Proceedings of the International Symposium for Production Research 2018 PDF

The conference aims at forming a unique platform to bring together academicians and practitioners from industrial engineering and management engineering as well as from other disciplines working on production function applying the tools of operational research and production/operational management. Topics treated include: computer aided manufacturing, industry 4.0, big data and analytics, flexible manufacturing systems, fuzzy logic, industrial applications, information technologies in production management, optimization, production economy, production planning and control, productivity and performance management, project management, quality management, risk analysis and management, supply chain management.

108 downloads 4K Views 125MB Size

Report

Download pdf

Recommend Stories

Empty story

Idea Transcript

Numan M. Durakbasa M. Günes Gencyilmaz Editors

Proceedings of the International Symposium for Production Research 2018

Proceedings of the International Symposium for Production Research 2018

Numan M. Durakbasa M. Günes Gencyilmaz •

Editors

Proceedings of the International Symposium for Production Research 2018

123

Editors Numan M. Durakbasa Faculty of Mechanical and Industrial Engineering Technische Universität Wien Vienna, Austria

M. Günes Gencyilmaz Faculty of Engineering Istanbul Aydın University Istanbul, Turkey

ISBN 978-3-319-92266-9 ISBN 978-3-319-92267-6 https://doi.org/10.1007/978-3-319-92267-6

(eBook)

Library of Congress Control Number: 2018950774 © Springer Nature Switzerland AG 2019 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Foreword

Dear Colleagues and Friends, It is a great pleasure to welcome you to the “International Symposium for Production Research—ISPR 2018” at the TU Wien, between 28 and 31 August, 2018, with the theme of “Impacts of Industry 4.0 on Production Systems”. We are particularly pleased with our collaboration with Society for Production Research, İstanbul, Turkey, in hosting this year’s Symposium in Vienna. The purpose of the symposium is to bring together researchers, scientists and experts at universities, companies, institutions, communities, associations and societies to share ideas and vision on production and operations management and technology and to provide a forum for experts and professionals to discuss the actual theme “Impacts of Industry 4.0 on Production Systems” and the further relevant developments in this area. We hope that this gathering will offer a venue where the participants will enrich their professional knowledge base and will present an opportunity to all the participants to build and deepen friendships with experts, industry and corporate professionals and authorities.

August 2018

Kurt Matyas Vice-rector for Academic Affairs

v

Preface

The 18th International Symposium for Production Research “ISPR2018” was held in Vienna, Austria, from 28 to 31 August 2018, with the theme of “Impact of Industry 4.0 on Production Systems”. This symposium was organized in Vienna, for the second year in a row, by TU Wien and Society for Production Research, Istanbul, Turkey. The generic theme of “Industry 4.0” was ﬁrst adopted in the symposium held in 2016 and maintained in the following two symposia in 2017 and 2018 but with an emphasis on the relevant developments and progress on the various aspects of this “fourth industrial revolution”, thus symposium adopting the same main theme but taking various aspects of it into account, with the purpose of drawing the attention of the researchers to the influences of this new industrial era on the production systems and production management. The world of science and technology is under increasing influence of the requirements of Industry 4.0. Transition to a new era seems inevitable for every sector of industry. Due to the importance of this theme, ISPR2018 hosted numerous distinguished speakers from both the academia and industry to hear their views on the impacts of Industry 4.0 on the various components of production systems. This volume of proceedings from the symposium contains 76 refereed papers in 18 categories shown in the contents of the proceedings. Participants from 11 countries attended the symposium. 12 invited talks and 76 papers were presented in 19 sessions. We are very grateful to our host institution, the Vienna University of Technology, for its invaluable support and hospitality and for enabling this symposium to be organized, for a second time, in their premises. In particular, we would like to express our gratitude to Vice-Rector Prof. Kurt Matyas, also the Honorary Chairman of the Scientiﬁc Committee of this symposium for his leadership and generous support. We also thank to Prof. Detlef Gerhard, Dean of the Faculty of Mechanical and Industrial Engineering, Prof. Friedrich Bleicher, Head of the Institute for Production Engineering and Laser Technology for their interest and support for this symposium.

vii

viii

Preface

We would like to thank all the keynote and invited speakers whose contributions enhanced the success of the symposium. In organizing this event, our colleagues in Vienna and Istanbul contributed endless hours of hard work, energy and wisdom to make this event the success it was. On the Vienna side, our sincere thanks go to the staff of the Department for Interchangeable Manufacturing and Industrial Metrology of the Institute for Production Engineering and Laser Technology, in particular, Mr. Erol Güclü. On the Istanbul side, we are very grateful to Prof. Zaim and Prof. Çebi who made available the resources of their departments, and to Ms. Hatice Camgöz Akdağ who was involved in every aspect of the organization from the very beginning. Our special thanks go to Ms. Tuğçe Beldek, and Mr. Kemal Konyalıoğlu, as well as our research assistants, for their hard and dedicated work. We would like to express our gratitude to the board members of the Society for Production Research in Istanbul for their strong support and dedication to make the symposium a success. Our very special thanks go to our colleagues, the participants of this symposium. Undoubtedly, they are the core component of this organization. We would like to recognize and thank our dear colleagues who graciously accepted to join the honorary and scientiﬁc committees or who served as peers in this event. Finally, no such event is possible without the generous support of patrons and sponsors. In this regard, we would like to thank Dr. Michael Ludwig, the Mayor of Vienna, for hosting a reception for the participants of this symposium, and to all the corporations and individuals who provided invaluable ﬁnancial and intellectual contributions. And last, but not least, we are grateful to Ms. Silvia Schilgerius, Senior Editor Applied Sciences from Springer Nature, for her able guidance, professionalism and patience. M. Güneş Gençyılmaz Numan M. Durakbaşa

Production of the Future

The Industry 4.0 is an innovative concept and model for future enterprises which is initiated with the aim to provide cost-effective, efﬁcient, agile and optimal ways to customer-driven design and production. Companies operating in various production areas in the future will be adapting more and more Industry 4.0 technologies that were originally modelled in smart manufacturing and multi-functional integrated factories (MFIF). The transition to the Industry 4.0 requires models to be integrated by utilizing advanced information analytics, artiﬁcial intelligence and interconnected Industrial Internet of things (IIoT) as a part of automated and robotic applications, networked intelligent machines and instruments. Intelligence is an essential feature of future development and production systems and intelligent production is a major component of future businesses, also according to the technological developments especially in the ﬁeld of precision engineering at micro/nano and pico scale production. Modern production engineering and production metrology and its industrial application started on the basis of the scientiﬁc, technical and organisational work of E. Abbe, William Taylor and F. W. Taylor and at the end of the twentieth century—the “Quality Control Century”—the development has reached nanotechnology and is proceeding to pico- and phytotechnology. To meet market demands in a global industrial world, manufacturing enterprises must be flexible and agile enough to respond quickly to product demand changes also according technological developments especially in the ﬁeld of precision engineering at micro-, nano- and pico-scale production with support of artiﬁcial intelligence (AI) and modern Information technology (IT), modern cost-effective customer-driven design and manufacturing. Adequate knowledge in the areas of intelligent production and especially intelligent metrology are important presuppositions to achieve waste-free production and low costs of manufacturing and accuracy at the same time within the sophisticated production systems. This is of extreme importance in the current and future worldwide competition in industry and production engineering and at the same time in the face of increasingly higher costs of energy and raw material. Learning with self-improving ability makes possible the way to “Zero Error” production. Fuzzy logic will be applied for quality function deployment (QFD) and ix

x

Production of the Future

for the monitoring and forecasting of maintenance of measuring instruments, and further on for intelligent design and for a high-level knowledge-based expert system for tolerancing and quality planning. The supervision of quality in the process chain as well as its optimization by means of knowledge and neuronal-network-based learning management system, and a self-learning system with neuronal networks can be used in the factory of the future according to Industry 4.0 concept to learn stepwise from deviations and to improve the processes continuously. On the basis of the technological developments, the manufacturing sector faces radical structural changes, with the digital transformation offering enormous opportunities but also presenting high-level challenges. The challenge today is implementing the concepts of the automation creatively with the possibilities of new sensors and instruments that allow metrological technologies and integrate telepresence production processes towards a complete platform industry 4.0 not only in large production complexes but mainly in the SMEs. Amid the challenges of competitive and digital technologies in the advanced manufacturing industry, the universities and industry cooperation are imperative. The new aspects of cooperation between the research institutions and the manufacturing industry will provide a development capacity for high-quality and innovative products in the frame of Industry 4.0 concept. August 2018

Günes Gencyilmaz Numan Durakbasa

Organization

ISPR2018 was organized by TU Wien, Austria, and the Society for Production Research, Turkey. The symposium took place on the TUtheSky Campus of the TU Wien.

Editors Numan M. Durakbaşa M. Güneş Gençyılmaz

Co-editors Peter Kopacek Ayhan Toraman Selim Zaim Jorge Martin Bauer Serpil Erol Semra Birgün Kemal Güven Gülen Andreas Otto Alptekin Erkollar Mahmut Tekin

Honorary Chair Kurt Matyas

TU Wien, Austria

xi

xii

Organization

Symposium Chairs Numan M. Durakbaşa M. Güneş Gençyılmaz

TU Wien, Austria Istanbul Aydın University, Turkey

International Honorary Committee Andrew Kusiak Duc Truong Pham Albert Weckenman Daniela Popescu Detlef Gerhard Sorin Popescu Vidosav Majstorovic Stanislaw Adamczak Friedrich Bleicher Lubomir Šooš Jozef Peterka

The University of Iowa, USA University of Birmingham, UK University of Erlangen-Nürnberg Germany Technical University of Cluj-Napoca, Romania TU Wien Technical University of Cluj-Napoca, Romania University of Belgrade, Serbia Kielce University of Technology, Poland TU Wien Slovak University of Technology in Bratislava, Slovakia Slovak University of Technology, Slovakia

Organizing Committee Ayhan Toraman, Turkey Selim Zaim, Turkey Güneş Gençyılmaz, Turkey Erol Güçlü, Austria Semra Birgün, Turkey Serpil Erol, Turkey Kemal Güven Gülen, Turkey Alptekin Erkollar, Turkey Eva Walcher, Austria Ferhan Çebi, Turkey Günther Poszvek, Austria Alp Baray, Turkey Hatice Camgöz Akdağ, Turkey

Prepared for Publishing by Tuğçe Beldek, Turkey A. Kemal Konyalıoğlu, Turkey

Haluk Soyuer, Turkey Hür Bersam Bolat, Turkey Ece Soyuer, Austria Ezequiel Simon, Austria Jorge Bauer, Austria Tuğçe Beldek, Turkey Kemal Konyalıoğlu, Turkey Ignacio Dorna, Austria Gökçen Baş, Austria Gamze Uğur Tuncer, Austria Oskar Mayer, Austria David Riepl, Austria Liane Höller, Austria

Organization

xiii

Scientiﬁc Committee S. Adamczak H. Camgöz Akdağ K. Altinel B. Baki Ş. Baray G. Baş N. Başoğlu J. Bauer A. Baykasoğlu E. Bayraktar F. Berto S. Birgün P. Blecha F. Bleicher F. Bozbura I. Budak A. Bulgak M. Bulu L. A. Crisan F. Cus F. Çebi D. Delen H. Demir M. Demirbağ B. Dengiz T. Dereli M. Dinçmen Á. Drégelyi-Kiss N. Durakbaşa B. Durmuşoğlu A. Erkollar S. Erol Ş. Esnaf S. Firat W. Fisher G. Gencyilmaz D. Gerhard S. Gözlü S. Grozav K. Gülen A. Güngör V. Gyula

Kielce University of Technology, Poland Istanbul Technical University, Turkey Bosphorus University, Turkey Karadeniz Technical University, Turkey Istanbul University, Turkey Vienna University of Technology, Austria İzmir Institute of Technology, Turkey National Technological University Buenos Aires, Argentina 9 Eylül University, Turkey The American University of the Middle East, Kuwait Norwegian University of Science and Technology, Norway Fenerbahçe University, Turkey Brno University of Technology, Czech Republic Vienna University of Technology, Austria Bahçeşehir University, Turkey Faculty of Technical Sciences in Novi Sad, Serbia Concordia University, Canada İstinye University, Turkey Universitatea Tehnica Cluj-Napoca, Romania University of Maribor, Slovenia Istanbul Technical University, Turkey Oklahoma State University, USA Yaşar University, Turkey University of Essex, UK Başkent University, Turkey İskenderun University, Turkey Istanbul Technical University, Turkey Óbuda University, Hungary Vienna University of Technology, Austria Istanbul Technical University, Turkey Sakarya University, Turkey Gazi University, Turkey Istanbul University, Turkey Marmara University, Turkey University of California, Berkley, USA Istanbul Aydın University, Turkey Vienna University of Technology, Austria Istanbul Technical University, Turkey Universitatea Tehnică din Cluj-Napoca, Romania Namık Kemal University, Turkey Pamukkale University, Turkey University of Miskolc, Hungary

xiv

E. Hekelová F. Holesovsky Z. Irani V. Işler C. Kahraman İ. Kara T. Katoka M. Klumpp D. Kocaoğlu T. Koç P. Kopacek M. Králik L. Kräuter C. Kubat U. Kula A. Kulakli J. Kundrak A. Kusiak V. Majstorovic I. Mankova A. Markopoulos K. Matyas M. Novák Y. Omurtag N. Özçakar D. Özdemir A. Özok E. Öztemel J. Peterka D. Pham M. Pokusová S. Popescu D. Popescu E. Pucher A. Sağbaş J. Sanz Juan A. Sharif A. Smith Ľ. Šooš A. Sorguç A. Soysal H. Soyuer K. Stepien B. Tan M. Tanyaş

Organization

City University Bratislava, Slovakia J. E. Purkyně University in Ústí nad Labem, Czech Republic University of Bradford, UK Hasan Kalyoncu University, Turkey Istanbul Technical University, Turkey Başkent University, Turkey Kindai University, Japan Duisburg University, Germany Portland State University, USA Istanbul Technical University, Turkey Vienna University of Technology, Austria Slovak University of Technology in Bratislava, Slovakia Vienna University of Technology, Austria Sakarya University, Turkey American University of the Middle East, Kuwait American University of the Middle East, Kuwait University of Miskolc, Hungary University of Iowa, USA University of Belgrade, Serbia Technical University of Kosice, Slovakia National Technical University of Athens, Greece Vienna University of Technology, Austria J. E. Purkyně University in Ústí nad Labem, Czech Republic Robert Morris University, USA Istanbul University, Turkey Bilgi University, Turkey Okan University, Turkey Marmara University, Turkey Slovak University of Technology, Slovakia University of Birmingham, UK Slovak University of Technology in Bratislava, Slovakia Technical University of Cluj-Napoca, Romania Technical University of Cluj-Napoca, Romania Vienna University of Technology, Austria Namık Kemal University, Turkey Polytechnic University of Valencia, Spain University of Bradford, UK University of Auburn, USA Slovakia University of Technology, Slovakia Middle East Technical University, Turkey Doğuş University, Turkey Ege University, Turkey Kielce University of Technology, Poland Koç University, Turkey Maltepe University, Turkey

Organization

F. Taşgetiren H. Taşkin M. Tekin A. Toraman J. Torgersen O. Torkul G. Ulusoy G. Varga K. Velíšek J. Wang A. Weckenmann M. Yalçintaş M. Yenisey M. Yurci S. Zaim W. Zębala M. Zerenler

xv

Yaşar University, Turkey Sakarya University, Turkey Selçuk University, Turkey Istanbul Technical University, Turkey Norwegian University of Science and Technology, Norway Sakarya University, Turkey Sabancı University, Turkey University of Miskolc, Hungary Slovak University of Technology, Slovakia National University of Singapore, USA University of Erlangen-Nürnberg, Germany Istanbul Commerce University, Turkey Istanbul University, Turkey Yıldız Technical University, Turkey Istanbul Şehir University, Turkey Cracow University of Technology, Poland Selçuk University, Turkey

Reviewers Baray, Ş. Bayraktar, E. Bayyurt, N. Bereketli-Z., İ. Berto, F. Beyca, Ö. Blecha, P. Bolat, H. Budak, I. Bulut, Ö. Camgöz Akdağ, H. Crisan, L. Çebi, F. Dragomir, M. Dregelyi-Kiss, Á. Ekinci, E. Erkollar, A. Erol, S. Esnaf, Ş. Gergin, Z. Gülen, K. Kazançoğlu, Y. Kesikburun, D. Kızılay, D.

Kopacek, P. Krauter, L. Küçükdeniz, T. Mankova, I. Mullaoğlu, G. Novak, M. Öner, A. Öner, E. Özcan, S. Özdemir, D. Öztürkoğlu, Y. Pokusova, M. Soyuer, H. Staiou, E. Stepien, K. Şahingöz, Ö. Tekin Temur, G. Torgersen, J. Üney Yüksektepe, F. Üstündağ, A. Varga, G. Yurtseven, C. Zaim, S. Zębala, W.

Contents

Artiﬁcial Intelligent Applications Multi-objective Optimization Study in Face Milling of Steel . . . . . . . . . János Kundrák, Angelos P. Markopoulos, Tamás Makkai, Nikolaos E. Karkalos, and Antal Nagy RSM and Neural Network Modeling of Surface Roughness During Turning Hard Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pavel Kovač, Mirfad Tarić, Dragan Rodić, Bogdan Nedić, Borislav Savković, and Dušan Ješić

3

16

Big Data and Analytics A Data Mining Approach to Predict E-Commerce Customer Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Büşra Altunan, Ebru D. Arslan, Merve Seyis, Merve Birer, and Fadime Üney-Yüksektepe Data Mining in Digital Marketing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mahmut Tekin, Mehmet Etlioğlu, Özdal Koyuncuoğlu, and Ertuğrul Tekin

29

44

Computer Aided Manufacturing Investigation of Flatness and Angularity in Case of Ball-End Milling . . . Balázs Mikó and Gábor Rácz

65

Decision Making A DBN Based Prognosis Model for a Complex Dynamic System: A Case Study in a Thermal Power Plant . . . . . . . . . . . . . . . . . . . . . . . . Demet Özgür-Ünlüakın, İpek Kıvanç, Busenur Türkali, and Çağlar Aksezer

75

xvii

xviii

Contents

An Assessment Model for Organizational Adoption of Industry 4.0 Based on Multi-criteria Decision Techniques . . . . . . . . . . . . . . . . . . . . . Fatma Demircan Keskin, İnanç Kabasakal, Yunus Kaymaz, and Haluk Soyuer

85

Analysis of Reactive Maintenance Strategies on a Multi-component System Using Dynamic Bayesian Networks . . . . . . . . . . . . . . . . . . . . . . 101 Demet Özgür-Ünlüakın and Ayşe Karacaörenli Comparison of Hydraulic Bending Machines for Proﬁle, Pipe and Beams in Manufacturing Companies with Electre Method . . . . . . . 111 Elif Çirkin, Aşkın Özdağoğlu, and Kevser Yılmaz Evaluation of Metal Forming Machine Alternatives Used for Production Activities with MOORA Method . . . . . . . . . . . . . . . . . . 124 Aşkın Özdağoğlu, Kevser Yılmaz, and Elif Çirkin Prediction of Industry 4.0’s Impact on Total Productive Maintenance Using a Real Manufacturing Case . . . . . . . . . . . . . . . . . . . 136 Ebru Turanoglu Bekar, Anders Skoogh, Nihan Cetin, and Osman Siray The Selection of a Process Management Software with Fuzzy Topsis Multiple Criteria Decision Making Method . . . . . . . . . . . . . . . . . 150 Arda Yiğit Şen, Neslihan Semiz, Buse Güneş, Duygu Algül, Zeynep Gergin, and Nurcan Demirok Dönmez Fuzzy Logic A Multimoora Method Application with Einstein Interval Valued Fuzzy Numbers’ Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Hatice Camgöz-Akdağ, Gökhan Aldemir, and Aziz Kemal Konyalıoğlu Healthcare Systems and Management The Comparison and Similarity Study Between Green Buildings and Green Hospitals: A General View . . . . . . . . . . . . . . . . . . . . . . . . . . 181 Hatice Camgöz-Akdağ, Aziz Kemal Konyalıoğlu, and Tuğçe Beldek Industrial Applications Accuracy of Ducts Made with Various Processing Strategies . . . . . . . . . 193 L. Nowakowski, M. Skrzyniarz, and E. Miko Acquisition of Measurement Data on a Stand for Durability Tests of Rolling Bearings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 Jaroslaw Zwierzchowski, Dawid Pietrala, Pawel Andrzej Laski, and Henryk Lomza

Contents

xix

Amplitude Surface Texture Parameters of Models Manufactured by FDM Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 Tomasz Kozior and Stanisław Adamczak Analysis of the Impact of Ball Bearing Track Waviness on Their Frictional Moment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 Stanisław Adamczak, Łukasz Gorycki, and Włodzimierz Makieła Application of Wavelet Transform to Determine Surface Texture Constituents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 Damian Gogolewski and Włodzimierz Makieła Assessment of the Accuracy of Laser Vibrometer for Measurement of Bearing Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 Stanisław Adamczak and Mateusz Wrzochal Challenges of Miniaturizing a Precision Gear . . . . . . . . . . . . . . . . . . . . 239 Numan M. Durakbasa, Jorge M. Bauer, Osman Bodur, and Günther Poszvek Cross Mark Coordinate Determination and Automatic Registration for Offset Printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 N. Gökhan Kasapoğlu, Zeynep Gergin, M. Güneş Gençyılmaz, Fadime Üney-Yüksektepe, Fatma Kutlu-Gündoğdu, and Ayşe Bilge Torbalı Design and Development of Human Machine Interface for Unmanned Aerial Vehicle Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Fatih Akkoyun, İsmail Böğrekci, Pınar Demircioğlu, and Salih Vardin Design of a Test Stand for Rolling Bearing Durability Testing . . . . . . . 275 Pawel Andrzej Laski, Dawid Sebastian Pietrala, Jarosław Zwierzchowski, and Henryk Łomża Implementation of a Real-Time Data Acquisition System with Wireless Sensor Network for Temperature Measurement . . . . . . . . . . . 280 İsmail Böğrekci, Fatih Akkoyun, Pinar Demircioğlu, and Salih Vardin Integrations Management and Product Development for New Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292 Semih Dönmezer Measurement Error on the Reconstruction Step in Case of Industrial Computed Tomograph . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309 Ágota Drégelyi-Kiss and Numan M. Durakbasa Precision Metrology for Additive Manufacturing . . . . . . . . . . . . . . . . . . 324 Binnur Sagbas, Tahir Hakan Boyacı, and Numan M. Durakbasa

xx

Contents

The Evaluation of a Contact Proﬁlometer Measuring Tip Movement on the Surface Texture of the Sample . . . . . . . . . . . . . . . . . . 333 Paweł Zmarzły and Stanisław Adamczak Industry 4.0 Applications Approach of Medıum-Sized Industry Enterprıses to Industry 4.0 a Research in Konya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 Leyla Polat and Güzide Karakuş Cultural Aspects in the Adoption of ERP Systems: A Holistic View . . . 355 Gülay Ekren, Tuğba Koç, Alptekin Erkollar, and Birgit Oberer Deﬁning the Pros and Cons of Cloud ERP Systems: A Turkish Case . . . 367 Tuğba Koç, Gülay Ekren, Birgit Oberer, and Alptekin Erkollar Design of Car Seat Mechanism for Disabled . . . . . . . . . . . . . . . . . . . . . 379 Michael Pasteka and Marian Králik Developments in Biomedical Techniques Through Digital Transformation of Industry 4.0 – Applications in Tissue and Dental Implants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 Gökcen Bas, Wolfgang Pirker, Lukas Kräuter, Erol Güclü, and Lara Durakbasa Digitized Production – Its Potentials and Hazards . . . . . . . . . . . . . . . . . 402 Petr Blecha, Numan Durakbasa, and Michal Holub Evaluation of Industry 4.0 Readiness Level: Cases from Turkey . . . . . . 412 Gül T. Temur, Hür Bersam Bolat, and Sıtkı Gözlü Industry 4.0 Scorecard of Turkish SMEs . . . . . . . . . . . . . . . . . . . . . . . . 426 Zeynep Gergin, Fadime Üney-Yüksektepe, M. Güneş Gençyılmaz, A. Tülin Aktin, Kemal Güven Gülen, Doğan Aybars İlhan, Uğurcan Dündar, Özay Cebeci, and Ali İhsan Çavdarlı Measurement Technology & Quality & Justicia in Industry 4.0 . . . . . . 438 Andreas Bauer, Gökcen Bas, Numan M. Durakbasa, Lukas Kräuter, and Gamze Ugur–Tuncer Performance Analysis of Vehicle-Speciﬁc Methods and Sensors for Autonomous Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451 Ernst Pucher, Andreas Gruber, Mathias Innerkofler, and Marco Buhmann Smart Factories: A Review of Situation, and Recommendations to Accelerate the Evolution Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 K. Öncü Şen, M. Numan Durakbasa, Merve Vildan Baysal, Gizem Şen, and Gökçen Baş

