, RFID can potentially bring benefits to product management and other areas, including 1.

2.

3.

Enhanced quality control within facilities – Each tagged container of raw materials or case of finished goods can provide important information about the life of individual products. Visibility to life as the products are stored and moved between facilities will allow managing product life more effectively and addressing potential quality issues, before products are moved further in the supply chain. Improved recall capabilities – The ability to track discrete cases (through the use of a serial number in the tag) of products at manufacturing and retailer locations would facilitate a much more effective recall process. While the incidence of recalls is relatively low, the cost of each recall can be substantial. In addition, given the visibility and importance of any consumer product recall, the ability to conduct quick, precise recalls is vital in limiting negative exposure in brand image and financial or legal risk. Improved market intelligence – The ability to have visibility to cases of product at retailer locations will provide a level of insight into product performance that is possible today only through external data gathering agencies. In the future, the cost of gathering market intelligence is significantly reduced as tagged case movement is automatically captured by RFID networks in a supply chain.

60

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

3.2.3 The Value Model – Process Model Linkage In the consumer product business investigated using the tool set, the indirect benefits captured by the process model manifest themselves in a reduction in average inventory carried in the supply chain and improved customer service in the form of a lower out-of-stock rate. These two performance measures are then used as input to the BVM. In the BVM, reduction in average inventory results in a lower inventory carrying cost, decreasing the SG&A cost in the P&L statement. It will also impact all inventory-related measures such as inventory turns. An improved out-of-stock rate will reduce the back-ordering cost (assuming that the customer will take back-orders), hence decreasing the SG&A cost, or will increase the revenue if we assume that a potential sale will be lost if the item is out of stock. In different industries, other indirect benefits may be realized, but eventually all indirect benefits will be translated into cost reduction or revenue increase in the BVM. For example, in a service industry, the availability of real-time data may decrease the total time a customer experiences in obtaining service. This improvement in customer experience may increase revenue because the enterprise can now serve more customers with the same resources or because the customer will bring more business either through more frequent visits or by bringing new customers through word of mouth. Clearly, some of these consequences are not trivial to quantify. The capacity increase can be estimated through queueing or simulation models; the improved revenue will be more difficult to estimate and is an entire research subject in itself. Once these models are built, their results can be incorporated into the BVM similar to those of the BPM in this study.

3.3 Application of the Business Value Modeling Tool Set The business value modeling methodology developed has been deployed in multiple business cases for leading manufacturers and retailers to investigate whether RFID technology should be adopted in their particular business. We consider the one-time cost of installation and ongoing tag and reader costs for maintenance and replacement, balanced against a number of one-time and ongoing benefits. Data inputs required cover metrics such as labor costs, locations (facilities and the number of discrete locations within facilities), transportation, inventory levels, shrink statistics, and product volumes, as required by the cost and benefit calculations described above. All costs and benefits identified are validated with process experts and functional heads within the enterprise, and the projected size of the expected benefits are adjusted if necessary to accommodate any assumptions or simplifications made. For each business case, the framework of benefit opportunities described is used to enable a comparison between our output summary and that of the industry level findings published by the FMI/GMA/NACDS and GCI, among others. Where benefits require process changes, we assume that appropriate processes will be in place to enable realization of these benefits. Benefits are phased over a 10year time frame in relation to the anticipated rate of retail adoption. Quantification of some benefits is in part based on documented industry

A Tool Set for Exploring the Value of RFID in a Supply Chain 61

assumptions. For example, the benefit opportunity related to a reduction in out-ofstocks leveraged industry studies on the relationship between product availability at the retail shelf and consumer purchase behavior. Most business cases have a planning horizon of multiple years; for RFID, this is particularly important. First, implementing RFID in a supply chain is not a trivial task, so the implementation or adoption time will be several years. The benefits have to be scaled based on the implementation rate. Scaling factors can be estimated using simple but reasonable assumptions, such as, that the benefits are directly proportional to the implementation rate in terms of the volume of products using RFID. Second, as the industry climbs the learning curve of producing RFID infrastructure items (tags, readers, etc.), the cost of purchasing these items will decrease significantly in the next several years. Such variable unit costs are directly taken into account in the investment input of the BVM. Our experience indicates that whether there is a business justification for RFID depends on the type of business and the products, the level of tagging (item, case, or pallet), the current operational efficiency of the supply chain, the business volume, and the rate of trading partner adoption. It is possible that direct benefits alone, although easier to estimate, may not provide adequate benefits to offset the investment cost. Indirect benefits are more complex and may involve changing certain business processes or decision-making logic. They are also not widely understood and are therefore a barrier to RFID adoption at present. One critical factor in all the business cases is the cost of the RFID tags. The unit tag cost is expected to decrease continually in the future, but the rate of decrease is uncertain. A useful application of the business value model is to analyze what rate of cost decrease will enable the investment to break even in the planning horizon (i.e., with a net present value of zero over the horizon). By approximating the rate of decrease with a constant rate (representing the average rate), we can find the break-even rate using a simple search (e.g., bisection search). Comparing this break-even rate to historical cost decrease rates of other technologies, we can gain insights into the chance of realizing a positive return on an RFID investment at different tagging levels.

3.4 An Illustrative Example In this section, we provide a hypothetical example to illustrate the use of the tool set. This example considers a simple manufacturer-retailer supply chain in the CP business, which was also used to conduct an in-depth exploration of indirect RFID benefits in Lee et al. (2005). The supply chain consists of three echelons: a manufacturer, a distribution center (DC, that belongs to the manufacturer), and a retail store. (See Figure 3.2.) The application of RFID technology is modeled in each echelon in the supply chain. At the manufacturer, we model RFID tag reading at the points of production completion and shipping. At the DC, we model tag reading at the receiving and shipping docks. At the retailer, we model tag reading at receiving, in the backroom, and on the shelf in the store. Various simplifications and assumptions are made to

62

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

capture the essence of supply chain behavior without making the model very complicated.

Manufacturing Decision

Shelf Replenishment Decision

Store Replenishment Decision

P20 output

P20 output

P20 output

Customer

Inventory

Manufacturer Manufacturing input P100

p101 output

Shipping-MFG input p102

p103 output

Receiving-DC input P104

DC Shipping-DC

P105 output input P8

P7 output

Receiving input P6

P5 output

Retailer BackRoom input P4

Shelf

P3 output input P2

RFID Tag Reader

Figure 3.2. Three-echelon supply chain model for a CP retail business

At the retail store, we model four products which are sold to customers with equal probability. Customers arrive with an inter arrival time characterized by a log-normal distribution, and their purchase quantity on each purchase occasion is uniformly distributed between 1 and 3. The store replenishment is based on an (s, S) policy: reorder point, s, and target inventory, S. Shelf replenishment is also based on an (s, S) policy. At the manufacturer, we assume that the daily production quantity for each product is based on a certain policy and is shipped to the DC once a day. Several different production policies are investigated. The lead time for shipment from manufacturer to DC is 1 day. At the DC, the products are pulled from the retailer based on the retailer’s replenishment policy and decision frequency. The lead time for shipment from the DC to the retailer is 1 day. The business value model was implemented in Microsoft Excel and the simulation model was implemented using the simulation engine of IBM WBI Modeler ® (IBM Corporation). Details of the simulation model are described in Lee et al. (2005). In this hypothetical example, we assume that the retailer is interested in exploring the value of RFID tags at the item-level. The revenue and major costs (before RFID implementation) of the retailer are shown in Figure 3.3.

A Tool Set for Exploring the Value of RFID in a Supply Chain 63

RFID - All Service Areas

Data Source

Financial Net Revenue ($K) Research and Development expense ($K) COGS ($K) SG&A ($K) Depreciation and amortization expense ($K) Effective tax rate (%) Pre-tax interest expense ($K) Cash and Marketable Securities ($K) Accounts Receivable ($K) Finished goods inventory ($K) WIP inventory ($K) Raw materials inventory ($K) Accounts Payable ($K) Property, Plant and Equipment ($K) Other Income or loss ($K) Other current assets ($K) Other fixed assets ($K) Other long-term liabilities ($K) Other current liabilities ($K) Long-term debt ($K) No. of Employees Weighted Average Cost of Capital (%) Inventory carrying cost (%/year) (not including financing cost) Obsolescence and Surplus Write-offs Annual cost of expired product write-offs ($K) Annual cost of seasonal product obsolescence ($K)

Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry Manual Entry

Annual cost of new product launch write-offs ($K) Annual cost of surplus writeoffs ($K) Direct Store Delivery Annual cost of check-in and invoice reconciliation ($K)

Manual Entry 1,000,000 35,000 500,000 105,000 22,000 39% 7,000 110,000 100,000 8,000 3,000 30,000 80,000 800,000 10,000 11,000 12,000 9,000 7,500 100,000 10 11%

Manual Entry Manual Entry Manual Entry

12% 1,458

Manual Entry

2,562

Manual Entry Manual Entry Manual Entry

2,304 2,268

Manual Entry

3,000

Figure 3.3. Partial list of input data items for the business value model

In this setting, RFID can contribute to the following benefits: 1.

2.

Labor a.

Reduction in labor due to elimination of scanning, counting, and manual data entry at shipping and receiving points, and store shelf b. Reduction in labor to manage returns c. Reduction in labor to perform physical/cycle counting d. Reduced labor to manage shortage/deduction claims Deductions and write-offs a. Improved cash flow and profit due to reduced deductions b. Reduction in expired product write-offs

64

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

3. 4. 5. 6.

7. 8.

Product diversion a. Increased margin from reduction in diversion Product returns a. Reduced returns as a result of accurate shipments Lead times a. Reduced shipping and receiving times Inventory a. Reduction in shrinkage, including damages theft, “paper shrinkage” (i.e., inventory wasted because it cannot be found when needed) b. Reduced safety stock from increased inventory accuracy c. Reduced safety stock due to shortened lead times d. Reduced inventory due to improved visibility and hence improved order quality e. Reduced inventory carrying costs due to the several sources of reduced inventory mentioned above f. Reduced back-order handling costs due to reduced fraction of time that is out of stock Transportation a. Reduction in transportation due to reduction in inventory Plant and equipment a. Elimination of bar code related fixed assets, e.g., bar code scanners, both handheld and conveyor mounted

Most of the above benefits are computed using the BVM. Labor benefits are easy to calculate. Deductions and write-offs, product diversions, and product returns all arise because of the tracking ability of RFID. How much the improvements will be, in terms of a fraction of the as-is performance, is based on expert input. Lead time reductions are estimated by eliminating all manual operations related to bar coding. Their ultimate impact in this example is to decrease the safety stock, which is computed using the standard safety stock formula of k-standard deviations of the lead time demand. Other inventory related benefits are more complicated; one of these is described below in detail. Transportation cost saving is a one-time saving of the shipping costs of the inventory that is reduced. Plant and equipment saving is straightforward. We now discuss a more interesting inventory benefit that is driven by shrinkage reduction. The direct effect of inventory shrinkage is calculated as the sum of shrinkages due to various causes, including theft, process failure, damage, and others. Each type of shrinkage is specified in terms of a percentage of sales. As-is values can be estimated from the store’s historical data. To-be values are estimated by an expert user based on experience and/or published information. The reduction in the total shrinkage can then be translated to either a decrease in COGS, assuming that sales remain the same, or an increase in sales, assuming that the goods that are recovered from shrinkage will be sold (with COGS remaining the same). The latter may be true for seasonal items where the supply is purposely planned to be below the estimated demand. The latter case is used in the example here.

A Tool Set for Exploring the Value of RFID in a Supply Chain 65

On the other hand, the indirect effects of inventory shrinkage cannot be calculated straightforwardly. Inventory shrinkage causes a discrepancy between the actual quantity of the physical inventory and the inventory quantity recorded in the computer system. When inventory replenishment decisions are made, they are based on the inventory information kept in the system (system inventory). The discrepancy between the actual and the system inventory, therefore, impacts the replenishment decisions and eventually the inventory level and out-of-stock performance. Such impact is estimated using the BPM. In general, the system inventory and actual inventory count are seldom synchronized due to shrinkage or stock loss, transaction error, inaccessible inventory, and incorrect product identification. The error in the system inventory accumulates over time and is not corrected until a physical inventory takes place. Physical counts of a product are conducted infrequently (at most few times a year) due to its labor-intensive nature. Raman et al. (2001) report that a staggering 65% of apparel items at the store level have mismatched inventory records. Now, RFID will not prevent product damage and may not be very effective against theft. But it will enable the system inventory to reflect accurately what is still physically there. By identifying where goods have been lost, it will help the enterprise prevent future losses. In this example, we make a simplified assumption that shrinkage will continue in the same amount, but the system inventory now takes into account the goods that are lost. Using the BPM, we can calculate the cost saving (if any) resulting from the visibility of shrinkage, which can then be communicated back to the BVM. In practice, RFID will enable visibility of goods lost for reasons other than shrinkage, such as misplacement. This discussion focuses on shrinkage alone, but the full benefit of RFID on inaccurate inventory records is potentially far beyond what is illustrated here. In this example, we assume that shrinkage at the retailer occurs at a rate of 1.6%. This shrinkage rate is used because it seems to represent what a typical U.S. retailer faces (as reported in Kang and Koh 2002). We simulate a scenario where RFID technology does not exist, and two scenarios where RFID technology is deployed. In the first scenario, RFID technology does not exist. Physical inventory of a product is done once every 3 months at the store (using cycle counting). The error in the system inventory accumulates over time until a physical count is carried out, at which time system inventory is synchronized with the actual inventory. The retail store’s replenishment policy is a continuous review (s, S) policy, with the reorder point (s) of 36 and the target inventory (S) of 48, based on the system inventory. Because the actual inventory is less than what the system thinks, we can expect that the customer service level will be lower than the target, even though all technical assumptions of the inventory replenishment policy may be satisfied. In the second scenario, shrinkage occurs as before, but RFID is deployed. With RFID, inventory is tracked more accurately and in real-time, and better replenishment decisions can be made. To illustrate this effect clearly, we assume that the accuracy of RFID is 100% and so the system inventory is same as the actual inventory. Simulation results indicate that the back-order quantity decreases to 1% of that in scenario one, the average inventory becomes 20% higher, but the

66

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

fluctuation of inventory is much smaller than that in scenario one. This presents an opportunity to decrease the inventory by modifying the replenishment policy, e.g., lowering the reorder point (s) and target inventory (S) without sacrificing customer service (i.e., without increasing the back-order quantity). We show one such case in the third scenario, with the reorder point decreased to 26 from 36, and the target inventory lowered to 38 from 48. The back-order quantity shows a 22% reduction from the first scenario, and the average inventory is reduced by 16%. Table 3.1 below illustrates the computation of the benefits described above in the BVM. The first section of the table shows the calculation related to the direct benefits of RFID with respect to inventory shrinkage, which is the reduction of the shrinkage itself. The second section computes the reduction of handling cost for back-orders due to improved visibility of inventory shrinkage. The last part of the table shows the reduction of finished goods inventory which is also the result of improved inventory visibility. Note that the highlighted numbers are the parameters obtained from the analysis using the BPM. Table 3.1. Computation details of RFID benefits related to inventory shrinkage Benefit Description

Driver/Input Description

Incremental revenue due to reduced shrinkage Shrinkage due to process failure as % of s Shrinkage due to theft as % of sales Shrinkage due to product damage as % of Shrinkage due to other reasons as % of sa Shrinkage as % of sales Net Revenue ($K) Increase in sales Net revenue accounting for increase Reduction of backorder handling cost due to improved vi

Case Level RFID

Impact

Financial Metric

1,000,000

1,005,000

5,000

Revenue Benefit

1.0% 1.0% 1.0% 1.0% 5.0% 1,000,000

0.50% 1.0% 1.0% 1.0% 4.5% 1,000,000 5,000 1,005,000 (2,063)

SG&A Benefit

(1,280)

Inventory Benefit

"As-is" case

1,000,000 9,375

Customer backorders as % of sales 10.0% % reduction in customer backorders Annual number of backorders 2,500,000 Net Revenue ($K) 1,000,000 Average order quantity per order 2 Average price per item ($) 20 Annual number of orders 25,000,000 Annual cost for handling backorders 9,375 Average processing time for backorders ( 0.25 Hourly labor rate 15 FGI Inventory % decrease in FGI due to reduction of shri Finished goods inventory ($K)

7,313 7.80% 22% 1,950,000 1,000,000 2 20 25,000,000 7,313 0.25 15

8,000

6,720

8,000

16% 6,720

Although ignored in this example, the reduction in back-order quantity will improve customer satisfaction, potentially increasing future revenue. The quantitative relationship is, however, a research subject in itself. Once this relationship is known, either from the literature or from empirical studies, it can be included in the BVM as well, with the amount of back-order reduction computed from the BPM.

A Tool Set for Exploring the Value of RFID in a Supply Chain 67

The investment or cost data for the RFID initiative is entered in the BVM. The investments include both one-time costs and recurring costs over the planning horizon. Table 3.2 below shows the investment data used in the example with a planning horizon of 8 years, with an initial pilot program conducted during the first year and the full implementation to be completed by the end of the third year. Table 3.2. Illustrative data of RFID costs Name Hardware ($K) Workstations and Servers Tags and Readers RFID / Label Printer Networking Infrastructure test Total Hardware Costs Software ($K) Software License Fees Software Maintenance Fees Total Software Costs Total Hardware / Software

Category

Investment Type

2005

2006

Hardware Hardware Hardware Hardware Hardware

Capital Capital Capital Capital Capital

456 310 600 100

72 250 600 100

1,466

Capital Expense

Capital Capital Capital Capital

Software Software

Services ($K) Professional Service Costs Software Installation and ConfiguraServices Hardware Installation and Configur Services Tagging Services Pilot and Phase-in Services Total Services

2007

2008

2009

2010

3,866 250 1,200 400

82 250 -

84 250 -

88 250 -

1,022

5,716

332

334

338

800 100 900 2,366

400 100 500 1,522

1,200 100 1,300 7,016

100 100 432

100 100 434

100 100 438

500 200 400 500 1,600

500 200 400 1,100

500 200 300 1,000

500 200 300 1,000

400 200 300 900

400 200 300 900

400 400 43,660

400 400 30,220

400 400 84,160

400 400 18,320

400 400 17,340

400 400 17,380

Others ($K) Other RFID Costs Total Others Total Implementation Costs

Others

Expense

Key Improvement Opportunities ($K)

Revenue Increase

1540

Cost of Sales reduction 5000 2700 250

SG&A reduction Working capital

Figure 3.4. BVM output: Key improvement opportunities

Based on the costs and benefits defined and computed, the BVM provides a high-level summary of the overall business impact of the RFID initiative. It includes a pie chart showing key sources of benefit at the level of the enterprise’s

68

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

profit and loss statement and a graph of projected positive and negative cash flows for the input scenario (Figures 3.4 and 3.5). Standard ROI metrics such as net present value, payback period, and internal rate of return are also reported. There are many other useful functions provided by the BVM. For example, it allows users to define “what-if” scenarios and compare projected financial performance for different scenarios, such as different product demands or different costs of RFID tags. This capability enables users to understand each initiative’s likelihood of success in an uncertain business environment. Another advantage of using the BVM is its ability to show the linkage of financial performance metrics to business value drivers using the decomposition approach described earlier in Section 3.2.1. A user can find the relative contribution of a particular value driver, for example, the impact due to shrinkage discussed above, on the financial performance of the enterprise. Projected Cash Flows ($K) 60,000 40,000 20,000 (20,000) (40,000) (60,000) (80,000) (100,000)

2005

2006

2007

2008

2009

2010

Time Period

Figure 3.5. BVM output: Projected cash flows

3.5 Guidelines to Practitioners This chapter presents a tool set, consisting of a spreadsheet based value model and a simulation based process model, to help analyze the business value of RFID technology in a supply chain. Because the benefits of RFID include both direct and indirect ones, the entire tool set is necessary to cover all potential gains. As we have learned, accounting for a subset of the benefits, for example, only the direct benefits which can be more easily calculated in a spreadsheet, is not adequate and may lead to an unwise decision. Indirect benefits are sometimes not obvious to identify; Lee et al. (2005) give suggestions on areas to look for in a practical situation. To realize some of these indirect benefits, significant changes in business processes or decision-making logic may have to be made, far beyond a drop-in replacement of an existing automatic identification method such as bar coding. One may argue that such process improvements enabled by RFID are the pinnacle of what RFID can offer. But it is important to realize that substantial

A Tool Set for Exploring the Value of RFID in a Supply Chain 69

effort is required to design and deploy the improvements beyond the mechanical replacement of bar coding. Because modeling the indirect benefits using simulation takes significant time and resources, we recommend the application of the tool set in two stages. First, the direct benefits are analyzed using the spreadsheet based value model alone. If a positive business case can be built, there is no need to go further. Otherwise, another study has to be launched to explore where RFID can provide value for that particular business setting, using a simulation model to help design and analyze associated changes. Such a two-stage process will reflect a typical road map for RFID implementation: First as a direct replacement of bar coding, then a redesign of business processes to take full advantage of what RFID can provide. Even though our focus has been on the evaluation of RFID technology, the value modeling tool set is applicable to other new technologies as well. It is increasingly important to focus on indirect benefits, as automation gains wide adoption and further labor savings will have minimal effect on total cost. As mentioned in Section 3.2, our process model is limited to the logistics of the supply chain. Other aspects, such as market intelligence which will lead to better decision-making in how to sell the products or how to respond to market signals quickly, can be influenced by the presence of RFID. These indirect benefits are even more difficult to quantify but can be very important in some businesses. More research is needed in the evaluation of “soft” or “strategic” benefits. These benefits are real and will eventually translate into tangible financial gains, but we do not yet know how to put a value on them. Recent research in estimating the value of flexibility in a supply chain using real options modeling (e.g., Nembhard et al. 2005) is along this line of thought. Once a method is devised to model the benefit, its results can be incorporated into the overall business value model, similar to those of the simulation model described here. We have limited our scope to analyzing the benefits of RFID tags applied to finished products. Tags can be used for raw materials or subassemblies as well. More advanced tags can be written with appropriate information as the material is transformed into the final product. The tool set will be applicable in that situation, although additional benefits will have to be developed in the framework described.

3.6 Acknowledgment Many IBM colleagues participated in various ways in the work reported here. In particular, we would like to acknowledge the contribution of Anthony Bigornia, Sean Campbell, Gerald Feigin, Bill Gilmour, Kathryn Gramling, William Grey, Roger Gung, Tarun Kumar, Erin Livedoti, Doug Maine, Christian Riemann, and Michael Sun.

3.7 References Agarwal V, (2001) Assessing the Benefits of Auto-ID Technology in the Consumer Goods Industry, White paper. Auto-ID Center, Massachusetts Institute of Technology.

70

Ying Tat Leung, Feng Cheng, Young M. Lee, and James J. Hennessy

Alexander K, Gilliam T, Gramling K, Kindy M, Moogimane D, Schultz M, Woods M, (2003) Focus on the Supply Chain: Applying Auto-ID Within the Distribution Center, White paper, Auto-ID Center, Massachusetts Institute of Technology. A.T. Kearney and Kurt Salmon Associates, (2004) Connect the Dots: Harnessing Collaborative Technologies to Deliver Better Value to Consumers, White paper sponsored by the Food Marketing Institute, Grocery Manufacturing Association of America, and National Association of Chain Drug Stores, February 2004. Bagchi S, Baseman RJ, Cheng F, Gung RR, Leung YT, Lin GY, Lu Q, Wu J-Z, (2005) A Business Value Modeling Tool Set for Service-After-Sales Outsourcing, IBM Research Report RJ10348, San Jose, California. Fleisch E, Tellkamp C, (2005) Inventory inaccuracy and supply chain performance: a simulation study of a retail supply chain, International Journal of Production Economics. 95(3):373. Forrester JW, (1958) Industrial dynamics – A major breakthrough for decision-makers, Harvard Business Review 36(4):37–66, July–August 1958. Gaukler G, Seifert RW, Hausman WH, (2004) Item-level RFID in the retail supply chain, to appear in Production and Operations Management. Global Commerce Initiative and IBM, (2005) EPC: A Shared Vision for Transforming Business Processes, White paper. Grey W, Katircioglu K, Shi D, Bagchi S, Gallego G, Adelhelm M, Seybold D, Stefanis S, (2003a) Beyond ROI, Supply Chain Management Review, March/April 2003:20–27. Grey W, Katircioglu K, Bagchi S, Shi D, Gallego G, Seybold D, Stefanis S, (2003b) An analytic approach for quantifying the value of e-business initiatives, IBM Systems Journal 42(3):484–497. Hardgrave B, Waller M, Miller R, (2005) Does RFID Reduce Out of Stocks? A Preliminary Analysis, Research Report, University of Arkansas, November 2005. Helfert EA, (2001) Financial Analysis Tools and Techniques: A Guide for Managers, McGraw-Hill, New York. Joshi Y, (2000) Information Visibility and Its Effect on Supply Chain Dynamics, Master degree thesis, Massachusetts Institute of Technology. Kambil A, Brooks JD, (2002) Auto-ID Across the Value Chain: From Dramatic Potential to Greater Efficiency & Profit, White paper, Auto-ID Center, Massachusetts Institute of Technology. Kang Y, Koh R, (2002) Applications Research, Research Report, Auto-ID Center, Massachusetts Institute of Technology. Lee HL, Padmanabhan V, Whang S, (1997) The bullwhip effect in supply chains, Sloan Management Review 38(3):93–102. Lee YM, Cheng F, Leung YT, (2005) A Quantitative View on How RFID Will Improve a Supply Chain, IBM Research Report RC23789, Yorktown Heights, New York. Luenberger DG, (1998) Investment Science, Oxford University Press, New York. Nembhard HB, Shi L, Aktan M, (2005) A real-options-based analysis for supply chain decisions, IIE Transactions 37:945–956. Raman A, Deoratius N, Ton Z, (2001) The Achilles’ heel of supply chain management, Harvard Business Review, May 2001:2–3. Shishko R, Ebbeler DH, Fox G, (2004) NASA technology assessment using real options valuation, Systems Engineering 7(1):1–12. Wal-Mart Stores, Inc., (2005) Wal-Mart Improves On-Shelf Availability Through the Use of Electronic Product Codes, press release, October 14, 2005.

4 The Effect of RFID On Inventory Management and Control Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

Abstract:

Our thesis is that the evolution of information technology (IT) facilitates the flow of information, which in turn may reduce the variance of an inventory system, and hence its cost. We use ™ˆ‹–T™Œ˜œŒ•Š Gidentification (RFID) as a paradigm. RFID is the latest application of IT to tracking goods and services or anything for that matter, including human beings. It is an evolution from bar code and palette technology, and, in this chapter, we present the argument that RFID is superior in reducing the mean and variance of inventory cycle times. As inventory cost is a function of these (among other variables, such as unit holding and shortage costs), we show that RFID reduces this cost. Also, because RFID leads to rapid transmission of data, it would help avoid excessive inventories and shortages, further reducing total inventory cost. We argue that RFID is superior to existing identification technologies according to mean-variance stochastic dominance. We discuss the ethical implications and the societal trade-offs inherent in RFID, as society must decide how much of its privacy it is willing to curtail in the pursuit of lower prices versus physical security.

4.1 Overview of RFID 4.1.1 Introduction Radio-frequency identification (RFID) is a technology that transmits the identity of an object using radio waves. A typical RFID system consists of a tag (containing the data), a reader, and middleware that interprets the tag information and communicates it to application software. According to Hsieh et al. (2005), RFID tags were invented in 1969 and patented in 1973. So the RFID technology has been around for some time, but only recently has its importance for improving supply chain performance been recognized and increasingly acted upon. RFID is

72

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

envisioned as the principal data capture means of “sense and respond” systems and autonomic, self-healing “adaptive supply chains” (Capone et al. 2004). RFID would improve supply chain operations because “inventory can be tracked more accurately in real-time, resulting in reduced processing time and labor” (Lee et al. 2004). Advertised in Scientific American (Want 2004) as “a key to automating everything,” It is predicted that RFID will experience slow, then exponential growth, from $1 billion in 2003 to $4 billion in 2008 to $20 billion in 2013. WalMart has required its top suppliers to use RFID tags, and the Department of Defense (DoD) has called on its suppliers to adopt RFID labeling. The buyers in this case enjoy monopsonistic power, so suppliers need to conform. The potential applications of RFID are as diverse and far-reaching as befit a truly ubiquitous technology. Wal-Mart sees the value of RFID in tracking inventory, the Food & Drug Administration (FDA) sees it as a potent tool to frustrate drug imitators, and the University of California at Los Angeles (UCLA) will implant chips in cadavers to reduce the illegal sale of body parts (Schwartz 2005). At present, RFID is an inchoate strand of technology with a host of data management problems, for example, coping with the large variety of tags and readers; design and software for data cleaning, filtering, and aggregating, similar in concept but not in goal to what is done in data warehousing (Chawathe et al. 2004). There is also the matter of the present lack of worldwide standards (Erdman 2005). For simplicity, we are using RFID and RFID technology synonymously, even though the terms are not strictly identical. On the one hand, RFID consists of just the tag and readers and perhaps the edgeware for filtering the data and the middleware for routing. On the other hand, the network architecture for sharing information within a supply chain and between supply chains goes beyond the RFID tag and is considered part of the larger RFID technology (Hozak 2006). In connection with RFID, we must mention “velocity management.” Introduced by the U.S. Army in 1995, the velocity management (VM) initiative ushered in a new era for the Army’s logistics management (Drummond et al. 2001). VM seeks to improve the responsiveness, reliability, and efficiency of U.S. Army logistics. This initiative, aptly named, focuses on improving the speed and accuracy of the order fulfillment process. This point is worth belaboring because velocity is more than speed: it is speed coupled with accuracy. Through the improvements accomplished in speed and accuracy, the VM initiative reduces the inventory requirements (read: stockpiles of military hardware) for the desired level of performance. In effect, the VM initiative substitutes velocity for mass as the primary means of supporting army operations. Thus, VM and RFID are often spoken of together because RFID can be a key enabler of the accuracy dimension of VM. One key objective of VM is reducing the mean and variance of order, shipping, and cycle times. The knowledge of what is where at any instant that RFID affords can contribute materially to both the speed and accuracy goals of VM. In logistics, time and cost are positively correlated; so an increase in velocity brought about by RFID will reduce cost.

The Effect of RFID On Inventory Management and Control 73

4.1.2 Organization of the Chapter RFID has so far stirred interest in both academia and industry and is listed as one of the top three technologies of 2004, along with biometrics and nanorobotics (www.rafsec.com). We do a literature review of it in Section 4.2. In Section 4.3, we present five propositions related to the potential impact of RFID on operational performance: We examine the effects of RFID on enhancing information flow in supply chains in terms of lead time length and variability, inventory levels, and inventory cost. In Section 4.4, which includes a sixth proposition, we show how RFID would stochastically dominate, in a mean-variance sense, existing automatic identification technologies, such as the bar code. Section 4.5 is the conclusion and Section 4.6 contains the guidelines for practitioners. Parts of this paper may be found in condensed form in Bagchi and Hayya (2005) and Hayya et al. (2005).

4.2 Literature Review Effective inventory management is a key contributing factor in maintaining a competitive advantage in supply chain operations, and RFID is emerging as a valuable and indispensable technology for efficient inventory management within supply chains. In this section, we review the impact of RFID technology on inventory management and control in supply chains. Reflections of scholars on the ethical implications of the use of RFID technology will also be addressed. 4.2.1 Historical Overview IT can enhance an organization’s competitive advantage through improvement and efficiencies in operational performance. Janoff (2001) says that 80% of CEOs claim that information technology is critical to the success of their companies. This fits in with inventory management and control because modern inventory management practices require accurate product identification and tracking, coupled with real-time information collection and electronic data interchange (EDI) between members of a supply chain network. The efficiency and quality of these actions depend on the type of IT in place. Firms need a persuasive logic for investing in inventory tracking technologies. For example, Raman et al. (2001) delve into the Gamma Corporation (actual name disguised), a big retailer with about 37,000 SKUs. They start by saying that “more than 65% of the inventory records were inaccurate at the store-SKU level (i.e., system inventory did not match physical inventory)” (p.137). Coupling their investigation at Gamma with data from two other leading retailers and interaction with others, they draw the conclusion that the two factors that hamper the effective use of IT in retail (according to Raman et al., retailers were spending $30 billions annually on IT) are inaccurate inventory records and misplaced SKUs. They go on to say that at the two retailers, these two factors were responsible for a 10% decline in profits. Now suppose that in terms of evolution, the precursor to RFID was the bar code. Since its inception in the late 1950s, bar coding has been the workhorse technology for information control of inventories in supply chains at the product

74

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

level. (A detailed analysis of the development and implementation of bar coding technology in inventory systems can be found in Manthou and Vlachopoulou 2001). But in the past decade, RFID technology has generated considerable interest among supply chains practitioners as an enabler of real-time product identification and tracking. Of course, RFID has distinct advantages over the bar code. Whereas bar coding exhibits control at the product level, RFID uses an electronic product code (EPC) to provide a unique identity to each individual product, thereby enhancing the tracking and control of inventories within a supply chain. RFID-tagged products allow retailers to know the exact location and quantity of inventories within a supply chain without resorting to physical inventories or cycle counts. A detailed comparison of RFID and bar code technologies is found in Jones et al. (2005). We argue that increased accuracy in inventory record keeping afforded by RFID will improve a firm’s ability to satisfy the customer by having the right product in the right place at the right time. For a summary of the technical aspects of RFID, the reader is referred to Finkenzeller (2003). 4.2.2 Studies of RFID in Supply Chain Management Cavinato (2005) reports that logistic, supply chain, and purchasing practitioners perceive RFID (and other wireless technologies) as a strategic initiative that will play a salutary role in the revenue and growth of their enterprises. Similar positions on the benefits and opportunities for RFID adoption are adopted in the retail trade literature and in several white papers, for example, NCR (2003) and IDI (2002). Spending on RFID technology in the United States, it is estimated, will top $1.3 billion in 2008 (Douglas 2005). Despite the appeal of this vast amount, formal research papers on the benefits of RFID technology in inventory and supply chain management have been scant. Several authors have examined the impact of RFID implementation. Prater et al. (2005) examine the effect of RFID implementation on supply chain management in the grocery industry. In a study of supply chain management practices in the children’s toy industry, Wong et al. (2005) report that main-order retailers within the industry view RFID technology as a key future supply chain initiative for reducing the risks associated with inventory. Walker (2005) outlines the contribution of RFID technology to the information architecture of supply chains. Kärkkäinen and Holmström (2002) discuss RFID as an enabler for efficient information-handling in support of sourcing and manufacturing operations within supply chains. Theoretical and simulation-based models have demonstrated how RFID technology can support initiatives to improve inventory and supply chain performance metrics. Kang and Gershwin (2005) investigated the causes and impact of inventory inaccuracy (e.g., the stock loss problem) on the performance of inventory systems. They developed both an analytical and a simulation-based stochastic model for the widely used single-item (Q,R) reorder point inventory control system. RFID was used to justify an errorless inventory system, thereby establishing the baseline for ordering decisions with perfect accuracy. The best performance, as measured by the stock-out rate and amount of inventory held, was

The Effect of RFID On Inventory Management and Control 75

achieved under the RFID errorless system. The findings of the analytical and simulation models revealed that failure to correct even small rates of inventory inaccuracy leads to severe stock-out penalties. Lee et al. (2004) use a simulation methodology to study the effects of inventory accuracy, shelf level improvement, and inventory visibility on inventory reduction and level improvement in a three-stage supply chain (manufacturer, distribution center, and retailer). The simulation comprises three case studies focused on inventory management practices with and without RFID. Within each case study, multiple simulation scenarios were run. We report the results of the scenarios with the greatest degree of RFID involvement compared to a non-RFID baseline scenario. The first case study determines the parameters of the retailer’s (s,S) system, with and without knowledge of the prevailing level of inventory accuracy (s is the reorder point, whereas S is the order-up-to level). Exact knowledge of the level of inventory accuracy is used as the proxy for the effect of RFID technology. In the simulation, average inventory held decreases by 16% and total back-orders decrease by 22% when the (s,S) policy decisions are made with accurate inventory information. The second case study investigates how the frequency of shelf replenishment affects shelf inventory, backroom inventory, and lost sales. A daily manual shelfmonitoring policy is compared to an RFID policy of continuously monitoring shelf inventory. Under the RFID policy for shelf monitoring, the total quantity of lost sales decreases by 84%, and average shelf inventory is reduced by 16%. Backroom average inventory decreases by 30%, and the retailer’s average inventory decreases by 23%. The third case study examines how the visibility of inventory data (periodic without RFID and in real-time with RFID) affects the manufacturer’s production lot-sizing decision. The manufacturer’s production quantity decision is determined with and without real-time RFID-enabled data from the distribution center. With RFID in this simulation, average inventory at the distribution center decreases by 47%, and back-orders are eliminated. As for other papers, Hardgrave et al. (2005) compares, over a 29-week period, the frequency of out-of-stock items for 4,554 different SKUs for 12 Wal-Mart stores using RFID technology and 12 equally matched stores without RFID. Out-of-stocks are reduced by 16% and faster shelf replenishments are achieved in RFID stores compared to non-RFID stores. Based on a review of the trade literature across several industries, Jogleker and Rosenthal (2005) summarize the status of RFID implementation in the supply chains of forty-six firms. No firm confessed to a full-scale implementation of RFID throughout its supply chain and that, for the most part, RFID implementation was being driven by mandates from customers. The most common advantage shared by thirteen of these firms is improved inventory visibility in the supply chain followed by increased efficiencies in supply chain management (12 firms) through reducing inventory-related costs. These companies presumably intend to do before and after-RFID analysis of the cost of monitoring inventory, and the authors conclude that “... extensive voluntary RFID application for high volume low-margin goods will probably arise only when the tag cost falls to five cents … from its present level of about twenty cents” (p.11).

76

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

4.2.3 Ethical Considerations Despite the many advantages of RFID, ethical concerns over the invasion of privacy by RFID are gathering. These concerns are directed against outside forces, such as marketers, DoD, organized crime, petty criminals, or just others who will abuse private information gathered through RFID. Weinburg (2006) summarizes the purported dangers of RFID to privacy across these three dimensions: Surveillance – fear that RFID technology allows geographic tracking of a person’s movement; Profiling – information collected from RFID tags found in a consumer’s possessions, clothing, identification documents, etc., can be used to create and maintain a profile of actions and behavior of that consumer; Business Action – profile and surveillance information could be integrated to create specific action plans, such as advertising and pricing campaigns, that could be directed at individual consumers. A fundamental criticism of RFID technology in the retail industry is that, attached or embedded, RFID tags on products can be used to collect personal information on buying behavior without the buyer’s knowledge. Hargraves and Shafer (2004) present a series of hypothetical scenarios that highlight the potential privacy risks ensuing from purchasing an RFID-tagged product. As mentioned before, privacy threats may come from a variety of agents, such as marketers, the government, and criminals who may continue to milk an RFID tag after the point of sale. In giving a list of reasons for privacy, Wayner (2002, pp. 2–3) adds to the list drug dealers, terrorists, child pornographers, and money launderers. Consumer backlash over privacy issues has now led to the creation of a number of advocacy groups, such as Consumers Against Supermarket Privacy Invasion and Numbering (CASPIAN), who, among numerous other groups easily identified by a simple internet search (for example, at www.spychips.com), are critical of RFID and view the technology as an aggression against their constitutional rights. The intensity of the backlash may have been amplified by some recent consumer research activities reported in the popular press. Wolinsky (2003) reports how remotely placed Proctor and Gamble researchers employed video technology to observe unknowing customers interacting with RFID-tagged lipsticks at a WalMart cosmetics counter. Tesco, the largest supermarket chain in Britain, conducted a 6-month study of RFID “smart shelves” in which unknowing shoppers had their images recorded when removing tagged Gillette razors from the retail shelf. Images were again recorded when the consumer completed payment at the checkout kiosk (Gilbert 2003). Dick Cantwell, vice-president of the Gillette division and head of the Proctor and Gamble RFID initiative, views the ability to track products that are popular items for pilferage as a strong argument for accelerated RFID deployment (Roberti 2006). Consumer concern has stimulated government legislation and also the development of RFID jamming devices. In 2004, proposals were introduced into the California State Legislature that, if passed into law, would require businesses to notify consumers if they are using an RFID system to track and collect information (Stanton 2005). In July 2005, the California Senate approved a ban on the use of RFID technology on state-issued forms of identification, such as driver licenses (in

The Effect of RFID On Inventory Management and Control 77

counterpoint, the British are tagging license plates.) Similar bills are under study in Missouri, Utah, and Virginia (Swedberg 2004). Technological innovations in the form of “blocker-tags” that act as RFID jamming tools are under development as a means to foil privacy invasion (Juels et al. 2003). Perhaps RFID tags will evolve to selectively choose which data are stored, which data are not, and which data are hidden and from whom (Wayner 2003). For example, critics wonder what companies adopting RFID would do with the massive data that RFID produces. But according to Wayner (2003), this eventuality could be resolved by the use of translucent technology algorithms that would produce just what is needed.

4.3 RFID, Lead Time, and Inventory Cost In what follows, we present our principal argument in the form of five propositions. Following the statement of each proposition, we discuss why the proposition should hold. Proposition 1 (P1) RFID will reduce average lead time. Most introductions of new technologies experience both expected and unexpected difficulties of varying degrees, which often result in a temporary dip in system performance. RFID will not be an exception. It will be the rare organization that avoids these bumps on the RFID implementation highway. But after the initial difficulties have been resolved and the system has settled to a steady state, it is difficult to imagine how RFID, barring a technology failure, can increase average lead time. Lead time may consist of at least one or more of the following: Order entry time, manufacturing time, shipping time, order receiving time, and plain waiting time when the order is waiting to be worked on. Indeed, waiting time may actually be the single largest portion of lead time. It is not plausible that RFID would increase any of the aforementioned components of lead time, and it is likely that RFID would decrease manufacturing and order receiving times. Furthermore, RFID should reduce order waiting time because information unavailability is often an important reason why orders have to wait. Examine, for example, the impact of RFID on the bullwhip effect (BE). RFID’s ability to reduce average lead time leads to a dampening of the BE. As an illustration, consider a simple two-stage supply chain consisting of a retailer who faces external customer demands and a manufacturer who supplies the retailer. Under assumptions that the retailer faces a fixed lead time, follows a simple periodic review policy, and uses a simple moving average of length p to forecast customer demand, Simchi-Levi et al. (2003) have shown that the BE can be expressed as BE

Var( Q ) 2 L( p  L ) , t 1 Var( D ) p2

(4.1)

78

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

where Var(Q) is the variance of orders placed by the retailer with the manufacturer, Var(D) is the variance of customer demand, L is the lead time, and p is the length of the moving average. What can be noted immediately from this formula is that the BE increases with the average length of the lead time, L. Hence, it is logical to deduce from Proposition P1 that RFID will reduce the BE. Note also that the lower bound of BE, call it BELB , depend on the two parameters, L and p. In Figure 4.1, we plot BELB as a function of L by fixing p. We see that BELB, which can be approximated by a power function, is quickly reduced with a decrease in the lead time, given a small number of observations in the moving average forecast. The power function shown in Figure 4.1 provides not only an excellent approximation for a quick calculation of BELB, but it is also a view of the relationship between BELB and L.

p 3 4 5 8 10

Approximation BELB = 1.5002*L 1.234 BELB = 1.2817*L 1.0888 BELB = 1.1701*L 0.9727 BELB = 1.0392*L BELB = 1.0088*L

R2 0.98 0.97 0.96 0.95 0.94

Figure 4.1. The relationship between the lower bound of the bullwhip effect and lead time

To see the joint effects of (p, L) on the retailer’s order variability, one needs to plot BELB as a function of both p and L. The plot will show that BELB has an inverse relationship with p and L; that is, reducing lead time, L, or increasing the number of observations, p, will both quickly reduce order variability.

The Effect of RFID On Inventory Management and Control 79

Proposition 2 (P2) RFID will reduce lead time variance. Proposition P2 seems even more axiomatic than Proposition P1. Note that RFID enhances the transmission of information along a supply chain, and variance and information availability are inversely related. Information can be defined as the lower bound of the reciprocal of the variance (Kendall and Buckland 1967); alternatively, information could be equated to “precision,” also the reciprocal of variance (Cochran 1977). For probabilistic systems, the most commonly used measure of information is entropy, which we now introduce. Let X be a discrete random variable with alphabet Ȍ and probability mass function p(x) = Pr{X = x}, x İȌ. Then the entropy H(X) of X is given by (Cover and Thomas 1991) H(X) = T ™ x İȌ {p(x) log p(x)} ,

(4.2)

Similarly, if X is a continuous random variable with support set S and probability density function f(x), then the entropy H(X) of X is given by (Cover and Thomas 1991)

³

H(X) = T S f ( x ) log f ( x ) dx ,

(4.3)

provided that the integral exists. Example 4.1 Suppose that X is the location of a piece of inventory. Let p be the probability that the inventory is in a particular warehouse at a particular time, (1 – p) that it is not. Now, the entropy of a Bernoulli random variable X with parameter p is H(X) = T p log(p) – (1 – p) log(1 – p).

(4.4)

and the variance of X is given by V(X) = p(1 – p).

(4.5)

The graphs of H(X) and V(X) against p are shown in Figure 4.2. As can be seen, both H(X) and V(X) reach their maximum at p = 0.5, and their minimum at p = 0 and p = 1. Clearly both entropy and variance tell the same story, and entropy is a measure of uncertainty (Applebaum 1996), just as the variance is. Consider now two situations: One in which a technology, such as RFID, affords perfect knowledge of X, and the other in which no such technology exists and p is subjectively estimated to be 0.5. The first situation would lead to both entropy and variance equal to zero, whereas in the second, both entropy and variance would be at their maximum value. This particular scenario makes the best case for RFID vis-à-vis its predecessor technologies because we have assumed a perfectly reliable RFID system in the first situation and complete ignorance of X in the second. But as long as RFID moves the estimate of p from the center of the spectrum in Figure

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

80

4.2 toward either end, P2 should hold.

Entropy/ variance

H(x) 1

V(x)

1/4

0

p 0

1/2

1

Figure 4.2. Entropy and variance of a Bernoulli random variable

Example 4.2 Suppose that lead time X is a random variable uniformly distributed between a and b, where b > a. Then the entropy of X (Applebaum 1996) is H(X) = log(b – a),

(4.6)

and the variance of X is V(X) = (b – a)2/12.

(4.7)

Again, as in Example 4.1, when entropy increases, so does the variance. Example 4.3 Suppose that lead time X is a normally distributed random variable

The Effect of RFID On Inventory Management and Control 81

with mean ȝ and variance H(X) = log [ı (2ʌ e)

1/ 2

V 2.

Then the entropy of X (Applebaum 1996, p. 159) is

].

(4.8)

Note that H(X) is a function only of ı, underscoring the strong relationship between entropy and variance. Proposition 3 (P3) RFID will reduce both the mean and variance of lead time demand. To see how P3 follows from P1 and P2, let X represent lead time demand, L stochastic lead time, and D stochastic demand rate. Let E and V denote the expected value and variance operators. Assuming that L and D are independent, we obtain (Bagchi et al. 1983) E(X) = E(L) . E(D),

(4.9) 2

V(X) = E(L).V(D) + [E(D)] .V(L).

(4.10)

It is evident that V(X) decreases if either E(L) or V(L) decreases. Proposition 4 (P4) RFID will reduce inventory levels. We advance three arguments in favor of P4. First, a reduction in average lead time should lead to a reduction in pipeline or transit inventory. Hence, P4 follows from P1. Second, a reduction in the variance of lead time demand should lead to a reduction in safety stock. It is common practice in inventory management to maintain safety stock levels based on a service level criterion, usually the fill rate or the probability of stocking out in a replenishment cycle (Silver et al. 1998, p. 245). The service level corresponds to some upper percentile of the convolution of lead time demand. If lead time demand were approximated by the normal distribution, then the reorder point is given by s = E(X) + k V X ,

(4.11)

where k is the safety factor, easily found in the standard normal variate probability table and ıX is the standard deviation of lead time demand. The safety stock is kıX, and thus P4 follows from P2. Third, and on a broader note, the justification for P4 is provided by the so called OM (operations management) triangle of Lovejoy (1998), the three corners of the triangle are information, inventory, and capacity (Figure 4.3). The OM triangle has its genesis in the Pollacek-Kinchine formula (Heyman and Sobel 1984), which can be stated as [(Inventory) (Capacity/ Demand - 1) / (Variability)] = constant.

(4.12)

82

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

Capacity (What we can do)

Information (What we know)

Inventory (What we have)

Figure 4.3. Capacity, inventory, and information as substitutable ingredients of system performance

If variability is viewed as the inverse of information, then for a fixed demand, the three quantities (inventory, capacity, and information) are substitutes in the following sense: If more of one is available, then less of one or both of the others is needed for the same level of system performance. The trade-off suggested by the OM triangle captures a fundamental truth about operations. We argue that RFID will increase the amount of useful data (good information) collected; so with capacity at the same level as before, it would be possible to provide the same system performance with lower inventory. Proposition 5 (P5) RFID will reduce the long-run average cost of inventory. If by inventory cost is meant the cost of holding inventory, then P5 follows from P4. On the other hand, if inventory cost includes, in addition to inventory holding cost, other costs, such as the cost of ordering and the penalty cost for stock-outs, then a more sophisticated argument is called for. Such an argument is provided by Song (1994) who considered exogenous lead times, compound Poisson demands, and long-run average inventory system cost composed of ordering, holding, and penalty costs. Ordering costs were proportional to the order quantity, holding costs were linear in inventory on-hand, all stock-outs were back-ordered, and penalty costs were linear in back-orders. Song shows that a more variable lead time always leads to higher long-run average inventory system cost for any fixed basestock policy. Thus P5 follows from P2. He et al. (2005) show that the impact on inventory cost of lead time variability is approximately linearly increasing in the standard deviation of lead time, with a loose upper bound on the increase in cost of

The Effect of RFID On Inventory Management and Control 83

D(bh)

0.5

VL ,

(4.13)

where D is the (deterministic) demand rate, b the shortage cost per unit per unit time, h the holding cost per unit per unit time, and V L the standard deviation of lead time. The following tight upper bound is given by Kim et al. (2004): 2

D (b+h)V(L)/2Q ,

(4.14)

where V(L) is the variance of lead time and Q is the order quantity.

4.4 RFID and Stochastic Dominance Stochastic dominance is concerned with the condition when a particular random prospect is better than another random prospect. By “better,” we mean preferable from the perspective of a reasonable person. So, if every reasonable person prefers a random prospect, say the RFID technology, to another random prospect, say bar code, then we would say that RFID stochastically dominates bar code. 4.4.1 Definitions First-order Stochastic Dominance: Given wealth (W), we say that system A with cdf (cumulative distribution function) FA (W ) dominates system B with cdf FB(W ) according to first-order stochastic dominance, if for all nondecreasing utility functions (Copeland and Westin 1988), FA (W ) ” FB (W ) , for all Wi; and FA (Wi ) < FB (Wi ) , for some Wi.

(4.15)

Example 4.4 Let the distribution of A be normal with mean 20 and variance 16, abbreviated as N(20,16), and that of B be N(10,36). Then, distribution A dominates B according to first-order stochastic dominance. The term “stochastic dominance” makes sense in this case because the returns of A have a higher mean but a lower variance than the returns of B. Given an arbitrary return wo, P(W t wo| A) t P(W t wo| B). See Figures 4.4 and 4.5.

(4.16)

84

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

Figure 4.4. N(20, 16) vs.N(10, 36)

Figure 4.5. First-order stochastic dominance: N(20,16) stochastically dominates N(10, 36)

First-order stochastic dominance applies to all types of increasing utility functions and is particularly useful when the cdf of one of the two random prospects under study is entirely to the right (that is to say, below) of the cdf of the other random prospect. But suppose that the utility function is nondecreasing and strictly concave, as it would be for a risk-averse person, and neither cdf is entirely to the right (or below) of the other. The first-order stochastic dominance criterion fails to

The Effect of RFID On Inventory Management and Control 85

provide a preference ordering of the two random prospects. This leads us to the notion of second-order stochastic dominance whose concern is ordering random prospects according to relative risk in terms of the spread of the probability mass of the cdf. Second-order Stochastic Dominance: System A dominates system B according to second order stochastic dominance, if (Levy 1992) Wi

³ [ F (W )  F (W )]dW t 0, for all W ; F (W ) z F (W ), for some W . B

f

A

B

i

A

i

i

(4.17)

Equation (4.17) means that at some Wi , the area from the left to Wi under the cdf of A must be less than or equal to the area from the left to Wi under the cdf of B. The reasoning behind this definition may be understood by considering two normal probability distributions of W with the same mean of 10 but with standard deviations 2 and 4. Now the areas from the left under the cdfs connote accumulated wealth, and thus, for all Wi , the accumulated wealth to the right of Wi ( • Wi ) under A is greater than the corresponding area under B. These are the areas of interest because they tell us that the prospects for system A dominate those for system B. See Figures 4.6 and 4.7. Note that first-order stochastic dominance implies second-order stochastic dominance, but not vice versa. What is more to our purpose is that the concept of spread that is inherent in second-order stochastic dominance is not, strictly speaking, the same as the concept of variance, although the two must be close. A consequence of the difference between spread and variance is that variance reduction, holding the mean constant, is no guarantor of an increase in expected utility. There are situations where a mean-preserving reduction in variance may decrease expected utility (Levy 1992). Nevertheless, because the mean and variance are often the only parameters used in practice to describe a random prospect, it is intuitively appealing and practically useful to have a notion of stochastic dominance that can order two prospects where the means are the same but the variances are different or the means are different but the variances are the same. With that in mind, we introduce a version of stochastic dominance called mean-variance stochastic dominance.

86

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

Figure 4.6. N(10,4) vs.N(10,16)

Figure 4.7. Second-order stochastic dominance: N(10,4) dominates N(10,16)

Mean-variance Stochastic Dominance: Given wealth (W), we say that system A with mean EA(W) and variance VA(W) dominates system B with mean EB(W) and variance VB(W) according to mean-variance stochastic dominance, if for all nondecreasing utility functions,

The Effect of RFID On Inventory Management and Control 87

E A (W) • E A (W) and V A (W) ” V B (W)

(4.18)

with at least one of the two inequalities holding strictly. The mean-variance approach to ordering random prospects is a basis for portfolio selection and capital asset pricing theories in corporate finance. It has been proposed for making decisions under price uncertainty (Meyer 1987) and used by Post (2001) to incorporate parameter uncertainty and risk aversion in data envelopment analysis (DEA). We note two facts. First, if the random prospects are completely characterized by mean and variance, as is the case for normally distributed random variables, then second-order stochastic dominance is equivalent to mean-variance stochastic dominance (Tobin 1958, 1969). However narrow this result may seem from a theoretical perspective, it is a windfall for the meanvariance approach. Second, Levy (1973) extended the previous result to include lognormal distributions, and Bigelow (1993) has characterized the entire class of random variables for which the mean-variance approach is consistent with expected utility analysis. This class of random variables is completely ordered by Rothschild-Stiglitz (1970) increasing risk after normalization by their means. It may also be noted that the variance can be used as the risk index when the utility function is quadratic. See Levy (1992) for a survey of stochastic dominance and expected utility. 4.4.2 Our Proposition We now present our main observation in this section in the form of a proposition. Proposition 6 (P6) RFID will dominate existing automatic identification technologies, including bar code, according to mean-variance stochastic dominance. Consider profit ( S ) as revenue (R) minus cost (C). We obtain E( S ) = E(R) - E(C) V( S ) = V(R) + V(C) – 2COV(R,C),

(4.19) (4.20)

where COV(R,C) is the covariance of revenue and cost. From P5 and (4.19), it is clear that RFID will likely increase E( S ). It is reasonable to assume that RFID will not increase V(C). To the contrary, RFID may reduce V(C), because cost is a system attribute, and we have argued before that, by enhancing information flow, RFID will reduce system variance. The impact of RFID on V( S ) is thus likely to be a decrease. P6 now follows from P5, because a smaller mean and variance for cost would translate into a larger mean and a smaller variance for profit. No distinction is made in this paper between the utility of wealth and net profit.

88

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

The objective in inventory management should be to maximize “net utility,” which is approximated by profit (Arrow et al. 1951). “Gross utility” would correspond to “gross revenue,” and cost is a disutility. As cost is a disutility, so is lead time because customers want shorter and more predictable lead times. Suppose that lead times under Technology A have mean EA(L) and variance VA(L), whereas lead times under Technology B have mean EB(L) and variance VB(L). Suppose further that EA(L) ” EB(L), VA(L) ” VB(L).

(4.21) (4.22)

and that one of the two above inequalities is strict. It is then possible to argue that Technology A stochastically dominates Technology B in the mean-variance sense. Hence, P1 and P2 together provide another argument for P6.

4.5 Conclusions The RFID technology is widely regarded to have the potential to revolutionize both manufacturing and service operations. Of all the motivations for this technology – efficiency, security, and accuracy, to name a few -- none captures the imagination of supply chain professionals more than the opportunity to reduce systemwide cost. One key component of supply chain system cost is inventory. In this chapter, we argue that RFID will reduce inventory cost. The impact of RFID on inventory will be the result of using good information in inventory management to deliver good customer service. And it will be the result of RFID’s impact on what customers perceive as good service, namely, shorter lead times. RFID should reduce average lead times as more speedy and accurate information enables a faster and more accurate response all along the supply chain. RFID will thus also reduce the variance of lead time, because of the inverse relationship between system variance and information availability. A reduction in either the average lead time or the variance of lead time would be enough to reduce the variance of demand during lead time, and, consequently, of safety stock requirements. A reduction in average lead time also leads to a reduction in pipeline or in-transit stock. We have attempted to show that RFID will dominate - not absolutely, inevitably, or in every instance, but stochastically, practically, and discretely – other existing identification technologies. In this chapter, we have been like a drone hovering above to get a glimpse of the terrain, and we think that the terrain in the direction of RFID will be a definite improvement.