Contents

xxi

Technology Selection for Digital Transformation: A Mixed Decision Making Model of AHP and QFD . . . . . . . . . . . . . . . . . . . . . . . 480 Hasan Erbay and Nihan Yıldırım The Innovation Performance Under the Shadow of Industry 4.0. . . . . . 494 Banu Ozkeser and Cüneyt Karaarslan The Role of IoT on Production of Services: A Research on Aviation Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503 Güzide Karakuş, Emre Karşıgil, and Leyla Polat Lean Production An Application of Kaizen in a Large-Scale Business . . . . . . . . . . . . . . . 515 Mahmut Tekin, Murat Arslandere, Mehmet Etlioğlu, and Ertuğrul Tekin An Application of SMED and Jidoka in Lean Production . . . . . . . . . . . 530 Mahmut Tekin, Murat Arslandere, Mehmet Etlioğlu, Özdal Koyuncuoğlu, and Ertuğrul Tekin Lean Manufacturing Implementations for Process Improvement in a Company Operating in FMCG Sector . . . . . . . . . . . . . . . . . . . . . . 546 Oğuz Emir, Samet Karataş, Eren Ay, Hümra Özker, and Zeynep Gergin Miscellaneous Topics A Model Suggestion for Entrepreneurial and Innovative University-Industry Cooperation in Industry 4.0 Context in Turkey . . . 565 Özdal Koyuncuoğlu and Mahmut Tekin An Investigation on Online Purchasing Preferences of Internet Consumers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581 Emel Celep, Ebru Özer Topaloğlu, and H. Serdar Yalçınkaya Environmental Risk Assessment of E-waste in Reverse Logistics Systems Using MCDM Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 590 Ferhat Duran and İlke Bereketli Zafeirakopoulos Grey Forecasting Model for CO2 Emissions of Developed Countries . . . 604 Asiye Özge Dengiz, Kumru Didem Atalay, and Orhan Dengiz The Examination of Complaints About the Health Sector by Text Mining Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612 Gamze Yildiz Erduran and Fatma Lorcu Operations Research Applications A Study About Affecting Factors of Development of E-commerce . . . . . 625 Mahmut Tekin, Haydar İnce, Mehmet Etlioğlu, Özdal Koyuncuoğlu, and Ertuğrul Tekin

xxii

Contents

Process Management Analyzing the Delivery Process with TOC . . . . . . . . . . . . . . . . . . . . . . . 645 Fatma Serab Onursal, Semra Birgün, and Ercan Mızrak Monitoring of Machining in the Context of Industry 4.0 – Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660 Wojciech Zębala, Grzegorz Struzikiewicz, and Emilia Franczyk The Potential Effect of Industry 4.0 on the Literature About Business Processes: A Comparative Before-and-After Evaluation Based on Scientometrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674 Güzin Özdağoğlu, Onur Özveri, Aşkın Özdağoğlu, and Muhammet Damar Production Planning and Control Production-Integrated Metrology with Modern Coordinate Measuring Machines Using Multisensor and X-Ray Computed Tomography Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693 Joachim Stopp, Raoul Christoph, and Ralf Christoph Productivity and Performance Management A Research on Financial Performance Analysis of Informatics Companies in the Scope of Industry 4.0 . . . . . . . . . . . . . . . . . . . . . . . . . 705 V. Özlem Akgün and Ali Akgün Analysis of the Relationship Between Enterprise Resource Planning Implementation and Firm Performance: Evidence from Turkish SMEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724 Ahmet Safa Kocaaga, Beyzanur Cayir Ervural, Omer F. Demirel, and Selim Zaim Investigation of System Productivity with Fuzzy Availability Analysis Considering Failure and Repair Times . . . . . . . . . . . . . . . . . . 737 Berna Dengiz, Merve Uzuner Sahin, and Kumru Didem Atalay Quality Management Problems of Mathematical Modelling of Rotary Elements . . . . . . . . . . . 747 Stanisław Adamczak, Dariusz Janecki, and Krzysztof Stępień The Effect of Service Quality and Offered Values on Customer Satisfaction and Customer Loyalty: An Implementation on Jewelry Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753 Meltem Diktaş and Mahmut Tekin

Contents

xxiii

With the Trio of Standards Now Complete, What Does the Future Hold for Integrated Management Systems? . . . . . . . . . . . . . . . . . . . . . . 769 Mihai Dragomir, Călin Neamțu, Sorin Popescu, Daniela Popescu, and Diana Dragomir Risk Analysis and Management A Bayesian Network Analysis for Occupational Accidents of Mining Sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 781 Fatma Yaşlı and Bersam Bolat Evaluation of Spatial Risks of Nursing Homes by Fuzzy Risk Analysis Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 800 Semra Boran, Didem Yılmaz, Zerrin Funda Ürük, and Seda Hatice Gökler Simulation and Modelling Optimisation of Machining in the Context of Industry 4.0 – Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 813 Andrzej Matras and Wojciech Zębala Capstone Projects A Cargo Vehicle Packing Approach with Delivery Route Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827 Uğur Eliiyi, Mert Bulan, and Emre Külahlı A New Demand Forecasting Framework Based on Reported Customer Forecasts and Historical Data . . . . . . . . . . . . . . . . . . . . . . . . 839 İlker Mutlu, Doğaç Sancar, Ege Naz Altın, Semih Balaban, Turan Can Cesur, and Önder Bulut An Application of Permutation Flowshop Scheduling Problem in Quality Control Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 849 Göksu Erseven, Gizem Akgün, Aslıhan Karakaş, Gözde Yarıkcan, Özgün Yücel, and Adalet Öner Daily Production Planning Problem of an International Energy Management Company . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 861 Elif Ercan, Pınar Yunusoğlu, Nilay Yapıcı, Sel Ozcan, and Deniz Türsel Eliiyi Design of a Decision Support System (DSS) for Housekeeping Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 872 Esin Acar, G. Şeyma Demir, Talya Temizçeri, and Levent Kandiller

xxiv

Contents

Efﬁciency Analysis in Retail Sector: Implementation of Data Envelopment Analysis in a Local Supermarket Chain . . . . . . . . . . . . . . 884 Ceren Kahraman, İrem Uluğ, Can Burak Othan, Yeşim Deniz Özkan-Özen, and Yiğit Kazançoğlu Model Sequencing and Changeover Time Reduction in Mixed Model Assembly Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 898 Derya Tataroğlu, Renan Dalkıran, Sueda Sezen, Damla Kesikburun, and Yiğit Kazançoğlu Repair Cost Minimization Problem for Containers: A Case Study . . . . 914 Merve Çamlıca, Gülce Çini, Ayşegül Eda Özen, Nilay Çınar, Sel Ozcan, and Deniz Türsel Eliiyi Routing Optimization for Container Dispatching Operations . . . . . . . . . 922 Hasibe Serap Baş, Ayşe Tolan, Mahmut Ali Gökçe, and Cansu Yurtseven The Distributor’s Pallet Loading Problem: A Case Study . . . . . . . . . . . 937 Selen Burçak Akkaya, Aykut Gül, Zeynep Coşkun, Coşku Karaman, Hande Öztop, and Gizem Mullaoğlu Three Dimensional Cutting Stock Problem in Mattress Production: A Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 949 Selin Altın, Tezcan Aydilek, Umut Şirvan, Damla Kesikburun, Adalet Öner, and Nejat Kutup Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 961

Artificial Intelligent Applications

Multi-objective Optimization Study in Face Milling of Steel János Kundrák1, Angelos P. Markopoulos2(&), Tamás Makkai1, Nikolaos E. Karkalos2, and Antal Nagy1 1

2

Institute of Manufacturing Science, University of Miskolc, Miskolc-Egyetemváros, Miskolc 3515, Hungary {kundrak,tamas.makkai,antal.nagy}@uni-miskolc.hu School of Mechanical Engineering, Section of Manufacturing Technology, National Technical University of Athens, Heroon Politechniou 9, 15780 Athens, Greece {amark,nkark}@mail.ntua.gr

Abstract. High productivity of parts manufacturing in industrial practice is closely related, not only to time efﬁciency, but also to the production of parts with high surface quality and considerable lifespan. Face milling is widely used for the efﬁcient creation of accurately flat surfaces, for a large variety of part sizes and materials. However, determining the process parameters, which can lead to the achievement of all required conditions can be considered as a multiobjective problem. This problem can be sufﬁciently solved using suitable optimization techniques. In the present work, it is attempted to determine the optimum parameters for face milling of steel parts, in order to achieve minimum cutting forces and surface roughness, as well as maximum possible material removal rate. For that reason, after regression models are derived to correlate process parameters with cutting forces and surface roughness, an optimization process is carried out with two different optimization methods, namely Genetic Algorithm and Fireworks Algorithm and after the determination of the optimum process parameters, results concerning the efﬁciency of optimization algorithms are discussed as well. Keywords: Face milling Design of experiments Genetic algorithm Fireworks algorithm

Optimization

1 Introduction Machining processes are essential in metalworking industries, as they can accurately and efﬁciently produce a large variety of part features. Especially, in the automotive and aerospace industries, turning and milling are widely used for the fabrication of a considerable part of the required components. When it is required to create high quality flat surfaces, face milling process is usually selected due to its simplicity and ability to achieve high material removal rates. For the study of machining processes, it is required to derive the correlation between process parameters and the outcome of the process, e.g. cutting forces, temperatures, surface quality and residual stresses, among others. In the relevant literature © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 3–15, 2019. https://doi.org/10.1007/978-3-319-92267-6_1

4

J. Kundrák et al.

of face milling, there exist many studies on cutting forces, surface quality and workpiece integrity [1–4]. Regarding surface quality, Mikó [5] developed a surface roughness predictive model for milling of a complex free-form surface using statistical methods. Mikó and Farkas [6] presented a study where both flatness and surface roughness were evaluated after face turning and face milling of various workpieces, with a view to determine strategies for the control of these deviations. An interesting approach concerning the theoretical prediction of surface roughness parameters was presented by Kundrák and Felhő [7], who used a CAD software to model the machined surface for face milling cases with square and dodecagonal inserts. In a later work, Felhő and Kundrák [8] developed also a theoretical model for the prediction of surface roughness in the case of face milling with octagonal and circular inserts. Finally, Mikó, Tóth and Varga [9] conducted an analysis regarding surface roughness for the case of ball end milling under various process conditions. Apart from cutting forces and surface quality, it is considerably important to obtain high material removal rates in industrial practice. However, as the goals of obtaining high surface quality, low power consumption, low tool wear and workpiece damage but also high material removal rate are contradictory, it is required to conduct an optimization process for the determination of suitable process conditions, which will enable the achievement of each goal at a sufﬁcient level. Wang, Liu and Wang [10] conducted a multi-objective optimization for turning using NSGA-II algorithm, emphasizing on energy consumption, machining cost and surface quality. Their approach was able to perform sufﬁciently for energy and cost objectives but improvement on surface quality was limited. Yan and Li [11] conducted multi-objective optimization for milling using grey relational analysis (GRA) and response surface methodology (RSM), focusing on surface quality, production rate and energy consumption. Anand et al. [12] also performed optimization on turning regarding energy consumption during machining of various metals such as steel, aluminum and brass. Sharma and Pandey [13] derived the optimum parameters regarding residual stress minimization during ultrasonic turning process by RSM. Moreover, Fu, Zhao and Liu [14] were able to determine the optimum cutting parameters in high-speed milling with a model combining GRA with principal component analysis. In the present work, a multi-objective optimization approach is presented, with the aim of obtaining process parameters, which will maximize material removal rate during face milling, under the constraints of maintaining lowest possible cutting forces and surface roughness. For that reason, the three components of cutting forces, two surface roughness indicators and material removal rate are included in the objective function. The methodology presented comprises of an efﬁcient DOE method which reduces considerably the experimental work, the derivation of regression models correlating input and output quantities of the face milling process and ﬁnally the optimization process using Genetic Algorithm (GA) and Fireworks Algorithm (FA) techniques. After the results are presented, discussion on the efﬁciency of the proposed method is conducted and useful conclusions are drawn.

Multi-objective Optimization Study in Face Milling of Steel

5

2 Methodology In this work, results from face milling experiments on steel workpieces with various process conditions are employed in order to determine optimum process parameters according to the desired goals. In the flowchart of Fig. 1, the steps followed in the present work are presented in detail. The desired goal for this work is the maximization of material removal rate under the constraint of maintaining cutting forces and surface roughness as low as possible. More speciﬁcally, the requirement for low cutting forces leads to minimum tool wear and power consumption of the machine tool, the requirement for minimum surface roughness leads to better surface quality and maximization of material removal rate leads to increase of production rate.

Fig. 1. Flowchart of the methodology followed for the determination of optimum process parameters

The 15 experiments were designed according to Box-Behnken method for 3 factors at 3 levels each, as can be seen in Table 1. For these experiments, a milling head with a single cutting insert was employed in a Perfect Jet MCV-M8 vertical machining center. The milling head diameter was 80 mm, the width of cut 58 mm and the workpiece material was normalized C45 (1.0503) carbon steel. Typical chemical composition for C45 is: 0.45% C, ð

T X

tqnt Þ dn

8n

ð10Þ

tqnt Þ 8n

ð11Þ

t¼1

Earln >dn ð

T X t¼1

X

2 Xnt 6 Capkt 5 n2N1 X

2 Xnt 6 Capkt 3 n2N4 X

2 Xnt 6 Capkt 3 n2N5 [ N6

8t; n 2 N1 ; k ¼ 1

ð12Þ

8t; n 2 N4 ; k ¼ 3

ð13Þ

8t; n 2 N5 [ N6 ; k ¼ 3

Xnt 2 Z þ [ f0g

8n; 8t

ð14Þ ð15Þ

866

E. Ercan et al.

Tardn ; Earln 2 Z þ [ f0g Ynt ; qnt 2 f0; 1g

8n

8n; 8t

ð16Þ ð17Þ

The objective function (2) minimizes total weighted tardiness and earliness (total lateness), where a denotes the tardiness coefﬁcient. Constraint set (3) ensures that the total order quantity produced on each line cannot exceed the capacity of that line on that day. Constraint set (4) controls whether the production is done for each customer order that day. Constraint set (5) guarantees demand satisfaction for each customer order. Constraint sets (6–9) control whether the order quantity is split or not and provide the completion date of each customer order when the production quantity is equal to the order quantity. Constraint set (10) calculates the tardiness of the order n which is the difference between completion date of the order and its due date. If the customer order is produced until the due date, tardiness of this order is equal to zero. Constraint set (11) calculates the earliness of the order n that is the difference between the due date of the order and its completion date. Constraint sets (12–14) are problemspeciﬁc constraints related with the capacity of two production lines. Constraint set (12) guarantees that daily production amount of a product group with a speciﬁc MRPC controller does not exceed the 40% of the production capacity of the corresponding production line. Constraint sets (13) and (14) ensure that the maximum production amount of two product groups should be 2/3 of the production capacity of the corresponding production line, respectively. Constraint sets (15), (16) and (17) deﬁne the decision variables. In the second stage, the mathematical model with the objective of minimizing the total number of orders split is solved. The model deﬁned in (3)–(17) is solved where the objective function is deﬁned (18) and, an additional constraint (19) is introduced. Min

N X T X

Ynt

ð18Þ

n¼1 t¼1 N X n¼1

aTardn þ

N X

ð1 aÞEarln 6L

ð19Þ

n¼1

The objective function (18) minimizes a total number of orders split, as the company prefers. In constraint (19), we limit the total weighted tardiness and earliness cannot exceed the optimal objective function value of the model deﬁned in (1)–(18). With constraint set (19), we guarantee the minimum lateness, denoted as L is maintained in the second problem, as well.

4 Heuristic Method Bitran and Yanasse show that the CLSP is NP-hard even without setup times; no approach is provided for gathering the optimality [3]. Since this problem is NP-hard, we provide an alternative solution approach, embedded into a user-friendly decision

Daily Production Planning Problem

867

support system developed in Excel VBA, so as to obtain fast and effective solutions for large problem sizes. We propose a heuristic method by combining the company’s expectations and heuristic method for CLSP that is proposed by Karimi [4]. The decision support system (DSS) ﬁrst starts with the elimination of the non-value-added activities within production planning process. Our DSS provides a near optimal production plan by minimizing the total lateness of customer orders and by using the capacity of each line in the most efﬁcient way. Flow chart of the heuristic method is given in Fig. 1.

Fig. 1. Flowchart of heuristic method

The inputs of the problem are partial production plan, new customer orders which are taken from SAP and daily production capacities of production lines. In the heuristic method, ﬁrstly the related data is obtained from the spreadsheets and validity of the input data is checked. Then, current and remaining capacity is calculated for each line. New orders are sorted according to the EDD (Earliest Due Date) rule and each order’s related production line is controlled. If the remaining capacity of the desired date is greater than or equal to the quantity of the order, the order is planned to the related production line. If there is not enough amount of remaining capacity, then the remaining capacity of line is controlled for one day before the desired day and if it ﬁts,

868

E. Ercan et al.

the order is planned to that day. This procedure repeats for ﬁve days before the desired day. After the ﬁfth day, if the remaining capacity of related line is not enough, the order cannot be planned and in order to inform the user, the system provides a report. Then, the capacity utilization is checked for three (3) consecutive days. If the capacity utilization is 100%, the plan is completed for that day. If it is not, then the customer orders that are proper quantity and produced for stock are shifted to that day in order to use the remaining capacity. The heuristic method was coded in Excel VBA to reduce the wasted time of current production planning process. A dynamic production planning interface was created so that the company can use for daily production planning activities.

5 Computational Study The proposed mathematical model is solved using IBM ILOG CPLEX Optimization Studio 12.6.3 on a computer with i7 processor and having 8 GB RAM. As a pilot study, 3 real-life instances of the problem are examined for validation and veriﬁcation of the proposed mathematical model. Properties of each problem instances and their solution duration and optimality gap are given in Table 1. Table 1. Properties and solution durations of the test problem instances Instance # 1 2 3

Number of orders 52 144 700

Number of product types 8 8 9

Number of production lines 6 6 6

CPU (min) 2.04 5.43 360

Optimality gap (%) 0 0 51.68

Note that, the solution times of relatively small instances e.g. instances 1 and 2 are less than a minute. However, as the number of orders increases, the solution time increases, concurrently. For example, the optimal solution cannot be obtained for instance 3 within 6 h and the optimality gap is around 52%. A sensitivity analysis is conducted to test whether the optimal solution is affected for several a levels. The optimal solution of each test problem, as well as their computational time performances, are reported in Table 2. We can conclude that a does not play an important role in the optimal solution. Table 2. Sensitivity analysis for various a levels a 0.2 0.5 0.7 0.9 0.995

Total earliness (in days) 3 3 3 3 3

Total tardiness (in days) 10 10 10 10 10

Number of early orders 1 1 1 1 1

Number of tardy orders 4 4 4 4 4

Daily Production Planning Problem

869

For three real life instances that are given in Table 1, we compare the computational time performances and solution qualities of the preemptive goal programming model and the heuristic approach in Table 3. In the existing system, the duration of planning 30 (thirty) customer orders takes 90 (ninety) minutes. Table 3. Comparison of the methodologies Instance #

# of Methodology orders

1

52

2

144

3

700

Heuristic method Preemptive goal Programming model Heuristic method Preemptive goal Programming model Heuristic method Preemptive goal Programming model

Total Total Number of Number of earliness tardiness early orders tardy orders (in days) (in days) 32 9 7 2 0 1 0 1

68 3

18 10

138 47 Out of memory

CPU (min) 0.92 2.04

23 1

8 4

2.31 5.43

82

30

7.2 360

For all instances, both the preemptive goal programming model and the heuristic method yield better solutions than the current system. In the current system, for instance, 1, total earliness is 127 days, tardiness is 14 days and a total number of early and tardy orders are 45 and 3 respectively. For instance 2, total earliness is 325 days, tardiness is 34 days, number of early orders are 104 and number of tardy orders are 15. There are 549 early and 90 tardy orders in the current system for instance 3 and total earliness and tardiness are 1463 days and 283 days respectively. Thus, the tardiness and earliness in customer orders are reduced with the use of both methodologies and ontime delivery is improved accordingly. We can conclude that the preemptive goal programming model deﬁnitely outperforms the heuristic method, however, it is computationally expensive. We report that, for instances 1 and 2, the number of early orders decreased by more than 90% with the preemptive goal programming and number of tardy orders reduced by nearly 70%. However, the preemptive goal programming model does not perform well in terms of computational time. For instance 3, the optimal solution cannot be obtained by the mathematical model, but heuristic method provides a solution for this instance in 7.2 min. The increase in the solution quality is relatively lower than the preemptive goal programming when the heuristic method is applied. The early and tardy orders are decreased by 80% and 35%, respectively once it is compared with the existing system. Although a clear dominance of the preemptive goal programming model over the heuristic method is observed in terms of solution quality, the daily production planning

870

E. Ercan et al.

duration is decreased by 94% with the heuristic method. This implies that the heuristic method provides effective solutions within reasonable computational time.

6 Conclusion In this study, the real-life daily production and capacity planning problem of a company is studied. In this perspective, the production lot size and production date of each customer order are determined according to due date and the quantity of the customer orders without exceeding the capacity of the production lines. Once the problemspeciﬁc constraints are employed, the problem differs from the well-known capacitated lot sizing problem. The problem is formulated as a preemptive goal programming model and solved in two stages. Moreover, a simple heuristic approach is developed to obtain high quality solutions within reasonable time. A user-friendly decision support system is developed in Microsoft Excel Visual Basic for Application. A heuristic approach is also employed into the decision support system, where the inputs of the problem are taken from the company’s ERP system and the daily production plans for each production line are constructed thereby. A computational study is carried out with the real-life instances gathered from the company’s past data. The computational study indicate that both the preemptive goal programming model and the heuristic method yield better production plans than the existing system. However, preemptive goal programming model deﬁnitely dominates the heuristic method in terms of solution quality, whereas relatively good solutions are achieved with the heuristic method in very short computation time. For small-sized instances, the optimal solution is found very quickly by the preemptive goal programming model. However, for larger instances, the run times for optimality seem to be unacceptable. Hence, heuristic method is preferred as it provides fast and effective solutions even for large instances. We believe that our decision support system can handle the company’s basic needs on planning and achieve drastic time savings for the planner as well as reduces user errors. As a future research agenda, our aim is to propose more sophisticated heuristic approaches where the quantities of make-to-stock orders are also determined with the consideration of inventory holding costs. Acknowledgment. We are grateful to the company for sharing their data with us to complete this work. This work cannot be completed without the assistance of Asst. Prof. Dr. Adalet Öner, Asst. Prof. Dr. Canan Pehlivan, Research Assistant Sinem Özkan, and students Alper Uyar, Ece Başar, Fatih Akamca and Irem Amaç. We are thankful for the contribution of them. We also thank The Scientiﬁc and Technological Research Council of Turkey (TUBITAK) for funding this study within 2209B-National/International Research Projects Fellowship Programme for Undergraduate Students.

Daily Production Planning Problem

871

References 1. Wang YM, Parkan C (2007) A preemptive goal programming method for aggregating OWA operator weights in group decision making. Inf Sci 177(2007):1867–1877 2. Sawik T (2003) Integer programming approach to production scheduling for make-to-order manufacturing. Math Comput Model 41(2005):99–118 3. Bitran GR, Yanasse HH (1982) Computational complexity of the capacitated lot size problem. Manag Sci 28(10):1174–1186 4. Karimi B, Fatemi Ghomi SMT, Wilson JM (2003) The capacitated lot sizing problem a review of models and algorithms. Omega 31(2003):365–378

Design of a Decision Support System (DSS) for Housekeeping Operations Esin Acar, G. Şeyma Demir, Talya Temizçeri, and Levent Kandiller(&) Department of Industrial Engineering, Yasar University, İzmir, Turkey [email protected], [email protected], {talya.temizceri,levent.kandiller}@yasar.edu.tr

Abstract. This paper conducts our senior project at Altın Yunus Hotel located in Çeşme, Turkey and it contributes to the improvement of operations management of the hotel, focusing on the front ofﬁce, reception and housekeeping services. The problem is determined as the housekeeping problem in consideration of other problems described in detail in the following sections. Required data are provided by the hotel management. The aim of the study is to develop a decision support system (DSS) covering and increasing the efﬁciency in service quality. Through the literature survey, system analysis and the developed mathematical models such as regression, time study, worker assignment, uniform parallel machine scheduling and routing optimization are presented. Finally, achieved time and cost savings are presented. Keywords: Regression Time study Worker assignment Uniform parallel machine scheduling Decision support system

1 Introduction In any hospitality establishment there are three departments particularly concerned withaccommodation: the reception department, whose staff sell and allocate the rooms; thehousekeeping department, whose staff plan, provide and service the rooms; themaintenance department, whose staff provide adequate hot and cold water, sanitation, heating, lighting and ventilation as well as maintaining and repairing individual articlesand area within the rooms operation. This study is among the ﬁrst studies in which the techniques of production research applied to the above operations at hospitality sector. The main motivation of this study is our contracted research at Altın Yunus Hotel which is a subsidiary business of Yaşar Holding and one of the largest facilities of Turkey, located on an area of 140.000 m2 at the Kalem Burnu in the Boyalık District of Çeşme, Izmir, Turkey. It was founded in 1972 and the construction was completed in 18 months and became operative in the 1974 summer season. Altın Yunus is the very ﬁrst “ﬁrst class holiday village in Turkey” which could accommodate a thousand people in 465 rooms. At the beginning of our study, a number of symptoms are observed and the problems are identiﬁed with the use of related data. One of the most important problems encountered in the hotel is the real-time scenarios created due to system dynamically changing demands. For example, the difﬁculty of managing the © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 872–883, 2019. https://doi.org/10.1007/978-3-319-92267-6_70

Design of a DSS for Housekeeping Operations

873

process of emptying rooms and accommodating customers in the cleaned rooms during the day between 12:00 and 14:00. During this hot period, while on one hand attention is paid to customer satisfaction, the teams need to be used efﬁciently on the other hand. In order to facilitate the solution of similar problems, a study has been conducted in the floor services unit in the hotel reception department. At the system analysis, a demand forecasting model is developed as well as decision models formulated at both the tactical level and the operational level. Our ﬁrst planning model performs annual personnel levelling. Our second planning model facilitates a daily workforce planning and scheduling including the hot time period mentioned above. The third tool provides a navigation and routing service for both housekeeping and technical services after solving TSP with network distances. Solution methods were developed for these three models and veriﬁed by small size problems and the coding of these solution algorithms is completed. An MS Excel-based DSS developed is available for the decision makers. The remainder of this paper is organized as follows. In Sect. 2, the literature review of studies is presented. In Sect. 3, the problem deﬁnition is stated. The problem formulation with observations and input analysis is presented in Sect. 4. In Sect. 5, veriﬁcation, validation and sensitivity analysis are reported. Computational results, decision support system and output analysis are proposed in the same section. Finally, the conclusions and recommendations for future research are presented in Sect. 6.

2 Literature Review A standard time is a time required for a work to be deﬁned under speciﬁed conditions [1]. A time study can also be deﬁned as a work measurement technique used to record the time and extent of a particular work item under certain conditions and to determine the time required for that work to be performed at a deﬁned work rate (performance) by analysing the collected data. Time standards are used in planning future work and in evaluating past work. The time study also requires the use of concentration techniques such as performance grading so that the working speed can be determined, and the working speed can be correlated with the standard working tempo. When working on the standard work schedule and using the appropriate rest periods, a worker will have reached the standard performance level during workday or shift. The standard time for a job, observing the repetition frequency of components causes all time to occur [2]. ‘Regression Analysis’, ‘Time Study’, ‘Travelling Salesman Problem’ ‘Parallel Machine Scheduling Problem’ and “Scheduling Method” topics are discussed briefly. The main purpose of the regression analysis, a statistical technique used to relate variables, is to create a mathematical model to relate dependent variables to independent variables [3]. In general, a regression model form deﬁnes a single algebraic equation [4]. The multiple linear regression models are used for stating the relation between two or more explicative variables and the response variable by identifying a linear equation between the observed dates. For each value of the independent variable x, it is associated a value of dependent variable y. The individual values of the registered explanatory variables within the linear regression X1 ; X2 ; . . .; Xn is deﬁned as: ly ¼ b0 þ b1 X1 þ b2 X2 þ . . . þ bp Xp .

874

E. Acar et al.

The parallel machine scheduling problem is one of the important and difﬁcult problems in literature. This problem consists of the scheduling of a set of independent jobs on identical (parallel) machines (processors) with the aim of minimizing maximum job completion time. The rooms to be cleaned by any housekeeper are the jobs and the housekeepers are the machines. Moreover, the goal is to schedule identiﬁed jobs and minimize the maximum completion time [5]. The Traveling Salesman Problem (TSP) historically is the core optimization problem studied extensively in the literature. The TSP problem deals with ﬁnding the shortest (closed) tour in an n-city situation where each city is visited exactly once [6].

3 Problem Deﬁnition In the current system, the service flow starts at the front desk (reception) when the registration operations are carried out on the system. Transaction entries by the front desk are automatically fed floor services and the check-in operations of customers are initiated. While these operations are taking place, manual preparation of these services causes inability to meet the demand and it leads to some problems. These observed symptoms are unsteady working methods in the housekeeping process and delays in the workflow as a result of those methods. One of the most important reasons for the disruptions in these processes is the difﬁculty in planning the check-in check-out and cleaning services during hot times (12:00 to 14:00). One of the difﬁculties observed is that the customer does not comply with the check-out time and continues to stay in the room when he/she is required to leave the room. In the face of such events, floor services get affected by this disruption and cannot complete room cleaning, and the people responsible for meeting customers (receptionist, bellboy etc.) cannot assign the rooms to the customers. This situation puts both the employees and the company in a difﬁcult situation against the customers and the mistakes made by the employees in the operation ﬁeld interrupt the following processes. In the direction of these symptoms and observations, we can say that there is no systematic arrangement for job assignments in the company, which causes some disruption in the ﬁeld of operations management. The company needs a decision support system to overcome these problems and to increase productivity. Along the direction requested by the ﬁrm, we are focusing the housekeeping department which the main problems are observed. The housekeeping department is responsible for cleaning all areas, especially the customer rooms. The main issues observed in this department are the enrollment of different employee types according to the seasonal intensity, the identiﬁcation of the employees who need to be ready for any job, the assignment of the employees in the floor services for the cleaning of the rooms to be prepared and the arrangement of the rooms according to a certain rotation system. As another study, the performance ratings are determined considering the experience of the employees (experienced, inexperienced, intern) and a time study is conducted to collect the preliminary data. After these observations, cleaning processes are recorded and the information about how long each employee cleans an area is obtained according to the time analysis.

Design of a DSS for Housekeeping Operations

875

According to these observations, reducing the cost, reducing the total time and reducing the total delay (which are our key performance indicators) are the key performance indicators.

4 Problem Formulation In this section, the four models mentioned above are described in detail together with the solution procedures. 4.1

Demand Forecasting

The timing model used in our ﬁrst model determines the number of floor personnel, whose estimated capacity will be equivalent. Observing the number of people coming to Altın Yunus and the number of the rooms for the last 4 years, regression analysis is conducted to estimate the year 2018. First, analyzing past years’ data, it is noted that the hotel is a seasonal hotel and more customers come on weekends. Thus, a week is divided into two half weeks as weekdays and weekend. It is determined as a half week from Monday to Thursday and a half week from Friday to Sunday, so one year is determined as the independent variable as 104 half-weeks. Our other independent variables are temperature averages, special days, terrorist incidents and their combinations. R2adj was found as 95%, thus the regression model is good to use. Due to hotel privacy, the speciﬁc regression equation is concealed. 4.2

Workforce Planning

This model makes cost-effective long-term team planning. The formulation reduces the total labor cost while meeting the demand for services. This is a simple aggregate service planning (team scheduling) model [6]. The teams that are ready for a certain planning period (half week) in the basic constraint are on the left side of the inequality (full-time employees, interns, part-time employees). On the right side of the equation, there are the requirement values determined by the regression model. As a solution method, CPLEX general solver has been used. Due to the fact that the results of the model are found meaningful by Altın Yunus Executives, proving the validity (Table 1). The long-term planning formulation is given below: Xn min ðCF Fw þ CP Pw þ CI Iw Þ ð1Þ w¼1 s.t. qF

Xw t¼wdF þ 1mod ðnÞ

Xw

I t¼wdI þ 1mod ðnÞ t

Fw þ qI

Xw

I t¼wdI þ 1mod ðnÞ w

uI ; P w uP ;

IW ; PW ; FW 0

þ qP Pw Rw

Xw t¼wdF þ 1mod ðnÞ

and

INT

Ft uF

w ¼ 1; . . .:; n

w ¼ 1; . . .:; n ð2Þ w ¼ 1; . . .:; n

ð3Þ ð4Þ

876

E. Acar et al. Table 1. Notation for workforce planning model

The objective function (1) minimizes the total annual cost. The main constraint states that the maximum number of rooms that the existing workers will clean must be more than the number of rooms that are needed to be cleaned during hot times. Constraint set (3) imposes the bounds set by the decision maker. Constraint (4) is the sign restriction. 4.3

Employee Assignment

This model aims to designate rooms to be cleaned at that date to the employees in the most appropriate way (Table 2).

Table 2. Notation for employee assignment model

The employee assignment formulation with three alternative objective functions is given below: Min

XW w¼1

Zw

ð5Þ

Design of a DSS for Housekeeping Operations

877

or Min

XR l¼1

ð6Þ

Lr

or ð7Þ

Min L s.t. L LR XW w¼1

XR r¼1

XW

Ypqw Xpw

w¼1

ð8Þ

Xrw ¼ 1 r ¼ 1; . . .; R

Xrw MZw

RTr Sr Sr þ

r ¼ 1; . . .; R

w ¼ 1; . . .; W

r ¼ 1; . . .; R

Prw Xrw Lr DDr

r ¼ 1; . . .; R

8p; q ¼ 1; . . . ; R; p 6¼ q; w ¼ 1; . . . ; W

Ypqw Xqw

8p; q ¼ 1; . . .; R; p 6¼ q; w ¼ 1; . . .; W

Sp þ Ppw Xpw Sq þ M 1Ypqw

8p; q ¼ 1; . . .; R; p 6¼ q; w ¼ 1; . . .; W

Sr ; Lr 0 Xrw ¼ 0 or 1 Ypqw ¼ 0 or 1

r ¼ 1; . . .; R

r ¼ 1; . . .; R

w ¼ 1; . . .; W

8p; q ¼ 1; . . .; R; p 6¼ q; w ¼ 1; . . .; W

Zw ¼ 0 or 1 w ¼ 1; . . .; W

ð9Þ ð10Þ ð11Þ ð12Þ ð13Þ ð14Þ ð15Þ ð16Þ ð17Þ ð18Þ ð19Þ

The objective function (5) minimizes the number of workers used; whereas the objective function (6) minimizes total lateness and the objective function (7) minimizes the maximum lateness. Constraint (8) ensures that the maximum lateness is equal or greater than the lateness of each room. Constraint (9) states that every room should be processed. Constraint (10) marks the usage of worker w. Constraint (11) states that the starting time of a room is later than, its release time. Constraints (12) calculates the lateness. Constraint (13) and (14) relate with x variables to y variables. Constraint (15) ensures that no overlap of rooms by a worker. Constraint (16)–(19) are sign restrictions. Since the problem is NP-Hard, we develop a heuristic to obtain good solutions in a reasonable time. The algorithm given in Fig. 1 runs as follows: The rooms to be vacated at 12:00 are shown in Group A (ready) and the rooms to be vacated later than 12:00 are shown in Group B (not ready). When the time is beyond ready time, these

878

E. Acar et al.

rooms are moved to the list A. The rooms are assigned to Group A, according to the time of delivery, the nearest employee in terms of distance and the available times of the employees. When the assignment is realized, the dependent time for cleaning the rooms is also taken into consideration and time advances. Our goal here is to assign the employees the most appropriate way by reducing walking and keeping the waiting times to the minimum. A sample problem given in Table 3 is given for illustrative purposes. A sample solution is depicted in Fig. 2.

Fig. 1. Our heuristic algorithm Table 3. Cleaning time form for rooms

Fig. 2. Distribution of rotation to employees

Design of a DSS for Housekeeping Operations

4.4

879

Routing Aid

Both housekepping operations and technical maintenance service operations over a large hospitality facility involving more than ten stops needs to decision support for route optimization in terms of time and distance travelled. The underlying decision problem is good old TSP. The formulation of the problem is intensionally omştted since the TSP problem is very well studied in the literature for the last decades. The underlying problem is standard TSP with network distances. In our study, a network distance matrix of more than 500 locations is compiled. As a solution procedure, the nearest neighbor heuristic is used as construction and two-way interchange heuristic is used for greedy improvement. Since we employed standard algorithms, the formulation and algorithmic details are not included here for saving space.

5 Veriﬁcation Validation and Sensitivity Analysis The goal of the project is to develop a dynamic decision support system that solves individual problems quickly. The employee planning problem is solved through CPLEX 12.8 Asus on a Core i7,4720 HQ, 2.60 GHz, 16 GB RAM computer. The scheduling problem and the TSP is solved by means of heuristics coded in Netbeans IDE 8.2 and VBA, respectively. The developed models are run with small-sized problems and it is observed that CPLEX results are signiﬁcant. By this way, the veriﬁcation process is completed and the solutions to the real size problem are obtained in the ﬁrst step for validation. These solutions are shared with the authorized people in Altın Yunus and it is seen that the results are reasonable. After the establishment of DSS at Altın Yunus towards the end of the project, daily performance is measured individually. Comparing these performances with the results suggested by the model, the key performance indicators we determined in the problem deﬁnition section become measurable under some circumstances. Sensitivity analysis is performed by enumerating different values to model parameters. The ﬁrst model proposes a cost-effective long-term planning. Main restriction of our model is the number of employees. Planning model includes necessary values, the number of rooms needed to be cleaned during hot times at each half week which calculated from the regression equation. The aim is to plan one year of employee planning with 104 half weeks. With the help of CPLEX and Ms Excel VBA codes, the DSS developed in very short time. At opening page of the program shown in Figs. 3 and 4, intern, full-time employee, part-time employee’s cost and the total cost are presented. In Table 4, number of rooms that need to be cleaned in hot times is presented. In Table 5 intern, full time and part time working employees, how many halfweek works in one year and required time to clean one room in hot and cold times and it can be explained the left of the equation which type of employee will be work in the 104 half weeks and how many employees will work. For the success of the program, the regression analysis results should yield the adjusted R2 values at least 90%.

880

E. Acar et al.

Fig. 3. User form 1

Fig. 4. Workers calculated Table 4. Hot time rooms

Table 5. Number of workers per week

The worker assignment model aims to assign employees to the best way to rooms to be cleaned. The model makes the most appropriate assignment, using the workers in the most appropriate way and reducing the delay times of the rooms to minimum values. Walking distances have been added to the processing times, calculated through the time study, so that the workers are assigned to the nearest suitable room and the walking distances are reduced to minimum values. The assignment model is coded in Netbeans IDE 8.2. Our program ﬁrst selects rooms to be cleaned as shown in Fig. 5. Then, in Fig. 6, the user enters the ready time and due date of the room. In Fig. 7, the number of workers required by the user (the data obtained from the aggregate planning model) is entered. As a result, our program transfers the outputs to Excel, reporting which order the workers will clean the rooms and the delay times of the rooms as illustrated in Fig. 2.

Design of a DSS for Housekeeping Operations

Fig. 5. Entering # of rooms to be cleaned

881

Fig. 6. Entering RT & DD

Fig. 7. Entering workers into the system

Another DSS module is set up on the routing system of rooms to be cleaned. For the distance calculations made in this decision support system, an interface is created by means of VBA codes in Excel. When we turn on this DSS, a worksheet shown in Fig. 8 is opened. For this output, the following operations are performed in order; with the ‘‘Add Room’’ button, the room numbers to be cleaned are entered by the user, a small matrix containing the distances of the related rooms is obtained with the ‘‘Find Matrix’’ button and then with the button ‘Construct’, the room to be entered for the list is made feasible. In this way, cleaning staff and technicians will be able to reach the room with minimum route with minimal effort. By developing the code, the gains obtained by the binary interchange method are calculated for all possibilities. This improved algorithm has been added to the work screen with ‘Binary Interchange’ button. This myopicimprovement algorithm aims to overcome the disadvantages of the nearest neighbor algorithm, especially the long return path that it uses when completing the tour.

Fig. 8. TSP modelling output

882

E. Acar et al.

Instead of, manually based estimation and assignment processes based on personal experience, we created a systematic system through the decision support system. Workers were assigned with the most appropriate route, time is minimized, and the minimum delays are provided. At the same time, according to estimation, the occupancy rate and the number of workers to be needed is planned the closest to reality. It is seen that the costs are reduced when the number of employees employed in the hotel is compared with the number of employees providing from the DSS. Finally, the 2018 occupancy rate cannot be reached due to the hotel’s privacy policy, so the comparison of 2018 is not made. As we mentioned in the problem deﬁnition, there are three types of employees in the hotel, full time, part time and intern. The most costly of these employees are parttime employees. The hotel was keeping up part-time employees and it caused high costs. By the help of the decision support system, worker assignments are carried out in the most efﬁcient manner, resulting in a 52% cost saving. The costs obtained from the Altın Yunus’s manual assignments (left side) and the costs from the decision support system (right side) are given in Table 6. Table 6. Total costs and percent saving dI ¼ 8; dF ¼ 96; uI ¼ 10; uF ¼ 1; up ¼ 25; n ¼ 104Þ

6 Conclusion This study is among the ﬁrst studies in which the techniques of production research applied to the housekeeping operations at hospitality sector. The study is motivated from a real life system, Altınyunus Resort and Thermal located in Turkey. A regression model is developed for demand forecasting producing point estimates and prediction intervals given a risk value. Together with a motion and time study conducted, the regression model generates the data required for two planning models at the tactical and operational level. The ﬁrst model balances the workforce through the year whereas the second model solves worker assignments at daily basis. A third decision problem helps routing within the premises. In this project, a new decision support system has been created, which is going to be live to assist operational managers. Last years’ data are used for validating the system and we helped to reduce the annual costs considerably and improve customer satisfaction.

Design of a DSS for Housekeeping Operations

883

The traditionally turbulent relationship between the front ofﬁce and housekeeping department is too often the headline act and, as with any sibling rivalry, is based entirely on the quality of communication between the two. Understanding the bottlenecks is important and this has only become possible with the rise of the smart production research that allows for useful software integrations. Dynamic planning, realtime status updates and instant communication can lead to unprecedented cost management for a hotel, let alone reduce heated arguments. Housekeeping is, arguably, the most inefﬁcient operation in any hotel. Amidst a labyrinth of rooms with enigmatic guests slipping in and out, clipboards and printed reports are too often the only compass housekeepers have to navigate through their daily chores. According to a workflow calculation by productivity management experts, the daily work of a maid consists of 18 tasks. Yet it is estimated that housekeepers spend about 10 to 15% of their time just trying to ﬁnd the next room to clean. Furthermore, excellent housekeeping is all important for guest satisfaction. According to a survey commissioned by cleaning products brand CLR and conducted by TNS, 86% of hotel guests cited cleanliness as the top criteria they look for when reading online hotel or holiday rental reviews. The survey also revealed that eight out of ten guests would rather give up internet access for the duration of their than stay in a dirty hotel or rental. In short, this means you should be able to run your hotel with less resources and deliver a higher quality experience for guests. The future of hotel housekeeping lies in mobile apps run by a cental decision support system whose model base is pure production research. Acknowledgment. This paper presents the works of a university-industry collaboration project, under the framework of the capstone design courses (IE 495 Systems Analysis and IE 496 Systems Design) taught in the Industrial Engineering Department of Yasar University. The authors acknowledge the support provided by Necip Atılgan and the project team members; İlber Gündüz, Gamze Urak, Nilay Yıldız, Emre Narin and Merve Kıran as students of Yaşar University, as well as the support of the Department of Industrial Engineering and with Altın Yunus Resort & Thermal Hotel, during the academic year 2017/2018.

References 1. Groover MP (2007) Work systems and the methods, measurements and management of work, University of Lehigh, UK, Chapter 13, vol 352 2. Meyers FE, Stewart JR (2002) Motion and time study for lean manufacturing, 3rd edn. Prentice Hall, New Jersey 3. Anghelache C, SACALÃ C (2016) Multiple linear regression used to analyse the corelation between GDP and some variables. Rom Stat Rev Suppl Rom Stat Rev 64(9):94–99 4. Anghelache C, Manole A, Anghel MG (2015) Analysis of ﬁnal consumption and gross investment influence on GDP – multiple linear regression model. Theor Appl Econ 22 (3):137–142 5. Hiller FS, Lieberman GJ (2010) Introduction to operations research, University of Stanford, California, Chapter 8, vol 335 6. Taha HA (2007) Operations research: an introduction, 8th edn. Person Education, Harlow. Chapter 9

Efﬁciency Analysis in Retail Sector: Implementation of Data Envelopment Analysis in a Local Supermarket Chain Ceren Kahraman, İrem Uluğ, Can Burak Othan, Yeşim Deniz Özkan-Özen(&), and Yiğit Kazançoğlu International Logistics Management Department, Yaşar University, Izmir, Turkey [email protected], [email protected], [email protected], {yesim.ozen,yigit.kazancoglu}@yasar.edu.tr

Abstract. Increased competitiveness leads the importance of efﬁciency analysis for organizations. There are many tools and techniques for performance measurement. Data Envelopment Analysis is one of the well-known nonparametric, direct programming technique that can deal with multiple inputs and outputs and used for efﬁciency analysis. In this study, data envelopment analysis is used in retail sector. This study is based on a real life problem and the aim is to measure the store performance of a local retail chain in Turkey. While, size of the store, number of employees, number of deliveries and total cost are used as inputs; number of customers, sales, and store evaluation of customers are used as outputs in this study. Main contribution of this study is using store evaluation of customers as an output by conducting an interview with each stores’ customers. Due to the dynamic environment of the retail sector, model that allows variable return to scale while maximizing inputs; input-oriented BCC is used in the study. At the end of the study, numerical results are obtained from a software called Frontier Analyst, and managerial suggestions are presented in the conclusion part. Keywords: Data envelopment analysis BCC model Efﬁciency analysis Performance measurement

Retail sector

1 Introduction Developments in service sector, increase in production amounts, shrinking product life cycles, increase in accessibility of variety of products reveals the importance of gaining competitive advantage in the market place. Therefore, continuous performance monitoring is essential for organizations to be successful. Different tools and techniques are used for measuring performance of organizations including balanced Scorecard [1, 2], EFQM Excellence Model [3] and Key Performance Indicators [4]. As one of a well-known mathematical model based technique, data envelopment analysis (DEA) is also used for measuring performance. DEA is used for assessing productivity and efﬁciency of decision making units (DMU) and it is a nonparametric © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 884–897, 2019. https://doi.org/10.1007/978-3-319-92267-6_71

Efﬁciency Analysis in Retail Sector

885

linear programming method [5]. Implementation areas of DEA have a high variety including service and manufacturing sectors. Basically, DEA is used to estimate the level of efﬁciency of similar units of organizations, so called DMU, which utilize the same inputs to produce the same outputs [6]. In this study, DEA is used for the retail sector, which is not new, and in the literature there are many studies. Most of these studies use inputs related to employees, size of the store, economic factor; and outputs related to sales, number of customers etc. However, to the best of our knowledge, using store evaluation of customers as an output while assessing the DMUs has not been conducted. Only [7] used customer satisfaction as an input in residential building management assessment. Therefore, originality of this paper is using store evaluation of customers as an output, in addition to other well-known inputs and outputs for retail sector. This study is conducted based on a real life problem, where an assessment was needed for a close down decision for a store in the retail chain. The objective of this study is to measure performance of the seven stores in a local retail chain in Turkey and to guide the organization about the close down decision according to the numerical results. This study is divided into six main parts. After introduction, ﬁrstly literature review related to general DEA usage and retail sector is presented. Then, methodology, implementation of the study and numerical results are shown. Finally, discussions and conclusion are presented.

2 Literature Review Data Envelopment Analysis can be deﬁned as nonparametric direct programming strategy for evaluating the productivity and efﬁciency of Decision Making Unit’s (DMU) [8]. DEA is used in many sectors for measuring the performance efﬁciency including; healthcare, ﬁnance, production etc. To start with the healthcare sector [9–12] can be given as examples where DEA is used for measuring efﬁciencies. DEA methods varies according to need of implementation environment, even though the sector remains the same. For instance; InputOriented CCR Model was used in [10] on the other hand, Output-Oriented CCR Model was used with [11]. Moreover, Dynamic Network and Black Box Model was used by [9] in healthcare sector. Moreover, education is another sector that DEA is used to measure performance. [13] used Ordinary Least Squares Method and Stochastic Frontier Analysis Model to measure UK universities performance. Output-oriented BCC model was used by [13, 14] to measure efﬁciencies in the UK and Chinese universities respectively. In the studies of [15, 16] DEA was used in production sector. In Ref. [15] inputoriented CCR was used for vendor evaluation for a baby food manufacturer. On the other hand, [16] used output-oriented network slack based model to evaluate ten Iranian soft drink manufacturer in e view of green supply chain management. Finance is another sector that DEA is used frequently. Reference [17] conducted a survey, which focuses on 80 Published DEA applications in 24 different countries. In addition, Regression Model was integrated with DEA and used by [18] to evaluate cost, technical and allocative efﬁciencies for Brazilian Banks Between 2000–2007.

886

C. Kahraman et al.

Moreover, both CCR and BCC Models were used by [19] to measure relative efﬁciencies of banks. In Table 1, summary of the literature review is given by including sector, focused area and method. Since this study focuses on performance measurement in retail sector by using DEA, in the following section studies that used DEA in retail sector are investigated. Table 1. Summary of the literature review Author(s) Banker et al. [11]

Sector Healthcare

Weber [15]

Production

Jemric and Vujcic [20]

Finance

Johnes [21]

Education

Avkiran [22]

Financial Service

Johnes [23]

Education

Johnes and Li [14]

Education

Nayar and Ozcan [10]

Healthcare

Staub et al. [18]

Finance

Focused area Comparing about hospital cost and production from 2 estimation models Evaluating the 6 vendor supplying an item to a baby food manufacturer Analyzing bank efﬁciencies in croatia between 1995 and 2000 Measurement of universities’ performance in UK, 1993 The multi stakeholder perspective on benchmarking rates ﬁrms’ performance Measuring the content of higher education in England, 2000–2001 Examining the efﬁciency in the production of 109 Chinese regular universities in 2003 and 2004 Performance measurements of quality in Virginia hospitals Evaluating cost, technical and allocative efﬁciencies for Brazilian bank between 2000–2007

Method Translog joint cost function, OutputOriented CCR Model Input-Oriented CCR

Both OutputOriented BCC and CCR Models Ordinary least squares method, stochastic frontier analysis model, multilevel model Input-Oriented and Output-Oriented CCR Model Output-Oriented BCC Model Output-Oriented BCC Model

Input-Oriented CCR Model

Regression model

(continued)

Efﬁciency Analysis in Retail Sector

887

Table 1. (continued) Author(s) Ram Jat and Sebastian [12]

Sector Healthcare

Paradi and Zhu [17]

Finance

Mirhedayatian et al. [16]

Production

Kawaguchi et al. [9]

Healthcare

Thomas et al. [24]

Industry

Othman et al. [19]

Finance

2.1

Focused area Measuring technical efﬁciency of 40 public district hospital in India Surveying 80 published DEA applications in 24 countries that focus on bank branches Evaluating 10 Iranian soft drinks companies in terms of green supply chain management Evaluating 9000 private and public hospitals in Japan Operational and environmental efﬁciencies of 47 prefectures in Japan Measuring relative efﬁciency of banks

Method Input-Oriented BCC Model

Additive and slackbased models (8 of them), BCC and CCR models (72 of them) Output-oriented network slack based model

Dynamic network and black box model Input-Oriented CCR

Both CCR and BCC models

DEA in Retail Sector

DEA method is a useful technique for performance evaluation in retail sector in terms of measuring multiple inputs and outputs. From retail perspective, DEA is used to ensure better service to customers over time in consequence of input-output analysis to perform new policies in line with the outcomes [24]. Nowadays, there is a perceptible rise and growth in competition in retail sector, therefore, increasing the performance of retail stores to gain competitive advantage and investigating the causes of bottlenecks are essential. In order to do that, measuring current performance is necessary. For efﬁciency analysis, DEA is very suitable in terms of evaluating multiple inputs and outputs. According to the needs of research, inputs and outputs that are used in DEA may vary. When the literature is investigated, it has been noticed that value of stock, floor space, and recurrent costs were used as inputs, annual sales, and customer satisfaction were used as outputs with Output-Oriented CCR model by [25]. However, customer satisfaction in [25] refers to the customers that visit the store more than twice and does not consider an evaluation of customers. In addition to that, equivalent number of vehicles used for delivery and total transport cost were used as inputs and number of customers served, number of orders ﬁlled and total revenue were used as outputs with Input-Oriented CCR and BCC Model in the study of [26].

888

C. Kahraman et al.

Moreover, [27] focused on ﬁnancial measures and used current ratio, stock turnover and ﬁnancial lever as inputs, and net proﬁt margin and marketing value as outputs, where Output-Oriented CCR and BCC Models were utilized. On the other hand [28] analysed retail sector from more macro perspective and used quality management applications and systems, internal check, documentation capacity of process and production, company management, design and development opportunities, performance of cost reduction as inputs; Quality Cost, Delivery, Performance of Cost Reduction as outputs. In the same study [28], DEA was integrated with AHP (Table 2). Table 2. Summary of literature review in terms of retail sector Author(s) Kamakura et al. [29]

Thomas et al. [24]

Methods Inputs OutputLabor, and customer Oriented CCR service area

Input-Oriented Labor, experience, CCR location related costs, and internal processes Donthu and Yoo [30] Input-Oriented Environmental CCR conditions, customers factors, retail ﬁrms managerial efforts, and employees factors Gemici [28] DEA and AHP Quality management applications and systems, internal check documentation, capacity of process and production, company management, design and development opportunities, and performance of cost reduction Mishra [25] OutputValue of stock, floor Oriented CCR space and recurrent costs Mostafa [31] OutputEmployees, and assets Oriented CCR Lau [26] Input-Oriented Equivalent number of CCR and BCC vehicles used for delivery, and total transport cost Gandhi and Shankar [32] CCR and BCC Cost of labor, and capital employed Geyikçi and Bal [27] OutputCurrent ratio, stock Oriented CCR turnover, and ﬁnancial and BCC lever

Outputs Cash Deposits, Other Deposits Pays (Volume of transit in branch), and MREV (Service fee) Sales and proﬁt

Financial or economic outcomes, and behavioral outcomes

Quality, cost, delivery, performance of cost reduction

Annual sales, and customer satisfaction Revenue, earn share, and market value Number of customers served, number of orders ﬁlled, and total revenue Proﬁt, and sales Net proﬁt margin, and marketing value

Efﬁciency Analysis in Retail Sector

889

As mentioned before, contribution of this study is using store evaluation of customers as an output, in addition to other well-known inputs and outputs for retail sector. In the following section, methodology of the study is explained.