4.6 Guidelines for Practitioners As an element of information technology, today RFID is what bar code technology was a few decades ago: promising, expensive, frustrating, and also inevitable. The future will be like the past, only more so. The technology is still amorphous, not

The Effect of RFID On Inventory Management and Control 89

yet sufficiently robust. Varying levels of implementation – pallet, case, or item tagging - have varying levels of costs and benefits. And companies trying to implement RFID will need to make significant investments in training. The costs for consulting, system integration, internal projects, tags, and readers can be substantial. Yet, the overall story of RFID should not be dissimilar to that of bar code technology. Fixed costs will come down, as will the operating costs, and the RFID technology will become increasingly robust. In effect, the cost of the new technology becomes the cost of doing business. We know that supply chains are becoming more and more data-driven. Higher data quality and reliability are crucial to attaining and maintaining competitive advantage. As discussed, RFID is superior to bar coding and manual collection methods, such as cycle counting, in capturing real-time product data within the supply chain. Currently, it may be too expensive for firms to adopt RFID technology throughout a supply chain. However, an experience curve effect in the manufacturing of RFID technology will drive the cost of the technology down, and, in the near future, the RFID technology will become more cost-feasible. Supply chain practitioners who are laggards in the implementation of RFID technology face the loss of competitive advantage. For a company that has been using bar code technology, but has no first-hand experience with RFID, we make several recommendations. At a minimum, have a surveillance program that scans the environment for RFID developments, that monitors significant developments, and that carefully tracks progressions that may require a competitive response. So, at first, undertake an RFID project of limited scope, thus reducing both the initial hardware investment and the impact on ongoing operations to learn about RFID firsthand, to generate data on the impact of RFID on lead time, inventory, and cost, and to build a business case for RFID. In particular, generating comparative data on bar code and RFID would be extremely useful. In doing a cost-benefit analysis, consider both the short term and the long term. If your suppliers or customers are experimenting with RFID, we recommend RFID projects in partnership with one of those suppliers and customers. If any of your competitors is experimenting with RFID, be vigorous in understanding the competitive implications for your business. The market share game is a zero-sum game, and a competitor who improves customer service with the help of RFID is likely to permanently wrest market share away from you. However, as in the case of the cost-benefit analyses of just-in-time (JIT) manufacturing techniques a few decades back, traditional capital budgeting methodologies alone may be inappropriate to justify the implementation of RFID technology. Often, the implementation of JIT technology was justified from a perspective of strategic survival. In this chapter, we posit six propositions regarding RFID that speak to the concerns of the supply chain practitioner, for example, reducing inventory cost and lead time and its variability. We introduce the concept of stochastic dominance to demonstrate the strategic benefit of integrating RFID technology in supply chain operations. A few decades from now, RFID will be ubiquitous, and supply chain practitioners will scarcely believe that past generations did without it. We think that the innovative use of RFID technology in supply chain management, health care, and security will continue to grow. Paralleling this growth will be an

90

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

increasing awareness and concern over the ethical use of RFID technology. Increased regulatory policy on the use of RFID technology is inevitable. For example, in retail, RFID tags may be mandated to be permanently deactivated at the point of sale. We close by hoping that the benefits espoused by RFID enthusiasts can be amicably balanced with the public’s twin concerns for privacy and safety.

4.7 Acknowledgment The last-named author acknowledges with thanks the support from the Center of Supply Chain Research, College of Business, Penn State University.

4.8 References Applebaum, D, (1996) Probability and Information. Cambridge University Press. Arrow KJ, Harris T, Marschak J, (1951) Optimal inventory policy. Econometrica 19: 250– 272. Bagchi U, Hayya JC, Ord JK, (1983) The Hermite distribution as a model of demand during lead time for slow-moving items. Decision Sciences 14: 447–466. Bagchi U, Hayya JC, (2005) RFID, Inventory costs, and velocity management. Proceedings of the IEEE International Conference on Service Operations and Logistics, and Informatics 1048–1053. Bigelow, JP, (1993) Consistency of mean-variance analysis and expected utility analysis. Economics Letters 43: 187–192. Capone, GD, Costlow D, Grenoble WA, Novack RA, (2004) The RFID-Enabled Warehouse. Working Paper, Center for Supply Chain Research, Smeal College of Business, Pennsylvania State University at University Park. Cavinato J, (2005) Supply chain logistics initiatives. International Journal of Physical Distribution and Logistics Management 35(3):148–151. Chawathe, SS, Krishnamurthy V, Ramachandran S, Sarma SE, (2004). Managing RFID data. Proceedings of the 30th VLDB Conference 1189–1195. Cochran, WG, (1977) Sampling Techniques. Wiley, New York, 3 Ed. Copeland TE, Westin JF, (1988) Financial Theory and Corporate Policy. Addison-Wesley, Reading, MA, 3 Ed. Cover, TM, Thomas, JA, (1991) Elements of Information Theory. Wiley-Interscience, New York. Douglas R, (2005) Bar Codes vs RFID. Toronto Globe and Mail, February 10. Dummond, J, Brauner M, Eden R, Folkeson JR, Girardini KJ, Keyser D, Pint EM, Wang M, (2001) Velocity Management: The Business Paradigm That Has Transformed U.S. Army. Rand Publications, MR-1108-A. Erdman Center for Operations and Technology Management (Spring 2005). Radio Frequency Identification – A Perspective on RFID and the Future of Supply Chains. Working Paper, School of Business University of Wisconsin-Madison. Finkenzeller K, (2003) RFID Handbook. Wiley, New York, 2 Ed. Gilbert A, (2003) Cutting-edge “Smart Shelf’ test ends. CNET News, August 22. Hardgrave BC, Waller M, Miller R, (2005) Does RFID reduce out of stocks? A preliminary analysis [on-line]. Working Paper, RFID Research Center-Information Technology

The Effect of RFID On Inventory Management and Control 91

Research Institute, Sam M. Walton School of Business, University of Arkansas. Available from http://itrc.uark.edu. Hargraves K, Shafer S, (2004) Radio Frequency Identification (RFID) Privacy: The Microsoft Perspective [on-line]. Available from http://www.microsoft.com/two. Hayya JC, Bagchi U, He XJ, Kim JG, Pan AC, (2005) Supply chain analysis: RFID & inventory control. Proceedings of the 36th Annual Meeting of the Decision Sciences Institute 11861–11866. He X, Kim J, Hayya JC, (2005) The cost of finite lead time variability in inventory analysis. International Journal of Production Economics 97(2): 130–142. Heyman D, Sobel M, (1984) Stochastic Models in Operations Research, Vol. I. McGrawHill, New York. Hozak K ([email protected]) (2nd February 2006). Personal email to J.C. Hayya ([email protected]). Hsieh C-T, Sharp T, Lin B, (2005) RFID and its values as perceived by college-age consumers. Proceedings of the 36th Annual Meeting of the Decision Sciences Institute 13841–13846. Interaction Design Institute , IDI (2003) RFID A Week Long Survey on the Technology and Its Potential [on-line]. White paper, 1–47. Available on http://people.interaction-

ivrea.it/ c. noessel/RFID/RFID_research.pdf Janoff B, (2001) Hot wired. Progressive Grocer 80(1):53–56. Jogleker N, Rosenthal S, (2005) Experimentation with RFID usage in supply chains. POMS Chronicle 12(1): 10–11. Jones M, Wyld D, Totten J, (2005) The adoption of RFID technology in the retail supply chain. The Coastal Business Journal 4(1): 29–42. Juels A, Rivest RL, Szydlo M, (2003) The blocker tag: Selective blocking of RFID tags for consumer privacy. In Atluri, V. Proceedings of the Eighth ACM Conference on Computer and Communications Security, ACM Press, 103–111. Kang Y, Gershwin SB, (2005) Information inaccuracy in inventory systems: Stock loss and stockout. IIE Transactions 37:843–859. Kärkkäinen M, Holmström J, (2002) Wireless product identification: Enabler for handling efficiency, customisation and information sharing. Supply Chain Management 7(4):242–252. Kendall MG, Buckland WR, (1967) A Dictionary of Statistical Terms. Hafner, New York. Kim JD, Sun D, He X, Hayya JC, (2004) The (s, Q) inventory model with Erlang lead time and deterministic demand. Naval Research Logistics 51(6):906–923. Lee Y, Cheng MF, Leung YT, (2004) Exploring the impact of RFID on supply chain dynamics. In Ingalls, RG., Rossetti, MD, Smith, JS, Peters BA, Proceedings of the 2004 Winter Simulation Conference 1145–1152. Levy H, (1973) Stochastic dominance among log-normal prospects. International Economic Review 14:601–614. Levy H, (1992) Stochastic dominance and expected utility: Survey and analysis. Management Science 38(4):555–593. Lovejoy WS, (1998) Integrated operations: A proposal for operations management teaching and research. Production and Operations Management 7(2):106–124. Manthou V, Vlachopoulou M, (2001) Bar code technology for inventory and marketing management systems: A model for its development and implementation. International Journal of Production Economics 71:157–164. Meyer J, (1987) Two-moment decision models and expected utility maximization. American Economic Review 77: 421–430.

92

Uttarayan Bagchi, Alfred Guiffrida, Liam O’Neill, Amy Zeng, and Jack Hayya

NCR Corporation (2003) 50 Ideas for Revolutionizing the Store through RFID [on-line]. White paper, 1–46. Available on http://www.ncr.com/en/repository/articles/articles.htm Prater E, Frazier GV, Reyes PM, (2005) Future impacts of RFID on e-supply chains in grocery retailing. Supply Chain Management 10(2):134–142. Post T, (2001) Performance evaluation in stochastic environments using mean-variance data envelopment analysis. Operations Research 49(2):281–292. Raman A, DeHoratius N, Ton Z, (2001) Execution: The missing link in retail operations. California management Review 43(3):136–152. Roberti M, (2006) P&G Adopts EPC advantaged strategy. The RFID Journal 24:1–2. Rothschild M, Stiglitz JE, (1970) Increasing risk: I. A definition. Journal of Economic Theory 2: 225–243. Schwartz J, (2005) Graduate Cryptographers Unlock Code of “Thiefproof” Car Key. New York Times, 29 January, A10. Silver EA, Pyke DF, Peterson R, (1998) Inventory Management and Production Planning and Scheduling. Wiley, New York. Third Edition. Simchi-Levi D, Kaminsky P, Simchi-Levi E, (2003) Designing and Managing the Supply Chain: Concepts, Strategies, and Case Studies. McGraw-Hill, New York. Second Edition. Song J-S, (1994) The effect of lead time uncertainty in a simple stochastic inventory model. Management Science 40:603–613. Stanton R, (2005) RFID – ripe for debate. Computer Fraud and Security, December: 12–14. Swedberg C, (2004) States move on RFID privacy issue. The RFID Journal, April 30:1–3. Tobin J, (1958) Liquidity preferences as behavior toward risk. Review of Economic Studies 25: 65–86. Tobin J, (1969) Comment on Borch and Feldstein. Review of Economic Studies 36:13–14. Walker WT, (2005) Emerging trends in supply chain architecture. International Journal of Production Research 43(16):3517–3528. Want R, (2004) RFID: A key to automating everything. Scientific American, January:56–65. Wayner P,(2002) Disappearing cryptography. Morgan Kaufman, New York. Wayner P, (2003) Translucent databases. Flyzone Sr LIc. Weinberg J, (2006) RFID, Privacy and Regulation. In RFID Applications, Security and Privacy, In Garfinkel S., Roseberg, B. Pearson Education, Inc., 83–97. Wolinsky H, (2003) P&G, Wal-Mart store did secret test of RFID, The Chicago Sun Times, November 10. Wong CY, Arlbjorn JS, Johansen J, (2005) Supply chain management practices in toy supply chains. Supply Chain Management 10(5):367–378. www.rafsec.com. www.spychips.com

5 Mobile Supply Chain Event Management Using AutoID and Sensor Technologies – A Simulation Approach Frank Teuteberg and Ingmar Ickerott

Abstract:

The goal of mobile supply chain event management (SCEM) is to monitor supply networks by observing specific events, disruptions, and exceptions in real-time, alerting decision-makers if problems have occurred and offering them good solutions. This chapter presents a framework for mobile SCEM based on software agents, auto-ID, sensor, and mobile computing technologies. Such technologies can be used together to proactively anticipate unexpected events, disruptions, and delays in the supply network before they lead to major problems and to provide useful alerts. Local human or artificial decision-handling agents can be proactively notified via mobile devices before delays arise or events fail to take place at all. A special focus is placed on a simulation approach. Herein, simulation is used to support agents by offering them good solutions for handling events. The authors of this chapter will show how to combine the above-mentioned technologies in mobile SCEM with simulation systems and open enterprise resource planning (ERP) systems to realize adaptive supply networks. Potentials and challenges in realizing adaptive supply networks by such technologies are also discussed.

5.1 Introduction and Motivation Enterprise resource planning (ERP) systems and advanced planning systems (APS) for supply chain management are primarily focused on long-term period planning and are not flexible enough to respond adequately to changes caused by disruptions in supply networks (Jung and Jeong 2005). Through mobile access to SCEM systems, decision-makers can act and react faster.

94

Frank Teuteberg and Ingmar Ickerott

This chapter describes a simulation approach for effective supply chain event management (SCEM) using agent and mobile technologies e.g., radio-frequency identification (RFID), sensors. A peer-to-peer-based multi agent system with access to back-office systems such as ERP and tracking and tracing systems (T & T) is outlined. The chapter is organized as follows: In the next section, basic concepts of mobile SCEM and enabling technologies are presented. Section 5.3 outlines the concepts of distributed simulation for decision support in SCEM. In Section 5.4, we present selected results of a literature survey based on a research database about mobile SCEM (which is available at http://www.mobilescm.de) to demonstrate the emergence of this research area and its current status. In the fifth section, a multilayer architecture of an agent-based mobile SCEM system is described. In Section 5.6, we introduce a simulation model of an adaptive supply network. In the seventh section, problems and issues for further research and development are outlined. In the final section, some conclusions are drawn. We also provide some recommendations for future research.

5.2 Mobile Supply Chain Event Management – Basic Concepts and Enabling Technologies 5.2.1 Basic Concepts The ARC Advisory Group (ARC, 2002) defines supply chain event management (SCEM) as part of supply chain process management to identify and monitor (unexpected) events in the supply chain, which is also called tracking. Mobile SCEM extends the concept of SCEM by using mobile computing technologies such as GPS or RFID for localization and automatic identification. Table 5.1 illustrates major tasks of mobile SCEM. The tasks are listed in linear order, but they are not necessarily sequential; they may also be performed in parallel. In the literature, a distinction is often made only between planned events (e.g., “shipment of goods”) and unplanned events (e.g., “supplier out of capacity”) which disrupt supply chain plans and their execution. However, there are many criteria to distinguish between different types of events in SCEM (Yufei and Detlor 2005).

Mobile Supply Chain Event Management

95

Table 5.1. Major tasks of mobile supply chain management Major Tasks

Key Roles

Enabling Key Technologies

1) Monitoring and reporting

Continuously monitor the supply network and quickly report to SCEM systems

2) Identification

Quickly identify the nature and scope of the event; support event classification

RFID, sensors, mobile computing (e.g., GPS), tracking and tracing systems, data warehouses

3) Notification

Rapidly notify the appropriate personnel

4) Organization

Establish teams with appropriate roles and responsibilities

SCEM systems in combination with mobile computing and agent technology

5) Simulation and Analyze action alternatives; explore planning multiple "what if" scenarios in advance

APS, simulation and SCEM systems in combination with mobile computing and agent technology

6) Execution and controlling

Coordinate information exchange between response personnel; provide onsite consulting; offer resource allocation and support; execute and control actions

SCEM systems in combination with mobile computing and agent technology

7) Assessment and investigation

Generate assessment reports; investigate major causes of disruptions; investigate effects and consequences on other supply network tiers

SCM-/SCEM systems

In Table 5.2, we classify events by several criteria in a morphological box. Morphological analysis was developed by Fritz Zwicky in the late 1960s (Zwicky 1969) for multidimensional, nonquantifiable, sociotechnical systems and problems. Not by reducing the number of criteria (variables) involved, but by eliminating the number of possible attributes (solutions), it is possible to reduce the complexity of such systems and problems in a so-called morphological box. For example, one may distinguish between unplanned and standard events (e.g., order change), unplanned and nonstandard events (e.g., a lost shipment), planned and standard events (e.g., an order shipment) and planned and nonstandard events (e.g., a transport strike). In 2000, for example, a fire damaged the plant of Nokia’s and Ericsson’s main supplier, which delivers mobile phone components (Lee 2004). For illustration, we classify this specific event "plant fire" by our morphological box for event classification (see Table 5.2).

96

Frank Teuteberg and Ingmar Ickerott

Table 5.2. Morphological box for event classification Criteria

Attributes

Category

Planned

Unplanned

Type

Standard

Nonstandard

Frequency of event

Minutely

Hourly

Daily

Weekly

Monthly

Yearly

Short

Medium

Long

Routine

Critical

Catastrophe

Risk level

Low

Medium

High

Probability of event occurrence

Low

Medium

High

Cost/event

Low

Medium

High

Time/event

Low

Medium

High

Resources/event

Low

Medium

High

Duration of event Impact level

Event producer Appropriate response personnel/experts

Unknown

Known

Internal

External

Event process level (SCOR)

Operations strategy (SCOR level 1)

Intra- and intercompany configuration (SCOR level 2)

Intra- and Intercompany process, Practice and system Configuration elements

Intra- and intercompany supply chain improvements (SCOR level 4)

(SCOR level 3) Event location SC planning influence

SC flow level

Near to suppliers Shortterm plans

Internal focus

MasAggregate ter plans plans

Information flow

Near to customers Logistics strategy plans

Goods flow

Dual focus (pooling of responsibilities) Business strategy plans

Corporate strategy plans

Cash flow

Mobile Supply Chain Event Management

97

5.2.2 Enabling Technologies Five technologies currently have a strong impact on mobile SCEM: Localization, auto-ID, sensor, agent technology, and Web Services/service-oriented Architectures (SOAs). Localization, auto-ID and sensor technologies help answer the central question: "Where is which object, and what is its current state?" Agent technology and Web Service help to integrate heterogeneous ERP systems or APS systems for enhanced supply network visibility. In the subsequent sections, we give a brief overview of these technologies. 5.2.2.1 Mobile Computing Technology for Localization Proximity is a commonly used localization technique, where cellular access points are monitored to determine the position of smart cards/mobile devices owners. Lateration and angulation of ultrasonic signals are other frequently used localization techniques, which rely on distance or angle measurements between some fixed points by means of signal strengths (Hightower and Borriello 2001). The Global Positioning System (GPS) is a widely used system that uses the lateration technique with one GPS receiver and four visible GPS satellites to determine the location of objects. However, GPS can only be used outdoors. As a fairly new technology, wireless real-time location systems (RTLS) enable clerks or managers to track (misplaced) assets, tools, and equipment, etc. also indoors. RTLS consist of locating devices (e.g., RFID tags), the infrastructure (e.g., antennas, RFID readers, and a wireless communication network) and application software as the "brain" of RTLS (Geier and Bell 2001). 5.2.2.2 Auto-ID Technology Auto-ID (automatic identification) technology such as a bar code, smart cards, biometric systems, or RFID (radio-frequency identification) is used to identify personnel, products, or delivery units. The most common auto-ID technology today is RFID. RFID tags are small computer chips that are attached to entities (e.g., products) that store the entity identifier and entity-related data. An antenna is connected to the RFID tag so that chips can be read out by RFID tag readers without the need for line of sight. RFID systems, therefore, are a good alternative to bar code systems, since they do not need human intervention or direct line of sight between the tag and the reader. The auto-ID Center established the electronic product code (EPC) network (auto-ID infrastructure) that usually consists of the components see illustrated in Figure 5.1 (Angeles 2005, pp. 52–56). The chip on an RFID tag contains the socalled EPC, serving as a unique identifier. The EPC is identified by means of RFID readers (step 1). The EPC is usually transmitted to a so-called savant server that filters and bundles data from the RFID tag readers (step 2) to query a so-called object naming service (ONS), where more information on the objects can be found (step 3). The ONS receives a uniform resource locator that references an EPC information service delivering product-related data (e.g., product name, supplier, etc.) of the observed objects (e.g., container, pallets, packages, articles) as physical

98

Frank Teuteberg and Ingmar Ickerott

markup language (PML) messages (step 4). PML (Flörkemeier et al. 2003) is a language used to describe objects in EPC networks. Existing implementations of the EPC network that are used in practice for product tracking are e.g., Sun’s EPC network architecture (SUN 2004) or the SAP auto-ID infrastructure (SAP 2005).

Figure 5.1. Auto-ID infrastructure

5.2.2.3 Mobile Sensor Technology Mobile sensor technology can detect changes in the environment, for example, thermal, acoustic, visual, infrared, magnetic, seismic, or radar sensors allow monitoring conditions such as temperature, humidity, vehicular movement, noise levels, the presence or absence of mechanical stress levels, or current characteristics such as the speed, direction, or weight of an object (Akyldiz et al. 2002). As illustrated in Table 5.3, a variety of sensors are available to detect a range of contexts such as an object's or user’s location (where?), the time (when?), or the identities of objects and users (what and who?). Object-identification sensors such as RFID detect the EPC or analyze the shape of objects by cameras. Time sensors provide the time whenever an event or disruption occurs. Time sensors (e.g., physical clocks) may represent the absolute time (e.g., time point, time interval, time zone) or the relative time that describes the relationship between timedependent events (e.g., before, after, at the same time).

Mobile Supply Chain Event Management

99

Table 5.3. Examples of sensors for event context detection Context

Output information

Sensor type

Who

User identity

Biosensors (e.g., retina, fingerprint scanner), camera

Where

Outdoor/indoor location coordinates

GPS, camera

What

Identity of object and attributes

Auto-ID technology (e.g., RFID), camera

How

Object's/user's (body) Conditions/physical status

E.g., weight-, temperature-, air pressure-, humidity-sensitive sensors, etc., or biosignal sensors (e.g., electrocardiogram, BVP (blood volume pressure), heart rate, GSR (galvanic skin response), respiration

When

Absolute/relative time

Clock

Why

Intention, user's gesture, facial expression, emotion, body conditions

Camera

Current research in sensor technology focuses on so-called wireless sensor networks (Akyldiz et al. 2002). Networking capabilities of sensors and technical advances in microelectromechanical systems allow the design of next generation sensors that are getting smaller, can be embedded in entities, consume less energy are no longer directly connected to a central controlling computer, and thus benefit from being autonomous. 5.2.2.4 Mobile Agent Technology The characteristics of agents (e.g., autonomy, proactivity, or adaptivity) lend themselves to model, simulate, and realize distributed supply networks. Agent technology has been successfully applied to a variety of supply network processes such as production planning, scheduling, and transportation (Fox et al. 2000). Agent platforms running on mobile devices have to be lightweight because today's mobile devices are constrained by small user interfaces, limited storage, limited battery life, and relatively low processing power. For this reason, the business logic of mobile agent platforms usually remains on a separate agent application server, whereas the mobile device is responsible only for presentation tasks. A very popular agent platform for mobile devices is JADe-LEAP (lightweight extensible agent platform) (Adorni et al. 2001). JADe-LEAP is a FIPA platform (foundation for intelligent physical agents; http://www.fipa.org) that can be used to enable software agents to spread across a heterogeneous network of mobile devices. JADe-LEAP agents run on personal digital assistants (PDAs) and Java-enabled mobile phones. Other examples of mobile agent platforms are the MobiAgent (Mahmoud 2001) platform and kSaci (Albuquerque et al. 2001).

100 Frank Teuteberg and Ingmar Ickerott

5.2.2.5 Service-oriented Architectures The integration of heterogeneous ERP systems is difficult because application programming interfaces (APIs) of ERP systems are usually proprietary. Web Services enable a platform-independent access mechanism to the ERP system's functionality (Jankowska and Kurbel 2005). Service-oriented architectures (SOAs) provide important functionalities such as discoverable and platform-independent Web Services to integrate heterogeneous ERP systems or APS systems for enhanced supply network visibility. SOAs comprise loosely coupled and highly interoperable software applications (services) that interoperate based on a formal definition language. The core components of an SOA (see Figure 5.2) are the following XML-based standards (W3C 2004): x x x

Simple Object Access Protocol (SOAP): SOAP defines a common syntax for data exchange between service applications. Web Service Description Language (WSDL): The public Web Service interface is described by WSDL, an XML-based service description of how to communicate with the Web Service. Universal Description, Discovery, and Integration (UDDI): The Web Service description is published using UDDI, which enables applications to look up Web Service descriptions to determine whether or not to use them.

Figure 5.2. Components of a service-oriented architecture

Mobile Supply Chain Event Management 101

5.3 Simulation for Decision Support in Supply Chain Event Management 5.3.1 Simulation Concepts and Approaches Simulation is a technique for using computers to imitate the operations of realworld facilities or processes that are represented in a model system (Law and Kelton 2000). The intention is to gain some understanding of the corresponding system’s behavior. Normally the system of interest is dynamic and stochastic. Simulation is an adequate approach for supporting business-related decisionmaking. Simulation has always played a vital role in supply chain management research. Going back to the early days of SCM research, Forrester analyzed the so-called bullwhip effect by using continuous simulation (Forrester 1961). Continuous simulation increments the simulation time in equidistant and very small steps. State variables change quasi-continuously with respect to time, and the rates of change are typically modeled in differential equations. A discrete simulation changes the state variables instantaneously at certain nonequidistant time. These times are those at which an event occurs. Table 5.4 shows the two main simulation strategies, their key concepts, and some important subclasses of the discrete-event simulation. Simulation is a versatile but often time-consuming tool. Using it for real-time decision support seems to be unrealistic at first glance. The success of simulation for this purpose depends crucially on a significant increase in modeling speed and the intelligent (re)us of business data from operating systems by applying appropriate interfaces to the simulation system. For this reason, traditional simulation methodologies have been expanded in different ways. Following the process-oriented and object-oriented view, a modeler chooses to define the changes in the state of a model by describing the life cycles of each individual component. These components are called active objects. In contrast, passive objects are simple data containers. They are passive because they are unable to affect other objects. Apart from the methodological object classes, simulation scenarios distinguish between physical and informational object classes. By applying object-oriented technology, the simulation technique became more flexible, and reusing source code and architecture has become much easier. It is natural to use the information stored conveniently in a database to accommodate the informational need of building and running simulation models (Witte 1990). The so-called data-driven simulation uses data from relational databases that are typically the basis for integrated operating systems such as the well-known ERP systems from SAP or Oracle. Most of the data necessary for a simulation experiment are available from databases of these companies and should be used as directly as possible.

102 Frank Teuteberg and Ingmar Ickerott

Table 5.4. Simulation concepts Time-advance mechanism

Simulation

Continuous

Continuous time advance

Discrete

Next-event time advance (event-oriented)

fixed-increment time advance (time-oriented)

Key concepts and components

Simulation subclass

Special concepts and specifics

System state variables, simulation clock, coupled differential equations with rates of change for state variables Process-oriented System state variables, simulation clock, Object-oriented events, event list, event routines, Agent-oriented timing routine

Processes, process state Passive objects, active objects, life-cycles Agents, beliefs, desires, intentions Quasi-continuous Fixed time increment

In recent years, computer technology has developed parallel and distributed computing environments in which individual computers can be linked together. This has also become part of the simulation research, taking the influence on the simulation tools into the market. There are many conceivable ways of splitting up a dynamic simulation to distribute its execution over different processors. One possibility is to allocate relatively autonomous parts of the model to different processors. Another way is to do this with the support functions for randomnumber generation or event-list handling. The logical execution of the discreteevent simulation will remain sequential, whether or not its computational structure is distributed. Communication, time-ordering, and synchronization become important issues. Driven by researchers of military applications, the IEEE1516 High Level Architecture (HLA) was developed. The HLA is a standard for distributed simulation and is controlled by the Simulation Interoperability Standards Organization (Taylor 2004). In our approach, the term distribution refers to two things. On the one hand, we integrate distributed master, stock, and dynamic data in the form of real business events from different firms within a cooperative supply chain. On the other, we distribute the system architecture itself by using agentbased modeling. Besides speeding up model building and the execution time of simulation experiments, there is the question of how to generate alternative solutions and to select the best ones within an appropriate time. Sophisticated and well-proven approaches come from the field of real-time planning and scheduling on a shop

Mobile Supply Chain Event Management 103

floor where accomplishing a good plan is not easy due to the concurrent flow of various parts and the sharing of different types of resources. These methods are also needed for the application of simulation in the domain of SCEM. Multipass shop scheduling, for example, is a well known method for solving the aforementioned problem (Yoo et al. 2004). 5.3.2 Simulation Software for Production, Logistics, and Supply Chain Management Since the late 1970’s, a large number of tools have been developed for the simulation of production and logistics systems. During the 1990’s, these simulation tools became an integral part of planning and realizing supply chain processes. Additionally, simulation has established itself as an assisting instrument for the ongoing management of operations. Figure 5.3 shows a taxonomy of simulation systems for supply chain management.

Figure 5.3. Taxonomy of simulation systems for production, logistics and supply chain management (see Wenzel and Noche 2000)

The tools differ in their grade of specificity. The range goes from generic programming languages to systems that are specialized for a small application domain. There is still a lot of development going on, focusing mainly on taskspecific simulators, open configurable simulation, and simulation development environments. New features for cost accounting or interfaces to CAD systems and databases are currently being added to existing systems. Taylor describes a useful common

104 Frank Teuteberg and Ingmar Ickerott

theme for simulation systems called commercial-of-the-shelf simulation packages (COTS) (Taylor 2004). COTS are used mostly by simulation modelers for model building, experimentation, animation, visualization, and reporting. COTS are typically based on some variant of the discrete event simulation paradigm. Models represent entities or objects that pass through networks of queues and workstations. A package has a couple of basic model elements, i.e., workstations, queues, sources, sinks, or resources, etc., that are used to build a model via drag-and-drop functionality. Model elements and element attributes, respectively, can be modeled by menu systems or by a programming language. There is no common internationally recognized naming convention for model elements and attributes. 5.3.3 Agent-based Simulation Tools Agent-based simulation applies the above-mentioned agent paradigm and seems to promote a natural form of modeling, as active entities in the original system are also interpreted as autonomous actors in the model. Developing, designing, and implementing agent-based simulation systems is not trivial. Modelers have to deal with high complexity due to an increased number of parameters and multilevel behavior of an agent-based system. The agent-based approach does not solve all problems faced by the established approaches, but it has a few advantages in domains that are affected by cooperating and autonomous decision-makers, as in supply networks. There are only a few tools available for agent-based simulations. Table 5.5 characterizes software tools that can be used to build agent-based models. Some tools are exclusively designed for simulation, whereas others are agent-based software systems that can be applied for building simulation systems. The review is limited and does not aim to be complete.

5.4 Related Research In this section we give an overview of research directions and existing agent-based systems for supply chain (event) management. This literature review is based on a research database developed at the University of Osnabrueck. The purpose of this database is to document today's state of the art of mobile SCM research by analyzing scientific journal articles. For example, current articles were analyzed with regard to the research areas, research designs and research methods (e.g., simulation, explorative/empirical research studies, case studies, prototype development, etc.) used to investigate mobile SCEM research topics. The research database (available at http://www.mobilescm.de) presently contains more than 100 articles on mobile SCEM published since the year 2000.

Mobile Supply Chain Event Management 105

Table 5.5. Tools for agent-based simulation Tools Authors and publication

JACK Howden et al. (2004)

Jadex Pokahr et al. (2003)

XRaptor Brund et al. (2003)

SeSAm

Daniels (1999)

Klügl (2001)

www.swarm. org

www.simses am .de/

www.agentsoftware.com

MAS development Network management, logistics

MAS development

MAS development

Simulation Simulation Simulation

General purpose

General purpose

Testbed for multiagent models

General purpose, artificial life

Biological systems, artificial life

Commercial or research use

Mainly research

Commercial

Research

Research

Research

Research

Language

Visual

Jack (Java)

Java

C++

ObjectiveC or Java

Lisp

Main objective Main model domains

www.inform atik. unimainz.de/ ~polani/XRa ptor/

Swarm

vsis-www. informatik. unihamburg.de/ projects/jade x/

labs.bt.com/ projects/agen ts/ zeus/

URL

Main system characteristics

ZEUS Nwana et al. (1999)

No No No Fixed time EventFixed time Time model simulation simulation simulation increments based increments system system system Process Yes Yes No No No Yes model SelfSelfSelfSelfSelfFramework developed developed JADE developed developed developed framework framework framework framework framework Beliefs, Beliefs, Java code Actions, Tasks and Reusable desires, desires, C++ code and activity modules goals intentions intentions MAML graphs BehaviorGoal- and based / No taskBDI Agent Reactive BDI specific or other architecture reflex oriented agent agents TCP/IP FIPA Interface based Agent ACL, available message FIPA commuKQML for the passing / No FIPA No FIPA conform nication and CORBA, FIPA predefined support support JADE coorditeam conform coplatform nation modeling JADE ordination framework platform strategies Interface DevelopDifferent System Different and ment Visual No visual visual visual access and tools modeling analysis environtools user support tools tools ment GUI ExtensibiliYes Yes Yes Yes Yes Yes ty

106 Frank Teuteberg and Ingmar Ickerott

Recent research relevant to our approach is concerned mainly with the following three research directions: x

Agent-based simulation in supply network environments: - Fung and Chen (2005) proposed a multiagent-based system for supply chain integration and simulated an illustrative example of supply chain planning. - Kleijnen (2005) discusses different types of simulation for supply chain management. - Reddy and Rajendran (2004) simulated a supply chain to evaluate the proposed dynamic order policies based on various heuristic settings.

x

Integrating agent technology with Web Services in supply networks: -

x

Wang et al. (2004) apply agent technology to deal with complex and distributed processes in exception management; Web Services are proposed for more interoperability in supply networks.

Agent-based architectures and prototypes for SCM: - Huhns and Stephens (2001) apply agent technology to promote the interchange of standard business documents and manage exceptions that might occur during supply chain execution. - Jung and Jeong (2005) proposed a decentralized productiondistribution planning system using collaborative agents for SCM. - Yung et al. (2000) developed a multiagent-based system (MAS) for SCM to support inventory management. Agents are used to monitor the inventory levels and to support managers’ decisionmaking.

The publications reviewed above illustrate the advantages of multiagent technology for decentralized collaboration in supply networks. However, they are deficient in providing a practical multiagent system reflecting the difficulties in negotiation and global information sharing in supply networks. The publications reviewed have either focused on cooperative decision support using multiagent technology or simulation. Our research simultaneously addresses both cooperative decision support and distributed simulation. Existing agent-based prototypes that address SCEM tasks are presented in Table 5.6 (Teuteberg and Schreber 2005). The MAS ECTL-Monitor and PAMAS are two research prototypes that address the basic concepts of tracking and tracing in supply networks. In contrast to these prototypes, DIALOG is an advanced tracking system that supports several sensor and auto-ID technologies such as RFID. It is the only agent-based prototype in SCEM that enables user access to the system via mobile devices. It is released under the Lesser General Public License (LGPL). A further but extended research prototype exists with PROVE, which solves problems in supply networks through agent-based negotiation. CoagenS is a complex MAS addressing several issues in SCEM such as disposition, monitoring,

Mobile Supply Chain Event Management 107

and resource management, applying heuristics and agent-based negotiation for problem solving (Dangelmeier et al. 2004). In the joint project, Agent.Enterprise (Frey et al. 2003), different supply chain management tasks such as production planning, distribution or tracking and tracing are addressed by several MAS (e.g., IntaPS, KRASH and FABMAS for production planning, ATT/SCC for T and T, and DISPOWEB for supply chain planning) that are integrated into one architecture. The ATT/SCC project seems to be an interesting approach in SCEM, but no prototype has been presented yet. An interesting work has also been presented by Krishnamurthy and Zeid (2004). They describe an agent-based architecture for information access to existing ERP or legacy systems by mobile devices. The architecture is based on Java technology. XML is used for interagent communication. However, they only consider the intraenterprise level, not the whole supply chain or network. The tendency to provide mobile access to enterprise information systems for the field staff is still supported by vendors of ERP systems such as SAP and Oracle. In practice, many solutions support only the intraenterprise level. The problem of integrating the ERP, APS and SCM systems of all members into a supply network is still to be realized by open standards such as XML, BPEL, or Web Services. A promising approach is to combine existing components, such as SAP’s Netweaver Platform, with SAP's Event Manager, SAP's auto-ID Infrastructure, and SAP's Mobile Infrastructure as an option for mobile SCEM from one source (Blanchard 2004). Nevertheless, this can be a disadvantage if ERP systems from other vendors still exist in a supply network or if the solution is not affordable, particularly for small and medium enterprises. Further solutions tend to focus on the intraenterprise level, e.g., NXC with its Xagent Business Activity Monitoring Solution (NXC 2004). Current approaches in both research and practice focus on tracking, tracing, and notification capabilities. However, more research is required to enhance supply network transparency and to solve identified disruptions in supply networks automatically, or at least semiautomatically, through the application of software agents and mobile computing technologies. Furthermore, it is desirable to anticipate disruptions in advance before they occur. Our approach described in the following sections covers this case.

5.5 Mobile SCEM Prototype – System Architecture and Functionality In this section, we introduce an agent technology and Web Service-based prototype currently under development for mobile SCEM in supply networks and illustrate its flexible architecture. Auto-ID, mobile, agent, and simulation technologies are combined within this prototype to realize permanent tracking and tracing of resources (e.g., products, vehicles) in supply networks, to visualize resourcerelated key performance measures, and to explore multiple "what if"-scenarios in advance without the cost and risk involved in making real-world changes.

108 Frank Teuteberg and Ingmar Ickerott

Table 5.6. Existing prototypes of MAS in supply chain event management Multiagent systems

ECTLMonitor

SCEM

Authors and Year Hofmann et al. (1999) of Publication

MAS architecture

DIALOG

PROVE

CoagenS

Zimmerman and Paschke (2003)

Kärkkäinen et al. (2003)

Szirbik et al. (2000)

Dangelmeier et al. (2004)

Area of scope

Tracking by Tracking and Tracking and means of tracing tracing RFID

Monitoring of products in VEs

Monitoring the whole supply chain

Architecture

Peer-to-peer

Peer-to-peer

Peer-to-peer

Not fixed

Peer-to-peer

Use of RFID

No User, retrieval, and service provider agents

Intended Coordination, surveillance, and wrapper agents

Yes

No Mobile monitoring agents, mediator agents

No

Agent coordination

Cooperative

Cooperative

Cooperative

Cooperative

Cooperative

Agent communication

Selfdeveloped language

FIPA-ACL

XML-based for future versions

XML-based language

XML-based language

Agent negotiation

Not supported

Not supported

Not supported

Bilateral

Multilateral

Framework

Selfdeveloped framework

FIPA-OS

Selfdeveloped framework

Selfdeveloped framework

In conformity with FIPA

Language

Java

Java

Java

Java

Java

Ontology

No explicit ontology

No explicit ontology

No explicit ontology

XML-based ontology

XML-based ontology

Planning algorithms

Not supported

Not supported

Not supported

Agent mediated negotiation

System access

Web interface

Java GUI

Agent roles

Agents' features

PAMAS

Agents are assigned to the tagged product

Work agents

Extensibility

No

Yes

Mobile devices, Web interface Yes

Yes

Based on rules, agent mediated negotiation Web interface, Lotus Notes GUI Yes

Adaptivity

No

No

No

Intended

Yes

Autonomy

Yes

Yes

Yes

Yes

Yes

Reactivity

Yes

Yes

Yes

Yes

Yes

Proactivity

No

Yes

Yes

Yes

Yes

Java GUI

Mobility

No

No

No

Yes

No

Social ability

Yes

Yes

Yes

Yes

Yes

Mobile Supply Chain Event Management 109

5.5.1 Generic Requirements Our architecture addresses a number of different aspects. For better understanding, we differentiate these into the categories of Mobile SCEM, technology, user, and simulation-related requirements: x

Mobile SCEM-related functional requirements: -

-

-

-

x

Event notification: Through tracking with RFID and mobile technology, resources (e.g., commodities, products, and vehicles) should be monitored along the whole supply chain. Exception management: When disruptions occur, agents should negotiate with each other to find a new solution to fulfill the disrupted process or at least provide advice to human users in the form of decision support. Organization: Roles and responsibilities within the supply network should be represented in the system architecture. Simulation and planning: Action alternatives should be explored by multiple “what if”-scenarios and suitable planning algorithms. Execution and controlling: Information about decisions and their consequences should be spread throughout the entire supply network if necessary. Assessment and investigation: The analysis of historical data of disruptions, irregularities, and other problems should be realized so that future planning processes can be more effective. A powerful tracing solution is, therefore, required.

Technology-related functional requirements: -

-

-

-

Peer-to-peer structure: According to the structure of supply networks, it seems feasible to implement a decentralized, peer-topeer-based system. This preserves the necessary independence of the enterprises within the supply network and provides a flexible mechanism for adapting to changes in the supply network structure (e.g., adding new network partners). Real-time event handling: The ability to process events in realtime and provide real-time status information to users is required. Supply chain visibility: All partners of the supply network should access actual data such as available stock, delivery status, actual lead times, and costs via Web interfaces and mobile devices. Openness: The system should follow standards in the field of Web, agent, and peer-to-peer technology. It should use XML for data handling and information interchange. Extensibility: The architecture should be extendable to fulfill new tasks. New agent platforms should be added when new members become partners of a supply network.

110 Frank Teuteberg and Ingmar Ickerott

x

User-related functional requirements: -

x

Personalization: Users should initialize agents to provide only the information that is necessary. Anticipating: Agents should be capable of anticipating users’ decision-handling activities to improve the generation of action alternatives by adopting learning algorithms.

Simulation-related functional requirements: -

-

-

-

Discrete event-based modeling: Due to the event orientation of mobile SCEM, it is straightforward to use a state-of-the-art discrete event-based model for the simulation of supply chain processes. Real business events should kick off simulation runs that are themselves driven by simulation events. Process-based: All activities within a simulation are related to events and should be modeled and encapsulated as simulation processes. The transformation of processes from one state (e.g., delayed) to another (e.g., executable) should be governed by simulation events. Database interface: Simulations within a mobile SCEM architecture handle real-time and mass data and, therefore, need a comprehensive data storage solution. Databases should keep all passive objects including historical data about previous simulation runs. We use several interfaces to OLTP and OLAP databases to realize the above-mentioned data-driven simulation approach. Analysis support: Being able to draw the right consequences from a simulation experiment depends heavily on the visualization and reporting capabilities of the simulation tool in use. For this reason, the simulation tool should have an adequate analysis module.

5.5.2 Simulation Definition and Event Definition Markup Languages

5.5.2.1 Simulation Definition Markup Language (SDML) The XML-based simulation definition markup lanaguage (SRML) was introduced for building, distributing, and running simulations using Web-based technologies. SRML was developed at Boeing Corp. as an enhanced eXtensible markup language (XML) that can specify the behavior of distributed simulation models (Reichenthal 2002). The goal of SRML is to enable distributed simulations described in a standard way using Web-based technologies. Although SRML looks very promising for describing, distributing, and running simulations using Web-based technologies, it also has its limitations. For instance, SRML is not suitable for computationally intensive simulation models, complex scenarios or for describing specific events in our application domain. Therefore, we developed the XML-based language SDML (simulation definition markup language) to define a flexible XML-based standard for

Mobile Supply Chain Event Management 111

representing simulations that enables organizations to speak a common process language, regardless of the software application (service) used. SDML can be considered a metalanguage with the ability to model and simulate business processes (events) in supply networks of varying complexity. Important simulation data considered in SDML are x x x x x

Simulation model metainformation e.g., simulation model, description, negotiation context/type (e.g., proposal, accept), etc., Simulation model structure (e.g., related objects and their relationship), Simulation model decision parameters and their data types, Simulation model input values, Simulation model constraints.

Figure 5.4 is an example of SDML.

… … … … …

Figure 5.4. Simulation definition markup language

5.5.2.2 Event Definition Markup Language (EDML) The event definition markup language (EDML) was developed as a general language for describing physical objects and events in supply networks that are capable of being observed or measured by remote sensors. EDML is heavily based on PML (Flörkemeier et al. 2003). Figure 5.5 is an example of an EDML file.

112 Frank Teuteberg and Ingmar Ickerott

E123C1 urn:epc:1:4.16.56 Warehouse 4 Osnabrueck 2006-02-10T14:23:55 urn:epc:1:3.24.500 10.5

Figure 5.5. Event definition markup language

5.5.3 System Architecture In this section, we propose the peer-to-peer agent-based architecture CoS.MA (cooperative and ubiquitous supply network monitoring agents) to use existing potentials and technologies in supporting business processes in SCEM. CoS.MA’s mobile agents are not limited to information gathering from legacy or ERP systems for the field staff. Moreover, agent-based decision support for mobile workers and autonomous agent-based negotiation between all members of a supply network will be possible to realize more effective processes. The challenge here is to integrate the distributed information of all members of a supply network so that transparency and access to heterogeneous ERP systems is achieved. Our architecture (see Figure 5.6) consists of four layers: x x x x

The presentation layer: Providing the user interface responsible for the interaction between the user and the systems via mobile devices (e.g., PDAs), The presentation logic layer: Containing presentation rules, user profiles, The service layer: Offering general services for the other layers (e.g., database services or communication services). The data layer: Represented by several databases (e.g., ERP, simulation, and/or T and T databases).

The core component that is installed as an extension to existing enterprise information systems by every partner in the supply network is a multiagent system for mobile SCEM, which is implemented using the Java agent development framework (JADE). JADE is an open source system that can easily be extended. Several extensions are available such as Jadex (Braubach et al. 2004) to facilitate practical reasoning agents, JADe-LEAP (Berger et al. 2003) to enable agents on

Mobile Supply Chain Event Management 113

mobile devices, or the Web Service Integration Gateway (Greenwood et al. 2005) to enable the invocation of Web services by agents. JADe-LEAP is used on mobile devices to allow a direct connection between agents on the user’s mobile device and the stationary agents. The JADe-LEAP agent, therefore, acts as an interface agent to the stationary MAS. As previously mentioned, the wrapper agents interact with the underlying systems by Web Services using the Web Service Integration Gateway. With the help of Web services it is possible to increase the compatibility between heterogeneous and distributed ERP and tracking and tracing systems. For each monitored resource (e.g., a product, an order, a vehicle, etc.), a monitor agent is triggered via a user agent. The monitoring agents search for desired status information in external applications (e.g., in ERP systems), databases, or tracking and tracing systems (layer 3) by wrapper agents that collect the corresponding status information (e.g., using Web Services). Wrapper agents make information and functionality of the underlying systems available to other agents, in that they transform the data into a standardized format that can be easily extracted by the other agents. The monitoring agent is a control unit that controls the intracompany data and exchanges messages with other monitoring agents from the remaining partners in the supply network to achieve visibility through the integration of distributed information from all monitoring agents.

Figure 5.6. Multiagent system architecture CoS.MA

114 Frank Teuteberg and Ingmar Ickerott

Resource agents are mobile agents that can negotiate with the resource agents on other CoS.MA agent platforms about prices, quantities, availability of resources, etc. The resource agent is designed as a mobile agent because of the complex negotiations that can arise in the problem-solving process. As illustrated in Figure 5.6, we use the open-source ERP system Compiere (http://www.compiere.org) in our prototype. A Web Services Façade was implemented to gain access to selected Compiere functionalities. This Web Services Façade is available at http://cvs.sourceforge.net/viewcvs.py/scmws/. In our Mobile SCEM architecture, the decisions on how to handle internal and external supply chain events are made by agents who themselves use a simulation tool called SCM.SIM for the generation of good actions and action alternatives (see Figure 5.7). From a technical point of view, the decision support functionality is encapsulated as a Web Service. A Web Service façade provides an open interface for internal and external events. Internal events come from a company’s on-line transaction processing (OLTP), on-line analytical processing (OLAP), or other systems (e.g., proprietary tracking tools). These systems implement an appropriate interface for event descriptions: The so-called EDML files (see 5.5.2.2). External events come from the firm's agent platform. The company’s agents can call the simulation by sending an EDML file to the Web Service façade of the simulation module. The event is then saved to a simulation database where all relevant data for the decision problem are stored.

Figure 5.7. Integration of the simulation system

Afterward, a simulation run is started to figure out the consequences of this event with the given decision parameter set. This initial parameter set reflects the current distribution, production, and transportation plans. An optimization tool

Mobile Supply Chain Event Management 115

changes the parameters with respect to some problem search strategy and starts a new simulation run. When a certain goal has been met, the best parameter set is taken as the basis for the new logistics plans. The results can be visualized and analyzed with an analysis tool before being sent to the other agents within the supply chain, together with a description of the simulation model and the simulation run in an SDML file (see 5.5.2.1). In most cases, the solution will have to be checked with the constraints of the supply chain partners. The endogenous decision parameters of one agent become exogenous parameters for the simulation runs of the other agents. Figure 5.8 shows an example of such a collaborative planning process.

Figure 5.8. Sequence diagram of a collaborative planning process

A disruption occurs in the supplier’s part of the SCM process. For instance, a sensor detects an event and informs an agent on his own agent platform. This agent invokes the simulation module by sending an event description file to the Web Service façade of the module. Alternatively, the EDML file comes directly from the supplier’s ERP system. In the example scenario, the disruption has no direct impact on the plans of the supplier, so that the simulation does not give back a new parameter set but only the initial disruption event. The event description file is then forwarded to the manufacturer’s agent because the disruption could cause a need for anticipation in the manufacturer’s part of the supply chain process. The receiving agent calls his own simulation module for decision support. A policy change is required so that a plan proposal is given back to the supplier, who in turn has to check the proposal with a new simulation run. All relevant information for

116 Frank Teuteberg and Ingmar Ickerott

the simulation run is described in an SDML file. This file covers the parameters of a new common logistics plan. If the supply chain partner agrees, after calling his simulation module with the SDML file as input, the proposed plan becomes the new active plan of the company. In the given example scenario, the parameter sets are then copied to the ERP systems of both the supplier and the manufacturer. Should the partner disagree, a new proposal would have to be made with the second or next best solution parameter set. It is expected that the overall performance of a supply network would be better in comparison to centralized control. The global goal of the whole system is to aggregate the local objectives by negotiation. Each local decision-maker starts negotiating with its best solution. However, there is no guarantee that the agents will converge to an “equilibrium” solution at the end of their negotiations. If the agents do not find a collaborative solution, the human decision-makers who are in charge of the supply network will be asked to continue the coordination process.

5.5.4 Simulation Tool AnyLogic™ Within our simulation system SCM.SIM, we use the simulation tool AnyLogic™, developed by XJ Technologies (http://www.xjtek.com/anylogic), to model, simulate, and visualize different types of logistics processes. AnyLogic™ is a general-purpose simulation environment based on the programming language Java for the development of discrete, continuous, and hybrid simulation systems. Simulation models developed by using AnyLogic™ can interoperate with any office software (e.g., MS Excel) or with custom software applications because they are written in Java™. Via the Java native interface (JNI), AnyLogic™ can also interoperate with software written in other languages. Furthermore, AnyLogic™'s open architecture can dynamically read and write data to databases or ERP systems or can be embedded in a real-time operational environment. Simulation models can be called from external programs, and vice versa, using the open application programming interface (API) of AnyLogic™; they also run on any Java™-enabled platform (such as Windows, Solaris, Linux, MacOS, etc.) and can even be placed on a Web site as applets, because they are written in 100% Java™. Furthermore, one can write one's own code to extend the features of AnyLogic™.

5.6 Simulation Scenario As a proof-of-concept prototype, an implementation of an agent-based system for mobile SCEM is ongoing, extending the ideas of the CoS.MA architecture. Several of these prototypes will be set up in a network to simulate a supply network, as shown in Figure 5.9. The intention is to simulate typical supply network problems such as negotiation with other partners in the case of a delay in delivery. The agents will then automatically solve these problems with the help of integrated and simulation-based planning modules, as shown in Section 5.5.3.

Mobile Supply Chain Event Management 117

Figure 5.9. Simulation scenario

In our current simulation scenario, we consider a three-tier supply network, consisting of a focal manufacturer who assembles module parts that are delivered by four suppliers and three retailers who sell the goods to the market (see Figure 5.9). The consumer demand differs from retailer to retailer. The manufactured product consists of three different parts. The third part is delivered by supplier three and also by supplier four (see Figure 5.9). Table 5.7 shows the most important parameters of our initial simulation model. Table 5.7. Model parameters

Exogenous (input) parameters

Retailer

Manufacturer

Supplier

Consumer demand

Order plan

Assembly plan

Transport time

Assembly lead time

Production lead time

Transport time Endogenous (decision) parameters

Order plan

Assembly plan

Production plan

Objective function (output parameters)

Profit

Profit

Profit

The endogenous parameters of one stage (tier) are exogenous through the eyes of the next stage. All remaining exogenous parameters are stochastic. There are probability distributions for the consumer demand of each retailer, the assembly and production lead times, and the transportation times for the material movements from and to the manufacturer. One of the most common distributions in economics and business, which is also very typical for processes in nature and technology, is the Gaussian distribution. A simplification of this model is the triangular

118 Frank Teuteberg and Ingmar Ickerott

distribution. For example, a demand can be modeled with the parameters, minimum demand, average demand, maximum demand, and standard deviation. We use these two probability functions to model the above-mentioned input parameters as well as the occurrence of disruptions. The agents of our scenario try to maximize their profit function. Since there is no impact on selling prices in our basic model, we can focus primarily on the cost function. There are penalty costs for lost and late delivery, stock keeping costs, and transportation costs. In our first simulation scenario, we consider a time horizon of 14 days, beginning at 00.00 on the first day and ending at 24.00 on the fourteenth day. We use days as the elementary time steps in the initial simulation runs, although we refined the model later. Table 5.8 illustrates a simple simulation run with a constant consumer demand. It is complex enough to illustrate two examples of disruptions that could happen in our simulation model. The first is a transportation delay of a material movement from supplier 2 to the manufacturer on the first Thursday. This event is not critical to the manufacturer because his stock of inventory is high enough to compensate for lost shipments. The second example is a 1 day disruption of the assembly line on the second Monday (indicated in bold in Table 5.8). At the end of week two, the situation has recovered because the retailers were able to buffer demand. In both examples, the affected agents do not try to compensate for lost shipments by ordering more than in the stable situation. The so-called bullwhip effect does not emerge. This very simple example scenario shows two major keys to success in modern supply networks. All experiments we have run so far show these results independently of additional complexity. The first conclusion is that the agents within a supply network give an account of disruptive events as quickly as possible. In our example scenario, the affected agents received information about the event and the estimated time of tardiness. Armed with this information, the agents were able to measure the impact of the event and find a solution that balances the material flow without causing a bullwhip effect. The second conclusion is that buffers play an essential role in balancing the supply network. Smart sizing of buffers became key questions in our research. The role of buffers can be adopted by changing suppliers. But this is obviously not feasible in every case. We examined this option in our model scenario by allowing the manufacturer to countervail lost shipments of supplier 3 with an additional order with supplier 4.

Mobile Supply Chain Event Management 119

Sat

Mon

Tue

Wed

Thu

Fri

Sat

5

5

5

5

5

5

Order plan

5

5

5

5

5

5

5

5

5

Buffer

5

5

5

5

5

5

0

3

5

Transport time

2

2

2

2

2

2

2

2

2

Consumer demand

6

6

6

6

6

6

6

6

6

6

Order plan

6

6

6

6

6

6

6

6

6

6

Buffer

6

6

6

6

6

6

0

4

6

6

Transport time

1

1

1

1

1

1

1

1

1

1

Consumer demand

4

4

4

4

4

4

4

4

4

4

Order plan

4

4

4

4

4

4

4

4

4

4

Buffer

4

4

4

4

4

4

0

3

4

4

Transport time

1

1

1

1

1

1

1

1

1

1

Demand

15 15 15 15 15

15 15 15 15 15 15

Delivery

15 15 15 15 15

15

0

25 20 15 15

Assembly lead time

1

1

2

1

1

Assembly plan

15 15 15 15 15

0

25 20 15 15 15

1

1

1

Buffer part 1

15 15 15 15 15 15

30 15 15 15 15 15

Buffer part 2

15 15 15 15

15

30 15 15 15 15 15

Buffer part 3

15 15 15 15 15 15

30 15 15 15 15 15

Supplier 1 (Part 1)

1

Production plan

15 15 15 15 15 15

15 15 15 15 15 15

Production lead time

1

1

1

1

1

1

1

1

1

1

1

1

Transport time

1

1

1

1

1

1

1

1

1

1

1

1

Supplier 2 (Part 2)

1

Production plan

15 15 15 15 15 15

15 15 15 15 15 15

Production lead time

1

1

1

1

1

1

1

1

1

1

1

1

Transport time

1

1

1

2

1

1

1

1

1

1

1

1

Supplier 3 (Part 3)

1

Production plan

10 10 10 10 10 10

10 10 10 10 10 10

Production lead time

1

1

1

1

1

1

1

1

1

1

1

1

Transport time

1

1

1

1

1

1

1

1

1

1

1

1

Production plan

5

5

5

5

5

5

5

5

5

5

5

5

Production lead time

1

1

1

1

1

1

1

1

1

1

1

1

Transport time

1

1

1

1

1

1

1

1

1

1

1

1

0

Sun

Fri

5

Sun

Thu

5

Wed

5

Tue

Consumer demand

Supplier 4 (Part 3)

Manufacturer

Retailer 3

Retailer 2

Retailer 1

Mon

Table 5.8. Demand and supply plans for a simplified example scenario

120 Frank Teuteberg and Ingmar Ickerott

5.7 Problems and Questions for Further Research In this section, we discuss a number of problems and topics for further research in mobile SCEM. Challenges still to be solved are x

x x

x

x

x

x

x

Security: Security issues concerning mobile agents and mobile devices still remain unsolved (Luck et al. 2003, p. 13). Today’s auto-ID systems, for example, are vulnerable to hijack attacks, which have to be addressed when developing more secure communication protocols. Trust and privacy: Beyond security, trust issues concerning autonomous mobile agents (Luck et al. 2003, p. 51), and privacy issues concerning autoID have to be solved (Twist 2005, p. 231). Available resources: Krishnamurthy and Zeid (2004, p. 177) note that mobile agents require significant resources for execution. Smart phones and PDAs have limited resources, so resource-intensive tasks should be executed by stationary agents, which are not located on mobile devices. Information overflow: The mass of products, machine data, and other resources that have to be scanned and transmitted in an EPC network have to be managed in time-critical processes due to constraints in available bandwidth and computing power (Angeles 2005, p. 55). Lack of standards: There is still a diverse multiplicity of competitive standards (e.g., diverse radio-frequency identification standards, codes for product identification, data exchange formats, and mobile computing standards, etc.) worldwide which differ substantially in their suitability for different industries, as well as with regard to their diffusion. These standards are still being improved and constantly expanded. Lack of cooperation: Although our proposed agent-based architecture provides a secure environment for fostering mobile information sharing in supply networks, peers still can refuse to share their information with their partners. Lack of honesty and hiding of information: Supply chain partners could hide important information from each other to optimize individual utility, or they might be dishonest. Or else they might believe that they can solve problems before they start affecting other supply network members and might thus wait instead of alarming other members immediately. Thus, trust is an important prerequisite in our approach. In contrast to centralized planning approaches, as realized, for example, in the current advanced planning systems, there is no need for a single integrated database and common objectives. Ontologies: Furthermore, the design and implementation of ontologies to realize semantic Web Services is required to establish the integration and exchange of machine-readable data between agents and heterogeneous information systems in a standardized way. OWL-S (Web ontology language for services) and WSMO (Web Service modeling ontology), for example, are emergent technologies to realize such semantic Web Services (Cabral 2004).

Mobile Supply Chain Event Management 121

x

Scalability: Another example for future work is to demonstrate the scalability of our distributed agent architecture beyond three tiers, the small number of supply chain partners, and a limited set of products, as described in Section 1.6, even though the proposed peer-to-peer-based architecture is designed to be inherently scalable.

5.8 Conclusions Today, automatic observation and management of disruptions and other irregularities in supply networks is limited mostly to single supply chain members. The peer-to-peer-based architecture CoS.MA presented in this chapter is intended to integrate data from single members so that all members can visualize pertinent data. Mobile agents investigate unexpected events on their behalf and are capable of giving decision support. Furthermore, with support from auto-ID and mobile technologies, permanent tracking and tracing of objects and events in the whole supply network will be possible. Critical business information and functionality can be made accessible to mobile workers no matter where and when disruptions occur. In the next steps, mobile agents capable of migrating to other agent platforms on mobile devices will be developed. We are currently simulating simplified supply network problems to let the multiagent system automatically handle such problems. Step by step we will make our simulation models more complex to be more realistic. In the future, it is intended to simulate business processes in supply networks based on real-world data from business partners.