3 Methodology DEA is a methodology that evaluates relative efﬁciency of DMUs with using multiple inputs and outputs. Also, DEA is used for analyzing the managerial performance of productive units. Although, there are different methods under DEA and integration between other methods is possible, CCR (Charnes, Cooper, Rhodes) and BCC (Banker, Charnes, Cooper) are the most well-known DEA methods. CCR model assumes that constant returns to scale whereas BCC model considers that variable returns to scale [33]. Because of this differentiation, CCR and BCC models are separated from each other. In addition, models are categorized as input-oriented and output-oriented models. The aim of the input-oriented model is maximizing the outputs. On the other hand, the goal of the output-oriented model is minimizing the inputs. Retail sector is a dynamic sector and the changes in inputs or outputs do not affect the other linearly. In other words, changing the values of inputs do not cause a change in outputs in the same amount. Therefore, a model which allows variable return to scale is needed. From this point of view, BCC model is more appropriate for retail sector. Furthermore, maximizing the outputs is the aim of the retail sector. For this reason, input-oriented BCC model was chosen. Linear programs for the input-oriented BCC model (primal) given below; Xp Qk ¼ Max l Y l rk r 0 r¼1 Subject to: Xm i¼1

Xp r¼1

lr Yrj

xi Xik ¼ 1

Xm i¼1

xi Xij l0 0

lr e xi e µ0: unconstrained j ¼ 1; . . .; n r ¼ 1; . . .; p i ¼ 1; . . .; m

890

C. Kahraman et al.

In here; µr: rth output is weighted by DMU of k, xi : ith input is weighted by DMU of k, Yrk: rth output is produced by DMU of k, Xik: ith input is used by DMU of k, Yrj: rth output is produced by jth DMU, Xij: ith input is used by jth DMU, ɛ: a sufﬁciently small positive number, µ0: variable that is related to the scale according to input direction. In the following section, details of the implementation of the study are presented by covering company details, inputs and outputs of the model and input oriented BCC model of the problem.

4 Implementation of the Study Implementation of the study was conducted in a local retail chain in Turkey and mathematical model was structured based on a real life problem. Name of the chain is called as ABC Retail chain in this study. ABC Retail chain currently has 20 stores in different locations. However, in this study 7 stores were selected by company manager for initial performance evaluation. Name of these stores are Narlıdere, İskele, Balçova, Bucakoop, Alaybey, Bayraklı and Karabağlar called DMUs in the mathematical model. As mentioned in the previous section, input oriented BCC model in this study, and in the following subsections ﬁrstly inputs and outputs for the study and mathematical model of the problem given respectively. 4.1

Inputs and Outputs of the Model

As mentioned in the literature review part, there are many different inputs and outputs related to retail sector that can be used to evaluate efﬁciencies of DMU’s. In this study 4 inputs and 3 outputs were selected according to the needs of the ABC retail chain. Except store evaluation of customers, all the inputs and outputs are in line with the previous literature. Inputs are modiﬁed from [24, 26, 29, 31] and outputs are taken from [24, 32]. Brief explanations of inputs and outputs are given below: Inputs: Size of the Land: It refers to the size of the each store in meter square. Number of Employees: It indicates the number of employees in with different jobs including cashiers, cleaning attendants, security etc. Number of Deliveries: It refers to number of deliveries that the store received in a month. Total Cost: It includes all the monthly costs including rent, transportation cost, taxes, staff salaries etc.

Efﬁciency Analysis in Retail Sector

891

Outputs: Number of Customers: It refers to the number of customer that visited the store for shopping in a month. Sales: It contains amount of sales for each stores. Store Evaluation of Customers: It indicates general feelings of customers about the store. In order to measure store evaluation of customers’ output 10 positive statement were prepared and short interviews were conducted with customers separately in each DMUs. Statements are given as below: – Design of the layout in the store is good, so I can ﬁnd the products that I want to purchase easily. – Employees are friendly and helpful. – Cashiers are experienced and well trained. – Number of open checkout corner is sufﬁcient, thus, I get served rapidly. – The store is clean and tidy. – Store has a good location; it is easy to arrive. – Price labels and discount tags are accurate on the shelves. – I can ﬁnd the products that I want to purchase in store inventory. – It is easy to return or change the products if it is necessary. – Number of employees are sufﬁcient. Customers were asked evaluate the statements with a 5 points scale, where 5 point has the highest rate and refers that customer strongly agree the statement, and 1 point is the lowest and indicates that customer does not agree with the statement. In total, 400 evaluations have been conducted, which were distributed among the stores according to the number of customers of each store. In Table 3, dataset of inputs and outputs that are going to be used to formulate the mathematical model, is given. 4.2

Mathematical Model of the Problem

According to the identiﬁed inputs and outputs, model of the problem was generated. In DEA, separate mathematical models are needed to be formulated for each DMU. However, only objective functions and ﬁrst constraints are changed according to the DMUs. Other constraints are the same for each DMU. The models are given below; DMU1: Narlıdere Maxg1 ¼ 24748l1 þ 621508l2 þ 46l3 l0 350x1 þ 30x2 þ 16x3 þ 116073x4 ¼ 1

892

C. Kahraman et al. Table 3. Input and output data

Name of the store

Inputs Land Number of employee Narlıdere 350 30 İskele 550 17 Balçova 350 12 Bucakoop 300 17 Alaybey 650 32 Bayraklı 350 25 Karabağlar 700 14

Number of Total delivery cost 16 116073 12 83487 12 50119 12 74500 24 107000 24 108768 12 72411

Outputs Number of customer 24748 14496 17420 21000 37768 49036 12272

Sales 621508 545600 397096 435200 856000 852000 267792

DMU2: İskele (Urla) Maxg2 ¼ 14496l1 þ 545600l2 þ 43l3 l0 550x1 þ 17x2 þ 12x3 þ 83487x4 ¼ 1 DMU3: Balçova Maxg3 ¼ 17420l1 þ 397096l2 þ 40l3 l0 350x1 þ 12x2 þ 12x3 þ 50119x4 ¼ 1 DMU4: Bucakoop Maxg4 ¼ 21000l1 þ 435200l2 þ 44l3 l0 300x1 þ 17x2 þ 12x3 þ 74500x4 ¼ 1 DMU5: Alaybey Maxg5 ¼ 37768l1 þ 856000l2 þ 38l3 l0 650x1 þ 32x2 þ 24x3 þ 107000x4 ¼ 1 DMU6: Bayraklı Maxg6 ¼ 49036l1 þ 852000l2 þ 35l3 l0 350x1 þ 25x2 þ 24x3 þ 108768x4 ¼ 1

Store evaluation of customers 46 43 40 44 38 35 22

Efﬁciency Analysis in Retail Sector

893

DMU7: Karabağlar Maxg7 ¼ 12272l1 þ 267792l2 þ 22l3 l0 700x1 þ 14x2 þ 12x3 þ 72411x4 ¼ 1 24748l1 þ 621508l2 þ 46l3 350x1 30x2 16x3 116073x4 l0 0 14496l1 þ 545600l2 þ 43l3 550x1 17x2 12x3 83487x4 l0 0 17420l1 þ 397096l2 þ 40l3 350x1 12x2 12x3 50119x4 l0 0 21000l1 þ 435200l2 þ 44l3 300x1 17x2 12x3 74500x4 l0 0 37768l1 þ 856000l2 þ 38l3 650x1 32x2 24x3 107000x4 l0 0 49036l1 þ 852000l2 þ 35l3 350x1 25x2 24x3 108768x4 l0 0 12272l1 þ 267792l2 þ 22l3 700x1 14x2 12x3 72411x4 l0 0 lr ; xi e [ 0ðr ¼ 1; 2; 3Þ ði ¼ 1; 2; 3; 4Þ

µ0: variable that is related to the scale according to input direction, ɛ: a sufﬁciently small positive number. In the following section, numerical results of the given mathematical models are presented by using a software called Frontier Analyst.

5 Numerical Results After generating the model, the problem was solved with Frontier Analyst which performs an analysis with linear programming to determine the relative efﬁciency of organizational units that manage. Before entering the data as shown in Fig. 1, maximizing outputs, and varying returns options are selected on the software to receive an input oriented BCC solution.

Fig. 1. Inputs and outputs

894

C. Kahraman et al.

Results in terms of efﬁcient and inefﬁcient DMUs are revealed as shown in Fig. 2. According to the results, only DMU7: Karabağlar, was found as inefﬁcient since its score is less than 1, in other words 100%. Rest of the DMUs revealed as efﬁcient.

Fig. 2. Efﬁciency scores of DMUs

Following analysis were conducted for only the inefﬁcient DMU7: Karabağlar Store with a score of 65.1%. Frontier Analysts suggested that DMU3: Balçova and DMU4: Bucakoop as reference to Karabağlar. In Fig. 3, speciﬁc results of Karabağlar is presented with actual and target values, potential improvement, and peer contribution results with DMU3: Balçova and DMU4: Bucakoop. According to the numerical results, shown in Fig. 3, size of the land for DMU7: Karabağlar needs be reduced from 700 m2 to 330 m2. Moreover, total cost should be reduced from 72411 TL to 59871 TL. On the other hand, changes in the number of employees, and number of deliveries were not suggested. In addition, when the output results are investigates; number of customers, total sales and level of store evaluation of customers should be increased to reach efﬁciency target.

Fig. 3. Frontier analyst solution of DMU7: Karabağlar

Efﬁciency Analysis in Retail Sector

895

6 Discussions and Conclusions As mentioned at the beginning of this paper, this study stands on a real life problem. During the initial meeting with the manager of ABC retail chain, their concerns about the DMU7: Karabağlar were revealed in the ﬁrst place. The numerical results of this study are found in line with their concerns. Therefore, some improvements in Karabağlar store should be conducted in order to avoid close down decision. Firstly, decreasing the size of the store can be suggested due to the results. This could be done by moving the store in a smaller size store alternative or using some parts of the store with different purposes. Moreover, it has been revealed that their customer and sales number is under the target value. In order to increase these numbers, different sales strategies and promotions can be implemented. Furthermore, opinions of customers were found negative according to the results of store evaluation, therefore Karabağlar store needs to focus on customer satisfaction more. Regular service quality surveys can be suggested to the store for improvement, which also helps to increase in number of customers and sales. In conclusion, this study aimed to make a performance evaluation in a local retail chain by using DEA method. As a contribution to the literature, store evaluation of customers were used as an output. By doing this, not only traditional numerical measures for inputs and outputs were used, but also opinions of customers about the store were considered. In future studies, performance evaluation can be conducted for all the stores of the retail chain and moreover, number of inputs and outputs can be increased.

References 1. Kaplan RS, Norton DP (1996) Using the balanced scorecard as a strategic management system 2. Ittner CD, Larcker DF, Meyer MW (2003) Subjectivity and the weighting of performance measures: evidence from a balanced scorecard. Account Rev 78(3):725–758 3. Wongrassamee S, Simmons JE, Gardiner PD (2003) Performance measurement tools: the balanced scorecard and the EFQM excellence model. Meas Bus Excell 7(1):14–29 4. Chan AP, Chan AP (2004) Key performance indicators for measuring construction success. Benchmarking Int J 11(2):203–221 5. Ji YB, Lee C (2010) Data envelopment analysis. Stata J 10(2):267–280 6. Mardani A, Zavadskas EK, Streimikiene D, Jusoh A, Khoshnoudi M (2017) A comprehensive review of data envelopment analysis (DEA) approach in energy efﬁciency. Renew Sustain Energy Rev 70:1298–1322 7. Chen WT, Tan PS, Fauzia N, Wang CW (2017) Performance assessment of residential building management utilizing network data envelopment analysis. In: Proceedings of the international symposium on automation and robotics in construction, ISARC, vol 34. Vilnius Gediminas Technical University, Department of Construction Economics & Property 8. Cook WD, Zhu J (2007) Classifying inputs and outputs in data envelopment analysis. Eur J Oper Res 180(2):692–699

896

C. Kahraman et al.

9. Kawaguchi H, Tone K, Tsutsui M (2014) Estimation of the efﬁciency of Japanese hospitals using a dynamic and network data envelopment analysis model. Health Care Manag Sci 17 (2):101–112 10. Nayar P, Ozcan YA (2008) Data envelopment analysis comparison of hospital efﬁciency and quality. J Med Syst 32(3):193–199 11. Banker RD, Morey RC (1986) The use of categorical variables in data envelopment analysis. Manag Sci 32(12):1613–1627 12. Ram Jat T, San Sebastian M (2013) Technical efﬁciency of public district hospitals in Madhya Pradesh, India: a data envelopment analysis. Glob Health Action 6(1):21742 13. Johnes J (2006) Data envelopment analysis and its application to the measurement of efﬁciency in higher education. Econ Educ Rev 25(3):273–288 14. Johnes J, Li YU (2008) Measuring the research performance of Chinese higher education institutions using data envelopment analysis. China Econ Rev 19(4):679–696 15. Weber CA (1996) A data envelopment analysis approach to measuring vendor performance. Supply Chain Manag Int J 1(1):28–39 16. Mirhedayatian SM, Azadi M, Saen RF (2014) A novel network data envelopment analysis model for evaluating green supply chain management. Int J Prod Econ 147:544–554 17. Paradi JC, Zhu H (2013) A survey on bank branch efﬁciency and performance research with data envelopment analysis. Omega 41(1):61–79 18. Staub RB, e Souza GDS, Tabak BM (2010) Evolution of bank efﬁciency in Brazil: a DEA approach. Eur J Oper Res 202(1):204–213 19. Othman A, Owen L (2001) Adopting and measuring customer service quality (SQ) in Islamic banks: a case Johnes 20. Jemric I, Vujcic B (2002) Efﬁciency of banks in Croatia: a DEA approach. Comp Econ Stud 44(2–3):169–193 21. Johnes J (2006) Measuring teaching efﬁciency in higher education: an application of data envelopment analysis to economics graduates from UK Universities 1993. Eur J Oper Res 174(1):443–456 22. Avkiran NK (2006) Productivity analysis in the service sector with data envelopment analysis 23. Johnes J (2006) Measuring efﬁciency: a comparison of multilevel modelling and data envelopment analysis in the context of higher education. Bull Econ Res 58(2):75–104 24. Thomas RR, Barr RS, Cron WL, Slocum JW Jr (1998) A process for evaluating retail store efﬁciency: a restricted DEA approach. Int J Res Mark 15(5):487–503 25. Mishra RK (2009) Benchmarking scheme for retail stores efﬁciency. Int J Mark Stud 1 (2):131 26. Lau KH (2013) Measuring distribution efﬁciency of a retail network through data envelopment analysis. Int J Prod Econ 146(2):598–611 27. Geyikçi UB, Bal V (2015) Veri Zarflama Analizi ile Borsa İstanbul AŞ’de Faaliyet Gösteren Toptan ve Perakende Ticaret Sektörü Firmalarının Etkinlik Analizi. Abant İzzet Baysal Üniversitesi Sosyal Bilimler Enstitüsü Dergisi 15(1):21–42 28. Gemici MF (2009) Tedarik Zincirinde Veri Zarflama Analitik Hiyerarşi Prosesi Yöntemiyle Perakende Sektöründe Tedarikçi Performans Değerlendirmesi, Doctoral dissertation, Fen Bilimleri Enstitüsü 29. Kamakura WA, Lenartowicz T, Ratchfrord BT (1996) Productivity assessment of multiple retail outlets. J Retail 72(4):333–356 30. Donthu N, Yoo B (1998) Retail productivity assessment using data envelopment analysis. J Retail 74(1):89–105

Efﬁciency Analysis in Retail Sector

897

31. Mostafa MM (2009) Benchmarking the US specialty retailers and food consumer stores using data envelopment analysis. Int J Retail Distrib Manag 37(8):661–679 32. Gandhi A, Shankar R (2014) Efﬁciency measurement of Indian retailers using data envelopment analysis. Int J Retail Distrib Manag 42(6):500–520 33. Cooper WW, Seiford LM, Zhu J (2004) Data envelopment analysis. In: Handbook on data envelopment analysis. Springer, Boston, pp 1–39

Model Sequencing and Changeover Time Reduction in Mixed Model Assembly Line Derya Tataroğlu1, Renan Dalkıran1, Sueda Sezen1, Damla Kesikburun1(&), and Yiğit Kazançoğlu2 1

Department of Industrial Engineering, Yaşar University, İzmir, Turkey [email protected], [email protected], [email protected], [email protected] 2 Department of International Logistics Management, Yaşar University, İzmir, Turkey [email protected]

Abstract. Changeover time is a time between last good product of a speciﬁc model, to a ﬁrst good product of another speciﬁc model. During this changeover time, materials, equipment and layout can be changed and this causes non-value added activities. This study aims the reduction of changeover time losses in the sixth line of a worldwide factory which produces combi boilers. Before developing the solution alternatives, causes of time losses were determined. After that, solutions have been searched with process design improvements. A simulation model and decision support system were generated to set the differences between proposed and current situation. While developing solutions, for speciﬁc symptoms, various technics were used such as 5S, Lean Manufacturing methods and Quality Function Deployment. Based on these studies, the current situation of the assembly line during changeover process has been improved. Keywords: Changeover improvement 5S Quality Function Deployment

Process design Lean manufacturing

1 Introduction This study is conducted in a world-wide combi boiler producer factory which is located in Manisa, Turkey. This report presents the solution methodology of model sequencing and changeover time reduction problem in a mixed model assembly line, which is the sixth lean production line of the company. While solving the problem, various methods were used. The steps that have been followed can be listed as the investigation of the literature to ﬁnd similar problems and their solution approaches, problem deﬁnition, presenting solution methodologies such as; Redesign of Material Placement, Mathematical Model, Poke Yoke, 5S, Quality Function Deployment (QFD), veriﬁcation of results, preparing the Decision Support System (DSS), by integrating Arena Simulation with Excel VBA. As a result, when compared with the current situation of the line, a signiﬁcant reduction in changeover time was achieved. © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 898–913, 2019. https://doi.org/10.1007/978-3-319-92267-6_72

Model Sequencing and Changeover Time Reduction

899

This project is held in the sixth line of the combi boiler producer factory which uses U type production system as it can be seen in Fig. 1. In this line, 4 models are being produced with 27 operators. These models are Model-1A, Model-1B, Model-2, Model3, Model-4 and the cycle times are 120, 120, 93, 93 and 80 s, respectively. While these models are being produced, the changeover time losses exist in sixth production line, because of the model transition. According to the production strategy of the combi boiler producer company, model transition occurs from larger cycle time to smaller cycle time. However, this strategy was not always applicable because of the variety of orders.

Fig. 1. Line layout with replacement of operators

The production process in the line should be ready when the model transition begins. Before the model transition, the location of materials, that will be used, should be changed according to new model by the logistic operators. In addition, the location of the operators must also be changed in various conditions and the hand tools for production process should be changed by the operators for producing new model in production line. Moreover, during the model transition, BOM list should be controlled, as well. At the end, these changes result in time losses and eventually time losses causes to less amount of production.

900

D. Tataroğlu et al.

2 Literature Review Changeover process in productive industries is a crucial problem that requires so many progresses to improve as in the combi boilers producer company. It causes huge losses to the company, when it is calculated as daily, monthly and especially annually. There are several studies that are associated with reduction of changeover times. Some common reduction procedure examples are, design for changeover (DFC), simulation and single minute exchange of die (SMED). In this study, the methods of changeover processes are investigated. It can be envisaged that achieving reduction in changeover times affects the total productivity. By supporting this argument, the following literature is included. In the article of Shingo [1], Single Minute Exchange of Dies (SMED) method is a necessary element for shorter setup times. The target of this approach is to reduce setup times less than 10 min to eliminate waste of time with four phases of SMED. Furthermore, this approach known as Quick Changeover of Tools. The methodology of SMED also tries to make an effort on minimizing the time required to ﬁnish all of the tasks that contain changeover. In the article of Almomani et al. [2] assumed these four phases as; setup process map, classifying activities as external or internal setups, transferring internal to external activities, and simplifying all internal and external activities. Furthermore, SMED approach is also initiated in the article of Sabadka et al. [3]. A video recording of the machine’s manufacturing process is made, based on which the individual operations of the redevelopment are analyzed and included in the individual categories and the speciﬁed times of the individual operations. Thirdly, within the topic of changeover process, according to the article of Mileham et al. [4], DFC improvement can also consist of four main areas, which are, product design, machine design, tooling design and system design and for all these areas, securing, location, adjustment, tool movement, cleaning and access subjects need to be considered. According to Mileham’s research to improve changeover times both design and methodology should be considered, since both has more potential time saving and it is more sustainable than only improving the methodology. For methodology improvement there is a list of procedures that consist of three stages, a strategic stage, a preparatory stage and an implementation stage. The understanding and implementation of the rules is relatively simple and not only reduces the transition times considerably, but it also provides the environment in which ecologist production can be done (Mileham et al. 1999). The DFC methods also can be useful for the combi boilers producer company sixth lean line, in logistics design and methodology during changeover and for station designs for tools and equipment. Also, in the article of Gungor and Evans [5], changeover operations can cause signiﬁcant burdens through product and time losses, additionally, water and energy use. Cleaning process during the changeover is also important event that cause unignorably losses. According to Gungor and Evans, true cost of a changeover can be calculated as sum of cleaning costs, product losses and time losses (2017). Gungor and Evans (2017) stated that some designs affect the changeover processes such as equipment design, standard operating procedure design and cleaning operations design.

Model Sequencing and Changeover Time Reduction

901

Moreover, McIntosh et al. [6], initiated that DFC is to facilitate rapid dismantling of equipment and subsequent rapid, accurate reassembly of the same equipment for purpose of changeover. It means that it can be used for decreasing losses of changeover processes by facilitating operators’ works and usage and maintenance of equipment. Finally, the topic of changeover process is related with simulation approach. According to Khalid and Kai, in order to be able to increase productivity, the current situation flows should be established (2017) [7]. Solution methods are suggested for problem solving in the current situation. The modeling is carried out using the simulation to show the current state of the flow and the state after the improvement. A case study of the production line of a production facility is used to verify the effectiveness of the proposed solutions and to demonstrate the basic features of the model being developed

3 Problem Deﬁnition In this study, the sixth line of the combi boiler producer company is examined, and the causes of time losses during the changeovers were detected. These causes are material placement, model 1 family transition, movement of operators during changeover, BOM control problem, hand tool and variation of test pads. A. Material placement is important for producing a new model. When a new model is produced, the location of materials is changed by logistic operators, but sometimes materials are not reachable for operators, therefore, operator change the place of material boxes in order to work in a comfort. Hence, it becomes more difﬁcult to ﬁnd the materials they need. This leads to loss of time. In addition, observations made on model 1 family suggest that there are further losses at Station-4, Station-5, Station-7, Station-8 and Packing Station-5 due to an improper location of materials during changeover. B. In model transitions, two of the operators change their places for producing a new model and this situation cause next stations to wait. In model-4, the two operators who are working in the hanging device station which can be seen in Fig. 1, are transferred to the combustion chamber stations when switching to model-1. When the changeover takes place, these two operators are needed in the combustion chamber stations or vice versa. However, since these operators are working in the hanging device station of the rotating model, they continue to work on the hanging device stations until the model-4 device is completely out of production. This place changing problem is also occurring when the models turn to model-2 and model-3. Buffer stock does not have a signiﬁcant impact on changeover losses. However, buffer stock is not recommended according to the lean production design. For this reason, the company aims to produce as few buffer stock as possible (or without buffer stock). C. For each product family that will be produced on the model change, the employees check the materials. In the material control; each product is made up of a list of materials prepared for the family and called Bill of Materials (BOM). Each operator, controls the materials in his station based on these lists. However, since

902

D. Tataroğlu et al.

this list includes all the materials of the combi boiler, it takes considerable amount of time for the operator to ﬁnd and control his own materials from the list. D. Hand tools change during the model change and operator cannot ﬁnd required hand tools because of untidy cabinets and variety of hand tools. This causes time loses in the line. There are hand tools that are used for each product family at each station where the assembly production line is operated. Hand tools are used for assembling the parts such as rivet guns and air torque tools. The hand tools are changed at each model based on tolerance ranges of the torque values. For that reason, these hand tools are subject to be changed at each changeover. During the exchange of hand tools, operator must change the tool from material cabinet or from other stations. In the meantime, traveling for the material cabinet and searching other stations for hand tools cause time loss during changeover. E. Test pads are varied and they are required for testing processes. Unsuitable test pads can damage the product during the test processes. It is time-consuming for the operator to ﬁnd the appropriate model from various pads. When the current situation and system are examined, it is observed that the planned time losses are signiﬁcantly exceeded. This situation is reflected as inefﬁciency and loss of time when the model conversion is performed in intra-family and inter-family devices in the sixth production line. In this study, it is aimed to reduce time losses and unforeseen losses during model conversion by keeping the number of operators and cycle times constant as they were designed. It is aimed that with the reduction of losses, the productivity of the production line will increase.

4 Problem Formulation and Solution Methodologies This study aims the reduction of time loss in model conversion. In order to do that, in this section, the following methods have been suggested, which are corresponding to each problem that was deﬁned in the previous section. 4.1

Redesign of Material Placement

In order to solve the logistic operators’ material positioning problem, material placement plans for each station were generated. After all the material list analysis have been made, the content of that plan was generated with all materials that were used in the process. When the list has been created, materials have been grouped as the Kanban Materials and Consumable Materials. Consumable materials are the most frequently used materials that come in large numbers such as screws, staples, gaskets and etc. Kanban materials are the materials has a critical role in the process that came with certain batches such as boiler covers, display screen and etc. The usage of materials by operators had been considered. After all these processes, material plans are completed and placed to the all stations as operators and logistics operators can see. As a result, the material feeding and usage of them has become easier.