5.9 Guidelines for Practitioners Practitioners (planners and schedulers) have major problems in maintaining an overall perspective on each planning object in supply networks. Furthermore, tracking and tracing objects is generally limited to a single supply network member. However, the flow of information is equally as important as the flow of raw materials and products to the success of a supply network. To overcome these limitations, we present a peer-to-peer based multiagent system supported by auto-ID and mobile computing technology that integrates data from several members of the network, so that each member can visualize the pertinent data. We propose a flexible and open multiagent architecture that can be easily reconfigured when the supply network structure is reorganized. Moreover, mobile access to information enables decision-makers to act and react more quickly. Most of the technologies we use, such as GPS and RFID, are already on the market at affordable prices, and are technically mature. The basic idea is to integrate these independently developed technologies in an agent-based system for mobile supply chain event management (SCEM).

122 Frank Teuteberg and Ingmar Ickerott

Thus, we implemented a prototype of a peer-to-peer based multiagent system with access to back-office systems to identify, monitor, and anticipate unexpected events in supply networks as a proof of concept. Our prototypical implementation and preliminary simulation results show that accessing multiagent systems on top of heterogeneous ERP systems through mobile devices is a promising approach for mobile SCEM. In the future, we intend to simulate business processes in supply networks based on real-world data from our business partners. This work is novel in that it demonstrates how agent-based mobile SCEM can be realized to respond in real-time to disruptions that might occur in supply networks and to propose corrective actions. It is a first step toward the creation of decentralized self-controlled and adaptive supply networks. Although our proposed system provides a secure environment for fostering mobile information sharing in supply networks, in real-world situations, supply network members may still refuse to share information with their partners. Thus, a critical factor for adopting agent-based systems for mobile SCEM and guaranteeing easier acceptance in real business environments will be the implementation of solutions for both security and privacy issues.

5.10 Acknowledgement This work is part of the research project “Mobile business processes and user interfaces based on Wireless Internet” (Mobile Internet Business) funded by the German Federal Ministry of Education and Research (http://mib.uni-ffo.de).

5.11 References Adorni G, Bergenti F, Poggi A, Rimassa G, (2001) Enabling FIPA Agents on Small Devices. Klusch M, Zambonelli F, eds. Cooperative Information Agents V. Proceedings of 5th International Workshop CIA 2001. Lecture Notes in Computer Science 2182, Springer, Heidelberg:248–257. Akyldiz I, Su W, Sanakarasubramaniam Y, Cayirci E, (2002) Wireless Sensor Networks: A Survey. Computer Networks Journal 4:392–422. Albuquerque RL, Hübner JF, de Paula GE, Sichman JS, Ramalho GL, (2001) KSACI: A Handheld Device Infrastructure for Agents Communication. Meyer JJ, Milind T, (eds.). Pre-proceedings of the Eighth International Workshop on Agent Theories, Architectures, and Languages:327–337. Alvarenga CA, Schoenthaler, RC, (2003) A New Take on Supply Chain Event Management. Supply Chain Management Review 2:29–25 Angeles R, (2005) RFID Technologies: Supply-Chain Applications and Implementation Issues. Information Systems Management 1:51–65. ARC Advisory Group (2005) Supply Chain Process Management Worldwide Outlook. Available at: http://www.arcWeb.com/research/ent/scpm.asp Berger M, Rusitschka S, Schlichte M, Toropov D, Watzke M, (2003) Porting Agents to Small Mobile Devices – The Development of the Lightweight Extensible Agent Platform. EXP – In Search of Innovation 3:32–41.

Mobile Supply Chain Event Management 123

Blanchard RJ, (2004) SCM Spotlight: RFID and the Adaptive Enterprise: What’s Real Now? What’s Next? Available at: http://www.sapinsideron-line.com/searchspi/search.htm?page=article&key=37914 Bose I, Pal R, (2005) Auto-ID: Managing Anything, Anywhere, Anytime in the Supply Chain. Communications of the ACM 8:100–106. Braubach L, Pokahr A, Lamersdorf W, (2004) Jadex: A Short Overview. Proceedings of the Main Conference Net.ObjectDays, AgentExpo (on CD). Brund G, Müssinger P, Polani D, Schmitt R, Spalt R, Uthmann T, Weber S, (2003) XRaptor – A Simulation Environment for Continuous Virtual Multi-Agent Systems – User Manual, Version 7.3.8a. Available at: http://www.informatik.uni-mainz.de/~polani/XRaptor/ Cabral L, Domingue J, Motta E, Payne T, Hakimpour F, (2004) Approaches to Semantic Web Services: An Overview and Comparisons. Lecture Notes in Computer Science 3053, Springer, Heidelberg:225–239. Chan FTS, Tang NKH, Lau HCW, Ip RWL, (2002) A Simulation Approach in Supply Chain Management. Integrated Manufacturing Systems 2:117–122. Chang Y, Makatsoris H, (2001) Supply Chain Modeling Using Simulation. International Journal of Simulation 1:24–30. Chu E, (2003) A Review of Simulation Studies on Supply Chain Management. Journal of the Academy of Business and Economics 1:22–27. Dangelmeier W, Pape U, Rüther M, (2004) Agentensysteme für das Supply Chain Management: Grundlagen, Konzepte, Anwendungen”. Deutscher Universitaets-Verlag, Wiesbaden. Roy D, Anciaux D, Monteiro T, Ouzizi L, (2004) Multi-agent Architecture for Supply Chain Management. Journal of Manufacturing Technology Management 8:745–755. Daniels M (1999) Integrating Simulation Technologies with Swarm. Available at: http://www.santafe.edu/~mgd/anl/anlchicago.html Flörkemeier C, Anarkat D, Osisnki T, Harrison M, (2003) PML Core Specification 1.0. White Paper, Auto-ID Center at MIT, Cambridge, MA. Forrester JW, (1961) Industrial Dynamics. MIT Press, Cambridge, MA. Fox M, Barbucceanu M, Teigen R, (2000) Agent-Oriented Supply-Chain Management. International Journal of Flexible Manufacturing Systems 2:165–188. Frey D, Mönch L, Stockheim T, Woelk P, Zimmermann R, (2003) Agent.Enterprise – Integriertes Supply Chain Management mit hierarchisch vernetzten MultiagentenSystemen. Dittrich et al., eds. Tagungsband der GI-Jahrestagung:47–63. Fung RYK, Chen T, (2005) A Multiagent Supply Chain Planning and Coordination Architecture. The International Journal of Advanced Manufacturing Technology 78:811–819. Geier J, Bell R, (2001) RTLS: An Eye on the Future. ID Systems 3. Available at: http://www.scs-mag.com/reader/2001/2001_03/rtls0301/index.htm Greenwood D, Calisti M, (2004) Engineering Web Service – Agent Integration. Proceedings of the IEEE International Conference on Systems, Man and Cybernetics 2:1918–1925. Hightower J, Borriello G, (2001) Location Systems for Ubiquitous Computing. IEEE Computer Magazine 8:57–66. Hofmann O, Deschner D, Reinheimer S, Bodendorf F, (1999) Agent-Supported Information Retrieval in the Logistic Chain. Proceedings of the 32nd Hawaii International Conference on System Sciences, Maui. Available at: http://csdl.computer.org/comp/proceedings/hicss/1999/0001/08/00018028.PDF Howden N, Rönnquist R, Hodgson A, Lucas A, (2004) JACK Intelligent Agents – Summary of an Agent Infrastructure. Proceedings of 5th International Conference on Autonomous Agents. Available at:

124 Frank Teuteberg and Ingmar Ickerott

http://www.agent-software.com/shared/resources/papers/JACK-infrastructure.pdf Huhns MN, Stephens LM, (2001) Automating Supply Chains. IEEE Internet Computing 4:90–93. Jankowska AM, Kurbel K (2005) Service-Oriented Architecture Supporting Mobile Access to an ERP System. Ferstl OK, Sinz EJ, Eckert S, Isselhorst T, (eds.). Wirtschaftsinformatik 2005: eEconomy, eGovernment, eSociety, Physica-Verlag, Heidelberg:371–390. Jung H, Jeong B, (2005) Decentralised Production-Distribution Planning System Using Collaborative Agents in Supply Chain Network. The International Journal of Advanced Manufacturing Technology 1-2:167–173. Jung JY, Blau G, Pekny J, Reklaitis G, Eversdyk D, (2004) A Simulation Based Optimization Approach to Supply Chain Management under Demand Uncertainty. Computer and Chemical Engineering 10:2087–2106. Kärkkäinen M, Främling K, Ala-Risku T, ( 2003) Integrating Material and Information Flows Using a Distributed Peer-to-Peer Information System. Jagdev HS, Wortmann JC, Pels HJ, (eds.). Collaborative Systems for Production Management:305–319. Kleijnen JPC, (2005) Supply Chain Simulation Tools and Techniques: A Survey. International Journal of Simulation & Process Modeling 1/2:82–89. Klügl F, (2001) Multiagentensimulation: Konzepte, Werkzeuge, Anwendungen. AddisonWesley, München. Krishnamurthy S, Zeid I, (2004) Distributed and Intelligent Information Access in Manufacturing Enterprises Through Mobile Devices. Journal of Intelligent Manufacturing 2:175–186. Law AM, Kelton WD, (2000) Simulation Modeling and Analysis. McGraw-Hill, Boston, Third Edition. Lee HL, (2004) The Triple-A Supply Chain. Harvard Business Review 10:103–112. Luck M, McBurney P, Shehory O, Willmott C, AgentLink community, (2004) Agent Technology Road map: Overview and Consultation Report. Available at: http://www.agentlink.org/road map/road mapreport.pdf Mahmoud QH, (2001) MobiAgent: An Agent-based Approach to Wireless Information Systems. Proceedings of Agent-Oriented Information Systems Workshop at Autonomous Agents 2001, Montreal. McGinity M, (2004) RFID: Is This Game of Tag Fair Play?. Communications of the ACM 1:15–18. Nurmilaakso JM, (2004) Supply Chain Scheduling Using Distributed Parallel Simulation. Journal of Manufacturing Technology Management 8:756–770. Nwana H, Ndumu D, Lee L, Collis J, (1999) ZEUS: A Toolkit and Approach for Building Distributed Multi-agent Systems. Etzioni O, Müller J, Bradshaw J, (eds.). Proceedings of the Third International Conference on Autonomous Agents, Seattle, WA:360–361. NCX (2004) Business Activity Monitoring. Available at: http://www.ncxinc.ca/documents/Xagent%20-%20BAM.pdf Pokahr A, Braubach L, Lamersdorf, W, (2003) Jadex: Implementing a BDI-Infrastructure for JADE Agents. EXP – In Search of Innovation 3:76–85. Reddy AM, Rajendran C, (2005) A Simulation Study of Dynamic Order-up-to Policies in a Supply Chain with Non-Stationary Customer Demand and Information Sharing. The International Journal of Advanced Manufacturing Technology 9/10:1029–1045. Reichenthal SW, (2002) SRML - Simulation Reference Markup Language, W3C Note, Dec. 2002. Available at: http://mirrors.Webthing.com/view=Medium/www.w3.org/TR/SRML SAP, (2005) SAP Auto-ID Infrastructure. Available at: http://www.sap.com/solutions/netweaver/autoidinfrastructure.epx

Mobile Supply Chain Event Management 125

Speyerer JK, Zeller AJ, (2004) Managing Supply Networks: Symptom Recognition and Diagnostic Analysis with Web Services. Proceedings of the 37th Hawaii International Conference on System Sciences. Available at: http://csdl.computer.org/comp/proceedings/hicss/2004/2056/07/205670207b.pdf Sun, (2004) The Sun EPC Network Architecture: A Technical White Paper. Available at: http://www.sun.com/aboutsun/media/presskits/networkcomputing04q2/epc_whitepaper. pdf Szirbik NB, Wortmann JC, Hammer DK, Goossenaerts JBM, Aerts ATM, (2000) Mediating Negotiations in a Virtual Enterprise via Mobile Agents. IEEE Computer Society, (eds.). Proceedings of the Academia/Industry Working Conference on Research Challenges Buffalo, NY:237–242. Taylor SJE, (2004) Distributed Simulation in Industry: Status and Perspectives. Mertins K, Rabe, M, (eds.). Experiences from the Future: New Methods and Applications in Simulation for Production and Logistics:21–42. Teuteberg F, Schreber D, (2005) Mobile Computing and Auto-ID Technologies in Supply Chain Event Management – An Agent-Based Approach. Bartmann D et al., (eds.). Proceedings of European Conference on Information Systems (ECIS), Regensburg (on CD). Twist DC, (2005) The Impact of Radio Frequency Identification on Supply Chain Facilities. Journal of Facilities Management 3:226–239. W3C, (2004) Web Services Architecture Requirements. Available at: http://www.w3.org/TR/2004/NOTe-wsa-reqs-20040211/ Wang M, Wang H, Xu D, Wan K, Vogel D (2004) A Web-service Agent-based Decision Support System for Securities Exception Management. Expert Systems with Applications 1:439–450. Wenzel S, Noche B, (2000) Tools in Production and Logistics – A Market Survey. Mertins K, Rabe M, (eds.) The new simulation in production and logistics: Prospects, views and attitudes, IPK-Berlin:424–432. Witte T, (1990) Object Oriented Simulation and Relational Databases. Schmidt B, (ed.). Modeling and Simulation:70–74. Yoo T, Kim D, Cho H (2004) A new approach to multi-pass scheduling in shop floor control. Ingalls R.G. Rosetti M.D, Smith JS., Peters BA, (eds.). Proceedings of the 2004 Winter Simulation Conference:1109–1114. Yufei Y, Detlor B, (2005) Intelligent Mobile Crisis Response Systems. Communications of the ACM 2:95–98. Yung SK, Yang CC, Yen J, (2000) Applying Multi-Agent Technology to Supply Chain Management. Journal of Electronic Commerce Research 4:119–132. Zimmermann R, Paschke A, (2003) PAMAS – An Agent-Based Supply Chain Event Management System. Proceedings of the Ninth Americas Conference on Information Systems, Tampa. Available at: http://ibis.in.tum.de/staff/paschke/docs/PAMAS_AMCIS_03.pdf Zwicky F, (1969) Discovery, Invention, Research – Through the Morphological Approach. Macmillan, Toronto.

6 Impact of Information Technology on Supply Chain Management Enver Yücesan

Abstract:

Information is said to be the glue that holds supply chains together. As a key infrastructure, Web-based technologies continue to have significant impact on supply chain strategies. On the coordination side, the Web provides a virtually free platform for enhancing transparency, eliminating information delays and distortions, and significantly reducing transaction costs. One should note, however, that, although information flow has accelerated considerably, material flow has not gained much speed. This phenomenon makes the coordination of material, information, and cash flows even more crucial for effective supply chain coordination. On the design side, current technology does not yet permit dynamic supply chain design in response to changing business environment. The adoption of Web Services represents a significant step in that direction.

6.1 Introduction The 1986 Annual Report of the Digital Equipment Corporation (DEC) set the ambitious goal “to connect all parts of an organization – the office, the factory floor, the laboratory, the engineering department – from desktop to data center. We can connect everything within a building; we can connect a group of buildings on the same site or at remote sites; we can connect an entire organization around the world. We propose to connect a company from top to bottom with a single network that includes the shipping clerk, the secretary, the manager, the vice president, even the president” (Garvin 1997). More importantly, this goal was not based on some vaporware, but on a concrete enabling technology, a new generation of super minicomputers based on a single computer architecture, VAX. From small desktop machines to computer clusters, VAX-based machines would be fully

128 Enver Yücesan

compatible, use a uniform operating system, and communicate across shared networks. While DEC was working on the hardware and the infrastructure, a relativelyyoung German company, SAP AG, was taking on a big gamble by transitioning its mainframe-based enterprise resource planning (ERP) software, the R/2, into the client-server architecture, the R/3, making the enterprise software accessible to thousands of organizations that did not necessarily wish to invest in mainframe computers. However, it took almost two decades during which the computing power has tremendously increased and the Internet has become ubiquitous for DEC’s 1986 vision of creating a networked organization to become a reality. Scholars of strategic management increasingly recognize that the source of value creation may lie in networks of firms (Dyer and Nobeoka 2000; Gulati et al. 2000). Amit and Zott (2001) further built on this line of thinking by postulating that value is created by the way in which transactions are enabled. Enabling such transactions requires a network of capabilities drawn from multiple stakeholders such as customers, suppliers, and complementors. Information plays a crucial role in enabling transactions in supply chains. Creating an adequate information infrastructure to interface the members of a supply network has always been challenging. Such an infrastructure must be able to satisfy simultaneously the following requirements (Upton and McAfee 1996): First, it must be able to accommodate members with varying degrees of IT sophistication. Second, it must provide a wide range of functionality ranging from simple data transmission to access to remote applications. Finally, it must be able to accommodate a constantly changing pool of suppliers and customers within varying stages of relationships. The Internet has emerged as an infrastructure technology that may simultaneously satisfy these three requirements. Johnson and Wang (2002) define e-business as “the marriage between the Internet and supply chain integration.” Lee and Whang (2002) divide e-business applications into three categories: eCommerce, e-procurement, and e-collaboration, as depicted in Figure 6.1. eCommerce enables “a network of supply chain partners to identify and quickly respond to changing customer demand captured over the Internet.” e-Procurement enables “companies to use the Internet for procuring direct or indirect materials as well as for handling value-added services.” e-Collaboration facilitates “coordination of various decisions and activities beyond transactions among the supply chain partners over the Internet.” ERP systems have played a crucial role in e-enabling companies. In a natural experiment conducted at a U.S. high-tech manufacturer, McAfee (2002) found evidence that the implementation of an ERP system has led to significant improvements in operational performance (e.g., lead times and on-time deliveries) after an initial performance dip. In a recent survey, however, supply chain professionals highlighted four shortcomings in current ERP systems (Akkermans et al. 2003):

Impact of Information Technology on Supply Chain Management 129

x x x x

lack of extended enterprise (EE) functionality, lack of flexibility in adapting to changes in the environment, lack of more advanced decision support functionality, and lack of (Web-enabled) modularity.

Although Web-based technologies enable the management of a portfolio of relationships in an effective and efficient manner by drastically reducing transaction costs (through e-procurement and e-commerce), the cost of establishing such relationships (through e-collaboration) remains relatively high. Furthermore, the developments in the computing and telecommunications industries made the transfer of information almost instantaneous, out manufacturing, warehousing and distribution technologies could not accelerate the movement of material to such phenomenal levels. The coordination of information, cash, and material flows has thus assumed increased importance for effective supply chain management. For instance, Randall et al. (2006) empirically illustrate the importance of investment in fulfillment capabilities for Internet retailers.

e-Procurement

e-Commerce

e-Enabled Firm

Supplier

Customer

e-Collaboration

Figure 6.1. Lee and Whang’s classification of e-business applications

The transition from the mainframe to the client-server architecture was the key technological breakthrough that unleashed the first ERP revolution. The adoption of the service-oriented architecture (SOA) holds the promise of triggering the next wave by enabling the dynamic reconfiguration of supply chains, making them readily adaptable to changing business models, growing competition and globalization, tighter regulations, and increased mergers and acquisition activities. Web Services are the key building blocks of SOA. A Web Service is a selfcontained, self-describing piece of application functionality that can be found and accessed by other applications using open standards. By using highly standardized

130 Enver Yücesan

interfaces to hide the implementation of the underlying functionality, Web Services enable interoperability and compatibility among various heterogeneous applications (Currie and Parikh 2006). ERP vendors are extending Web Services standards and SOA principles to develop composite applications to support new business processes or scenarios. For example, SAP’s enterprise services architecture aggregates Web Services into business-level enterprise services, providing higher level building blocks for the task of dynamically adapting the IT infrastructure to evolving business conditions. In this chapter, we review the impact of the ERP revolution, the Internet, and other Web-based technologies on supply chain strategies. The impact of information technology on supply chain coordination has been undeniably positive, out some reservations remain regarding its impact on supply chain design. The remainder of the chapter is organized as follows. Section 6.2 briefly introduces working definitions and key trends in supply chain management (SCM). Section 6.3 illustrates the role of IT in SCM and IT’s ability to keep up with these trends. Section 6.4 discusses the challenges of dynamic supply chain design. Section 6.5 concludes the chapter with practical guidelines for practitioners.

6.2 Supply Chain Management 6.2.1 Working Definitions As illustrated in Figure 6.2, a supply chain is a network consisting of suppliers, manufacturers, distributors, retailers, and customers. The network supports three types of flows that require careful planning and close coordination: x Material flows, which represent physical product flows from suppliers to customers as well as the reverse flows for product returns, servicing, and recycling; x Information flows, which represent order transmission and order tracking to coordinate the physical flows; and x Financial flows, which represent credit terms, payment schedules, tax considerations, and other contractual arrangements. Note that all three flows are bidirectional. Traditional thinking focuses on goods and services flowing from suppliers downstream through a series of valueadding steps all the way to the end customer. Increasingly, manufacturers are responsible for taking back their product at the end of the product’s life cycle and disposing of it in an environmentally responsible fashion. Closed-loop – not linear – supply chains must, therefore, be designed right from the outset. Similarly, information was thought to flow from the market upstream to all the tiers in a supply chain. Innovative organizations (e.g., UPS and Dell among others) have shown the value of providing customers with real-time information about the status of their requests. Finally, financial flows are no longer simply based on payment terms. Facilities ranging from consignment stock to various risk-sharing arrangements have also made financial flows bidirectional. The supply chain, a platform to coordinate these three flows, is constructed with three key building blocks:

Impact of Information Technology on Supply Chain Management 131

x x x

Processes, which encompass such value-adding activities as order fulfillment, new product development, and after-sales service; Organizational structures, which encompass a range of relationships from total vertical integration to networked companies as well as performance measurement and reward schemes to reinforce such structures, and Enabling technologies, which encompass both process and information technologies.

Financial flows Information flows Material flows Suppliers o Manufacturers oDistributors oCustomers

Processes

Organizational structures

Enabling Technologies

Figure 6.2. A working definition of the supply chain

The above definitions hide some crucial subtleties. First and foremost, modern SCM has significantly benefited from a series of improvement initiatives. Over the past few decades, the waves of just-in-time, total quality management, and business process reengineering produced a significant impact on the individual components of a supply chain, eliminating nonvalue-adding activities, enhancing productivity, and streamlining workflows, ultimately enabling us to focus on the interfaces among these individual components. Second, the advances in communication and computation technologies have made it possible to collect, analyze, transmit, and deploy huge amounts of data necessary to run operations on a global scale. Finally, infrastructure investments along with developments in global transportation have greatly facilitated the movement of goods. Supply chains perform two principal functions (Fisher 1997): The physical function of transformation, storage, and transportation, and the market mediation function of matching demand and supply in a volatile and uncertain world. The physical function has been extensively studied within the production control and inventory management literature, but innovative approaches have recently been emerging in the market mediation function. These approaches are classified in Figure 6.3. Supply chain coordination is concerned with the coordination of the three types of flows in the network. Effective coordination strategies combine a range of approaches for enhancing supply chain transparency through information sharing

132 Enver Yücesan

(e.g., sharing point-of-sales data with the manufacturer) and information deployment [(e.g., centralizing decision rights through vendor-managed inventories (VMI), efficient consumer response (ECR), and collaborative planning, forecasting and replenishment (CPFR) initiatives] as well as for increasing operational flexibility [e.g., replacing make-to-stock systems with assemble-to-order (ATO) and make-to-order (MTO) systems] to be able to react to timely information. These approaches may facilitate new forms of organizational structures (e.g., process orientation, e-collaboration) and new forms of interorganizational collaboration (e.g., outsourcing via third-party service providers, e-procurement, and e-commerce). Information and communication technologies facilitating closer collaboration and promoting supply chain transparency are crucial for effective coordination. Innovative product and process designs are a prerequisite for operational flexibility. Most of the innovative supply chain coordination practices (e.g., postponement) are therefore enabled by innovative product, process, and supply chain design.

Market Mediation

Supply Chain Design - Network configuration - Capability building

Supply Chain Coordination

Information Sharing

Information Deployment

Operational Flexibility

- POS data - Joint forecasting - Schedule sharing

- Contracts - VMI/ECR - CPFR

- Postponement - ATO, MTO

Figure 6.3. Matching demand and supply in a supply chain

Supply chain design is thus concerned not only with the determination of the optimal network topology, but also with the prioritization of the capabilities to be developed and retained internally, the selection of those capabilities to outsource, and forging new partnerships with other entities along a supply network. According to Fine (1998), supply chain design ought to be thought of as “the capability to design and assemble assets, organizations, skill sets, and competencies for a series of competitive advantages, rather than a set of activities held together by low transaction costs.” This dynamic view is necessary in a fastevolving world where new products and emerging distribution channels necessitate a continuous review of supply chain design decisions. Just as product design has an enormous impact on manufacturing performance, superior supply chain design offers significant benefits in supply chain coordination.

Impact of Information Technology on Supply Chain Management 133

The fundamental law of supply chain coordination is the bullwhip phenomenon (Lee et al. 1997, Fiala 2005). This is the amplification of demand volatility along the network as one moves upstream away from the market. The key drivers of the bullwhip phenomenon include the lack of information sharing, communication, and collaboration among the supply chain entities resulting in information distortion as well as delays in information and material flows. The fundamental law of supply chain design is industry clock-speed, the rate with which products, processes, and organizations evolve over time, necessitating a rethinking of the existing supply chain solution (Fine 1998). Technology, competition, and regulatory initiatives appear to drive industry clock-speed, which, in turn, determines the validity (or shelf life) of a supply chain design. Evolving products, processes, or industry structures necessitate adaptable supply chain solutions. 6.2.2 Key Trends in SCM Akkermans et al. (2003) reported a ranked list of key SCM trends generated by a group of European SCM professionals. As reproduced in Table 6.1, the panel of experts sees further integration of activities between suppliers and customers across the entire chain as one of the three biggest trends in SCM. This view coincides with a strong trend toward mass customization. Both trends may have a similar root cause, i.e., increased competition driven by growing consumer power helped by increasing transparency of the global marketplace. Rapidly changing customer requirements not only tolerate very little inventory in the supply chain, but also require drastic modifications in supply chain topologies. This requirement poses a tough challenge to ERP systems in maintaining sufficient flexibility as supply chain needs keep changing. SCM experts recognize the difficulty for a single organization to satisfy the changing requirements of consumers. They expect that supply chains will consist of several enterprises and that noncore activities such as physical distribution and finance and administration (F&A) will be increasingly outsourced. An important issue for our panel then becomes the determination of “who will be sitting in the driver’s seat” in this chain, since conventional command-and-control structures no longer apply in a network of independent firms. Greater and fasterchanging demands from customers will need to lead to faster and more comprehensive information exchanges among all the players in the chain. In terms of technology, this will not just mean better ERP systems but, in general, enhanced ITtools to integrate the different parties in the supply chain. Internet technology is most likely to provide the technological means for doing so. This will make possible distributed architectures, in which standardization takes place mainly at the level of information definitions and processes, so that local flexibility in information usage can be maintained. Needless to say, all these developments are taking place on a global scale. Hence, IT for SCM in general, and ERP systems in particular, will have to be developed globally.

134 Enver Yücesan

Table 6.1. Trends in supply chain management

Key trends in SCM

% Votes

1. Further integration of activities between suppliers and customers across the entire chain 2. How to maintain flexibility in ERP systems to deal with changing supply chain needs? 3. Mass customization: complex assortments, shorter cycle times, less inventory 4. Who will be in the driver's seat in supply chain coordination? 5. Supply chains consisting of several enterprises 6. Full exchange of information with all the players in the chain 7. Further outsourcing of activities such as physical distribution, finance and administration 8. Enhancements of IT tools required to integrate the different parties in the supply chain 9. Globalization: How to build worldwide ERP systems? 10. Greater transparency of the global marketplace 11. Internet technology will be the backbone to connect systems of partners in the chain 12. Standardization of processes and information definitions, the rest is IT infrastructure

87% 57% 39% 35% 35% 35% 30% 30% 26% 26% 26% 22%

Source: Akkermans et al. 2003

6.3 IT in Supply Chain Coordination Information is said to be the glue that holds supply chains together. The bullwhip phenomenon, the key challenge in supply chain coordination, is driven by delayed and distorted information as well as by transaction costs promoting local optimization (Lee et al. 1997). Information technologies (IT) have the greatest impact on supply chain coordination by eliminating information delays and distortions, by reducing transaction costs, and, ultimately, by enabling ecollaboration, which is defined as business-to-business interaction facilitated by the Internet. These interactions include such advanced activities as information sharing, decision sharing, process sharing, and resource sharing (Johnson and Wang 2002). As indicated earlier, creating an adequate information infrastructure to interface the members of a supply chain has always been challenging. Such an infrastructure must be able to satisfy simultaneously the following requirements (Upton and McAfee 1996): First, it must provide the lowest common denominator of IT sophistication among partners. Second, it must provide a wide range of functionality ranging from simple data transmission to access to remote applications. Finally, it must be feasible at any stage of relationship between any

Impact of Information Technology on Supply Chain Management 135

two companies; that is, it must be able to accommodate a constantly changing pool of suppliers and customers. Within this framework, Figure 6.4 reproduces these three dimensions of electronic connectivity, as introduced by Upton and McAfee (1996). The utility of potential infrastructure technologies in enabling e-collaboration can then be assessed by how well “they fill the cube.” For example, electronic data interchange (EDI) has been the most widely used tool for connecting manufacturers with their suppliers. From a functionality perspective, EDI affords simple data transmission under a particular file format over a dedicated communication channel. Rudimentary computer skills are required for maintaining EDI connectivity. Given the dedicated communications infrastructure and the proprietary standards, however, EDI necessitates significant up-front investment and considerable expense for maintenance. Such an investment is difficult to justify at the early stages of a buyer-supplier relation, where the buyer is in the process of assessing supplier capability and, therefore, is unwilling to provide any long-term commitment.

data access naive

2. Lowest common denominator of IT sophistication among partners

simple data transmission

experienced

Most Difficult

3. Level of functionality

telepresence/access to applications

Least Difficult integrated collaborative

exploratory

1. Stage of relationship between any two companies Figure 6.4. Upton and McAfee’s framework for electronic connectivity

Groupware extends the limited information transmission functionality of EDI into a (limited) collaborative platform. However, the required computer skills to take full advantage of groupware’s capabilities are even higher. Furthermore, groupware requires even a higher initial investment than EDI, which, once again, makes the tool more appropriate at more advanced stages of a supply chain relationship. One can obtain further IT functionality by establishing a wide-area network. Such a platform can be used not only for information transmission, but also for collaboration, and granting access to application programs, i.e., telepresence, for the suppliers. Unfortunately, the increased functionality comes at

136 Enver Yücesan

a higher level of initial investment. One can also envisage such an infrastructure only among close partners along the supply chain. In summary, traditional technologies always entail a trade-off: Improvements in one dimension (e.g., enhanced functionality) come at the expense of further complications in at least one of the other two dimensions (e.g., higher initial investments). ERP systems, which arguably had the most significant impact on SCM, could not initially break this trade-off. In the functionality dimension, ERP systems offer the opportunity for unparalleled transparency across the organization, making a single database visible to all the stakeholders. This visibility may, in turn, enable closer cooperation. These capabilities, however, can be deployed through long and expensive implementation processes, which typically require advanced levels of IT sophistication. Furthermore, best-practice templates that often guide ERP implementations limit the scope of applicability of these systems to advanced stages of relationships among collaborating organizations. In short, traditional ERP systems fail to fully “fill the cube” as well. This failure is discussed in further depth in the next section. 6.3.1 Shortcomings of Current ERP Systems for SCM Table 6.2 summarizes the principal shortcomings in ERP along with a pointer to the key SCM trends summarized in Table 6.1, as reported by Akkermans et al. (2003). x Lack of EE functionality: Extended enterprise functionality entails the ability to share internal data efficiently with supply chain partners and to accommodate the data made available by the partners. Data sharing can be deployed either for operational decision-making or for calculating overall supply chain performance measures. Moreover, EE functionality enables business processes to be distributed over multiple organizational entities, as a first step toward e-collaboration. x Lack of flexibility in adapting to changing supply chain needs: A single organization might have different types of relationships with its supplier and customer base. Its ERP system should be sufficiently flexible to accommodate a multitude of relationships. Some suppliers may have adopted VMI, and others may still be engaged in a classical vendor/buyer relation. The ERP system should be able to accommodate all these different modes of collaboration simultaneously and be able to change efficiently from one mode to another. Gartner Institute emphasizes that the ability to engage into — and disengage from — collaborative relationships is of critical importance. x Lack of advanced decision support capabilities: A recent trend in ERP is the emergence of advanced planning and scheduling systems (APS). In itself, planning with longer time horizons and across different business units is nothing new for ERP. However, as it becomes increasingly apparent that supply chains, rather than individual organizations, compete, there is an increasing demand for collaborative architectures in decision support software.

Impact of Information Technology on Supply Chain Management 137

x Lack of open, modular system architecture: Current ERP systems lack modular, open, and Internet-like system architecture. Basically, this shortcoming is the reverse side of some of the generic advantages of ERP, whereby ERP was originally intended to replace a multitude of local legacy systems; a great deal of emphasis was, therefore, placed on its integrated architecture. In the new, networked economy, this former strength has rapidly become a weakness. 6.3.2 Web Services and Service-oriented Architecture The recasting of ERP systems within the service-oriented architecture (SOA) not only addresses these shortcomings, but also eliminates the electronic connectivity trade-off of Figure 6.3. SOA is enabled through Web Services, software components that can interact with one another dynamically via standard Internet technologies. For instance, TCP/IP provides a universal communication standard over the Internet for connecting disparate computer systems. A universal communication standard, in turn, significantly reduces the up-front investment needed to connect various players along the supply chain regardless of the type of computing platform they possess. Low entry and exit costs make the Internet and Web-based applications affordable at any stage of a supply chain relationship. Given the flexibility to customize the interface over the Web, one can customize the communication channels for each supplier. Web-based technologies also provide a full portfolio of functionality ranging from simple information transmission to telepresence. Standardized interfaces and application development software make it easy even for the uninitiated to start using the system quickly, further reducing the lowest common denominator of IT sophistication among the supply chain entities. In short, SOA enabled by Web Services fills the cube. Wider acceptance of open standards, cheap and powerful computing, increased bandwidth, enhanced security, accumulated expertise, and higher familiarity with the technology are bound to increase the utility of such technologies in supply chain coordination. With increased connectivity, the Web provides a virtually free platform for enhancing transparency, eliminating information delays and distortions, and reducing transaction costs, ultimately mitigating the bullwhip phenomenon. The transition from the mainframe to the client-server architecture was the key technological breakthrough that unleashed the first ERP revolution. The adoption of Web Services and SOA holds the promise of unleashing the next wave by enabling the dynamic reconfiguration of supply chains. The enabling technology is Web Services, which denote a group of technologies that allow accessing business processes or information over the Internet through application-to-application interaction (Moitra and Ganesh 2005). Web Services are loosely coupled, dynamically bound, accessible over the Web, and standards based. The W3C Architecture Working Group defines Web Services as “software applications identified by a Universal Resource Indicator, whose interfaces and bindings are capable of being defined, described, and discovered as XML artifacts.” In layman’s terms, Web Services can be published, located, and invoked by other applications

138 Enver Yücesan

over the Internet, thereby integrating applications written in different languages and operating on disparate platforms.

Table 6.2. Shortcomings of current ERP systems

Shortcomings of current ERP systems mentioned, grouped Key SCM Trends by common threads from Table 1 1. Lack of extended enterprise functionality: The ability to support operations across multiple organizations 1. (Integration) x Extended enterprise functionality 4. (Driver seat) x Extended enterprise functionality 7. (Transparency) x ERP systems miss linking across the boundaries of enterprises 7. (Transparency) x ERP systems don’t interconnect easily with other than partner systems x Information exchange between parties is underdeveloped 1. (Integration) x Ability to support multiple coding system to enable cross- 1. (Integration) company implementations 2. Lack of flexibility in adapting to ever-changing supply chain needs 3. (Customization) x Flexibility to adapt to changing business models 7. (Transparency) x Flexibility to adapt to changes in business processes 3. Lack of more advanced supporting functionality beyond transaction management x Flow-based information exchange instead of order-based 1. (Integration) 1. (Integration) x MRP-based instead of finite capacity; ERP+ required 3. (Customization) x Advanced planning systems with proven functionality 4. (Driver seat) x Connections with tactical decisions 4. (Driver seat) x From transactions to information for decision-support 4. Lack of open, modular, Internet-like system architectures 4. (Driver seat) x Modular set of systems 4. (Driver seat) x Module manager for the supply chain 3. (Customization) x Connectivity 6. (Info exchange) x Web-enabled ERP 5. Various x IT (network technology, big, shared databases, XML,…) 6. (Info exchange) 1. (Integration) x Customization will remain necessary 6. (Info exchange) x Identification of barriers and developing business cases to overcome these Source: Akkermans et al. 2003

A Web Service is thus a self-contained, self-describing piece of application functionality that can be found and accessed by other applications using open standards. A Web Service is self-contained in that the application using the Web Service does not have to depend on anything other than the service itself. It is self-

Impact of Information Technology on Supply Chain Management 139

describing in that all the information on how to use the service can be obtained from the service itself. The descriptions are centrally stored and accessible through Web-standards-based mechanisms to all applications that wish to invoke the service. By using highly standardized interfaces to hide the implementation of the underlying functionality, Web Services enable interoperability and compatibility among various heterogeneous applications. Web Services make available open and standardized interfaces, allowing for the encapsulation and the software and applications breaking into component. Hence, they enable easy configuration and reconfiguration of software applications. In other words, instead of requiring programmers to establish and maintain links between applications, Web Services are loosely coupled, enhancing flexibility and promoting reuse. Changes can be made in the underlying implementation or in the program calling the Web Service as long as the behavior of the Web Service stays the same. The underlying philosophy behind Web Services is based on service-oriented Architecture. SOA would allow service providers (such as ERP vendors and their complementors) to publish services, which may be accessed by the service consumers (such as members of a supply network) resulting in a high degree of service reuse (McGovern et al. 2006). Consider, for example, the functionality “delete order” that may necessitate cross-application activities, including sending a confirmation to the customer, removing the order from the production plan, releasing materials allocated to the order, notifying the invoicing department, and changing the order status and deleting it from various systems. Instead of the hard wiring illustrated in Figure 6.5, each of these activities may be a Web Service offered by different systems. The ability to build a complex end-to-end solution to cancel an order would be a powerful enterprise-level business service. Web Services, however, are too granular to be used as efficient building blocks for enterprise business scenarios. In fact, ERP vendors are extending Web Services standards and SOA principles to develop composite applications to support new business processes or scenarios. Aggregating Web Services into business-level enterprise services provides more meaningful building blocks for the task of dynamically adapting the IT infrastructure to evolving business conditions. This is what Evgeniou (2002) calls an “adaptive enterprise” where IT provides both visibility and flexibility. For example, SAP’s enterprise services architecture, depicted in Figure 6.6, attempts to provide solutions to the current ERP shortcomings of lack of EE functionality and lack of flexibility in adapting to changing supply chain needs through composite applications, of lack of advanced decision support capabilities through analytics, and of lack of open modular system architecture through an open composition platform. One should note, however, that wider acceptance of open standards hinges on the business user’s adoption of competing platform technologies such as IBM’s WebSphere, Microsoft’s .NET, Oracle’s Enterprise Manager, and SAP’s Netweaver.

140 Enver Yücesan

Figure 6.5. Canceling an order in three-tier client-server architecture

6.4 IT in Supply Chain Design The impact of Web-based technologies is less convincing for supply chain design. Three-dimensional concurrent engineering (3D-CE) is a rich framework for dynamically guiding supply chain design. As introduced by Fine (1998), 3D-CE encourages simultaneous design of products, processes, and supply chains and explicitly considers the interfaces among these three dimensions. Some of these interfaces are well understood. For instance, the DFx (e.g., design for manufacturability or design for localization) captures the interdependence of product and process design decisions (Stoll 1986). Similarly, the interface between product and supply chain design is well understood by trying to match agile supply chains with innovative products and efficient supply chains with functional products (Fisher 1997). One of the key, but less well-understood interfaces in 3DCE is between process design and supply chain design, where the principal decision is what to produce in-house and what to procure elsewhere. The outsourcing decision, which is based on a company’s competitive capabilities, is thus based on a company’s needs for additional manufacturing capacity or for external capability (and/or technology). Toy manufacturers rely on outsourcing for additional capacity for short life-cycle products, but equipment manufacturers may rely on suppliers for tight-tolerance machining capability they may not have inhouse. 6.4.1 e-Procurement To appreciate the impact of Web-based technologies on make-or-buy decisions, hence, on the procurement process, consider Forrester’s B2B digital transaction

Impact of Information Technology on Supply Chain Management 141

models reproduced in Figure 6.7. Web-based technologies enable different formats of relationships among potential buyers and vendors. This is consistent with our earlier assessment that the Web provides a platform that simultaneously satisfies all three dimensions of electronic connectivity requirements (Figure 6.3), including the stage of the relationship between any two companies, the lowest common denominator of IT sophistication among partners, and the desired level of functionality.

Figure 6.6. SAP’s enterprise services architecture

Based on transaction theory, Malone et al. (1987) suggest that IT reduces transaction and coordination costs, and will therefore lead to an overall shift from hierarchical coordination to market coordination. In fact, total savings in eprocurement are estimated at around 13% to 28%. Most of these savings come from reduced costs of search (need identification 11%, vendor selection 27%, and vendor approval 23%) and of coordination (order processing, billing, and payment processing 18%, tracking and logistics administration 21%). According to Jupiter

142 Enver Yücesan

Communications, in 2004, on-line transaction volume reached $6 billion in emarkets in the United States, defined as IT-based governance or coordination mechanisms (Malone et al. 1987).

SELLERS

B U Many Y E R S

One

Many

AUCTION Ingram Micro E-Chemicals Poultry First Fast Parts

AGGREGATORS SciQuest Chemdex PlasticNet EXCHANGES E-Steel Altra Paper Exchange

EDI/EXTRANET One

Dell Cisco

BIDDING ReverseAuction FreightWise OnSale

Figure 6.7. Business-to-business digital transaction models

e-Markets, however, pose a dilemma for buyers and vendors. Most manufacturers spent the past two decades establishing close relationships with their suppliers under such different initiatives as strategic sourcing and supply base rationalization. e-Markets, on the other hand, signal a dramatic shift toward an arm’s length relationship focusing solely on cost reduction. Within this shift, the promised transaction cost reductions associated with vendor selection and vendor approval necessitate closer scrutiny. From this perspective, the items purchased by a manufacturer can be classified in two broad categories: Manufacturing (direct) inputs, goods that go directly into a product or a process, and operating (indirect) inputs, usually referred to as MRO (maintenance, repair, and operations). Manufacturing inputs vary widely from industry to industry and hence purchased from industry-specific suppliers, but MRO is not necessarily industry specific and can therefore be purchased from vendors serving many industries. Similarly, procurement practices can be classified under two broad categories: Systematic sourcing, where long-term contracts are negotiated with prequalified suppliers, and spot sourcing, where an

Impact of Information Technology on Supply Chain Management 143

immediate need is fulfilled at the lowest possible cost perhaps from anonymous parties. Putting these two dimensions of what to purchase and how to purchase it on a B2B matrix (Figure 6.8) offers some interesting insights (Kaplan and Shawney 2000). W.W. Grainger is a distributor of MRO supplies in the United States, offering over 200,000 products ranging from nuts and bolts to machine lubricants through its Web site. Such Web sites can be further extended into hubs with multiple suppliers greatly expanding the number of items offered. MRO hubs constitute an example where the Internet is ideally suited for eliminating the inefficiencies of the current channel by lowering transaction costs, by integrating lower tier suppliers, by eliminating the duplication of data entry, and by expanding the product portfolio.

Operating Inputs

Systematic Sourcing

MRO HUBS

Spot Sourcing

YIELD MANAGERS

Manufacturing Inputs

CATALOG HUBS

EXCHANGES

Figure 6.8. The B2B trading matrix

MRO procurement can be further extended by yield managers. While buyers are seeking further cost reductions, e-markets are providing the necessary IT infrastructure for conducting on-line auctions. As early as 1999, FreeMarkets reportly conducted some 30 “competitive bidding events” (CBE) for United Technologies (UTC) totaling around $250 million of purchase volume. Around the same time frame, UTC reported cost reductions of 10% to 70% on such diverse categories as rivets and studs, logistics services, telephone services, and tax preparation. For manufacturing inputs, e-markets provide two key benefits. First, catalog hubs offer the possibility of bringing together a virtually unlimited number of offers from different suppliers on a global scale. Such an infrastructure would greatly reduce the search cost for the buyer. For the supplier – in particular, for the small supplier – the platform offers unparalleled access to potential buyers. This is the first step in systematic sourcing. As for spot sourcing, e-procurement provides, for the first time, yield management capabilities for manufactured products. A manufacturer stuck with low capacity utilization in a particular month can bid for orders to fill up its fixed production capacity, or a buyer with an unexpected shortage of manufacturing input can bid for the material available on the market in

144 Enver Yücesan

the same way airlines price their seats or hotels price their rooms (McAfee and McMillan 1987). Unlike MRO procurement, however, manufacturing inputs are industry specific, where longer term contracts are negotiated with specific, typically prequalified, suppliers. In this domain, the procurement process is typically divided into three stages: Strategic sourcing, supplier management, and day-to-day purchasing. Strategic sourcing includes supplier identification, certification, and selection. Supplier management is concerned with supplier integration, supplier performance evaluation, and contract management. Beyond these two stages, we find the dayto-day purchasing activities, including order request, logistics coordination, and payment management. There is no doubt that e-procurement drastically reduces day-to-day purchasing costs. Our hesitation, therefore, concentrates on the first two stages of the procurement process. Although e-markets provide support for request-for-information (RFI) or request-for-quote (RFQ) preparation, supplier performance evaluation, and contract management, the crucial activity of supplier identification and certification is still affected by the richness/reach trade-off (Evans and Wurster 1999). Reach refers to the number of people and products that are accessible quickly and cheaply in virtual markets; richness refers to the depth and detail of information that can be accumulated, offered, and exchanged among market participants. For supplier selection and certification, the trade-off is depicted in Figure 6.9. During the quality movement of the past two decades, a large number of companies have undergone the ISO certification process. As a result, in RFIs or RFQs, ISO certification has become a natural requirement. The certification, therefore, has achieved worldwide recognition, resulting in great reach. Many buyers, however, have quickly discovered that ISO certification is a necessary but not sufficient assessment of the process capability of a potential supplier. In the automotive and aerospace industries, in particular, manufacturers launched their own certification processes ensuring critical process capability at potential suppliers. These supplier certification programs possess the desired depth, but they have limited reach due to their intensive resource requirements. In most cases, companies have been devising multiyear strategic sourcing programs to reduce and certify their supply base. 6.4.2 e-Collaboration Though necessary, technology is not sufficient in and of itself for dynamic supply chain design. Many researchers have suggested that it is not sufficient that eprocurement and e-commerce provide the potential for higher transaction volumes, lower transaction costs, and better market mediation. Other factors such as transaction complexity and frequency, decision powers, existing market structure, and incomplete contracts may impact the formation and sustainability of markets or hierarchies (Wang and Benaroch 2004).

Impact of Information Technology on Supply Chain Management 145

Richness Supplier certification programs

ISO certification

Reach Figure 6.9. Reach versus richness trade-off in supplier certification

However, one big challenge remains. The shift from hierarchical coordination to market coordination replaces industry structures dominated by vertical integration with networked organizations or loosely coupled ecosystems. In the absence of a clear command and control structure, coordination among the members of a supply chain is not trivial, necessitating the implementation of incentive schemes for aligning the divergent economic interests of the members (Tsay et al. 1999). To render e-collaboration attractive and sustainable, Wang and Benaroch (2004) propose not only coordinating contracts between buyers and vendors, but also economic incentives to encourage supplier participation in the form of a cap on the transaction fees imposed by the market maker and/or the payment of a premium by the buyer. Similarly, Bailey and Bakos (1997) suggest four particular roles for electronic intermediaries to promote e-collaboration: Facilitation of transactions, trust building to prevent opportunistic behavior, matching of buyers and sellers, and aggregation of supply and demand. Others propose additional new roles such as designing innovative procurement practices and providing novel types of transactions. Such innovative procurement practices and novel types of transactions should therefore include contractible initiatives (e.g., buyback policies, quantity flexibility, price protection, and options contracts) to ensure Pareto improving supply chain designs by aligning the economic incentives of buyers and suppliers.

146 Enver Yücesan

6.5 Guidelines for Practitioners Web-based technologies have a significant impact on supply chain strategies. Swaminathan and Tayur (2003) provide an overview of analytical research models developed to address critical issues such as procurement and supplier management, visibility and information sharing, pricing and distribution, customization and postponement, and enterprise software and real-time decision technologies within the context of e-commerce. Sodhi (2001) emphasizes the opportunity to further improve planning and execution by extending the decision horizon for planning within the enterprise, by broadening the physical scope beyond the enterprise to customers and suppliers, and by expanding the functional scope to include product design, marketing, and customer relationship management through the deployment of operations research techniques in Web-enabled supply chains. On the coordination side, the Web provides a virtually free platform for enhancing transparency, eliminating information delays and distortions, and significantly reducing transaction costs. In a striking example, Sivakumar (2004) describes the eChoupal initiative in India, which is projected to connect 25 million farmers in 100,000 villages to world markets by 2010, generating $2.5 billion worth of transactions. The deployment of Web Services and SOA principles constitutes an equally big step in enabling the IT infrastructure to evolve with changing business conditions, making them readily adaptable to changing business models, growing competition and globalization, tighter regulations, and increased mergers and acquisition activities. For companies, which have already invested millions of dollars in their enterprise systems, however, taking the next step to adopt SOA still requires significant additional investment. Companies should view this investment as a real option that would enable them to painlessly create agile supply ecosystems as the competitive landscape evolves. The quantification of the flexibility afforded by SOA should also provide a fruitful research area for the academic community. A recent example of such academic work is Hackney et al. (2006), who provide an evaluation framework for the adoption of Web Services. On the design side, current technology does not yet permit the mitigation of the trade-off between richness and reach in 3D-CE for the crucial challenge of dynamic supply chain design. More technology (i.e., higher levels of technology adoption), however, should not automatically be equated which better business performance. In a study of the adoption of the Quick Response (QR) program in the specialty retailing industry, Palmer and Markus (2000) found an association between adoption of QR (at a minimal level) and firm performance, and a high level alignment between the IT adopted and strategic goals. However, they could not find any support for the hypothesis that higher levels of QR adoption lead to better business performance or that a poor IT fit with strategy leads to poorer performance. The conclusion should therefore not be “more” technology, but “adequate” technology. As Davenport (1998) states, “putting an enterprise into an enterprise system – instead of the other way around – is clearly problematic, constraining, and thus lowering the performance of the organization as a whole.”

Impact of Information Technology on Supply Chain Management 147

6.6 References Akkermans HA, Bogerd P, Yücesan E, Van Wassenhove LN, (2003) The impact of ERP on supply chain management: Exploratory findings from a European Delphi study. European Journal of Operational Research 146: 284–301. Amit R, Zott C, (2001) Value creation in e-business. Strategic Management Journal 22: 493–520. Bailey JP, Bakos YJ, (1997) An exploratory study of the emerging role of the electronic intermediaries. International Journal of Electronic Commerce 1: 7–20. Clemons EK, Row MC, (1992) Information technology and industrial cooperation: The changing economics of coordination and ownership. Journal of MIS 9: 9–28. Currie WL, Parikh ML, (2006) Value creation in Web services: An integrative model. Journal of Strategic Information Systems 15: 153–174. Davenport T, (1998) Putting the enterprise into the enterprise system. Harvard Business Review Jul–Aug, 121–131. Dyer JH, Nobeoka K, (2000) Creating and managing a high-performance knowledgesharing network: The Toyota case. Strategic Management Journal 21: 345–367. Evans P, Wurster TS, (2000) Blown to Bits. HBS Press. Evgeniou T, (2002) Information integration and information strategies for adaptive enterprises. European Management Journal 20: 486–494. Fiala P, (2005) Information sharing in supply chains. Omega 33: 419–423. Fisher M, (1997) What is the right supply chain for your product? Harvard Business Review March–April, 105–116. Fine CH, (1998) Clockspeed: Winning Industry Control in the Age of Temporary Advantage. Perseus Books. Garvin DA, (1997) DEC: The endpoint model (A), Harvard Business School Case 9-688059 Gulati R, Nohria N, Zaheer A, (2000) Strategic networks. Strategic Management Journal 21: 203–215. Hackney R, Xu H, Ranchhod A, (2006) Evaluating Web Services: Toward a framework for emergent contexts. European Journal of Operational Research 173: 1161–1174. Johnson ME, Whang S, (2002) e-Business and supply chain management: An overview and framework. Production and Operations Management 11: 413–423. Kaplan S, Shawney M, (2000) e-Hubs: The new B2B marketplaces. Harvard Business Review May–June, 97–103. Lee, H.L., P. Padmanabhan, and S. Whang. 1997. The bullwhip effect in supply chains. Sloan Management Review, Spring, 93–102. Lee HL, Whang S, (2002) Supply Chain Integration over the Internet in: Supply Chain Management: Models, Applications, and Research Directions (Geunes, Pardalos, and Romeijin, eds.). Kluwer Academic. Malone, TW, Yates J, Benjamin RI, (1987) Electronic markets and electronic hierarchies. Communications of the ACM 30: 484–497. McAfee A, (2002) The impact of enterprise information technology adoption on operational performance: An empirical investigation. Production and Operations Management 11: 33–53. McAfee RP, MacMillan P, (1987) Auctions and bidding. Journal of Economic Literature 25: 699–738. McGovern J, Sims O, Jain A, Little M, (2006) Enterprise Service Oriented Architectures: Concepts, Challenges, Recommendations. Springer. Dordrecht, The Netherlands.

148 Enver Yücesan

Moitra D, Ganesh J, (2005) Web services and flexible business process: Toward the adaptive enterprise. Information and Management 42: 921–933. Palmer JW, Markus ML, (2000) The performance impacts of Quick Response and strategic alignment in specialty retailing. Information Systems Research 11: 241–259. Randall T, Netessine S, Rudi N, (2006) An empirical examination of the decision to invest in fulfillment capabilities: A study of Internet retailers. Management Science 52: 567– 580. Sivakumar S, (2004) eChoupal Experience Sharing: Workshop on ICT for Poverty Alleviation in India – Financing Models and Scaling up Opportunities. www.iimahd.ernet.in/egov/ifip/april2004/eChoupal.pdf [accessed on July 10, 2006] Sodhi MS, (2001) Applications and opportunities for operations research in Internet-enabled supply chains and electronic markets. Interfaces 31: 56–69. Stoll, (1986) Design for manufacture: An overview. Applied Mechanical Review 39:9 Swaminathan JM, Tayur SR, (2003) Models for supply chains in e-business. Management Science 49: 1387–1406. Tsay A, Nahmias S, Agrawal N, (1999) Modeling supply chain contracts: A review. In: Quantitative Models for Supply Chain Management (Tayur, Ganeshan, and Magazine, Eds.), Kluwer Academic. Boston. 299–336. Upton DM, McAfee A, (1996) The real virtual factory. Harvard Business Review July– August, 123–133. Wang CX, Benaroch M, (2004) Supply chain coordination in supplier centric B2B electronic markets. International Journal of Production Economics 92: 113–124.

7 An Agent-based Approach to Enhance Supply Chain Agility in a Heterogeneous Environment Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Abstract:

Businesses are undergoing a major paradigm shift, moving from traditional management into a world of agile organizations and processes. An agile enterprise should be able to respond rapidly to market changes. For this reason, enterprises have been seeking to develop numerous information technology (IT) systems to assist with their business processes. However, by their very nature, information/knowledge in the systems is disparate and heterogeneous and can be represented in various ways (text, pdf, html, etc.) and can be either structured or unstructured. It is, therefore, difficult to acquire, organize, or distribute information/knowledge using only traditional information technology methods such as e-mail or file servers. Because of the autonomous and collaborative aspects inherent in agent-based technology, this may be a possible solution to the problem of heterogeneity. Agent technology radically alters not only the way in which computers interact, but also the way complex processes, e.g., supply chains are conceptualized and built. This chapter proposes an agent-based system and formulates agile interaction for the collaboration of supply chain entities in a heterogeneous environment. The agent-based system enhances supply chain agility through the approach in which the agents autonomously plan and pursue their objectives and subgoals to cooperate, coordinate, and negotiate with others and to respond flexibly and intelligently to dynamic and unpredictable situations. Annotation and articulation mechanisms are developed to address and solve the heterogeneity problem of information/knowledge resources.

7.1 Introduction Supply chain managements (SCM) have continuously focused on change and innovation to maintain a competitive advantage over their rivals. However, the rate of change in the current environment is more rapid than that at any previous

150 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

time. The new paradigm known as “agility” is being promoted as the solution for maintaining competitive leadership in this new environment (Goldman and Nagel 1993; Kidd 1994). Agility means using market knowledge and a virtual corporation to exploit profitable changes in a volatile marketplace (Naylor et al. 1999). By becoming agile, a supply chain (SC) is positioning itself for long-term profitability due to its innate ability to excel in a changing environment (Dove 1994a). The ability to respond rapidly to changing market opportunities by using agile business processes and information technology is a key attribute of an agile SC. However, establish and maintain long-term competitive leadership, manufacturing enterprises in a SC are presented with the task of creating agile business processes, e.g., agile supply chain processes. These enterprises require methods and tools to help configure these processes and evaluate the performance using agility enablers (Devanna and Tichy 1990; Burgess 1994; Dove 1994b, 1995; Goldman et al. 1995; LaBarre 1995; Meade and Sarkis, 1999). In this chapter, a supply chain is viewed as a distributed network of entities interacting to deliver products or services to the end customer, linking flows from raw material supply to final delivery (see Figure 7.1) (Ellram 1991; Lee and Ng 1997). The agile interaction is focused and formulated to increase the collaboration of supply chain entities and to avoid the nature of fragment. In multientities (enterprises), interoperability with heterogeneous knowledge sources expands real-time efficiency and remains an open problem in SCM. The problems of interoperability between interacting information systems have been well documented (Cui et al. 2001). Sheth (1998) classifies different kinds of interoperability problems including systemic, syntactic, structural, and semantic levels of heterogeneity. Most of the structure of the information/data is not evident to a robot browsing the Web (Berners-Lee 1998), and data/information’s syntax and semantics are still handled in different ways, which leads to interoperability problems. For example, a business in a SC adapting electronic data interchange (EDI), portal systems, ebXML (URL 0; Hofreiter et al. 2002; Cole and Milosevic; 2001, Han et al., 2001), BizTalk (URL 1; He and Milosevic 2001; Herring and Milosevic 2001), e-cl@ss (URL 2; Leukel et al. 2002), UNSPSC (URL 3; Leukel, et al. 2002), Open Applications Group (URL 4), and RosettaNet (URL 5; Bussler 2002; Huhns and Stephens 2001) has difficulty in solving the problem of semantic interoperability (Berners-Lee 1998). These specifications can be used successfully to specify an agreed set of labels with which to exchange product information, it is wrong to assume that these solutions also solve the problems of semantic heterogeneity (Cui et al. 2001). In addition, due to information overload, achieving semantic interoperability between a knowledge source and a knowledge receiver (e.g., a decision-maker, application, or peer agent) is more critical than ever. Only a few papers focus on the interoperability problem of knowledge sharing, i.e., dealing with heterogeneous sources.

An Agent-based Approach to Enhance Supply Chain Agility 151 Direction of flow of demand

$

Direction of flow of product

Customers

$

Retailers

Tier-II Supplier

$

Customers

$

Customers Retailers

Tier-II Supplier

Customers

Manufacturers $

$

Tier-II Supplier

Retailers

Manufacturers

Customers $

$

Retailers

Tier-II Supplier

Customers $

$

Raw Material Vendor

Tier-II Suppliers

Tier-I Suppliers

Manufacturers

Distribution Centers

Retailers

Customer Zones

Figure 7.1. Supply chain viewed as network related architecture

To enhance agility in a heterogenous envirnment, a supply chain of enterprises should be able to respond rapidly to market changes while IT plays a crucial role. However, it is frequently difficult to control “information/knowledge resources,” because of characteristics such as invisibility, changeability, and nonlinearity, as well as systemic, syntactic, structural, and semantic levels of heterogeneity (Corby and Dieng 1996; Satyadas et al. 2001; Sheth 1998). Therefore, it is important for business to address the problem of how to transfer, manage, and share information/knowledge effectively in heterogeneous systems of a supply chain. Three main heterogeneity problems are involved in the deployment of information/knowledge management in a supply chain. First, in the interorganizational business environment, each organization develops its own ontology. The semantic interoperability of various ontologies between two or among more different business entities is required. Second, in most organizations, knowledge is held among many individuals, departments, and data stores. It is difficult to access, share, and distribute knowledge gathered from different sources in a coordinated fashion. Third, the heterogeneous issues of information/knowledge sources also present a challenge in business. Information/knowledge is produced and dispersed daily throughout an organization in the form of business or technical documents, information manuals, legacy databases, e-mails, etc. Specifically, they are represented in various formats (text, pdf, html, etc.) and can be structured, semistructured, or unstructured in different businesses. In addition, supply chains of general industry have a long tradition of collaborative working between members. To aid these actions, the ITs that are currently available have been used. These efforts have yielded some success but are hampered by the problems posed by the use of heterogeneous ITs and the lack of effective collaboration between ITs. In particular, there are very few ITs

152 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

available to support distributed asynchronous collaboration except for distributed artificial intelligence (Anumba 2002). The commonly implemented distributed artificial intelligence, which is usually in the form of intelligent agents, offers considerable potential for the development of such ITs. An agent is an encapsulated computer system that is located in some environment and is capable of flexible and autonomous action in that environment to meet its design objectives (Jennings 1995; Jennings et al. 2000). Wide application domains in which an agent solution is being applied or investigated include electronic commerce (Karacapilidis and Moraïtis 2001), a genetic agent-based model (Choi et al. 2001), supply chain management (Kaihara 2001), distributed discrete-event simulation techniques (Logan and Theodoropoulos 2001), software engineering (Bresciani et al. 2001), interfaces (Descamps and Ishizuka 2001), computer-supported projectoriented production planning (Mayr et al. 2001), artificial intelligence (Ferber 1999), and knowledge-based systems and expert system (Luo et al. 2001). The agent technology is increasingly used in a wide range of industrial and commercial domains. To satisfy the set of feature requirements and allow agents as the strongest solution candidate, the SCM problem in a heterogeneous environment should be addressed by a team of specialists or intelligent agents – (1) distributed object systems have the necessary encapsulation, but not the sophisticated reasoning required for social interaction or proactive behavior; and (2) distributed processing systems deal with the distributed aspect of the domain but not with the autonomous nature of the entities. Agent technology radically alters the way in which computers interact and also the way to SC entities. Furthermore, collaboration among agents for streamlining and integrating the entire process of a supply chain is desired (Ito and Salleh 2000). The literature of agent technology applied in the supply chain area is numerous, for example Wu (2001), Gerber et al. (2001), Fischer et al. (1999), Ivezic et al. (2000), Yung et al. (1999), Pathak et al. (2000), Felfernig et al. (2000), and Ito and Salleh (2000). Swaminathan et al. (1998) describe supply chain models composed of software components that represent the types of supply chain agents. However, few previous papers focus on the enhancement of agility, the trend paradigm. The aforementioned studies did not provide a whole picture of agent-based SCM crossing all functions and proposed a global architecture for agents, specifically focusing on enterprise agility. Furthermore, they do not consider the heterogeneity problems while the agent technology is applied. The objective of this chapter is to develop an agent-based system to enhance supply chain agility and solve heterogeneity related problems. The agents in the system autonomously plan and pursue their actions and subgoals to cooperate, coordinate, and negotiate with others; and to respond flexibly and intelligently to dynamic and unpredictable situations in a virtual way. Simulation results provide adaptation to the operational parameters and the structure of the supply chain to manage the supply chain agilely in a fast-changing environment. The remainder of this chapter describes the work undertaken to conceptualize supply chain activities and interaction with a collection of intelligent agents. The intelligent supply chain agent system is introduced in Section 7.2. Section 7.3 classifies intelligent supply chain agents. Section 7.4 presents the way the agent

An Agent-based Approach to Enhance Supply Chain Agility 153

handles an incoming message. Section 7.5 proposes the basic agent architecture. Section 7.6 illustrates a detailed quantitative measure with simulation and depicts an application in industry. The implementation procedure is practically recommended. Section 7.7 concludes the paper.

7.2 Intelligent Supply Chain Agent System The approach in this work is to use multiagents, called an intelligent supply chain agent (ISCA) system for modeling and analyzing supply chains. Different types of agents are defined and each agent can use a set of control elements. The control elements help in decision-making on structural problems by the agent by using various policies (derived from analytical models such as inventory policies, just-intime release, sales, and routing algorithms) for demand, supply, information, and materials control among the supply chains. To deal with nonstructural problems, problem-solving agents using decision-making resources in the agency are developed. The basic unit of the ISCA system comprises an agent and the services under its management. Each of these units would be able to negotiate with any other agent when providing a service. Furthermore, an agent could negotiate with any other agent to enhance its services or create new services. A service corresponds to some problem-solving activities of the supply chain. The simplest service represents a problem-solving atomic activity in the ISCA system, e.g., transport a batch of products from location A to B. These atomic activities can be combined to form complex services, e.g., distribution of products by adding ordering constraints. The nesting of services can be arbitrarily complex and at the topmost level, the entire supply chain process can ultimately be viewed as a service. A grouping of services is known collectively as an agency. Service requirements are issued either from other departments, e.g., market teams through an Intranet, or from external suppliers or customers through the Internet (see Figure 7.2). Services are associated with one or more agents that are responsible for the management and execution of those services. Each service is managed by an agent, although the execution of its subservices may involve a number of other agents. Since agents are autonomous, there are no dominating/dominated dependencies between them. Therefore, if an agent requires a service, which is managed by another agent, it cannot simply instruct that agent to render the service. Rather, the agents must come to a mutually acceptable agreement about the terms and conditions under which the desired service will be performed. The mechanism for making agreements is negotiation - a joint decision-making process in which the parties verbalize their (possibly contradictory) demands and then move toward the agreement by a process of concession or search for new alternatives.

154 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Figure 7.2. The environment of an agent-based system

In this organizational SC model, servant agents reside in an agency. These servant agents are loosely coupled to each other but tightly coupled to the dominating agent. The dominating agent provides access to the world outside its agency. Agents within an agency may negotiate only with external agents and a servant agent. A dominating agent will normally be a loosely coupled agent in a higher level agency. Agents in an agency will usually be the dominating agents of lower level agencies. This leads to a hierarchical organization of agencies reflecting the logical structure.

7.3 Classification of Intelligent Supply Chain Agents Supply chain dynamics is complicated to model due to the presence of heterogeneous entities, multiple performance measures, and complex interaction effects. The variety of supply chains poses a limitation on reusability of processes across them (Tan 2001). To capture the dynamics of supply chains across a wide spectrum, a classification of agents is presented first. Swaminathan et al. (1998) classify the agents in supply chains into two broad categories: structural agents and control agents. Structural agents are involved in actual production and transportation of products, whereas control agents help in coordinating the flow of products in an efficient manner by using messages. Structural agents are further classified into two basic sets of agents, production and transportation agents. Control agents are classified into inventory control, demand control, supply control, flow control, and information control agents.

An Agent-based Approach to Enhance Supply Chain Agility 155

Extended to Swaminathan et al.’s (1998) classification, ISC agents, which enable the modeling and analysis of a large variety of SC activities, include additional problem-solving agents (see Figure 7.3). These problem-solving agents deal with semistructural or nonstructural problems in decision-making, e.g., planning, monitoring, controling, and problem-solving. Despite the three types of agents, information and interface agents are used as an information/knowledge center to integrate and transfer all information/knowledge and play all roles in a supply chain to solve heterogeneity issues.

Figure 7.3. Structual, control, and problem-solving agents

7.3.1 Information and Interface Agents Information and interface control agents are essential for coordination within a supply chain. Two types of information flow are considered: x x

Directly accessible: Directly accessible information transfer refers to the instantaneous propagation of information. Periodic: Periodic information updates may be sent by different production and transportation agents to indicate changes in business strategy, price increases, introduction of new services or features in products, and the introduction of new production agents, etc.

To solve heterogeneity issues, annotation and articulation mechanisms are developed in the information and interface agent. Annotation mehanism The annotation mechanism provides an annotation process which relates to the unstructured formats of knowledge documents and allows knowledge resources to be accessed efficiently so that these heteroformatted or unstructured knowledge documents can be retrieved. The annotation includes three types of descriptions – basic description of annotated knowledge (AnB), descriptive information of

156 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

annotation (AnD), and relationship description (AnR) among the documents, ontology, and generated knowledge. AnB includes general statements about the knowledge document and its source. AnD presents the 5W1H (What, Where, Who, When, Why, and How) to represent annotated knowledge. The machine-understandable annotated metadata in this chapter are based on Zachman’s framework (Inmon et al. 1997). The Zachman framework provides a systematic representation for unstructured knowledge in organizations. The use of the 5W1H model (i) allows the representation of knowledge to be consistent and flexible, and (ii) improves the sharing of knowledge in a meaningful manner over the Web (Huang and Kuo 2003; Inmon et al. 1997).The Zachman framework defines the six dimensions of a problem as x x x x x x

entities (what? things of interest) activities (how? the method) locations (where? places of interest) individuals (who? individuals and organizations of interest) times (when? things occur) motivations (why? reasons and rules)

5W1H is able to present these six dimensions in relation to a particular event. AnR presents two types of relationship – (i) the relationship between the annotated knowledge document and other heterogeneous /unstructured documents and (ii) the relationship between the annotated knowledge and ontology. The proposed annotation process is presented in Figure 7.4. The knowledge documents in heterogeneous formats are annotated and stored as follows (Heflin and Hendler 2000). Step 0: Unstructured knowledge which may be presented in various formats (doc, sound, or graphic files) is prepared as input to the annotation process. Step 1: The information and interface agent and the knowledgeable worker extract the knowledge contents from unstructured or heterogeneous knowledge documents. The agent is responsible for inserting XML-description tags into the knowledge contents. The XML-based metadata for these knowledge contents are then generated. Step 2: The agent is responsible for identifying the metadata of knowledge contents as (AnB, AnD, AnR) and for transforming them into XML format documents. The annotated knowledge (KA), represented by the 5W1H model, is generated. A conception schema is also created in this step. A conception schema is formed in an XML file, which describes the structure of the concepts and the properties of the annotated knowledge.

An Agent-based Approach to Enhance Supply Chain Agility 157

Figure 7.4. Agent-based annotation process

Step 3: The agent stores the annotated knowledge document together with the conception schema in the metadata repository. The agent registers the link (to the original document) in the register. The metadata repository validates the consistency between the elements in the benchmark ontology and the metadata of the original source documents. The registry validates the consistency of the links between the annotated knowledge document and the original knowledge source. Using the annotation process with the agent, the heterogeneity problem of knowledge sources in different formats and structure levels is solved since the (AnB, AnD, AnR) of unstructured or semistructured documents are annotated and become machine-readable and understandable by experts. Articulation mechanism Articulation is a mechanism for supporting the interoperability of various sources, and solves the problems of semantic interoperability. The articulation mechanism proposed in this section (i) efficiently resolves the heterogeneity of ontologies used by various business entities; and (ii) appropriately intersects, unifies, and differentiates heterogeneous ontologies with the benchmark ontology, according to purpose - for example, “intersection” for clarifying the interoperating part of the two ontologies and “difference” for validating articulation rules, respectively. Figure 7.5 depicts the structure of the articulation mechanism and the interaction among the three parts-ontologies, the articulation generator, and the knowledgesharing platform. This mechanism recognizes the need for heterogeneous ontologies, generates articulation rules, supports semantic interoperation, and aids the knowledge-sharing platform to query desired information/knowledge.

158 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Figure 7.5. Components of articulation mechanisms

In the mechanism, the graphical conceptual model with articulation rules (dotted lines) is applied to integrate two different ontologies. In the graphical conceptual model, the label of an edge, whose links cross two conceptual graphs, corresponds to the name of a semantic relationship between the concepts. The label may be null if the relationship is unknown. In the graphic conceptual model, the semantic relationships “SubclassOf,”, “PropertyOf,”, “InstanceOf,”, “ValueOf”, and “S”emantic “I”mplication are represented by edges, labeled “S,”, “P,” “I,”, “V”, and “SI,” respectively. The individual ontology may contain other binary relationships between concepts other than the above-mentioned relationships. The set of logical rules R (including articulation rules), associated with ontology, are expressed in a logic-based language. An articulation rule is a statement, (for example, “match concept A in ontology 1” and “concept B in ontology 2”), which expresses matching between two equivalent concepts. The matching relationship further indicates the semantic relationship between the two concepts, which is also called the semantic implication (SI) (Mitra et al. 2000). For example, the articulation rule (match “concept A in ontology 1’ and “concept B in ontology 2”) is expressed by (A, “SIBridge,” B), which means, “concept A in otology 1 semantically belongs to concept B in ontology 2”. The articulation rules are generated by a semiautomatic articulation method. The semiautomatic articulation method (Mitra et al. 2000) uses an automatic articulation generator to determine whether the match between the concepts of the two ontologies is satisfactory. The articulation generator is a modular agent, which applies a hybrid heuristic method and linguistic matcher algorithm (Mitra and Wiederhold 2002) to generate the scores of semantic similarity between the

An Agent-based Approach to Enhance Supply Chain Agility 159

concepts of two ontologies. A human expert, knowledgeable about the semantics of concepts in both ontologies, validates the matches generated by the articulation generator using a graphic user interface tool. The expert may reject a generated match if it is irrelevant to the supply chain domain. The expert can also add new matches that the articulation generator might have missed. The articulation is processed iteratively until all of the matched nodes in the graphical conceptual model have been checked, and knowledge workers, as well as experts, are satisfied. In summary, a heterogeneous information/knowledge document is generated according to a heterogeneous ontology. The information and interface agent annotates this heterogeneous document using articulation rules. The experts’ feedback on, and updates of, the articulation are stored. The feedback is supported by a mechanism to enable further articulation, on which other users impose this heterogeneous ontology. This learning process improves the quality of the articulations. Since the conceptual model is simple and the articulation architecture is modular, great scalability can be achieved with few problems. 7.3.2 Problem-solving Agent Problem-solving agents deal with nonstructured problems, which have no structured phases of intelligence, choice, and decision (Turban and Aronson 2001), and must be solved with both standard solution procedures and human judgment (see Figure 7.6). Problems Other computerbased systems

Internet, intranets, extranets

In ormation Inter acing agent

Structure problems Semi- or Un-structure problems Database

Procuremen t mechanism Dispatching mechanism Logistic mechanism

Structure event agent

Decision ma ing agent

Data managemen t agent

Other mechanism

Database External models agent

no ledgemanagement agent

Decision mechanism

Scheduling mechanism Expediting mechanism

Optimi ation agent

Database Organi ational no ledge ase odel-base

Figure 7.6. Problem-solving agents

imulation agent

Rule bas e

160 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Data management agent: Manages all data related to decision-making, information, materials and finance in a supply chain. According to all conditions in the supply chain, the agent updates the database on a real-time basis. Simulation agent: The simulation agent possesses the following three objectives: 1. Incorporate the decision-making functions of manufacturing systems in a simulation in the form of intelligent agents; 2. Represent the hierarchical structure (net) of supply chain systems within a simulation; 3. Provide facilities for intelligent agents to receive information from the simulation at different levels of detail. By using the concept of intelligent simulation multiagent tools (ISMAT) (Gajanana and John 1993), the simulation agent has the following modules: 1. Intelligent-agent description: The main focus of IMSAT is the development of this module. It defines a rule language to implement the decision-making rules for intelligent simulation agents. 2. Hierarchical model specification: This module defines classes to represent the structure of a manufacturing system in IMSAT simulations. 3. Product-flow definition and abstraction-mechanism specification: The product flow definition module implements a simulation kernel called the object-oriented manufacturing simulation language (OMSL). It provides the class definitions necessary to model machines, queues, stores, and manufacturing-process plans. The abstraction mechanism defined in this module provides a way of summarizing the process information for use by intelligent agents. 4. Simulation management: The simulation management module maintains the simulation calendar. It also interacts with the knowledge management agent for running simulation models. The sets of agents defined above along with the customer agent that generates the demand for the system constitute our framework. The roles of agents are clearly defined, and their interaction processes are also clarified. The well-defined roles and standardized interaction processes can support the supply chain activities in organizations and also let the agents know which agent it should communicate with and what message it will receive or send. In this way, the agile collaboration of SC entities is enhanced.

7.4 Message Handling Part of the agent architecture and the handling process to handle incoming message is illustrated in Figure 7.7. An incoming message, which is sent from an external source (e.g., Web application or Intranet application, or other agent) is accepted by the message-input processor of the agent. The action is to be exercised only after the incoming message has been interpreted by the services-message sense processor and the contents of the message been recognized by the knowledge representation mechanism incorporated with the knowledge base of the agent.

An Agent-based Approach to Enhance Supply Chain Agility 161

Corresponding to each incoming message, the contents may have been transformed and described with the XML-based knowledge representation language, e.g., KQML (knowledge query manipulation language) or DARPA (defense advanced research projects agency) agent markup language + ontology inference layer (DAML + OIL) (Hendler and McGuinness, 2000) and the contents recognized. In this case, they are stored in the knowledge base. In the other case, the message contents are not in this format and cannot be recognized. Through (i) interaction with Internet Web Services, (ii) filtration by the resource gate, (iii) the inference of ontology, and (iv) transformation of the message contents, they have to be incorporated with the knowledge base in this agent and also the knowledge base of external agents and then are stored in the knowledge base and wait for processing. The knowledge base in each agent includes the event-handling knowledge base and the domain knowledge base. The event-handling knowledge base stores the knowledge related to the way this agent handles incoming messages. The domain knowledge base stores the knowledge related to how to interact with other agents while handling incoming messages (e.g., pinpointing the operation schedule, interaction of firing sequence of agents, etc…). Both knowledge bases support XML, format output (XML vocabulary) and can store numerous details of previous events. Functional mechanism

Event selection mechanism Priority selection function

Other agents

Agency

Goal process mechanism Scheduling Goal negotiator

Situation Dispatching Current Stituation control Update Situation control

Coordination Allocating

Message input processor

Ξ Ξ

Response manipulation

Incoming message

Response message Message transfer

Message output processor

Output message

Ontology Inference and mapping

ServicesMessage sense processor

˜́˹̂̅̀˴̇˼̂́ʳ˹̂̅̀˴̇ʳ ˷˸̃˿̂̌̀˸́̇ʳʹʳ̃˴̅̆˸̅

Event-handling Knowledge base Knowledge representation mechanism(DAML)

DAML->RDF Schema translator

Domain Knowledge base support XML

Term-Conceptual relational network Term A Explorer mechanism

AgentsϗKnowledge (XML Vocabulary)

˧˸̅̀ʳ˖

˧˸̅̀ʳ˕

˧˸̅̀ʳ˗

˧˸̅̀ʳ˘ Resource Gate Resources of Internet based Web services & SOAP tech

˧˸̅̀ʳ˚

˧˸̅̀ʳ˙

Figure 7.7. Part of agent architecture corresponding to message handling processing

162 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

While the new transformed XML-based incoming message is retrieved from the knowledge base and is processed, the agent retrieves relevant “knowledge resource” from the event-handling knowledge base, and defines a commonly used set of vocabulary with the XML-based knowledge representation language. In this way, one can describe the incoming message, including statement, function, constant of the contained knowledge, and the 5W1H framework. Second, the collected knowledge resources are aggregated and the vocabulary is extended and widened; in this approach, a logic statement and axiom, which are accepted by all agents and relevant application resources are created. Third, the knowledge base is updated and the ontology is formed while the domain knowledge is reorganized corresponding to each incoming message. Fourth, the inference of the contents in the new incoming message is made through the mapping and matching of the ontology map, which is produced from various knowledge bases in various agents. Next, through the inferences, the facts and a solution approach to the problem that the agent needs to solve, are recognized and planned. At last, the plan is coordinated and disseminated through the message output processor. Because of the functionality of a semantic Web, the incoming message can be interpreted and recognized clearly, agilely, and efficiently. In case the incoming jobs are multiple, the required operations are as follows: The selection of incoming messages corresponding to these jobs is made by the priority selection function of the event selection mechanism, which will be discussed later. Then the incorporation of the message-input processor with the agent negotiator (the gate of goal process mechanism) needs to negotiate and communicate with other agents, which is based on the results generated from the goal process mechanism, including scheduling jobs, dispatching resources, and coordinating plans. In this architecture, operation of each agent and the global goal in the multiagent environment are optimized because this designated agent architecture can: (1) negotiate with other agents, (2) update the agency through the Explorer Mechanism and situation control mechanism, (3) prioritize the processing messages, (4) map and match by the ontology map of the knowledge base to filter and retrieve relevant information and knowledge, and (5) recognize the facts and solution approach to solve the problems which the agent faces. Based on the command of the goal process mechanism, the response manipulation mechanism informs the functional mechanism about the way to implement and send the output message to the message output processor. The message output processor formulates the output messages and then sends them to the external resource or another agent. In the whole process, the agent updates the message output processor and sends updated (updated verification) messages to the external agency concurrently. In this way, the agent is able to respond and achieve the objectives of real-time and synchronization. The contents of a message are represented with the six dimensions of the Zachman framework and coded with XML (Huang et al. 2002). The content in each message is constructed by the six dimensions (5W1H) of the Zachman framework, including solution approach information, domain knowledge and know-how, and has an impact on agents. The Zachman framework provides a

An Agent-based Approach to Enhance Supply Chain Agility 163

systematic approach to externalizing unstructured contents in messages of agents, although not all contents can be represented in this format. Designing the knowledge structure of messages is critical for message interpretation, and a standard storage system provides a place for the messages to reside. Through the exchange platform of ontology, the knowledge representation and inference language are used and incorporated with the RDF and RDFS to communicate and share the knowledge. The RDF and RDFS language are based on XML. Therefore, the representation language is transformed as XML-based format, and knowledge can be easily presented. It also aims to offer quick transmission of the information communication and exchange. Moreover, XML shares many semi-structured features. For example, its structure can be irregular, is not always known in advance, and may change frequently and without any notice. Therefore, this chapter delineates the knowledge in each message using XML because XML-based documents are easy to store and manage. Figure 7.8 illustrates an XML-based documentation of contents for the knowledge in each message, which is constructed by the Zachman framework.

Figure 7.8. An example of an XML-based documentation representing knowledge

7.5 Basic Agent Architecture Agent descriptions provide the ability to specify both static and dynamic characteristics of various supply chain entities. Each agent is specialized according

164 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

to its intended role in the supply chain. All ISCAs have the same basic architecture (see Figure 7.9). This involves an agent body that is responsible for managing the agent's activities and interacting with peers and an agency that represents the solution resources for problems of the supply chain, including (1) other agents, (2) services, and (3) an agent repository: Set of attributes, knowledge about other agents, interaction constraints, performance measures, control agents, message processing semantics, and selector with control policy. The body has four basic mechanisms responsible for each of its main activities: an Event selection mechanism, a functional mechanism, a negotiation mechanism, and an explorer mechanism. This internal architecture is broadly based on the Grate (Jennings et al. 1996a) and Archon (Jennings 1996b) agent models. The agency includes an agent repository and services and also other ISCAs. The latter allows constructing a nested (hierarchical) agent system in which higherlevel agents realize their functionality through lower level agents (the lower level agents have the same structure as the higher level agents and, can, therefore, have subagents as well as selected jobs in their agency). For example, the higher level agent may represent a manufacturing department whose work is carried out by a number of manufacturing teams (the lower level agents). This structure enables modeling flat, hierarchical, and hybrid organizations in a single framework. The artitecture includes the following components: Peer Agents Agent

Agent

Agent

Agent

Information sharing and communication Incoming message

Agent body Other agentsϗ Knowledg e base

Event Selection mechanism

Access

Schedule & monitor

Exception Handle

Handle

Functional mechanism Status message

Access Resource Gate

Retrieval

Explorer mechanism

Extrrnal objects

Negotiation mechanism

Maintain & deliver

Execution & presentation

Resources of Internet based Web services & SOAP tech

Agency Repository

Agent

Domain Knowledge base ˞́̂̊˿˸˷˺˸ʳ ˵˴̆˸ʳ

Figure 7.9. The ISCA Architecture

Service

An Agent-based Approach to Enhance Supply Chain Agility 165

Event selection mechanism: According to the aforementioned knowledge base, ontology, and message handling process, incoming messages are selected based on the rule of event selection mechanism such as first priority, first served (FPFS). Each message type has a message handler or a script that determines how the message will be processed. The message handler is characterized by the control policies that are used by the agent. Next, the mechanism schedules the selected job operations responsible for assessing and monitoring the agent's ability to meet (i) the service agreement that is already agreed upon and (ii) the potential service agreement that it may agree to in the future. This involves two main roles: Scheduling and exception-handling. The former involves maintaining a record of the availability of the agent's resources, which can then be used to determine whether the service agreement can be met or a new service agreement can be accepted. The exception-handler receives exception reports from the functional mechanism during service execution (e.g., “service may fail,” “service has failed,” or “no customer agreement in place”) and decides upon the appropriate response. For example, if a service is delayed, then the mechanism may decide to locally reschedule it, to renegotiate its service agreement, or to terminate it altogether. Functional mechanism: Each type of agent has the specified functionality to complete the jobs. The functionalities of partial agents are summarized in Table 7.1). Table 7.1. Function of each type of agent Agent name

Function Description

Structural agents Manufacturing plant agent The main focus here is on optimal procurement of components (particularly common components) and on efficient management of inventory and manufacturing processes. Supply agent

Updates the orders and decides the delivery time corresponding to the current status of supply-demand relationship in the market. Updates the orders of raw materials and determines the delivery time based on the status of the raw material supply.

Distribution center agent

Reduces the inventory carried and maximizes throughput. In a standard distribution center, products come in from the manufacturing or supplier plants.

Retailer agent Transportation agent Marketing agent

The main focus here is on reducing the cycle time for the delivery of a customer order and minimizing stock-outs Move the products from one production agent to another. Interacts with retailers and consumers through internet and provides a mechanism that can trigger additional demand for products in seasonal, random, or permanent in numerous ways, including advertisements, discounts, coupons, and seasonal sales.

External supply agent

Focuses on low turn-around time and inventory. Their operation is characterized by the supplier contracts which determine the lead time, flexibility arrangements, cost-sharing, and information-sharing with customers. Responsible negotiating a reasonable price for raw materials.

Control agents Production control agent

Maximizes the production efficiency and minimizes the throughput and delay of order delivery incorporated with other agents. In addition to maximizing production efficiency, the agent cooperates different mechanisms from other agents. For example, the agent shortens the duration between placing order(s) and shipping product(s) and controls the delay time of the order(s) efficiently.

Inventory control agent

Minimizes the inventory and response time of demand; conveys the delivery time to customers. Controls overinventory as much as possible and minimizes the response time projected for the purchase order. Determines the possibilities for prompt delivery if the inventory is sufficient; otherwise, determines the next available delivery time.

Supply control agent

Supply control agents dictate terms and conditions for delivery of the material once orders have been placed. Supply control agents support the desired context of the orders and delivery time based on the determination of a particular order from supply agent. Supply control agent provides the optimal terms and delivery time and condition based on the determination of a particular order from supply agent.

Flow control agent

Coordinates flow of products between production and transportation elements.

Problem-solving agents Data management agent

Updates the data, including material, cash, information flows, and the data about the interaction among SC entities; supports the updated information to the decision-making agent. Promptly updates all records from the physical agents (e.g., material flows, cash flows, information flow and any records of interaction) in the supply chain. Instantly provides the latest information to the data management agent.

Simulation agent

The simulation agent is exemplified for the agent-based system and the simulation results are presented to demonstrate the feasibility of the ISCA system.

166 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

For example, the marketing agent provides a mechanism that can trigger additional demand for products in seasonal, random, or permanent in numerous ways, including advertisements, discounts, coupons, and seasonal sales. Three main roles such as service execution management (optimizing executed jobs as specified by the agent's service agreements), solution presentation (routing solutions between servers, clients, and other agents), and exception-handling (monitor the execution of jobs and services for unexpected events and then react appropriately) are involved. Negotiation mechanism: An agent cannot simply instruct other agents to start the service. Rather, the agents must come to a mutually acceptable agreement about the terms and conditions under which the desired service will be performed. The negotiation mechanism makes agreements by a joint decision-making process in which the parties verbalize their (possibly contradictory) demands and then move toward agreement by a process of concession or search for new alternatives. The mechanism 1. 2.

maintains the agent repository where related information is stored, delivers the status messages of active services between the functional mechanism and the clients, between an agent and its agency, and between peer agents, and 3. communicates x

between the functional mechanism and clients within the agency related to job management activities (e.g., activate, suspend, or resume a job), and x between agents within that agency or peer agents relating to service execution management (e.g., an instruction to start service, service finished, and service results). Explorer mechanism: The mechanism searches the resources that are located in other distributed databases and performs the role of managing, querying, or collating related information from many distributed sources. Based on the Web Services (http://www.w3.org/) and through the Universal Description Discovery Integration (UDDI) registry functionality (mechanism) and service description mechanism of Web Services Description Language (WSDL), relevant service and resources can be explored. The service and resource in need are retrieved through the resource gate and integrated into the external knowledge base, in which all agents share the knowledge. Such an explorer mechanism can traverse the WWW by using Web Services, gather information, and report what it retrieves to a home location.

7.6 A Case Study In this section, one case in supply chain application is studied and the improvements are presented to demonstrate the feasibility of the ISCA system. The proposed agent-based system is applied to a personal computer (PC) OEM

An Agent-based Approach to Enhance Supply Chain Agility 167

group. The supply chain under consideration had 11 different types of end PC products, 1200 different parts in the bill of material, and 200 inventory locations, including both internal divisions as well as external suppliers located worldwide. The introduction procedure of ISCA system is as follows (Shelly 1995): Step 1 System concept development. Define the scope or boundary of the ISCA concept. Includes systems boundary document, cost-benefit analysis, risk management plan, and feasibility study. Step 2 Planning. Develops a ISCA project management plan and other planning documents. Provide the basis for acquiring the resources needed to achieve a solution. Step 3 Requirements analysis. Analyze user needs and develop user requirements. Create a detailed functional requirements document. Step 4 Design. Transform detailed requirements into complete, detailed ISCA system design document. Focus on delivering the required functionality. Step 5 Development. Convert a design into a complete information system. Include acquiring and installing systems environment; creating and testing databases/preparing test case procedures; preparing test files; coding, compiling, refining programs; performing test readiness review and procurement activities. Step 5 Integration and Test. Demonstrate that the developed ISCA system conforms to requirements as specified in the functional requirements document, conducted by the Quality Assurance staff and users. Produce test analysis reports. Step 6 Implementation. Include implementation preparation, implementation of the ISCA system into a production environment, and resolution of problems identified in the integration and test phase. Step 7 Operations, Learning, and Maintenance. Describe tasks to operate and maintain information systems in a production environment. Include postimplementation and in-process reviews. At the learning stage, the information and interface control agents in Section 1.3 evaluates and verifies detailed quantitative measures associated with various quantitative alternatives before the strategic alternative is chosen for implementation. Consequently, the quantitative alternatives of agility enhancement can be prioritized with the analytical network approach (Meade and Sarkis 1999) by the problem-solving agent. Evaluation of alternative configurations provides the manager with insights into how changing the supply chain might affect the performance in agility. The ability to fine-tune the system and evaluate performance under different scenarios makes this application useful for evaluating short-term (e.g., setting inventory levels, changing inventory control policies) as well as long-term (e.g., changing a supplier, adding a distribution center) reengineering efforts. Moreover, it is impossible to have tractable analytical models for these problems under realistic assumptions. One might be interested in evaluating quantity alterations to the supply chain in various ways (such as using various operational parameters, e.g., some agility enablers in Meade and Sarkis 1999) to improve performance. The ability to modify the operational parameters and the structure of the supply chain and to evaluate the effect of these modifications is extremely useful in effectively, and agilely, managing the supply chain in a fast-changing environment. In the learning, by setting various operational parameters (some agility enablers), the global and local performance of the supply chain are observed

168 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

simultaneously. Empirical studies have shown that taking a global perspective may be sometimes harmful to some of the entities in the supply chain (Cash and Konsynski 1985; Swaminathan et al. 1998). Therefore, separating local performance measures (PMi) from global performance measures (GPM) is necessary. A global performance measure may be an appropriate yardstick for an intraorganizational supply chain (most of the entities belong to the same organization). However, local performance becomes an important measure for interorganizational supply chains. Note that there is a very strong link between the goals of the agent in terms of the performance measures PMi and priorities Qi. The learning process that is verified by the simulation is illustrated in Figure 7.10, where the optimization of agility is carried out with the policies in the inventory control agents, marketing agents, and forecast agents corresponding to inventory control, sales, and production planning, respectively. Based on the focuses of agents presented earlier, the data are collected and evaluated to determine the optimal quantitative alternative in the simulation. The focus of agents is presented as follows: Retailer agent: Reducing the cycle time for the delivery of a customer order and minimizing stock-outs. Distribution center agent: Reducing the inventory carried and maximizing throughput. Manufacturing plant agent: Optimizing procurement of components (particularly common components) and efficiently managing inventory and the manufacturing process. External supplier agent: Lowering turn-around time and inventory. Transportation vehicles agent: Maximizing resource utilization. Flow control Agents: Optimal routines. All agents: The performance measures PMi. BOM, Demand, Lead Time, Transportation Time, Supply Chain Network, Cost, Supplier Reliability

Simulation

Supply Chain Data

Customer

Agents determine the optimal quantitative alternative

Service Inventory Levels

Simulation

Fill Rates, Inventory Costs, Work-InProcess, Order Turnaround Time

Figure 7.10. The provision of learning

Optimization

An Agent-based Approach to Enhance Supply Chain Agility 169

Simulation under different starting conditions should be performed to obtain robust outputs. The simulation scenario is used while making those decisions by developing different simulation models with various operational parameters (agility enablers) for alternative configurations. A comparison of the performance of alternative configurations provides the manager with information about the expected benefit from each quantitative alternative. The manager would choose one among the various quantitative alternatives based on their estimated agility and their measured performance in the simulation. In addition to providing all the advantages of simulation, the user enables modeling a broader set of supply chain issues under reduced development time. After lauching a simulation of 6 months, simulation outcomes are analyzed to show the superiority of the proposed approach. In the simulation process, (i) The queues of all processes and the storage behind the processes are collected to show the inventory level, (ii) After each simulation run, the problem-solving agent adjusts the parameters of machine setup time. The cycle time is changed and learning is initiated to recognize the desired parameters. Consequently, the cycle time is collected to verify the reduction of lead time of machining without changing any process time and manufacturing procedures, but setup time only through learning, (iii) The surveys are collected to show the customer satisfaction level on the B2B Web site after each service is completed. The poll is provided to encourage customers to feedback their feelings to the service, (iv) According to each order, a parameter of throughput is collected at the end of each process, (v) The log file collects each service duration to show the effectiveness after the agentbased approach is applied, (vi) To demonstrate the usage level of information/knowledge after the annotation and articulation mechanisms are used, clip ratio and duration time on the (information/knowledge) Web pages are collected and analyzed aided by log files to compare to the usage level when heterogenous sources are not adopted, (vii) Each week, the user satisfaction level is analyzed and aggregated while the end users complete their working report at the same time, (viii) The size of the repository in the agency is measured to explore the increase of accumulated knowledge sources. Statistically, the effectiveness of the proposed approach is confirmed. The significant results in the personal computer (PC) OEM group are reduction of inventory by 24.5%, reduction of lead time by 12.3%, reduction of user’s complaints by 55%, increase of throughput by 19.5%, increase of effectiveness by 23%, and user satisfaction increase by 34%. After the heterogeneity problems are solved, the available information/knowledge sources increases by 203%. The proposed agent-based system shows great promise for supply chain management and solves the issues of heterogeneity. Although the validation outcomes are encouraging, several issues deserve attention during the initial stage: (1) The construction and maintenance of the knowledge repository in the ISCA system requires a substantial amount of commitment and work by knowledgeable experts. It may require involving experts in a variety of fields to provide their comments. Experts may not be able to participate due to time limitation. This is a continuing process and may require a lot of effort. Furthermore, the issue of why people adopt or use information systems from different perspectives deserves further investigation. For example, the TAM (technology acceptance model)

170 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

focused on the perceived usefulness and perceived ease of use, and shows that it is the main factor influencing people to use information systems (Davis 1989). Research based on social cognitive theory (SCT) shows that behavior (use of information systems) is influenced by personal (e.g., outcome expectancies or selfefficacy) and environmental (support or other’s use) factors (Compeau & Higgins 1995). Research based on social exchange theory (SET) shows costs. Extrinsic and intrinsic benefits may be important factors to influence knowledge exchange (Kankanhalli et al, 2005). (2) Other managerial issues required further consideration to make the adoption decision for this system. Factors to adopt/develop the system, such as potential advantages, compatibility, and complexity are complicated (Rogers 1995). Therefore, during the period of launching the system, the development team has recognized that in developing/operating a real system, the system must (i) show perceived usefulness to the users and ease of use by users (TAM’s perspective); (ii) have a fair reward mechanism to enhance extrinsic and intrinsic benefits and be well-designed to reduce experts costs (SET’s perspective); (iii) training, education, and company’s support are also important factors (SCT perspective); (iv) at the organizational level, this new system must have substantial advantages such as low complexity, compatibility with previous/other systems (e.g., database or hardware), and easily observed outcomes; (v) technical and managerial skills must be equally important in developing practical IT systems. (3) Inspite of the gains from implementing this ISCA system, there are still limitations associated with it. The wise see knowledge and action as one. However, knowledge does not always result in direct action, thereby creating value. In addition, knowledge processes do not stand still. A knowledge repository is continually in danger of providing outdated material. (4) Finally, even if organizations can provide access to substantial quantities of knowledge, creativity still achieves ultimate breakthroughs. Knowledge alone does not guarantee a creative response in decision-making situations. In general, the business and managerial issues are crucial in launching a new IT system and they should be considered at the initial stage. In this case study, six organizational strategies to achieve ISCA system success from related studies are recommended and they have been revised in the implementation procedure (Miller 1987; Grover 1993, King & Sabherwal 1992; Tallon et. al 2000, and Ang et al. 2001): Decision-making structure: Previous studies found decentralized decision-making one of the strongest facilitators of customer-based information inter-organizational system (CIOS) adoption (Grover 1993) and IT use in large and complex organizations (Boynton et al. 1994). On the other hand, studies indicated that highly centralized and organizational design can anticipate better managerial effectiveness of end user computing (Brown & Bostrom 1994), and they tend to adopt more successful strategic information systems applications (King and Sabherwal 1992). Top management support: Based on the important role of top-level managers in organizations, top management support has been one of the most widely discussed organizational factors in ISCA system success from the previous literature. For example, top management support has been investigated in several studies linking

An Agent-based Approach to Enhance Supply Chain Agility 171

its influence on IT/IS use (Jarvenpaa and Ives 1991; Boynton et al. 1994; Ang et al. 2001), IT/IS adoption (Grover 1993), CBIS implementation (Mohd Yusof 1999); strategic use of IS (King and Teo 1996); IS success (Igbaria et al. 1997), and other related IS studies. Goal alignment: Goal alignment involves linking business goals and the ISCA system goal. According to Saunders and Jones (1992), to promote the achievement of organizational goals, IS planning must be tied to organizational planning. Accordingly, the current trend toward this issue has gained interest among researchers and practitioners in both the public and private sectors (Swain et al. 1995; Watson et al. 1997; King and Teo 1996; Tallon et al. 2000). Managerial IT knowledge enrichment: Managerial IT knowledge refers to senior management experience and knowledge, especially in agent technology. This attribute involves the backgrounds of managers, their experience and awareness of IT/IS activities, their recognition of IT/IS potentials, as well as their ability to plan strategically (Jarvepaa and Ives 1991; Boynton et al. 1994; Ang et al. 2001). Jarvenpaa & Ives (1991) found that executives with relevant skills and knowledge background tend to be more productive, more proactive, become more participative in IT/IS projects, and have more favorable views of IT. Desired management style: Management style deals with the way in which the management tends to influence, coordinate, and direct people’s activities toward group objectives (Aldag and Sterns 1991; Robbins 1994). For example, at the development stage and the maturity stage, both people-oriented and task-oriented styles had positive and significant relationships with system success. On the other hand, at the initiation stage, both styles have no effect on system success. They argued that at the initial stage, the computer is being introduced to the organization and users must learn the new technology on their own. This has ended up in dissatisfaction among users. An important component of management style is leadership. Lu and Wang’s study (1997) seems to support the major finding in the study by Igbaria et al. (1990) on the relationship between leadership style and user satisfaction. Both studies found that leadership style and system success correlate significantly and positively. However, the authors agreed that more issues need to be explored involving the many styles of management and leadership. Sufficient resources allocation: The last factor is concerned with allocating resources. According to Ein-Dor and Sejev (1978), resources include money, people, and time that are required to complete the project successfully. Resources lead to a better organizational commitment, and also overcome organizational obstacles (Beath 1991; Tait and Vessey 1988). A sufficient resource also leads to organizational implementation success and project implementation success (Wixom and Watson 2001). Wixom and Watson (2001) found a significant relationship between resources and IT project implementation. They observed that having sufficient fund, appropriate people, and enough time have had positive effects on a project’s outcome.

172 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

7.7 Conclusions This chapter described the work undertaken to conceptualize supply chain activities and interaction with a collection of intelligent agents. The intelligent agent framework was presented and detailed quantitative measure was evaluated with simulation and illustrated with an example of a personal computer (PC) OEM group. The agents in the system autonomously plan and pursue their actions and subgoals to cooperate, coordinate, and negotiate with others and to respond flexibly and intelligently to dynamic and unpredictable situations. Agility is enhanced by using this virtual agency to exploit profitable changes in a volatile marketplace. Future work includes x

x

Universal implementation of the knowledge enriched in a supply chain is essential. For example, Andraski (1994) has stated that only about 7% of U.S. retail supply chains operate effectively. He further argues that the main reason for this result is that supply chains are 20% technology problems, 80% people or knowledge-sharing problems. The challenge of enriching the supply chain with knowledge management is on the way. Communication among agents for system efficiency/effectiveness is crucial. A conversation policy to formulate the communication procedures is needed.

7.8 Guidelines to Practitioners 1.

Key Insights and Contributions This chapter proposes an agent-based system and formulates agile interaction for the collaboration of supply chain entities in a heterogeneous environment. The agent-based system enhances supply chain agility through the approach in which the agents autonomously plan and pursue their actions and subgoals to cooperate, coordinate, and negotiate with others and to respond flexibly and intelligently to dynamic and unpredictable situations. Annotation and articulation mechanisms are developed to address the heterogeneity problem of information/knowledge resources.

2.

How to Apply the Methodology to Real-World Situations First, the user must launch a supply chain management system based on ISCA architecture. Second, the user needs to implement simulations under different scenarios. Third, the user has to develop different simulation models with various operational parameters (agility enablers) for alternative configurations. Fourth, evaluation of alternative configurations

An Agent-based Approach to Enhance Supply Chain Agility 173

provides the manager with insights into changing the supply chain might affect the performance in ways agility. The ability to fine-tune the system and evaluate performance under different scenarios makes this application useful for evaluating short-term (e.g., setting inventory levels, changing inventory control policies) as well as long-term (e.g., changing a supplier, adding a distribution center) reengineering efforts. 3.

Future Possible Extension of Research for Practitioners x x

Since communication among agents to make the systems efficiency/effectiveness is crucial, a conversation policy to formulate communication procedures is needed. Universal implementation of the knowledge enriched in a supply chain is essential. More and more efforts could be invested to cope with the challenge of enriching the supply chain with knowledge management.

7.9 References Aldag, R.J., Stearns, T.M. (1991). Management. South-Western, OH. Andraski, J. C. (1994). Foundations for a successful continuous replenishment programme. International Journal Logistics Management, 5, 1–8. Ang, C.L., Davies, M.A., Finlay, P.N. (2001), An empirical model of IT usage in the Malaysian Public Sector. Journal of Strategic Information Systems, 10, 159–174. Anumba, C.J., Ugwu, O.O., Newnham, L., and Thorpe, A. (2002), Collaborative design of structures using intelligent agents, Automation in Construction, 11(1), January, 89–103. Beath, C.M. (1991). Supporting the information technology champions. MIS Quarterly, 15, 355–372. Berners-Lee, Tim (1998). Semantic Web road map. Available on the Internet at http://www.w3.org/DesignIssues/Semantic.html. Boynton, A.C., Zmud, R.W., Jacobs, G.C. (1994). The Influence of IT management practice on IT use in large organizations, MIS Quarterly 18(3), 299–318. Bresciani, A. P., Paolo G., and Giunchiglia F., (2001). Level software engineering methodology for agent oriented programming. Proceedings of the Fifth International Conference on Autonomous Agents. Brown, C. and Bostrom, R. (1994). Organization Designs for the Management of End-User Computing: Re-examining the Contingencies. Journal of Management Information Systems, 10 (4), 183–211. Burgess, T. F., (1994). Making the leap to agility: defining and achieving agile manufacturing through business process redesign and business network redesign. International Journal of Operations and Production Management, 14, 23–34. Bussler, Christoph (2002). Modeling and executing semantic B2B integration. Proceedings of the 12th International workshop on Research Issues in Data Engineering: Engineering e-Commerce/ e-Business Systems(RIDE’02), 69–74. Cash, J. I., and Konsynski, B. R., (1985). IS redraws competitive boundaries. Harvard Business Review, 63(2), 134–142. Choi, S. P.M., Liu, J., and Chan, S-P., (2001). A genetic agent-based negotiation system, Computer Networks, 37(2), 195–204.

174 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Cole, J and Milosevic, Z (2001). Extending support for contracts in ebXML. Proceedings Workshop on, Information Technology for Virtual Enterprises, 2001. ITVE 2001., Gold Coast, Qld., Australia, 119–27. Compeau, D. R. and Higgins, C. A. (1995), Application of Social Cognitive Theory to Training for Computer Skills, Information Systems Research, 6(2), 118–143. Corby, O. and Dieng, R. (1996). Cokace: A centaur-based environment for CommonKADS conceptual modeling language. In Proceedings ECAI-96 (W. Wahlster, Ed), 418–422. Cui, Zhan, Jones, Dean, and O'Brien, Paul (2001). Issues in ontology-based information integration, Proc. of the IJCAI-01 Workshop on Ontologies and Information Sharing, Seattle, USA, 141–146. Davis, F. D. (1989), Perceived usefulness, perceived ease of use, and user acceptance of information technology, MIS Quarterly, 13(3), 319–340. Descamps S. and Mitsuru Ishizuka (2001). Bringing affective behavior to presentation agents. Proc. 3rd Int'l Workshop on Multimedia Network Systems (MNS2001) (IEEE Computer Soc.), Mesa, Arizona, 332–336. Devanna, M. A. and Tichy, N. (1990). Creating the competitive organization of the 21st century: The boundaryless corporation. Human Resource Management, 29, 455–471. Dove, R. (1994a). The meaning of life and the meaning of agile. Production, 106 (11), 14– 15. Dove, R. (1994b). Plumbing the agile organization. Production, 106 (12), 12–14. Dove, R., (1995). Measuring agility: The toll of turmoil. Production, 107 (1), 14–15. Ein-Dor, P. Segev, E. (1978). Organizational Context and Success of Management Information Systems. Management Science, 24 (10), 1064–1077. Ellram, L. M. (1991). Supply chain management: the industrial organization prospective. International Journal of Physical Distribution and Logistics Management, 21(1), 13–22. Ferber, Jacques, (1999) Multi-Agent Systems: An introduction to distributed artificial intelligence 1st, Addison-Wesley Pub Co, New York. Felfernig, A., Friedrich, G., Jannach, D., Zanker, M., (2000). Cooperating configuration agents supporting supply chain integration of customizable products. MultiAgent Systems, 2000. Proceedings. Fourth International Conference on, 385–386. Fischer, K., Chaib-draa, B., Muller, J. P., Pischel, M. and Gerber, C. (1999). A simulation approach based on negotiation and cooperation between agents: a case study. Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions, 29(4), 531–545. Franciszek Seredynski, Anna Swiecicka, Albert Y. Zomaya., (2001). Discovery of parallel scheduling algorithms. Cellular Automata-Based Systems 132. IPDPS 2001. Gajanana, N. and John, E.B. (1993). Intelligent manufacturing-simulation a gents Tool (IMSAT). TOMACS, 3(1), 42–65. Gary B. Shelly Thomas J. Cashman Judy Adamski Joseph J.Adamski (1995) Systems Analysis and Design, Second Ed. Boyd&Fraser. Gerber, Andreas, Klusch, Matthias, Russ, Christian, Zinnikus, Ingo (2001). Holonic agents for the coordination of supply Webs. The Proceedings of the 5th International Conference of Autonomous Agents, Montreal, Canada, 184–189. Goldman, S. L. Nagel, R. N., (1993). Management, technology, and agility: the emergence of a new era in manufacturing. International Journal of Technology Management, 8, 18– 38. Goldman, S. L., Nagel, R. N. Preiss, K. (1995). Agile Competitors and Virtual Organizations, New York: Van Nostrand Reinhold. Grover, V. (1993). An empirically derived model for the adoption of customer-based interorganizational systems. Decision Sciences, 24(3), 603–639.

An Agent-based Approach to Enhance Supply Chain Agility 175

Han, Woo-Yong, Park, Cheon-Shu, lim, Shin-Young, and Kang, Ji-Hoon (2001). An XML digital signature for Internet e-business applications. Proc. ICII 2001 - Beijing. 2001 International Conf. on, Beijing, China, 6, 23–29. He, Ning Milosevic, Z. (2001). B2B contract implementation using Windows DNS. Proc. Workshop on, Information Technology for Virtual Enterprises, 2001. ITVE 2001, Queensland Univ., Qld., Australia, 71–79. Heflin, J. Hendler, J. (2000). Semantic interoperability on the Web. In: Proceedings of Extreme Markup Languages 2000. Graphic Communications Association, Alexandria, VA, 111–120. Hendler, J. McGuinness, D.L., (2000). The DARPA Agent markup language, IEEE Intelligent Systems, 16(6), 67–73. Herring, C. Milosevic, Z. (2001). Implementing B2B contracts using BizTalk. Proc. of the 34th Annual Hawaii International Conf. on System Sciences, 2001, 4078–4087. Hofreiter, Birgit, Huemer, Christian, Klas, Wolfgang (2002). ebXML: status, research issues, and obstacles. Proc. of the 12th Int.l Wrkshp on Research Issues in Data Engineering: Engineering e-Commerce/ e-Business Systems (RIDE.02), San Jose, CA, USA, 7–16. Huang, C. C. Kuo, C. M. (2003). Transformation and searching of semi-structured knowledge in organizations. Journal of Knowledge Management, 7(4), 106–123. Huang, Chun-Che, Tzu-Liang (Bill) Tseng, Min-Chiun Chen (2002). The agent-based collaboration information system for agile supply chain management, The 6th World Multiconference on Systemics, Cybernetics, and Informatics, July 14–18, ~ Orlando, Florida, 503–508. Huhns, Michael N. Stephens, Larry M. (2001). Automating supply chains. IEEE Internet Computing, 5(4), 90–93. Igbaria, M., Zinatelli, N.; Cragg, P. Cavaye, A. (1997). Personal Computing acceptance factors on small firms. MIS Quarterly, 21 (3), 279–302. Inmon, W. H., Zachman, J. A., Geiger, J. G., (1997), Data stores, data warehousing, and the Zachman Framework: Managing enterprise knowledge, McGraw-Hill. New York. Ito, T., Salleh, M. R., (2000). A blackboard-based negotiation for collaborative supply chain system. Journal of Materials Processing Technology, 107, 398–403. Ivezic, Nenad, Barbacci, Mario, Libes, Don, Potok, Tom, Robert, John (2000). An analysis of a supply chain management agent architecture. Fourth International Conference on MultiAgent Systems (ICMAS-2000), 10–12 July 2000, Boston, Massachusetts. Ferber J., (1999) Multi-Agent Systems: An introduction to distributed artificial intelligence, 1st ed., Addison-Wesley. New York. Jarvenpa, S.,Ives,B. (1991) Executive Involvement and Participation in Progressive Use of IT, MIS Quarterly, June 1991. Jennings, N. R., (1995). Controlling cooperative problem solving in industrial multi-agent systems using joint intentions. Artificial Intelligence, 75(2), 195–240. Jennings, N. R., Corera, J., Laresgoiti, I., Mamdani, E. H., Perriolat, F., Skarek, P., Varga, L. Z., (1996b). Using ARCHON to develop real-word DAI applications for electricity transportation management and Particle acceleator Control. IEEE Expert, 6(5), 64–70. Jennings, N. R., Faratin, P., Johnson, M. J., O'Brien, P., Wiegand, M. E., (1996a). Using intelligent agents to manage business processes. Proceedings of the First International Conference on The Practical Application of Intelligent Agents and Multi-Agent Technology, London, UK, 345–360. Jennings, N.R., Sierra, C., P. Faratin., (2000). A service-oreinted negotiation model between autonomous agents. Collaboration between Human and Artificial Societies Coordination and Agent-Based Distributed Computing (ed. J. Padget), Lecture Notes in Artificial Intelligence, Volume 1624, Springer Verlag, 201–220.

176 Chun-Che Huang, Tzu-Liang (Bill) Tseng, Horng-Fu Chuang, and Yu-Neng Fan

Kaihara, T., (2001). Supply chain management with market economics. International Journal of Production Economics, 73(1), 5–14. Kankanhalli, A., Tan, Bernard C. Y. Wei, K.K. (2005), Contributing knowledge to electronic knowledge repositories: An empirical investigation, MIS Quarterly, 29(1), 113–143. Karacapilidis, N. Moraïtis, P., (2001). Building an agent-mediated electronic commerce system with decision analysis features. Decision Support Systems, 32(1), 53–69. Kidd, P. T., (1994). Agile manufacturing: forging new frontiers, Addison Wesley, New York. King, W. R., Teo, T. H. (1996). Key dimensions of facilitators and inhibitors for the strategic use of information technology. Journal of Management Information Systems,12(4), 35–53. King, W.R., Sabherwal, R.(1992). The factors affecting strategic information systems applications: An empirical assessment. Information & Management, 23(4), 217–235. Kochikar, V. P., (1998). Object-Powered Web. IEEE Software, 15(3), 57–62. LaBarre, P., (1995). The seamless enterprise. Industry Week, June 19, 22–34. Lee, H. L., Ng, S. M., (1997). Introduction to special issue on global supply chain managment. Production and Operation Management, 6(3), 191–192. Leukel, Joerg, Schmitz, Volker, Dorloff, Frank-Dieter (2002). Modeling and exchange of product classification systems using XML. Proc. of the 4th IEEE Int’l Workshop on Advanced Issues of e-Commerce and Web-Based Information Systems (WECWIS 2002), 242–244. Logan, B. S. Theodoropoulos, G.., (2001). An approach to interest management and dynamic load balancing in distributed simulation. Proceedings of the 2001 European Simulation Interoperability Workshop (ESIW'01), Simulation Interoperability Standards Organisation and Society for Computer Simulation, 565–571. Lu, H.P. Wang, J.Y. (1997). The Relationship between Management Styles, User Participation, and System Success over MIS growth stages. Information & Management, 32 (3), 203–213. Luo, X., Zhang, C. Leung, H., (2001). Information sharing between heterogeneous uncertain reasoning models in a multi-agent environment: A case study, International Journal of Approximate Reasoning, 27, 27–59. Mayr, Heinrich C., Lazanský, Jirí, Quirchmayr, and Vogel, Gerald Pavel (2001). In: Proceedings of 12th DEXA International Conference 2001. Munich, Germany. Meade, L. M. Sarkis, J., (1999). Analysis organizational project alternatives for agile manufacturing process: An analytical network approach, INT. J. PROD. RES., 37(2), 241–261. Mitra, P. Wiederhold, G. (2002). An algebra for semantic interoperability of information sources. The 2nd IEEE International Symposium on Bioinformatics and Bioengineering (BIBE'01), 174. Mitra, P., Kersten, M., Wiederhold, G. (2000). A graph-oriented model for articulation of ontology interdependencies. Proceedings of EDBT. Mohd-Yusof, M. (1999). A multiple perspective approach to diagnosing computer-based information systems implementation. Asia-Pacific Journal of Management Sciences, 5, 17–35. Naylor B., BERRY D., and BERRY D. (1999). Leagality: Integrating the lean and agile manufacturing paradigms in the total supply chain. International Journal of Production Economics, 62, 107–118. Pathak, S.D., Nordstrom, G., Kurokawa, S., (2000). Modeling of supply chain: a multi-agent approach. Systems, Man, and Cybernetics, 2000 IEEE International Conference, 3, 2051–2056.