Model Sequencing and Changeover Time Reduction

903

Fig. 2. Material placement plan example for a station (Color ﬁgure online)

Observations made on model 1 family suggest that there are further losses at Station-4, Station-5, Station-7, Station-8 and Packing Station-5 due to an improper location of materials during changeover. The time loss was calculated by the following calculation; “40 s * 6 (station) = 240 s for the line” since 40 s. spend at material changeover when the other stations spend 20 s. As an opportunity cost, this calculated time equals to producing 20 devices. The expectation of the company in the scope of this project study is that this lost time shall be reduced to 120 s for the line. This reduction allows to earn 10 devices. The loss of this time during changeover was tested with the new materials plan and a decrease in time were observed. A list of all materials that were used in model-1 sub models has been compiled and optimization analysis was performed. Only the locations of the kanban materials that are common to the families have been ﬁxed. All other materials, kanban or consumable were returned to the line with hourly material feed. This improved the duration of the model changes compared to the previously set goal, and the reduction in time reflected onto the simulation model. 4.2

Losses Due to Movement of Operators During Changeover

One of the most important losses is, the movement of operators, during changeover for Model 1, 2, 3 and 4. This movement cause time wastes when operators move to their new stations for producing new models. In case of the changeover, switching models from large cycle time to small cycle time generates time wastes. In Model 4, two operators that work in the hanging device station, move to the combustion chamber stations for switching to Model 1. In Model 2 and 3, two operators that work on the combustion chamber station 1 and 2, move to eighth and tenth stations. These movements retard next stations and the production line. The total time losses are calculated by multiplying number of waiting stations and cycle times of models. Table 1 shows the model change with minimum time waste. For instance, ﬁrst

904

D. Tataroğlu et al.

model change does not waste time until the change from Model 4 to Model 1B as shown in Table 1. This change retards eighteen stations with the cycle time of Model 4. Table 1. Optimization of model conversion Changeover (Type to Type) Calculation of total loss times (sec) 1A-3-2-4-1B 18 * 80 = 1440 1A-4-1B-3-2 18 * 80 = 1440 1A-1B-2-3-4 0* 1A-2-3-1B-4 18 * 93 = 1674 1B-3-2-4-1A 18 * 80 = 1440 1B-4-1A-3-2 18 * 80 = 1440 1B-1A-2-3-4 0* 1B-2-3-1A-4 18 * 93 = 1674 2-3-1A-1B-4 18 * 93 = 1674 2-1B-3-4-1A (18 * 93) + (18 * 80) = 3114 2-4-1A-1B-3 18 * 80 = 1440* 2-3-4-1A-1B 18 * 80 = 1440* 3-2-1A-1B-4 18 * 93 = 1674 3-1B-2-4-1A (18 * 93) + (18 * 80) = 3114 3-4-1A-1B-2 18 * 80 = 1440* 3-2-4-1B-1A (18 * 93) + (18 * 80) = 3114 4-3-2-1A-1B (11 * 80) + (18 * 93) = 2554 4-1A-1B-3-2 18 * 80 = 1440* 4-1B-3-2-1A (18 * 93) + (18 * 80) = 3114 4-2-1A-1B-3 (11 * 80) + (18 * 93) = 2554

A mathematical model is formulated for losses due to movement of operators during changeover. (1) Parameters Sets and Decision Variables This model includes four sets and the deﬁnition of these sets are; n ¼ Permutation Indices m ¼ Model Indices s ¼ Model Change Indices r ¼ Sorting Indices

n ¼ 1; 2; 3; . . .; 256 m ¼ 1; 2; 3; 4 s ¼ 1; 2; 3; 4 r ¼ 1; 2; 3; 4

The model parameters are speciﬁed as shown below. DTms = The difference between the number of old and new stations that operators were located m, s CTm = Cycle Times of Models m OTm = Order Time m Tn = Lost time due to movement of operators n

Model Sequencing and Changeover Time Reduction

905

Nmr = The location of model in changeover m, r Pn(mr) = Permutations of model change n Decision variables are speciﬁed as shown below. Tn ¼

1; waste time occurs 0; otherwise

OTm ¼ Nmr ¼

1; Nmr 0; otherwise 1; Pnðmr Þ 0; otherwise

(2) Mathematical Model The mathematical, model that we have formulated is given below. Min

X

Tn

X

OTm

X

NmrðPnðmrÞ Þ

ð1Þ

DTms 30

ð2Þ

Tn ¼ DTms CTm

ð3Þ

1 Nmr 4

ð4Þ

OTm 0; Tn 0; CTm 0; DTms 0; PnðmrÞ 0

ð5Þ

The objective function (1) calculates the weekly production planning according to the orders where there is no time loss in model changes because of the operator changes. The difference between number of stations (2) should be less than or equal to 30, since there are 30 operations in total. Time loss due to movement of operators between models (3) is calculated by multiplying the difference between the number of old and new stations which operators were located and the cycle time of the model. Constraint (4) states that the location of model should be between 1 and 4 during changeover. Model parameters should be greater than or equal to zero (5). Furthermore, Model 1A and Model 1B are deﬁned as Model 1 in common. The reason is that, according to the results of the calculations, the time loss due to the movement of the operator are the same for both Model 1A and 1B.. The mathematical model became more understandable and contributed to have faster results on Excel VBA. 4.3

BOM Control Problem

The Bill of Material lists are consisting all materials used in the line, and the operators were checking the materials, it was too long to control and it was taking approximately 52 s. In order to deal with this problem, Poka-Yoke is used, it is a lean manufacturing technique that aims to prevent, correct, or reduce the errors of misunderstanding,

906

D. Tataroğlu et al.

carelessness, or indecisiveness of operators before they happen. After the improvements, the lists become unique to each station and it takes 30 s to control. Also, coloring of materials in the product list has been suggested to make the job even easier. The BOM list is given to the operator of ﬁrst station by the logistics operator in order to start the work. BOM control is performed by operators who check the codes on the boxes. The signature page at the back is signed and after that BOM control list is ﬁxed with a magnet on the combi boiler. Transition of this list to the next station is ensured and it is expected to be done at every station. Previously, the material codes had to be searched through the whole material lists and these codes were not speciﬁed for models and were not in order of importance, for example, before the new list, Kanban and consumable materials was in a mixed order. Thereby, now it became a list that is easily reached and followed by the operator. This study does not only improve production efﬁciency and quality, but also decreases human error. 4.4

Hand Tool Suggestions

To overcome the problem in hand tool, again Poka-Yoke is used, it is a lean manufacturing technique that aims to prevent, correct, or reduce the errors of misunderstanding, carelessness, or indecisiveness of operators before they happen. The hand tools used in the production line have speciﬁc torque values. As shown in Table 2, combination of torque values was done to minimize the worker error and the time loss while searching for correct hand tools. Some torque values with wide range are narrowed, reducing the number of hand tools to 9.

Table 2. Combination of torque values Hand tools Torque Screwdriver Torque wrench Torque device Vertical screwdriver Torque wrench Vertical gun

Torque values Value -1 Value -2 Value -3 Value -4 Value -5 Value -6 Value -7 Value -8 Value -9 Value -10 Value -11 Value -12

Number of hand tools 1 1 2 2 3 3 2 1 1 1 5 1

Combination of torque values Value -1

Number of common hand tools 1

Value -3

2

Value -6

3

Value -8

1

Value -10

1

Value 12

1

Model Sequencing and Changeover Time Reduction

907

Taking hand tool from cabinet or station is one of the signiﬁcant loss for the model transformation. By using the 5S technique, the hand tools were examined and classiﬁed for each model on a station-by-work basis, based on their torque values. As a result, hand tools will be unique for each station. Figure 2 shows design of hand tool cabinets. Styrofoam was used in the design and it was cut into to the shape of the hand tools. The same hand tools that has different torque values was labeled and colored to be separated from each other (Fig. 3).

Fig. 3. Hand tools cabinet

4.5

Test Pads Design Suggestions

For the solution of the test pad problem, the Quality Function Deployment (QFD) method has been studied and adapted to the problem. With this method, operations were separated and graded according to their importance and priorities. Then, the roof is drawn, and priority order has decided by rating according to relations. Then, it has been decided to think of the operator as a customer, and to reveal the claims of the operator and to check their feasibility using the quality function deployment method. According to the QFD results, as shown in Fig. 4, ergonomic design found out to be the leading one followed by test subdivisions and new shelf system design, therefore, we have designed a 4-layer test bench rack system to overcome the mess in test pads and sorting of the other bench in the 9th station during the change of the test bench. The shelf system, shown with the visuals given below, has all the test panels of the models and reduces the time loss by eliminating the need for the operator to put the test pads and order them or place them on the shelves. Test pads will be painted with different colors for each model and the shelf in the same color will be sorted and we will be able to overcome the problem of mixing any test pad, as shown in Fig. 5.

908

D. Tataroğlu et al.

Fig. 4. QFD method usage for test pad design

Fig. 5. Test pads design (Color ﬁgure online) Table 3. Comparison of changeover loss times after improvements Model conversion Apparatus exchange Material regulation BOM reading-checkingsignature Total

First situation in main models (Sec) 55 42 52

Last situation in main models (Sec) 45 35 30

149

110

5 Veriﬁcation of the Application and Generating a Decision Support System In this section, the numerical results obtained from the solution of the problem have been analyzed, measured and the veriﬁcation of the proposed system have been made. The main reason of losses and average times measured can be seen in the Table 3.

Model Sequencing and Changeover Time Reduction

909

These measurements were calculated by the average of the measured times at each station during model conversion. After all the reductions, the efﬁciency of the sixth line has been calculated and the detailed calculations has given below: Average changeover in a shift ¼ 10 ! in 3 shifts : 3 10 ¼ 30 Initial Situation: Daily lost time: 30 * 149 s (Maximum Changeover Loss) = 4470 s Percentage Production for Models 9 Model-1 : %30 > > = Model-2 : %15 ð0:3 120Þ þ ð0:15 93Þ þ ð0:3 93Þ þ ð0:4 80Þ Model-3 : %15 > > ; Model-4 : %40 ¼ 95; 9 Weighted Average Cycle Time Scheduled run time = 415 s Production piece in a shift: ð415 60Þ=95:9 ¼ 259 combi boiler ! Daily : 3 259 ¼ 778 combi boiler Number of lost devices per day: 4470/95.9 = 46 combi boiler Final Situation: Daily lost time: 30 * 110 = 3300 Number of lost devices per day: 3300/95.9 = 34 combi boiler Gain: 46 – 34 = 12 combi boiler Annual Productivity Increase: 12/778 = 1.5% The results are simulated by Arena Simulation Software before generating the DSS. With simulated results, the possible effects were observed. The station-based time losses for all possible models are transferred directly from Decision Support System to Arena Simulation with Read-Write module. The simulation program uses time losses as a variable table in the program itself. Thus, this table reflects non-value-added time in terms of changeover time. The determination of which model should be produced is given by the company. According to this data, simulation runs for one shift and simulates changeovers. These changeovers can be seen from an Andon system which is used by the combi boiler producer company, inside the production line. In this study Excel VBA is used to make updates based on user input, results of the improvements, comparative graphics showing the initial situation and the ﬁnal situation. The weekly production plan was arranged to prevent the losses due to the shift of the operator based on the mathematical model. The home page shown in Fig. 6, provides information to the user about how to use the system and this information is reachable at any time.

910

D. Tataroğlu et al.

Fig. 6. Home page

In the update section as shown in Fig. 7, the user can update cycle times, changeover times which are based on stations and instructions.

Fig. 7. Update page

Fig. 8. Comparison dashboard page

Model Sequencing and Changeover Time Reduction

911

In comparison section, user can easily understand the time losses during conversion between models. The bottleneck based time losses of initial situation and improved ﬁnal situation are shown on Fig. 8 with the comparison graphs. Arena Update section is aimed to be changed dynamically and the data table that was used in the simulation program, so that Arena and Excel VBA were synchronized. In production ordering section, as shown in Fig. 9, the models are sorted without any problem and this section creates a weekly plan by using Excel VBA with the mathematical model that is based on decreasing time losses due to movements of operators and expected orders.

Fig. 9. Production order page

6 Conclusion and Future Work As a conclusion, several solution methodologies has been used to decrease the changeover losses in the sixth lean production line at the combi boilers producer company. These solution methodologies are Mathematical Modelling, Poke Yoke, 5S, Quality Function Deployment (QFD), Decision Support System (DSS), by integrating Arena Simulation with Excel VBA. These solution methodologies and approaches have been applied in the factory. The material placement plans have been applied by analyzing material lists based on each station to decrease changeover losses. Model 1 family transition time has been reduced from 240 s to 120 s. Losses due to movement of operators during changeover have been studied based on mathematical model. BOM checking problem has been solved by developing colored product lists for each station. Hand tools has been improved with Poke Yoke and 5S. Test pads have been redesigned based on QFD. After all, DSS has been built as a user-friendly tool to update the data, to compare the initial and the ﬁnal situation, and to plan the production orders with minimizing time losses. DSS is synchronized with Arena Simulation by changing or updating the changeover times. Finally, changeover losses has been decreased, current situation has been improved signiﬁcantly, and the productivity of the sixth line has been increased by 1.5%.

912

D. Tataroğlu et al.

Future works may concentrate on deeper analysis of new mechanisms and methods. During this study main focus was on how to reduce the changeover time losses without changing any labor force. However, following suggestions can be analyzed to reduce the changeover loss times in the future: • In BOM list checking problem, the Industry 4.0 methods can be used and BOM lists can be programmed regarding to warehouse management and can be digitalized with barcode systems. This system can reduce the checking time and it also creates a sustainability since no more papers will be used. • There are stations that must use the trash box according to the operations performed at the stations in the production line. The stations, where waste containers are used, differ in each model and there is a limited number of waste containers inside the line. It has been observed that these waste containers were moved to stations where they will be used during the changeover. These unnecessary movements also cause time losses. • Samples can be taken from materials coming from the suppliers these samples can be inspected before they were sent to the production line for quality control purposes. During this process, if there is a defective material among the raw materials and component, there is a risk that it may cause production delay or even production stoppage. Therefore, statistical quality control methods can be applied in the future. • Products are transported by conveyor system till the 10th station. In the 10th station combi boilers are assembled to the trolleys in order to execute testing and packing operations. During this assembly, trolley pin settings are needed to be changed for each model to ﬁt the combi boilers perfectly. These changes are carried out by workers and cause time losses. There are 17 trolleys on the sixth production line, and during the changeover, the average run time is 5 s. Therefore, more improvement methods can be studied in this situation, as well. Acknowledgement. We are grateful to the company for sharing their data with us to complete this work. This work cannot be completed without the assistance of Cumhur Kerem Güven, Aygen Aytaç, Emre Akkuzu, and Yasemin Erdem. We are thankful for the contribution of them.

References 1. Shingo S (1985) A revolution in manufacturing: the SMED system. Productivity Press, Cambridge 2. Almomani MA, Aladeemy M, Abdelhadi A, Mumani A (2013) A proposed approach for setup time reduction through integrating conventional SMED method with multiple criteria decision-making techniques. Comput Ind Eng 66(2):461–469 3. Sabadka D, Molnar V, Fedorko G (2017) The use of lean manufacturing techniques – SMED analysis to optimization of the production process. Adv Sci Technol 11(3):187–195 4. Mileham A, Culley S, Owen G, Mclntosh R (1999) Rapid changeover – a pre-requisite for responsive manufacture. Int J Oper Prod Manag 785–796 5. Gungor ZE, Evans S (2017) Understanding the hidden cost and identifying the root causes of changeover impacts. J Clean Prod 167:1138–1147

Model Sequencing and Changeover Time Reduction

913

6. McIntosh RI, Culley SJ, Mileham AR, Owen GW (2000) A critical evaluation of Shingo’s SMED (Single Minute Exchange of Die) methodology. Int J Prod Res 38:2377–2395 7. Mustafaa K, Chenga K (2017) Improving production changeovers and the optimization: a simulation based virtual process approach and its application perspectives. In: 27th international conference on flexible automation and intelligent manufacturing, FAIM 2017, 27–30 June 2017, Modena. Procedia Manufacturing, vol 11, pp 2042–2050

Repair Cost Minimization Problem for Containers: A Case Study Merve Çamlıca, Gülce Çini, Ayşegül Eda Özen, Nilay Çınar, Sel Ozcan(&), and Deniz Türsel Eliiyi Department of Industrial Engineering, Yaşar University, Bornova, Turkey [email protected], [email protected], [email protected], [email protected], {sel.ozcan,deniz.eliiyi}@yasar.edu.tr

Abstract. With the increase of global trade activities and the decrease in transportation costs, the volume of international trade has grown considerably, yielding to an increase in the logistics activities as well as the number of containers used in transport. Hence, containers gradually deteriorate over time and logistics companies start to diminish the repair costs of the containers, as well. This study addresses the repair cost minimization problem for containers where the objective is to minimize the total transportation, repairing and delay time costs. The problem is formulated as a variant of the well-known transportation problem, where we assume that the broken containers transport to repair depots, and the ﬁxed containers are delivered from the repair depots to the customers until the requested due date. In order to monitor the performance of the proposed model, realistic test instances are constructed. Optimal solutions are achieved within 40 min for relatively small and moderate test instances. Keywords: Logistics Transshipment problem Cost minimization Delay cost

Broken container repair

1 Introduction Along with the change of the world, the methods of competition of the enterprises also change rapidly. A growing importance of speed, agility and ﬁerce competition in the global supply chain force companies to reconsider using traditional logistics services. As a result of the increase in both market competition and service level expectation of the customers, logistics service providers are forced to re-evaluate and concurrently improve their business services. The main goal in logistics is to reach a high level in customer service, to optimize the use of resources and investments, and to gain competitive advantage in this way. In order to establish the freight transportation in global manner, logistics providers should serve for complex networks with a large number of routing alternatives, which are mainly carried out by different transportation channels, including a collection of truck, rail, barge, air, and ship. Through containerization, all competitors have potentially the same level of access to an efﬁcient and global freight distribution system through ports [1]. Therefore, containerization has become the main driver for global intermodal freight transport, which involves the © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 914–921, 2019. https://doi.org/10.1007/978-3-319-92267-6_73

Repair Cost Minimization Problem for Containers

915

transportation of freight in containers of standard dimensions with the consideration of high level safety and environmental issues. Many containers are made from materials such as steel, aluminum, and these large boxes gradually deteriorate over time. Considering the damage level and repair alternatives as well as the coordination with the repair centers, the management of the repair operations for containers is a complex issue. Motivated from a real-life problem of a leading logistics company located in Izmir, Turkey, this study is focused on the assignment of broken containers to related repair centers and sending them to the related customers at a minimum cost. The rest of the paper is organized as follows. In Sect. 2, literature review is proposed. The problem and its characteristics are deﬁned in Sect. 3. The mathematical model is presented in Sect. 4. Section 5 includes the computational study where the solutions of sample problems are listed, tabulated and discussed. Finally, Sect. 6 includes our concluding remarks and future work.

2 Literature Review Reinhardt et al. studied the drayage problem with the objective of reducing the efﬁciency of pre- and end-haulage bottlenecks. In this study, the pre- and end-haulages of containers are scheduled using techniques of vehicle routing problems since the problem is based on the movements between the customers and the depots. It has been shown that the model can be easily solved using column enumerating since the number of possible paths in the problem is limited. The effect of side constraints on overall cost has also been analyzed [2]. Häll et al. design vehicle routes and schedules for a dial-a-ride service where some part of each request may be performed by a ﬁxed route service. Passengers can go from one place to another without changing the vehicle, however they can also exchange vehicle. Each request contains one or several passengers and requires a certain capacity in a vehicle, for the persons and any wheelchairs, walkers, luggage etc. They assume that all vehicles have the same capacity and the aim is to minimize the total routing cost [3]. Kiremitci et al. study one of the most important types of vehicle locating problems. The multi-vehicle distribution and collection problem with time windows, where the objective is minimization of total transport costs as well as the number of vehicles required, balancing routes for travel time and vehicle load. The number of variables is reduced by using real values as much as possible. New algorithms approaches are compared with old ones on some problems in the literature and it is observed that the proposed algorithm gives relatively better results [4]. Hlayel and Alia studied the transportation problem with the objective of reducing the transportation cost and time. The most important way of achieving the solution by the best candidate method (BCM) is to start with selecting the most suitable candidates and reduce the solution combinations accordingly. Combinations can be obtained without any means of intersection. By comparing the results obtained, the implementation of BCM in the proposed method achieves the best ﬁrst applicable solution for a transport problem and achieves faster than current methods with minimal computation time and less complexity [5].

916

M. Çamlıca et al.

Eliiyi et al. model the transfer problem of a company with a transfer warehouse with many subcontractors and customers in the transfer sector. Taking into account the supply times and the customer’s deadlines distinguishes the work from others [6].

3 Problem Deﬁnition As the need for containers increases during transport, the container damages increase accordingly. Logistics companies are trying to repair the broken containers in the shortest time and with the minimum cost and deliver them to the customers. Motivated from the real-life problem of the logistics company, who is serving as a pioneer of the maritime transportation business in Turkey, the decision of where and when to repair the broken containers is a complex issue. The problem is formulated as a transshipment problem, which is a variant of a transportation problem, where the source nodes are the points where the container is broken, and the sink nodes refer to the customer locations. The intermediate nodes between the source and sink nodes, i.e., transshipment points, denote the repair centers. Our problem differs from the classical transshipment problem, since time windows should also be considered. Time window restrictions imply that the customer containers need to arrive within a certain period of time. Moreover, each customer has a due date and additional penalty costs are incurred for each delay. In this direction, a solution method should be developed to minimize the problem of container repair, which is a problem that arise in real life.

4 Mathematical Model The following section describes our mathematical model. The problem is formulated as a transshipment problem where the objective is to minimize the total transportation, repairing and delay time costs. The mathematical model decides on how to transfer the broken containers from the breakdown points to one of the suitable depots and then to one of the candidate customers. We assume that there are three different vehicle types used for carrying containers from/to depots, breakdown points and customers. The indices and parameters used in the model of transshipment problem are deﬁned as below: i: breakdown point index j: depot point index k: customer point index Si : The number of total containers in breakdown point i Dk : The number of demand of customer k RTj : Repairing time of depot j TTij : Transportation time from breakdown point i to depot j for truck TTjk : Transportation time from depot j to customer k for truck STij : Transportation time from breakdown point i to depot j for ship STjk : Transportation time from depot j to customer k for ship Transportation time from breakdown point i to depot j for train RTij :

Repair Cost Minimization Problem for Containers

RTjk : WT:

917

Transportation time from depot j to customer k for train Waiting time of container, including the waiting time of the vehicle and in the queue Due date which is given by customer k Delay time of container Cost of delay The transportation and repairing cost for truck from i to j The transportation and repairing cost for ship from i to j The transportation and repairing cost for train from i to j The transportation cost for truck from j to k The transportation cost for ship from j to k The transportation cost for train from j to k The number of broken container which is transported by truck from i to j The number of ﬁxed container which is transported by truck from j to k The number of broken container which is transported by ship from i to j The number of ﬁxed container which is transported by ship from j to k The number of broken container which is transported by train from i to j The number of ﬁxed container which is transported by train from j to k

DDk : t: tc: Cij : Xij : Yij : Cjk : Xjk : Yjk : QTij : QTjk : QSij : QSjk : QRij : QRjk :

Decision variables are deﬁned as follows. ( Tij ¼ ( Sij ¼ ( Rij ¼

( Tjk ¼ ( Sjk ¼ ( Rjk ¼

1; 0;

if the container which is transported by truck from breakdown point i to depot j otherwise

1; 0;

if the container which is transported by ship from breakdown point i to depot j otherwise

1; 0;

if the container which is transported by train from breakdown point i to depot j otherwise

1;

if the container which is transported by truck from depot j to customer k

0;

otherwise

1;

if the container which is transported by ship from depot j to customer k

0;

otherwise

1;

if the container which is transported by train from depot j to customer k

0;

otherwise

918

M. Çamlıca et al.

Based on the deﬁnitions above, the MILP formulation of the problem is as follows. X X X X X X X X Min C QTij þ C QTjk þ X QSij þ X QSjk i j ij j k jk i j ij j k jk X X X X þ Y QRij þ Y QRjk þ tc t i j ij j k jk

ð1Þ

s.t. Tij þ Sij þ Rij 1;

8i; j

ð2Þ

Tjk þ Sjk þ Rjk 1;

8j; k

ð3Þ

X j

X j

QTij þ QSij þ QRij Si ;

QTjk þ QSjk þ QRjk Dk ; X i

X i

Si ¼

X k

8i

ð4Þ

8k

ð5Þ ð6Þ

Dk

X QTij þ QSij þ QRij ¼ QTjk þ QSjk þ QRjk ; k

8j

ð7Þ

QTij M Tij ;

8i; j

ð8Þ

QSij M Sij ;

8i; j

ð9Þ

QRij M Rij ;

8i; j

ð10Þ

QTjk M Tjk ;

8j; k

ð11Þ

QSjk M Sjk ;

8j; k

ð12Þ

QRjk M Rjk ;

8j; k

ð13Þ

RTj þ TTij Tij þ TSij Sij þ TRij Rij þ TTjk Tjk þ TSjk Sjk þ TRjk Rjk þ WT ðDDk þ M ð1 Tij þ Sij þ Rij ÞÞ t; 8i; j; k

ð14Þ

Tij ; Tjk ; Sij ; Sjk ; Rij ; Rjk binary; QTij ; QTjk ; QSij ; QSjk ; QRij ; QRjk ; t 0 integer;

ð15Þ

8i; j; k

The objective function in (1) minimizes the total transportation, repairing and delay time costs. Constraint set (2) enforces that every container must be transported with at most one transportation way (either with land route, shipping way or railway) from breakdown point i to depot point j. Constraint set (3) satisﬁes that every container must be transported with only one transportation way (land route, shipping way or railway) from depot point j to customer k. Constraint sets (4) and (5) ensure that the broken containers on the breakdown point i are transferred to the relevant customer point k. Constraint set (6) satisﬁes the supply and demand equality at breakdown and customer points. If a container transports to depot j, it should be send to the customer at

Repair Cost Minimization Problem for Containers

919

depot j. Constraint set (8–13) calculates the total number of containers transferred via each vehicle type. Constraint set (14) ensures that the total repair time for all depots, transshipment times between breakdown points and depots and customers should be smaller or equal to the due date of customers k. Finally, constraint set (15) deﬁnes the decision variables.