An Agent-based Approach to Enhance Supply Chain Agility 177

Robbins, S., (1994). Management, 4th ed., Prentice-Hall. Rogers, E. M. (1995), Diffusion of Innovations, 4th ed., Free Press, New York. Shelly, Gary B., Cashman, Thomas J., Adamski Judy, and Adamski, Joseph J., (1995) , Systems Analysis and Design, Boyd&fraser publishing company, new York. Satyadas, A., Harigopal, U., Cassaigne, N. P. (2001). Knowledge management tutorial: An editorial overview. IEEE Transactions on Systems, Man and Cybernetics, Part C: Applications and Reviews, 13(4), 429–437. Saunders, C.S. Jones, J.W. (1992). Measuring Performance of the Information Systems Function. Journal of Management Information Systems, 8(4), 63–73. Sheth, A. P. (1998). Changing focus on interoperability in information systems: From system, syntax, structure to semantics. In Interoperating Geographic Information Systems, (M. F. Goodchild, M. J. Egenhofer, R. Fegeas, and C. A. Kottman, ed.), Kluwer Academic, Boston. Swain, J.W., White, J.D., Hubbert, E.D. (1995). Issues in Public Management Information Systems, American Review of Public Administration, 25 (3), 279–296. Swaminathan, J.M., Smith, S.F., and Sadeh, N.M., (1998). Modeling supply chain dynamics: A multi-agent approach, Decision Sciences, 29(3), 607–632. Tait, P. and Vessey, I. (1988), The Effect of User Involvement on Systems Success: A Contingency Approach, MIS Quarterly, 12(1), 91–108. Tallon, P.P., Kraemer, K.L., Gurbaxani, V. (2000). Executives’ Perceptions of the business value of information technology: A process-oriented approach. Journal of Management Information Systems, 16(4), 145–173. Tan, Keah Choon, (2001). A Framework of Supply Chain Management Literature, European Journal of Purchasing & Supply Management, 7 , 39–48. Turban, E. Aronson, J.E., (2001). Decision support systems and intelligent systems, Prentice Hall, New Jersey. URL 0: http://www.ebxml.org/ URL 1: http://www.microsoft.com/biztalk/evaluation/overview/biztalkserver.asp URL 2: http://www.eclass.de URL 3: http://www.un-spsc.net/ URL 4: http://www.openapplications.org Watson, R.T., Kelly, G.G., Galliers, R.D. Brancheau, J.C. (1997). Key issues in Information systems anagement: An international perspective. Journal of Management Information System, 13(4), 91–115. Wixom, B., Watson, H. (2001), An Empirical Investigation Of the Factors Affecting Data Warehousing Success. MIS Quarterly, 25(1), 17–32. Wu, D.J. (2001). Software agents for knowledge management: coordination in multi-agent supply chains and auctions, Expert Systems with Applications, 20(1), 51–64.

PART II. TRENDS IN METHODOLOGIES

8 Design of Reverse Logistics Networks for Multiproducts, Multistates, and Multiprocessing Alternatives Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

Abstract:

This chapter proposes a modeling methodology for designing reverse logistics networks. The model aims at determining the location and missions of sites for the recovery of unused products from ultimate consumers, valorization or clean disposal of recovered products, redistribution of reusable materials, and attribution of new or reusable (valorized) products. Valorization activities refer to repair, refurbishing, reassembling, product disassembly for reusable material recovery (cannibalization), and recycling. The proportion of recovered product volumes to orient to valorization and clean disposal activities is not known a priori but is determined according to demand and return volumes, site capacities, and the general anticipated state of recovered product volumes. This model may be used to evaluate the impact of reintegrating valorized products (finished products and spare parts) into current supply chains initially designed only for distribution and maintenance of new products. The chapter discusses key parameters such as the localization and estimation of potential returns and demands for new and reusable (valorized) products, as well as the probability that a returned product be in a specific state, which could lead to one or many processing alternatives (repair, disassembly, clean disposal, etc.). This mathematical model is inspired by the recent healthcare allocation and valorization of the wheelchair policy of the Province of Quebec (Canada), governed and managed by a governmental agency.

8.1 Introduction Economic and environmental pressures to reduce the consumption of nonrenewable resources increase organizational responsibility regarding return of end of life products. The Kyoto Protocol, recent directives on electronic waste in

182 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

Europe, and significant return volumes in the electronic industry are some examples. Direct reuse, valorization activities, which refer to repair, refurbishing, reassembling, product disassembly for reusable materials recovery and recycling (Thierry et al., 1995), and even clean disposal are being considered with increasing interest. For healthcare systems, product reuse is seen as an economical alternative to reduce continuously rising costs and to fulfill increasing demand while ensuring a high service level. A recent study concerning wheelchairs conducted by the Régie de l’assurance maladie du Québec (RAMQ) shows that valorization activities, realized autonomously by mandated rehabilitation centers, can improve accessibility to such devices while reducing expenditures (Côté et al., 2003). Thus, the RAMQ and rehabilitation centers in the Province of Québec (Canada) are now interested in reviewing the actual configuration of their logistics network, which was initially designed only for allocation and maintenance of new products. The RAMQ is also involved in the impact evaluation of such a configuration on logistics costs and service level. New modeling approaches of logistics networks design must be investigated to consider the recovery of unused products from ultimate consumers and the use of valorized products (finished products and spare parts) instead of new. A generalized approach is proposed in this chapter. Strategic decisions on reverse logistics network design concern recovery, processing, and redistribution. Sites have to be located. All logistics activities need to be assigned to suitable sites, while respecting capacity and ensuring effective and efficient response to requests expressed to networks. Different costs and service levels will be met according to choices made. Product orientation to processing alternatives can influence the network configuration in which business units are dedicated to specific processing alternatives and even to one or several products. Products can be used to meet network needs in a variety of ways: They can be repaired, refurbished, or reassembled for finished product supply, or disassembled for spare parts supply. If preferable, products can even be recycled or simply cleanly disposed of. Product state, demand and return expressed to the network and site capacities all determine how product volume will be processed and whether processed products will be used to respond partly or completely to network needs. The objective of this chapter is to suggest a methodology for designing the logistics network while evaluating the strategic proportion of product to direct toward each of these alternatives in an efficient and effective way to meet network needs. This chapter presents various models suggested for specific reverse logistics networks design. In light of these models, a logistics network reengineering process is suggested to represent networks flows in a multiproduct, multistate and multiprocessing alternatives context. It deals with the design of reverse networks, while considering current supply chain networks, to efficiently supply processed products as alternatives to new. To complete it and to widen horizons, new parameter definition approaches and a mathematical model are proposed.

Design of Reverse Logistics Networks

183

8.2 Related Reverse Logistics Design Models Several models for reverse logistics networks design are proposed in the literature (Table 8.1). The formulation of these models proposes some distinctive modeling methods in addition to those traditionally developed for supply chains (Jayaraman and Pirkul 2001; Martel 2001; Geoffrion and Graves 1974). 8.2.1 Localization and Determination of Demand and Return Volumes One significant distinction modeling reverse logistics networks is underlined by Fleischmann (2001). Within the supply chain context, models are generally developed so that networks satisfy final demand from consumers (pull flows), whereas flows must be directed, as well as possible, in reverse logistics networks (push flows). In some cases, processed products can be reintroduced into the supply chain to satisfy demand partially or completely, thus causing disequilibrium between supply and demand. Most of the recent models take this situation into account with demand, recovery, and site capacity constraints. A closed or open supply loop, with products reintegration in the original supply chain or in alternative markets, is then considered. Some proposed models consider that unused products from ultimate consumers are already in established recovery centers, whereas others use the concept of user zones. For each of these zones, a demand and a return volume is generally defined up front (Listes 2007; Lu and Bostel 2007; Fleischmann 2001), or return volume is represented as a fraction of the demand volume (Fandel and Stammen 2003). They are used to locate recovery centers, processing centers, and warehouses and to evaluate transportation costs, while allowing consideration of service level. Zones can be defined according to the distance separating user zones from sites offering products and services (service centers). 8.2.2 Product Families and Bill of Materials Few models deal with multicommodity networks. Some tackle it for product decomposition. It is notably approached by Fandel and Stammen (2003) with a reverse bill of materials or by Spengler et al. (1997) and Shih (2001) with mass relation between materials. The former approach specifies which product families can be recovered from others and which disassembly sequence is to be used. 8.2.3 Processing Conditions and Product States Generally, only one (Listes and Dekker 2005; Fandel and Stammen 2003; Jayaraman et al. 2003; Shih 2001; Krikke 1998) and sometimes two options for processing alternatives (Listes 2007; Lu and Bostel 2007; Fleischmann 2001) are considered in proposed models. When more than one processing alternative is considered, the proportion of products directed toward one or another alternative is generally determined and fixed a priori. Fleischmann (2001) proposes a lower bound on material quantity eliminating and thus considers technical or economical

184 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

infeasibilities related to product reintroduction into the market. A certain degree of freedom is then considered as to processing alternatives. According to the state of recovered product volumes, different processing alternatives may occur. Listes and Dekker (2005) consider this aspect by assigning states with proportions of recovered products. A processing alternative is associated with each state, and processed products are used to fulfill a given demand. Typically, recovered products are reintroduced in networks in a state similar to those in the supply chain. Hence, new or like-new product reuse or sales of only valorized products represent distribution activities in those networks. In this way, customers do not have a valorized product alternative to new or as new products. Valorized products, especially in a public setting, may represent an economical supply source and a compromise solution allowing improved performance in service delay and satisfaction. It may also have an impact on logistics network environmental performance. Little work evaluating these impacts has been done. Several methods have been proposed to establish recovery and processing strategies and to evaluate their impact on costs and benefits to an organization (Teunter 2005). Some contributions at the strategic level, such as Krikke (1998), aim at defining the proportion of product to be directed to each processing alternative a priori, according to the quality of returned product and their constitutive parts, and then to configure the network in consequence. Teunter (2005) has adapted this preliminary approach by considering multiple disassembly processes and partial disassembly in addition to product quality. Some reverse logistics specificities are mentioned in the literature; however, these references offer few suggestions on how to integrate them explicitly in the modeling stage of logistics networks. When specificities are introduced into models, parameters are generally defined as deterministic, and little flexibility is suggested regarding recovery, processing, and redistribution. Most of the models proposed in the literature do not approach the fact that the same recovered product can be used to meet various needs, in particular in terms of valorized finished products and spare parts. Different processing alternatives may also arise according to product condition and the capacity and needs in a network. Ideally, more than one processing alternative should be considered to improve value recovery possibilities and to limit environmental impacts. However, product flow orientation is complicated in such a context. Additional efforts are then required to define products, customers, and organizational activities, and thus, to detail all potential network flows in a simple way. This chapter aims to contribute to the development of an extended model tackling these issues. The characteristics of this model, compared to those identified in the literature, are summarized in Table 8.1. Table 8.1. Main characteristics of related reverse logistics design models

Design of Reverse Logistics Networks

Table 8.1. (continued)

185

186 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

Table 8.1. (continued)

Design of Reverse Logistics Networks

187

Table 8.1. (continued)

[1] Spengler et al. 1997; [2] Krikke 1998; [3] Fleischmann 2001; [4] Shih 2001; [5] Fandel and Stammen 2003; [6] Jayaraman et al. 2003; [7] Bloemhof-Ruuward et al. 2004; [8] Listes and Dekker, 2005; [9] Listes 2007; [10] Lu and Bostel 2007; [11] Chouinard et al. (Appendix). a b c d e f g h *

Recovery only Reassembly considered by the replacement of unusable components or assembly module (component or assembly module in state s=3 and s=4) Multirecycling processes Reusable products reintegration according to process capacities Process localization Minimize economic costs, energy use, and residual waste Fixed cost of selecting a process Bound on number of realizations of a process at a location Work in progress

8.3 Logistics Network Reengineering Process First, the necessary information to dress a total portrait of the organization, as regards its products, ultimate consumers, capacities, constraints, etc., is identified. Such a parameter definition step is critical in the design methodology. In his book, Shapiro (2001) extensively discussed the importance of parameter definition in supply chain design. Additional aspects are to be considered with a view to their adaptability to reverse logistics integration. Parameters considered here include: (i) demand and recovery zones (user zones) and demand and return volumes for each zone (service forecasts); (ii) product families; (iii) bill of materials, including recovery; (iv) processing conditions and product states (Sections 8.3.2–8.3.5). The next step of this reengineering process approached here is the development of a mathematical model. It shows how reverse logistics characteristics can be

188 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

conceptualized and integrated (Section 8.3.6 and Appendix) with the use of these parameters. Additional required flow conservation constraints are detailed with figures in the chapter. Constraints of the mathematical model, detailed in the Appendix, are presented with notations using square brackets in these figures ([Equation]). 8.3.1. Context Studied The network (Figure 8.1) consists of service centers already in place satisfying ultimate consumers within user zones defined geographicly. These centers offer new and valorized products as well as maintenance and recovery services. New products are delivered by established external suppliers. Each service center can also play the role of recovery center, using its private vehicle fleet or a service logistics provider. Voluntary returns from ultimate consumers are possible, following product replacement or maintenance activities, and recovered products must then be forwarded to a suitable processing center, for finished product repair, components disassembling and refurbishing, or recycling and clean disposal. Product orientation to processing alternatives depends on product state, site capacities, and demand for valorized products (finished products and spare parts).

Figure 8.1. Reverse networks business units, products, and potential flows

Valorized products generated in these centers are stored to adequately meet future service and processing centers needs to repair products. Both new and valorized

Design of Reverse Logistics Networks

189

spare parts can be used for product repair. Recovery and processing centers and warehouses are to be located to improve accessibility to valorized materials and to minimize costs of integrating such reverse logistics network with the current supply chain. Use of valorized products is cheaper than new. The network reported here is a generalization from the wheelchair allocation problem in Quebec, (Chouinard et al., 2005; Chouinard, 2003). 8.3.2 Localization and Determination of Demand and Return Volumes The number of ultimate consumers (users) is too high for each to be considered individually at the design stage of the logistics network. They are represented by restricted geographic, here called user zones, which allow localizing and forecasting demand and return volumes. Zones can also be assigned to proper service and recovery centers according to the distance separating them. It is useful to estimate transportation costs, particularly when more than one transportation mode may be considered to fulfill demand or to recover products. For supply chains, user zones are defined as demand zones. A general approach to locate these zones is to weight customer coordinates (Xi, Yi) with their demand volume (Wi), known and fixed a priori, as follows: Xj =

¦ iV j Wi X i

¦ iV j

Wi

j = 1, …, m ;

Yj =

¦ iV j WiYi

¦ iV j Wi

j = 1, …, m,

where Vj is zone j gathering some customers i. Clustering methods are generally recommended for the aggregation process (Ballou 1994). Some general rules are suggested in the literature to control the location, size, and number of zones for supply chains (Ballou 1994). It is difficult to apply this approach to reverse logistics as products in circulation represent potential returns, which can feedback to the organization. In most reverse logistics cases, return origins and quantities are not known a priori. Where reverse logistics are integrated into the supply chain, modeling of such user zones must consider two types of flow, direct (demand) and reverse flows (recovery). These two types of flows can be treated by distinct or common geographic zones (Figure 8.2). The choice will depend on the studied context. Five principal factors can influence this choice: x x x x

Localization of service centers and recovery centers in common and/or distinct sites: Common sites can lead to common zones for both direct and reverse flows, Different territory covered by service centers and recovery centers: For common sites, as regards service centers and recovery centers, the same territory cover can lead to common zones, Link between direct and reverse flows: Dependent direct and reverse flows can lead to common zones, Service costs (delay, transportation costs) to evaluate only direct or reverse flows or for the two types of flows jointly: Distinct zones can

190 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

x

facilitate evaluation of service costs, particularly when evaluated differently for direct and reverse flows, Users or customers and products common to demand and recovery processes: Different users and products as regards demand and recovery can require distinct zones.

In addition, various service types can arise in networks, both for demand and recovery. Demand can come from existing consumers, but also from new consumers. It can occur during acquisition (new product or material addition to a product in circulation) or product replacement (exchange or maintenance). Demand can be satisfied by new or valorized products. Recovery can arise following voluntary return (replacement or unused product) on behalf of the user or from recovery undertaken by the organization (private vehicle fleet or logistics service providers). Products in different states can be involved and require the use of distinct resources. Distinct geographic areas can be considered and particular sites can be used for each service type. Figure 8.2 presents different service types that can occur between user zones and sites (service and recovery centers). According to the situation, product and ultimate consumer (user) characteristics can influence organizational activities. Certain information known by the organization that links users and products, such as contract conditions, can also be used to evaluate needs expressed in the network. Relevant information can then be used to forecast possible services accurately in a dynamic and highly uncertain environment caused by reverse logistics integration to the supply chain: x x

x

User status: Age, gender, localization, condition and intensity of use, consumer behavior, social class, etc. Product status: Product family, bill of materials, end of guarantee, value, aesthetic and technical quality when put on the market (e.g. new or valorized) or in circulation, age, failure law (residual lifespan), etc. Organization: Localization, operating context of service and recovery centers, condition at market entry (sale or renting), contract conditions, service type, product state during activities, etc.

The use of this information to create forecasts requires data and, consequently, ultimate consumers (users) and product families (Section 8.3.3) in circulation need to be localized beforehand. Therefore, it is preferable to initially locate users and product families in circulation in restricted geographic areas and then establish or distribute forecasts to these zones according to their composition. However, the approach adopted must consider potential errors related to zone definition: x

Positioning errors (Hillsman and Rhoda 1978 in Ballou 1994): -

x

Total estimated service cost compared to real service cost met when all consumers (for each service type) of a zone are served individually; Inappropriate allocation of consumers to sites and erroneous location of these sites due to an aggregated rather than individual demand.

Service forecast errors:

Design of Reverse Logistics Networks

-

191

Increased forecast errors due to poor consumer aggregation or poor service characterization.

Figure 8.2. Possible scenarios of user zone

The level of details relating to the extent of geographic areas, by individually approaching or not direct and reverse flows and by separately considering or not each service type, has an impact on the evaluation of network performance. Different users and products can be involved according to the situation and will require products and services at different costs and service levels.

192 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

8.3.3 Product Families Product variety managed by an organization can be relatively significant. To facilitate later stages of logistics network design, products (finished products, assembly modules, and components) are generally gathered in families. A family contains products with similar characteristics which use equivalent processes throughout the logistics network. Products are gathered, for logistics networks design, on the basis of logistics processes characteristics. Definition of product families is usually done with ABC classification methods, analytic hierarchy process (AHP), or clustering algorithms (Ramanathan 2005; Srinivasan and Moon 1999; Flores et al. 1992; Ernst and Cohen 1990). For the supply chain, products are gathered according to the following process characteristics (Martel 2004): x x x x

Demand profile (risks, seasonality, ...); Production, storage, and handling technologies used; Distribution channels and service levels required; Means of transportation considered.

However, to extend these approaches to reverse logistics, it is necessary to consider that products can be reintroduced in the network at several levels and may be subject to modifications before reintegration. Additional characteristics then need to be considered (Chouinard 2003; Krikke 1998): x x x x x

Technical feasibility to recover and process products; Commercial feasibility to redistribute reusable products; Environmental feasibility to recover and process products; Material flow (demand and return volume, product volume state); Economical and environmental costs and benefits in product life-cycles.

Distinct families can be defined for supply chains and reverse logistics, however, considering both at once would complicate subsequent analysis steps of logistics network design. Some links may also occur between different products gathered in different families, and product substitutability may be allowed in some network stages. The suggested methods can then require explicit use of bills of materials, but they must be adapted to a reverse logistics context. 8.3.4 Bill of Materials In a supply chain context, different volumes of material (raw material, component or assembly module) families can be assembled at different stages of the manufacturing system to fulfill demand for finished product families. Regarding reverse logistics, different volumes of material families can be disassembled from recovered finished product families to recover those in good condition or to replace the defective ones. These activities can occur at different stages of the logistics network and may involve many sites. For logistics networks design, a bill of materials (BOM) is used to specify material needs or supplies and assembly or disassembly sequence at each stage of the network (Figure 8.3).

Design of Reverse Logistics Networks

193

Some papers suggest clustering and association mining methods to define BOM. Romanowski et al. (2005) propose a method that clusters design of products, represented as BOM, into product families. Srinivasan and Moon (1999) suggest an approach to define product families, which takes into account BOM and the fact that products may be used at different levels of the network. Contrary to BOM considered for supply chains, which is presented by an acyclic oriented graphic, cycles can occur between graph nodes in the context of reverse logistics (Figure 8.3). This characteristic arises with material flows that can occur in both directions, to consider (re)assembling and disassembling. Moreover, nodes can be presented in two generic forms, such as new products or as recovered and valorized products. These later nodes represent, respectively, products that can be disassembled in material (raw material, component or assembly module) families in various states (Section 8.3.5) and products that can be used as an alternative to new products, but of poorer quality. As for a supply chain, the quantity associated with each link of the BOM indicates the product volume portion, which can be involved in a particular network stage. Bold arrows in Figure 8.3 represent requirements of new products for assembly, dotted arrows represent product requirements for reassembly or product supplies while in disassembly. It is considered here that only new products can be used in new product composition, whereas new and valorized products can be used to valorize recovered product. The quantity characterizing each link ensures flow conservation. For new product assembly, it is then possible to ensure new materials (raw material, component and assembly module) availability for production. For product reassembly, in particular during maintenance or valorization activities, the role is similar. However, contrary to new product assembling, reassembling can be done with either new or valorized materials, according to the given case. Flow conservation constraints intervene here to ensure that the volume for a given material, of new and valorized material necessary for product reassembling, does not exceed the specified quantity characterizing the link between the two nodes involved. For recovered product disassembly, flow conservation constraints determine the probability of recovering a volume of material (raw material, component, or assembly module) family from a product family. An additional parameter must be considered to specify if it is possible to disassemble the targeted product family, as suggested by Fandel and Stammen (2003). This aspect takes technical, economic, and other constraints (disassembly feasibility) into account that can occur while disassembling (Figure 8.3). In our context studied, recovered product families can generate only valorized materials, but in various states; therefore, other constraints are necessary if this aspect is to be considered (Sections 8.3.5 and 8.3.6.1–8.3.6.2). 8.3.5 Processing Conditions and Product States Product families and bills of materials are defined for the supply chain by considering the flow of material in a identical state to produce finished products respecting the quality standards of the organization. However, product families in different states can occur in reverse networks.

194 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

During the processing of a recovered product, the product state will determine if it can be reused directly or has to be oriented to some other processing alternative. Subfamilies then need to be defined and assigned to product families in the mathematical model, according to their state. These states depend on the average general condition of a product family in circulation within a zone (Section 8.3.2). They also depend on the processing alternative retained for a specific disassembled product family of a given state, since some alternatives may be more destructive. These states can be fixed a priori, however, it is probably more convenient to assign states to product families with probability distribution functions integrated into the model. This approach then considers uncertainties regarding processing and reuse of recovered product volumes. Within the framework of this chapter, four states can be assigned to products: x x x x

New: s = 1; Used and dedicated to repair (valorized): s = 2; Used and dedicated to disassembly and spare parts refurbishing: s = 3; Used and dedicated to recycling or clean disposal: s = 4.

Probability distribution functions (Figure 8.4) may describe the quantity and state of all recovered materials in the network from ultimate consumers such as those generated following disassembly. These functions would be defined for each product family considered in the model (Section 8.3.3). Material volumes produced while disassembling will have to respect the bill of materials (Figure 8.3). Raw material, Component family (cC)

1N*

2N

1V



3N

cN

Product family (pP)

h3(c+2)=0 (c+1)N

(c+2)N

(c+1)V

… (c+2)V

Assembly module family (mM)

gcp=1 h(c+2)(m-1)=1 (m-1)N

(m-1)V

mN gmp=2

Finished product family (aA)

(m+1) N

aN

… aV

pN

LEGEND: Possible material flows while assembling and disassembly in the network:

gpp’ : hpp’ :

: need ( ) or supply ( ) of valorized products. : need ( ) of new products. quantity of product family p involved in transformation activities (ass./disassembly) of product family p’. feasibility to disassemble product family p from product family p’ (hpp’ = 1 if possible, 0 otherwise).

* N and V index refer respectively to new materials or to recovered and valorized materials.

Figure 8.3. Partial bill of materials for modeling logistics networks

Design of Reverse Logistics Networks

195

A given state will predestine product families to a certain processing alternative. However, the model suggested in this chapter offers a certain degree of flexibility in this regard (Section 8.3.6.2).

Figure 8.4. Valorized material production function

8.3.6 Location-allocation Model The mathematical model developed, and introduced in this section consists primarily in determining a reverse network configuration while considering the use of current supply chain sites and additional business units. It concerns the location of recovery and processing centers and warehouses. Material flows are summarized and represented in figures, which indicate the related model constraints. Details model equations are given in the Appendix. The objective function of the model is to minimize x

Fixed costs (recovery and processing centers and warehouses) + variable processing costs + material flow costs (handling, storage and transportation) between sites [Equation 8.1].

Material flows are subjected to the following constraints: x x x x x x x x x

Demand [Equation 8.2–8.6] (Section 8.3.6.1); Recovery [Equation 8.7–8.9] (Section 8.3.6.1); Preliminary product flows orientation [Eq. 8.10–8.11] (Section 8.3.6.1); Elaborated product flows orientation [Eq. 8.12–8.15] (Section 8.3.6.2); Disassembly [Equation 8.16–8.23] (Section 8.3.6.2); Recycling or clean disposal [Equation 8.24–8.27] (Section 8.3.6.2); Material replacement [Equation 8.28] (Section 8.3.6.2); Valorized products supply and demand [Equation 8.29–8.30] (Section 8.3.6.2); Capacity (transportation, processing, storage) [Equation 8.31–8.34].

Emphasis is placed on specific constraints related to product flow orientation according to network conditions (supply and demand volumes, site capacities) and general product volume state (Figure 8.5).

196 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

Demand

Demand for new products

Demand volumes Minimal demand fulfillment with valorized products Recovery

Preliminary orientation Maximal proportion of products in good condition

Return volumes

Product with valorization potential

Demand for valorized products

Elaborated orientation, disassembly and material replacement Maximal proportion of finished products Repaired in good condition finish products Refurbished spare parts

Minimal proportion of unusable products Recycling or clean disposal

Unusable products

Minimal proportion of unusable material

New delivered products Valorized product supply and demand Valorized delivered products Valorized recovered products

Unusable and unvalorized products

LEGEND :

Constraint:

Possible product volume:

Possible product flow:

Figure 8.5. Summarized relation between material flows and model constraints

8.3.6.1 Potential Logistics Network The potential network for the considered context (section 8.3.1) is presented in Figure 8.1 and is detailed in the text starting from user zones, where reverse flows are initiated. Users and product families in circulation are aggregated in restricted geographic areas, indicated as user zones (section 8.3.2). Recovery centers are supposed to be located in a restricted number of service centers in this chapter. Location of recovery centers will be evaluated by scenarios analysis, for which user zones will consequently be assigned according to the distance separating them. Material recovery initiates reverse flows in the network. Aggregated flows of recovered product families are split among the different processing centers, which are either valorization centers or recycling and/or clean disposal centers. The former provide effective product reintegration into the original logistics network (closed supply loop) and the latter handles end of product family life-cycles within this same network (open supply loop). Valorization centers can use internal logistics network units, such as service centers with necessary expertise and resources to deal with valorization activities or centers dedicated to these activities, or third parties. Recycling and clean disposal centers are third parties. Valorized material volumes, which can extend product family lifecycle within the original network, are stored before their delivery to ultimate consumers, and both public and private warehouses are considered. Warehouse and processing center location and capacity need to be determined to ensure better delivery for valorized products in comparison with new products from suppliers, according to their distance from service centers. These sites must

Design of Reverse Logistics Networks

197

be able to meet network needs effectively, at both service (acquisition and replacement, including maintenance) and processing centers (valorization activities to repair recovered products). Supply of new products comes from external suppliers, which deliver new finished products and spare parts on demand. Each supplier offers a specific product variety, represented by product families. Transportation costs are included in acquisition costs. Transportation between various nodes within the network (excluding suppliers) is carried out by logistics service providers or private vehicle fleets from service centers. Transportation is not considered for product acquisition and in some return circumstances, specifically for replacement, since ultimate consumers have to visit service centers. Recovery services are offered by the organization to ensure a greater possibility of recovery, which can be particularly significant when external pressures are exerted on the organization (by governmental agencies or other). 8.3.6.1 General Material Flows Orientation In addition to demand and recovery volume within the potential logistics network sites and their capacities, material flows between each site are also split according to product family condition. Details of these flows are schematized in Figure 8.6. Material flows within logistics networks, for both direct and reverse flows, are initiated at the user zone level. In this chapter, demand expressed by each user zone and, consequently, to the associated service center arises under two circumstances: Acquisition and product replacement (including maintenance). For these two situations, customers or users have to visit service centers where demand can be filled by either new or valorized materials. However, a portion of needs must be filled by valorized materials, according to customer or user requirements and according to organizational policies. As with demand, two recovery situations can arise: Voluntary returns and recovery steps undertaken by the organization. Different recovery costs are then generated, according to return type and also according to recovery resources needed (Section 8.3.2), since a proportion of unused product cannot be recovered from ultimate consumers by the organization, particularly when they cannot be found. No penalty costs are considered for these products. As previously indicated, four states can be assigned to product families (Section 8.3.5); each state requires different processing and resources. To avoid unnecessary material flows in the network at a strategic level, preliminary orientation (Figure 8.6) is carried out at the recovery center level (service centers). The volume of product families recovered is oriented in the network based on the general state of returned products (finished products or spare parts recovered during maintenance activities, for example). This volume is then separated into two categories. All materials with reuse potential are directed toward valorization centers, whereas all materials in deteriorated condition are directed toward recycling or clean disposal centers.

198 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

8.3.6.2 Elaborated Product Flows Orientation and Processing The general state of a product family volume can determine its direction toward a given processing alternative, but this alternative is not necessarily retained for the entire volume. According to network needs and site capacities (processing centers and warehouses), it is also possible that a portion of this volume is directed toward other alternatives (Figures 8.7). In this chapter, only used products (s = 2, s = 3 and s = 4) are recovered by service centers. The volume of these products is then split among the processing centers. ALLOCATED PRODUCTS

¦iI FNpsij

¦lL FVpslj

[1.6]

Product volume p in state s (new [s=1] and valorized [s=2] products) allocated (acquisition and replacement) to user zone d by service center j

¦ d D j Dapsd + ¦ d D j Drpsd

[1.2-1.5]

FLEET OF PRODUCTS IN CIRCULATION

Fleet of product p in user zone d of service center j UNUSED PRODUCTS

Product volume p in state s unused from user zone d of service center j

xunpsd FPVpsdj +

FRpsdj [1.7, 1.8]

¦tTj FLSpsdjt

UNRECOVERED PRODUCTS

PRELIMINARY ORIENTATION OF RECOVERED PRODUCTS

[1.9, 1.31]

Product volume p in state s unrecovered from user zone d by service center j

Product volume p in state s recovered at recovery center j (voluntary returns and recovery steps)

¦dDj (FPVpsdj+ ¦tTj FLSpsdt+FRpsdj)

xnrpsd ELABORATED ORIENTATION OF RECOVERED PRODUCT

¦jJ FPpsjk

¦jJ FEpsjk [1.11]

[1.10]

Product volume p in state s oriented to processing center k

Qrep(s=2)k + Qdp(s=2)k + Qdp(s=3)k + Qepsk

¦iI FNpsik

Qrepsk

[1.28]

[1.13]

Product volume p in state s repaired by processing center k

Xrepsk

¦kK FVps

¦lL FVpslk [1.29]

[1.12, 1.26]

Xdpsk

¦kK FEpskk’ [1.26]

Valorized product volume p in state s in transit in warehouse l ¦kK FVpskl [1.34]

¦lL FVpslj

Qdpsk [1.14-1.15]

Product volume p in state s disassembled by processing center k [1.16-1.25, 1.32-1,33]

[1.28]

¦kK FEpskk’

Product volume p in state s recycled/ disposed by processing center k

Xepsk

[1.30]

LEGEND:

Possible product volume:

Figure 8.6. network

Possible product flow:

Principal annual aggregated material flows within the potential logistics

Design of Reverse Logistics Networks

199

The state of a product family volume directs it to a given processing alternative. However, it can be reoriented toward a lower processing alternative, presenting a lower recovery value. In this chapter, three processing alternatives are distinguished and given in descending recovery value: x x x

Repair [s = 2 – finished products only in this work]; Disassembly and refurbishing [s = 2–3 – finished products and spare parts]; Recycling or clean disposal [s = 2–4 – finished products and spare parts].

A higher ranked processing alternative is initially considered; however, when circumstances do not allow this (demand, overflow of recovered products at the processing centers and warehouses), a portion of product family volume can then be directed toward the lower ranked alternative (Figure 8.7). The last possible processing alternative, for all recovered materials, is recycling or clean disposal. End of product family life-cycle is reached when these alternatives are taken. Only recycling can represent here a source of recovery value. This decision rule for processing applies to both finished product and spare part families. The volume of products directed toward valorization centers can be handled in two ways: it can be repaired (s = 2) or disassembled for spare parts refurbishing (s = 2 and s = 3). Repair alternative involves here only finished product families, whereas the disassembly and spare parts refurbishing alternative involves finished product as well as materials (component and assembly modules) families. With the repair alternative, no change is made in the product’s original shape. Only unusable materials (s = 4) are replaced to restore the product quality level to organizational standards. New (s = 1) or valorized (s = 2) materials can be used as spare parts. Finished products are cleaned. A volume of valorized finished product families (s = 2) thus obtained. The disassembly alternative is mainly a question of breaking up the volume of recovered product families to obtain a sufficient volume of valorized spare parts to meet all network needs efficiently. Materials in various states are generated from this alternative (s = 2, s = 3 and s = 4). Unusable materials (s = 4) are directed toward adequate recycling or clean disposal centers, whereas other materials can be used to feed the network with valorized spare parts (s = 2) or can be further disassembled (s = 2 and s = 3) to generate other materials. This alternative includes some adjustments (repair and/or cleaning) before reintroducing this volume of valorized material into the network. The volume of valorized product families (s = 2) obtained (materials and finished products) are used to fill, partially or completely, needs expressed in the network at service and processing centers. Needs can be filled either with new (s = 1) or valorized (s = 2). According to the service type or processing alternative involved, a portion of needs must be filled only with valorized products. The volumes of reusable products resulting from processing alternatives are termed valorized products (s = 2). The volumes of recycled or cleanly disposed products (s = 4) represent the other result of processing alternatives, since volume of materials dedicated to disassembly (s = 3) cannot be used to meet customer needs and can only be further disassembled. If products cannot be further disassembled, they are recycled or cleanly disposed of.

iI

lL

[1.28]

+ ¦ FNpslk

¦ FVpsik

[1.13]

Xdp(s=2)k

[1.18, 1.291.30, 1.33]

Xrea(s=2)k

[1.16, 1.291.30, 1.32]

VALORIZED

[1.21, 1.23]

PRODUCTS

[1.16, 1,20, 1.22, 1.24]

Xrep(s=4)k

PRODUCTS

REJECTED

[1.17]

REFURBISHING

DISASSEMBLY AND

FEp(s=4)kk’

PRODUCTS ORIENTED TO RECYCLING/DISPOSAL

Figure 8.7. Decisions for product sorting at processing centers

[1.19, 1.33]

Xdp(s=4)k

PRODUCTS

Xdp(s=3)k

PRODUCTS

[1.26]

[1.21]

[1.18, 1.33]

Vdp(s=4)k [1.21, 1.25]

Vdp(s=3)k

REJECTED

p(s=3)jk

Zdp(s=4)jk

PRODUCTS

UNVALORIZED

[1.12]

Q p(s=4)k

e

DISASSEMBLED

Z

d

[1.17]

Zdp(s=3)k - Xdp(s=3)k

[1.15]

PRODUCTS AVAILABLE FOR DISASSEMBLY

Qdp(s=3)k

FPp(s=3)jk + FP(pC‰M)(s=2)jk

DISASSEMBLY AND REFURBISHING

tTj

¦ (FPVpsdj + ¦ FLSpsdt + FVRpsdj)

dDj

PRODUCTS DEDICATED TO

Zdp(s=2)k - Xdp(s=2)k

Vdp(s=2)k

Zdp(s=2)jk

PRODUCTS AVAILABLE FOR VALORIZATION

[1.14]

Qdp(s=2)k

[Eq. 1.14]

FPa(s=2)jk - Qrea(s=2)k

PRODUCTS

VALORIZED

FPa(s=2)jk

PRODUCTS DEDICATED TO REPAIR

Qrp(s=2)k

REPAIR

[1.10]

* The term product in this figure is used as volume of a product family (finished product and material).

ELABORATED ORIENTATION - Products oriented through processing activities

PRELIMINARY ORIENTATION – Product* oriented to a processing center

[1.27]

Xep(s=4)k’

PRODUCTS

RECYCLED/DISPOSED

FEp(s=4)jk’

TO RECYCLING/DISPOSAL

PRODUCTS DEDICATED

RECYCLING/ CLEAN DISPOSAL

[1.11]

LEGEND:

Results from processing alternatives:

Material replacement:

Possible product flow reorientation:

Product flow orientation:

Possible product flow orientation:

Possible product volume:

200 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

Design of Reverse Logistics Networks

201

8.4. Conclusion and Future Work The suggested logistics network reengineering aims at structuring a reverse network, while considering supply chain activities. It is inspired by current work on wheelchair allocation and valorization in the Province of Quebec (Canada). Activities related to new product distribution, maintenance, recovery, processing, and redistribution are considered. At the end of this reengineering process, a mathematical model is obtained to deal with decisions related to location of recovery and processing centers, distinguished here as valorization centers and recycling or clean disposal centers, as well as warehouses. Decisions relate also to activities allocation. Processing includes repair of finished products, disassembly, and spare parts refurbishing as well as recycling or clean disposal. New modeling approaches are proposed to evaluate the impacts of using valorized products as alternatives to new products on network performance from an economic and customer service (material accessibility and delivery delay) point of view. Service level is evaluated with the modeling of user zones. An approach needs to be developed to locate and forecast services according to user and product family status. Services include demand (acquisition and product replacement) and material recovery (voluntary return and recovery steps undertaken by the organization). New and valorized products can be used in response to demand. Different values and delivery delays can be associated with these two material types, following the implicated sites. It is even considered that a certain proportion of the demand be filled only by valorized materials, according to ultimate consumer requirements or according to established organizational policies. For recovery steps, the model aims at determining how collection must be carried out with the use of private vehicle fleets and/or logistics service providers. To ensure material flow conservation, a reverse bill of materials is proposed. Probability distribution functions are suggested to consider the fact that different volumes of products in various states can be produced from each processing alternative. It is thus possible to reflect the uncertain character of processing products. The processing alternative is selected on the basis of the recovered product family state and also by considering site capacity, supply, and needs expressed to the network. Greater flexibility is then considered for decisions relating to product volume orientation in reverse networks. Some additional product flow conservation constraints have been proposed to deal with this aspect. Work has been undertaken to develop methodologies to define processing conditions and user zones and to forecast services (demand and recovery), following users and products in circulation status. These methodologies and the mathematical model suggested in this chapter will be validated with data in hand on the wheelchairs in Quebec. Although some new modeling approaches are proposed, thus covering new logistics functions neglected until now, others could be considered. Notably, technology selection would be interesting to approach. Resolution complexity of developed models should also represent an exciting challenge for researchers in the coming years.

202 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

8.5 Guidelines for Practitioners A methodology for designing logistics networks integrating reverse logistics is suggested. It includes the development of a deterministic mathematical programming model. Two major stakes are raised by this methodology: Data collection and processing and resolution of large-scale problems. Data collection and processing needed for the definition of all model parameters represent a significant stage of the methodology. They represent nearly 70% of all modeling efforts. The quality of the parameters will dictate the quality of location/allocation model decisions. Some data might be difficult to obtain such as those related to demand, by distinguishing new and valorized products, recovery and product orientation in networks according to their general state. They are not always collected by current information systems. When they are, they cannot be used directly because of lack of standardized processes, notably for sorting and grading, and in follow-up of product life-cycles. Reverse logistics is generally characterized by a high uncertainty level regarding quantity, quality, variety, and timing of recovery. Information systems and decision-making tools must then be considered to ensure data capture and also to attenuate variability by supporting material flows adequately (Chouinard et al. 2005). Proportions or probability distribution functions that feed the model would thus be established with more precision. Taking incertitude into consideration while designing logistics networks may require the conversion of a determinist model to a stochastic model. However, this raises large-scale problems. Recent sampling strategy, notably the sample average approximation (SAA), may be adapted to the proposed model for solving a huge number of scenarios (Santoso et al. 2005). It would then be possible to evaluate the impact of demand and return volume and of product flow orientation, according to their general state, on the network configuration.

8.6 Acknowledgements The research work reported here was completed thanks to a scholarship of the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Fonds québécois de la recherche sur la nature et les technologies (FQRNT). The authors would also like to thank the management of the Assistive Technology Department (ATD) of the Quebec City Rehabilitation Institute (QCRI), the management of the programs outside Quebec and Technical Aids Program of the Régie de l’assurance maladie du Québec as well as the Société de l’assurance automobile du Québec for their contribution to this work. The authors are indebted to the anonymous reviewers for their many helpful and thoughtful suggestions and comments on an earlier version of this chapter. They would also like to thank François Routhier, Eng. PhD at the ATD of the QCRI, and Dr Angel Ruiz from the CENTOR for their participation in this work.

Design of Reverse Logistics Networks

203

8.7 Appendix 8.7.1 Notation 8.7.1.1 Inferior index C: Component family set (cC). M: Assembly module family set (mM). A: Finished product family set (aA). P: Product family set (pP = C‰M‰A). I: New product supplier set (iI). J: Service and recovery center set (jJ). D: User zone set associated with service or recovery center j (dDj). Tj: Set of logistics service providers for recovery associated with service center j (tTj). K: Processing center set (kK). R: Recycling and clean disposal center set (rR). L: Valorized product warehouse set (lL). N: Network unit set (nN = I‰J‰D‰K‰L). U: External network unit set (uU = I‰D). V: Internal network unit set (vV= J‰K‰L). S: Product family state set (sS). 8.7.1.2 Superior index a: Product acquisition demand. d: Disassembly and refurbishing activity. e: Recycling or clean disposal activity. E: Product flows directed toward recycling and clean disposal centers. LS: Product flows recovered by a logistics service provider. P: Product flows directed toward valorization centers. PV: Product flows recovered by a private vehicle fleet. N: New product family flows required by service centers and processing centers. nr: Unrecovered product. r: Product replacement demand. R: Recovered product family flows following replacement or voluntary return. re: Repair activities. rv: Voluntarily returned product. un: Unused product. V: Valorized product family flows required by service and processing centers. w: Storage activities. z: Processing (repair, disassembly & refurbishing, recycling or clean disposal) and storage activities (z = d‰e‰re‰w). 8.7.2 Data av : Fixed cost of using internal network unit v. bzpsv : Capacity related to product family p in state s of the internal network unit v.

204 Marc Chouinard, Sophie D’Amours, and Daoud Aït-Kadi

dapd; drpd :Expected demand (acquisition, replacement) for product family p in user zone d. fpsnn’ : Unit cost for flow of product family p in state s between node n and n’. gpp’ : Quantity of product family p involved during activities (assembly/disassembly) on product family p'. hpp’ : Feasibility to disassemble product family p from p' (hpp’ = 1 if possible 0 otherwise). qp : Size of product family p in standard unit. rrpsd; rrvpsd : Expected quantity of product family p recovered in state s from user zone d (replacement, voluntary return). vpsv : Unit processing cost for product family p in state s at the internal network unit v. xunpsd : Expected quantity of unused product family p in state s in user zone d. xnrpsd : Expected quantity of unused products family p in state s not-recovered from user zone d. Įapsd; Įrpsd : Expected minimal demand proportion (acquisition, replacement) for product family p answered with valorized materials (s = 2) in user zone d. ijrepsk; ijdpsk : Expected minimal proportion of product family p to be recycled or cleanly disposed of (s = 4) following valorization activities (repair, disassembly, and refurbishing) at processing center k. Ȗrepsk; Ȗdpsk : Expected maximal proportion of product family p dedicated to disassembly (s = 3) generating component or assembly module family which can be reused following valorization activities (repair, disassembly, and refurbishing) at processing center k. 8.7.3 Decision Variables Dapsd; Drpsd : Demand response (acquisition, replacement) to product family p in state s in user zone d. FNpsiv; FVpslv : Quantity of product family p in state s exchanged between supplier i (s = 1) or warehouse l (s = 2) and one of the internal network units v, specifically service center j or processing center k. FRpsdj; FPVpsdj; FLSpsdjt : Quantity of product family p in state s coming from user zone d recovered at service center j (replacement/voluntary return, private vehicle fleet, logistics service provider t). FEpsvk’ : Quantity of product family p dedicated to recycling or clean disposal activities (s = 4) directed to internal network unit v, at service center j or processing center k, toward processing center k'. FPpsjk : Quantity of product family p dedicated to valorization activities (1 K, where K is the total ordering cost when both suppliers are used simultaneously and Ki (i = 1, 2) is the ordering cost when only supplier i is used. Moreover, on the basis of numerical results, they conjecture that when the conditions for convexity of the objective function are not satisfied (i.e., the cost function is not convex everywhere), sole sourcing is optimal. For the newsboy model, concavity of the expected profit function is shown, necessary

Supply Uncertainty and Diversification: A Review

355

optimality conditions are developed, and an approximate solution technique for obtaining the (optimal) order quantities is proposed. They also develop closedform expressions for the approximate solution. Comparing the cost of the approximate solution with the optimal objective value in a few numerical examples, they find that the error does not exceed 10%. As we see, Parlar and Wang (1993) study dual sourcing in the newsboy (i.e., single period) model with random yield. On the other hand, Anupindi and Akella (1993) consider random yield and/or variable lead time, where they analyze dual sourcing in multiperiod inventory models and provide optimal ordering policies. More specifically, Anupindi and Akella (1993) address the operational issue of quantity allocation between two uncertain suppliers and its effects on the inventory policies of the buyer. They consider three models for the supply process as follows. Model I (uncertain lead time): This is a one-delivery contract with all of the order quantity delivered either in the current period with probability ȕ, or in the next period with probability 1 – ȕ. Here, it is assumed that there is a single delivery of all the requirements. Thus, each supplier i, with probability ȕi supplies 100% of the order quantity with zero lead time, and with probability 1 – ȕi supplies nothing. If there is no delivery in this period, all of the order quantity is delivered next period. Therefore, the uncertainty in the supply process translates to a specific form of uncertain lead time (i.e., zero or one period lag) for the buyer. Model II (random yield): This is also a one-delivery contract with a random fraction of the order quantity delivered in the current period, that is, each supplier i delivers a random fraction pi  [0, 1] of the order quantity in this period. The portion of the order quantity not delivered is canceled. This is equivalent to a pure random yield problem. The single period version of this model is similar to the newsboy model analyzed in the second part of Parlar and Wang (1993), where they assume the initial inventory is zero, so they get optimal policies that place orders with both suppliers. Model III (random yield and uncertain lead time): This model is similar to Model II with the remaining quantity delivered in the next period. Here, the firm has to deal with the combined effect of random yield and uncertain lead time. Anupindi and Akella (1993) derive the optimal ordering policies that minimize the total ordering, holding, and penalty costs with back-ordering. They analyze the single period and multiperiod versions of the three models and show that the optimal ordering policy in period t for each of these models is characterized by three regions and two critical numbers, ut and vt, as follows. Denoting the on-hand inventory by x, there exist two order points, ut and vt, such that if x • ut, order nothing; if vt ” x < ut, use only one supplier (order from the cheaper supplier); and if x < vt, order from both suppliers. For the limiting case (as backlog increases) in the single period version of Model I, they derive conditions under which one would continue ordering from one or the other or both suppliers, that is, they show that for a sufficiently large backlog, if the shortage penalty is not too high, then extra units will be ordered only from the supplier with larger marginal benefit (i.e., sole sourcing). Otherwise, if the shortage penalty is high, then one will continue ordering from both suppliers (i.e., dual sourcing). For Model II, they give sufficient conditions for not using the second (more expensive) supplier when the demand and yield distributions have some special form. For the single period

356 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

version of Models II and III with equal marginal ordering costs, they show that when demand is exponential and yields are either normal or gamma, the optimal order quantities follow the ratio rule described by (15.1). Therefore, they extend the result obtained by Gerchak and Parlar (1990) for the EOQ case. While the above papers consider only inventory (replenishment) systems, Gurnani et al. (1996) study an assembly problem where two critical components are required by a firm to assemble the final product whose demand is stochastic. The components can be ordered separately from individual suppliers or in a set (a set refers to the components in the required ratio) from a joint supplier. The assembly stage is assumed to be free, i.e., the firm procures and stores the components and sells complete sets. They consider a supply process identical to the one-delivery contract of Model I in Anupindi and Akella (1993), that is, delivery timing of the suppliers is uncertain (zero or one period lead time). Both single-period and multiperiod problems are analyzed. In the single-period problem, if no order is placed with the joint supplier, the order quantities from the individual suppliers follow an order-up-to policy structure with identical order levels. However, it is optimal to diversify (i.e., order from both the individual and the joint supplier) when the inventory level is below a certain threshold. Under some conditions, the policy structure for the multiperiod problem is shown to be similar to that of the single-period problem, except that the order-up-to levels are not the same. Gurnani et al. (2000) consider the single-period assembly system of Gurnani et al. (1996) with a different supply process. Here, the firm agrees to accept partial shipment (i.e., a random fraction) of the order quantity (random yield), as opposed to delivery in a single shipment with uncertainty in the timing of the deliveries (uncertain lead time). The supply contract assumed in this chapter is identical to the one-delivery contract of Model II in Anupindi and Akella (1993). Since the exact cost function is analytically complex to deal with, they analyze a modified cost function. When all suppliers (i.e., the individual suppliers and the joint supplier) are unreliable in delivery, the authors determine the conditions under which diversification is optimal. All the articles, reviewed so far in this subsection, use dual sourcing, except for Gerchak and Parlar (1990) who introduce an EOQ model with n identical suppliers. The paper by Agrawal and Nahmias (1997) seems to be the first work that considers a random yield situation with multiple (not necessarily identical) suppliers, and tries to deal with supplier selection. Agrawal and Nahmias (1997) address the issue of order splitting in the newsboy model, where the yield of the product delivered from each supplier is random and demand is deterministic. They introduce a model to determine optimal lot sizes and the optimal number of suppliers. The key trade-off addressed in this model is between the reduction in yield uncertainty and the increase in fixed ordering costs, both of which result from placing small orders with a large number of suppliers. Now, we present a slightly modified version of their model. The total order quantity Q is split among n suppliers, i.e., Q ¦in 1 Qi , where Qi > 0, i = 1,…, n, are the order sizes from each ¦in 1U i Qi , where Ui, i = 1,…, n, are assumed to be independent, normal random variables with mean ȝi and

supplier. The total yield from all suppliers is Y

Supply Uncertainty and Diversification: A Review

standard deviation ıi such that 0 d P i r 3 V i d 1 .

357

As a result, Y is normally 2

2

distributed with mean ¦in 1 P i Qi and standard deviation ¦in 1 V i Qi . For each order of Qi > 0 units placed with supplier i, there is a fixed cost Ki and a variable cost ci proportional to the order size; thus, the total purchasing cost is obtained as ¦in 1 ( K i  ci P i Qi ) . The authors analyze two cases, identical suppliers and nonidentical suppliers. When suppliers are identical (i.e., ȝi=ȝ, ıi=ı, ci=c, and Ki=K, for i=1,…, n), the total order size Q is equally split among the suppliers selected, i.e., Qi = Q/n. In this case, the mean of Y is P Q and its standard deviation is V Q n , indicating that yield uncertainty decreases as the number of suppliers increases. They show that the expected profit function denoted by E[Ȇn(Q)] is strictly concave in the order quantity Q, for every n = 1, 2,…, and that the optimal expected profit E{Ȇn[Q*(n)]} is concave in n, where Q*(n) is the profit maximizing order size for a given n. For K = 0, the optimal number of suppliers is infinity and for K > 0, there will be an optimal 1 ” n < ’. They conjecture that when K > 0, the optimal n will be moderate in most cases, and hence, they recommend increasing n by one until the incremental profit is less than or equal to the fixed cost K. In their sensitivity analysis, they show that the optimal profit increases as the parameters of the yield process, ȝ and ı, improve. When suppliers are nonidentical, they show the concavity of expected profit for n = 2 and conjecture that concavity holds for n > 2 as well. Then, for a selected set of suppliers (i.e., a fixed n), they provide the necessary optimality conditions for Qi’s. Moreover, when the unit purchase price from all suppliers is the same, they obtain a result, similar to that of Gerchak and Parlar (1990) for the EOQ model, as follows: 2

V j Pj

V i Pi

Qi

Qj

2

, i, j

1,  , n .

(15.2)

Using a numerical example with two suppliers, they show that diversification (dual sourcing) could increase the expected profit. As we see, Agrawal and Nahmias (1997) consider multiple unreliable suppliers in a model with deterministic demand. Recently, Chen et al. (2001) studied a model with multiple unreliable suppliers and random demand as follows. There are n suppliers, indexed by i = 1,…, n, and each supplier i has a yield ratio of 1 – pi. Specifically, a proportion, pi, of any quantity ordered from supplier i may not be delivered; thus, an order of Qi units placed with supplier i will result in an expected delivery of Qi (1 – pi) units. The suppliers are indexed in increasing order of pi values, i.e., 0 < p1 < … < pn ” 1. Purchasing cost c(pi) and salvage value v(pi) are assumed to be decreasing and convex functions of the defective rate pi. This model incorporates a new feature that is not present in previous random yield models: At a premium of a + bpi per unit (above and beyond the purchasing cost), the supply (delivery) can be guaranteed. Hence, out of Qi units from supplier i, for which we pay a purchasing cost of Qi c(pi), we may choose to guarantee a delivery of qi units

358 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

by paying an additional premium of qi (a + bpi). As indicated in the paper, this feature is analogous to upgrading a defective item to a perfect item through inspection and repair. First, Chen et al. (2001) analyze a single period model in which the decision variables are the order quantities from n suppliers, (Q1,…, Qn), and the number of units to guarantee from each supplier through paying premium, (q1,…, qn). The objective is to minimize the expected total net cost, i.e., purchasing, premium, and penalty costs minus profit and salvage value. Assuming that the order quantities



(Q1,…, Qn) are given, they develop an algorithm to obtain ( q1 ,  , q n ) , the optimal number of units guaranteed from each supplier. Then, they extend the earlier model to include the order quantities (Q1,…, Qn) as decision variables, where they assume that supplier 1 is perfectly reliable: p1 = 0. They are able to develop a greedy algorithm to obtain the optimal order quantities. Letting supplier 1’ be the supplier that has the lowest combined purchasing cost and premium among all the suppliers other than supplier 1, they show that it is optimal: either not to pay any premium to any supplier (if it is economical to select supplier 1); or to pay premiums for all units ordered from supplier 1’ and not to pay any premium to any other supplier (if supplier 1 is ignored and replaced by supplier 1’), that is, it is optimal to guarantee a perfect yield from at most one supplier and not to pay any premium to any other supplier. More importantly, when c(p) and v(p) are linear functions, it is optimal to use only two suppliers, 1 (or 1’) and n, i.e., dual sourcing is optimal. Note that supplier 1 (or 1’ whose units all have guaranteed delivery) provides a perfect yield and supplier n offers the lowest unit purchasing cost. Using numerical examples, they show that even when c(p) and v(p) are not linear, dual sourcing is still very close to optimality, which supports the general practice of dual sourcing. We recall that the model of Agrawal and Nahmias (1997) mainly trades off the reduction in yield uncertainty and the increase in fixed ordering costs; however, there is no fixed ordering cost here. The key trade-off in this model is between the high-quality supply and the low-cost system. Secondly, the single period problem is extended to optimal replenishment over an infinite horizon, where any unsatisfied demand is lost and the objective is to minimize the long-run average cost. Instead of salvage value, they assume that there is a holding cost h(pi), a decreasing and convex function of the defective rate pi. They show that an order-up-to policy is optimal and that to find the optimal replenishment policy in the infinite-horizon case amounts to solving a single period problem. In particular, the optimal order-up-to level can be derived from the greedy algorithm there. Furthermore, when both c(p) and h(p) are linear, dual sourcing (from suppliers 1 and n) is optimal in the infinite-horizon case, just as in the single period case. Burke et al. (2004) and Yang et al. (2005) also study the newsboy model in which the newsboy can purchase the product from multiple suppliers with random yields and different purchasing prices. As in Chen et al. (2001), they address optimal supplier selection.

Supply Uncertainty and Diversification: A Review

359

15.3.2 Random Supplier Availability

In most inventory models, one of the implicit assumptions is continuous supply availability; however, supply availability might be subject to random fluctuations that arise from suppliers equipment breakdowns, labor strikes, embargoes, etc. Whenever the product is purchased from an outside supplier, the supply can be cut off at random times for random durations, or the product may be unavailable due to, among other factors, machine breakdowns or preventive maintenance policies in the production process of the supplier. The reader is referred to Parlar and Perry (1996), Gurler and Parlar (1997), and Parlar (1997) that present examples of random supply availability in practice. Before reviewing the papers in this subsection, one should note that random supply availability can be considered a special case of random yield. To see this, consider an all-or-nothing yield model as follows: YQ = UQ such that Pr{U = 1} = 1 – Pr{U = 0} = p, or equivalently, Pr{YQ = Q} = 1 – Pr{YQ = 0} = p, that is, either the entire order quantity, Q, is delivered with probability p or nothing is delivered with probability 1 – p. Now, if the supplier has random durations of availability and unavailability, and the current state of the supplier is known at any time, then the random availability can be characterized as an all-or-nothing yield model. Here, the all outcome {U = 1} and the nothing outcome {U = 0} are specified by the supplier’s current state, available and unavailable, respectively. Parlar and Perry (1996) study the (s, Q) inventory models where the availability of supply is subject to random interruptions. They analyze both single and multiple supplier models where demand rate is deterministic and the availability status of supply is denoted by ON (available) and OFF (unavailable) periods. It is assumed that the lengths of ON and OFF periods of each supplier are independent exponential random variables and that each supplier alternates between ON and OFF periods independently of other suppliers. Moreover, they assume that the suppliers are identical except for a possible difference in the distribution parameters of their ON and OFF durations. For the sake of exposition, we first explain the policy used for the single supplier model in which the state of the system can be either ON or OFF at any time. When the inventory level drops to s, if the supplier is ON, Q units are ordered, which increase the inventory to s + Q. If the supplier is OFF when the inventory level hits s, the inventory manager has to wait until the supplier becomes available (ON) at which time an order is placed and the inventory is increased to s + Q. In other words, an (s, Q) policy is used when the supplier is available and an (s, S) policy, with S = s + Q, is used when the supplier is unavailable (and becomes available again). When there are two suppliers, the status of the system is determined by one of following four states: x x x x

State 0: both suppliers are ON. State 1: supplier 1 is ON and supplier 2 is OFF. State 2: supplier 1 is OFF and supplier 2 is ON. State 3: both suppliers are OFF.

The policy chosen is denoted by (s, Q0, Q1, Q2) and works as follows. When the inventory drops to the reorder point s and the state of the system is i = 0, 1, 2 (i.e., at least one supplier is available), an order for Qi units, i = 0, 1, 2, is placed (when

360 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

both suppliers are available, Q0 is the total units ordered from either one or both suppliers since purchase prices are the same). On the other hand, if neither of the suppliers is available when the reorder point s is reached (i.e., state 3), then nothing can be ordered. Once the system leaves state 3 and supplier i becomes available, the inventory is increased to s + Qi units, i = 1, 2. They develop the average cost function, consisting of fixed ordering, holding, and shortage (back-ordering) costs, for both single and two supplier models. Using a numerical example, they show that the two supplier model could be cost effective compared to the single supplier model, which again supports the idea of dual (or multiple) sourcing in the presence of supply uncertainty. Furthermore, they analyze a multiple supplier problem, assuming that the suppliers have identical availability characteristics (i.e., n identical suppliers). They develop the average cost function and show that when the number of suppliers becomes large, the cost function of the multiple supplier model reduces to that of the EOQ model. Gurler and Parlar (1997) extend the two supplier model of Parlar and Perry (1996) to the case where quite general forms for the ON (available) and OFF (unavailable) periods are allowed. In particular, they assume that the inventory manager deals with two suppliers that may be either individually ON or OFF, while the distributions of the ON periods for the two suppliers are assumed to be Erlangian and the distributions of the OFF periods for each supplier are assumed to be general. As suggested by the authors, a two supplier market with randomly available suppliers could be justified for products that are imported mainly from two countries. For instance, the United Sates and Japan are the two suppliers for many high-tech products such as computers and communication systems. For various reasons, supply of these products might be interrupted, among which temporary shift of production to other items, increase in exports to other countries, and reduction of production due to local economic changes could be mentioned. They identify the regenerative cycles of the inventory level process and employ the renewal reward theorem to construct the long-run average cost per time. Moreover, they obtain analytic solutions for the large (asymptotic) order quantity where the cost function takes a very simple form. 15.3.3 A General Model

Dada et al. (2003) study the newsboy model with multiple unreliable suppliers, where they introduce a very general notion of supplier unreliability. Here, we briefly explain their model. Consider the newsboy model in which the product can be procured from n independent suppliers offering different purchase prices. Define x x x x

Qi=quantity ordered from supplier i, Ui=reliability of supplier i, Si (Qi , Ui)=supplier i’s supply function, Pi (Qi , Ui)=supplier i’s production output function.

¦in 1 Qi . The supplier reliability Ui is an exogenous construct modeled as either fixed or random.

The total order quantity Q is split among the n suppliers, i.e., Q

Supply Uncertainty and Diversification: A Review

361

The supply function represents the quantity supplied or delivered by supplier i such that Si (Qi , Ui) ” Qi, and is defined as follows: Si (Qi , Ui) = min{Qi , Pi (Qi , Ui)}. The production output function explicitly constrains the quantity that the supplier can deliver; moreover, it allows us to model supplier i’s production capability as either endogenous (i.e., dependent upon Qi) or exogenous, as well as either deterministic or random. Here are two examples: 1. 2.

If Pi (Qi , Ui)=Ui Qi and 0 ” Ui ” 1 is a random variable, then Si (Qi , Ui)=Ui Qi. This is a random yield model. If Pi (Qi , Ui)=Ui and Ui is a random variable, then Si (Qi , Ui)=min{Qi , Ui}. This is a random capacity model. Note that, in the presence of a single supplier, random capacity has been studied by Ciarallo et al. (1994), and the combination of random capacity and random yield has been analyzed by Wang and Gerchak (1996).

The goal is to determine the vector of order quantities (Q1,…, Qn) so that the

expected profit is maximized. If Qi ! 0 , supplier i is selected (active); otherwise,

if Qi

0 , supplier i is not selected (inactive). This translates into the supplier

selection problem, just as in Chen et al. (2001). The objective function is not guaranteed to be concave; however, the necessary (KKT) optimality conditions are established and analyzed. They discuss interesting properties of optimal supplier selection, some of which are mentioned here: 1. 2.

If a given supplier is inactive, then all more expensive suppliers will be inactive. If a given supplier is perfectly reliable (i.e., ˜Si (Qi , Ui) / ˜Qi = 1 for all values of Qi and Ui), then all suppliers more expensive than the perfectly reliable supplier will be inactive. Hence, no more than one perfectly reliable supplier will be active.

They find that, in general, although reliability may influence how much is ordered from an active supplier, cost takes priority over reliability in selecting suppliers. Furthermore, for two active suppliers (dual sourcing), they investigate the impacts



of purchasing price and reliability on the optimal solution (Q1 , Q2 ) , service level, and the total order quantity Q*.

15.4 Discussion Classifying the uncertainty sources in supply, we have reviewed the articles that use multiple sourcing in inventory models to decrease the severe effects of uncertainty in supply. We can identify three major sources of supply uncertainty, uncertainty in supply timing (lead time), uncertainty in supply quantity, and uncertainty in purchase price. When lead time is stochastic, the statistical approach shows that order splitting reduces the mean and variance of the effective lead time

362 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

and consequently, decreases safety stocks. Furthermore, the cost minimization approach shows that order splitting could provide significant savings in holding and shortage costs and thereby total cost. Uncertainty in supply quantity can be categorized as follows: random yield, random supplier availability, and random capacity. Minimizing total cost, researchers suggest that multiple sourcing could be economical in dealing with this type of uncertainty. As one would expect, uncertainty on the supply side complicates analytical analysis of inventory models. In this chapter, we have restricted our attention to single-stage models with upstream uncertainty; however, we should remember that supply chains consist mostly of multiple stages and different sources of uncertainty exist between any two stages. The models reviewed here could be building blocks for modeling and analyzing more complex supply chains. Furthermore, diversification is not the only way to tackle supply uncertainty; multiple sourcing is considered an indirect approach to variability reduction. Firms may prefer to have a long and close relationship with only a single supplier, which is usually advocated by the just-in-time practice. In some cases, there might be only a single supplier (or few suppliers) able to deliver a product. For example, to acquire hightech products such as communication systems, one may not have many choices. If multiple sourcing is not an appropriate approach, firms usually resort to the direct approach to variability reduction, which is investing in uncertainty reduction. For instance, the transportation mode might be replaced by a more expensive one to reduce variability in lead time. Another example is to obtain more stable quality and/or quantity from a reliable but expensive supplier. Our review reveals that most multiple sourcing inventory models consider uncertainty in supply lead time, supply yield, or supplier availability. However, when there is uncertainty in the supplier capacity or purchase price offered by a supplier, diversification has received very little attention in the literature. To our knowledge, the paper by Dada et al. (2003) is the only work that addresses multiple sourcing in the presence of random capacity, and the only studies using multiple sourcing in the presence of random price changes are by Tajbakhsh et al. (2005a, b). The latter studies, which consider random discount prices (a special case of uncertain purchase price), again show that multiple sourcing can be a cost effective approach in the presence of upstream uncertainty, and thus, will be reviewed briefly in this section. We believe that multiple sourcing in the presence of uncertainty in capacity or price deserves more work. In addition, we introduce two topics in this section that could provide new avenues for future research: (1) uncertainty in both lead time and supply quantity; (2) supplier selection. 15.4.1 Uncertainty in Purchase Price

Uncertainty in purchase prices could be an incentive for firms to diversify their supply sources. Intuitively, this would provide savings in purchasing costs, and consequently, inventory costs. Here, we elaborate on this issue by reviewing two recent studies. Tajbakhsh et al. (2005b) consider a continuous review inventory system for a single product with a constant demand rate. There are n suppliers who offer a supplier specific discount price at random times. More specifically, supplier i

Supply Uncertainty and Diversification: A Review

363

( i  {1,  , n} ) offers a discount price ci according to a Poisson process with a rate Ȝi independently of the other suppliers. The product can always be procured at a regular price cL from any of the n suppliers. It is assumed that the regular and discount purchases incur different fixed ordering costs and that replenishment lead times are negligible. Moreover, they assume that the discount offer evaporates quickly so that the inventory manager has to make a decision instantly upon offer. They employ a replenishment policy, denoted by (R, s, Q1,…, Qn), as follows. When a discount offer is made by supplier i, if the on-hand inventory level is less than or equal to a threshold level s, then a replenishment order is placed to raise the inventory to a supplier specific target s + Qi. If the inventory is depleted and no discount is offered, then the inventory manager orders R units at the list (or regular) price. Notice that a list price purchase happens only when the inventory level hits zero. They construct the expected total cost per unit time, which consists of fixed ordering, variable purchasing, and holding costs. They develop necessary

and sufficient conditions under which Qi ! 0 for all i = 1,…, n. Assuming that these conditions are satisfied, they provide closed-form expressions for the optimal





policy parameters ( R , s , Q1 ,  , Qn ) and the minimum expected total cost per

unit time. The expressions derived for Qi ’s and the minimum cost resemble the

EOQ (square root) formulas. Moreover, they show that if ci = cj, then Qi



Qj ,

and if ci < cj, then Qi ! Q j , which are intuitive results. They conduct numerical experiments to examine the cost savings of multiple sourcing compared to sole sourcing. Their results suggest interesting managerial insights some of which are presented here: x x x

Having more suppliers willing to provide discount prices is not necessarily beneficial unless their discount prices are comparable. Savings from multiple sourcing are insensitive to regular fixed order cost. Savings from multiple sourcing can be as much as 55%.

Furthermore, the application of the developed model to optimal supplier selection is addressed. To find the optimal (i.e., most economical) subset of suppliers, they suggest that one could apply the multisupplier model repeatedly to all possible combinations of suppliers since in practice the set of suppliers is expected to be moderate in size. This complete enumerative approach is also suggested by Agrawal and Nahmias (1997). Using a numerical example, the authors show that when one or more of the suppliers are much better than the others in terms of discount parameters (i.e., ci’s and Ȝi’s), the suppliers with higher discount prices must be removed. On the other hand, if all suppliers are comparable in discount parameters, then it is beneficial to increase the number of suppliers. As mentioned before, Guo and Ganeshan (1995) suggest a similar decision rule for determining the optimal number of suppliers among n identical suppliers, when lead times are uniformly or exponentially distributed. They suggest that if lead time reduction is of considerable importance, then it may be preferable to use as many suppliers as possible.

364 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

In the model analyzed by Tajbakhsh et al. (2005b), discount offerings of each supplier follow a Poisson process, but discount price is a deterministic constant. Thus, there is only an uncertainty in the arrival of discount offers. Tajbakhsh et al. (2005a) extend this model to a more general case where they also allow uncertainty in the discount prices. Specifically, in addition to Poisson arrivals of discount offers, suppliers discount prices are assumed to be discrete random variables with known, supplier-specific distributions. They obtain analytical and numerical results similar to those of Tajbakhsh et al. (2005b). 15.4.2 Uncertainty in Both Lead Time and Supply Quantity

In reality, firms have to deal with variability in both supply lead time and supply quantity. Although each of these two sources of uncertainty has been studied separately, combinations of them have received very little attention in the literature. To our knowledge, the paper by Anupindi and Akella (1993) is the only work that addresses this issue. As mentioned before, their third model deals with a very special case of combined random yield and random lead time, where lead time is either zero or one period. We now present an approximate model for a special class of items to demonstrate the impacts of variability in both timing and quantity of delivery. Consider a continuous review inventory model with a constant, deterministic demand, where a reorder point-order quantity (s, Q) control policy is used. Furthermore, suppose that replenishment lead time and supply quantity are independent random variables. For simplicity, we assume that the item under consideration is a “B” item; therefore, as argued in Silver et al. (1998), it is reasonable (although not optimal) to use a prespecified value for Q, and then, determine s. For B items, a sequential determination of Q and s might be justified. If we assume that the quantity delivered follows the random yield model(s) suggested by Silver (1976), we can easily determine the value of Q. Then, using one of the methods introduced in Chapter 7 of Silver et al. (1998), we can determine the value of s. Roughly speaking, in this simple model, yield variability affects the order quantity, thereby cycle stock, whereas lead time variability affects safety stock. As expected, two sources of uncertainty would have more severe effects on inventory costs. In the presence of stochastic lead time, most researchers advocate order splitting for both cycle stock and safety stock reduction and suggest that as lead time variability increases, savings of multiple sourcing will also increase (e.g., see Kelle and Silver 1990a and Ramasesh et al. 1991). Furthermore, order splitting in the presence of yield variability can be cost effective (e.g., see Parlar and Wang 1993). Therefore, one would expect that when both lead time and yield are random, multiple sourcing could result in significant savings. We believe that diversification in more complex models such as the aforementioned model deserves more attention. 15.4.3 Supplier Selection

As shown in the literature, increasing the number of suppliers does not necessarily decrease total costs. Specifically, due to supplier management costs, fixed

Supply Uncertainty and Diversification: A Review

365

ordering cost increases as more suppliers are used, while holding and shortage costs are expected to decrease. To find the leastcost subset of suppliers, the cost savings should be balanced with the increase in ordering costs, and therefore, the optimal subset of suppliers usually consists of a finite number of suppliers. Examples of studies that have addressed the supplier selection problem are as follows: x x x

In the presence of stochastic lead time: Guo and Ganeshan (1995) and Sedarage et al. (1999) In the presence of random yield: Gerchak and Parlar (1990), Agrawal and Nahmias (1997), Chen et al. (2001), and Dada et al. (2003) In the presence of random discount offers: Tajbakhsh et al. (2005a, b)

To obtain analytical results, most researchers have tried to simplify the supplier selection problem. This simplification could be one or more of the following: considering only two suppliers, considering multiple, identical suppliers, and ignoring fixed ordering cost. For more detail, the reader is referred to the paper by Agrawal and Nahmias (1997). Due to the importance of sourcing decisions, we believe that the problem of combined inventory management and supplier selection needs more attention; especially, when both timing and quantity of deliveries are uncertain, optimal supplier selection could provide the inventory manager with significant savings.

15.5 Guidelines for Practitioners The first thing that comes to mind when we talk about uncertainty in supply chains is usually demand uncertainty. This is why many studies in the literature on supply chain management try somehow to address demand uncertainty. Recently, however, it has also been shown that supply uncertainty is of great concern to supply chain managers and practitioners. When procuring raw materials and components from outside suppliers, timing of delivery, quantity and quality of delivery, and purchasing costs are usually the most important factors that management keeps in mind to make wise purchasing decisions. Therefore, upstream uncertainty could translate into three major aspects of supply: (1) uncertainty in timing of supply (known as lead time variability in the literature); (2) uncertainty in the quantity (and/or quality) of supply; (3) uncertainty in the price of the item supplied. There are various reasons why these uncertainties exist: war, political issues, weather conditions, labor strikes, and machine breakdowns in the suppliers’ plants, to name a few. For instance, war and political instability in the Middle East may cause an unprecedented uncertainty in the supply of crude oil. Consequently, the price of crude oil might become uncertain. As another example, we can consider Canada’s automobile industry, where different components are purchased from U.S. suppliers and carried to Canada using road transportation. In this case, extreme weather conditions in winters can easily cause high variability in the replenishment lead time of components, or a truck driver strike can cause a complete interruption in the supply of components.

366 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

To deal with uncertainty on the supply side, many researchers have advocated multiple sourcing (or diversification). In this chapter, we have tried to present a comprehensive review of single stage inventory management models, where multiple sourcing is used to alleviate the effects of upstream uncertainty. As can be seen from the literature, supply uncertainty by itself complicates the analysis of inventory systems, and having multiple suppliers further adds to the problem. Hence, our purpose is to provide practitioners with the simple, theoretical results available in the literature. Some of these models are too simple to be readily applicable to real problems, and by no means can solve problems encountered in complex supply chains in practice; however, they provide managerial insights for practitioners in charge of decision-making and lay out a building block for the analysis of real-world problems. In addition, we have suggested new directions for further investigation. First, models considering uncertainty in both timing and quantity of supply need to be further developed and analyzed because in practice these two sources of uncertainty are most likely present simultaneously. Second, since firms could take advantage of multiple supplier models in selecting their suppliers and making their purchasing decisions, we suggest that the combination of inventory models with upstream uncertainty and the supplier selection problem need further attention from researchers and practitioners.

15.6 References Agrawal N, Nahmias S, (1997) Rationalization of the supplier base in the presence of yield uncertainty. Production and Operations Management 6:291–308. Anupindi R, Akella R, (1993) Diversification under supply uncertainty. Management Science 39:944–963. Bollapragada S, Morton TE, (1999) Myopic heuristics for the random yield problem. Operations Research 47:713–722. Burke GJ, Carrillo JE, Vakharia AJ, (2004) Sourcing decisions with stochastic supplier reliability and stochastic demand. Working Paper, Warrington College of Business Administration, University of Florida. Chen J, Yao DD, Zheng S, (2001) Optimal replenishment and rework with multiple unreliable supply sources. Operations Research 49:430–443. Chiang C, Benton WC, (1994) Sole sourcing versus dual sourcing under stochastic demands and lead times. Naval Research Logistics 41:609–624. Ciarallo FW, Akella R, Morton TE, (1994) A periodic-review, production planning model with uncertain capacity and uncertain demand - optimality of extended myopic policies. Management Science 40:320–332. Dada M, Petruzzi NC, Schwarz LB, (2003) A newsvendor model with unreliable suppliers. Working Paper, College of Business, University of Illinois at Urbana-Champaign. Fong DKH, Gempesaw VM, Ord JK, (2000) Analysis of a dual sourcing inventory model with normal unit demand and Erlang mixture lead times. European Journal of Operational Research 120:97–107. Fong DKH, (1992) A note on exact moment computation for normal lead times in the twosupplier case. The Journal of the Operational Research Society 43:63–69.

Supply Uncertainty and Diversification: A Review

367

Ganeshan R, Tyworth JE, Guo Y, (1999) Dual sourced supply chains: The discount supplier option. Transportation Research Part E 35:11–23. Gerchak Y, Parlar M, (1990) Yield randomness, cost trad-offs, and diversification in the EOQ model. Naval Research Logistics 37:341–354. Guo Y, Ganeshan R, (1995) Are more suppliers better? The Journal of the Operational Research Society 46:892–895. Gurler U, Parlar M, (1997) An inventory problem with two randomly available suppliers. Operations Research 45:904–918. Gurnani H, Akella R, Lehoczky J, (1996) Optimal order policies in assembly systems with random demand and random supplier delivery. IIE Transactions 28:865–878. Gurnani H, Akella R, Lehoczky J, (2000) Supply management in assembly systems with random yield and random demand. IIE Transactions 32:701–714. Kabak IW, Weinberg CB, (1972) The generalized newsboy problem, contract negotiations and secondary vendors. AIIE Transactions 4:154–157. Kelle P, Silver EA, (1990a) Safety stock reduction by order splitting. Naval Research Logistics 37:725–743. Kelle P, Silver EA, (1990b) Decreasing expected shortages through order splitting. Engineering Costs and Production Economics 19:351–357. Lau H-S, Lau AH-L, (1994) Coordinating two suppliers with offsetting lead time and price performance. Journal of Operations Management 11:327–337. Lau H-S, Zhao L-G, (1993) Optimal ordering policies with two suppliers when lead times and demands are all stochastic. European Journal of Operational Research 68:120–133. Lau H-S, Zhao L-G, (1994) Dual sourcing cost-optimization with unrestricted lead time distributions and order-split proportions. IIE Transactions 26:66–75. Minner S, (2003) Multiple-supplier inventory models in supply chain management: A review. International Journal of Production Economics 81-82:265–279. Pan AC, Ramasesh RV, Hayya JC, Ord JK, (1991) Multiple sourcing: The determination of lead times. Operations Research Letters 10:1–7. Parlar M, Perry D, (1996) Inventory models of future supply uncertainty with single and multiple suppliers. Naval Research Logistics 43:191–210. Parlar M, Wang D, (1993) Diversification under yield randomness in inventory models. European Journal of Operational Research 66:52–64. Parlar M, (1997) Continuous-review inventory problem with random supply interruptions. European Journal of Operational Research 99:366–385. Ramasesh RV, Ord JK, Hayya JC, Pan A, (1991) Sole versus dual sourcing in stochastic lead time (s, Q) inventory models. Management Science 37:428–443. Ramasesh RV, Ord JK, Hayya JC, (1993) Note: dual sourcing with nonidentical suppliers. Naval Research Logistics 40:279–288. Sculli D, Shum YW, (1990) Analysis of a continuous review stock-control model with multiple suppliers. The Journal of the Operational Research Society 41:873–877. Sculli D, Wu SY, (1981) Stock control with two suppliers and normal lead times. The Journal of the Operational Research Society 32:1003–1009. Sedarage D, Fujiwara O, Luong HT, (1999) Determining optimal order splitting and reorder level for n-supplier inventory systems. European Journal of Operational Research 116:389–404. Silver EA, Pyke DF, Peterson R, (1998) Inventory Management and Production Planning and Scheduling. Wiley, New York, Third Edition. Silver EA, (1976) Establishing the order quantity when the amount received is uncertain. INFOR 14:32–39.

368 M. Mahdi Tajbakhsh, Saeed Zolfaghari, and Chi-Guhn Lee

Tajbakhsh MM, Lee C-G, Zolfaghari S, (2005a) An inventory model with random discount offerings. Working Paper, Department of Mechanical and Industrial Engineering, University of Toronto. Tajbakhsh MM, Lee C-G, Zolfaghari S, (2005b) A multi-supplier inventory model with random discount offers. Technical Report, Department of Mechanical and Industrial Engineering, University of Toronto. Wang Y, Gerchak Y, (1996) Periodic review production models with variable capacity, random yield, and uncertain demand. Management Science 42:130–137. Yang S, Yang J, Abdel-Malek L, (2005) Sourcing with random yields and stochastic demand: a newsvendor approach. Working Paper, Stillman School of Business, Seton Hall University. Yano CA, Lee HL, (1995) Lot sizing with random yields: A review. Operations Research 43:311–334. Zhao L-G, Lau H-S, (1992) Reducing inventory costs and choosing suppliers with order splitting. The Journal of the Operational Research Society 43:1003–1008.

16 Quantitative Robustness Index Design for Supply Chain Networks Ming Dong and F. Frank Chen

Abstract:

As an event-driven system, a supply chain network will face uncertainties inside the supply chain and also unexpected events outside the supply chain network such as contingency, disruption, and disaster. These uncertainties and unexpected events have negative impacts on the survival and performance of supply chain networks. As a dynamic system, a supply chain network will evolve over time. Nodes and links may be added and deleted, or part of networks can be disconnected. The functional performance of nodes and links of supply chain networks may deteriorate quickly under unexpected events. Providing resilient capability against failures is a critical issue for supply chain networks since a single failure may propagate along the chain and cause a series of severe losses in network performance and structure. Robustness of a supply chain network is an important research issue, a vulnerable supply chain network may not be able to operate at all. The goal of this chapter is to investigate the relationship between robustness metrics and basic network parameters. A systemwide approach is presented to quantifying the robustness of supply chain networks. This approach considers both network structural and network functional parameters. Metagraphs are employed to calculate the structural robustness of nodes, and a topological index is used to capture the robustness of the overall network structure. Finally, the integrated systemwide network robustness index can be obtained by the proposed algorithm. Several hypothetical supply chain networks are employed to demonstrate the proposed approach.

16.1 Introduction Nowadays, a supply chain network will face uncertainties inside the supply chain and also unexpected events outside the supply chain network such as contingency, disruption, disaster, and terrorism. These uncertainties and unexpected events have negative impacts on the survival and performance of supply chain networks. For

370 Ming Dong and F. Frank Chen

example, machine breakdown in a manufacturing plant will delay product delivery and may result in lost sales since customers might not receive their orders on time. The earthquake in Taiwan disrupted wafer fabrication, consequently delayed computer assembly in the United States due to insufficient material supply. The 9/11 terrorism attack in the United States closed boundary between the U.S. and Canada several days for security reasons. This further led Ford Motor Company shut down its five plants temporarily since it could not get enough parts from its suppliers located in Canada (Sheffi 2001). These facts show that the robustness of a supply chain network is an important research issue, a vulnerable supply chain network may not be able to operate at all. An important step for solving this problem is to maximize its chances of overall survival by adapting its configuration to environmental pressures. In this regard, the development of corresponding quantitative metrics on a supply chain’s ability to cope with deviation and disruption is necessary. The survival of a supply chain network depends on the ability of each node to communicate or interact with all other nodes through different links in an efficient and robust manner. The interactions can be through material flow, information flow, or capital flow. Here, two critical measures of a network can be identified, efficiency and robustness. By robustness we mean the extent to which the network is able to carry out its functions despite some damage done to it, such as the removal of some of the nodes and/or links in a network. However, these are often conflicting objectives. In recent years, much attention has been paid to efficiency improvement and lean technique employment to reduce inventory costs and improve overall supply chain performance. However, this transformation to closer coordination between supply chain members has left companies more vulnerable to unanticipated events. Although supply chains have become leaner, they have also become less robust. While there has been extensive work on the measurement of supply chain performance, little of this work has focused on measuring a supply chain’s robustness, i.e., its ability to cope with deviation and disruption. Gaonkar and Viswanadham (2003) classified the risks in supply chains into three categories: 1). Deviation: One or more parameters such as demand and lead time stray from their expected or mean values without any changes in the underlying supply chain structure. Examples include variations in supply/demand processes and production/transportation lead time variability. 2). Disruption: A certain part of production, warehousing and distribution, or transportation facilities becomes unavailable due to unexpected events caused by humans or nature, and a certain part of the supply chain structure is changed. An example is production disruption due to a fire in a Toyota’s supplier’s factory. 3). Disaster: A temporary irrecoverable shutdown of the supply chain network occurs due to unforeseen catastrophic systemwide events, and the structure of the supply chain is radically transformed. Typical examples include the terrorist attack on the 11th of September 2001, disruption of IC chip production due to the earthquake in Taiwan, and recent hurricanes, Katrina and Rita. Gaonkar and Viswanadham (2003, 2004) argue that supply chains need to be robust at three levels, strategic, tactical, and operational, and they need to handle minor regular operating deviations and major disruptions at each of these levels.

Quantitative Robustness Index Design for Supply Chain Networks

371

Sheffi (2001) provides a dual sourcing approach to handle supply risks in the face of unforeseen events in the supply chain caused by international terrorism. To achieve just-in-time delivery in supply chains, a systematic approach is needed to reduce the variability in different stages of supply chains; then the robustness of supply chains can be improved. Yu and Luh (2004) apply a variance control technique to calculate the variability of lead times. However, only lead time variability is considered in their paper. Pai et al. (2003) argue that risk analysis can be broadly classified into three categories, vulnerability assessment, consequence analysis, and countermeasure analysis and implementation. A complete causal tree structure and an inference engine are required to determine the most probable path and the relative probabilities of occurrence for any chain of events. Inference can be done using three techniques: Bayesian network, fuzzy logic, and hybrid networks. A more general conclusion is given by Souter (2000) and Stauffer (2003): Disruptions affect an individual organization and also affect the different stages of a supply chain along the chain. From a broader perspective, although disruptions come from different sources, they will affect different components of supply chains. Although there are some papers discussing different issues on disruptions in supply chains, there is a lack on the development of quantitative robustness metrics of supply chain networks. This study presents a three-stage systemwide approach to developing the necessary robustness-related indices. First, a network functional robustness index that considers node service level and link lead time is established. At the second stage, network structural robustness indices characterizing a network’s total connectivity and topological structure are developed. Third, a comprehensive integrated network robustness index is constructed from structural and functional robustness indices. This chapter is organized as follows: Section 16.2 provides the definition of supply chain network robustness. The framework used to characterize the robustness metrics is presented in Section 16.3. Section 16.4 is devoted to the development of functional robustness of supply chain nodes and links. Structural robustness indices of networks are provided in Section 16.5. Section 16.6 gives an integrated robustness index of whole supply chain networks. Example supply chain networks are used to demonstrate the proposed robustness indices in Section 16.7. Finally, Section 16.8 presents conclusions and points out future research lines.

16.2 Definition of Supply Chain Network Robustness The vulnerability of a supply chain to a disruptive event consists of two quantifiable measures, the magnitude of the event and the likelihood of its occurrence. Reducing either of them could reduce vulnerability and correspondingly, increase the robustness. Vulnerability can be viewed as the impact of a demand surge or disruption conditioned by the probability of its occurrence. An aggregate measure of vulnerability takes into account the relevant vulnerabilities at each node of a supply chain network.

372 Ming Dong and F. Frank Chen

Two concepts can be used to characterize a supply chain’s robustness to demand surge or disruption. (1) The magnitude of a performance measure deviates from its expected value under the surge demand or disruption. (2) The likelihood of a performance measure’s deviation from its expected value at a particular magnitude occurs in a specified interval. Surge demand or disruption will cause some supply chain system performance measures to deviate from their expected values. Supply chains with different degrees of robustness have different behaviors under surge demand or disruption. In Figure 16.1, the performance of all three supply chain will deteriorate with an increase in the magnitude of disruption. However, they exhibit different behaviors: supply chain 1’s performance deteriorates slowly, and therefore, it has the largest robustness; it has the smallest distance from the expected performance (i.e., deviation 1); supply chain 3’s performance deteriorates rapidly and therefore it has the smallest robustness; it has the largest distance from the expected performance (i.e., summation of deviation 1, deviation 2, and deviation 3); the performance of supply chain 2 is in between; its performance deteriorates linearly with an increase in the magnitude of disruption, it has a distance to the expected performance (i.e., the summation of deviation 1 and deviation 2). It can be seen that the concave shape of performance deteriorating behavior (e.g., supply chain 1) is the most desired. And the robustness of a supply chain is determined by the distance between its desired performance curve and actual performance curve under surge demand or disruption. Expected performance

Performance Deviation 1 Deviation 2

Supply chain 1 Supply chain 2

Deviation 3 Supply chain 3 No disruption

Magnitude of

(normal condition)

disruption

Figure 16.1. Performance curves of supply chains under surge demand or disruption

There are many decision variables whose changes affect supply chain performance. Some are critical to supply chain performance, some are not. Surge demand or disruption will change the values of some decision variables such as order quantity and reorder point. Correspondingly, the changes in values of these variables from their optimal points will worsen supply chain performance. Figure 16.2 illustrates this situation: variable 3 is the most sensitive variable for supply

Quantitative Robustness Index Design for Supply Chain Networks

373

chain performance; a small change in variable 3 from its optimal value will greatly deteriorate supply chain performance; on the other hand, variable 1 is insensitive to supply chain performance, its changes almost do not alter the performance values; variable 2 is in between. Performance

Expected performance

Variable 1 Variable 2 Variable 3

Optimal values of variables

Deviations of variables from optimal values

Figure 16.2. Sensitivity of variables to supply chain performance

Expected performance

Performance

Curve 1

0

Curve 2

Curve 3

Time needed to resile

Figure 16.3. Resiliency of different supply chains

Under surge demand or disruption, the performance of supply chain will deviate from their expected values; supply chains with large robustness should be able to resile to their original performance with in a short time (see curve 1 in Figure 16.3), supply chains with small robustness will take a much longer time to resile to their expected performance (see curve 3 in Figure 16.3). Therefore, the time needed to resile from disruption is another indicator of supply chain robustness.

374 Ming Dong and F. Frank Chen

16.3 Framework for Characterizing Supply Chain Network Robustness Supply chains may differ in their network structure (serial, parallel, assembly and arborescent distribution), product structure (levels of bill of materials), transportation modes, and degree of uncertainty that they face. However, they have some basic elements in common (Dong and Chen 2005). 16.3.1 Nodes

A supply chain network can be viewed as a network of functional nodes connected by different links (see Figure 16.4). Generally, there are four types of nodes: (1) supplying nodes: They either provide raw materials from outside suppliers or deliver finished products to retailers; (2) fabrication nodes: They transform raw materials into components; (3) assembly nodes: They assemble the components into semifinished products or finished goods; and (4) distribution nodes: They deliver the finished products to warehouses or to customers. These four types of nodes can be treated as the building blocks for modeling the whole supply chain. All manufacturing nodes (either fabrication nodes or assembly nodes) in the network are capable of building parts, subassemblies, or finished goods in either a make-to-stock or make-to-order mode. The service level such as the fill rate at a node (fill rate is the fraction of customer orders that is filled by on-hand inventory) can be used to characterize its performance reliability (or functional robustness). The structural robustness of a node can be defined by its failure rate under disruptive events. 16.3.2 Links

All nodes in the supply chain network are connected by links that represent supply and demand processes. Three types of flow can go through the links: Material flow, information flow, and capital flow. The material-flow link represents that an upstream node provides replenishments to the specified downstream node (see Figure 16.4). The reliability of the links plays a key role in the robustness and resiliency of a supply chain under surge demand or disruption. The functional robustness of a link could be described by its transit-time reliability to the destination downstream node. The failure rate under disruptive events could be used as an indicator of link structural robustness. 16.3.3 Modeling Supply Chains Using Metagraphs

A metagraph is a graphical structure that represents directed relationships between sets of elements. The theory of metagraphs is available in Basu and Blanning (1994, 1997, 2000).

Quantitative Robustness Index Design for Supply Chain Networks

375

Assemble-to-order Store28

R

F

R

Store23

Store22

Store14

Supplier1

A Store11

Store27

R

F

R

Store21

Store20

Store13

Product differentiation

A

Supplier2

Store10

Make-to-stock

R Store26

Store12

R

Supplier3

Store19

A

R

Store17

Store18

Store25

R

R

Store9

A

Supplier4

R Store24

R

F Store15

Store6

Store3

R

D

Store7

Store8

Store16

D

Assembly

Store5

Store2

R

D

Supplier5

Fabrication

Store4

Store1

Distribution

Icon representation

Activity, operation, task R: Receive; A: Assemble; F: Fabrication; D: Distribution. Stores that perform some operations and stock a finished good stock keeping unit Stores that perform some operations and stock a work-in-process stock keeping unit Customers

Material flow

Raw material suppliers

Sites

Figure 16.4. An integrated modeling framework for supply chain networks

16.3.3.1 Basic Definitions Given a finite generating set X={xi, =1...I}, a metagraph is an ordered pair M = in which E ={ek, k = 1...K} is a set of edges. Each xi  X is called an element.

376 Ming Dong and F. Frank Chen

Each edge is an ordered pair ek = in which Vk Ž X is the invertex of the edge ek and Wk Ž X is the outvertex of ek. The coinput of any x  Vk is Vk\{x} and the cooutput of any x  Wk is Wk\{x}. A metagraph is a representational tool and also an analytical one. By defining edges as pairs of sets of elements rather than its elements (as in digraphs) or as single sets of elements (as in hypergraphs), a mathematical structure with substantial properties can be obtained. These structural properties such as paths and connectivity provide an analytical foundation far richer than that provided by digraphs and hypergraphs. 16.3.3.2 Supply Chain Network Modeling by Metagraphs Metagraphs can be used to model supply chain networks with different structures. Figure 16.5 gives the metagraph models of six basic supply chain network structures. Figure 16.5(a) represents that product B can be obtained by processing component A; in Figure 16.5(b), B can be obtained from A, and D can be obtained by processing C; Figure 16.5(c) represents the assembly operation, i.e., product C can be obtained by assembling components A and B; A, B, and C represent different suppliers in Figure 16.5(d); in Figure 16.5(e), distribution center A provides products to different locations: B, C, and D; Figure 16.5(f) is a disassembly structure in which components A and B can be obtained by disassembling product C.

A A

A

D

B

C C

(a) Serial processing

(b) Parallel processing

A B

B

B

(c) Assembly

B D

C

A

C D

(d) Multiple supplying

(e) Distribution

A C B

(f) Disassembly

Figure 16.5. Metagraph models of basic supply chain network structures

Quantitative Robustness Index Design for Supply Chain Networks

377

16.4 Functional Robustness of Nodes and Links 16.4.1 Functional Robustness Index of Nodes

A node’s fill rate is an indicator of its service level or performance reliability. When a surge demand or disruption happens in a node, its average back-order level will increase rapidly. The back-order level will affect its service level and supply lead time. If the (Q, r) inventory control policy is adopted at this node, then its average back-order level can be expressed as (Hopp and Spearman 2000) 1 Q

B (Q, r )

r Q

¦ B( x)

x r 1

1 [ B (r  1)  ˜ ˜ ˜  B (r  Q)] , Q

(16.1)

where x is the customer demand at a particular time, r is the reorder point, and Q is the order quantity (or production lot size). Correspondingly, the expected service level S is as follows: S (Q, r ) 1 

1 [ B (r )  B (r  Q)] . Q

(16.2)

Under a base stock inventory policy, the average back-order level is given by (Hopp and Spearman, 2000) f

B( R)

¦ ( x  R) p( x)

Tp( R )  (T  R)[1  G ( R)] ,

(16.3)

x R

where x is the customer demand at a particular time, T is mean the lead time demand, R is the base stock level, LT is the lead time, p(x)= Prob{demand during lead time LT equals x} and G(x)= Prob{demand during lead time LT is less than x}. Note that the base stock level R = r+1, therefore, in general the average backorder level is a function of Q and r. The corresponding expected service level S under a base stock inventory policy is given by

S ( R)

P( X  R)

­G ( R), if G is continuous ® ¯G ( R  1) G (r ), if G is discrete.

(16.4)

16.4.2 Functional Robustness Index of Links

Transit time is a key measure to describe the performance robustness of links. The longer the transit time, the greater the probability that a disruptive event will

378 Ming Dong and F. Frank Chen

intervene to cause problems. Suppose that there is a “need by” date which will be either the original date or a new one agreed upon between supplier and customer. In either case, for any given time needed to meet an order, the longer the transit time, the more vulnerable will be the system. The critical measure is the ratio of “transit time for the materials needed to satisfy the order” to “time available to meet the surge order.” The smaller this ratio, the better the chances of meeting demand under a disruptive environment. By Little’s law (here, wait W is analogous to cycle time, back-order level B(Q, r) is analogous to WIP, and demand rate DR is analogous to throughput), delay (denoted as W) due to back-ordering can be described by (Hopp and Spearman 2000) B(Q, r ) . DR

W

(16.5)

Let l be the nominal transit time, L be the transit time (including back-order delay) for an order from an upstream node to a downstream node. Then the effective transit time is E[ L]

" W .

(16.6)

Under disruptive situations, there are two possibilities: Either order sees no delay and transit time is l, or it encounters a back-order delay and has transit time l+y (y is a deterministic delay). Since the probability of the order encountering a back-order is 1-S, E[ L ]

( S )(")  (1  S )("  y )

"  (1  S ) y .

(16.7)

From Equations (16.6) and (16.7), we can obtain y

W . (1  S )

(16.8)

Further, we can get the variance of L in the following: E[ L2 ]

S" 2  (1  S )("  y ) 2 ,

2

Var ( L)

E[ L2 ]  E[ L] 2

S (1  S ) y 2

§ S ·§ B (Q, r ) · ¸ . ¨ ¸¨ © 1  S ¹© D ¹

(16.9)

Therefore, the functional robustness of a link can be characterized by the mean and the standard deviation of L. Here, Std(L) is the standard deviation of L, which can be calculated from (16.9). When L increases, the performance robustness of a

Quantitative Robustness Index Design for Supply Chain Networks

379

link will decrease. In detail, the link’s transit time robustness will decrease if service level S and back-order level B increase.

16.5 Structural Robustness of Supply Chain Networks 16.5.1 Structural Robustness of Nodes

Accessibility is a key element of nodes in supply chain networks since it is a direct expression of mobility either in terms of material, information, or capital. Accessibility can be defined as the measure of a node to be reached by or to reach different nodes. The most basic measure of accessibility involves network connectivity where a network is represented as a connectivity matrix (C1), which expresses the connectivity of each node with its adjacent nodes. The number of columns and rows in this matrix is equal to the number of nodes in the network. A value of 1 is given for each cell where there is a connected pair and a value of 0 for each cell where there is an unconnected pair. The summation of this matrix provides a very basic measure of accessibility, also known as the degree of a node: N node

¦c

C1

ij

,

(16.10)

j 1

where C1=degree of a node, cij =connectivity between node i and node j (either 1 or 0), and Nnode=number of nodes. However, the connectivity matrix does not take into account all possible indirect paths between nodes. Under such circumstances, two nodes could have the same degree but may have different accessibilities. To consider this attribute, the total connectivity matrix (T) is used to calculate the total number of paths in a network, which includes direct as well as indirect paths (Rodrigue 2003). The total connectivity matrix can be used to measure the structural robustness of nodes. Its calculation involves the following procedure: D

T

¦C

k

,

(16.11)

1 ij

1 ˜ c kji1 , (k z 1) , cij is the first-order connectivity matrix,

k 1 n

where C k

n

¦¦ c i 1 j 1

c kji1 is the (k-1)th order [(k˰ 1)˰ linkages paths] connectivity matrix. D = the

diameter of the network. Diameter is defined as the length of the shortest path between the most distant nodes of a network. D measures the extent of a network and the topological length between two nodes.

380 Ming Dong and F. Frank Chen

However, the total connectivity matrix doesn’t take link failure rates into account. In this regard, the adjacency matrix of supply chain metagraph models plays an important role. The adjacency matrix A of a metagraph is a square matrix with one row and one column for each element in the generating set. One purpose of the matrix is to indicate whether there are any edges connecting a row element to a column element, that is, whether there are any edges whose invertex contains the row element and whose outvertex contains the column element. It differs from the adjacency matrix for digraphs in that each of its components conveys more information than merely the existence or nonexistence of an edge connecting the row and column elements. It also conveys information about other elements of the edges. Each aij element is a set of zero or more ordered triples, one for each edge connecting xi to xj; the first component of the ordered triples is the set of all elements in the invertex other than xi, the second component is the set of all elements in the outvertex other than xj , and the third component denotes the edge ek connecting xi and xj. Thus, if there is no edge connecting xi to xj, aij=I; if one or more edges connect xi to xj, aij will contain one triple corresponding to each such edge. The definition of the adjacency matrix of a metagraph is given as follows: Definition 1. Given a generating set X={xi, =1...I} and a metagraph M = with E ={ek, k = 1...K}, the adjacency matrix, A, of M is an I u I matrix with each aij for i, j^1...I} defined as follows (Basu and Blanning 1994): K

aij

 (D

)

ij k

k 1

where

(D ij ) k

­ Vk \ {xi }, Wk \ {x j }, ek , ® ¯I

if both xi  Vk and x j  Wk otherwise.

Metagraphs defined on the same generating sets can be combined, and a multiplication operator has been defined in terms of the adjacency matrix A of a metagraph, which can also be used to calculate the powers of an adjacency matrix. Each member of An, anij, is a set of zero or more triples, one for each path of length n connecting xi to xj. The first component of the triple is the coinput of xi in the path, the second component is the cooutput of xj, and the third component is the path. Another useful construct is the closure A* of the adjacency matrix, which is formed by adding successive powers of the matrix. Thus the closure is A+A2+A3+A4+... (at some point all successive An will have null elements and the summation can be terminated, since the multiplication operator does not allow more than one traversal of a cycle). Thus, the matrix A* discloses all paths of any length connecting any two elements, and it can also be used to identify different paths. The calculation of A* is as follows (Basu and Blanning 1994): Definition 2. Consider a generating set X={xi, i=1,…I} and a metagraph M= with adjacency matrix A. Then anij =I iff there is no path of length n from xi to xj. Otherwise, for each triple in anij there is a path from xi to xj such that (1) The first member of anij is the set of coinputs of the path, (2) The second member of anij

Quantitative Robustness Index Design for Supply Chain Networks

381

is the set of cooutputs of the path, (3) The third member of anij is the path and all such paths are represented by a triple in anij. Under above conditions, the closure A* of the adjacency matrix can be calculated as follows: A*

A  A 2  ...  A

N edge

,

where Nedge is the number of edges in M.

(16.13)

In analyzing adjacency matrices and their powers, some operators are needed to be defined on triples to identify their components. These operators include In( ), Out( ), Path( ), and Num(Path( )). For example, suppose that we have the square 2 consists of two ordered triples, corresponding to the two paths of A2=AuA, and a16

^ a , n 1,2` . We further suppose that a I , ^x , x `, e , e and a ^x `, ^x , x , x `, e , e . Then, Path a e , e and Num Path a 2 .

2 length 2 from x1 to x6, that is a16 2 16 1

3

2 16 2

4

1

1

2 16 n

2 16 2

3

2

3

4

5

1

2

2 16 2

2

An algorithm used to obtain the metagraph model-based total connectivity matrix (TM) is given as follows: Step 1. For a given supply chain network, construct the corresponding metagraph model using the basic building blocks provided in Figure 16.2. Step 2. Determine every element of the adjacency matrix A for a given metagraph-based supply chain network model by using Equation (16.12). Step 3. Calculate the closure A* of the adjacency matrix A by using Equation (16.13).

^ a , n

Step 4. For each element a ij* (i.e., aij*

* ij n

`

1,2,..., N ) of the closure A*

of the adjacency matrix A, if a ij* z I , we first get the path of a triple using operator Path( ), then we find the number of edges in the path using operator Num(Path( )), i.e., Ne=Num(Path( )). If edge ek’s failure rate is Pek , then the structural robustness Ne

of the path Rp can be calculated as R p

– 1  P . ek

Thus, for each a ij* z I , the

k 1

corresponding connectivity index can be given as follows: * Num Path a º ij n N ª « CI aij* 1  Pek », where e k  Path aij* n . « » n 1« k 1 ¬ ¼» An example supply chain network with eleven nodes and ten links is used to illustrate the computation of the total connectivity matrix TM in Figure 16.6. Table 16.1 gives the failure rates of the example supply chain network.

¦

–





382 Ming Dong and F. Frank Chen Table 16.1. Failure rates of edges of the first example supply chain network Edges

e1

e2

e3

e4

e5

e6

e7

e8

e9

e10

Failure Rate

0.03

0.07

0.04

0.03

0.07

0.11

0.05

0.06

0.07

0.05

Based on Step 2 of the above algorithm, the adjacency matrix A of the example supply chain network in Figure 16.6 is provided in Table 16.2. By using (16.13), the closure A* of the adjacency matrix A is given in Table 16.3. From Step 4 of the developed algorithm, the corresponding total connectivity matrix TM and the structural connectivity weights of nodes are given in Table 16.4. Distribution center 1 e1

A

e3

D

H

Retailer 1

F

e8

Manufacturer 1

Supplier 1

e7

B

Retailer 2

e6

e5 Transshipment links

I

J

e9

Supplier 2

e2

C

e4

E

Retailer 3

G

Manufacturer 2

Distribution center 2

Supplier 3

e10

K

Retailer 4

Figure 16.6. The first example supply chain metagraph model Table 16.2. The adjacency matrix A of the first example supply chain network A

B

C

D

E

F

G

H

I

J

K

A

I

I

I

¢I, I, ¢e1²²

I

I

I

I

I

I

I

B

I

I

I

I

I

I

I

I

I

I

C

I

I

I

I

¢C, I, ¢e2²² ¢B, I, ¢e2²²

I

I

I

I

I

I

D

I

I

I

I

I

¢I, I, ¢e3²²

I

I

I

I

I

E

I

I

I

I

I

I

I

I

I

I

F

I

I

I

I

I

I

¢I, I, ¢e4²² ¢I, I, ¢e5²²

¢I, I, ¢e7²²

¢I, I, ¢e8²²

I

I

Quantitative Robustness Index Design for Supply Chain Networks

383

Table 16.2. (continued) G

I

I

I

I

I

¢I, I, ¢e6²²

I

I

I

¢I, I, ¢e9²²

¢I, I, ¢e10²²

H I J K

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

Table 16.3. The closure A* of the adjacency matrix A of the first example supply chain network A

B

C

I

I

I

D ¢I, I, ¢e1²²

E

A

F ¢I, {D}, ¢ e1, e3²²

G ¢I, {D, F}, ¢e1, e3, e5²²

H ¢I, {D, F}, ¢ e1, e3, e7²²

I ¢I, {D, F}, ¢ e1, e3, e8²²

B

I

I

I

I

¢C, I, ¢e2²²

¢{C}, {E, G}, ¢e2, e4, e6²²

¢{C}, {E} ¢e2, e4²²

I

¢B, I, ¢e2²²

¢{B}, {E, G}, ¢e2, e4, e6²²

¢{B}, {E}, ¢e2, e4²²

I

I

I

¢I, I, ¢e3²²

¢I, {F}, ¢ e3, e5²²

¢{C}, {E, G F} ¢e2, e4, e6, e7²² ¢{B}, {E, G F} ¢e2, e4, e6, e7²² ¢I, {F}, ¢e3, e7²²

¢{C}, {E, G F} ¢e2, e4, e6, e8²² ¢{B}, {E, G F} ¢e2, e4, e6, e8²² ¢I, {F}, ¢e3, e8²²

C

I

I

I

D

I

I

E

I

I

I

I

I

¢I, {G}, ¢e4, e6²²

¢I, I, ¢e4²²

I

I

I

I

I

¢I, I, ¢e5²²

¢I, {G, F}, ¢e4, e6, e7²² ¢I, I, ¢e7²²

¢I, {G, F}, ¢e4, e6, e8²² ¢I, I, ¢e8²²

F

I

I

J ¢I, {D, F, G }, ¢ e1, e3, e5, e9²² ¢{C}, {E, G}, ¢e2, e4, e9²²

K ¢I, {D, F, G }, ¢ e1, e3, e5, e10²² ¢{C}, {E, G}, ¢e2, e4, e10²²

¢{B}, {E, G}, ¢e2, e4, e9²²

¢{B}, {E, G}, ¢e2, e4, e10²²

¢I, {F, G}, ¢e3, e5, e9²² ¢I, {G}, ¢e4, e9²²

¢I, {F, G}, ¢e3, e5, e10²² ¢I, {G}, ¢e4, e10²²

¢I, {G}, ¢e5, e9²²

¢I, {G}, ¢e5, e10²²

384 Ming Dong and F. Frank Chen Table 16.3. (continued) G

I

I

I

I

I

¢I, I, ¢e6²²

I

¢I, {F}, ¢e6, e7²²

¢I, {F}, ¢e6, e8²²

¢I, I, ¢e9²²

¢I, I, ¢e10²²

H I J K

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

I I I I

Table 16.4. Total connectivity matrix TM1 of the first example supply chain network Node

A

B

C

D

E

F

G

H

I

J

K

Weight

A

0

0

0

0.97

0.00

0.93

0.87

0.89

0.88

0.81

0.82

0.16

B

0

0

0

0.00

0.93

0.80

0.90

0.76

0.76

0.84

0.86

0.16

C

0

0

0

0.00

0.93

0.80

0.90

0.76

0.76

0.84

0.86

0.16

D

0

0

0

0.00

0.00

0.96

0.89

0.91

0.90

0.83

0.85

0.14

E

0

0

0

0.00

0.00

0.86

0.97

0.82

0.81

0.90

0.92

0.14

F

0

0

0

0.00

0.00

0.00

0.93

0.95

0.94

0.87

0.88

0.12

G

0

0

0

0.00

0.00

0.89

0.00

0.85

0.84

0.93

0.95

0.12

H

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

I

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

J

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

K

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

16.5.2 Topological Robustness Index of the Overall Network Structure

The topological index (TI) was proposed by Hosoya (1971) to classify isomers in molecular chemistry. The TI can be used to show the differences in the topological structures of isomers. In this section, the TI is employed to evaluate supply chain network robustness from the overall structure. When two or more edges do not share the same node, such edges are nonadjacent. G(M, q) is defined as the number of subsets of edges consisting of q nonadjacent edges in metagraph M. G(M, q) is the “nonadjacent number of class q.” The topological index of metagraph M is defined as follows (Sakakibara et al. 2004): m

TI ( M )

¦ G(M , q) ,

(16.14)

q 0

where m = Nnode/2 if Nnode is even, m = (Nnode˰ 1)/2 if Nnode is odd, and Nnode is the number of nodes in the metagraph.

Quantitative Robustness Index Design for Supply Chain Networks

385

A “dispersed” supply chain network reduces the risk of functional isolation in the event of a disaster because it provides more possibilities of connection with neighboring districts than a “concentrated” network. Therefore, the TI can be used as a quantitative index of dispersiveness/concentration of the supply chain network. Different metagraphs have different numbers of nodes and edges. To compare the TIs of different metagraphs, we define the average TI for metagraph M as follows: m

TI avg ( M )

TI( M ) N node  N edge

¦ G (M , q) q 0

N node  N edge

,

(16.15)

where Nedge is the number of edges in the metagraph, and G(M, 0) and G(M, 1) are fixed to 1 and Nedge, respectively. Take the first supply chain network M1 as an example (see Figure 16.6). The calculation of TI when q equals 4 is given in Figure 16.7. The calculation of its average TI is given in the following:

TI avg ( S1 )

1  10  24  18  4 11  10

2.7143 .

(16.16)

Figure 16.7 Calculation of the TI of M1 when q equals 4

16.6 Integrated Robustness Index of Supply Chain Networks To evaluate the overall robustness of a supply chain network, we need to consider the functional robustness (i.e., performance reliability) of individual nodes and links and also the structural robustness (i.e., structural connectivity and topological index). Therefore, a new systemwide, integrated robustness index for supply chain networks should be developed. The computation of the integrated robustness index can be divided into three stages. In the first stage, the accessibility matrix, which

386 Ming Dong and F. Frank Chen

includes simultaneously the concept of link transit time weighted by the performance reliability of a node can be measured as follows: N node

¦

A( M )

N node

PRi 

i 1

¦ PR

j

/ TTij ,

(16.17)

j 1

where, A(M) = accessibility matrix of network M, TTij = transit time between node i and node j, the PRi = performance reliability of node i (e.g., service level). The most accessible node has the lowest summation of transit times. For each node i, its accessibility can be calculated as N node

AN i

¦a

ij

,

(16.18)

j 1

where, ANi = accessibility of node i and aij = elements of accessibility matrix A(M). At the second stage, the weights of nodes based on structural connectivity can be determined from the total connectivity matrix TM. The structural connectivity weight of node i can be derived as follows: ª N node N node º t ij / « t ij » , «¬ i 1 j 1 »¼ 1

N node

CN i

¦ j

¦¦

(16.19)

where CNi = connectivity weight of node i and tij = elements of total connectivity matrix TM. Finally, the integrated systemwide network robustness index, which considers both functional robustness (i.e., accessibility) of nodes and links and network structural robustness (i.e., connectivity and topological structure), can be measured as follows: N node

RI ( M )

TI avg ( M ) ˜

¦ CN

k

˜ AN k ,

(16.20)

k 1

where RI = robustness index of the network, CNk = connectivity weight of node k, the ANk = accessibility of node k. Once a network’s robustness index is determined, a link l’s robustness index is given by the decrease of network robustness due to removing link l: LRI l

RI  RI l ,

(16.21)

where, LRIl = robustness index of link l, RI = network robustness index, RI-l = network robustness index without link l.

Quantitative Robustness Index Design for Supply Chain Networks

387

16.7 Examples In the following, two example supply chain networks are used to illustrate the proposed robustness index for supply chain networks. The topological structures of two example supply chain network are given in Figures 16.6 and 16.8, respectively. It can be seen that the two example supply chain networks have the same number of nodes and links but different topological structures. The corresponding total connectivity matrix of the first supply chain network M1 and the structural connectivity weights of the nodes are given in Table 16.4. The given data on performance (service level and transit time) and failure rates of nodes and links for both example supply chain networks are provided in Table 16.5. Distribution center 1

e1

A

F

e8

Manufacturer 1

Supplier 1

B Supplier 2

e2

C

e9 E

e4

Manufacturer 2

Supplier 3

I Retailer 2

e5

e6

H Retailer 1

e3

D

e7

Retailer 3

G Distribution center 2

J

e10

K Retailer 4

Figure 16.8. The second example supply chain metagraph model

From the proposed algorithm, the total accessibility matrix TM2 of the second example supply chain network is given in Table 16.6. The calculation of TI of the second example supply chain network (M2) when q equals 4 is given in Figure 16.9. The calculation of its average TI is given in the following: TI avg ( S 2 )

1  10  27  20  4 11  10

2.9524 .

(16.22)

From (16.20), the integrated systemwide robustness index for the first example supply chain network is: RI(M1)= 2.7143 u 1.0126= 2.7486. Similarly, the integrated system-wide robustness index for the second example supply chain network is RI(M2)= 2.9524 u 1.0106 = 2.9837.

388 Ming Dong and F. Frank Chen

It can be seen that M2 has a higher robustness value than M1. This makes sense since it is obvious that links e5 and e6 in M2 play more important roles than links e5 and e6 in M1. Table 16.5. Given data on performance and failure rates of nodes and links ʳ ʳ

Q

r

Backorder

Service

level

level

failure rate (Pf)

Structural

Effective

Transit time

service level

ୄhours୅

1˰ Pf

A

22

15

0.318

0.889

0.03

0.97

0.86233

e1

B

50

15

0.14

0.951

0.07

0.93

0.88443

e2

C

26

17

0.126

0.948

0.04

0.96

0.91008

e3

D

30

80

0.992

0.839

0.03

0.97

0.81383

e4

E

44

82

0.483

0.916

0.07

0.93

0.85188

e5

F

120

4

0.009

0.993

0.11

0.89

0.88377

e6

G

230

5

0.002

0.998

0.05

0.95

0.9481

e7

H

25

19

0.007

0.995

0.06

0.94

0.9353

e8

I

24

87

0.339

0.929

0.07

0.93

0.86397

e9

J

15

6

0.012

0.987

0.05

0.95

0.93765

e10

K

22

15

0.318

0.889

0.03

0.97

0.86233

Table 16.6. The accessibility matrix TM2 of the second example supply chain Node

A

B

C

D

E

F

G

H

I

J

K

Weight

A

0

0

0

0.97

0.00

0.93

0.86

0.89

0.88

0.80

0.82

0.19

B

0

0

0

0.00

0.93

0.87

0.90

0.82

0.81

0.84

0.86

0.18

C

0

0

0

0.00

0.93

0.87

0.90

0.82

0.81

0.84

0.86

0.18

D

0

0

0

0.00

0.00

0.96

0.89

0.91

0.90

0.83

0.85

0.16

E

0

0

0

0.00

0.00

0.93

0.97

0.88

0.87

0.90

0.92

0.17

F

0

0

0

0.00

0.00

0.00

0.00

0.95

0.94

0.00

0.00

0.06

G

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.93

0.95

0.06

H

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

I

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

J

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

K

0

0

0

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

Quantitative Robustness Index Design for Supply Chain Networks

389

Figure 16.9. Calculation of the TI of M2 when q equals 4

16.8 Conclusions Aiming at disruptions in supply chain networks, this chapter attempts to develop some robustness measures for supply chains. These measures can be used to evaluate both supply chain network robustness and individual node and link robustness. In this study, the integrated systemwide network robustness index is proposed. This index considers network structural robustness (i.e., connectivity and topological index) and also functional robustness (i.e. accessibility) of individual nodes and links. In this three-stage approach, first, the total connectivity matrix is constructed, the accessibility matrix is built in the second stage. Finally, after integrating the information from the connectivity matrix, topological structure, and accessibility matrix, the systemwide network robustness index is developed. Once the network robustness index is determined, the robustness index for individual links can be obtained. This chapter provides a first attempt at the construction of quantitative robustness indices of supply chain networks. A further study could be the development of robust policies based on the proposed robustness index. A robust policy is one that works well most of the time. This differs from an optimal policy, which is the best policy for a specific set of conditions. An optimal policy may work extremely well for the set of conditions for which it was designed, but performs very poorly for many others. Under an unsafe environment, robust policies are desired. The proposed robustness indices could be incorporated into a larger decision support tool that handles a much bigger scope of supply chain management with unexpected events.