5 Computational Results The performance of the proposed mathematical model is tested on several test instances generated by realistic assumptions. The proposed mathematical model is solved using IBM ILOG CPLEX Optimization Studio 12.7.1 on a computer with an i7 processor and 16 GB of RAM. The smallest test instance includes 2 different container breakdown points, 3 different repair depots and 2 different customer points. Then, the instance is extended with 3 and 4 broken containers at each point. Figure 1 visualizes all possible combinations of the smallest instance tested.

Fig. 1. Network diagram of the toy data solution

The performance of the proposed model is observed in larger test instances. The properties of the large test instances are reported in Table 1. The case of where 10 breakdown, repair and customer points are considered can be denoted as small, and the remaining two cases, i.e., with 20 and 30 points in each, are labeled as moderate and large, respectively. We also expand our computational experiment by employing the number of broken container at each breakdown point as 1, 2 and 3, respectively. All test instances are solved optimally by CPLEX and Tables 2, 3 and 4 report the computational time performance of each problem tested. Note that, the computational time gradually increases as the problem size increases, as expected. The computational time performances of small and moderate cases are comparable (see Tables 2 and 3). Especially with 30 breakdown points, 30 depots and 30 customer points the solution time increased exponentially as the number of broken containers increased.

920

M. Çamlıca et al. Table 1. Characteristics of the large test instances

Number of breakdown points 10 10 10 20 20 20 30 30 30

Number of repair depots 10 10 10 20 20 20 30 30 30

Number of customer points 10 10 10 20 20 20 30 30 30

Number of broken container 1 2 3 1 2 3 1 2 3

Table 2. Computational results for the test instances with 10 points Number of broken container Time (in sec) 1 85 2 753 3 1298

Table 3. Computational results for the test instances with 20 points Number of broken container Time 1 322 2 545 3 2379

Table 4. Computational results for the test instances with 30 points Number of broken container Time 1 2376 2 7510 3 21072

6 Conclusions In this study, minimization of the transportation and repair costs of containers, which is one of the real-life problems, has been discussed. The problem is formulated as a variant of the transshipment model and solved optimally by IBM ILOG CPLEX Optimization Studio 12.7.1. The test problems with various sizes are constructed so as to reflect the real-life complexity. We can deduce that the model yields similar performances on relatively small and moderate test instances, whereas the computational time performance is low for larger test problems.

Repair Cost Minimization Problem for Containers

921

Since assignment of the broken containers is an operational level decision making as the future extension, sophisticated heuristic methods can be applied to obtain good quality solutions within reasonable time. Moreover, the formulation can be extended by using the real-time availability on each repair depot. Acknowledgment. First of all, we would like to thank the Professor Dr. Deniz Türsel Eliiyi for providing this research course opportunity to us. Additionally, we would like to thank to our instructors Sel Ozcan and Hande Öztop for their guidance and enlightenment. We feel enthusiasm and get excited about this special research thanks to them.

References 1. Rodrigue JP, Notteboom T (2015) Looking inside the box: evidence from the containerization of commodities and the cold chain. Marit Policy Manag 42(3):207–227 2. Reinhardt LB, Pisinger D, Spoorendonk S, Sigurd MM (2016) INFOR: information system and operational research: optimization of the drayage problem using exact methods. I N F O R J 54(1):33–51 3. Häll CH, Andersson H, Lundgren JT, Värbrand P (2009) The integrated dial-a-ride problem. Public Transp 1:39–54 4. Kiremitci B, Kiremitci S, Keskintürk T (2015) A real valued genetic algorithm approach for the multiple vehicle pickup and delivery problem with time windows. Istanb Univ J Sch Bus 43:391–403 5. Hlayel AA, Alia MA (2012) Solving transportation problems using the best candidates method. Comput Sci Eng Int J (CSEIJ) 2:23–25 6. Türsel Eliiyi D, Yurtkulu EZ, Yurdakul Şahin D (2010) Supply chain management in apparel industry: a transshipment problem with time constraints

Routing Optimization for Container Dispatching Operations Hasibe Serap Baş, Ayşe Tolan, Mahmut Ali Gökçe(&), and Cansu Yurtseven Industrial Engineering, Yaşar University, Bornova, Turkey [email protected], [email protected], {ali.gokce,cansu.yurtseven}@yasar.edu.tr

Abstract. An optimal container dispatch planning model for Alkon Logistics is proposed. Alkon Logistics handles container movement between three large ports and a logistics center nearby, connected to major highways and railroad systems. As a solution approach, a novel integer mathematical model has been developed to minimize total cost of operations. Our model is developed to serve three different ports, which are T.C Ege Gübre, Nemport, and APM Terminal. In addition to making daily operational decisions optimally, the proposed model also assists in making critical long-term decisions, such as vehicle investments. Our model is applicable to any transport company, which carries singlecommodity due to its flexible structure. Proposed model is solved using IBM ILOG CPLEX Optimization Studio 12.8 program and tested with real-life data from the company. The results show signiﬁcant improvement over the current situation. To ensure efﬁcient usage of the proposed approach, a decision support system (DSS) is designed and implemented that allows studying different scenarios. Created DSS is dynamic and practical for users and enables making analysis by using mathematical model with different fleet sizes. Results from experimentation show that compared to current situation, number of rented trucks goes down to less than 10 from 17 corresponding to a decrease in renting cost by 65% and a decrease of 25% in total cost. Proposed solution methodology provides a container dispatch planning with improved truck utilization and optimal cost. Keywords: Routing optimization Container terminal Logistics Decision Support System

Dispatch planning

1 Introduction The amount of container transportation in the world has been increasing at an amazing pace. Starting with 50 million twenty feet equivalent unit (TEU) in 1985, world container turnover has reached more than 350 million TEU in 2004 [1]. That trend continued to increase over the years, without any sign of slowing down. Today, number of cargo ships in the world is approaching 60000 and there are about 20 million containers traveling the oceans every day. For these reasons, container transportation has great importance for logistics companies, which handles the movement of containers between different modes of transportation. An efﬁcient container dispatch planning © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 922–936, 2019. https://doi.org/10.1007/978-3-319-92267-6_74

Routing Optimization for Container Dispatching Operations

923

enables faster and safer dispatching of commodities. In addition, efﬁcient container dispatching operations affects performance of logistics companies positively and, it provides strong relationship with customers. In container dispatching operations, there are many operational constraints and costs involved. There are also peak times for demand, based on strict timeline of ships’ and freight cars’ arrivals and departures. These peak times and congestions due to port trafﬁc lead to inefﬁcient trucking operations. This type of situations causes ﬁnancial loss for logistics ﬁrms. Our proposed solution provides an optimal and implementable container-dispatching plan, which can handle all customer demands under strict time constraints. The objective of the problem is minimizing total dispatching cost with increased truck utilization. To provide that, a novel mathematical model is developed on a time-space graph, which discretizes the passing time into manageable short time slots. Proposed model determines optimal routes of trucks and containers on graph.

2 General System Analysis Alkon Logistics is located in Biçerova, Izmir, Turkey. It manages container dispatching operations between Biçerova Container Terminal and three ports nearby Izmir. All three ports are located in Aliağa, İzmir. Company works in two-shifts; ﬁrst is between 08:00-16:00 and, second is 16:00-24:00. All trucks must return to Biçerova Container Terminal at the end of each shift. Company operates with its own truck fleet which consists of 15 trucks. Besides, company rents additional trucks at the beginning of each year with an agreement for peak time operations. The agreement is for one year period and currently 17 extra trucks are used. Containers are carried to the ports via open-bed trucks and they must arrive at given ports before the cut-off time of vessels. Cut-off time is the latest entrance time for dispatched containers to ports to be able to make it onto the vessel. Not being able to meet this constraint has signiﬁcant costs associated with it and hence this is taken as a hard constraint. Containers, which are delivered to ports earlier than cut-off times, can be stored at ports for a certain amount of time at no cost. This is called free storage period and is determined by port management on a company basis. If containers are stored at ports longer than free storage period, company incurs storage cost that must be paid to port management. Containers are dispatched from Biçerova Container Terminal for inland transportation via freight cars. Departure time of freight cars is 23:00 p.m. every day excluding Sundays. Possible routes of trucks, transportation modes of containers and, locations of Biçerova Container Terminal and three ports are shown in Fig. 1. As is the case in the world, both 20-feet and 40-feet containers are used and must be transported. Each truck can carry exactly one container between pairs of locations (port-terminal, port-port), whether the container is full or empty. In total, approximately 1000 containers are dispatched within a week, with a tendency of fast growth over the next years.

924

H. S. Baş et al.

Fig. 1. Locations of company and three ports

2.1

Observation and Symptoms

There are two main observations made. Firstly, the company has serious difﬁculty in applying appropriate dispatch planning during peak times. This situation leads inefﬁcient truck operations; meaning more trucks traveling without containers to demand locations for handling customer demands. Second symptom related to the ﬁrst one is inadequacy of the truck fleet. Number of trucks belonging to company can be insufﬁcient for dispatching operations during peak times. In such a case, company often rents additional trucks unnecessarily. These situations cause waste of money and time, decreasing performance of the company.

3 Problem Identiﬁcation There are a number of operational decisions, which have to be made on a daily basis. The main decisions are determining which truck takes which container, at what time and, from which location to which location. In other words, optimal flows of trucks and containers must be done subjects to all constraints. The objective function of the problem is minimizing total operational cost. Total cost includes traveling cost of trucks, renting cost of extra trucks and storage cost, if incurred (Figs. 2 and 3). There are two strict time related constraints that must be met. Delivery of containers must meet: • Cut-off time of vessels at ports: and • Departure time of freight cars in Biçerova:

Routing Optimization for Container Dispatching Operations

Fig. 2. Operation process

Fig. 3. Example of time-space graph

925

926

H. S. Baş et al.

If containers are dispatched from Biçerova via freight cars, they must arrive in departure time of freight cars in Biçerova.

4 Literature Review In literature, we ﬁnd some similar example problems that have been examined, such as inter-terminal transportation, multimodal transportation and, vehicle assignment. To the best knowledge of the authors, no exact model for this problem has been published before. Therefore, this work ﬁlls an important void in literature with an exact model that can be adapted to many different situations. However, the most relevant problem is a mathematical model of inter-terminal transportation [2] and our work is based on the idea from this work. Tierney et al. [2], developed an integer mathematical model for analyzing interterminal transportation. Containers can be dispatched between terminals by using different transportation modes such as rail, sea, and road. Inter-terminal transportation leads lateness of container delivery and congestion in peak times. Tierney et al., generated a system which aims minimizing penalized lateness of deliveries. Chung et al. [3] examine workflow of container transportation and develop mathematical models, which compound various characteristics and classiﬁcations of containers and trucks. Objective of this study is to minimize fleet size, total operational cost of vehicles, and total transportation cost. They solved this problem by splitting into three stages. For the ﬁrst stage, fleet size is tried to minimize by applying the Multiple Traveling Salesman Problem (MTSP) standard formula. For the second and third stages, operational costs are grouped for three different vehicle types and, total operating and transportation cost are optimized by applying insertion heuristics algorithm. Kim and Nguyen [4] proposed a solution approach for vehicle dispatching at port container terminals. A real-time vehicle dispatching algorithm is developed to assign delivery orders of containers to vehicles considering uncertain travel time of vehicles. This study also can be called as a scheduling problem for vehicles. To study different scenarios and test the dynamic environment, a simulation study is conducted. Their study can be helpful for increasing truck utilization with the developed algorithm. A mathematical model of inter-terminal transportation [2] was used as a guidance for solving the problem of Alkon Logistics. It is helpful for establishing a solution approach in handling constraints of this problem. The main contributions are allowing of cross-way movements, peak time routing, analysis of capital investment (truck purchasing) under a variety of scenarios and a decision support system that allows running the model.

5 Mathematical Model Weekly container dispatch planning is modelled on a time-space graph, which includes specially designed constructions. Created model decides routes of trucks during the day; it also allows cross-way movement. In cross-way movement, a truck travels to port from Biçerova with a container. After delivery, truck travels to another port for carrying

Routing Optimization for Container Dispatching Operations

927

another container and return to Biçerova. Cross-way movements are modelled in this work for the ﬁrst time. These cross-way movements allow better utilization of trucks and lead to much more efﬁcient operation. Besides, model includes components, which can decide how many extra trucks need to be rented. Extra trucks are rented in situations such as inadequate truck fleet for peak operating times. 5.1

Assumptions

First assumption is that capacities of all trucks are same; each truck can carry exactly one container from a location to another location. Second assumption is that, there is no difference in carrying full or empty containers. Third assumption is that daily operating time of two shifts are limited, same and, no overtime. 5.2

Construction of Time-Space Graph

First, a non-temporal base graph is designed. This base graph identiﬁes connections between Biçerova Container Terminal and three ports. Assume that n is the number of nodes. Thus, the base graph can be constructed as G ¼ ðN; AÞ, where N ¼ f1. . .ng is the set of nodes and A is the set of arcs ði; jÞ. Using the base graph, a time-space graph is designed. This graph is required to model the problem. A time period is needed as a discretized way of representing passing time. Assume that s is the number of time periods. In this way, the time-space graph can be constructed as GT ¼ ðN T ; AT Þ, where N T ¼ f1. . .s:ng is the set of all nodes in the time-space graph and AT is the set of all arcs ði; jÞ in the time-space graph. 5.3

Stationary Arcs

Stationary arcs are deﬁned to allow trucks and containers to remain in the same place along the time periods. These arcs connect each node to itself with an arc for the next time periods. Assume that AS is the set of stationary arcs. AS can be deﬁned as: [ AS ¼ 1 i n: ðs1Þ fði; i þ nÞg

5.4

Arcs and Node Properties

There are sets of nodes for accessing to each time-space node i with arcs (i 2 N T ). These sets can be divided into two different types of set. First is the sets of nodes which can enter to each node i with arcs fInðiÞg: Second is the sets of nodes which can leave from each i node with arcs fOutðiÞg: • InðiÞ ¼ fðj; iÞ 2 AT ^ j\ig • OutðiÞ ¼ fði; jÞ 2 AT ^ j [ ig At the start of the optimization, each node i is associated with a number of trucks which present at its location (i 2 N T ). In other words, si determines origins of the truck

928

H. S. Baş et al.

flows. It is expected that all trucks must return to Biçerova at the end of the day. Thus, at the start of the optimization, the only node, which has truck at time step one, is Biçerova Terminal node. 5.5

Demand

Container demands arrive in groups. Each demand group includes one or more containers. This is due to a number of containers for a customer arrive or leave with a single vessel. Assume that D is the number of demand groups D ¼ f1 d Dg. The following parameters are deﬁned for each demand group. • Origin and destination nodes; od 2 N, dd 2 N. • Release time step; rd 2 f1. . .sg • Due time step; ud 2 f1. . .sg 5.6

Example of Time-Space Graph

Following graph represents an example for 2 h 15 min period. As it was mentioned before, there will be 40 time-space nodes for 10 time step (TS) and 4 speciﬁc nodes. All nodes in ﬁrst column specify Biçerova Container Terminal and the other columns specify Nemport, T.C. Ege Gübre, and APM Terminal, respectively. Dashed lines represent stationary arcs which provide containers and trucks remain in same place through the time periods. Assume that travel time is 30 min between locations and there are two demand groups, which must be dispatched; group A and B. Node 5, which is shown as solid blue, represents release location of demand group A. Also, node 5 is the member of time step two, this means release time step of demand group A is second time step. At the same time node 18, that is shown as pattern blue, represents delivery location of demand group A. Node 18 is the member of ﬁfth time step, thus due time step of demand group A is ﬁfth time step. In the same way, release and due time steps of demand group B is ﬁfth and eighth time step, respectively. In accordance with these information, some InðiÞ and OutðiÞ sets can be deﬁned as: • • • •

Inð5Þ ¼ 1 Inð18Þ ¼ 9; 14; 11; 12 Outð5Þ ¼ 9; 14; 15; 16 Outð18Þ ¼ 25; 22; 27; 28

Inð5Þ set includes only node 1 as stationary because; time interval does not ﬁt with the travel time for node 2, 3 and 4. Inð18Þ set includes node 14 as stationary and also includes node 9, 11, and 12 for respecting travel time restriction. 5.7

Modeling

Created model routes all trucks on time-space graph while minimizing total cost. Total cost includes loaded and empty travel costs of all trucks, cost of extra rented trucks and storage cost.

Routing Optimization for Container Dispatching Operations

929

Sets and Indices N= Set of nodes in the base graph, N ¼ f1; 2; 3; 4g (More ports can be added, when it is required) s= Number of time periods NT = Set of nodes in the time-space graph, N T ¼ f1. . .s:N g T Set of arcs in the time-space graph A = D= Set of demand groups 1 d D W = Set of days 1 w W InðiÞ = Set of nodes which can enter to each node i with arcs, i 2 N T OutðiÞ = Set of nodes which can leave from each i node with arcs, i 2 N T Parameters Amountd = od = dd = rd = ud = std = cij = cfij = stcost = si = zij = Hirecost =

Number of containers in demand group d Origin node of demand group d, od 2 N Destination node of demand group d, dd 2 N Release time step of demand group d, rd 2 f1. . .sg Due time step of demand group d, ud 2 f1. . .sg Time step which gives starting of free storage time for demand group d Transportation cost between node i and node j, i 2 N T , j 2 N T Cost difference among loaded and empty travel between node i and node j, i 2 N T , j 2 N T Storage cost per time period Number of trucks present at node i at the beginning of the optimization, i 2 NT A parameter which can take the value of 0 or 1. if arc ði; j) is stationary arc, the parameter will be 0, otherwise it will be 1. Hiring cost of extra truck

Decision Variables Decision variable xij , decides that how many trucks will travel on arc ði; jÞ and, it routes trucks on time-space graph. Decision variable yijd , determines number of containers that carried on arc ði; jÞ, it also provide flow of the containers on time-space graph. (Hiredw ) A decision variable, which determines how many trucks need to be rented from outside, is deﬁned for situations such as inadequate truck fleet. Model determines the value of this variable at time step one, which is the beginning of the optimization. At the beginning of the optimization, all rented trucks must be in Biçerova. Thus, the only node, with any rented trucks at time step one, is terminal node Biçerova. In the following time periods, decision variable hij routes rented trucks on time-space graph. xij = yijd = hij = Hiredw =

Number Number Number Number

of of of of

trucks on arc ði; jÞ, ði; jÞ 2 AT container on arc ði; jÞ for demand group d; ði; jÞ 2 AT rented trucks on arc ði; jÞ, ði; jÞ 2 AT rented trucks at node, w 2 W

930

H. S. Baş et al.

Objective Function X X X X Min ðxij þ hij Þ : cij þ Hiredw : hirecost þ y : cfij þ i;j2AT w 2W i;j2AT d2D ijd X X y : stcost i;j2AS ^j\d þ ðst 1Þ : N d2D ijd d

d

ð1Þ Constraints X

z d2D ij

: yijd xij þ hij ; ði; jÞ 2 AT

X

x j2OutðiÞ ij

X j2OutðiÞ

X j2OutðiÞ

X X X

x k2InðiÞ ki

X

þ

x k2InðiÞ ki

X

hij hij

X

ð4Þ

hki 0; i 2 N T nf1g

ð5Þ

hki ¼ s1 þ Hiredw ; w 2 W

ð6Þ

X k2InðiÞ

k2InðiÞ

ð3Þ

hki Hiredw ; i ¼ 1

k2InðiÞ

X

si ; i 2 N T

¼ Amountd d 2 D; 2 N T ; i ¼ od þ ðrd 1Þ : N

ð7Þ

¼ Amountd d 2 D; j 2 N T ; j ¼ dd þ ðud 1Þ : N

ð8Þ

y j2OutðiÞ ijd

y i2InðiÞ ijd

ð2Þ

y j2OutðiÞ ijd

X

y k2InðiÞ kid

¼ 0 d 2 D; j 2 N T

i 6¼ od þ ðrd 1Þ : N and i 6¼ dd þ ðud 1Þ : N

ð9Þ

Objective function (1), minimizes the total cost as summation of traveling cost of all trucks, renting cost of extra trucks and, storage cost. Constraint (2) provides that number of containers on an arc cannot be more than number of trucks on same arc. At the same time, if the arc is stationary arc, zij takes the value 0. In this case, container can remain at same place and no truck needs to carry the container. Constraint (3) is truck flow balance constraint, which allows the trucks to travel on time-space graph. In accordance with constraint, number of trucks leaving a node cannot be more than summation of number of trucks entering the node and number of trucks that start at the node. This constraint also allows trucks to travel without container on time-space graph. Constraint (4) determines the number of rented trucks leaving from Biçerova at the start of the optimization. Constraint (5), is rented truck flow balance constraint which allows rented trucks to travel on time-space graph. Constraint (6) enforce that all trucks return to Biçerova at the end of each shift. Index i takes values, which represents end of each shift through the week in time-space graph. Constraint (7)–(9), flow the containers on time-space graph. Constraint (7) speciﬁes a starting node on time-space graph for all demand group. In other words, this constraint binds the origin node of the

Routing Optimization for Container Dispatching Operations

931

demand group. Constraint (8) enables that all demands arrive at the destination node on time. Constraint (9) is balance constraint of containers. It enforces that number of containers leaving a node must be equal to number of containers entering the node excluding origin and destination nodes for all demand groups.

6 Sensitivity Analysis Real-life data is obtained from the company to test the proposed model with ILOG IBM CPLEX Program. Table 1 shows obtained results from different scenarios based on this data. If results are examined, required number of trucks can be determined according to changing demand information. As shown in Table 1, different scenarios were created by changing number of trucks for 900 containers and 1000 containers. Prepared table includes scenario number, number of containers, number of trucks, and number of rented trucks, storage cost, cost of extra rented trucks, travel cost, total cost and solving time. Number of containers and number of trucks are foreknown information. Total cost and number of rented trucks are determined by the model. Figures 4 and 5 show percentages of all different costs in total. Table 1. Optimization results Scenario number 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Number of container 900 900 900 900 900 900 900 1000 1000 1000 1000 1000 1000 1000

Number Number of of trucks rented trucks 12 4 13 3 14 2 15 1 16 0 17 0 18 0 12 6 13 5 14 4 15 3 16 2 17 1 18 0

Storage cost (TL) 6.319 6.328 6.328 6.121 6.328 6.308 6.287 9.414 9.414 9.414 9.413 9.414 9.414 9.414

Cost of renting Travel extra trucks (TL) cost (TL) 2.714 67.184 2.036 69.890 1.357 69.890 678 69.889 0 69.211 0 69.211 0 69.211 4.072 79.761 3.393 79.760 2.714 79.760 2.036 79.761 1.357,46 79.761 678,73 79.760 0 79.761

Total cost (TL) 76.218 78.254 77.575 76.689 75.539 75.519 75.498 93.247 92.568 91.889 91.210 90.532 89.853 89.175

Solving time 00:05:18 00:06:52 00:05:30 00:07:01 00:04:11 00:04:14 00:04:07 00:07:00 00:06:51 00:06:53 00:06:38 00:05:18 00:07:13 00:05:01

Based on the results; it is clear that storage cost depends on congestion that arises because of large number of demanded containers. In peak times, model dispatches some containers earlier than free storage time to handle all demands in optimal way. Besides, our model can solve the problem for more than 1000 containers with average 5 min solving time; it shows that our model is reasonable for operations of Alkon Logistics.

932

H. S. Baş et al.

Fig. 4. Cplex result graph for 900 containers

In the current system, company operates with 15 own trucks and 17 additional trucks. It causes crucial ﬁnancial issues for company in long-term. Our model guarantees solving the problem with minimum number of trucks and total operational cost. Furthermore, 16 trucks can handle the dispatching operations for 900 containers, while 18 trucks are enough for 1000 containers. It should be considered that number of trucks required will change as demand information changes.

Fig. 5. Cplex result graph for 1000 containers

Routing Optimization for Container Dispatching Operations

933

7 Decision Support System Decision support system (DSS) provides efﬁcient usage of proposed mathematical model. DSS was created by integrating IBM ILOG CPLEX Optimization Studio 12.8 with Visual Basic for Applications (VBA). DSS allows to respond to all different scenarios that can occur. When Excel ﬁle is opened, user is redirected to Excel spreadsheet “Home”. “Home” page includes specially designed buttons that is shown in Fig. 6. First, user must select “Change Number of Trucks Belonging to Company” button to decide for the number of trucks to be used in dispatching. This change option is provided to enable that user make changes when number of trucks belonging to company change.

Fig. 6. Excel spreadsheet “Home”

Created DSS is dynamic; this means it can respond to the changes quickly. When user presses “Data Input” button, userform “Data Input” is displayed that is shown in Fig. 7. One of the signiﬁcant point is that demand information must be recorded as weekly to this form. Critical times are entered for all demand groups such as container release time, container due time and free storage period to”Data Input” form. These time information are converted into time steps on a hidden Excel page. Conversion process is generated in a hidden page for preventing users make any changes in time step information. After entering demand information is completed, data is recorded by pressing “Ok” button. Weekly dispatch planning can be done after all required demand information are entered. Created mathematical model is resolved by IBM ILOG CPLEX Optimization Studio, when “Weekly Container Dispatching Plan” button is pressed. For displaying results, user should return to “Home” page and press “View Results” button. After, user is directed to Excel sheet called “Dispatching Plan”. This page contains dispatching plan for entire week excluding Sunday and it includes different information about trucks and dispatched containers. These information are provided for trucks:

934

H. S. Baş et al.

Fig. 7. Userform “Data Input”

(i) (ii) (iii) (iv) (v)

Origin location and starting time of traveling Destination location and ﬁnishing time of travelling Number of trucks traveling Travel type such as loaded, empty or waiting Number of rented trucks

If trucks carry containers, these information are provided for dispatched containers: (i) Name of dispatched demand group (ii) Number of carried containers (iii) Demand delivery day and delivery time After displaying weekly dispatch planning, user can press “Cost Analysis” button to view costs. User is directed to “Cost Analysis” page after button is pressed. There are two tables on this page. Table 2 shows number of trucks belonging to company, number of rented trucks, number of total dispatched containers, cost of renting extra trucks, traveling cost of trucks, total storage cost and, total cost. If user Table 2. Cost table on page “cost analysis”

Routing Optimization for Container Dispatching Operations

935

Table 3. Analysis table on page “cost analysis”

presses the “Save the Current Data” button on this page, these information will be saved on Table 3. Thus, user can easily perform sensitivity analysis. User can view the cost information by updating number of trucks that is used in dispatching operation and number of dispatched containers. Therefore, user can easily analyse this table, which is composed of different scenarios. 7.1

Implementation Plan for DSS

Decision support system is user friendly and can be used to respond to different scenarios that occur in the system. User can plan weekly container dispatch by updating number of trucks. First weekly container demand must be entered by pressing the “Data Input” button. After data entry is complete, user should press “Weekly Container Dispatching Plan” button. A warning message will be displayed; this message states that Excel ﬁle will be saved automatically. After, Excel ﬁle will close and model will run. After solution is obtained, user must open Excel ﬁle to view results. Finally, user can access information on weekly container dispatch planning easily by using “View Results” and “Cost Analysis” buttons.