390 Ming Dong and F. Frank Chen

16.9 Guidelines for Practitioners This chapter provides a unified framework for quantifying the robustness of a supply chain network subject to unexpected events, whereas current literature just pay attention to qualitative aspects of supply chain network robustness. The robustness indices developed in this chapter are aim general supply chain network structures (serial, parallel, assembly, and arborescent distribution). Therefore, they can be applied to any types of supply chain networks and different industrial supply chains. To develop robustness indices, three stages are required. In the first stage, the network functional robustness index that considers node service level and links’ lead time is established. In the proposed approach, a (Q, r) inventory control policy is adopted at a node. However, the network functional robustness index can also be built using other inventory control policies such as (S, s) and base-stock inventory control policies. The basic concepts are the same for different inventory control policies. The calculation procedures for finding node service level and link lead time can be easily implemented by programming tools such as Excel in the Microsoft Office package. At the second stage, network structural robustness indices characterizing a network’s total connectivity and topological structure are developed. In this step, the calculation for closure A* of the adjacency matrix of supply chain metagraph models is laborious. In this regard, computer implementation for automatically finding closure A* of the adjacency matrix can provide a great help. To compute the topological robustness index of the overall network structure, the topological index is needed. Finding the topological index can be greatly simplified through computer implementation. In the third stage, a comprehensive integrated network robustness index is constructed from structural and functional robustness indices. This index can be easily calculated from (16.20). It can be seen that the applications of proposed robustness indices to real-world supply chains are relatively straightforward. Although the proposed approach just provides an initial attempt to quantifying supply chain network robustness indices, it can be used to identify the key factors that will affect the network robustness under unexpected events. As indicated in the conclusions, a further study could be the development of robust policies based on the proposed robustness index. In this regard, substantial time and effort are expected for collecting and validating data in developing a planning model for supply chain networks under an unsafe environment. The robustness indices we described here are planned to be incorporated into a larger planning tool that handles a much bigger scope of supply chain management with unexpected events.

16.10 Acknowledgements The work presented in this paper has been supported by grants from the National Natural Science Foundation of China (70571050).

Quantitative Robustness Index Design for Supply Chain Networks

391

16.11 References Agrawal N, Nahmias S, (1997) Rationalization of the supplier base in the presence of yield uncertainty. Production and Operations Management 6:291–308 Anupindi R, Akella R, (1993) Diversification under supply uncertainty. Management Science 39:944–963 Basu A, Blanning RW, (1994) Metagraphs: A tool for modeling decision support systems. Management Science 40:1579–1600 Basu A, Blanning RW, (2000) A formal approach to workflow analysis. Information Systems Research 11:17–36 Basu A, Blanning RW, Avraham S, (1997) Metagraphs in hierarchical modeling. Management Science 43:623–639 Bollapragada S, Morton TE, (1999) Myopic heuristics for the random yield problem. Operations Research 47:713–722 Burke GJ, Carrillo JE, Vakharia AJ, (2004) Sourcing decisions with stochastic supplier reliability and stochastic demand. Working Paper, Warrington College of Business Administration, University of Florida Dong M, Chen FF, (2005) Performance modeling and analysis of integrated logistic chains: An analytic framework. European Journal of Operational Research 162:83–98 Gaonkar R, Viswanadham N, (2003) Robust supply chain design: A strategic approach for exception-handling. Proceedings of the IEEE Robotics and Automation Conference 2:1762–1767 Gaonkar R, Viswanadham N, (2004) A conceptual and analytical framework for the management of risk in supply chains. Proceedings of the IEEE Robotics and Automation Conference 3:2699–2704 Hopp W, Spearman M, (2000) Factory Physics, McGraw-Hill/Irwin, New York, Second Edition Hosoya H, (1971) A newly proposed quantity characterizing the topological nature of structural isomers of saturated hydrocarbons. Bulletin of the Chemical Society of Japan 44:2332–2339 Pai R, Kallepalli V, Caudill R, Zhou M, (2003) Methods toward supply chain risk analysis. Proceedings of the IEEE Systems, Man and Cybernetics Conference 5:4560–4565 Rodrigue JP, (2003) Graph theory: measures and indices. http://people.hofstra.edu/geotrans/, Date accessed 16/4/06 Sakakibara H, Kajitani Y, Okada N, (2004) Road network robustness for avoiding functional isolation in disasters. Journal of Transportation Engineering 130:560–567 Sheffi Y, (2001) Supply chain management under the threat of international terrorism. The International Journal of Logistics Management 12:1–11 Souter G, (2000) Risks from supply chain also demand attention. Business Insurance 34:28– 28 Stauffer D, (2003) Supply-chain risk: Deal with it. Harvard Business School Yu D, Luh PB, (2004) Achieving reliable delivery in supply chains: the control of uncertainties. Proceedings of the IEEE Robotics and Automation Conference 3:2693– 2698

17 Impact of Reducing Uncertainty in European Supply Chains1 Paul Childerhouse and Denis R. Towill

Abstract:

In this chapter, we show that reducing supply chain uncertainty increases responsiveness and thereby benefits bottom line performance as assessed via total cycle time reduction. This positive effect we term the uncertainty reduction principle. As the key enabler, we use the uncertainty circle to focus on the sources to be eliminated. We also show that these sources of uncertainty can react and magnify in a flywheel effect caused by poor value stream management. A supply chain audit methodology is described for identifying and codifying uncertainty. Our proposition is that smooth material flow leads to and statistically correlates with uncertainty reduction. Examples are given of good real-world supply chain practices thereby identified. Transferability of the uncertainty reduction principle is assured by establishing readily assimilated “best practice” guidelines via the study of “exemplar” operating characteristics.

17.1 Introduction In this chapter, we will show that reducing supply chain uncertainty increases responsiveness and thereby benefits bottom line performance as assessed via total cycle time reduction. This effect we term the uncertainty reduction principle. To enable uncertainty reduction, we use the uncertainty circle to focus on the sources

1

This chapter is a condensed version of the paper “Reducing Uncertainty in European Supply Chains,” Childerhouse, P. and Towill, D.R. - Journal of Manufacturing Technology Management; (2004) Vol. 15, No. 7, pp. 585-598. It is reproduced with the kind permission of Emerald Group Publishing Limited at: http://www.emeraldinsight.com/jmtm.htm

394

Paul Childerhouse and Denis R. Towill

to be eliminated. We also show that these sources of uncertainty can react and magnify in a flywheel effect caused by poor supply chain management. A supply chain audit methodology is described for identifying and codifying uncertainty. The proposition advanced herein is that smooth material flow leads to and statistically correlates with uncertainty reduction. Examples are given of good real-world supply chain practice thus identified and subsequently improved. Transferability of the uncertainty reduction principle is assured by establishing readily assimilated “best practice” guidelines via the study of “exemplar” operating characteristics. Our view of the terminology is that a supply chain consists of a bundle of one or more value streams. Hence, in many instances, these two terms may be regarded as interchangeable. We also concentrate on the product delivery process (PDP) herein (Parnaby, 1995). In other words, the pipeline is concerned only with pre-specified products. It is a cornerstone of our research that the uncertainties experienced with the value stream can be tracked back to the four segments of the uncertainty circle. Furthermore, these sources may be codified using a four-point Likert scale selected to minimize regression toward the mean. We shall show via a statistical analysis that there is a high degree of correlation between these uncertainty sources and the presence (or otherwise) of unwanted symptoms observed by the supply chain audit team. Micklethwait and Woolridge (1996) have stated that for any proposed element of management theory to be valid, it must be transferable between companies, and between market sectors. Our statistical analysis of the 32 value streams establishes a cluster of “exemplar” supply chains. These exemplars further demonstrate considerable commonality between observed good practice the cluster. This is extremely unlikely to be coincidental, since it is equally possible to show the bad practice emerging from the poorly performing end of the sample (Childerhouse 2002). Finally, as an aid to transferability, these established “best practices” are put in a format readily recognized in supply chain management. The proven “well trodden path” by which the uncertainty reduction via best practice is actually enabled in real-world supply chains is also identified herein.

17.2 The Uncertainty Circle Concept Supply chains are complex systems (usually unnecessarily so). Hence a systems approach to detecting and codifying uncertainty has been adopted by Mason-Jones and Towill (1998). This follows directly from the seminal work by Parnaby (1979) in model manufacturing systems. It is an approach based on problem identification via the study of flows across business interfaces, including material, orders, cash, and capacity movements. So it has much in common with conventional inputoutput analysis. van de Vorst and Beulens (2002) provide a useful definition of supply chain uncertainty as follows “Supply chain uncertainty refers to decisionmaking situations in the supply chain in which the decision-maker does not know definitely what to decide as he is indistinct about the objectives; lacks information about (or understanding of) the supply chain or its environment; lacks information processing capabilities; is unable to accurately predict the impact of possible

Impact of Reducing Uncertainty in European Supply Chains

395

control actions on supply chain behavior; or, lacks effective control actions (noncontrollability).” The uncertainty circle method is based on the control of our own value-added process responding to customer demand and in turn our placing orders on to our suppliers. Uncertainty can be seriously introduced by our own process not yielding the required products on time due to process breakdowns. etc. Supplier interface uncertainty results from noncompliance with our orders. Demand interface uncertainty is compounded by lack of marketplace transparency, “rogue” orders, etc. Finally, and perhaps surprisingly, further uncertainty is induced by poor system controls based on the wrong decision rules and stale, noisy, or incomplete information.

Figure 17.1. Concept of the uncertainty circle

The uncertainty circle concept is illustrated in Figure 17.1 where each source (our process: supply: demand: control) is depicted as a segment of the circle. For a “traditional” supply chain, these uncertainty sources are arbitrarily set equal to each other. As BPR programs are successfully implemented, these segments will reduce in size. We have also shown elsewhere that value streams tend to progress and reduce uncertainty via a sequential “well-trodden path.” The associated BPR scenario is first to set one’s own house in order by improving the reliability of our value-added process. This experience and acquired expertise is then put to good use in improving supplier performance. At this point, a common error occurs. This is to continue focusing on improving the process and supply interface, however there are diminishing returns to this approach. Rather, we should then exploit our new capabilities by getting closer to our customers and in particular to

396

Paul Childerhouse and Denis R. Towill

respond to accelerating rates of marketplace change, as identified by Harvey et al. (2000). Note that control uncertainty is progressively reduced because improved information flow and smoother material flow simplifies the task of synchronization and coordination and hence facilitates better decision-making.

17.3 The Symptoms of Value Stream Uncertainty It might be wrongly thought that the advent of customized supply chains (Christopher and Towill 2000) means that the top producers can arbitrarily cope with uncertainty whatever the source may be. This is not the case and top-class value stream operations must be engineered into the delivery system. For example, to cope with a demand for customized personal computers, great care is taken to eliminate upstream uncertainty. This is achieved by a combination of good product design (including modularization) and good supply chain design and execution. Hence, only the “front end” of assembly-to-order has high (and unavoidable) uncertainty. Thus the VDU supplier may have to cope with day-today order variations of 10 to 1 which the vendor in turn partially manages by removing uncertainty from within his own supply chain upstream of VDU final assembly.

Figure 17.2. The “vicious” nature of the uncertainty circle and how it affects our delivery performance

The potentially “vicious” nature of the uncertainty circle must be emphasized. This is shown in flow diagram format in Figure 17.2 and will become apparent in a European automotive industry example to be discussed later. In the Dell computer

Impact of Reducing Uncertainty in European Supply Chains

397

supply chain, it may be argued that all of the uncertainties in Figure 17.2 are minimal except the demand uncertainty induced by marketplace volatility. In practice, this is enabled by the need for identifying and then implementing “leagile” delivery. But inspection of Figure 17.2 yields three potentially volatile feedback loops each capable of destabilizing operations. Real-world supply chain instability can then result via poor design, poor implementation, poor management, or a combination of these influences. Table 17.1. What the value stream analysts look for in real-world supply chains – the symptoms of complex material flow (based on Towill 1999) Class of symptoms

Dynamic behavior

Physical situation

Operational characteristics

Organisational characteristics

Features observed in complex material flow x x x x x x x x x x x x x x x x x x x x x x x x

“Systems induced” behavior observed in demand patterns. System behavior often unexpected and counterintuitive. Causal relationships often geographicly separated. Excessive demand amplification as orders pass upstream. Rogue orders induced from within by system “players.” Poor and variable customer service levels. Large and increasing number of products per pound turnover. High labor content. Multiple “Ad Hoc” production and distribution points. Large pools of inventory throughout the system. Complicated material flow patterns. Poor stores control. Shop floor decisions based on batch-and-queue concept “Interference” between competing value streams. Causal relationships often well separated in time. Failure to synchronize all orders and acquisitions. Failure to compress lead times. Variable performance in response to similar order patterns. Decision-making by functional groups. Excessive quality inspection. Multiple independent information systems. Overheads allocated across product groups, and not by activity. Excessive layers of management between CEO and shop floor. Bureaucratic and lengthy decision-making processes

In seeking to establish the degree of uncertainty in an individual value stream, we postulate that complex material flow is a primary lead indicator. As a consequence, it is then possible to produce checklists for supply chain analysts to use when monitoring the behavior of existing systems. Table 17.1 shows how complex material flow symptoms may be classified into four groups as identified via our industry based research: dynamic behavior, physical situation, operational characteristics, and organisational characteristics. The presence of uncertainty will result in high frequency of these complex material flow symptoms. As will be seen later, the analyst works through the individual checklists via a combination of data modeling, activity sampling, process mapping, and structured questionnaires.

Paul Childerhouse and Denis R. Towill

398

17.4 Correlation Analysis In much of what follows, we make considerable use of the results obtained from the study of 32 European value streams by Childerhouse (2002). For example, codification via Likert scales subsequently allows us to evaluate the statistical relationship between these four classes of uncertainty symptoms and the four uncertainty circle segments. These are shown in Table 17.2. The important correlations for this particular sample are thereby seen to be: x

Own process uncertainty

x x x

Supply side uncertainty Demand side uncertainty Controls side uncertainty

~ Physical situation (at 99% level) ~ Operational characteristics (at 98% level) ~ Operational characteristics (at 95% level) ~ Dynamic behavior (at 95% level) ~ Dynamic behavior (at 99% level) ~ Physical situation (at 98% level) ~ Operational characteristics (at 99% level)

Table 17.2 illustrates the statistical relationships between each of the four sources of uncertainty and the four categories of symptoms resulting from complex material flow. In each case, the correlation coefficient and resultant significance level have been highlighted the shaded quadrants illustrate those which exceed 95%. The prediction strengths of simple linear regression models are indicated by the respective R2 values. Further, a brief statement, where appropriate, of feasible causal relationships has been provided in Table 17.2. For example, it can be reasoned that high control uncertainty in the form of competition between value streams, batch and queue decisions, and a lack of synchronization of time buckets along the chain will result in operational symptoms of complex material flow. Note here how the importance of system controls has emerged from this statistical analysis. Also that the control situation is strongly influenced by the dynamic behavior symptoms, the physical situation symptoms, and the operational characteristics symptoms. Whereas the first correlation is intuitively obvious, the others are far less likely to be recognized by inexperienced managers. Our results do, however, support the desirability of adopting a business systems engineering (BSE) approach with emphasis on process mapping and improvement (Parnaby, 1995) to supply chain design that is both essential and potentially profitable. Examination of Tables 17.1 and 17.2 thus confirms the complex interactive nature of real-world value streams thus justifying the BSE treatment if valid solutions are to be found and implemented.

Impact of Reducing Uncertainty in European Supply Chains

399

Table 17.2. Highlighting the correlation between complex flow symptoms and the uncertainty circle quadrants assessed over the 32 value stream sample

Figure 17.3 shows the presence of the uncertainty symptoms as observed over the 32 value stream sample. Figures 17.3(a), (b), (c), (d), respectively, illustrate the four classes of dynamic behavior, physical situation, operational symptoms, and organizational symptoms which the analysts are instructed to seek out. In each case, these are broken down to the individual characteristic level previously defined in Table 17.1.

Functional decision-making Excessive quality inspection Multiple information systems Indirect costs not activity based Excessive layers of management Bureaucratic decision-making

x Organisational characteristics x x x x x

Large number of products High labor content Multiple production and distribution points Large pools of inventory Complicated material flow Poor stores control

x x x x x x

Batch-and-queue decisions Competing value streams. Time separated causal relationships Failure to synchronize Failure to compress lead times. Variable performance to same input

Systems induced Unexpected and counterintuitive Geographicly separated Demand amplification Rogue orders Poor service levels

x x x x x x

Symptoms observed in complex material flow

x Operational characteristics x x x x x

Physical situation

Dynamic behavior

Class of Symptoms

As present in sampled “baseline” value streams

Up 20%

Down 26%

Down 33%

Down 15%

As present in sampled “lean thinking” value streams

Down 33%

Down 87%

Down 54%

Down 69%

As present in sampled “exemplar” value streams

400 Paul Childerhouse and Denis R. Towill

Table 17.3. Comparison of complex material flow symptoms observed in clusters representing “baseline,” “lean thinking,” and “exemplar” value streams

Impact of Reducing Uncertainty in European Supply Chains

401

Demand amplification

Poor stores control

Poor service levels

Large no. of products

Unexpected or counter-intuitive

Complicated material f low

Symptoms

Symptoms

All of the symptoms are present to a very marked degree. Each bar chart is laid out in the standard Pareto format of increasing frequency of occurrence. Thus the four “top” (or most commonly observed) features are geographic separation between supply chain echelons, large pools of inventory, failure to synchronize, and product group costings based on averaging and not on value stream activity. Yet many of these symptoms were specifically recommended for action for elimination over two decades ago in “Five Rules to Avoid Bankruptcy” (Burbidge 1983).

Rogue orders System Induced

Multiple prod. dist. points High labour content

Geographically separated

Large inventory pools

0

25

50

75

100

0

25

Average % presence of symptom

(a) Dynamic Behaviour Symptoms

75

100

(b) Physical Situation Symptoms

Competing value streams

Excessive layers of management

Faliure to compress lead times

Excessive quality inspection

Time seperated causal relationships

Symptoms

Symptoms

50

Average % presence of symptom

Batch & Queue decisions Variable performace to same input

Multiple information systems Functional decisional making Bureaucratic decision making

Faliure to synchronise

Product group cost allocation

0

25

50

75

Average % presence of symptom

(c) Operational Symptoms

100

0

25

50

75

100

Average % presence of symptom

(d) Organisational Symptoms

Figure 17.3. Relative occurrence of complex material flow symptoms across the 32 value stream sample

To exploit the audit results further, we have correlated two separately derived indices for the value stream sample, as shown in Figure 17.4. The vertical axis is an index computed by averaging the observed complex material flow symptoms score for an individual value stream. In contrast, the horizontal axis refers to the Euclidean norm score for uncertainty also estimated for each value stream. This Euclidean norm estimates the distance of all individual value streams from the target score set at the fully integrated (and hence minimum uncertainty) value stream. The first point to note is the high degree of correlation between these two metrics (established as significant at the 99.9% level). This means that the analysts are consistent in their judgment on the uncertainty experienced by any specific value stream. The second important point is that we can examine clusters in various regions of the graphs to further examine the working practices of value streams which in operational terms may be regarded as baseline, lean, and

Paul Childerhouse and Denis R. Towill

402

exemplar. Hence we propose to use this statistical analysis to identify and highlight known “best practice.”

Average % Presence of Complex Material Flow Symptoms

100

Correlation coefficient = 0.715 Significant at the 99.9% level

75

B L

50

25

E

t fit Bes

line

Key E=Exemplar L=Lean B=Baseline

0 0 Seamless Supply chain

1

2

3

4

Euclidean Norm Uncertainty

5

6 Traditional Supply chain

Figure 17.4. Correlation between complex material flow symptoms and value stream uncertainty scores established for 32 value stream sample

17.6 Discussion of Audit Results To provide insight into the collective health of our value stream sample, their uncertainty scores may be put in the pie chart format shown in Figure 17.5. Here the 32 value streams are categorized as follows: x x x x

Minimal control of our process (~ 10%) Still struggling with “lean thinking” concepts (~ 45%) Evidence of some good practice (~ 35%) Exemplars (~ 10%)

The break points have been selected with reference to the well-established Stevens (1989) supply chain integration model (Childerhouse, 2002). In relating Figure 17.4 to Figure 17.5 we suggest that “baseline” equals “minimal control of our process.” Also the “lean” value stream cluster in the former is roughly at the intersection of the “struggling with lean thinking” and “evidence of some good practice” status in the latter grouping. But can we now identify more precisely what these differences in uncertainty scores are caused by? To help understand this scenario, we relate the clusters identified within Figure 17.4 to the list of uncertainty symptoms defined in Table 17.1. A three-value

Impact of Reducing Uncertainty in European Supply Chains

403

stream cluster is thus highlighted in Figure 17.4 for each category arbitrarily considered to be of exemplar, lean, and baseline performance.

Figure 17.5. Pie chart summarizing uncertainty scores ranking of 32 value stream sample Large product variety

Design specification alterations

Scheduled build alterations

Poor stock control

Large product range

Inaccurate B.O.M

Schedule alterations placed on suppliers

Can not build what want to

Lost stock at point of use

High risk of obsolescence

Limited finished goods safety stock

Poor schedule adherence

Lack of components when required

Large set-up times and costs

Poor internal stock visibility

Lack of raw materials when required

Production capacity constraint

Lack of scheduling flexibility

Large batch production

Key

Customer Supplier

Figure 17.6. How interaction between customer and supplier increases uncertainty via the generation of the houlihan “flywheel effect”

The average scores for each cluster have then been calculated for each uncertainty symptom. So these scores can take only the values of 0, 1, 2, or 3, the latter means that the symptom was present in all three value streams making up that particular cluster. The results of this analysis are shown in Table 17.3. Note how the exemplars are much better than even the lean thinking value streams. Very

404

Paul Childerhouse and Denis R. Towill

importantly, 15 out of 24 targeted uncertainty symptoms have been completely eliminated by the exemplars. They epitomize the merit of matching the supply chain to the marketplace (Mason-Jones et al 2000). We saw in Figure 17.2 how the uncertainty circle could become “vicious,” i.e. lead to chaotic behavior. Our QSAM audit clearly identified one particular value stream where this problem actually occurred in practice. This is shown in schematic form in Figure 17.6 where a potentially unstable feedback loop is identified between OEM and vendor. Specifically excessive (and arguably avoidable) schedule alteration (as typified by Harrison 1997 and Hiebar et al 1998) placed on suppliers leads to poor schedule adherence causing lack of components when required. This in turn caused the OEM to shortfall on output leading to even more schedule disruption and so on, i.e., the Forrester flywheel effect. This is a certain recipe for creating an enduring blame culture between all “players” in the chain.

17.7 Impact of Reducing Uncertainty on Supply Chain Responsiveness It is reasonable to question whether reducing value stream uncertainty actually has a significant beneficial impact on supply chain performance. This we have assessed via the various lead times observed in particular value streams. It is already well established that cycle time compression leads to substantial improvements in the “bottom line” (Thomas, 1990). A further big advantage of using cycle time as a performance metric is that it is simple and unambiguous. It may be argued that part of the success of the “machine that changed the world” (Womack et al 1990) is due to their using such transparent and transferable indicators. Of course the exact impact of total cycle time reduction on bottom-line performance is varied. But as Thomas (1990) has shown, even conservative estimates show significant improvement. We have been exceptionally fortunate during this research program to be able to assess the performance of value streams both before and after extensive BPR programs. Both of these real-world reengineering exercises were based upon the well-established principles of smooth material flow (Burbidge 1962; Forrester 1961). Furthermore, Childerhouse and Towill (2003) have shown how the material and information flows were drastically improved during these BPR programs by increasing adherence to 12 simplicity rules. There is comprehensive published material available relating to both BPR programmes (Lewis 1998; McCullen and Towill 2002). In addition to evaluating the four sources of uncertainty, the extent to which the complex flow symptoms were reduced. To provide further insight into the improvement in value stream performance, distribution, manufacturing and supplier lead times are presented in addition to total cycle time in Table 17.4. These can give vital clues to any nonvalue-added activities remaining to be eliminated.

Impact of Reducing Uncertainty in European Supply Chains

405

Table 17.4. Examples of the impact of reducing uncertainty on supply chain responsiveness

Supply chain criteria

QS audit scores

Complex material flow symptoms

Observed improvement in cycle times

Performance attribute

Change in material flow integration level Uncertainty score Overall % of symptoms present Dynamic Physical Operational Organizational Distribution lead time Manufacturing lead time Supplier lead time Total cycle time

Example supply chains A. B. Mechanical Construction precision sector products products Functional to Internal to external external

Down from 4.80 to 1.73 45% decrease down to 13% 67% decrease 20% decrease 80% decrease 16% decrease Cut by 50%

Down from 3.00 to 2.24 58% decrease down to 21% 17% decrease 66% decrease 66% decrease 83% decrease Cut by 84%

Cut by 83%

Cut by 50%

Cut by 75% Cut by 78%

Cut by 81% Cut by 81%

Table 17.4 shows the results for the two supply chains for which “before” and “after” performance have been determined. In Case A (mechanical precision products) the BPR programs cover the large value stream transition from functional to external integration (as defined by the Stevens 1989 change model). The uncertainty score has decreased from 4.80 to 1.73 and all the complex material flow symptoms are reduced. The dynamic and operational contributions are shrunk very substantially. As expected, the lead times are but a fraction of the initial values. There is a consequential reduction in bullwhip of 50% and minimum stock turn improvements of 2 to 1 and increasing further with operating experience of the new system. Case B is a value stream supplying construction sector products. In this instance, the recorded BPR starting point was more advanced, since on the Stevens (1989) scale, the change in integration level is only from internal to external. Hence the uncertainty score was only 3.0 to start with, and post BPR is reduced to 2.24. Nevertheless, the consequential reductions in all lead times are still substantial. Furthermore, this value stream is now so responsive that it offers a unique guarantee. If the product is not delivered on site within 24 hours of schedule, then the product is delivered free. There is also a reported increase of 10% in profit margin over the BPR period (Childerhouse 2002).

406

Paul Childerhouse and Denis R. Towill

17.8 Transferability of Best Practice Micklethwait and Woolridge (1996) have argued that a criterion against which a proposed contribution to management theory may be judged is transferability between companies and between market sectors. Our view is that the clustering of value streams based on codified scores plus the improved responsiveness resulting from uncertainty reduction goes some way to establishing the “uncertainty reduction principle” within management theory. However, a detailed examination of the material analyzed during our value stream analysis does enable us to describe “best practice” in a meaningful and transferable way. For example, the cluster of three “exemplars” shown in Figure 17.4 yields many common factors observed during our research. As listed in Table 17.5, these are grouped under the convenient operations management headings of product variant mix, demand variability, information systems, inventory policy, supply chain relationships, and simplified material flow characteristics. Table 17.5. Transferable “best practices” shared by the three “exemplar” value streams

Area of good practice

Simplified material flow

Supply chain relationships

Inventory policy Information system Demand volume stability Variant mix

Exemplar value stream characteristics

x

x x

Visible continuous improvement toward the seamless supply chain objective Customer oriented operations Single piece material flow in place Pull systems based on kanbans Use of the shortest possible planning period Active process time compression Generic partnerships with key suppliers Dominant “player,” i.e. “product champion” who manages the synchronization and coordination of the supply chain Single point of control Direct line feed and consignment stocking

x

Totally integrated information systems

x

Average schedule variability of the three exemplars is 4% over a 1-month forecast, compared to 45% for the entire 32 value stream sample Average number of product variants is significantly lower for the three exemplars in comparison with the remainder of the sample

x x x x x x x

x

Impact of Reducing Uncertainty in European Supply Chains

407

An important feature of Table 17.5 is that it does interpret the uncertainty reduction principle in focused terms readily understood by the practitioner. In our view, it does provides a suitable checklist for management auditing purposes. It is also wide-ranging in raising awareness of important SCM issues. At the cultural level, it is concerned with appropriate supply chain relationships. The necessary enabling IT is covered and interacts with information transparency and inventory policy in the control of materials, production, and delivery. Reducing variant mix involves enhanced product design capability and possibly introducing the concept of postponement (Van Hoek 1998; Pagh and Cooper 1998). Finally, the simplified material flow guidelines directly relate to shop floor processes and their operating environment.

17.9 Conclusions and Guidelines for Practitioners We have seen that an analysis of 32 European value streams supports the concept of the uncertainty reduction principle as a mechanism for identifying performance improvement. This was done via an industrial audit which provided codified scores representing uncertainties associated with our own value-added process, the supply side, the demand side, and the control system. It was then found possible via statistical analysis to correlate these uncertainty scores with separate measures indicating the level of complex material flow observed in any given value stream. This gives confidence that there is consistency between these two sources of information used in supply chain analysis. The statistical correlation is highly significant at the 99.9% level. Furthermore, these results enabled the value streams to be ranked according to their uncertainty level, hence indicating those value streams which could be clustered and thereby regarded as baseline; lean; and exemplars. In this research a great deal of emphasis has been placed on organizational, operational, physical scenario, and dynamic behavior characteristics of the value streams. These provide the raw data used in the analysis. A crosscheck shows there is consistency between these observations and the subsequent classification of that particular value stream. In other words, this information is of potential value in the definition of levels of performance for use as a holistic metric which assesses total supply chain operations. Although it is of academic interest to demonstrate the usefulness of uncertainty reduction principle in supply chain design and operation, this concept is difficult to transfer to supply chain management without a supporting infrastructure. This is provided herein by the identification during the research of a set of operational features common across the exemplar value streams. Hence these transferable “best practice” features represent realistic guidelines on how to improve supply chain performance. Figure 17.7 summarizes how our 32 value streams may be shown in various states of transition between baseline and exemplar. It also gives a thumbnail description of the sequencing in “the well-trodden path” followed by value streams moving along this trajectory and hence improving their capability.

408

Paul Childerhouse and Denis R. Towill

Figure 17.7. Moving toward best practice : conclusions drawn from the 32 value stream sample

Finally, we have been fortunate to be able to assess performance in two complex real-world supply chains before and after substantial BPR programs have been executed. In both cases, there were substantial reductions in the uncertainty scores and in the presence (or otherwise) of observed characteristics associated with complex material flow. These BPR programs resulted in major compression of supplier lead time, manufacturing lead time, and distribution lead time cycles. Thus these supply chains are now much more responsive than hitherto. Published data indicate a corresponding improvement in bottom-line performance and a highly significant bullwhip reduction of 50% across a sample range of products. Further case examples relating to other industrial sectors are required to fully validate the generic nature of our proposed approach to reducing supply chain uncertainties. Other research currently underway involves categorization of the 12 simplicity rules and expansion of the sample to include non-European supply chains.

17.10 Acknowledgements We wish to acknowledge the support of EPSRC via the Supply Chain 2001 project and the industrial collaborators who enabled us to assess the performance of the 32 value streams. Thanks are also due to our colleagues in the Cardiff Logistics Systems Dynamics Group who participated in Quick Scan Audits as part of Supply Chain 2001.

Impact of Reducing Uncertainty in European Supply Chains

409

17.11 References Burbidge JL, (1983) Five golden rules to avoid bankruptcy. Production Engineer 62(10):13–14. Burbidge JL, (1962) The Principles of Production Control. Macdonald and Evans, London. Childerhouse P, (2002) Enabling seamless market-orientated supply chains. PhD Thesis, LSDG, Cardiff University. Childerhouse P, Towill DR, (2003) Simplified material flow holds the key to supply chain integration. OMEGA 31:17–27. Christopher M, Towill DR,(2000) Supply Chain migration from lean and functional to agile and customised. International Journal of Supply Chain Management 5(4): 206–213. Forrester JW, (1961) Industrial Dynamics. MIT Press, Cambridge MA. Harrison A, (1997) Investigating the sources and causes of schedule instability. International Journal of Logistics Management 8(2):75–82. Harvey M, Griffith D, Novicevic M, (2000) Development of “timescapes’ to effectively manage global inter-organisational relation communications. European Management Journal 18(6):646–662. Hieber R., BrĦtsch D, Frigo-Mosca F, (1998) How to manage your supply network to get better results. Proceedings of International Conference of the Manufacturing Value Chain, Ed. Bititci, U.S. and Carrie, A.S., Troon, 289–295. Lewis JP, (1998) Mastering Project Management – Applying Advanced Concepts of Systems Thinking, Control and Evaluation, and Resource Allocation. McGraw-Hill, New York. Mason-Jones R, Towill DR, (1998) Shrinking the supply chain uncertainty circle. Control, The Institute of Operations Management 24(7):17–22. Micklethwait J, Woolridge A, (1996) The Witch Doctors – What the Management Gurus are Saying, Why it Matters, and How to Make Sense of it. Mandarin Books, London. McCullen P, Towill DR, (2002) Diagnosis and reduction of bullwhip in supply chains. Supply Chain Management: An International Journal 7(3):164–180. Pagh JD, Cooper ML, (1998) Supply chain postponement and speculation strategy: How to choose the right strategy. Journal of Business Logistics 19(2):13–33. Parnaby J, (1979) Concept of a Manufacturing System. International Journal of Production Research 17(2):123–135. Parnaby J, (1995) Systems engineering for better engineering. IEE Engineering Management Journal 5(6):256–266. Stevens GC, (1989) Integrating the supply chain. International Journal of Physical Distribution and Materials Management 19(8):3–8. Thomas PR, (1990) Competitiveness Through Total Cycle Time. McGraw-Hill. New York. Towill DR, (1999) Simplicity Wins : Twelve Rules for Designing Effective Supply Chains. 'Control' - the Journal of the Institute of Operations Management 25(2):9–13. van der Vorst JGAJ, Beulens AJM, (2002) Identifying sources of uncertainty to generate supply chain redesign strategies. International Journal of Physical Distribution and Logistics Management 32(6): 409–430. van Hoek R, (1998) Reconfiguring the supply chain to implement postponed manufacturing. International Journal of Logistics Management 9(1). Womack JP, Jones DT, Roos D, (1990) The Machine that Changed the World. Mandarin Books, London.

18 Analyzing the Effectiveness of the Availability Management Process Young M. Lee

Abstract:

Availability management is a major factor in successful supply chain management since it influences key supply chain performance metrics such as customer service level and inventory. The availability management process involves generating availability outlook, scheduling customer orders against the availability outlook, and fulfilling the orders. The process is also associated with many uncertainties such as customer demand, customer preference of product configuration, and changes in supply constraints and various supply chain policies, which also affect the supply chain performance. As e-commerce is becoming a major part of business transactions it is much easier for customers to compare availability and services from many different sellers. Therefore, it is important for sellers to process customer orders in real-time, promise ship dates, fulfill the orders as promised, and to have availability of resources to be able to promise customers desirable ship dates. In today’s competitive and dynamic business environment, companies need to continually evaluate the effectiveness of the availability management process and supporting IT system and look for ways to transform the process to achieve better customer service and profitability. To do that, there is a need for an easy-to-use modeling tool that can accurately assess the effectiveness of the existing availability management process, evaluate the impact of potential changes in the process and identify opportunities for improvement. In this chapter, we describe an availability management simulation tool that was developed at IBM to support the continuous effort to improve the availability management process. The simulation model has become a critical tool in making strategic business decisions that impact customer service and profitability at IBM.

412 Young M. Lee

18.1 Introduction Promising customers a desirable delivery date and fulfilling the orders as promised are an important aspect of customer service. The recent surge and widespread use of e-commerce means that shoppers can now more than ever easily assess and compare customer service quality in addition to quality of goods and price among different vendors. This creates a very competitive business environment, thus making customer service a critical factor for the success and survival of many companies. Competitive pressures are forcing companies to constantly look for ways to improve customer service by evaluating and redesigning supply chain processes. The availability management process (AMP) is a key process that impacts customer service since it determines promised ship (or delivery) dates, the accuracy of the promised ship date, order scheduling delay, and order fulfillment rate as well as inventory. This work was motivated by supply chain processes of IBM’s computer hardware businesses. At IBM, businesses are being managed as on-demand businesses, where business strategies, policies, and processes are continually evaluated and changed to meet increasingly demanding needs of customers. These changes are called “business transformations” at IBM. Various business transformation ideas are generated, evaluated, and deployed to improve the effectiveness of the businesses especially in the area of the supply chain. The availability management process (AMP) is one such area where transformation ideas are constantly evaluated and implemented. When a change in AMP is sought, for example, a change in the available-to-promise (ATP) generation method, an order scheduling policy, or an order fulfillment policy, the impact of such a change has to be accurately assessed before it is implemented because changes are typically expensive and time-consuming to implement in large enterprises such as IBM. Other examples of changes in AMP can be moving from a make-to-order (MTO) to a configured-to-order (CTO) business, a change in demand classes, refresh rate of the availability database system, and supplier flexibility, etc. In addition, AMP is tightly associated with other exogenous changes in a supply chain such as customer demand, customer preference of product configuration, and changes in supply constraints. These exogenous changes in a supply chain often force an enterprise to transform its availability management process. Changes in the AMP typically require careful analyses to support the decision whether or when to implement the changes. For example, it is necessary to determine how a change in order promising policy would affect the customer service improvement and inventory positions, and how expensive the supporting IT systems would be before the change is deployed in the business. Therefore, there is clearly a need for a tool that is readily available to simulate the affected supply chain and quantify the performance metrics fast and accurately before making a costly investment. In this chapter, we describe a supply chain simulation tool called AMST (availability management simulation tool) developed at IBM by the author for the purpose. The AMST is a simulation framework that allows rapid customization of the tool for simulating various availability management situations. We also describe several case studies where AMST has been used successfully in evaluating availability management transformation opportunities at IBM.

Analyzing the Effectiveness of the Availability Management Process

413

Availability management involves generating an availability outlook, scheduling customer orders against the availability outlook, and fulfilling the orders. Generation of availability outlook is a push side of the availability management process. It allocates availability into ATP (available-to-promise) quantities based on various product and demand characteristics and planning time periods. Order scheduling is the pull side of the availability management process. It matches the customer orders against the availability outlook, determines when the customer order can be shipped, and communicates the promised ship date to the customer. Order fulfillment is executing the shipment of the order on promised ship date. Even if an order is scheduled for shipment on a certain date based on the availability outlook, the resources that are required to ship the product on the promised ship date may not actually be available when the ship date comes. A key role for an effective availability management process is to coordinate and balance the push side and pull side of ATP. Ball at al. (2004) gave an overview of the push side (availability Planning) and pull side (availability promising) of ATP with examples from Toshiba, Dell, and Maxtor Corporation. They stressed the importance of coordinating the push and pull sides of availability management for supply chain performance by making good use of available resources. Although ATP functions have been available in several commercial ERP and supply chain software such as SAP’s APO, i2’s Rhythm, Oracle’s ATP Server, and Manugistics’ SCPO modules, etc. for several years (see Ball et al. 2004 for details), those ATP tools are mostly fast search engines for an availability database, and they schedule customer orders without any sophisticated quantitative methods. Research on the quantitative side of ATP is still at an early stage, and there are only a limited number of analytic models developed in supporting ATP. For the push side of ATP, Ervolina and Dietrich (2000) developed an optimization model as the resource allocation tool and described how the model is used for a complex configured-to-order (CTO) environment of the IBM Server business. They also stress how the push side (availability promising) and pull side (availability planning) have to be work together for overall availability management performance. For the pull side of ATP, Chen et al. (2002) developed a mixed-integer programming (MIP) optimization model for a process where order promising and fulfillment are handled in a predefined batching interval. Their model determines the committed order quantity for customer orders that arrive with requested delivery dates by simultaneously considering material availability, production capacity as well as material compatibility constraints. They also studied how the batching interval affects supply chain performance with different degrees of resource availability. Moses et al. (2004) also developed a model that computes an optimal promised ship date considering availability and also other order-specific characteristics and existing commitments to previous scheduled orders. Pan et al. (2004) also developed a heuristics-based order promising model but with ecommerce environment in mind. They modeled a process where customer orders arrive via the Internet and earliest possible shipment dates are computed in realtime and are promised to customers.

414 Young M. Lee

All the previous work described above deals with either the push side of ATP or the pull side of ATP, but not together. There has not been any quantitative tool that looks at both the push and pull sides simultaneously as well as other dynamic factors in a supply chain, and evaluates the effectiveness of the overall availability management process. Some of the work described above uses simulation experiments to measure the effectiveness of their solutions, but their simulation work was capable of simulating only a very specific supply chain environment, focusing on only one aspect of the ATP process. In this chapter, we describe an availability management simulation tool that evaluates how various components (generating availability outlook, scheduling customer orders, and fulfilling the order) impact supply chain performance either by themselves or in coordination with others in various supply chain environments. The components may be simple business logics with arithmetic calculations that can be easily modeled inside the simulation model, but they also can involve complex ATP optimization tools, in both the pull and push sides of ATP, that interface with the simulation model. Discrete-event simulation has been around for many years in simulating supply chain management (SCM) processes to evaluate its effectiveness. McClellan (1992) used simulation to study the effect of the MPS method, variability of demand/supplier response on customer service, order cycle, and inventory. Hieta (1998) analyzed the effect of alternative product structures, alternative inventory, and production control methods on inventory and customer service performance. Bagchi et al. (1998) evaluated the design and operation of SCM using simulation and optimization, and analyzed SCM issues such as site location, replenishment policies, manufacturing policies, transportation policies, stocking levels, lead time, and customer service. Yee (2002) analyzed the impact of automobile model and option mix on primary supply chain performance such as customer wait time, condition mismatch, and part usage. However, there hasn’t been any simulation modeling work that analyzes both the pull and push sides of the availability management process as well as the coordination between them, in stand-alone mode or in an interface with ATP optimization models. Development of simulation models for the supply chain such as the availability management process can be time-consuming. The availability management simulation tool (AMST) is a simulation modeling framework that can be easily adapted to simulate various availability management situations. The AMST has been used at IBM for several years for developing availability simulation models rapidly by pulling together modeling components described in subsequent sections and modifying only necessary components of the tool. The AMST also contains a data interface utility, which uploads data necessary for a simulation model, and a tool interface utility, which can call ATP optimization models. The availability management simulation tool (AMST) is essential in supporting the continuous effort to improve the availability management process. The tool is playing a critical role in evaluating various supply chain transformation initiatives and making strategic business decisions that impact customer service and profitability at IBM. This chapter also describes several realistic business transformation case studies which used the model at IBM. The rest of chapter is organized as follows. In Section 18.2, we describe the availability management process based on our experience in IBM’s hardware

Analyzing the Effectiveness of the Availability Management Process

415

businesses. In Section 18.3, we describe how the simulation model works and what it is capable of. In Section 18.4, we describe several supply chain transformation case studies that we conducted with AMST, its impacts, and results. Section 18.5 provides conclusion and remarks.

18.2 Availability Management Process In IBM hardware businesses, availability management consists of three main tasks: (1) generating an availability outlook, (2) scheduling customer orders against the availability outlook, and (3) fulfilling orders. There are two types of IBM’s hardware supply chain environment that we studied in this work. The first one is a HVEC (high valued easy configured) business, which manufactures commoditylike hardware products such as personal computers. The second type is a CCHW (complex configured hardware) business, which manufactures more expensive, server-type computers. For the HVEC business, customer orders typically arrive without any advance notice, requesting early fulfillment of the orders, usually in a few days. For this type of business, products consist of a few components and can be assembled rather quickly to fulfill customer orders. For the CCHW business, on the other hand, customers place orders in advance of their actual needs, often a few months in advance. Typically, CCHW customers place orders as early as 3 months before the requested delivery (due) dates, and early delivery and payment are not allowed. For this environment, products usually consist of a hierarchy of complex components, and require longer supply planning. Many buyers in this environment purchase products based on careful financial planning, and they typically know when they want to receive the products and make payment. Customer orders in this environment are typically highly skewed toward the end of a quarter, e.g., only a small portion of orders are placed in the first week of a quarter, and the orders gradually increase. Finally, as much as 60–70% of orders are placed in the last 2 weeks of a quarter. Generation of an availability outlook, is the push side of the availability management process. It preallocates ATP quantities, and prepares a searchable availability database for promising future customer orders. For the HVEC business, the availability outlook is typically generated by daily buckets, and the availability planning horizon goes out to a few weeks into the future. For the CCHW business, the availability outlook is allocated by weekly buckets, and the availability is planned on a much longer horizon, often a quarter (3 months) into the future. The ATP quantity is called the availability outlook for this reason. The availability outlook is typically generated based on product type, demand classes, supply classes, and outlook time buckets. The product type can be the finished goods (FG) level for a make-to-stock (MTS) business or a components (Comp) level for make-to-order (MTO) or configured-to-order (CTO) businesses. Demand classes can be geographic sales locations, sales channels, customer priority, sensitivity to delivery dates, profitability, and demand quantity. Supply classes can be degree of constraints and value of products. Outlook time buckets can be daily buckets or weekly buckets. Availability is preallocated into an availability outlook bucket

416 Young M. Lee

based on the dimension described above and rolled forward daily or weekly. The availability outlook is determined based on the availability of components, finished goods, WIP (work-in-process), MPS (master production schedule), supplier commitment, and production capacity/flexibility. When customer orders arrive, the availability outlook is searched in various ways according to scheduling polices to determine the ship (delivery) date that can be promised to customers. Customer order scheduling is a pull side of availability management. It reacts to customer orders and determines ship dates for the orders. For an HVEC business, customers usually request that the order be shipped as early as possible, and they would also like to know when the order will be shipped. CCHW customers usually request the orders be shipped (or delivered) on specified future dates. And they would like to know whether the requested due date can be met or how long the delay will be if the due (requested) date can’t be met. Customer orders arrive with various information such as product types, demand classes, customer classes, and due dates. The order scheduler then searches through the availability outlook database, and identifies the availability that meets the characteristics. The scheduling can also be done by an ATP engine that uses a certain algorithm to optimize the scheduling, considering various resources, policies, and constraints. The scheduler then reserves specific availability against each order and decrements the availability according to the purchase quantity of the order. The ship date of the order is determined from the time bucket where the availability is reserved, and it is promised to customers. Depending on the business environment, various rules and policies are applied in this order scheduling process. Examples are first-come-first-served policy, customer priority-based scheduling, and revenue (or profit)-based scheduling, etc. In a constraint environment, a certain ceiling can also be imposed to make sure the products are strategically distributed to various demand classes. Order fulfillment is executing the shipment of the product on the promised ship date. Even if an order is scheduled with a specific promised ship date based on the availability outlook, the availability (ATP quantity) may not actually exist when the ship date comes. There are several reasons why orders cannot be fulfilled on the promised date. One reason is the quality of availability outlook generation. In a CTO environment, the availability outlook is often generated based on finished goods availability, which is estimated based on supplier commitments on components and forecasted configuration of the finished goods. Since the component availability changes often and there is certain error in a configuration forecast, the components that are required to assemble a certain finished good may not be available when it is time ship the product to the customer. Another source for the fulfillment problem is IT system that supports the availability management process. Order scheduling is done based on the availability outlook data in an IT system, which is typically refreshed periodically since it is very expensive to update the database in real-time. The availability information kept in the IT system (system availability) is not always synchronized with the actual availability (physical availability). Due to the potentially inaccurate view of availability, an unrealistic ship date can be promised to a customer. Therefore, for certain customer orders, the necessary ATP quantity may not be there when the promised ship date arrives, thus creating dissatisfied customers. The impact of IT on

Analyzing the Effectiveness of the Availability Management Process

417

fulfillment is discussed in detail in Section 18.4.3. Therefore, a key role for an effective availability management process is coordinating and balancing the push side and pull side of ATP as well as IT resources.

18.3 Simulation Modeling of Availability Management In this section, we describe the AMST (availability management simulation tool). The model simultaneously simulates the three components of the availability management process, generating availability outlook, scheduling customer orders, and fulfilling the orders, as well as the effect of other dynamics such as customer shopping traffic, uncertainty of order size, customer preferences for product features, demand forecast, inventory policies, sourcing policies, supply planning policies, manufacturing lead time, etc. The simulation model provides important statistical information on the promised ship date, accuracy of the ship date determination, scheduling delay, and fulfillment rate as well as inventory. 18.3.1 Modeling of Availability Outlook

The availability outlook (also called availability quantity) is modeled by a multidimensional data array which represents various attributes of availability such as product type, demand class, supply class, and planning period. The product type can be either finished goods or components depending on whether the business is MTO or CTO. For a simple example, for a process where there are two attributes of availability (product type and time period), the availability outlook is represented by a two-dimensional data array shown as cylinders in Figure 18.1. The availability outlook is time-dependent, e.g., there is availability for the current period (t = 1), and there is an availability quantity for future periods (t = 2, 3, …) as more availability quantity is expected to exist through production or procurement on future dates. The availability time periods can be daily buckets or weekly buckets depending on the business environment. For example, in Figure 18.1, three of component 1 are available on the current day, five more are expected to be available a day after, and 10 more are expected be available for day 3, and so on. The availability outlook can be determined by simple business logic involving demand forecast and supply contracts, etc., but it can also be computed by a push side ATP optimization tool. The availability outlook is used in computing the ship date of customer requests and orders. The availability quantity changes as a result of many events in the business. In this work, we models four main events that affect the availability, and they are explained below. 18.3.2 Simulation of Order Promising

Figure 18.1 shows an example of how the ship date calculation is simulated in this work. Customer orders or ATP requests arrive in certain stochastic intervals, usually modeled as a Poisson process. Each order has one or more line items, and each line item has one or more quantities. The order quantities are modeled with probability distribution functions which are derived based on historical data. The

418 Young M. Lee

line items and quantities are determined as the order is generated in the order generation event (details described in the next section). For each line item, certain components are selected as the building blocks of the product using a distribution function representing the customer preference for component features. For example, in Figure 18.1, line item #3 of order # 231 requires components 1, 3, and 4, one unit each. Different order scheduling policies can also be applied here in selecting specific components.

Order 231 Item1: qty:10 Item2: qty:10 Item3: qty:10

preference distribution of component features

sourcing policies

t=1

t=2

t=3

t=4 ……

t=N-1 t=N

3

5

10

10

....

comp1:

0

2

3

10

....

comp2:

10

5

5

0

....

comp3:

8

5

5

5

....

comp4:

....

comp M:

Order 232 Item1: qty:10 Item2: qty:10 Item3: qty:8

distribution of order arrival

components availability (3 days) + mfg lead time (2 days) = item 3 ship date (5 days) ….. item 2 ship date (3 days) item 1 ship date (10 days) ---------------------------------------------------total order ship date = 10 days

mfg lead time distribution

Figure 18.1. Simulation of order scheduling and ship date calculation for as early as possible orders

For the orders that are requested to be fulfilled as early as possible, as described in the earlier section as the HVEC business at IBM, the simulation model looks for a specified quantity of a chosen component starting from the first time period to latter time periods until the availability of the total quantity is identified. In this example, the time periods (buckets) are in days. Component #1, the requested quantity of 10 is identified in the first 3 days; 3 in day 1 (t = 1), 5 in day 2 (t = 2), and 2 in day 3 (t = 3). Therefore, for line item #3, the required quantity of component 1 is available by the third day. A similar search is carried out for component #3, which is available on the first day, and for component #4, which is available by the second day. Therefore, the component availability of line item #3 of order #231 is the 3rd day. In this example, let’s assume that the availability calculated for line item #1 is the 8th day, and that of the line item #2 is the 1st days. When all the components are available, the product is assembled or manufactured, which takes a certain amount of time. The manufacturing lead time can be a fixed number of days, or it can be described with a distribution function. The lead time to ship date is then calculated by adding the manufacturing (assembly) lead time to the availability lead time. Assuming that the manufacturing lead time for this

Analyzing the Effectiveness of the Availability Management Process

419

example is 2 days, the partial ship date for item #1 is the 10th day, for item #2 is the 3rd day, and for the item #3 is the 5th day, if the customer is willing to receive partial shipments. And the total order ship date is the 10th day from the date of order or request. Therefore, the lead time to ship date for the order #231 is 10 days for this example. When this order is scheduled, availability quantities are reserved (e.g., the availability is decremented) for the order. Typically, for each order, availability is reserved as late as possible so the availability in an earlier time bucket can be used for generating favorable ship dates for future orders. In this example as shown in Figure 18.1, the quantity of 10 for component 1 is reserved in t = 3, and the quantity of 10 for component 3 is reserved in t = 3. However, for component 4, the quantity of 5 is reserved for t = 1, and another 5 is reserved t = 2 instead of quantity 8 being reserved for t = 1 and 2 for t = 2. Having the availability of 3 for t = 1 is more valuable than t = 2 for scheduling and fulfilling future orders. The scheduling logic can vary based on business rules and policies. The scheduling can also be carried out by a pull side ATP optimization engine that optimizes order scheduling simultaneously considering inventory costs, backlog cost, and customer service impact, etc.

Order 231 Item1: qty:10 Item2: qty:10 Item3: qty:10

preference distribution of component features

sourcing policies

t=1

t=2

t=3

t=4 ……

t=N-1 t=N

3

5

10

10

....

comp1:

0

2

3

10

....

comp2:

10

5

5

0

....

comp3:

8

5

5

5

....

comp4:

....

comp M:

Order 232 Item1: qty:10 Item2: qty:10 Item3: qty:8

Due date (requested ship date)

components availability (t=4) = scheduling delay of 1 day

distribution of order arrival

Figure 18.2. Simulation of order promising and ship date calculation for advance orders

For orders with an advance due date, described earlier as the CCHW business of IBM, the simulation model looks for a specified quantity of a chosen component starting from the time period of due date (requested ship date), searches backward into the earlier time periods, and then forward to later time periods until the availability of all quantities are identified, as shown in Figure 18.2. For this example, item 3 of the order #231 requires the quantity of 10 of components #1, #2,

420 Young M. Lee

and #3. However, in this case, the order comes with a requested ship date of t = 3, say 3 days from the time of order. For component #1, the simulation model finds the availability of 10 on t = 3, and reserves the availability. For component 2, it finds the quantity of 3 on t = 3, then it searched backward to find 2 more on t = 2 and then moved forward to find 5 more on t = 4. But, in this case, the simulation reserves availability quantity of 10 all on t = 4 making the availability quantity intact for t = 2 and t = 3 for future orders. For component 3, the simulation model finds availability of 5 on t = 2 and t = 3 each, and reserves them. In this case the overall availability date is t = 4, a day after the due date. Therefore, the order scheduling delay here is 1 day. 18.3.3 Event Generation

In this work, the availability outlook changes as the result of four events; (1) demand event, (2) supply event, (3) roll-forward event, and (4) data refresh event, as shown in Figure 18.3. Each event changes the availability outlook; the demand event decrements the availability, the supply event increments the availability, the data refresh event refreshes the availability, and the roll-forward event shifts the availability, as explained in the next section. The events are generated independently using probability distribution functions or fixed intervals. The model can be easily extended to include more events depending on the supply chain environment being modeled. 18.3.3.1 Demand Event The demand event is a pull side of availability management, and it includes order scheduling and fulfillment. The demand event is triggered when customer orders are generated, and it decrements the availability outlook (quantity) when it schedules customer orders. Customer orders are generated in certain stochastic intervals, usually as a Poisson process. At this time of the order generation, each order is assigned one or more attributes such as quantity, product type, demand class, supply class, and due dates. This assignment of attributes is modeled with probability distribution functions based on historical sales data or expected business in the future. The orders travel through the business process as defined in the simulation model. When the orders reach a task which simulates the scheduling of customer orders, specific availability quantities are searched in the availability outlook, which are then reserved for the order and are decremented from the availability outlook. The reservation (consumption) of specific availability can be decided by the various policies and rules, such as the sourcing policy, scheduling polices, and fulfillment policies. The reservation of availability outlook can also be determined by availability promising engines described earlier. The ATP engines can be connected to the simulation model and communicate the optimal ATP reservation to the simulation model.

Analyzing the Effectiveness of the Availability Management Process

421

18.3.3.2 Supply Event The supply event is the push side of availability management, and it involves availability generation through schedules of production and procurement. The supply event is triggered in a certain interval, e.g., weekly or daily, and it increments the availability outlook. As finished products or building block components are reserved when customer orders are scheduled and fulfilled, additional availability is added to the availability outlook through production or procurement. This activity, supply event, is planned in advance, e.g., months, weeks, or days before the availability is actually needed to accommodate the lead time for production and procurement. As a result of supply planning, the availability outlook is updated and replenished. The replenishment frequency can be a fixed interval such as daily, weekly, etc, or it can be modeled using a distribution function. The replenishment quantity is typically determined based on the forecast of customer demand. The frequency and size of the replenishment are also decided by various replenishment policies. The allocation of the availability outlook can also be determined by availability planning engines, some of which were described in Section 18.1. These ATP engines can be connected to the simulation model and communicate the optimal ATP allocation to the simulation model. Availability Planning Engine

Roll forward event Supply event shift availability increment availability

static view of availability dynamic view of availability

... error

... Availability Promising Engine

decrement availability refresh availability Demand event Data Refresh event

Figure 18.3. Multiple events that affect availability

18.3.3.3 Roll Forward Event As the simulation clock moves from one time bucket to another, the availability of products or components that have not been consumed are carried forward to an earlier time bucket. For example, the availability quantity of the 2nd day moves to the availability quantity of the 1st day, and that of the 3rd day becomes that of the 2nd day, etc. Also, the availability quantity not consumed on the 1st day stays on

422 Young M. Lee

the same day, assuming it is nonperishable. The roll-forward event can be generated in a fixed interval, e.g., daily or weekly, depending on the business environment. 18.3.3.4 Data Refresh Event There are two instances of availability outlook; one represents the availability quantity in real-time (dynamic view of availability, or physical availability), and another represents availability recorded in the availability database (static view of availability, or system availability). The system availability is the one that is used for scheduling of customer orders, and it not always accurate. The system availability is synchronized with physical availability only periodically because it is expensive to have IT architecture that allows real-time synchronization. This synchronization between physical availability and system availability is modeled in the data refresh event. For example, the static view of availability is refreshed every few minutes, every hour, or even every few days. The discrepancy between the physical availability (dynamic view of availability) and the system availability (static view of availability) causes inaccurate ship date calculation. In our simulation model, the ship date is computed using both dynamic and static views of the availability, as shown in the Figure 18.3, and the inaccuracy of the ship date calculation from the system availability is estimated. The inaccuracy of the ship date calculation is an important indication of customer service level. The data refresh event can be modeled as a fixed interval event or a randomly generated event described by a distribution function. The analysis of the way the refresh rate impacts the ship date accuracy is described in Section 18.4.3.

18.4 Case Studies of AMST at IBM 18.4.1 Availability Management based on CTO vs. MTO Environment

In this scenario, one of IBM’s hardware businesses was interested in moving from a MTO environment to CTO environment, but they weren’t sure what would be the impact on the supply chain, specifically the customer service level and inventory. For this case, we have used the AMST to evaluate the benefits of managing availability on the component level vs. the finished goods level for the business. Using the model, we were able to quantify the improvement of supply chain performance with respect to order fulfillment rate, lost sales, and inventory. This business requires a high level of customer service. Customers configure products from available components when e-shopping or placing an order, and they expect a quick response time for a ship date quotation. The size of customer orders are highly skewed toward the end of the quarter, that is, a relatively small number of orders arrives in the beginning of a quarter (3 months), and more and more orders arrive toward the end of the quarter. The peak production and supply capacity are constrained. There are many uncertainties in this business including uncertainties in demand, order configuration, and supply. The availability

Analyzing the Effectiveness of the Availability Management Process

423

planning and order scheduling have been done at the FG (Finished Goods) level in the past, but a switch to the component level was being planned. For this study, we focused on key configurable components that are typically constrained. We have not included many other components that are an integral part of the product, a computer, but not constrained, such as power cable, case, and keyboard. We considered three key component types for the modeling; hard drive (HD), system memory (SM), and system processor (SP). We modeled six specific HDs, four SMs, and five SPs. Components in this environment are called ales building blocks (SBB), and preconfigured finished goods are called MTM (machine type model). For this simulation model, the availability outlook was allocated only by product type and time period. For modeling the preconfigured FG-based availability management (as-is process), we allocated the availability to 120 (HD x SM x SP) buckets, each representing a specific configuration of using 1 HD, 1 SM, and 1 SP for each time period. The planning time period was by week, and its horizon was 13 weeks. For the SBB-based availability management (to-be process), we allocated the availability to 15 (HD + SM + SP) buckets for each time period. For simplicity, we assumed uniform probability of each component to be picked for a FG configuration. This is a HVEC business, which was described earlier, and customer orders arrive without any advance notice requesting as early as possible fulfillment. Figures 18.4 and 18.5 show the simulation results on promised ship date for the FG-based availability management and component-based availability management, for the simulation duration of 1 quarter (90 days). As can be seen in Figure 18.4, for the FG-based scenario, the lead time of promised ship dates fluctuates mostly between 1 week and 5 weeks from the time of order arrival, then goes up to the 13 weeks and beyond toward the end of quarter. In contrast, as seen in Figure 18.5, for the component-based scenario, the lead time of promised ship date is mostly 1 week, and only occasionally 2 weeks, much better customer service. Table 18.1 summarizes and compares the order scheduling rate for the two scenarios. For SBB based availability management, 95.77% of orders was scheduled for the first week, 99.94% for the second week, and 100% by the third week, while for the FGbased, 74.22% for the first week, 89.89% for the second week, and 94.91% by the third week. The simulation result clearly demonstrates that the order fulfillment rate is higher with SBB-based scheduling than MTM-based scheduling. For MTM-based scheduling, 3.29% of orders couldn’t be scheduled within 5 weeks of order, and assuming that customers are not willing to wait 5 week, 3.29% is considered lost sales. For SBB-based scheduling, all orders that are scheduled can be fulfilled because specific components are reserved for orders when they are scheduled, and the components are available when it is time for the orders to be fulfilled. But for the MTM-based scheduling, 0.27% of orders that are scheduled couldn’t be fulfilled because MTM availability is reserved for orders when they are scheduled, but the required components to assemble the MTM are not available when it is time to fulfill the orders. The inventory of SBB for SBB-based scheduling is 3,791 (for all SBBs and first weekly bucket only), while that for the FG-based scheduling is 5,016, which is 24.5% higher than that of SBB-based scheduling. The simulation result clearly

424 Young M. Lee

shows that inventory can be kept lower with component based scheduling than with finished goods based scheduling. Scheduled Ship (MTM Based Scheduling)

Scheduled Ship (Week)

25

20

15

10

5

0 0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

90.0

Days

Figure 18.4. Promised ship date profile for FG-based availability management

Scheduled Ship (Component Based Scheduling) 25

Scheduled Ship (Week)

20

15

10

5

0 0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

90.0

Days

Figure 18.5. Promised ship date profile for component-based availability management

Analyzing the Effectiveness of the Availability Management Process

425

Table 18.1. Order scheduling rate

Comp-Based Scheduling (cumulative)

FG-Based Scheduling* (cumulative)

Week 1

95.77% (95.77%)

74.22% (74.22%)

Week 2

4.17 % (99.94%)

15.68% (89.89%)

Week 3

0.06% (100%)

5.02% (94.91%)

Week 4

0.00% (100%)

1.29% (96.21%)

Week 5

0.00% (100%)

0.50% (96.71%)

> Week 5

0.00% (100%)

3.29% (100%)

We also simulated the two scenarios (FG-based and component-based) with two different supply variabilities, e.g., uncertainty of supplier commitment. We used 5% and 10% standard deviations of supplier commitment for requested supply quantity and computed lost sales quantity and unfulfilled orders that are already scheduled. As seen in Tables 18.2 and 18.3, the lost sales are larger when the supply variability is larger, but component-based scheduling is more tolerant to supply variability than FG-based scheduling. Also, order promising accuracy is worse when supply variability is larger, but component-based scheduling is more tolerant supply variability than FG-based scheduling. Table 18.2. Lost sales for two different supply variability Comp-Based Scheduling

FG-Based Scheduling

Supply Std Dev = 5%

0.00%

3.29%

Supply Std Dev =10%

0.06%

4.29%

Table 18.3. Unfulfilled orders for two different supply variabilities Comp-Based Scheduling

FG-Based Scheduling

Supply Std Dev = 5%

0.00%*

0.27%

Supply Std Dev =10%

0.00%*

0.48%

The simulation results from this case study clearly show that component-based order scheduling is better than the finished goods-based order scheduling, and yields higher rates of order scheduling and fulfilment, lower inventory and potentially less lost sales or back-orders. Component-based order scheduling is also tolerant to supply variability.

426 Young M. Lee

18.4.2 Availability Management Based on Different Demand Classes

In this scenario, one of IBM’s hardware businesses was interested in managing availability based on new demand class, and they didn’t know how the new demand class would impact their supply chain performance, specifically their customer service and inventory cost. The business wanted to change from a demand class #1 representing four geographic demand regions to a new demand class #2 representing eight new geographic demand regions. For this case, we also used the AMST model to evaluate the impact of the demand class change on supply chain performance. We modeled and simulated four different scenarios based on different ways of availability allocation and order scheduling as shown in Table 18.4. Scenario 1 is the old (as-is) availability management process, where availability outlook is allocated based on 19 product types, 4 sources of supply, 4 elements of demand Class #1, and 13 weekly buckets. When an order is generated, the order is assigned with attributes, e.g., a product type, a source of supply, a demand class, and the customer requested ship date (also called due date). For scenario 1, the simulation model tries to schedule each order by searching for availability for a specific product, a source of supply, and a demand class, and then the weekly bucket that corresponds to the customer requested ship date. If no availability is found, the model goes back to earlier weekly buckets until it find availability. If availability is still not found, the simulation model looks for availability in later weeks until it finds availability. If no availability is found in any of 13 weekly buckets, the order is considered backlogged. For this case study, we simulated more than 100,000 orders which represent customer orders for the business for a year. From the simulation, we estimated the customer service and inventory holding costs for the scenario. Table 18.4. Four simulated scenarios for case study #2 Allocation of Availability Outlook

Constraint on Order Scheduling

Scenario 1

Product Type (19)

No constraint

(As-Is)

Source of Supply (4) Demand Class1 (4)

Scenario 2

Weekly buckets (13) Product Type (19)

(To-Be 1)

Source of Supply (4)

No constraint

Demand Class2 (8) Scenario 3

Weekly buckets (13) Product Type (19)

(To-Be 2)

Source of Supply (4)

Ceiling imposed by Product Type, Demand Class2 and Quarter

Weekly buckets (13) Scenario 4

Product Type (19)

(To-Be 3)

Source of Supply (4) Weekly buckets (13)

No constraint

Analyzing the Effectiveness of the Availability Management Process

427

Scenario 2 is the new (to-be) availability management process that the business would like to evaluate. For this scenario, availability outlook is generated based on 19 product types, 4 sources of supply and 13 weekly buckets. But, in addition, it is generated based on eight elements of Demand Class #2, which represent new geographic demand regions. Scenario 3 is another new (to-be) availability management process that the business would like to evaluate. For this scenario, availability outlook is generated based on 19 product types, 4 sources of supply, and 13 weekly buckets. It is not generated based on either Demand Class #1 or Demand Class #2. However, in this case, a constraint is imposed when scheduling an order. The constraint is a ceiling, which is a maximum allowed quantity for scheduling a specific product type and a specific demand class #2. The ceiling is usually imposed with a predetermined flexibility, 2%, etc. Scenario 4 is another new (to-be) availability management process that is similar to scenario 3, but there isn’t any ceiling imposed for the scheduling. Some of key data used in the simulation model are as follows. Customer orders are highly skewed toward the end of the 13 week period. The number of orders in the first week of the quarter starts at about 4% of quarterly volume, gradually increases, and for last two weeks of the quarter, the number of weekly order goes up to about 15% of quarterly orders. In addition to the weekly skew of orders, the weekly demand itself has variability. The variability of component supply is also modeled. The customer requested ship date (due date) is also skewed in that a large portion of orders arriving in the early part of the quarter request orders to be shipped in the latter part of the quarter, and the orders arriving in the latter part of the quarter request the orders to be shipped within a few weeks before the end of the current quarter. One of the key performance metrics we wanted to measure for this case study was scheduling delay. For this business, customer orders come with requested arrival dates (due date). Since the transportation lead time is known in advance based on the service level agreement with carriers, it is easy to figure out when the order should be shipped (requested ship date) so that the product arrives at customer’s place on the requested arrival date. The scheduling delay here, therefore, is defined as the difference between scheduled ship date and requested ship date. Figures 18.6-18.9 show the scheduling delays for the four scenarios for one product type. It is clear in Figures 18.6 and 18.7 that the scheduling delay gets worse when the demand class is changed from one that has less members (Demand Class #1) to one that has more members (Demand Class #2). This is obvious because when availability buckets are bigger, it is easier to schedule orders against them than when the availability buckets are smaller. As can be seen in Figure 18.8, the scheduling delay is substantially reduced when the demand class is dropped from the availability allocation. However, the ceiling creates significant constraint in scheduling toward the end of the quarter. Obviously, when the ceiling is dropped (Figure 18.9), the scheduling delay at the end of the quarter disappears. The scheduling delays for the four scenarios are summarized in Table 18.5.

428 Young M. Lee

Scenario 1 (As-Is) Order Scheduling Delay 14

weekly buckets

12 10 8 6 4 2 0 0

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

2000

2200

orders (for 3 months)

Figure 18.6. Order scheduling delay of scenario 1 (as-is)

Scenario 2 (To-Be 1) Order Scheduling Delay 14

weekly buckets

12 10 8 6 4 2 0 0

200

400

600

800

1000

1200

1400

1600

1800

orders (for 3 months)

Figure 18.7. Order scheduling delay of scenario 2 (to-be 1)

Analyzing the Effectiveness of the Availability Management Process

429

Scenario 3 (To-Be 2) Order Scheduling Delay 14

weekly buckets

12 10 8 6 4 2 0 0

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

2000

2200

orders (for 3 months)

Figure 18.8. Order scheduling delay of scenario 3 (to-be 2)

Scenario 4 (To-Be 3) Order Scheduling Delay 14

weekly buckets

12 10 8 6 4 2 0 0

200

400

600

800

1000

1200

1400

1600

1800

orders (for 3 months)

Figure 18.9. Order scheduling delay of scenario 4 (to-be 3)

Another the key performance metric for this case study was inventory holding cost. We assumed here that holding a product for one year costs 20% of the sales value. Table 18.6 compares inventory holding costs of the four scenarios. Scenario 2 would cost $2.827 million more than scenario 2 (as-is). However, scenarios 3 and 4 would generate substantial savings compared with the as-is scenario.

430 Young M. Lee

Table 18.5. Order scheduling delay for four scenarios

Order Scheduling Dealy wk0 wk1 wk2 wk3 wk4 wk5 wk6 wk7 wk8 wk9 wk10 wk11 wk12 >wk12 (backlogged)

Scenario 1 Scenario 2 Scenario 3 Scenario 4 (As-Is) (To-Be 1) (To-Be 1) (To-Be 1) 72.10% 70.74% 78.25% 78.26% 12.25% 11.57% 10.38% 10.42% 4.64% 4.85% 2.73% 2.74% 2.71% 2.99% 2.66% 2.70% 2.87% 2.97% 3.03% 3.18% 2.18% 2.04% 1.50% 1.61% 1.33% 1.23% 0.57% 0.75% 0.62% 0.78% 0.16% 0.19% 0.14% 0.59% 0.03% 0.02% 0.12% 0.28% 0.02% 0.01% 0.12% 0.25% 0.05% 0.03% 0.17% 0.33% 0.09% 0.02% 0.23% 0.46% 0.11% 0.03% 0.52% 0.95% 0.41% 0.04%

Table 18.6. Inventory holding costs for four scenarios Scenario 1

Scenario 2

Scenario 3

Scenario 4

(As-Is)

(To-Be 1)

(To-Be 2)

(To-Be 3)

Inventory Holding Cost

$13.135 million

$15.962 million

$9.405 million

$8.673 million

Inventory Holding Saving*

--

-$2.827 million

$3.730 million

$4.462 million

w.r.t. Scenario (As-Is)

The simulation results from this case study clearly show that demand class and the number of demand classes negatively impact customer service and inventory. More generally, the larger the ATP quantity, the better the supply chain performance. 18.4.3 Balancing Accuracy of Promised Ship Date and IT Costs

In an ideal e-business environment, when a customer order is scheduled and a ship date is computed, the availability should immediately be reserved and not be available for future orders. However, in reality, the availability data that are used for scheduling orders are not real-time availability (physical availability), but they are availability information stored in an IT system (system availability). The availability data in the IT system (static view of availability) is typically refreshed (synchronized with real-time availability) only periodically since it is very expensive to update the database in real-time. Due to the potentially inaccurate view of availability, some orders can’t be fulfilled on the promised ship date. Thus, for certain customer orders, products are shipped later than the promised ship date thus resulting in a degree of customer dissatisfaction. Therefore, one of the key

Analyzing the Effectiveness of the Availability Management Process

431

decisions in availability management is to properly balance the IT system (e.g., IT expense) and, customer service level. In this work, we studied how the availability refresh rate (IT system) impacts the customer service level. The simulation study helped the business group make a critical business decision on the refresh rate of availability and avoided the expensive investment of deploying a new IT system. Inventory Holding Cost ($) $18,000,000 $16,000,000 $14,000,000 $12,000,000 $10,000,000 $8,000,000 $6,000,000 $4,000,000 $2,000,000 $0 Scenario 1

Scenario 2

Scenario 3

Scenario 4

Figure 18.10. Inventory holding costs for four scenarios

The lead time to shipment is determined and provided to a customer at multiple times during the customer shopping process, at “Web-speed.” Customers make decisions on purchase based on the availability information (promised ship date) in addition to other criteria such as price and quality of goods. Once an order is placed, the customer expects the product to be delivered on the promised date. Since the promised ship date and its accuracy are directly related to customer service, it is very important to project them accurately before a new business process or its change is implemented. In this case study, we also used AMST to evaluate how the frequency of availability data refresh affects the accuracy of availability information given to customers. Figure 18.11 shows the ship date error profile for a three month period for a demand class when the frequency of availability data refresh is once a day. The figure shows that there are quite a few occurrences of ship date error, whose magnitude is mostly 1 week. The magnitude of the ship date error increases to 2 weeks toward the end of the quarter. Figure 18.12 compares ship date errors for four refresh frequencies, for orders arriving with three different demand classes for a specific business setting of the IBM hardware business. Table 18.7 also summarizes the simulation results. On average, the ship date error went down to 1.4% from 3.2% as the refresh frequency increases from once a day to four times a day. However, the ship date error does not decrease substantially as the refresh rate increases beyond 3 times a day.

432 Young M. Lee

Ship Date Error

3

Week

2

1

0 0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

orders

Figure 18.11. Ship date error for DM class 1 with once a day refresh

Ship Date Error vs Refresh Frequency 6.00

Ship Date Error (%)

5.00 4.00 3.00 2.00 1.00 0.00 1/day

2/day

3/day

4/day

refresh frequency DM Class 1

DM Class 2

DM Class 3

Weighted Average

Figure 18.12. Ship date error for three refresh frequencies

Figure 18.13 shows the trade-off between ship date error and IT cost for refreshing the availability outlook in the IT system. As is shown, as the refresh rate increases from once a day to four times a day, the IT costs increase substantially from $1.2 million to $2.3 million. Although the general relationship

Analyzing the Effectiveness of the Availability Management Process

433

between ship date error and IT costs not a surprise, the quantification of the tradeoff is key information that business leaders need to have to make sound business decisions on the availability management process. The right decision is balancing the ship date error (customer service) and IT costs that is reasonable for a business at the time of analysis. The simulation results from this case study clearly show that an IT system that handles the availability management process impacts the accuracy of ship date calculation when customer orders are processed. Simulation is a key tool in determining the trade-off between IT costs and supply chain performance. Table 18.7. Ship date error summary for various refresh frequencies Ship Date Error GEO

Once a day Refresh

Twice a day Refresh

3 Times a day Refresh

4 Times a day Refresh

Demand

2.156%

1.511%

1.011%

1.044%

EMEA

3.254%

1.841%

1.333%

1.127%

AP

5.000%

3.000%

2.463%

2.389%

Weighted Average

3.225%

1.997%

1.485%

1.419%

4.00

2.4

3.50

2.2 2

3.00 1.8 2.50 1.6 2.00 1.4 1.50

1.2

1.00

1 1/day

2/day Ship Date Error (%)

3/day

4/day IT Costs

Figure 18.13. Trade-off between ship date error and IT costs

IT Costs ($ million)

Ship Date Error (%)

Ship Date Error vs IT Costs

434 Young M. Lee

The simulation models described above for the cases studies were all validated by examining the simulation outputs of the as-is cases. After validation of the as-is cases, simulation models of to-be cases were used for analysis.

18.5 Concluding Remarks Availability management directly influences key supply chain performance such as customer service and inventory. In the current dynamic, competitive business environment where an organization has to adopt continually to accommodate customer’s need it is very important to be able to assess the existing availability management process and explore various ways to improve it. In this chapter, we described the availability management simulation tool (AMST) that was developed for IBM’s computer hardware businesses. The model simultaneously simulates the three main components of the availability management process; generating the availability outlook, scheduling customer orders, and fulfilling the orders, as well as the effect of other dynamics in the supply chain. The AMST is a simulation modeling framework for availability management processes, and it has reusable components and methods, which are easily adapted for various availability management environments. The tool has been instrumental in evaluating and deploying several availability management transformation opportunities in IBM.

18.6 Guidelines for Practitioners The AMST (availability management simulation tool) was developed using the simulation engine of IBM, WBI Modeler ® (IBM Corporation). However, the capability can also be built using other commercially available discrete-event simulation modeling tools such as ProModel (http://www.promodel.com/), Arena (http://www.arenasimulation.com/), Witness (http://www.lanner.com/home/the_value_of_knowing.php), to name a few. Although several ATP engines have been developed by other researchers to optimize the push (availability generation) and pull (order promising) sides of ATP, there hasn’t been any tool that evaluates the effectiveness of such ATP tools as well as various policies in various order management environments. AMST is such a tool that can simulate the availability management process by simultaneously modeling many components and dynamics, including ATP tools. The core of the simulation model is described in the Figure 18.3, which is part of the order processing process through which customer orders travel. Figure 18.14 shows an overview of a simple availability simulation model we developed. In this sample model, the rectangles represent various tasks (and events), circles represent several views of the availability outlook, and the arrows represent the movement of artifacts (customer orders in this case). Generation of orders (or online shopping) is modeled in the first rectangle on the left side of Figure 18.14, and the general availability of product, features, and price are also displayed to customer here. The orders then proceed to the next task where a specific product is

Analyzing the Effectiveness of the Availability Management Process

435

configured from the availability of components. Ship date is also determined here in the availability check (shop) task, which accesses the IT system that contains availability outlook data. If the customer is satisfied with the ship date, the order moves to the next step, the availability check (buy) task, and is submitted. A promised ship date is calculated again here using the availability outlook data and order scheduling policies. The submitted order goes through the order processing task in the back-office and order fulfillment process, where the availability is physically consumed. The tasks specified as rectangles in Figure 18.14 can have certain processing time. They can also require certain resources such as an IT server, a part of whose resource is tied up in processing orders. As is common for much modeling work, substantial time and effort are expended for collecting and validating data in developing a simulation model of the availability management process. The availability management simulation capability we described here is planned to be incorporated into a larger simulation tool that handles a much bigger scope of supply chain management.

RollForward Availability Data (Rollforward)

SupplyPlanning (Supply) P9

P21

Availability Data Refresh (Refresh)

Availability Data Repository (System)

P22

Check Availability of Configuration (Shopcheck)

P10

Order Fulfillment & Execution Availability Data Repository (Physical)

Confirm Availability of Configuration (Buycheck)

Catalog input P33

input P37

P38 output

Web Tool

Availability Check (Learn)

P7

P6

OrderEntry (Learn)

P35 P5

Availability Check (Shop)

P4

b

P36

P31

OrderSubmit (Buy) P100

P32

Users

P34 output

P3

Availability Check (Buy)

P2

p9

Order Processing in Back Office

Configuration

Figure 18.14. A sample availability management simulation model

18.7 Acknowledgements The author would like to thank Joe DeMarco, Brad Howland, Barun Gupta, Bill Armstrong, and Ken Cyprys of the IBM Integrated Supply Chain (ISC) group for sharing their knowledge and experience in IBM’s availability management processes, and colleagues in IBM Research; Tom Ervolina, Soumyadip Ghosh,

436 Young M. Lee

Feng Cheng, and Steve Buckley, for many useful discussions and contributions to the work.

18.8 References Bagchi S, Buckley S, Ettl E, Lin G, (1998) Experience using the IBM supply chain simulator, Proceedings of the 1998 Winter Simulation Conference, (eds.) Medeires DJ, Watson EF, Carson JS and Mani-vannan MS: 1387–1394. Ball MO, Chen C-Y, Zhao Z-Y, (2004) Available To Promise. Handbook of Quantitative Analysis: Modeling in an E-Business Era, (eds.) Simchi-Levi D, Wu D, Shen ZM, Kluwer: 446–483. Chen C-Y, Zhao Z-Y, Ball MO, (2002) A model for batch advanced available-to-promise. Production and Operations Management 11(4): 424–440. Ervolina T, Dietrich B, (2001) Moving toward dynamic available to promise. In supply Chain Management: Status and Future Research, Gass S and Jones A (eds) preprints produced by R. H. Smith School of Business, U. of MD and Manufacturing Eng. Lab., NIST. Hieta S, (1998) Supply chain simulation with LOGSIM-simulator. Proceedings of the 1998 Winter Simulation Conference, (eds.) Medeires DJ, Watson EF, Car-son JS, Manivannan MS: 323–326. McClellan M, (1992) Using simulation to facilitate analysis of manufacturing strategy. Journal of Business Logistics 13(1): 215–237. Moses S, Grand H, Gruenwald L, Pulat S, (2004) Real-time due-date promising by build-toorder environments. International Journal of Production Research 42: 4353–4375. Pan Y, Shi L, (2004) Stochastic on-line model for shipment date quoting with on-line delivery gurantees. Proceedings of the 2004 Winter Simulation Conference, (eds.) Ingalls RG, Rossetti MD, Smith JS, Peters BA. Yee S-T, (2002) Establishment of product offering and production leveling principles via supply chain simulation under order-to-delivery environment. Proceedings of the 2002 Winter Simulation Conference, (eds.) Yücesan E, Chen CH, Snowdon JL, Charnes JM: 1260–1268.

About the Editors

Hosang Jung is currently a chief researcher at the Samsung Economic Research

Institute, Seoul, Korea. He received B.S. and Ph.D. degrees in industrial and systems engineering from Yonsei University, Seoul, Korea. He was a postdoctoral research fellow at the Grado Department of Industrial and Systems Engineering, Virginia Tech, USA from 2004 to 2006. He was awarded the Korean Government Scholarship for his postdoctoral program in 2004 and received the best research paper award from the Korean Supply Chain Management Society in 2002. Also, he has been listed in Marquis Who’s Who in Science and Engineering since 2006.

F. Frank Chen is currently the John L. Lawrence Endowed Professor of

Manufacturing Systems Engineering at the Grado Department of Industrial and Systems Engineering, Virginia Tech, USA. He received a B.E. degree from Tunghai University, Taiwan, and M.S. and Ph.D. degrees in industrial engineering from the University of Missouri-Columbia, USA. He was one of the NSF (National Science Foundation) nominated engineering professors in the U.S. who received the 1996 Presidential Faculty Fellows Award at the White House.

Bongju Jeong is currently a professor at the Department of Information and

Industrial Engineering, Yonsei University, Seoul, Korea. He received B.S. and M.S. degrees in industrial engineering from the Seoul National University and Ph.D. degree from Pennsylvania State University, USA. He is an executive member of the Korean Supply Chain Management Society. He worked as a manufacturing systems consultant in Andersen Consulting & Co. and as a senior researcher in the Semiconductor Division, Samsung Electronics & Co.

Index

A

Abstraction-mechanism specification, 160 Accessibility, 182, 188, 201, 379, 386, 387, 389, Activity sampling, 397 Advanced planning systems, 93, 120, 138 Agency, 153, 154, 162–164, 166, 169, 172, 181, 218, 232, 325 Agent collaborative agent, 106, 124, 290, 291, 294 manufacturer agent, 299 distributor agent, 299 control agents, 154, 155, 163, 165, 167, 168 flow control agent, 165, 168 inventory control agent, 165, 168 production control agent, 165 supply control agent, 165 problem-solving agents, 153, 155, 159, 165 data management agent, 159, 165 simulation agent, 160, 165 structural agents, 154, 165 distribution center agent, 165 external supply agent, 165 manufacturing plant agent, 165, 168

marketing agent, 165, 166, 168 retailer agent, 165, 168 supply agent, 165, 327 transportation agent, 154, 155, 165 Agent architecture, 105, 121, 123, 153, 160–163, 175, 325, 326, 328 agent body, 163 agent repository, 163, 164, 166 agent-based annotation process, 157 basic agent architecture, 153, 163 information and interface agent, 155, 156, 159 intelligent agents, 123, 152, 160, 172, 173, 175, 311, 330 multiagents, 153 Agent platform, 99, 109, 114, 115, 121, 122 Agent technology, 95, 97, 99, 106, 107, 124, 125, 149, 152, 171, 175, 312, 325, 326, 330, agile business processes, 150 agile supply chain processes, 150 agility enablers, 150, 167, 169, 172 Algorithm, 19–22, 37, 77, 108–110, 153, 158, 174, 192, 211, 291, 311, 315, 317, 321, 327–329, 349, 358, 369, 381, 382, 387, 416 Alternative configurations, 167, 169, 172 Annotation mechanism, 155 5W1H, 156, 162

440

Index

annotated knowledge, 155–157 annotation process, 155–157 Articulation rules, 157–159 Articulation mechanism, 149, 155, 157, 158, 169, 172 Assemble-to-order, 132, 375 Assembly, 6, 7, 22, 23, 27, 36–38, 46, 47, 117–119, 185, 187, 192, 193, 199, 203, 204, 207, 243, 269–271, 352, 356, 367, 370, 374, 376, 390, 396, 418 operation, 47, 376 Asset tracking, 10, 23, 38 Auction, 143, 147, 177, 225, 232, 291, 309, 315, 317–322, 324, 326, 327, 329, 330 Audit results, 401, 402 Authentication, 27, 38, 39, 227, 232 Auto ID, 25, 26, 32 auto ID center, 25, 26, 32 Automated pipeline inventory and order based production control system (APIOBPCS), 334 Automated storage and retrieval system (AS/RS), 15, 22 Automatic articulation generator, 158 Autonomy, 99, 108, 226

B

Backroom, 33, 34, 61, 75 Balance-production, 269, 271, 274, 276–278, 282, 284 Bar code, 4, 30, 31, 33, 37, 41, 44, 46, 50, 54, 56, 57, 64, 71, 73, 74, 83, 87–91, 97, 254 Base stock, 377 level, 377 Baseline, 6, 21, 23, 74, 75, 334, 339, 342, 400–403, 407 Batch sizes, 335, 340 Batch-and-queue, 264, 397, 399, 400 Bayesian, 371 network, 371 Behavior performance deteriorating, 372

Benchmark ontology, 157 direct benefits, 50, 52, 54, 56–58, 61, 66, 68, 69 indirect benefits, 54, 58, 59, 60, 61, 68, 69 Best practice, 238, 309, 393, 394, 402, 406–408 Bidding, 143, 147, 318–321, 326, 329 Bill of materials (BOM), 183, 185, 193, 194, 201 Building block, 129, 130, 139, 241, 249, 250, 259, 265, 266, 362, 366, 374, 381, 418, 421, 423 basic, 265, 381 Bullwhip effect, 10, 58, 70, 77, 78, 101, 118, 147, 248, 313, 315, 325, 327, 328, 331, 338 Bullwhip phenomenon, 133, 134, 137 Business case, 4, 24, 31, 49–53, 60, 61, 69, 89, 138 Business process model, 49, 50, 58 Business process reengineering (BPR), 131, 239 Business process transformation, 58 Business systems engineering (BSE), 398 Business value model, 49–51, 53, 58, 60–63, 69, 70 Business-to-business (B2B), 134, 142, 216

C

Capital flow, 370, 374 Cash flows, 51, 67, 68, 127, 165 Causal relationships, 397, 398 Centralized supply chain, 7, 22, 291, 293, 296 planning, 293, 296 Certification, 144, 145 ISO, 144 Chemistry, 384 Client-server architecture, 128, 129, 137, 140 Clockspeed, 147 Closure, 380–383, 390

Index

Clustering, 189, 192, 193, 210, 406 Coefficient of variation, 338 Coinput, 376, 380 Collaborative agent, 106, 124, 290, 291, 294 Collaborative planning, forecasting and replenishment (CPFR), 132 Communications, 3, 5, 12, 25, 26, 91, 123–125, 129, 135, 142, 147, 175, 217, 226, 229, 238, 253, 409 network, 97, 312 routing protocol, 11, 14, 25, 27 scheduling protocol, 10 wireless, 5, 13, 97 Competitiveness, 241, 252, 265, 313, 322–324, 409 Complex flow symptoms, 397, 399– 405 Complex material flow, 397, 398, 400–402, 405, 407, 408 Computation, 56, 58, 66, 102, 110, 131, 273, 281, 282, 284, 285, 307, 311, 312, 313, 319, 324, 325, 327, 328, 350, 366, 381, 386 computational ecosystem, 312, 328 Computing technology industry association, 4 Conception schema, 156, 157 Configuration management, 36 Connectivity, 13, 14, 135, 137, 138, 141, 255, 371, 376, 379, 380, 381, 382, 384, 385–387, 389 index, 381 structural, 255, 385–387 Construction sector, 405 Content provider, 224 Contingency, 177, 369 Contracts, 142, 144, 145, 148, 174, 175, 417 Conveyor, 7, 9, 15, 22, 23, 42, 64, 256 Consequence analysis, 371 Countermeasure analysis, 371 Coordination, 4, 6, 44, 105, 108, 116, 123, 127–137, 141–148, 155, 174–177, 210, 215, 219, 262,

441

290–294, 309–329, 343, 370, 396, 406, 414 Cooutput, 376, 380 Correlation, 394, 398, 399, 407 Cost reduction, 60, 142, 143, 234, 318, 321–324, 351 Cost-benefit analysis, 12, 53, 89, 167 Cost-time-profile, 257, 258 Counterfeit protection, 33, 38, 39 Cross docking, 250, 251, 267 Customer service, 36, 57, 59, 60, 65, 66, 88, 89, 201, 234, 313, 314, 331, 332, 334, 337, 340–342, 350, 352, 397, 411, 412, 414, 419, 422, 423, 426, 430, 431, 434 Cycle time, 56, 71, 72, 134, 165, 168, 169, 213, 214, 251, 256, 270, 271, 273, 274, 277–279, 281, 285, 287, 378, 393, 404, 405

D

Data analysis, 337 Data integration, 230, 235, 236 Data layer, 112 Data modeling, 397 Data warehousing, 72, 175, 177 Decentralized decision-making, 170, 290, 291, 307 Decentralized model, 319 Decentralized supply chain planning, 289–292, 294, 295, 299, 307–309 Decision support system (DSS), 125, 334 Decision support tools, 49 Decision theory, 3, 49, 71, 127, 241, 269, 311, 345, 369 Decision trees, 51 Decoupling point, 271, 273, 285 Demand rate, 81, 83, 250, 359, 362, 378 Demand volumes, 195 Demand forecast, 417 Demand surge, 371, 372 demand volatility, 133 Design for localization, 140, 287

442

Index

Design for manufacturability, 140 Deviation, 17, 64, 81, 83, 85, 118, 334, 357, 370, 371, 378, 425 operating, 371 Digital divide, 230, 235 Digital equipment corporation, 127, 210 Disassembly, 181–185, 193–195, 199, 201, 203–205, 207, 211, 376 structure, 376 Disaster, 369, 370, 385, 391 Disruption, 93, 95, 98, 107, 109, 115, 118, 121, 122, 369, 370–374, 377, 389, 404 Distributed artificial intelligence, 152, 174, 175, 327 Distributed discrete-event, 152 Distributed simulation techniques, 152 Distribution, 7, 8, 15, 17, 18, 25, 34, 39, 56, 59, 61, 70, 75, 165, 167, 168, 173, 250, 269, 270, 271, 273, 282–284, 290, 292, 296–299, 302, 306, 312, 337, 374, 376, 390 arborescent, 374, 390 center, 7, 8, 15, 17, 18, 25, 34, 39, 56, 59, 61, 70, 75, 165, 167, 168, 173, 250, 269, 270, 271, 273, 282–284, 290, 292, 296– 299, 302, 306, 312, 337, 376 chain, 269, 270–272, 274, 282, 284–286 Distributor-driven supply chain, 290, 291, 293, 299, 300, 303, 305, 308, 309 District, 385 neighboring, 385 Diversification, 345, 346, 353, 354, 356, 357, 362, 364, 366, 367, 391 Dock door, 7, 15–18, 22, 23 Domain knowledge base, 161 Dominant strategy, 317, 320 Dual sourcing, 349, 350, 354–358, 361, 366, 367, 371 Dynamic behavior, 58, 332, 337, 397–399, 407

Dynamic lot-sizing, 311, 314–317, 325, 326

E

e-Administration, 214, 219 e-Business, 70, 128, 129, 147, 148, 173, 175, 213, 216–218, 223, 224, 227, 234, 235, 239, 329, 430, 436 Echoupal, 146, 148 e-Collaboration, 128, 129, 134–136, 144, 145 e-Commerce, 128, 129, 132, 144, 146, 173, 175, 176, 218, 219, 226, 234, 411–413 Economies of scan, 41, 44 Ecosystems, 145, 146, 328 e-Delivery, 218, 225, 233, 273, 355, 356 e-Democracy, 218, 219, 228, 231 Edge, 20, 90, 158, 376, 380, 381 Effective lead time, 346–349, 351, 361 Efficiency improvement, 234, 370 Efficient consumer response (ECR), 132 e-Governance, 217, 218, 219, 220 e-Government, 217–231, 233–239 framework, 214, 227 portal, 228 strategic value chain, 213, 222, 233 strategy, 214, 220–223 vision, 214, 220, 221, 233, 234 Electronic data interchange (EDI), 73, 135, 150 Electronic product code (EPC), 9, 32, 70, 74, 97 Electronic public service delivery, 224, 225, 237 Electronics industry, 343 e-Markets, 142–144 Enterprise resource planning (ERP), 93, 128 Entropy, 79–81 Environmental pressure, 181, 370

Index

e-Participation, 228 e-Pedigree, 38, 39 e-Procurement, 128, 129, 132, 140, 143, 144, 214, 219, 224, 225, 231–233, 237, 238 e-Purchasing, 224 e-Readiness, 230 e-Services, 214, 219, 221, 223, 229 e-Supply chain, 27, 92, 213, 215, 216, 239 Ethical considerations, 76 Euclidean norm, 401 Event definition markup language, 110–112 Event, 14, 44, 49, 56, 93–96, 98, 99, 101–111, 114, 115, 118, 121–125, 152, 156, 161–165, 227, 232, 319, 322, 369, 371, 377, 385, 414–422, 434 disruptive, 118, 371, 374, 377 systemwide, 370 unexpected, 93, 121, 122, 166, 261, 369, 370, 390 Event-handling knowledge base, 161 Event selection mechanism, 162, 164, 165 Every day low pricing (EDLP), 339 Exchange curve, 36, 37, 351 Expected performance, 372, 373 Expected value, 81, 372, 373 Explorer mechanism, 162, 164, 166 Exponential smoothing, 21 Extensible markup language (XML), 23, 100, 107–110, 137, 138, 150, 156, 160–163, 173–176, 254, 255

F

Fabrication, 269–275, 278–281, 285– 287, 352, 370, 374 Facility, 23, 37, 186, 210, 233, 251, 256, 257, 291, 297, 299–303, 306, 311, 313 Fill rate, 81, 256, 281, 282, 334–336, 350, 351, 374, 377 First-order, 83–85, 379

443

stochastic dominance, 83– 85 Five rules to avoid bankruptcy, 401 Forecasting, 7, 21, 22, 132, 189, 253, 254, 266, 327 Forrester, 58, 70, 101, 123, 140, 404, 409 flywheel effect, 404 Fourth party logistics (4PL), 293 Frequency, 14, 18–20, 26–32, 50, 59, 62, 75, 96, 144, 265 397, 401, 421, 431 Full service provider, 224 Functional mechanism, 162, 164–166 Functional risk, 4, 7, 10–12, 18, 19, 22 Functional product, 140, 294 Fuzzy logic, 371

G

Gen2 standard, 32 Geographic separation, 401 Global performance measures (GPM), 168 Global positioning system (GPS), 97 Government value chain, 213, 230, 231, 233, 235, 236 Government-to-business (G2B), 222– 224, 233 Government-to-citizen (G2C), 222– 224, 233 Government-to-employee (G2E), 222, 223, 233 Government-to-government (G2G), 222, 223, 233 Graph theory, 19, 391 Grocery sector, 331, 332, 337, 342 Groupware, 135

H

Heterogeneity, 149–152, 155, 157, 169, 172, 226, 230, 329 Heterogenous envirnment, 151 Hierarchical model specification, 160

444

Index

High frequency (HF), 30, 397 Human decision-maker, 116, 338 Hurricane, 370 Hypergraph, 376

I

IBM, 5, 25, 27, 33, 52, 62, 69, 70, 139, 237, 411–426, 431, 434–436 Incentive compatibility, 315 , 317, 326 schemes, 145 Indicator, 137, 230, 256, 373, 374, 377, 397, 404 Indirect path, 379 Inference, 161, 162, 163, 371 engine, 371 Information sharing, 4, 8, 10, 26, 27, 91, 106, 120–124, 131–134, 146, 147, 174, 176, 218, 260, 290–295, 308–315, 321, 322, 325–329 flow, 3, 4 , 5, 9–11, 24, 44, 46, 47, 73, 87, 96, 124, 127, 130, 155, 165, 216, 227, 249–252, 257, 260, 370, 374, 396, 404 real-time data, 60, 10, 11, 27 confidentiality, 321, 322, 323, 324 revelation, 318, 319, 321, 326 security, 311, 312, 313 privacy, 290, 291 system, 25, 47, 59, 107, 112, 120, 122, 124, 126, 147, 148, 150, 167, 169, 170–177, 202, 210, 219, 230–239, 252–255, 259, 262, 265, 327, 329, 391, 397, 399, 400, 406 Information and communication technologies (ICT), 132, 148, 213–219, 223, 226, 228–236 framework, 214, 215, 226, 228, 234, 235 Information technology (IT), 7, 71, 73, 90, 127, 130, 147, 149, 150, 173–175, 216, 226, 237, 253, 273, 290

Input-output analysis, 394 Integrated index, 385, 386, 388–391 Intelligent product, 8, 27 Intelligent simulation multi-agent tools, 160 Interference, 9, 13, 14, 17, 18, 19, 20, 23, 24, 33, 42, 397 Intermediary, 224 Internal rate of return, 51, 67 International standards organization, 32, 50 Interoperability, 9, 32, 102, 106, 130, 139, 150, 151, 157, 175–177, 225–229, 235, 239 Interval, 256, 372, 413, 417, 420–422 Interviews, 21, 337 Intragovernment processes, 224, 225 Inventory accuracy, 36, 64, 75 control, 18, 21, 35, 36, 44, 74, 91, 154, 165, 167, 168, 173, 333, 334, 336, 340, 343, 346, 348, 377, 390 cost, 71, 73, 77, 82, 83, 88–90, 250, 251, 273, 296, 294, 313, 316, 318–322, 346, 351, 362, 364, 368, 370, 419, 426 policy, 90, 351, 377, 406, 407 management, 4, 8, 15, 20–27, 71, 73, 75, 81, 88, 92, 106, 131, 328, 353, 365–367 sharing, 8, 26 shrinkage, 58, 59, 64, 65, 66 Invertex, 376, 380 Incapacitated production target (IPT), 278, 279 International standards organization (ISO), 32, 50 Isolation, 10, 12, 43, 45, 385, 391

J

Just-In-Time (JIT), 89, 153, 247, 250, 333, 343, 362, 371 delivery, 247, 371

Index

K

Key performance measures, 54, 107 Knowledge repository, 169, 170 Knowledge resource, 149, 151, 155, 161, 162, 172 Knowledge-based government (kgovernment), 235 Knowledge-based systems, 152 Knowledge-sharing platform, 157 Knowledge representation mechanism, 160 Key performance index (KPI), 54, 107

L

Lead time variability, 56, 83, 91, 347, 348, 350, 364, 365, 370, 371 Lean production, 247, 251, 253, 267, 333 Lean supply, 247, 248–263, 265–268 Lean technique, 266, 370 Lean logistics, 241, 248, 249, 250, 252, 259, 265, 266, 267 Lean manufacturing, 243–248, 250, 251, 261, 262, 265, 268 Lean supply chain, 241–243, 246– 268 Lean thinking, 241–243, 246–248, 261, 263–266, 268, 400, 402, 403 Likert scale, 394, 398 Linkage, 60, 67, 260, 379 Little’s law, 271, 378 Localization, 94, 97, 140, 181–190, 287, Location-allocation model, 195 Logistics, 4, 8, 9, 12, 15, 25, 26, 33– 39, 43–45, 48, 59, 69, 72, 90, 91, 96, 103, 105, 115, 116, 125, 141– 144, 173, 174, 181–198, 201–206, 210, 211, 215, 219, 236–239, 241, 248–253, 259, 264–267, 271, 284, 290–295, 309, 310, 325–328, 337, 341–346, 351, 366, 367, 391, 408, 409, 436

445

Lost sale, 2 Lot size, 327, 329, 352, 356, 377

M

Machine, 23, 120, 143, 156, 157, 169, 243, 256, 274, 275, 281, 286, 329, 343, 365, 370, 404, 409, 423 breakdown, 359, 365, 370 Make-to-order, 132, 291, 374, 412, 415 Make-to-stock, 132, 275, 374, 415 Management theory, 394, 406 Manufacturer, 6, 7, 22, 23, 31, 32, 37, 39, 42, 43, 46, 50–52, 57, 60–62, 75, 77, 78, 115–119, 128, 130– 135, 140–144, 243, 247, 264–267, 270, 289–296, 299–304, 307, 308, 331, 337, 338, 340, 382, 387 Manufacturer-driven supply chain, 290, 292, 293, 299, 300, 303, 304, 308 Manufacturing execution, 4, 10, 253 Market dynamics, 319 mediation, 131, 132, 144 sectors, 394, 406 Mass customization, 133, 134 Material flow, 18, 23, 41, 44, 46, 118, 127, 129, 130, 133, 165, 192–198, 201–205, 215, 254, 290, 314, 370, 374, 375, 393, 394, 396–409 Matrix, 143, 379, 380–384, 386–390 adjacency, 380, 383, 390 Mean, 8 Mean lead time, 351 Mean value, 370 Measure, 53, 79, 118, 153, 168, 172, 257, 279, 280, 338, 342, 348, 351, 371, 372, 377–379, 414, 427 aggregate, 371 basic, 379 critical, 370, 378 key, 377 quantifiable, 371 Mechanical precision products, 405 Medium access control (MAC), 13

446

Index

Message handling, 160, 161, 164 Message-input processor, 160, 162 Metadata, 111, 156, 157 Metadata repository, 157 Metagraph, 369, 374–376, 380, 381, 384, 385, 390, 391 Metric, 14, 256, 257, 340, 404, 407, 429 Microsoft, 62, 91, 139, 177, 390 .NET, 139 Microwave, 27, 32 Military supply chain, 251 Milk run, 251, 252 Minimum information sharing, 294, 295, 308 Mobile devices, 93, 97, 99, 106–109, 112, 113, 120–124 Mobile sensor technology, 98 Mobile supply chain event management, 93, 94, 121 Mobile supply chain management, 95 Mobility, 13, 14, 108, , 379 Monolithic model, 311, 313, 315, 316, 318, 322, 324 Maintenance repair and operating (MRO), 142–144 Multi-agent system, 106, 108, 112, 113, 121, 122, 174, 175, 310, 311–317, 325, 326, 329, 330 Multiple sourcing, 345, 346, 353, 361–367 Multiple supplier inventory models, 346

N

Negotiation, 106–108, 111, 112, 116, 153, 164, 166, 173–175, 291, 296, 304, 309, 327 mechanism, 164, 166 Net present value, 37, 51, 61, 67 Network, 124, 125, 173, 174, 181, 187, 196, 203, 210, 259, 260, 369, 370, 371, 374, 376, 379, 381 Bayesian, 371 hybrid, 371

integrated, 371, 390 distribution, 210, 273, 290, 292, 296, 297, 327 manufacturing, 309 performance reliability, 374, 377, 385, 386 Normal distribution, 62, 81, 87

O

Object-oriented manufacturing simulation language (OMSL), 160 Ontology, 108, 120, 151–158, 161– 164, 174, 176 Operational characteristics, 397–399 Operational parameters, 152, 167, 169, 172 Operator, 6, 13, 21, 23, 39, 81, 264, 380, 381 multiplication, 380 Oracle, 101, 107, 139, 413 Enterprise Manager, 139 Order quantity, 65, 66, 82, 83, 300, 309, 316, 346, 347–349, 352–361 Organization, 6, 9, 24, 32, 50, 73, 77, 95, 102, 109, 127, 128, 133, 136, 146, 151, 154, 168, 171–174, 184, 187–192, 194, 197, 201, 206, 217, 224, 229, 249, 255, 264, 285, 291, 325, 346, 371, 434 Networked, 128, 145 individual, 136, 371 Organizational characteristics, 397, 400 Out-of-stock, 33, 34, 53, 59, 60, 61, 65, 75 Outvertex, 376, 380

P

Pair, 375, 376, 379 connected, 379 ordered, 376 unconnected, 379 Parallel processing, 376

Index

Payback period, 51, 67 Performance curve, 372 actual, 372 desired, 372 Performance improvement, 217, 407 Personal computers, 396, 415 Physical markup language, 396 Physical situation, 397–400 Planning horizon, 53, 61, 67, 278, 296, 297, 299, 311, 314, 316, 322, 415 Policy, 15, 36, 62, 65, 66, 75, 77, 83, 90, 115, 164, 172, 173, 181, 217, 218, 229, 235–239, 273–275, 278, 284, 285, 287, 377, 389, 390, 406, 407, 412, 416, 420 best, 390 optimal, 363, 389 robust, 389 Pools of inventory, 397, 400, 401 Portfolio analysis, 52 Possibility, 51, 55, 102, 143, 197, 221, 292, 305 Postponement, 132, 146, 407, 409 Presentation layer, 112 Price discount, 296, 307, 308 Privacy, 35, 40–43, 48, 71, 76, 77, 90–92, 120, 122, 228–230, 235, 290, 291, 294, 308 Probability, 19, 20, 38, 62, 79, 81, 85, 90, 96, 117, 181, 193, 194, 201, 202, 216, 335, 347–349, 355, 359, 371, 372, 377, 378, 417, 420, 423 Process mapping, 337, 397, 398 Processing condition, 183, 187, 193, 201 Product differentiation, 375 disassembling, 376 finished, 43, 69, 181, 182, 184, 185, 188, 192, 193, 197, 199, 201, 203, 207, 312, 374, 412 product families, 183, 187, 190, 192–197, 199, 206 product mix, 262 product states, 183, 187, 193

447

Production chain, 269–272, 274, 284, 285, 286 Production lead time, 117, 119, 246, 296 Profit and loss (P&L), 53, 67 Profit shares, 324 Profit margin, 294, 405 Programmable logic controller (PLC), 22 Property, 158, 323, 352 Pull-distribution, 269, 272, 274, 284 Pull-production, 269, 271, 274, 280, 282, 284 Purchase price, 264, 345, 347, 349– 352, 354, 357, 360–362 Push-distribution, 269, 272, 274, 283 Push-production, 271, 274, 275, 277, 282, 284

Q

Quality of service (QOS), 13, 26, 326 Quantitative alternative, 167–169 Quantitative index, 385 Questionnaires, 337, 397 Quick response, 146, 148, 422 Quick scan audit methodology (QSAM), 337

R

Radio-frequency identification (RFID) antenna, 7, 17, 18, 23, 24 business case, 4, 24, 31, 49–53, 60, 61, 69, 89, 138 business justification, 61 business value, 49–53, 58, 60–63, 67–70, 177 data-based decision-making, 3, 5, 12 reader, 6, 7, 10, 18, 29, 31, 33, 34, 37–39, 42, 97

448

Index

tag, 5, 6, 9, 13, 16, 18, 19, 21, 24, 29–33, 35–43, 45–72, 76, 77, 90, 91, 97 technology, 3, 4, 6, 8, 9, 12, 17, 22, 25, 26, 29, 33, 40–44, 47, 50– 53, 60, 61, 65, 68–71, 73–76, 83, 88v91 Random capacity, 345, 361, 362 Random supplier availability, 345, 359, 362 Random yield, 345, 352–359, 361, 362, 364–368, 391, 345, 352–359, 361, 362, 364–368, 391, Raw material, 6, 7, 23, 59, 69, 121, 150, 165, 185, 192, 193, 247, 250, 269, 271, 312, 331, 338, 353, 365, 374 Readiness assessment, 220, 223, 233, 236 Real options, 51, 69, 70 Realistic guidelines, 407 Reorder point, 335 Replenishment, 34, 36, 38, 40, 58, 59, 62, 65, 66, 75, 81, 132, 173, 251, 265, 317–321, 326, 327, 332, 333, 337, 338, 343, 347, 356, 358, 363–366, 374, 414, 421 plan, 317–321, 326 Request-for-information (RFI), 144 Request-for-quote (RFQ), 144 Resiliency, 373, 374 Resource valuation, 319 Responsiveness, 72, 219, 273, 393, 404–406 Retailer, 7, 26, 31, 39, 40, 43, 50, 51, 53, 56–62, 65, 73–78, 117–119, 129, 130, 148, 165, 168, 250, 264, 270, 312, 337–342 , 347 Return volumes, 181–183, 185, 187, 189 Return-on-investment (ROI), 51, 53 Revenue, 53, 54, 57, 60, 62, 66, 74, 87, 88, 186, 216, 271, 272, 284, 285, 353, 416 Reverse flows, 130, 189, 191, 196, 197, 210, 216

Robustness, 16, 311, 369–374, 378, 379, 381, 384–391 index, 369, 371, 377, 384–390

S

Safety stock, 36, 64, 81, 88, 313, 333, 339, 347, 348, 351, 362, 364, 367 SAP, 98, 101, 107, 124, 128, 130, 141, 139, 413 AG, 128 Netweaver, 139, 107 R/2, 128 R/3, 128 Scalability, 121, 230, 325 Scenario analysis, 51, 331 Second-order stochastic dominance, 86 Security, 38, 48, 91, 92, 120, 229 Semantic interoperability, see heterogeneity, 150, 151, 157, 175, 176 Semiconductor supply chain, 270, 273 Sensor, 40, 93, 98, 99, 122 Sequence diagram, 115 Serial processing, 376 Serialized products, 36 Service execution management, 166 Service forecast errors, 190 Service layer, 122 Service level, 21, 36, 57, 59, 65, 81, 182, 183, 191, 192, 201, 225, 256, 332–335, 337–340, 342, 348–350, 352, 361, 371, 374, 377, 379, 386–388, 390, 397, 411, 422, 427, 431 Service-Oriented Architectures, 97, 100 Selling, general, and administrative expenses (SG&A), 54 Shape, 20, 98, 199, 372 Shared infrastructure, 224 Shelf replenishment, 34, 62, 75 Shop floor control, 3, 8, 15, 22, 125 Shortest path, 14, 379

Index

Simple object access protocol, 100 Simplicity rules, 404, 408 Simulation definition markup language, 110, 111 Site, 182, 375 Situation control mechanism, 162 SKU (stock keeping unit), 73, 75, 334, 375 Slap and ship, 45 Smooth demand, 333 smooth material flow, 393, 394, 404 Software engineering, 152, 173, 327, 328 Sole sourcing, 350, 354, 355, 363, 366 Spreadsheet, 49–51, 58, 68, 69, 334 Staged models, 223 Standard deviation, 17, 64, 81, 83, 85, 118, 334, 357, 378, 425 Stochastic dominance, 71, 83–87, 89, 91 Stochastic lead time, 81, 345–347, 349–351, 364, 365, 367 Stock level, 81, 333, 338, 348, 349, 377 Stockman, 5, 27, 30, 48 Store, 8, 23, 31, 33–35, 40, 42, 46, 47, 53, 57, 58, 59, 61, 62–65, 73, 91, 92, 97, 161, 163, 226, 229, 265, Strategic level, 184, 197 structural robustness, 369, 371, 374, 379, 381, 385, 386, 389, 390 Structured questionnaire, 397 Subset, 54, 68, 218, 363, 365, 384 Summation, 372, 379, 380, 386 Supplier, 6, 7, 22, 23, 36, 42, 43, 46, 52, 94, 95, 97, 115–118, 135–137, 143, 144–146, 148, 165, 167, 168, 173, 197, 203, 204, 241, 242, 251, 252, 260, 272, 316–326, 331, 334, 338, 339, 340, 341, 342, 345–368, 370, 378, 391, 395, 396, 403–405, 408, 412, 414, 416, 425 selection, 144, 348, 356, 358, 361, 362–366

449

Supply lead time, 56, 349, 351, 362, 364, 377 Supply network-centric, 3, 5, 6, 7 Supply quantity, 299–307, 345, 346, 352, 361, 362, 364, 425, Supply timing, 345, 346, 361 Supply uncertainty, 345, 346, 361, 362, 365, 366, 367, 391, 399 Supply chain event management, 93, 94, 101, 108, 121, 122, 125 Supply chain management, 93, 95, 101, 103, 106, 107, 122–125, 127, 129, 130, 134, 147–149, 152, 169, 172, 174–177, 210, 214–216, 233, 235–239, 241–243, 246–248, 252, 253, 267, 268, 273, 286, 287, 294, 309–314, 326–331, 343, 344, 365–367, 390, 391, 394, 407, 409, 411, 414, 435, 436 Supply chain dispersed, 385 hypothetical, 369 member, 121, 247–249, 252, 254, 258, 370 network, 73, 124, 182, 211, 269, 285, 329, 369–371, 374, 376, 379, 381, 382–390 performance, 49, 52, 72, 70, 74, 136, 248, 331, 334, 336, 341, 370, 372, 373, 404, 407, 411, 413, 414, 422, 426, 430, 434 risk, 391 robustness, 373 structure, 329, 370 design, 127, 130, 132, 133, 140, 144, 145, 146, 187, 391, 396, 398, 407 coordination, 44, 127, 131–134, 137, 148, 290, 327 Survivability, 325, 326, 328 Survival, 89, 289, 290, 369, 370, 412

450

Index

T

Tag ID, 31 item-level, 9, 31, 33, 34, 39, 41– 44, 47, 52, 54, 57, 58, 62, 70 Tagging, 6, 9, 31, 33, 41, 43, 48, 52, 54, 58, 61, 77, 89 Tags, 4–6, 9, 10, 13, 14, 16, 18, 19, 21, 29v54, 57, 58, 61, 62, 67, 69, 71, 72, 76, 77, 89–91 , 97, 156 Target balance (TB), 280 Target wip, 271, 277–279 Target-production, 272, 278, 279, 280, 282, 284 Technology acceptance model, 169 Technology Transfer, 4, 13 Tellkamp, 52, 70 attack, international, Terrorism, 369, 370, 371, 391 Third-party logistics (3PL), 252, 290, 310 Three-dimensional concurrent engineering (3D-CE), 140 Throughput, 44, 165, 168, 169, 256, 270–277, 280–284, 378 Time buckets, 398, 415, 435 Time-sensitive materials, 21, 27 Topological index, 369, 384, 385, 389, 390 Topological length, 379 Total accessibility, 387 Total connectivity, 371, 379–390 Total cycle time (TCT), 393, 404 Toyota, 147, 243, 244, 267, 370 Toyota Production System (TPS), 244, 243 Traffic congestion, 312, 336 Transaction costs, 127, 129, 132, 134, 137, 143, 144, 146 Transferability, 393, 394, 406 Transformation, 53, 58, 59, 110, 131, 161, 175, 185, 213, 214, 215, 217, 223, 227, 230, 231, 233–237, 243, 245, 261, 265, 289, 370, 412, 414, 415, 434

Transit-time, 374, 377, 378, 379, 386, 387, 388 Transmission control protocol /internet protocol (TCP/IP), 105, 137 Transport, 95, 117, 119, 153, 210, 282, 292, 296, 331, 332–338, 340–334 Transportation, 7, 34, 35, 40, 43, 57, 60, 64, 99, 114, 117, 118, 131, 154, 155, 165, 168, 175, 183, 186, 189, 192, 195, 197, 205, 210, 250–253, 256, 259, 261, 265, 274, 282–285, 292, 296–299, 301, 302, 306, 312, 343, 344, 350, 362, 365, 366, 370, 374, 391, 414, 427 Transshipment, 8, 26, 27, 282, 283 Triangulated evidence, 337 ordered, 340 Triple, 124, 380, 381 Trustworthiness, 315, 319, 321, 322 Trustworthy multiagent system, 316, 317

U

U.S. department of defense, 4, 9, 23, 43 Ultra high frequency (UHF), 18, 30, 32 reduction principle, 393, 394, 406, 407 Uncertainty, 7, 12, 14, 51, 79, 87, 92, 124, 202, 211, 319, 339, 345, 346, 349, 352–358, 360–362, 364–367, 374, 391, 393–409, 417, 425 Universal description discovery integration, 166 Unstable feedback loop, 404 Utility function, 83–87

Index

451

V

W

Value decomposition, 55 Value net integrator, 224 Value stream, 241, 244–252, 257, 265, 267, 337, 341, 393, 394–408 Map, 245, 247, 249, 257, 267 Value tree, 54, 55 performance, 54, 58–60, 64, 65, 67, 70 Value decomposition, 55 Value net Integrator, 224 Variance, 52, 71–73, 78–92, 277, 346–352, 361, 371, 378 Velocity management (VM), 72 Vendor managed inventory (VMI), 7, 8, 15, 22, 251, 331–333, 343, 344 Vickrey auction, 317, 320 Virtual models, 12, 13 Volatile feedback, 397 Value stream map (VSM), 247, 249, 257 assessment, 249 Vulnerability, 371

Wafer fabrication, 270, 287, 370 Wal-mart, 4, 7, 9, 23, 43, 50, 53, 70, 72, 75, 76, 92, 248, 250 Warehouse, 7, 14–17, 23, 26, 30, 31, 34, 41, 45, 52, 53, 56, 79, 90, 95, 183, 188, 195–199, 201, 203–209, 227, 246, 254, 264, 269–271, 274, 282–284, 312, 374 Web presence, 230, 231 Web services, 97, 100, 106, 107, 113, 114, 120, 123–130, 137, 139, 146, 147, 161, 166, 228, 229, 239 description language, 100, 166 Web digital government (WebDG), 228 Well-trodden path, 395, 407 What-if analysis, 51 Winner determination, 318, 319, 326 Work-in-process (WIP), 250, 270, 271, 274–284, 286, 287 , 343, 378, 399, 416