8 Conclusion In this paper, a solution approach is provided for solving the optimal dispatch planning problem for containers. Proposed solution approach includes several constructions that can model real-world elements such as cross-way movements, peak times and congestion. The proposed model and the DSS developed is helpful making long-term critical decisions such as purchasing of new trucks. To ensure this, a novel integer mathematical model was developed to minimize total cost. Proposed model discretizes passing time subject to hard constraints and, it provides optimal dispatch planning for containers. It prevents renting extra trucks unnecessarily by planning ahead of time and considering trade-off between storage, transportation and rental costs. The model is tested with real-life data from the company. The solution from the model provides efﬁcient usage of all trucks under strict time constraints with use of almost half of current number of trucks. Solution from the model is a 65% improvement in rental cost and 25% improvement in total cost. Proposed model has flexible structure; it means that number of shifts and their operational times can be changed. Based on

936

H. S. Baş et al.

experimentation, higher cost savings are possible by making rental agreements on a weekly basis rather than annual. To efﬁcient usage of mathematical model, also a decision support system (DSS) is generated by integrating IBM ILOG CPLEX Optimization Studio 12.8 with Visual Basic for Application (VBA). Created DSS is dynamic and practical for users. It provides to change number of trucks belonging to company and, to make analysis with provided costs tables. DSS can be used as a guidance for establishing critical long-term investments such as purchasing new trucks. Acknowledgment. This study was supported by Yaşar University Faculty of Engineering’ Department of Industrial Engineering. We also would like to thank Duygu Barış, Duygu Şentürk, Hüseyin Suçsuz and Yasin Özkabak for their input into this work. We especially want to thank to Adalet Öner, as the coordinator of Senior Design Projects this year at Yaşar University Industrial Engineering Department.

References 1. Günther HO, Kim KH (2006) Container terminal and terminal operations. OR Spectr 28:437– 445 2. Tierney K, Voß S, Stahlbock R (2013) A mathematical model of inter-terminal transportation. Eur J Oper Res 235:448–460 3. Chung K, Ko C, Shin J, Hwang H, Kim K (2007) Development of mathematical models for the container road transportation in Korean trucking industries. Comput Ind Eng 53:252–262 4. Nguyen VD, Kim KH (2010) Dispatching vehicles considering uncertain handling times at port container terminals, progress in material handling research. In: Proceedings of the 11th international material handling research colloquium, Milwaukee, WI, pp 210–226

The Distributor’s Pallet Loading Problem: A Case Study Selen Burçak Akkaya, Aykut Gül, Zeynep Coşkun, Coşku Karaman, Hande Öztop, and Gizem Mullaoğlu(&) Department of Industrial Engineering, Yaşar University, İzmir, Turkey [email protected], [email protected], [email protected], [email protected], {hande.oztop,gizem.mullaoglu}@yasar.edu.tr

Abstract. The pallet loading problem considers the loading of rectangularshaped boxes with known dimensions into the pallets. In this paper, a real-life pallet loading problem of a beverage company is studied. In this problem, the aim is to maximize the total loaded box volume on pallets by employing a threedimensional approach, where overlapping of any two boxes is not allowed and the fragility relationship is regarded. As a solution approach, a heuristic algorithm and a mathematical model have been proposed and veriﬁed with a computational experiment. A sensitivity analysis is also performed for the parameters of the problem. Furthermore, a user friendly decision support system, which is integrated with IBM ILOG CPLEX, is developed using Excel VBA interface. The proposed solution approach is embedded in this decision support system. It is estimated that proposed solution method will prevent the ﬁeld loss due to the improper loading and reduce the ﬁnancial damage, as it reduces the dependency on operators and the decision-making duration. As this problem can be seen in various companies, the proposed solution approach can also be employed by different sectors and companies. Keywords: Distributor’s pallet loading problem Three-dimensional loading Loading models with fragility constraints Heuristic algorithm Decision support system

1 Introduction The pallet loading problem (PLP) aims to determine the best pattern for loading a set of rectangular boxes with known dimensions into a pallet. We consider a real-life pallet loading problem of a beverage company in Izmir, Turkey. The aim is to maximize the total volume of the loaded boxes while ensuring that the boxes do not overlap and fragility constraints are regarded. In this study, a three-dimensional approach is implemented in order to reflect the real world conditions as much as possible. In the beverage company, products are offered to the customers in various forms of packages such as bottles, cans and barrels. Based on the order lists given by the sales department, different types of products are packed with an associated packaging material and loaded into the pallets by shipping operators. After the loading operation, © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 937–948, 2019. https://doi.org/10.1007/978-3-319-92267-6_75

938

S. B. Akkaya et al.

pallets are loaded into the trucks to be delivered customers using pre-determined routes. The sizes of the packaging materials loaded on the pallets may vary due to product type. However, the dimensions of the pallets are ﬁxed. Furthermore, the dimensions of the trucks and the sections of the trucks, in which the pallets are to be placed, are ﬁxed and known for every truck. Currently, in the company, the shipping operators act intuitively based on their experience and do not follow any standard loading procedure. This issue leads to time and ﬁnancial loss for the company. Due to the non-standard pallet loading operation, the desired pallet structure is not always obtained. For this reason, damaged products and unloaded customer orders occur. As there is a fragility relationship between different types of packaging materials, damages due to the improper placement of boxes cause company to suffer from ﬁnancial loss. The proper placement is crucial for the loading operation, as only one defective product in a pallet can affect the whole pallet. In addition, the extra shipments due to the unloaded demand, especially in summer seasons, cause additional costs to the company. As the shipment operators perform the loading operation based on a trial-and-error method without using any scientiﬁc approach, there is a low efﬁciency in terms of time. Furthermore, in case of an operator resigns, it takes 6 months for a new employee to experience and learn the pallet loading operation. This also creates high dependency on workforce. The main objective of this study is to develop a user-friendly decision support system for optimizing the pallet structure by deciding which products should be placed on which pallet in which location. In other words, the aim is to standardize and simplify the pallet loading operation by following speciﬁc steps and instructions. As mentioned above, it is crucial to determine the most suitable product placements on the pallets, in order to manage the space and use company’s resources effectively. Currently, due to the high dependency level on workforce, duration of the decision making process is not same for every operator. Therefore, introducing a scientiﬁc approach to pallet loading process is very important for performing the operation efﬁciently and reducing the loading time. As some of the product packaging materials have a risk of being broken in the wrong pallet placement, consideration of the fragility properties during the organization of pallets is also another signiﬁcant issue for this study. Consequently, the main performance measures of this study can be summarized as follows: the utilization of pallets, service level of order lists and the duration of the decision process. Therefore, the aim is to improve these performance measures, in other words, to increase the used pallet volume, reduce the amount of unloaded products on the order lists and reduce the duration of the decision making process of shipping operators. The rest of the paper is organized as follows. A comprehensive literature review on the PLP is presented in Sect. 2. A binary linear programming model is presented in Sect. 3, as well as the assumptions and limitations of the problem. An efﬁcient heuristic solution approach, which is combined with the mathematical model, is presented in Sect. 4. A user-friendly decision support system is also explained in Sect. 4. Section 5 presents the computational results, as well as the sensitivity analysis results for the parameters of the problem. Finally, concluding remarks and future work suggestions are given in Sect. 6.

The Distributor’s Pallet Loading Problem: A Case Study

939

2 Literature Review The studied problem can be seen with different names in the literature, such as “bin packing problem”, “3D bin packing problem”, “cutting stock problem”, “container loading problem” and “pallet loading problem”. There are two main variants of the pallet loading problem in the literature, namely, the manufacturer’s pallet loading problem and the distributor’s pallet loading problem. In the manufacturer’s pallet loading problem, products are packaged in identical boxes and these boxes are loaded into the identical pallets. Then, the formed pallets are loaded into the trucks that have standard dimensions. The aim of this problem is to choose dimensions of the boxes and pallets that maximize the volume of loaded products in the trucks. Hodgson (1982) deﬁnes the distributor’s pallet loading problem as follows [1]: The orders of customers are packaged in boxes with varying dimensions. The problem is to load the boxes on a standard pallet, such a way that the volume placed on each pallet is maximized, i.e., the number of pallets used is minimized. Our problem is similar to the distributor’s pallet loading problem, since order lists are pre-deﬁned, products are packaged in boxes with varying dimensions and the aim is to maximize the volume placed on each standard pallet. One of the earlier studies done within this context was published in 1982. In this study, Hodgson [1] studied the two-dimensional pallet loading problem and aimed to improve the transportation operation of US Airforce military equipment. Hodgson’s observations show that US Airforce ofﬁcials usually place the biggest box to origin point and place the rest of the boxes around this box. It is revealed that this intuitive method dramatically decreases the computational time. In another study, Kang and Park [2] studied the variable size box placement problem and aimed to minimize the total cost of used bins under the assumption that the cost of unit size of each bin does not increase as the bin size increases. This study focuses on variable sized bins and some heuristics were applied for the problem, such as ﬁrst ﬁt decreasing, best ﬁt decreasing, iterative ﬁrst ﬁt decreasing and iterative best ﬁt decreasing. In our heuristic algorithm, we inspire from some of these rules. Later, Lel et al. [3] proposed a heuristic algorithm to pallet packing problem of a beverage manufacturer. The objective is to determine the loading sequence of products with boxes of different sizes and the number of pallets required for the placement. In their algorithm, initially, products with similar cube size are grouped, and full pallet and partial pallet assignments are made in order to decrease the placement combinations. Results of their numerical analysis showed that the proposed algorithm solves the pallet loading problem efﬁciently within a reasonable computational time. Junqueira et al. [4] published an article about three-dimensional pallet loading models with consideration of cargo stability and load bearing constraints. This article focuses on box orientation, complete shipment of box groups, box priorities, complexity of arrangement, fragility constraints, and weight distributions. In this study, an optimization model is proposed as a solution methodology for minimizing the empty volume percentage on the pallets and the duration of the loading time of the pallets. Recently, Sheng et al. studied a container loading problem with the consideration of expiration of the products [5]. In this problem, all products on a given order must be

940

S. B. Akkaya et al.

placed in one container. The aim is to maximize the utilization of volume on the container. A heuristic algorithm is proposed to standardize the packing process. The literature review is summarized in Table 1. In the table, each article is classiﬁed according to studied problem, objective function and proposed solution method. Table 1. Literature review Authors Thom J. Hodgson

Date 1982

Park and Kang

2002

Vu et al.

2005

Morabito et al.

2010

Sheng et al.

2017

Problem Two-dimensional pallet loading problem Variable size box placement

Three-dimensional pallet loading problem Three dimensional container loading with load balance, transport and fragility constraints Container loading with multiple constraints

Objective function Maximize pallet occupancy Minimize the total cost of boxes used in variable sizes

Maximize the volume of products in the pallet Maximize the total volume of loaded boxes or minimize remaining free space Maximize the volume of loaded boxes

Solution approach Dynamic programming and heuristic approach Greedy algorithm and loading rules (best ﬁt decreasing, ﬁrst ﬁt decreasing etc.) Heuristic approach

Mixed integer programming model and heuristic approach Integer programming model and heuristic approach

The study of Junqueira et al. [4] can be used as a reference as it contains a threedimensional approach to pallet loading problem and proposes an optimal placement policy. Furthermore, this study integrates fragility issue into the model. However, their model considers the placement of boxes only on a single pallet. Therefore, we extend their model by changing the restrictions on possible location sets and adapting the multiple pallets.

3 Problem Formulation In this section the mathematical model is presented for the aforementioned problem. Due to the complexity of the problem, below assumptions are made. • All values are assumed to be integers. • Boxes can only be placed orthogonally, i.e., either parallel or perpendicular, into the pallet. • Orientation of the boxes is ﬁxed.

The Distributor’s Pallet Loading Problem: A Case Study

941

• Each box could be moved down and/or forward and/or to the left, until its bottom, front and left-hand faces are adjacent to other boxes. • All boxes and the pallets are assumed to be rectangular prisms. The parameters and decision variables are deﬁned as follows. J = {1, …, n} is the set of pallets and I = {1, …, m} is the set of product box types. Each product box type i 2 I: length li , width wi , height hi , volume vi and a maximum quantity bi values are deﬁned. These boxes can be loaded into a standard pallet with length L, width W and maximum allowed height H. To ﬁnd the possible positions of the boxes, a Cartesian coordinate system is adopted, where the origin point (0, 0, 0) represents the front-leftbottom corner of the pallet. As in Junqueira et al. [4] sets of possible locations (x, y, z) are deﬁned as follows: X ¼ f0; 1; 2; . . .; L mini ðli Þg; Y ¼ f0; 1; 2; . . .; W mini ðwi Þg; Z ¼ f0; 1; 2; . . .; H mini ðhi Þg; Two binary decision variables aijxyz and pj are deﬁned as follows, to decide which box will be placed on which pallet at which position and which of the pallets will be used, respectively:

aijxyz ¼

8 > > <

1; if a box i is placed on pallet j with its front - left - bottom corner at positionðx,y,zÞ 8i 2 I; x; x0 2 X; y; y0 2 Y; z; z0 2 Z so that 0 x L li ; 0 y W wi ; 0 z H hi > > : 0; otherwise

pj ¼

1; if pallet j is used ;j 2 J 0; otherwise

The proposed solution approach solves the pallet loading problem with the consideration of the fragility relationship of the boxes. However, this issue is handled within the heuristic algorithm, which is explained in Sect. 4, before the model solution step. Therefore, it is not included in the model as a constraint. In the pallet loading problem of the beverage company, due to the seasonality of the demand, the objective function changes according to the relationship between customer demand and supply, where supply refers to the capacity of trucks. During the summer seasons, the demand usually increases due to the increase in beverage consumption. As the demand is greater than the supply, the objective function is to maximize the volume used on pallets. On contrary, the demand usually decreases in winter seasons due to the decrease in beverage consumption. As supply exceeds the demand in this case, the objective function is to minimize the number of pallets used. The optimization model is adapted from the study of Junqueira et al. [4] by introducing a set for pallets, a new binary decision variable for determination of which

942

S. B. Akkaya et al.

pallets are used and adding pallet information to the current decision variables. An additional constraint is also deﬁned in order to link these two types of decision variables. The proposed mathematical model is presented below, for the case of demand exceeds the supply amount: max

Xn Xm X j¼1

i¼1

X

x2X

X y2Y

X

z2Z

aijxyz vi

ð1Þ

8i

ð2Þ

Subject to Xn X j¼1

Xm X i¼1

x2Xjx0 li x x0

;

X

x2X

X y2Yjy0 wi y y0

Xm X i¼1

X y2Y

;

aijxyz bi

X

0

z2Zjz0 hi z z0

X x2X

z2Z

X

y2Y

z2Z

0

0

; aijxyz 1 8j; x 2 X; y 2 Y; z 2 Z

aijxyz Mpj

8j

ð3Þ ð4Þ

aijxyz 2 f0; 1g; i 2 I; j 2 J; x 2 X; y 2 Y; z 2 Z

ð5Þ

pj 2 f0; 1g; j 2 J

ð6Þ

The objective function (1) aims to maximize the volume used on the pallets, in this case. Constraint (2) restricts the maximum number of packed boxes and constraint (3) prevents overlapping of two boxes. Constraint (4) allows the placement of boxes on a pallet only if that pallet is being used, where M is a large integer. Constraints (5) and (6) deﬁne the domain of the decision variables. For the case of supply exceeds demand, only the objective function and constraint (2) changes as follows: min

Xn j¼1

ð7Þ

pj

Subject to Xn X j¼1

x2X

X

X y2Y

z2Z

aijxyz ¼ bi ; 8i

ð8Þ

(3)–(6) The objective function (7) aims to minimize the number of used pallets. Constraint (8) ensures that all products on the order list are fully loaded, as there is enough volume on the pallets in this case.

4 Solution Approach and Implementation Since the studied problem is NP-hard [6], a heuristic solution approach is employed in the proposed solution procedure. The details of the heuristic algorithm is shown in Fig. 1.

The Distributor’s Pallet Loading Problem: A Case Study

943

Fig. 1. Algorithm flow chart

Algorithm starts with reading the order list. Firstly, if the order amount of one product type is enough to fulﬁll a pallet, the same products are placed on the pallet and a full pallet is formed. The full pallets are preferred by the company. In the case of determining the pallet loading structure of the remaining products, a three-dimensional pallet loading optimization model given in Sect. 3 is used. Before listing the products to be used in the model, case type of products are controlled and placed to the pallets in the form of identical blocks. The main reason of this block system is to satisfy the fragility constraint, as it is not allowed to place other types of products above/under these case type of products. The height of generated blocks is assumed to be equal to the height of the pallets so that any placement of other packaging material types above or under these blocks is avoided. Block system is not used for remaining products as placement above or under them is allowed according to fragility relationships. After this step, the total volume of the remaining products is calculated and compared with the remaining volume of the pallets that can be used for placement. If the demand exceeds supply, the objective function that maximizes the utilization rate of each pallet is used in the model. In the other case, objective function that minimizes the number of used pallets is used. Finally, the mathematical model is solved under 5 min time limit for on the pre-determined objective by using IBM ILOG CPLEX Optimization Software. A time limit is set for the mathematical model, as this pallet loading operation is performed daily in the company. Consequently, we develop a user-friendly decision support system (DSS) which uses the above solution methodology. This DSS is developed using the Excel VBA interface and connected with the IBM ILOG CPLEX Optimization Studio, in order to obtain results in short time. At the main screen of the program, there are two main parts namely “Inputs” and “Outputs”, as shown in Fig. 2. In “Inputs” part, users can automatically extract an existing order list in Excel format or create a new order list manually. In “Outputs” part, detailed information of loaded products and loading instructions are reported for each pallet, after running the proposed algorithm. This application is designed for enabling users to use the program dynamically. After getting the data from the user, the pallet structure is obtained employing the aforementioned algorithm shown in Fig. 1. The usages of each pallet are summarized

944

S. B. Akkaya et al.

in a report on the main page. Users are able to add or remove products; create a pallet loading scheme; update the information of pallets or products in database; and report the detailed pallet loading information and loading instructions. Moreover, users are able to print the results and loading instructions, send them by e-mail to other users and export the summary in Excel or PDF format as in Fig. 3. Finally, users can access user’s guide so that they can easily use the DSS (Fig. 2).

Fig. 2. DSS main screen

Fig. 3. DSS output – loading report

The Distributor’s Pallet Loading Problem: A Case Study

945

5 Computational Results The computational results of the proposed solution methodology are presented in this section. 5.1

Comparison and Improvements

In order to validate the accuracy of the solution methodology, 15 different order lists for each objective function case are obtained from the company and solved by the suggested method. It is observed that the results are as desired by the company and the model responds quickly and effectively to both objective functions. Furthermore, results of the proposed solution methodology are compared with the results of the current system used by the company. The computational results are shown in Table 2 for the minimization problem instances, and in Table 3 for the maximization problem instances. For each instance, number of product types and the total number of products are listed. Current method and proposed method are also compared for each instance and improvements are reported in terms of number of used pallets, average pallet utilization and unloaded amount of products. According to Table 2, improvements are observed in 9 out of 15 instances. On the average, a rate of %10 improvement is obtained in terms of number of pallets used. In addition, results show that the proposed solution method gives better results than the current method in terms of performance measures mentioned in Sect. 1. According to Table 2. Minimization model comparison results Minimization problem instances Order information No # of product Total # of type prod. 1 17 227 2 17 182 3 16 208 4 11 343 5 14 368 6 10 377 7 11 237 8 15 350 9 11 219 10 16 93 11 9 366 12 10 246 13 10 470 14 10 449 15 11 271 Average

# of used pallets Current Proposed method method 5 4 4 4 3 3 7 6 8 7 7 7 6 5 7 6 5 4 2 2 6 7 7 6 8 7 8 7 5 4 6 5

Improvement (%) 20 0 0 14 13 0 17 14 20 0 0 14 13 13 20 10

946

S. B. Akkaya et al. Table 3. Maximization model comparison results

Maximization problem instances Order information No # of Total # product of prod. type

Avg. pallet utilization Unloaded amount Current Proposed Imp. (%) Current Proposed method method (%) method method (%) (Amount) (Amount)

1 15 2 15 3 11 4 10 5 13 6 13 7 15 8 10 9 12 10 11 11 14 12 16 13 13 14 13 15 15 Average

81 90 91 88 99 86 79 90 94 95 76 93 86 74 80 87

577 554 645 697 600 689 740 767 780 620 819 815 758 748 394

91 90 100 95 98 97 94 94 89 91 91 93 94 90 99 94

13 0 9 8 0 12 18 4 0 0 20 0 10 22 23 9

88 0 113 104 18 69 90 54 21 0 108 92 74 61 51 63

0 0 54 56 40 39 45 47 46 0 71 97 42 33 34 40

Imp. (%)

100 0 52 46 0 43 50 13 0 0 34 0 43 46 33 31

the observations, the proposed algorithm yields results in maximum 5 min while the current decision duration of the operators is approximately 10 min. In this regard, the decision-making period is improved considerably. According to the results of the maximization model instances, a signiﬁcant decrease in the number of unloaded products is observed with the increase in the average utilization rate of pallets. While the average utilization of the pallets is improved by 9%, the improvement rate of the unloaded amount is 31%. Similarly, the decisionmaking period is also improved considerably by the proposed solution method, as in minimization instances. 5.2

Sensitivity Analysis

A sensitivity analysis is also conducted to observe how the proposed solution method reacts to different scenarios. Two examples of maximization and minimization cases are solved for 8 different settings using real data. In these settings, one of the dimensions of the pallet is increased by 10 or 20 cm, which the company allows it as overflow proportion. The results are shown in Table 4. Results for minimization cases demonstrate that, in the case of dimensions of the pallets are increased, the number of pallets used decrease for most of the situations, as expected. If the height of the pallet is increased while keeping the floor area constant, the algorithm gives better solutions in minimization cases. Moreover, it is observed that the height and the length values are restrictive for the minimization objective function, whereas the width value does not affect the objective function value for most of the cases.

The Distributor’s Pallet Loading Problem: A Case Study

947

Table 4. Sensitivity analysis results No Minimization # of total Dimensions products (10 cm) 1

184

2

205

12,10,18* 12,11,18 12,10,19 13,10,18 13,10,19 12,11,19 13,11,18 13,11,19 12,10,18* 12,11,18 12,10,19 13,10,18 13,10,19 12,11,19 13,11,18 13,11,19

# of used pallets 3 3 3 3 2 3 3 2 7 7 6 6 6 6 6 6

Maximization # of total Dimensions products (10 cm) 739

620

12,10,18* 12,11,18 12,10,19 13,10,18 13,10,19 12,11,19 13,11,18 13,11,19 12,10,18* 12,11,18 12,10,20 13,10,18 13,10,20 12,11,20 13,11,18 13,11,20

Avg. Unloaded utilization (%) amount (%) 97 97 99 98 100 99 97 100 91 91 100 91 100 100 91 100

29 31 21 22 20 28 22 20 10 10 0 10 0 0 10 0

*Base case

For maximization cases, in the case of dimensions of the pallets are increased by 10 cm, it is concluded that the number of unloaded orders is reduced and the average pallet utilization rates are increased. On the other hand, for the second order list shown in Table 4, no improvement is observed in the number of unloaded products when the pallet dimensions are increased by 10 cm. The main reason of this situation is that adjusted dimensions do not allow any placement of new products, since dimensions of these products are larger than 10 cm. However, in the case of the dimensions are increased by 20 cm, improvements are observed and the order list is fully placed. However, since the altitude limit is legally restricted in national highways, this change is not possible for height value but it may be considered as a recommendation for other dimensions.

6 Conclusions The aim of this study is to create a pallet loading system that increases the efﬁciency, removes the dependency on the workforce and reduces the decision-making time and additional costs. We consider a real-life problem of a beverage company, in which operators perform the pallet loading operation based on their experience level, without any scientiﬁc method. In this study, a mathematical model, a heuristic algorithm and a convenient decision support system, which makes the pallet placement with a scientiﬁc approach, are developed for this problem. The output of the developed user-friendly decision support system is a summary report that shows the locations of the products on

948

S. B. Akkaya et al.

the pallets as well as the utilization rates of pallets. Computational results indicate that the proposed solution approach performs much better than the current system, as described in Sect. 5. By applying the proposed solution methodology, the used volume of pallets has been increased in the direction of the observed instances, while the amount of unloaded products, the dependency on the work power and the duration of the decision making period have been reduced. Note that a %10 improvement is obtained in terms of number of pallets used, while the average utilization of the pallets is improved by 9% and the rate of unloaded amount is improved by 31%. As a future study, the proposed algorithm and model can be extended by allowing rotation of the boxes along x-y axis. In this manner, additional improvements can be observed in the pallet structures. Furthermore, additional constraints such as load bearing and stability can be adapted into proposed solution methodology. Acknowledgment. This work cannot be completed without the assistance of Atakan Yurttutan, Serhat Özbıçakçı and Merve Dirik. We are thankful for their contribution. Furthermore, we are grateful to the company for their co-operation.

References 1. Hodgson TJ (1982) A combined approach to the pallet loading problem. IIE Trans 14(3):175– 182 2. Kang J, Park S (2003) Algorithms for the variable sized bin packing problem. Eur J Oper Res 147(2): 365–372 3. Lel VT, Creighton D, Nahavandi S (2005) A heuristic algorithm for carton to pallet loading problem. In: 3rd IEEE International Conference on Industrial Informatics, INDIN 2005, pp 593–598 4. Junqueira L, Morabito R, Yamashita DS (2012) Three-dimensional container loading models with cargo stability and load bearing constraints. Comput Oper Res 39(1):74–85 5. Sheng L, Xiuqin S, Changjian C, Hongxia Z, Dayong S, Feiyue W (2017) Heuristic algorithm for the container loading problem with multiple constraints. Comput Ind Eng 108:149–164 6. Garey MR, Johnson DS (1979) Computers and intractability. W.H. Freeman, New York

Three Dimensional Cutting Stock Problem in Mattress Production: A Case Study Selin Altın, Tezcan Aydilek, Umut Şirvan, Damla Kesikburun, Adalet Öner(&), and Nejat Kutup Department of Industrial Engineering, Yaşar University, İzmir, Turkey [email protected], [email protected], [email protected], {damla.kesikburun, adalet.oner,nejat.kutup}@yasar.edu.tr

Abstract. This study involves in a real-world application of 3-D cutting stock problem in a local bedding company. The company produces various types of mattresses (beds). The bill of materials of a mattress usually includes foam rubber (sponge) components. Each mattress has different foam rubber components in the form of orthogonal rectangular prisms with different dimensions. Those components should be cut from big foam rubber blocks. The properties of current cutting machinery used in the company imposes “guillotine” cuts. The aim is to minimize total waste in cutting operations. There are mathematical models to solve cutting stock problems, however they require “cutting patterns” to be generated in advance. Cutting patterns represent potential combinations of how rectangular prisms are ﬁtted into foam rubber blocks. Column Generation method and a Branch and Bound algorithm are used to generate cutting patterns. Mathematical model then solves the problem provided that the cutting patterns are supplied as input. A user friendly Decision Support System (DSS) has been developed in order to incorporate proposed procedures to solve the problem. It enables the user in the company to prepare efﬁcient cutting plans easily and quickly. Keywords: Optimization Cutting stock problem Guillotine cut Cutting patterns Column generation Branch and bound Decision Support System

1 Introduction Foam rubber is used as the cushioning material in mattresses. In the production of mattresses, big foam rubber blocks should be cut into different types of orthogonal rectangular prisms that are the components of a mattress such as head bars, side bars, and comfort layers as shown in Fig. 1. In this form, the problem is a standard threedimensional cutting stock problem (3DCSP). A local bedding company in Izmir produces mattress and suffers a high level of waste in foam rubber cutting operations. Currently, there isn’t any systematic or scientiﬁc approach for planning of cutting operations. A planning engineer develops cutting plans manually, depending on his/her own judgment and experience. Company © Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 949–960, 2019. https://doi.org/10.1007/978-3-319-92267-6_76

950

S. Altın et al.

Fig. 1. Foam rubber components of a mattress

management is not happy with the current level of waste. The purpose of this study is to investigate and compile techniques and methods of three-dimensional cutting stock problem and then develop a planning tool that will be used to generate efﬁcient cutting plans. Therefore, the focus of the study is an industrial application of three-dimensional cutting stock problem.

2 Problem Deﬁnition The standard 3-D cutting stock problem (3DCSP) can be deﬁned simply as follows: There is an unlimited quantity of identical big foam rubber block B = (L, W, H) as raw material in producing mattress, where L, W, and H deﬁne the length, width and height of the blocks respectively. On the other hand, there is a set of n types of components or items (l, w, h, d) to be cut from big blocks B. The problem is to determine how to cut the smallest possible number of blocks B so as to produce di units of each items type i, i = 1,…, n. A basic mathematical model can be deﬁned for the standard 3DCSP as follows. Decision Variables xj: Number of Stock material to cut according to pattern j Cutting patterns represent potential combinations of how components (items) are ﬁtted into foam rubber blocks B. Parameters n = Number of items m = Number of patterns i = index for item, i = 1, …. n j = index for pattern, j = 1, …. m pij = Number of occurances of the ith item in the jth pattern di = Demand for item i

Three Dimensional Cutting Stock Problem in Mattress Production

951

Mathematical Model

min Z ¼

m X

xj

ð1Þ

j

Subject to m X

pij xj di

8 i 2 f1; ::ng

ð2Þ

j

xj 2 Z þ

8 j 2 f1; ::mg

ð3Þ

The objective function minimizes total number of cutting patterns to be used. Constraint (2) ensures that the items are produced in desired amounts. Constraint (3) indicates that an integer solution is requested. It is obvious that cutting patterns should be determined in advance as the input of the problem. The hard part of the problem is to generate valid patterns. A cutting pattern shows how many items of each type are cut from stock material. An example of generating cutting patterns are explained below in Figs. 2 and 3. For simplicity, the idea is shown on two dimensions. However, it can be extended easily for three dimensions. Assume that there are four types of items and they should be cut from stock material as shown in Fig. 2.

Fig. 2. Example case for generating cutting patterns

The items can be placed on the stock material in different combinations, and each combination represents a pattern. Figure 3 shows that one of the alternative patterns such that ﬁve instances of item 1, one instance of item 2 and two instances of item 3 could be placed on the stock material. This placement is just one of many possible combinations. In order to solve the problem optimally, all valid pattern should be determined and listed.

952

S. Altın et al.

Fig. 3. Generating a cutting pattern

There are many variants and extensions of standard 3DCSP depending on the environment and technological requirements. In this study, we consider the real-world requirements in the bedding company, which can be described as follows: • Technical speciﬁcations and settings of blades of every cutting machine used in the company impose guillotine cuts. A guillotine cut is a cut that is parallel to one of the sides of the object and goes from one side to the opposite one. • There are two types of cutting machines, and each type is set to cut along different Cartesian planes. On one of the machine type, the block B is loaded on the machine and the machine cuts “layers” of foam rubbers parallel to x-y plane that is called “horizontal” cut. The orientation of the block B on the machine is ﬁxed and can’t be changed. The other type of machine cuts parallel to y-z plane, and it is called “vertical” cut. In this type of machine, the orientation of the objects may be changed as desired. The cutting process is sequential such that any block B is ﬁrst cut horizontally and generated layers are then cut vertically as many times as required to get components to be used in mattress production. In such a case, cutting process is called k-staged cutting in literature [5]. Guillotine cuts are applied in both stages. The requirements described above leads to a special variant of 3DCSP. The problem turns out to be a 2-staged cutting problem constrained with guillotine cuts. This study focuses on this variant of the problem, and the aim is to develop a solution procedure and then embed it in a decision support tool.

3 Literature Review In literature, the ﬁrst known deﬁnition and formulation of the cutting stock problem (CSP) were made by Russian economist Kantorovich [1]. Gilmore and Gomory [2] proposed column generation method in order to solve single dimension CSP. Generating a column means generating a cutting pattern. Then, they have extended their study to describe how to use column generation method in order to solve multistage CSP of two and more dimensions [3, 4]. However, the proposed method doesn’t provide integer solution. Queiroz et al. [5] provide an extensive survey on the 3DCSP.

Three Dimensional Cutting Stock Problem in Mattress Production

953

There are various studies in literature that can be used in order to formulate and solve 2-staged 3DCSP. Among others, some mathematical models are the one-cut concept by Dyckoff [6], the arc-flow concept by Valerio de Carvalho [7], the DP-flow concept by Cambazard and O’Sullivan [8], and the general arc-flow with graph compression by Brandao and Pedroso [9]. There are also exact methods proposed for CSP depending on the branch & bound and branch & cut algorithms such as Hadjiconstantinou and Christoﬁdes [10] and Martello and Toth [11]. In meta-heuristics realm, Kampke [12] and Alvim et al. [13] implemented simulated annealing a tabu search methods respectively.

4 Modelling and Solution Methodology This study focuses on 2-staged 3DCSP with guillotine cuts as described in “Problem Deﬁnition” section. Technological requirements lead us to divide the 3D problem into two consecutive stages. The ﬁrst stage contains a collection of two-dimensional cutting stock problems (2DCSP), and the second stage contains a one-dimensional cutting stock problem (1DCSP). However, these two stages are not independent of each other, since the output of the ﬁrst stage (2DCSP) is used as the input in the second stage (1DCSP). Figure 4 shows the relations between the stages.

Fig. 4. Problem setting of 2-Staged 3DCSP

In modeling and solution process, the sequence of stages is reversed when compared to the sequence of cutting process in practice. The ﬁrst stage of the solution procedure corresponds to the second stage of cutting process in practice, which is the vertical cutting stage.

954

S. Altın et al.

In this problem setting, ﬁrst, the set of items (components) to be cut are grouped by their heights. The items with the same heights will be in the same group. For each group of items, a 2-D CSP is solved. It leads more than one 2-D problems such that each one corresponds to a particular height. The 2-D problems are independent of each other since they correspond to distinct heights. The solution of each 2-D problem gives the minimum number of “layers” to be cut while ensuring required numbers of items would be produced in a given height. The solutions of those 2-D problems also provide the information of how the items are located on the layers. That information is actually the cutting plan for the machines that performs vertical cuts. (See Fig. 4.) The output of independent 2-D problems are collected to make a list of layers along with their heights. The layers in that list are to be cut from blocks B. The list of layers is used as the input of the second stage which is a 1DCSP. The solution of the 1-D problem gives the minimum number of blocks B to be used. The solution also provides the information of how the layers are located in block B. That information corresponds to the cutting plan for the machines that performs horizontal cuts. For the solution procedure, it has been decided to use the model and formulation given in (1)–(3) and it is coded in Lingo 17.0. However, it was required to ﬁnd a systematic way to determine valid cutting patterns. As mentioned earlier, Gilmore and Gomory [2–4] proposed a column (pattern) generation method. In column generation method, the problem is divided into two problems: master problem and the auxiliary problem. First, a set of feasible cutting patterns are found and placed in the simplex method as the starting feasible solution. Then a special procedure is applied to ﬁnd a promising pattern if any. That procedure creates the auxiliary problem which is a knapsack problem whose input is the shadow prices from current basis in the simplex method. The solution of the auxiliary problem is then fed to master problem for the next iteration of simplex method. Column generation method is easy to implement for 1DCSP but not for multidimensional problems. Therefore, other relevant pattern generation methods have been reviewed in the literature. Malaguti et al. [14] present an extensive survey for such methods. Pattern generation methods are usually based on the branch and bound algorithm. Among many others, the methods proposed by Suliman [15] and Rodrigo et al. [16] have been scrutinized, and the latter has been chosen to implement in solution procedure because it is easy to implement and embed into decision support tool. The selected pattern generation methods which are column generation method and the algorithm of Rodrigo et al. [16] are coded in Lingo 17.0 and Excel VBA, respectively. The proposed solution procedure is described in Fig. 5. A sample problem is prepared below to show that how the proposed solution procedure works. It is assumed that the dimensions of the foam rubber blocks are 200 180 100 cm that corresponds to length, width and height (L, W, H) respectively. Assume also that production plan is received and it is ordered to produce seven different mattress types. The types and order amounts of mattresses are shown in Table 1. The BOM data indicates the number and dimensions of the components for each mattress type. Usually, each mattress must have two comfort layers, two head bars, and two side bars (see Fig. 1). Using BOM data, the dimensions and numbers of all the components (items) may be listed easily. For simplicity, comfort layers component were excluded from the list, therefore only head and side bars were considered. The list

Three Dimensional Cutting Stock Problem in Mattress Production

955

Fig. 5. Proposed solution procedure Table 1. Sample problem – production orders Mattress type Demand A 15 B 18 C 21 D 14 E 13 F 12 G 20

of the items are then divided into sub-lists depending on their heights. Each sub-list contains the items with the same heights. The production plan and BOM data lead us to have three item sub-lists shown in Tables 2, 3 and 4. Each sub-list is the input for separate and independent 2DCSPs. The lengths and widths of items are considered in those two-dimensional problems whereas the heights are ignored. The branch and bound method in Rodrigo [16] has been used to generate the cutting patterns and then cutting patterns have been used to ﬁnd the optimal solution in model deﬁned in (1)–(3). Tables 5, 6 and 7 presents the output of the two-dimensional problems.

956

S. Altın et al. Table 2. Item sub-list 1 (All heights = 8 cm) Mattress type Component A Head bar Side bar B Head bar Side bar C Head bar Side bar

Length (cm) Width (cm) Height (cm) Demand 130 16 8 30 170 16 8 30 150 14 8 36 190 14 8 36 190 20 8 42 200 20 8 42

Table 3. Item sub-list 2 (All heights = 13 cm) Mattress type Component D Head bar Side bar E Head bar Side bar

Length (cm) Width (cm) Height (cm) Demand 130 16 13 28 170 16 13 28 190 20 13 26 200 20 13 26

Table 4. Item sub-list 3 (All heights = 17 cm) Mattress type Component F Head bar Side bar G Head bar Side bar

Length (cm) Width (cm) Height (cm) Demand 130 16 17 24 170 16 17 24 150 14 17 40 190 14 17 40

Table 5. Output of two dimensional problem for sub-list 1 (All heights = 8 cm)

Three Dimensional Cutting Stock Problem in Mattress Production

957

Table 6. Output of two dimensional problem for sub-list 2 (All heights = 13 cm)

Table 7. Output of two dimensional problem for sub-list 3 (All heights = 17 cm)

Once two-dimensional problems are solved, the output speciﬁes selected cutting patterns. Those patterns have merged into a cutting plan for the machines that perform vertical cuts. The output also provides an information on the total numbers of patterns to be used. Then, the total numbers of patterns along with their respective heights have fed into the problem in the second stage, which is a 1DCSP. Tables 8 and 9 show the input and output of the 1DCSP, respectively. For the sample problem, the waste level is calculated as 6.25%. The solution procedure has been tested for 5 different problems in local bedding company. As a result of solving these problems, reduction in waste levels are shown in the Table 10. It has been observed that the solution time of test problems vary between 1 and 2 min.

Table 8. Input of one dimensional problem Height 8 13 17

Demand 19 10 10

958

S. Altın et al. Table 9. Output of one dimensional problem

Table 10. Reduction in waste level for test problems Test problem Reduction in waste level 1 2.79% 2 2.61% 3 4.16% 4 3.28% 5 1.42%

5 Decision Support System The solution procedure has been implemented in a user friendly decision support tool. The tool enables the user to develop the cutting plans easily and efﬁciently. The interface of the tool is shown in Fig. 6. The interface incorporates three main sections. The ﬁrst section is related to managing the production plan which contains the production orders of the mattresses. In this section, it is possible to list, delete, add or edit the entries in the production plan. The order management screen is shown in Fig. 7. It helps the user to have the updated version of the production plan. The second section takes the entries in production plan and process them with BOM data in order to generate item list to be cut and then solves the problem. At this point, the user is

Fig. 6. Main interface of planning tool

Three Dimensional Cutting Stock Problem in Mattress Production

959

informed about the stages of the solution. Finally, the third section allows the user to see and review the cutting plans. The outcome can be displayed in different format and in different detail levels. It is also possible to print the outcome in selected format.

Fig. 7. Management of the production plan

6 Conclusion and Future Work In this study, 3DCSP has been solved in two consecutive stages, since the block orientation is not allowed in the company. The ﬁrst stage contains a collection of twodimensional cutting stock problems (2DCSP), and the second stage contains a onedimensional cutting stock problem (1DCSP). The important point in this study is that the output of the ﬁrst stage (2DCSP) is used as the input in the second stage (1DCSP). According to the results of 5 sample data, the decrease in waste level is between 1.4% and 4.2%. A flexible and user friendly tool has been developed for the company to enable the planning engineer to prepare quick and efﬁcient cutting plans. It has been a useful tool for the company since they don’t have such a tool before this study. It manages the input of the problem and present the solutions to the user through a user-friendly interface. The solution process and outcomes has been veriﬁed and validated. An observable decrease (in between 1.4%–4.2%) in waste level has been reported. On the other hand, the company now have a means of data collection and history of cutting process. An important point can be studied in the future, if the orientation of the blocks B can be changed. Because, more decrease in waste levels may be attained by using some other solution methods, once the orientation of the blocks is allowed within the company. Acknowledgment. This study is supported by TUBITAK (The Scientiﬁc and Technological Research Council of Turkey) in the program of “2209-B Undergraduate Thesis Support Program for Industrial Applications”. We would like to thank our colleagues Buğra Akboy, Tevﬁk Ercan, Simay Karaşahin and Yusuf Rıdvanoğulları for their help and contribution to the study.

960

S. Altın et al.

References 1. Kantorovich LV (1960) Mathematical methods of organizing and planning production. Manag Sci 6:366–422 2. Gilmore PC, Gomory RE (1961) A linear programming approach to the cutting stock problem. Oper Res 9:849–859 3. Gilmore PC, Gomory RE (1963) A linear Programming approach to the cutting stock problem: Part II. Oper Res 11:863–888 4. Gilmore PC, Gomory RE (1965) Multistage cutting stock problems of two and more dimensions. Oper Res 13:94–120 5. Queiroz TA, Miyazawa FK, Wakabayashi Y, Xavier EC (2012) Algorithms for 3D guillotine cutting problems: unbounded knapsack, cutting stock and strip packing. Comput Oper Res 39:200–212 6. Dyckoff H (1981) A new linear programming approach to the cutting stock problem. Oper Res 29:1092–1104 7. Val´erio de Carvalho JMV (1999) Exact solution of bin packing problems using column generation and branch and bound. Annal Oper Res 86:629–659 8. Cambazard H, O’Sullivan B (2010) Propagating the bin packing constraint using linear programming. In: Principles and practice of constraint programming – CP 2010. LNCS, vol 6308, pp 129–136 9. Brandao F, Pedroso JP (2016) Bin packing and related problems: general arc-flow formulation with graph compression. Comput Oper Res 69:56–67 10. Hadjiconstantinou E, Christoﬁdes N (1995) An exact algorithm for general, orthogonal, twodimensional knapsack problems. Eur J Oper Res 83:39–56 11. Martello S, Toth P (1990) Knapsack problems: algorithms and computer implementations. Wiley, Chichester 12. Kämpke T (1988) Simulated annealing: use of a new tool in bin packing. Annal Oper Res 16:327–332 13. Alvim ACF, Ribeiro CC, Glover F, Aloise DJ (2004) A hybrid improvement heuristic for the one-dimensional bin packing problem. J Heuristics 10:205–229 14. Malaguti E, Durán RM, Toth P (2014) Approaches to real world two-dimensional cutting problems. Int J Manag Sci 47:99–115 15. Suliman MA (2001) Pattern generating procedure for the cutting stock problem. Int J Prod Econ 74:293–301 16. Rodrigo WNP, Daundasekera WB, Perera AAI (2012) Pattern generation for twodimensional cutting stock problem. Int J Math Trends Technol 3:54–62

Author Index

A Acar, Esin, 872 Adamczak, Stanisław, 208, 218, 232, 333, 747 Akgün, Ali, 705 Akgün, Gizem, 849 Akgün, V. Özlem, 705 Akkaya, Selen Burçak, 937 Akkoyun, Fatih, 265, 280 Aksezer, Çağlar, 75 Aldemir, Gökhan, 171 Algül, Duygu, 150 Altın, Ege Naz, 839 Altın, Selin, 949 Altunan, Büşra, 29 Arslan, Ebru D., 29 Arslandere, Murat, 515, 530 Atalay, Kumru Didem, 604, 737 Ay, Eren, 546 Aydilek, Tezcan, 949 B Balaban, Semih, 839 Baş, Gökçen, 391, 438, 464 Baş, Hasibe Serap, 922 Bauer, Andreas, 438 Bauer, Jorge M., 239 Baysal, Merve Vildan, 464 Beldek, Tuğçe, 181 Bereketli Zafeirakopoulos, İlke, 590 Birer, Merve, 29 Birgün, Semra, 645 Blecha, Petr, 402 Bodur, Osman, 239 Böğrekci, İsmail, 265, 280 Bolat, Bersam, 781

Bolat, Hür Bersam, 412 Boran, Semra, 800 Boyacı, Tahir Hakan, 324 Buhmann, Marco, 451 Bulan, Mert, 827 Bulut, Önder, 839 C Camgöz-Akdağ, Hatice, 171, 181 Çamlıca, Merve, 914 Çavdarlı, Ali İhsan, 426 Cebeci, Özay, 426 Celep, Emel, 581 Cesur, Turan Can, 839 Cetin, Nihan, 136 Christoph, Ralf, 693 Christoph, Raoul, 693 Çınar, Nilay, 914 Çini, Gülce, 914 Çirkin, Elif, 111, 124 Coşkun, Zeynep, 937 D Dalkıran, Renan, 898 Damar, Muhammet, 674 Demircan Keskin, Fatma, 85 Demircioğlu, Pınar, 265, 280 Demirel, Omer F., 724 Dengiz, Asiye Özge, 604 Dengiz, Berna, 737 Dengiz, Orhan, 604 Diktaş, Meltem, 753 Dönmez, Nurcan Demirok, 150 Dönmezer, Semih, 292 Dragomir, Diana, 769

© Springer Nature Switzerland AG 2019 N. M. Durakbasa and M. G. Gencyilmaz (Eds.): ISPR 2018, Proceedings of the International Symposium for Production Research 2018, pp. 961–963, 2019. https://doi.org/10.1007/978-3-319-92267-6

961

962 Dragomir, Mihai, 769 Drégelyi-Kiss, Ágota, 309 Dündar, Uğurcan, 426 Durakbasa, Lara, 391 Durakbasa, M. Numan, 464 Durakbasa, Numan M., 239, 309, 324, 438 Durakbasa, Numan, 402 Duran, Ferhat, 590 E Ekren, Gülay, 355, 367 Eliiyi, Deniz Türsel, 861, 914 Eliiyi, Uğur, 827 Emir, Oğuz, 546 Erbay, Hasan, 480 Ercan, Elif, 861 Erkollar, Alptekin, 355, 367 Erseven, Göksu, 849 Ervural, Beyzanur Cayir, 724 Etlioğlu, Mehmet, 44, 515, 530, 625 F Franczyk, Emilia, 660 G Gergin, Zeynep, 150, 254, 426, 546 Gogolewski, Damian, 224 Gökçe, Mahmut Ali, 922 Gökhan Kasapoğlu, N., 254 Gökler, Seda Hatice, 800 Gorycki, Łukasz, 218 Gözlü, Sıtkı, 412 Gruber, Andreas, 451 Güclü, Erol, 391 Gül, Aykut, 937 Gülen, Kemal Güven, 426 Güneş Gençyılmaz, M., 254, 426 Güneş, Buse, 150 H Holub, Michal, 402 I İlhan, Doğan Aybars, 426 İnce, Haydar, 625 Innerkofler, Mathias, 451 J Janecki, Dariusz, 747 Ješić, Dušan, 16 K Kabasakal, İnanç, 85 Kahraman, Ceren, 884

Author Index Kandiller, Levent, 872 Karaarslan, Cüneyt, 494 Karacaörenli, Ayşe, 101 Karakaş, Aslıhan, 849 Karakuş, Güzide, 345, 503 Karaman, Coşku, 937 Karataş, Samet, 546 Karkalos, Nikolaos E., 3 Karşıgil, Emre, 503 Kaymaz, Yunus, 85 Kazançoğlu, Yiğit, 884, 898 Kesikburun, Damla, 898, 949 Kıvanç, İpek, 75 Koç, Tuğba, 355, 367 Kocaaga, Ahmet Safa, 724 Konyalıoğlu, Aziz Kemal, 171, 181 Kovač, Pavel, 16 Koyuncuoğlu, Özdal, 44, 530, 565, 625 Kozior, Tomasz, 208 Králik, Marian, 379 Kräuter, Lukas, 391, 438 Külahlı, Emre, 827 Kundrák, János, 3 Kutlu-Gündoğdu, Fatma, 254 Kutup, Nejat, 949 L Laski, Pawel Andrzej, 201, 275 Łomża, Henryk, 201, 275 Lorcu, Fatma, 612 M Makieła, Włodzimierz, 218, 224 Makkai, Tamás, 3 Markopoulos, Angelos P., 3 Matras, Andrzej, 813 Mikó, Balázs, 65 Miko, E., 193 Mızrak, Ercan, 645 Mullaoğlu, Gizem, 937 Mutlu, İlker, 839 N Nagy, Antal, 3 Neamțu, Călin, 769 Nedić, Bogdan, 16 Nowakowski, L., 193 O Oberer, Birgit, 355, 367 Öner, Adalet, 849, 949 Onursal, Fatma Serab, 645 Othan, Can Burak, 884

Author Index Ozcan, Sel, 861, 914 Özdağoğlu, Aşkın, 111, 124, 674 Özdağoğlu, Güzin, 674 Özen, Ayşegül Eda, 914 Özgür-Ünlüakın, Demet, 75, 101 Özkan-Özen, Yeşim Deniz, 884 Özker, Hümra, 546 Ozkeser, Banu, 494 Öztop, Hande, 937 Özveri, Onur, 674 P Pasteka, Michael, 379 Pietrala, Dawid Sebastian, 275 Pietrala, Dawid, 201 Pirker, Wolfgang, 391 Polat, Leyla, 345, 503 Popescu, Daniela, 769 Popescu, Sorin, 769 Poszvek, Günther, 239 Pucher, Ernst, 451 R Rácz, Gábor, 65 Rodić, Dragan, 16 S Sagbas, Binnur, 324 Sahin, Merve Uzuner, 737 Sancar, Doğaç, 839 Savković, Borislav, 16 Semiz, Neslihan, 150 Şen, Arda Yiğit, 150 Şen, Gizem, 464 Şen, K. Öncü, 464 Seyis, Merve, 29 Şeyma Demir, G., 872 Sezen, Sueda, 898 Siray, Osman, 136 Şirvan, Umut, 949 Skoogh, Anders, 136 Skrzyniarz, M., 193 Soyuer, Haluk, 85 Stępień, Krzysztof, 747 Stopp, Joachim, 693 Struzikiewicz, Grzegorz, 660

963 T Tarić, Mirfad, 16 Tataroğlu, Derya, 898 Tekin, Ertuğrul, 44, 515, 530, 625 Tekin, Mahmut, 44, 515, 530, 565, 625, 753 Temizçeri, Talya, 872 Temur, Gül T., 412 Tolan, Ayşe, 922 Topaloğlu, Ebru Özer, 581 Torbalı, Ayşe Bilge, 254 Tülin Aktin, A., 426 Turanoglu Bekar, Ebru, 136 Türkali, Busenur, 75 U Ugur–Tuncer, Gamze, 438 Uluğ, İrem, 884 Üney-Yüksektepe, Fadime, 29, 254, 426 Ürük, Zerrin Funda, 800 V Vardin, Salih, 265, 280 W Wrzochal, Mateusz, 232 Y Yalçınkaya, H. Serdar, 581 Yapıcı, Nilay, 861 Yarıkcan, Gözde, 849 Yaşlı, Fatma, 781 Yıldırım, Nihan, 480 Yildiz Erduran, Gamze, 612 Yılmaz, Didem, 800 Yılmaz, Kevser, 111, 124 Yücel, Özgün, 849 Yunusoğlu, Pınar, 861 Yurtseven, Cansu, 922 Z Zaim, Selim, 724 Zębala, Wojciech, 660, 813 Zmarzły, Paweł, 333 Zwierzchowski, Jarosław, 201, 275

Proceedings of the International Symposium for Production Research 2018

Recommend Stories

Idea Transcript

Helpful Links

Smile Life

Get in touch